Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Population-based evaluation of two regimens for emergency contraception : a pharmacoepidemiologic study Leung, Vivian Wing Yan 2012

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2012_fall_leung_vivianwingyan.pdf [ 2.8MB ]
Metadata
JSON: 24-1.0073326.json
JSON-LD: 24-1.0073326-ld.json
RDF/XML (Pretty): 24-1.0073326-rdf.xml
RDF/JSON: 24-1.0073326-rdf.json
Turtle: 24-1.0073326-turtle.txt
N-Triples: 24-1.0073326-rdf-ntriples.txt
Original Record: 24-1.0073326-source.json
Full Text
24-1.0073326-fulltext.txt
Citation
24-1.0073326.ris

Full Text

POPULATION-BASED EVALUATION OF TWO REGIMENS FOR EMERGENCY CONTRACEPTION: A PHARMACOEPIDEMIOLOGIC STUDY  by  VIVIAN WING YAN LEUNG  B.Sc.(Pharm.), The University of British Columbia, 1999 Pharm.D., The University of British Columbia, 2006  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Pharmaceutical Sciences)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  October 2012  © Vivian Wing Yan Leung, 2012  ABSTRACT The Yuzpe (YZP) and levonorgestrel (LNG) emergency contraceptive (EC) regimens can prevent pregnancy after unprotected intercourse (UPI). This investigation was aimed at measuring EC users’ pregnancy rates under routine clinical conditions, comparing the effectiveness of the two regimens, reporting effectiveness using informative effect measures, and evaluating the impact of the timing of receiving ECs on the pregnancy rate. This cohort study included women who received pharmacist-prescribed ECs within a 25-month period. Treatment consent forms were matched to prescription records, which were linked to community and hospital medical records, to identify pregnancy-related administrative codes. Three experts adjudicated the data for time-compatible outcomes, and an observed pregnancy rate was determined for each regimen. The regimens were compared using logistic regression modeling, with adjustment for covariates. To evaluate EC effectiveness, observed pregnancy rates were compared against estimated expected pregnancy rates. The impact of EC timing was assessed by estimating the absolute risk reduction over time, and with logistic regression. Among 7493 women in the cohort, there were 99 (2.2%) and 94 (3.1%) compatible pregnancies in the LNG and YZP groups, respectively (p = 0.02). The crude odds ratio of pregnancy (LNG vs. YZP) was 0.71 (95% CI: 0.53 to 0.94). The adjusted odds ratio was ~0.65 (95% CI: 0.48 to 0.89) in the main analysis and sensitivity analyses. The expected pregnancy rate was ~4.1% in both groups. For LNG, the relative and absolute risk reductions were 47.6% and 2.0%, respectively [number needed to treat (NNT) = 50]. For YZP, the relative and absolute risk reductions were 24.4% and 1.0%, respectively (NNT = 100). Absolute risk reductions in the combined cohort decreased from 2% for EC dispensed within 24 hours to <1% beyond 72 hours after UPI. The adjusted odds of pregnancy increased by 1.27 (95% CI: 1.04-1.57) for each additional day after UPI. In  ii  conclusion, the LNG regimen was found to be superior to the YZP regimen for preventing pregnancy in routine clinical use. Given the limited effectiveness and treatment timeframe of these oral ECs, alternative forms of EC and regular contraception should be considered, and personalized according to each woman’s preferences.  iii  PREFACE This research was approved by the University of British Columbia Clinical and Children’s and Women’s Research Ethics Boards (certificate number CW05-0150/H05-70347). Previously published materials on the effectiveness and mechanism of action of hormonal emergency contraceptives, and the practice of emergency contraception in Canada, have been integrated into the Introduction and Discussion sections of this dissertation, and referenced accordingly. The publications are as follows: 1.  Leung VWY, Soon JA, Levine M. Measuring and reporting of the treatment effect of hormonal emergency contraceptives. Pharmacotherapy 2012;32:210-21.  2.  Leung VWY, Levine M, Soon JA. Evaluating emergency contraceptives using the preovulatory/postovulatory method. Pharmacotherapy 2011;31:496e-497e.  3.  Leung VWY, Levine M, Soon JA. Mechanisms of action of hormonal emergency contraceptives: a literature review. Pharmacotherapy 2010;30:158-68.  4.  Leung VWY, Soon JA, Levine M. Emergency contraception update: a Canadian perspective. Clin Pharmacol Ther 2008;83:177-80. For all of the above articles, the Ph.D. Candidate contributed substantially to the  conception and design of the contents, analysis and interpretation of the data, initial drafting of the manuscripts, revisions of the manuscripts for important intellectual contents, and approval of the final, published versions. In addition, the Ph.D. Candidate co-authored the following conference proceedings, based on abstract presentations of preliminary data on the investigations described in the Methods, Results, Discussion, and Conclusions sections of this dissertation:  iv  1.  Leung VWY, Soon JA, Lynd LD, Marra CA, Levine M. Comparison of pregnancy rates after two hormonal emergency contraceptives. Contraception 2011;84:333-4.  2.  Leung VWY, Soon JA, Marra CA, Lynd LD, Levine M. History of regular hormonal contraceptive use among emergency contraceptive users in British Columbia. J Popul Ther Clin Pharmacol 2011;18:e200-1.  3.  Leung VWY, Soon JA, Levine M, Lynd LD, Marra CA. Teen pregnancy rates following pharmacist-initiated emergency contraceptives. Pharmacoepidemiol Drug Saf 2009;18:S9.  4.  Leung VWY, Soon JA, Levine M, Lynd LD, Marra CA. Socioeconomic status of women obtaining emergency contraceptives from pharmacists in British Columbia. Can J Clin Pharmacol 2009;16:e228.  5.  Leung VWY, Soon JA, Levine M. Pregnancy-related outcomes among Canadian women obtaining two regimens for emergency contraception. Pharmacoepidemiol Drug Saf 2008;17:S260-1.  6.  Leung VWY, Soon JA, Levine M. Pregnancy and abortion rates among women obtaining emergency contraceptives from pharmacists in British Columbia. Can J Clin Pharmacol 2008;15:e496.  v  TABLE OF CONTENTS ABSTRACT ................................................................................................................................... ii PREFACE ..................................................................................................................................... iv TABLE OF CONTENTS ............................................................................................................ vi LIST OF TABLES ....................................................................................................................... xi LIST OF FIGURES ................................................................................................................... xiv LIST OF ABBREVIATIONS ................................................................................................... xvi ACKNOWLEDGEMENTS .................................................................................................... xviii DEDICATION............................................................................................................................ xix INTRODUCTION......................................................................................................................... 1 Emergency contraception............................................................................................................ 1 Mechanism of action of hormonal emergency contraceptives ................................................... 2 Statistical investigations.......................................................................................................... 4 Investigations of surrogate outcomes...................................................................................... 5 Investigations of pregnancy status .......................................................................................... 5 Conclusions on the mechanisms of action of hormonal emergency contraceptives ............... 6 Timing of hormonal emergency contraceptives ......................................................................... 6 Access to hormonal emergency contraceptives .......................................................................... 8 The population effect of expanding women’s access to hormonal emergency contraceptives 12 Estimating the magnitude of effect of hormonal emergency contraceptives ............................ 13 vi  Observed pregnancy rate....................................................................................................... 14 Expected pregnancy rate and the effectiveness estimate ...................................................... 16 Reporting the magnitude of effect of hormonal emergency contraceptives ............................. 20 Comparative effectiveness between the Yuzpe and levonorgestrel regimens .......................... 23 Pharmacist prescribing of emergency contraceptives in British Columbia and the present programme of research ............................................................................................................. 25 AIMS ............................................................................................................................................ 27 METHODS .................................................................................................................................. 28 Overview of research ................................................................................................................ 28 Ethics approval.......................................................................................................................... 28 Cohort definition ....................................................................................................................... 29 Data sources .............................................................................................................................. 29 Consent form data ................................................................................................................. 32 PharmaNet data ..................................................................................................................... 32 Medical Services Plan data ................................................................................................... 33 Hospital Separation data ....................................................................................................... 33 Canada Census data .............................................................................................................. 34 Initial data management ............................................................................................................ 34 Data linkage .............................................................................................................................. 35 Linking PharmaNet records and medical records ..................................................................... 35  vii  Matching PharmaNet records and consent records ................................................................... 35 Outcome ascertainment ............................................................................................................. 41 Screening for possible outcome-related codes ...................................................................... 41 Time profiles ......................................................................................................................... 42 Expert adjudication of compatibility .................................................................................... 45 Power estimation ....................................................................................................................... 45 Aim 1: To estimate the observed pregnancy rate (emergency contraceptive failure rate) associated with the levonorgestrel or Yuzpe regimen under conditions of routine use in the community ................................................................................................................................ 47 Aim 2: To compare the effect of the levonorgestrel and Yuzpe regimens on pregnancy under conditions of routine use, with adjustment for potential confounding ..................................... 48 Hypothesis............................................................................................................................. 48 Assessing and controlling for confounding .......................................................................... 48 Potential confounders in this study ....................................................................................... 48 Exploratory analyses ............................................................................................................. 50 Stratified analyses ................................................................................................................. 53 Multivariate logistic regression analyses .............................................................................. 53 Aim 3: To estimate an expected pregnancy rate and use both relative and absolute effect measures to report the apparent effectiveness of the levonorgestrel and Yuzpe regimens under conditions of routine use ........................................................................................................... 57  viii  Aim 4: To investigate the effect of timing of emergency contraceptives on pregnancy under conditions of routine use ........................................................................................................... 59 RESULTS .................................................................................................................................... 60 Results common to all four aims .............................................................................................. 60 Aim 1 ........................................................................................................................................ 63 Aim 2 ........................................................................................................................................ 67 Exploratory analyses ............................................................................................................. 68 Stratified analyses ................................................................................................................. 84 Multivariate logistic regression analyses .............................................................................. 95 Sensitivity analyses ............................................................................................................. 106 Aim 3 ...................................................................................................................................... 109 Aim 4 ...................................................................................................................................... 112 DISCUSSION ............................................................................................................................ 118 Discussion common to all four aims ....................................................................................... 118 Aim 1 ...................................................................................................................................... 122 Aim 2 ...................................................................................................................................... 124 Aim 3 ...................................................................................................................................... 133 Aim 4 ...................................................................................................................................... 138 Future directions ..................................................................................................................... 141 CONCLUSIONS ....................................................................................................................... 142  ix  REFERENCES .......................................................................................................................... 145 APPENDICES ........................................................................................................................... 164 Appendix A Pregnancy outcome data in clinical studies of the Yuzpe regimen for emergency contraception, containing a combination of ethinyl estradiol and either dl-norgestrel or levonorgestrel .......................................................................................................................... 164 Appendix B Pregnancy outcome data in clinical studies of the levonorgestrel regimen for emergency contraception ........................................................................................................ 174 Appendix C Reproduction of the standardized consent form completed by the women who received emergency contraceptives and by the prescribing pharmacists ................................ 180 Appendix D Methodology for correctly utilizing the age estimated from year of birth information in the matching procedure ................................................................................... 182 Appendix E Pregnancy-, prenatal care- and delivery-related administrative codes (excluding pregnancies with abortive outcomes)...................................................................................... 185 Appendix F Abortion-related administrative codes (not necessarily induced abortion codes) ................................................................................................................................................. 190  x  LIST OF TABLES Table 1 Oral hormonal emergency contraceptive products in Canada ........................................ 11 Table 2 Effectiveness of the Yuzpe and levonorgestrel regimens expressed as relative risk reduction, absolute risk reduction, and the number needed to treat in an investigation of emergency contraceptive studies .................................................................................................. 21 Table 3 Pertinent data from consent forms, the PharmaNet, Medical Services Plan, and Hospital Separation datafiles ....................................................................................................................... 30 Table 4 Power estimation over a range of plausible pregnancy rates in the Yuzpe group .......... 46 Table 5 Variables in the study and their data type ....................................................................... 52 Table 6 Characteristics of cohort records and excluded records .................................................. 62 Table 7 Characteristics of the women in the study cohort ........................................................... 64 Table 8 Sensitivity of the odds ratio (and 95% CI) of pregnancy for the levonorgestrel regimen group against the Yuzpe regimen group to the number of pregnancies in each group ................. 66 Table 9 Pregnancy outcome data based on the majority vote of the three experts ...................... 67 Table 10 Observed odds and log odds of pregnancy by age categories ....................................... 71 Table 11 Observed odds and log odds of pregnancy by cycle day quintiles ............................... 75 Table 12 Observed odds and log odds of pregnancy by time categories ..................................... 78 Table 13 Observed odds and log odds of pregnancy by income quintiles ................................... 81 Table 14 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the age variable .................................................................... 86 Table 15 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the cycle day variable .......................................................... 87  xi  Table 16 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the time variable .................................................................. 88 Table 17 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the income variable.............................................................. 89 Table 18 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of pregnancy variable ........................................ 90 Table 19 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of gynaecological condition variable ................. 91 Table 20 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of emergency contraceptive variable ................. 92 Table 21 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of hormonal contraceptive (excluding concurrent use) variable .................................................................................................................................. 93 Table 22 The unadjusted risk ratio for emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pooled risk ratios stratified on one covariate at a time .............................................. 94 Table 23 Odds ratio of pregnancy predicted by univariate regression modeling......................... 96 Table 24 Multivariate regression model-building to estimate the odds ratio of pregnancy for emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) as the explanatory variable of interest, adjusted for potential confounders ................................................................................ 105 Table 25 Adjusted odds ratio for emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) under various logistic regression model specifications for controlling the age and cycle day variables ............................................................................................................................... 107  xii  Table 26 Effect of confounding on the emergency contraceptive type-pregnancy odds ratio based on an array of assumptions ......................................................................................................... 108 Table 27 The magnitude of the treatment effect of emergency contraceptives expressed in a number of ways ........................................................................................................................... 111 Table 28 Expected and observed pregnancy rates by the time category (between unprotected intercourse and emergency contraceptive dispensation) and the absolute risk reduction .......... 112 Table 29 Multivariate regression model-building to estimate the odds ratio of pregnancy for the time between unprotected intercourse and emergency contraceptive dispensing as the explanatory variable of interest (in units of days: day 1, day 2, day 3, and beyond day 3), adjusted for potential confounders .............................................................................................. 115 Table 30 Multivariate regression model-building to estimate the odds ratio of pregnancy for the time between unprotected intercourse and emergency contraceptive dispensing as the explanatory variable of interest (in units of days: day 1, day 2, and beyond day 2), adjusted for potential confounders .................................................................................................................. 116  xiii  LIST OF FIGURES Figure 1 Effectiveness of hormonal emergency contraceptives, calculated by comparing observed pregnancy rates of 2.0% among Yuzpe regimen treatment arms and 1.7% among levonorgestrel regimen treatment arms against a range of expected pregnancy rates .................. 19 Figure 2 The number needed to treat to prevent one pregnancy, calculated by comparing observed pregnancy rates of 2.0% among Yuzpe regimen treatment arms and 1.7% among levonorgestrel regimen treatment arms against a range of expected pregnancy rates .................. 22 Figure 3 Algorithm for selecting PharmaNet records with unique combinations of age, emergency contraceptive type, dispensing date, and pharmacy local health area ........................ 37 Figure 4 Algorithm for selecting consent records with unique combinations of age, emergency contraceptive type, dispensing date, and pharmacy local health area ........................................... 38 Figure 5 Sample time profile ....................................................................................................... 44 Figure 6 The procedure for matching PharmaNet records of emergency contraceptive (EC) prescriptions and consent forms.................................................................................................... 61 Figure 7 Observed pregnancy rate by age categories .................................................................. 70 Figure 8 Log odds of pregnancy by age categories ..................................................................... 72 Figure 9 Observed pregnancy rate by nine cycle day categories ................................................. 73 Figure 10 Observed pregnancy rate by cycle day quintiles ......................................................... 74 Figure 11 Log odds of pregnancy by cycle day quintiles ............................................................ 76 Figure 12 Observed pregnancy rate by time categories ............................................................... 77 Figure 13 Log odds of pregnancy by time categories .................................................................. 79 Figure 14 Observed pregnancy rate by income quintile .............................................................. 80 Figure 15 Log odds of pregnancy by average population neighbourhood income quintiles ....... 82  xiv  Figure 16 Log odds of pregnancy by median household neighbourhood income quintiles ........ 83 Figure 17 Probability of pregnancy predicted by incorporating age in univariate regression as a linear term as well as a quadratic term........................................................................................ 101 Figure 18 Log odds of pregnancy predicted by incorporating age in univariate regression as a linear term as well as a quadratic term........................................................................................ 102 Figure 19 Probability of pregnancy predicted by incorporating cycle day in univariate regression as a linear term as well as a quadratic term ................................................................................ 103 Figure 20 Log odds of pregnancy predicted by incorporating cycle day in univariate regression as a linear term as well as a quadratic term ................................................................................ 104 Figure 21 The number of pregnancies expected after unprotected intercourse on menstrual cycle Day 1 to Day 40 among women in the cohort ............................................................................ 110 Figure 22 Expected and observed pregnancy rates by the time category (between unprotected intercourse and emergency contraceptive dispensing) ............................................................... 113 Figure 23 Odds ratios (and 95% CI) of pregnancy for the time between unprotected intercourse and emergency contraceptive dispensing as the explanatory variable of interest (in categories of days: day 1, day 2, and beyond day 2), adjusted for emergency contraceptive regimen type, age, and cycle day............................................................................................................................... 117  xv  LIST OF ABBREVIATIONS AIC  Akaike information criterion  ARR  Absolute risk reduction  BCLHD  British Columbia Linked Health Database  CCP  Canadian Classification of Diagnostic, Therapeutic, and Surgical Procedures  CI  Confidence interval  EC  Emergency contraceptive  HC  Hormonal contraceptive  ICD-9  The International Classification of Diseases, Ninth Revision  IUD  Intrauterine device  LH  Luteinizing hormone  LHA  Local health area  LMP  Last menstrual period  LNG  Levonorgestrel  mg  Milligram  MSP  Medical Services Plan  NNT  Number needed to treat  OR  Odds ratio  OTC  Over-the-counter  PHN  Personal health number  RR  Risk ratio  RRMH  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method  xvi  SD  Standard deviation  SE  Standard error  UPI  Unprotected intercourse  VIF  Variance inflation factor  WHO  World Health Organization  YZP  Yuzpe  xvii  ACKNOWLEDGEMENTS I am very grateful for the tutelage of my supervisor, Dr. Marc Levine, whose wisdom and passion for science inspired me to pursue knowledge with an inquisitive mind, and meet challenges with enthusiasm. I also sincerely appreciate my research committee members, Dr. Thomas Chang (chair), Dr. Judith Soon, Dr. Carlo Marra, Dr. Larry Lynd, and Dr. Jean-Paul Collet, for their mentorship and insightful guidance. The constant support and encouragement from Dr. Judith Soon helped me develop as a researcher, and made the past few years so much more rewarding. I would like to thank the clinical experts who assisted with outcome ascertainment and provided me with invaluable advice on this research project, Ms. Maja Grubisic for assistance with power estimation, and Mr. Pat McCrea of the Blue Thorn Research and Analysis Group for technical support with the data linkage. I am also grateful for the scholarship support from the Canadian Institutes of Health Research, Michael Smith Foundation for Health Research, Association of Faculties of Pharmacy of Canada, and the University of British Columbia. Funding for this research project was generously provided by the Canadian Institutes of Health Research, British Columbia Medical Services Foundation, and British Columbia Ministry of Health. My love and thanks go to my Mom, Dad, sister Karen, and fiancé Tyler. They are the greatest blessings in my life.  xviii  DEDICATION  To my grandparents  xix  INTRODUCTION Emergency contraception Emergency contraception, also called postcoital contraception, is a method of birth control. It is different from other methods in that it is indicated for use after an act of unprotected intercourse (UPI). Urgency is implied in the term “emergency contraception,” because currently available medications and devices are effective only if they are used shortly after UPI, and the methods are more effective the earlier that they are initiated.1 The modern-day concept of emergency contraception dates back to the 1920s.2 Early studies experimented with estrogenic ovarian extracts and synthetic estrogens in animals and in humans.2 The first major trial was conducted in 1963 using diethylstilbestrol or ethinyl estradiol postcoitally.3 In the 1970s, Canadian physician Albert Yuzpe and his team piloted the use of an estrogen-progestin combination.4,5 The Yuzpe (YZP) regimen (two doses, each containing ethinyl estradiol 0.1 mg and levonorgestrel 0.5 mg or norgestrel 1.0 mg, 12 hours apart) became a commonly used oral emergency contraceptive (EC). Danazol was investigated for possible use as an EC in the early 1990s but it was not consistently effective in studies and has since been abandoned.6-8 Researchers began to investigate the use of a progestin-only method of emergency contraception in 1973.3 In Canada and the United States, the levonorgestrel (LNG) regimen (two doses of LNG 0.75 mg 12 hours apart) became an approved progestin-only EC in the late 1990s.9 The dosing of the LNG regimen was subsequently modified to a single dose of 1.5 mg.9 Over the past decade, the YZP and LNG regimens were the most commonly used ECs in Canada and the United States and are the medications under study in the present programme of research.  1  In Canada, emergency contraception is available as oral medication and also as an intrauterine device (IUD).10 Compared to oral medications, IUDs are a far less popular alternative because they need to be inserted by a physician, are more invasive, and have the potential to cause discomfort, bleeding, and pelvic inflammatory disease.10,11 However, postcoital IUDs have the added benefit of providing ongoing contraception when they remain in place.10,12 It should be noted that mifepristone (RU486) can also be used as an EC,10 but it is more commonly used for medical abortion because of its antiprogesterone activity. For this reason, its availability is restricted in many countries. Recently, ulipristal acetate, a selective progesterone receptor modulator, has become available in Europe and in the United States, with indication for emergency contraception. It is licensed for use up to 120 hours after UPI, but it is currently not as easily accessible to women as the LNG regimen, and it will likely remain a prescription medication while safety data accumulate.13 Other potential EC modalities include a five-day regimen of meloxicam, and a vaginal ring containing ethinyl estradiol and Nesterone® (a trademarked progestin).14,15 Emergency contraceptives are currently available in more than 140 countries.16 The reasons for women accessing ECs include not having used a contraceptive before or during intercourse, using a contraceptive inconsistently or incorrectly, and method failure or deficiency.17  Mechanism of action of hormonal emergency contraceptives As the general public and clinicians became aware of the availability of oral hormonal ECs, there were questions and conflicting opinions about their mechanism of action.18-23 Some women perceive ECs to be abortifacients and refuse to use them on moral or religious grounds,24  2  and some pharmacists and other clinicians have refused to provide ECs based on a similar rationale.25-29 The debate over whether ECs affect established pregnancies depends on how ECs are believed to act, as well as when pregnancy is thought to begin.30 Theoretically, ECs can interfere with a number of steps in the female reproductive process.3 These include, but are not limited to, ovulation, sperm migration, and hormonal and endometrial effects. Follicular development and the eventual release of an ovum from the follicle are regulated by hormonal changes. Public objection to ECs mainly concerns effects other than those that lead to disruption of ovulation, especially potential postfertilization activity. For people who define pregnancy as beginning at implantation, an abortifacient is an agent that interferes with subsequent processes. Although theoretical concerns have been expressed by some for using ECs, which are high doses of exogenous hormones, there are currently no clinical data indicating that the YZP or LNG regimen will disrupt an implanted embryo or terminate an established pregnancy,11,31 nor have high doses of routinely administered oral contraceptives been shown to do so.31 Therefore, EC can be seen as a preventive strategy, for those who believe that pregnancy begins at implantation.32 For those who consider pregnancy to begin with the completion of fertilization, an abortifacient is an agent that interferes with any post-fertilization event, including implantation.32,33 A number of researchers have attempted to clarify whether ECs act by postfertilization mechanisms.32,34-38 From a clinical perspective, many people would consider ECs acceptable for use if they have no effect after fertilization has occurred.32,37 For those who object to the use of ECs because there is insufficient evidence to completely exclude the possibility of interference with implantation, ECs would not be acceptable contraceptives.32,39  3  To summarize the data, a literature search was conducted using PubMed and EMBASE to identify articles on the mechanisms of action of ECs.32 The goal was to select studies in humans and review articles intended to explain the mechanisms of action of the YZP or LNG regimen, as well as studies that had pertinent data on the effects of these regimens on ovulation, sperm activity, hormones, and the endometrium. Studies were classified into one of three investigative approaches: (1)  Statistical investigations examine results from studies designed to estimate the effectiveness of ECs, and they use statistical models to determine whether ECs would be as effective as reported if they act only by preventing or delaying ovulation.  (2)  Investigations of surrogate outcomes use empirical data from direct observation to determine whether ECs interfere with ovulation, sperm migration, hormonal levels, and endometrial receptivity to implantation.  (3)  Investigations of pregnancy status follow EC users prospectively for their pregnancy status, and then evaluate EC effectiveness among women who received EC before ovulation and after ovulation, respectively.  Statistical investigations The results of statistical investigations have suggested that ECs would not be as effective as has been estimated in clinical studies if they work only by an anovulation mechanism.34-36,40 However, there is growing evidence to suggest that the expected rates of pregnancy used to derive effectiveness in these studies have been overestimated.20,36,37,41,42 If this is the case, the gap between observed EC effectiveness and the theoretical effectiveness by an anovulation mechanism alone would be narrower, or nonexistent. Also, a number of assumptions are required  4  in determining the theoretical effectiveness of ECs, given the limited information the investigators had available about the clinical studies they cited. Other possible explanations include pre-fertilization mechanisms other than ovulatory disturbance, and the possibility that ECs may exert postfertilization actions, including alteration of endometrial receptivity, and interference with corpus luteum function and implantation.  Investigations of surrogate outcomes The effects of ECs on ovulation have been observed using ultrasonography. The evidence strongly indicates that suppression or delay of ovulation is a mechanism when ECs are used before the luteinizing hormone (LH) surge.43-48 In the interval between the LH surge and ovulation, the data are less clear. It appears that the closer to the time of ovulation that ECs are administered, the less likely they are to interfere with ovulation.45,48 Other potential mechanisms include interference with sperm migration, changes in hormone levels, and physical and biochemical changes that can alter endometrial receptivity to implantation. However, the data in these areas are sparse.32 The strongest data on physiological processes have indicated a delay or suppression of ovulation as a mechanism. It seems that even if ECs have a direct inhibitory effect on implantation, the earlier-stage effect of anovulation or delayed ovulation would have prevented fertilization; if fertilization is prevented, downstream processes will not occur.3  Investigations of pregnancy status Two studies37,38 used what was termed the preovulatory/postovulatory method to investigate the mechanisms of action of the LNG regimen.49 In these studies, the investigators evaluated EC effectiveness among women who received it before ovulation and after ovulation,  5  respectively, by comparing observed pregnancies to theoretical numbers of pregnancies expected. In the five days before ovulation, the data suggest that the LNG regimen is effective in preventing pregnancy.49 On the day of ovulation and in the (postovulatory) luteal phase, the results suggest that the regimen would be ineffective.49 The studies conducted so far have included small numbers of women; however, this approach has the potential to definitively address the question.49  Conclusions on the mechanisms of action of hormonal emergency contraceptives Taken together, the evidence strongly indicates disruption of ovulation as a mechanism of action when ECs are taken before ovulation, and this mechanism precludes fertilization and the downstream process of implantation.32 After ovulation, the available data suggest that EC administration is ineffective, and postovulatory effects are unlikely.49 For women who object to contraceptive methods for which there is insufficient evidence to completely exclude the possibility of postfertilization interference, ECs are not acceptable contraceptives.32,39 In contrast, for those who consider implantation or later events to be the beginning of pregnancy, there is substantial evidence for a non-abortive mechanism.32  Timing of hormonal emergency contraceptives Media reports often refer to oral ECs as “morning-after pills.” This term misrepresents ECs because it seems to imply that a woman needs to wait until the morning after an act of UPI to initiate EC, and it also suggests that the medication is effective only if taken the morning after.2 A number of studies have investigated the period of time within which ECs may be effective after an act of UPI.  6  In 1996, Trussell et al. reviewed nine trials that reported the number of pregnancies for women treated with the YZP regimen on the first, second, and third day after unprotected intercourse.50 The pooled data showed no difference in pregnancy rate over these three days, suggesting that the effectiveness of the YZP regimen might not diminish over time within the first 72 hours. However, the results of this study were contested in a subsequent report. In 1998, a large trial was conducted on the YZP and LNG regimens.51 The data from that trial were reanalyzed for the relationship between the timing of EC and pregnancy rate.52 Using the combined data from the YZP and LNG groups, the investigators identified a consistent linear relationship in the pregnancy rate, which was, at the lowest, 0.5% among women who received EC within 12 hours after UPI.52 In the time interval between 61 and 72 hours after UPI (the latest category), the pregnancy rate was 4.1%.52 After adjusting for age, weight, body mass index, gravidity, cycle length, day of the menstrual cycle on which UPI occurred, and previous EC use, the timing of EC (in an unspecified unit of time) was significantly associated with pregnancy, with an odds ratio of 1.46 [95% confidence interval (CI): 1.20-1.77].52 Early studies of the YZP and LNG regimens tended to exclude women who presented to clinics beyond 72 hours after UPI.5,51,53 Consequently, the YZP and LNG regimens were approved for use up to 72 hours after UPI. However, the 72-hour cut-off was arbitrary and so, based on the data available in the late 1990s and early 2000s, researchers investigated the possibility of extending the treatment time frame to 120 hours. Rodrigues et al. analyzed the pregnancy rates of women who sought EC within 72 hours after UPI and those who presented between 72 and 120 hours.54 In both groups, treatment with the YZP regimen was associated with a significantly lower pregnancy rate than the rate that  7  would be expected in the absence of EC.54 However, the two groups were not directly compared because the study was insufficiently powered for such a comparison.54 Subsequently, Ellertson et al. compared early presenters (within 72 hours) with an otherwise similar group of late presenters (between 72 and 120 hours).55 Both groups were offered the YZP regimen and were found to have similar pregnancy rates.55 As was expected, due to a small sample size of only 111 women, the late-presenting group had a wide confidence interval for its pregnancy rate: 0.2-6.8% with perfect use and 0.9-9.0% with typical use of EC.55 Based on these results, the authors concluded that the 72-hour time frame appeared to be “needlessly restrictive.” Recently, Piaggio et al. analyzed the combined results from four studies comprising a total of 6794 women treated with the LNG regimen.56 The observed pregnancy rate was stratified by the timing of EC. On the first, second, third, and fourth day, the pregnancy rate was 1.0%, 0.7%, 1.6%, and 0.8%, respectively. On the fifth day, the pregnancy rate was 5.2% (the implications of these results for the effectiveness of the regimen will be explained in the section below on the magnitude of effect). The authors also performed logistic regression analysis to estimate the daily odds ratio of pregnancy, relative to the first day, but the results were not adjusted for potential confounders. On the fifth day, the odds ratio was 5.8 (95% CI: 2.9-11.8). Current Canadian guidelines recommend that hormonal ECs be offered up to 5 days after UPI.10,11 The effect of timing on EC effectiveness in the routine clinical setting is investigated in the present programme of research.  Access to hormonal emergency contraceptives Until 6-7 years ago, women in Canada and the United States could only access ECs by obtaining a prescription from a physician, walk-in clinic, hospital emergency department, or  8  specialized clinics for family planning or adolescents. In Canada, regulatory changes in three provinces gave pharmacists prescriptive authority to prescribe ECs to women of all ages: as of December 2000 in British Columbia, December 2001 in Quebec, and September 2003 in Saskatchewan.19 In April 2005, Health Canada announced nationwide approval for the LNG regimen to be available without a prescription from a physician or pharmacist, but kept behind the pharmacy counter. This policy was implemented in British Columbia as of May 2007. Access to the LNG regimen was further enhanced following a decision by the National Association of Pharmacy Regulatory Authorities in May 2008 to recommend full over-the-counter (OTC) access directly from community pharmacy shelves. Despite the availability of OTC access, some women opt to have the transaction documented on PharmaNet (the province-wide prescription claims database)57 for coverage and insurance company reimbursement purposes; pharmacist- or physician-initiated prescriptions are required in those instances. In the United States, intense lobbying from professional and lay groups for enhanced access to ECs brought about regulatory changes as well. In April 2006, the United States Food and Drug Administration approved non-prescription access to the LNG regimen by women aged 18 years and older.58 The approval was extended to those 17 years of age in April 2009.58 In Canada and the United States, packages containing the YZP or LNG regimen have enhanced the convenience of oral hormonal ECs. Currently, the LNG regimen is available in brand and generic packages of the medication (Plan B®, NorLevo®, Next Choice®); whereas, the dedicated EC product for the YZP regimen (Preven®) had been discontinued in the early 2000s. However, women have been able to access the YZP regimen using tablets from packages of regular oral contraceptives, in numbers that supply the equivalent content.59 Until the discontinuation of Ovral 21® and Ovral 28®, women could access the YZP regimen by taking  9  immediately two tablets from either product, followed by another two tablets 12 hours later. Currently, the YZP regimen is available from Alesse®, Aviane®, Min-Ovral®, or Portia® contraceptives in the numbers shown in Table 1.  10  Table 1 Oral hormonal emergency contraceptive products in Canada Product name Ovral 21®, Ovral 28®*  Active ingredient(s)  Directions for use  Ethinyl estradiol 0.05 mg +  2 tablets q12h x 2 doses  levonorgestrel 0.25 mg Preven®*  Ethinyl estradiol 0.05 mg +  2 tablets q12h x 2 doses  levonorgestrel 0.25 mg Alesse 21®, Alesse 28®  Ethinyl estradiol 0.02 mg +  5 tablets q12h x 2 doses  levonorgestrel 0.10 mg Aviane 21®, Aviane 28®  Ethinyl estradiol 0.02 mg +  5 tablets q12h x 2 doses  levonorgestrel 0.10 mg Min-Ovral 21®, Min-Ovral 28®  Ethinyl estradiol 0.03 mg +  4 tablets q12h x 2 doses  levonorgestrel 0.15 mg Portia 21®, Portia 28®  Ethinyl estradiol 0.03 mg +  4 tablets q12h x 2 doses  levonorgestrel 0.15 mg Seasonale®  Ethinyl estradiol 0.03 mg +  4 tablets q12h x 2 doses  levonorgestrel 0.15 mg Plan B®  Levonorgestrel 0.75 mg  2 tablets x 1 dose  Norlevo®  Levonorgestrel 0.75 mg  2 tablets x 1 dose  Next Choice®  Levonorgestrel 0.75 mg  2 tablets x 1 dose  * Ovral 21®, Ovral 28®, and Preven® had been discontinued by the manufacturers and are currently not available  11  The population effect of expanding women’s access to hormonal emergency contraceptives At the turn of the century and over the ensuing decade, many reproductive health advocates believed that widespread use of ECs could substantially reduce the incidence of unintended pregnancy.60 In many countries, EC use was enthusiastically promoted,60 and the aforementioned regulatory changes of EC scheduling were regarded as important interventions toward lower rates of unintended pregnancy and subsequent abortion.22 However, expanded access to ECs has not been found to meet these expectations. A systematic review conducted by Raymond et al. in 2007 indicated that increased access to ECs did not reduce unwanted pregnancy or abortion rates.61 The authors reviewed 23 studies that compared the effect of different levels of access to ECs. Fourteen of these studies assigned women to either an increased EC access (intervention) group or a control group and followed up prospectively to ascertain outcomes. The other nine were observational studies that compared data before and after implementation of an intervention designed to increase access (e.g., legislated change to an OTC status, advanced provision so that women had tablets on hand, a telephone prescription service, etc.). None of the 23 studies observed a statistically significant reduction in pregnancy or abortion rates,61 thus calling into question the ability of ECs as a preventive contraceptive strategy to impact the health and social problem of unintended pregnancy. To date, only one study has reported a population effect on abortion rate.62 The ecological study from Utah in the United States found that statewide increase in the distribution of ECs coincided with a decrease in abortion rate over a 7-year period from 2000 to 2006.62 There are several plausible explanations for the apparent lack of effect of ECs in the population.60 First, ECs would not have an appreciable impact if they are only “weakly efficacious.”60,61 Second, some of the studies in the aforementioned systematic review61 were 12  methodologically limited, including being insufficiently powered to detect a relatively small change in pregnancy rate.60 Third, ECs were likely underused relative to the number of acts of UPI, despite expanded access in recent years and intensified promotion to the public. This last explanation is particularly noteworthy because it may reflect women’s knowledge about ECs and their perceived risk of pregnancy.60,61,63-66 If women who are aware of ECs and have requested a regimen on at least one occasion do not use them every time they are indicated, then untreated acts of intercourse can lead to an overestimation of the EC failure rate, and ECs may appear to be an ineffective contraceptive strategy in a population over time.  Estimating the magnitude of effect of hormonal emergency contraceptives In principle, the efficacy of ECs should be assessed by comparing the observed pregnancy rate (i.e., the EC failure rate) against the pregnancy rate in a placebo group.67 However, no placebo group has ever been used in EC studies. After more than three decades of clinical experience with these medications, the research community generally agrees that ECs prevent pregnancy at least some of the time and concedes that a placebo-controlled study involving EC seekers would be impossible in the absence of clinical equipoise.61,68,69 Indeed, a recent survey concluded that the design of an ethical placebo-controlled trial would require inclusion criteria so stringent that the trial would be practically infeasible to carry out.70 Without a placebo arm, pregnancy rates expected in the absence of treatment cannot be derived directly from EC studies.67 In place of a placebo group, EC studies typically use a theoretical comparator to indirectly derive the baseline (untreated) pregnancy risk.19,31,69,71 Measured this way, the  13  magnitude of effect is considered to be an effectiveness estimate, rather than an efficacy estimate. In studies of ECs, effectiveness is typically calculated using this formula:67,71 1 - (observed pregnancy rate / expected pregnancy rate) The resulting percentage is the effectiveness rate.71 It has also been referred to as the prevented fraction—the proportion of pregnancies averted by EC treatment.31 As can be seen from this formula, the only variables in estimating EC effectiveness are the observed pregnancy rate and the expected pregnancy rate. Each of these two variables is discussed in detail below.  Observed pregnancy rate The observed pregnancy rate in any given EC study is the number of pregnancies divided by the number of women at risk, or the number of treated menstrual cycles. In a recent investigation, a literature search was performed to identify studies that measured the observed pregnancy rate after EC use, and the data were pooled to derive a summary estimate.67 In this review, studies were selected for analysis if they had an explicit method of follow-up to systematically evaluate pregnancy outcome after a single course of EC treatment within a given menstrual cycle. Studies that restricted participation to women in the fertile period of their menstrual cycle were excluded because the pregnancy rates from these subsets could bias the pooled pregnancy rate upward. The observed pregnancy rate of each included study was abstracted. To optimize accuracy of the resulting summary estimates, women whose pregnancies were known to investigators as having resulted from intercourse other than the index act that prompted EC use6,51 were excluded from the numerator and denominator of the observed pregnancy rate.  14  A total of 34 studies were selected for summary and analysis.5,6,8,13,37,51,53-55,72-97 Three of the studies contributed data to the analyses of both EC regimens.51,85,90 The results67 are summarized in Appendix A for data in YZP treatment arms and Appendix B for LNG treatment arms, respectively. In the primary analysis,67 pregnancy data among subjects with known outcome were pooled to estimate summary observed pregnancy rates and their 95% CIs using the beta-binomial model, which accounted for heterogeneity across studies.98 The analysis was performed using SAS® version 9.2 (SAS Institute Inc., Cary, NC) and Ian Wakeling’s macro for fitting a betabinomial model.98 Pooled pregnancy rates were obtained for the YZP and LNG regimens. Among study subjects with known outcome, pooled observed pregnancy rates for studies of YZP and LNG treatment arms were 2.0% (95% CI: 1.5-2.5%) and 1.7% (95% CI: 1.2-2.2%), respectively.67 An unusually high pregnancy rate of 8.3% was reported in the YZP-treated group (n = 60) in one study.90 When this study was excluded (both the YZP and LNG arms), the pooled pregnancy rates were 1.9% (95% CI: 1.5-2.4%) and 1.8% (95% CI: 1.2-2.3%) for the YZP and LNG regimens, respectively.67 In eight studies, the investigators confirmed that the study participants were not already pregnant at the time of enrolment.13,51,54,87,90,92,95,97 One of those studies also excluded pregnancies that were known to have resulted from intercourse subsequent to the index act based on gestational dating.13 Under these more stringent conditions, the pooled pregnancy rates were 2.7% (95% CI: 1.0-4.4%) among four YZP study arms and 1.4% (95% CI: 0.8-2.0%) among six LNG study arms.67 A sensitivity analysis was performed to examine the impact of including women with unknown outcome over a range of pregnancy risk assumptions.67 The proportion of study participants with no outcome data was found to range from 0 to 33% (median 5%).67 When  15  women with unknown outcome were assumed to be entirely non-pregnant after treatment, the overall pooled pregnancy rates were 1.9% (95% CI: 1.4-2.3%) and 1.6% (95% CI: 1.2-2.1%) among YZP- and LNG-treated women, respectively.67 If 5% of these women were assumed to have become pregnant, then the overall pooled pregnancy rates would be 2.3% (95% CI: 1.82.8%) and 1.9% (95% CI: 1.4-2.4%), respectively.67 If 10% of the women became pregnant, the pooled pregnancy rates would be 2.7% (95% CI: 2.1-3.3%) and 2.1% (95% CI: 1.6-2.7%), respectively.67  Expected pregnancy rate and the effectiveness estimate While the observed pregnancy rate is directly available from an EC study, the expected pregnancy rate needs to be estimated using external information. The expected pregnancy rate in a cohort of women is the overall probability of pregnancy,19 calculated by multiplying the probability of pregnancy for UPI that occurred on each given day of the menstrual cycle, with the number of women who had UPI on each corresponding day. Depending upon the expected pregnancy rate being used, the effectiveness estimate derived from the above formula will vary. Suppose that, in a hypothetical study, the observed pregnancy rate after EC use is 2%. If the women in the study are expected to have an 8% pregnancy rate without EC, EC effectiveness would be 75%. However, if the expected pregnancy rate is 4%, then, at the same observed pregnancy rate of 2%, EC effectiveness would be 50%. Therefore, the major challenge in estimating EC effectiveness is to quantify and justify the pregnancy rate that would be expected among EC seekers or users, if no contraception had been used.67 Over the years, multiple sets of cycle day-specific conception probability estimates have been developed and applied to cohorts of EC seekers to estimate an expected pregnancy rate.99-  16  102  As a result, there is substantial variability in the reported EC effectiveness, ranging from 52%  to 94%.10,51,69,71,103,104 In early studies, probabilities of pregnancy were estimated for intercourse that occurred around the day of ovulation.71,102,103 Pooled data from studies that used an older set of probabilities suggested that the YZP and LNG regimens prevent 74%71,104 and 79%104 of expected pregnancies, respectively. However, these estimations relied upon self-reported cycle day and an estimated day of ovulation, which may not be accurate.19,37,41,42,105 Using this method, the expected cycle day of ovulation was assumed to be 13 or 14 days before the end of a cycle.71 However, the actual cycle day of ovulation is known to vary from cycle to cycle, and only women with regular menstrual cycles could be included in the analysis,42,101,106,107 thus limiting the usefulness of this approach. Another method was proposed by Wilcox et al. in 2001;101 it became known as the Wilcox Method or Wilcox Approach.67,104 Instead of estimating the probability of pregnancy with an act of intercourse relative to ovulation, it was estimated relative to intercourse on a given cycle day, counting from onset of the last menstrual period (LMP) and with adjustment for natural variation in the day of ovulation.101 Using this approach, Trussell et al.69 estimated that the YZP regimen is only 46.8-53.0% effective. More recently, a different approach was introduced by Mikolajczyk and Stanford.107 The authors estimated the daily probability of pregnancy with reference to the last day of the cycle (the day before subsequent menses), while taking into account estimated daily fecundity around the peak day and distribution of the peak day.107 This method would require that the previous cycle length be known, and it assumes that the length of the luteal phase is independent of the overall cycle length.107  17  Of all the proposed methods for estimating the expected pregnancy rate, the Wilcox Method101 is feasible in the absence of information on the day of ovulation, cycle length, and other predictors.67 In two large cohorts that were studied using the Wilcox Method, the expected pregnancy rates among women seeking ECs were 3.9% and 4.1%, respectively.69,108 At the observed pregnancy rates of 2.0% and 1.7% (reported above in the pooled analysis) for the YZP and LNG regimens, and at an expected pregnancy rate of 4.0%, the relative EC effectiveness would be 50.0% and 57.5%, respectively. These results are far lower than has been reported in earlier studies that used expected pregnancy rates of 7-8%.67 As shown in Figure 1, the estimated effectiveness rate of each of the two EC regimens can vary from approximately 50.0% to 78.8%, depending on the expected pregnancy rate used to derive it.67 Estimation of EC effectiveness is dependent on the observed and expected pregnancy rates, and the latter would be different from one method to another. For this reason, effectiveness estimates from different studies are not easily comparable. The expected pregnancy rate also affects the assessment of EC effectiveness with respect to time of EC use after UPI. In the study by Piaggio et al. (mentioned above in the section on the timing of ECs), the observed pregnancy rate of women treated with the LNG regimen on the first, second, third, fourth, and fifth day after UPI was 1.0%, 0.7%, 1.6%, 0.8%, and 5.2%, respectively.56 The expected pregnancy rate was not reported in this study; if it is approximately 4%,69,108 then the regimen can be seen as being effective when taken within the first four days, but ineffective by the fifth day after UPI.  18  Figure 1 Effectiveness of hormonal emergency contraceptives, calculated by comparing observed pregnancy rates of 2.0% among Yuzpe regimen treatment arms and 1.7% among levonorgestrel regimen treatment arms against a range of expected pregnancy rates (Reproduced with permission of John Wiley and Sons)67  19  Reporting the magnitude of effect of hormonal emergency contraceptives As explained above, EC effectiveness (or prevented fraction) is typically measured using the formula, 1 - (observed pregnancy rate / expected pregnancy rate).67,71 This approach produces a relative measure of effect that is, essentially, an estimate of the relative risk reduction, in the context of emergency contraception. In scientific reports and the clinical literature, EC effectiveness is almost exclusively reported as an estimated relative risk reduction—as the proportional reduction of pregnancy risk from an expected rate to the observed. The magnitude of effect can be estimated in absolute terms as well. The absolute risk reduction is the arithmetic difference between the control group event rate and the treatment group event rate.109 In the context of emergency contraception, the absolute risk reduction is the expected pregnancy rate minus the observed pregnancy rate. A clinically useful effect measure, the number needed to treat (NNT), can be derived from the absolute risk reduction.110 In EC studies, the NNT to prevent one pregnancy is the inverse of the absolute risk reduction. Recall that in the literature review described above the observed pregnancy rate was 2.0% and 1.7% in the YZP and LNG groups, respectively.67 The estimated relative risk reduction, absolute risk reduction, and NNT would vary, depending on the expected pregnancy rate (e.g., 4.0% vs. 8.0%).67 As shown in Table 2, the relative risk reduction appeared to communicate a more impressive magnitude of effect than the absolute risk reduction.67 The NNT was smaller when the expected pregnancy rate is assumed to be 8%, compared to 4%. However, the absolute difference in the NNT between the YZP and LNG groups would be small with a higher expected pregnancy rate (Figure 2).67  20  Table 2 Effectiveness of the Yuzpe and levonorgestrel regimens expressed as relative risk reduction, absolute risk reduction, and the number needed to treat in an investigation of emergency contraceptive studies (Reproduced with permission of John Wiley and Sons)67 Regimen  Yuzpe  Levonorgestrel  Expected pregnancy rate  Observed pregnancy rate  Relative risk reduction  Absolute risk reduction  Number needed to treat  4.0%  2.0%  50.0%  2.0%  50  8.0%  2.0%  75.0%  6.0%  17  4.0%  1.7%  57.5%  2.3%  43  8.0%  1.7%  78.8%  6.3%  16  21  Figure 2 The number needed to treat to prevent one pregnancy, calculated by comparing observed pregnancy rates of 2.0% among Yuzpe regimen treatment arms and 1.7% among levonorgestrel regimen treatment arms against a range of expected pregnancy rates (Reproduced with permission of John Wiley and Sons)67  22  Comparative effectiveness between the Yuzpe and levonorgestrel regimens The YZP and LNG regimens have been compared in three randomized trials. The World Health Organization (WHO) Task Force on Postovulatory Methods of Fertility Regulation conducted two comparative trials in the 1990s. In the first, an open-label study, Ho and Kwan randomized 880 women seeking ECs to receive the YZP or LNG regimen.85 The observed pregnancy rates were 3.5% and 2.9% in the YZP- and LNG-treated groups, respectively.85 After excluding women who had acts of intercourse subsequent to EC treatment, the pregnancy rates were 2.7% and 2.4%, respectively, and the difference was not statistically significant.85 Using an external set of conception probabilities developed by Dixon et al.102, effectiveness of the regimens was estimated to be 59.0% and 59.6%, respectively. The second investigation was a randomized, double-blind, multi-national trial that included close to two thousand women.51 Pregnancy rates of 3.2% (95% CI: 2.2-4.5%) and 1.1% (95% CI: 0.6%-2.0%) were observed in the YZP- and LNG-treated groups, respectively.51 Four women were found to have been pregnant at the time of enrolment; when these women were excluded, pregnancy rates were 2.9% (95% CI: 1.9-4.1) and 1.0% (95% CI: 0.5-1.9), respectively [relative risk 0.36 (95% CI: 0.17-0.73)]. Based on expected pregnancy rates derived from a 1995 study by Wilcox et al.,100 the effectiveness of the YZP and LNG regimens was estimated to be 57% and 85%, respectively.51 A subgroup of 1157 women was considered to have used ECs correctly based on reported compliance to the investigators’ instructions to take the first dose within 72 hours of UPI and the second dose within 24 hours of the first, abstaining from intercourse until the next menses, and refraining from using other hormonal contraception during the rest of the cycle. In these women, EC effectiveness was 76% and 89% in the YZP and  23  LNG groups, respectively, based on externally derived expected pregnancy rates that were probably too high. The results of the second trial generated enthusiastic petitioning for increasing women’s access to ECs, which led to the regulatory changes outlined in the section above on access to ECs. Based on the trial results, the LNG regimen has frequently been described as an EC that prevents 85% of pregnancies, as reported in the trial in relative terms.58,111 This magnitude of effect continues to be circulated in scientific publications in spite of subsequent data suggesting that the effectiveness estimates in that trial have been overstated.69 A third and most recent randomized trial was conducted by Farajkhoda, et al.90 In this trial, the investigators reported observed pregnancy rates of 8.3% and 0% in the YZP and LNG groups, respectively. This study was limited by a small sample size of only 122 women. These three clinical trials are the only direct comparisons of the YZP and LNG regimens in the literature. In all three randomized trials, inclusion was restricted to women with regular menstrual cycles and with reportedly only one act of intercourse within 48 or 72 hours of requesting EC, thus impairing generalizability to women who seek ECs in the community. There was indeed some inconsistency between the observed pregnancy rates in the trials and the rate derived from pooling both experimental and observational studies. In the pooled analysis of studies, the observed pregnancy rate was 2.0% and 1.7% in the YZP and LNG groups, respectively.67 In contrast, the clinical trials, including the two larger ones, have reported observed pregnancy rates outside of the range of 1.7-2.0%. In light of the inclusion criteria and trial protocols, the trial results may not be generalizable to the routine clinical practice setting. There was an opportunity in British Columbia to conduct a population-based study on the  24  pregnancy outcomes of women who accessed ECs in the community; it is explained in the next section.  Pharmacist prescribing of emergency contraceptives in British Columbia and the present programme of research In Canada, advocacy efforts supporting expanded access to ECs inspired the first related health policy change in British Columbia.19 As of December 2000, pharmacists who have undergone standardized training and been certified as EC Providers were granted the authority to independently prescribe oral ECs.112,113 Community pharmacies in British Columbia became new points of direct access to ECs, in addition to existing avenues: physician offices, clinics, and hospitals. Consequently, the initiative increased women’s opportunity to access ECs conveniently and in a timely manner.112 Compared to physician offices and clinics, pharmacies often have extended hours of operation, including evenings, weekends, and holidays.112,114 Pharmacist EC providers in British Columbia were trained to assess women for appropriateness of oral ECs and ensure that the women received adequate information. Depending on the situation, pharmacists prescribed and dispensed oral ECs with applicable documentation about the prescription and/or referred women to physicians or clinics. Women who received EC from pharmacists were asked to acknowledge receipt of information about ECs and provide consent for receiving pharmacist-prescribed EC on standardized consent forms. The consent forms were designed to document the above, as well as pertinent information about the EC seeker, the prescription, and the pharmacy. Documentation included the date of the first day of the most recent menstrual period, and the date and time of UPI, among others. In British Columbia, all prescriptions dispensed in community pharmacies are documented in a database called PharmaNet. Each health services user in the province is 25  uniquely identified by a personal health number (PHN). Investigators of approved research projects can use the patient identifier to link various databases, so that the health outcomes of drug therapy can be evaluated at the individual level. In the context of ECs, the effectiveness of the YZP and LNG regimens has not been evaluated and compared under the conditions of routine practice in the community. There was an opportunity to investigate the effectiveness of ECs on a large scale, using data on women who received ECs prescribed by community pharmacists in British Columbia. As explained in an earlier section, estimation of EC effectiveness requires an observed pregnancy rate and an expected pregnancy rate. To estimate the observed pregnancy rate among women who received ECs from community pharmacists, linked data can be used to identify pregnancy events and determine whether they are time-compatible with UPI as evidenced by EC prescriptions on a case by case basis. The expected pregnancy rate, however, is not directly available because it is the counterfactual pregnancy rate in the absence of EC. It is not equivalent to the pregnancy rate in the general population of women, because EC seekers have a different level of baseline pregnancy risk.115 However, the expected pregnancy rate can be estimated if indicator(s) of the probability of pregnancy is/are available. Among women who received ECs prescribed by pharmacists in British Columbia, the menstrual cycle day on which UPI occurred was discernable from the information on the treatment consent forms. The cycle day information can be used to estimate individual probability of pregnancy, and therefore, the expected pregnancy rate in the cohort.  26  AIMS This programme of research had four main aims:  Aim 1: To estimate the observed pregnancy rate (EC failure rate) associated with the LNG or YZP regimen under conditions of routine use in the community.  Aim 2: To compare the effect of the LNG and YZP regimens on pregnancy under conditions of routine use, with adjustment for potential confounding.  Aim 3: To estimate an expected pregnancy rate and use both relative and absolute effect measures to report the apparent effectiveness of the LNG and YZP regimens under conditions of routine use.  Aim 4: To investigate the effect of timing of ECs on pregnancy under conditions of routine use.  27  METHODS Overview of research A retrospective, population-based cohort study was designed to address the four aims, using data related to women who received ECs prescribed by pharmacists in British Columbia between December 2000 and December 2002. This study period was selected because, under the pharmacist prescriptive authority, consent for treatment was required from women requesting ECs from pharmacists during this period. The standardized consent forms contained pertinent information for estimating EC effectiveness. Data linkage is a key component in this pharmacoepidemiologic study. De-identified data from consent forms, PharmaNet (the provincial prescription database), and medical records of community and/or hospital services were abstracted for each woman in the study. Information on the explanatory variable (EC regimen type) was collected from consent forms and PharmaNet. To classify the outcome status (the presence or absence of subsequent pregnancy), data from medical records were retrieved and assembled on a timeline for adjudication by a panel of experts. The expert-adjudicated outcome data were used to investigate all four aims. In the comparison between the LNG and YZP regimens, data on potential confounders were derived from consent forms, PharmaNet, medical records, and Census data. The YZP regimen—the older of the two—was the reference to which the LNG regimen was compared.  Ethics approval This research was approved by the University of British Columbia Children’s and Women’s Research Ethics Board (certificate number CW05-0150/H05-70347).  28  Cohort definition The study cohort comprised the first EC prescription of each woman who received a pharmacist-initiated EC prescription in British Columbia between December 2000 and December 2002 and whose prescription record was matched to a consent form using the matching procedure described below.  Data sources In British Columbia, eligible residents can access publicly funded health services in hospitals and in the community. Substantial amount of administrative information collected in the provision of health services are maintained in the British Columbia Linked Health Database (BCLHD) at the British Columbia Ministry of Health. The data used in this research were derived from a number of sources, including de-identified EC treatment consent forms from the College of Pharmacists of British Columbia (hereafter referred to as the College of Pharmacists), and three provincial datafiles: the PharmaNet, Medical Services Plan (MSP), and Hospital Separation datafiles (Table 3). In addition, data from the 2001 Canada Census were used to derive a socioeconomic status indicator.  29  Table 3 Pertinent data from consent forms, the PharmaNet, Medical Services Plan, and Hospital Separation datafiles Consent  Age  PharmaNet  Medical Services Plan  Hospital Separation  StudyID derived from patient’s PHN  StudyID derived from patient’s PHN  StudyID derived from patient’s PHN  Year of birth  Date of birth  Date of birth  First day of the last menstrual period (date) Unprotected intercourse date and time Patient’s postal code Dispensing date and time  Dispensing date  Service date  Drug name  Admission date and separate date  Drug name Drug Identification Number Drug strength Drug dosage form Drug quantity Prescriber’s profession  Pharmacy code  Pharmacy LHA Clinic or hospital code  30  Consent  PharmaNet  Medical Services Plan  Hospital Separation  Immediate use or advance use ICD codes  ICD codes  MSP-specific Additional Diagnostic Codes Fee service codes CCP codes CCP ICD LHA MSP PHN  Canadian Classification of Diagnostic, Therapeutic, and Surgical Procedures International Statistical Classification of Diseases and Related Health Problems Local health area Medical Services Plan Personal health number  31  Consent form data Pharmacists who were certified to prescribe ECs agreed to file a standardized consent form with the College of Pharmacists, the governing body of pharmacists and pharmacies in the province. The consent forms were designed primarily for administrative purposes: to document the EC seeker’s or requester’s consent for receiving treatment and information related to the request. A blank, sample of the consent form is shown in Appendix C. The front side of the form had blank fields or check boxes for the pharmacist to document information, including the woman’s age, first date of the last menstrual period (LMP), date and time of the act of UPI for which EC was sought, EC product trade name, dispensing date and time, pharmacy identifier, and whether the EC was requested for immediate use (after the index UPI) or advance use (after a future act of UPI). Pharmacists were required to transmit a facsimile of the front side of the form to the College of Pharmacists. This de-identified information was approved for use in this research. Pharmacists had been instructed to refrain from transmitting the reverse side of the form, which contained confidential patient identifiers. Consent forms associated with ECs for advance use were excluded in this study.  PharmaNet data PharmaNet is a province-wide database that captures all out-patient prescription drug dispensations regardless of insurance coverage. Pharmacists are required to submit to the PharmaNet system information on all prescriptions dispensed in the community setting. PharmaNet maintains the prescription drug profile for each health service user under his or her 10-digit PHN. In British Columbia, a PHN unique to each health service user identifies him or her across various health service providers. For the purpose of this research, the Ministry of  32  Health converted each woman’s PHN into a StudyID. By making this conversion, PHNs remained confidential while StudyIDs could be used for linkage of datafiles. The fields of PharmaNet data used in this research included the woman’s StudyID, year of birth, geographical local health area (LHA) of residence, as well as the drug’s generic name, trade name, Drug Identification Number, strength, dosage form, quantity, dispensing date, pharmacy LHA, and the prescriber’s profession.  Medical Services Plan data Residents of British Columbia are required to enroll with MSP to receive insured medical services. The MSP datafile contains records of health services provided and billed for by physicians who are registered to receive fee-for-service. The fields of information used in this research included the woman’s StudyID (derived from her PHN, as above), date of birth, as well as International Classification of Diseases, Ninth Revision (ICD-9) diagnostic codes, MSPspecific Additional Diagnostic Codes, and fee service codes associated with the service encounter. Fee service codes are hereafter referred to as Fee items.  Hospital Separation data The Hospital Separation datafile contains records of health services provided in clinic facilities and hospitals. The fields of information used in this research included the woman’s StudyID (derived from her PHN, as above), admission date, separation date, ICD-9 codes, and procedure codes listed under the Canadian Classification of Diagnostic, Therapeutic, and Surgical Procedures (CCP).  33  Canada Census data Data from Statistics Canada’s 2001 Census were used to derive neighbourhood income at the dissemination area level for use as a socioeconomic indicator.116-119  Initial data management The datafiles were initially read into SPSS® version 12.0 (Chicago, IL) for checking, subsetting, and recoding. Consent and PharmaNet records were selected if the dispensing date was between December 2000 and December 2002. In the PharmaNet database, pharmacist-prescribed records were selected by excluding records with other prescriber codes. In the consent database, a time variable was created by taking the difference (in hours) between the EC dispensing date and time and the UPI date and time. This variable represented the time delay to receive EC after the index act of UPI. In addition, a cycle day variable was created by taking the difference between the UPI date and the LMP date, and then rounding up to the nearest integer. This variable represented the menstrual cycle day number on which UPI occurred, counting from the beginning of the LMP (Day 1 was the first day of the current menstrual cycle). In a previous project,120 dissemination area-level income data were linked to postal codes in British Columbia. Average and median neighbourhood income data were available from Census population and household universes. These data were linked to the postal code associated with each woman’s MSP registration. Average population neighbourhood income was analyzed in the main analysis, because data were available for 90% of the women in the study cohort; whereas, median household neighbourhood income was available for only 83% of women. Sensitivity analysis was conducted, using median household neighbourhood income.118,119  34  Data linkage PharmaNet records were linked to the MSP database and also to the Hospital Separation database. Also, PharmaNet records were matched to consent form records. As a result, a timeline of events could be created for each episode of EC dispensing: from the onset of a woman’s LMP to her reported date of UPI (the index UPI), to EC dispensing, and subsequent pregnancy-related events. The linked data were then used in the process of estimating an observed pregnancy rate. Also, prescription history and medical history in the linked data were used as covariates in the statistical procedures when comparing the YZP and LNG regimens.  Linking PharmaNet records and medical records As mentioned above, the StudyID derived from the PHN of each woman was a common identifier across health services. Using the StudyID, PharmaNet records were linked anonymously to MSP and Hospital Separation records for each corresponding woman.  Matching PharmaNet records and consent records In theory, each PharmaNet record of pharmacist-prescribed EC should have a corresponding consent form to link to. However, the two databases were not directly linkable. To maintain anonymity, patient identifiers on the consent forms were intentionally excluded among the data fields available for this research project. The linkage was performed by matching PharmaNet records with consent forms on four characteristics: EC dispensing date, EC product type (i.e., Ovral®, Preven®, or Plan B®), age of the woman (in years), and the geographical LHA of the pharmacy from which the prescription originated. The 3-digit pharmacy code that uniquely identifies each pharmacy in the province could be converted into the less specific variable of  35  pharmacy LHA, but not vice versa. Therefore, in the consent records, the pharmacy code was used to locate each pharmacy and assign the corresponding LHA. In the populated area of metro Vancouver, matching was performed to the detailed sub-LHA level. In order to match consent and PharmaNet records unambiguously by the four criteria, the records on each side of the match had to include exclusively unique combinations of the four criteria. Two algorithms were created to derive unique combinations of dispensing date, age, EC type, and pharmacy LHA from the PharmaNet records and from the consent records (Figures 3 and 4). The objectives were to make exact, unambiguous matches, and to maximize the number of matches.  36  Figure 3 Algorithm for selecting PharmaNet records with unique combinations of age, emergency contraceptive type, dispensing date, and pharmacy local health area  37  Figure 4 Algorithm for selecting consent records with unique combinations of age, emergency contraceptive type, dispensing date, and pharmacy local health area  38  On the PharmaNet side, the main data issues concerned: (1)  Invalid pharmacy LHAs (000, 998)  (2)  Missing information for age  (3)  Records that were identical on the four characteristics for matching (“non-unique records”) For every PharmaNet record that had an invalid LHA, the record itself and all non-unique  records (with the same dispensing date, EC type, and age) were excluded from the matching because it was not possible to accurately match the record with a consent form. Furthermore, the PharmaNet database available for this project included the women’s year of birth but not the specific date of birth to match with that in the consent database. Since StudyIDs are common across PharmaNet and MSP databases, the exact age for most women (for whom data were available) was derived from the date of birth available in the MSP database. In cases where the date of birth was not available from the MSP database, age was estimated from the year of birth. The method for correctly utilizing the estimated age in matching is described with example in Appendix D. The reason for using the estimated age was to minimize exclusion of non-unique records (with the same dispensing date, EC type, and pharmacy LHA) as a result of unknown age. On the consent form side of the matching, the main data issues were: (1)  Consents for ECs prescribed for advance use  (2)  Missing information for age, EC type, or pharmacy LHA Consent records of ECs indicated for advance use were excluded from the matching. In  records where information was missing for age, EC type, or pharmacy LHA, the records themselves and similarly non-unique records were excluded from the matching.  39  After the above exclusions, there remained some records with non-unique combinations of the four matching characteristics, especially in the PharmaNet database. In the PharmaNet database, women were distinguishable by their StudyIDs. In the consent database, women were roughly distinguishable by their reported dates of LMP and UPI. The remaining non-unique records were adjudicated as follows: (1)  If the ECs were prescribed for different women, both EC records were excluded because it was not possible to perform accurate matching on these records.  (2)  If the ECs were prescribed for the same woman, given that the records had identical dispensing date, EC type, age, and pharmacy LHA, one of the records was retained and included in the matching. As a result, each of the PharmaNet and consent databases was left with records that had  unique combinations of the four characteristics. PharmaNet records and consent records were then matched unambiguously by the four characteristics. In the matched set of EC records, the first prescription of each woman was included in the study cohort. The data set generated from all data linkages incorporated de-identified, patient-specific demographic and reproductive health information, corresponding prescription data, and medical diagnoses and procedures.  40  Outcome ascertainment To investigate the four aims of this study, the outcome of each woman in the cohort was determined by first identifying possible pregnancy-related events in her medical records, and then presenting the data to a panel of experts for adjudication.  Screening for possible outcome-related codes The MSP and Hospital Separation records of each woman in the cohort were screened for the presence of pregnancy-related administrative codes (Appendix E) within 42 weeks after the EC dispensing date. This screening period was selected based on published data from the British Columbia Perinatal Database Registry indicating that only approximately 0.1% of pregnancies in the province exceeded 42 weeks’ gestation during the years 2000-2002.121 Abortion-specific codes (Appendix F) were screened for up to 20 weeks from the EC dispensing date because, according to local experts, the clinic that provided abortion services at the latest provided them up to 20 weeks’ gestation. Although pregnancy time is clinically expressed in terms of gestational age, this parameter could not be accurately determined in the study data. Therefore, the reference point of screening was the EC dispensing date and not the reported LMP date. Three types of codes were screened for: (1) ICD-9 diagnostic codes from the MSP and Hospital Separation databases, (2) physician Fee-For-Service billing codes (fee items) from the MSP database, and (3) CCP procedure codes from the Hospital Separation database. The versions of codes available during the study period were considered. Clinical experts were interviewed to verify the lists of relevant codes to screen for. The experts included family physicians with training and experience in maternity care and an obstetrics and gynaecology  41  specialist. They were knowledgeable in the related field of clinical practice and familiar with relevant fee billing in the province.  Time profiles For each woman who screened positive for pregnancy-related codes, her MSP and Hospital Separation records were reviewed to manually document the date of the codes originally screened for (Appendices E and F). In addition, a number of pregnancy-related codes not originally screened for were manually abstracted: ICD-9 code 626 “Disorders of menstruation and other abnormal bleeding from female genital tract,” Fee items 15120 “Pregnancy test, immunological,” 8655 “B scan – obstetrical (under 14 weeks gestation),” 8651 “B scan – obstetrical (14 weeks gestation or over),” and MSP-specific codes 34A “Contraceptive advice,” 17B “Consultation re abortion,” 23B “Insertion/removal of IUD,” 30B “Prenatal care,” and 38B “Threatened abortion.” Time profiles were created to illustrate the following milestones in a standardized fashion: LMP date, UPI date, EC dispensing date, and the dates and definitions of pregnancyrelated codes identified during screening, as well as those abstracted after screening (listed above). A sample time profile is shown in Figure 5. Any oral EC dispensation subsequent to the index EC was included on the time profile as additional milestone(s), regardless of the profession of the prescriber. The milestones were depicted proportionately with respect to time; in other words, a 6-month interval appeared exactly twice as wide as a 3-month interval on a time profile. Clinic and hospital facilities that provided abortion services had guidelines for the gestational age range within which they performed surgical abortions. Local experts were interviewed to document the gestational age range within which abortion services were provided  42  circa 2001-2002. If the codes suggested an abortive outcome and if the facility was known, the gestational age range of service was noted as “Additional Information” on the time profile.  43  Figure 5 Sample time profile  1/20/2002 EC 1/17/2002 UPI  3/29/2002 ICD 635 "Legally induced abortion" Fee 4111 “Therapeutic abortion less than 14 weeks gestation” CCP 87.1 “Vacuum aspiration for termination of pregnancy”  1/1/2002 LMP  1/1/2002 - 1/17/2002 16 days  1/17/2002 - 3/29/2002 71 days  Additional information: The facility at which abortion was performed provided abortion services at 42-97 days’ gestation.  Comments:  EC case # Patient #  Your adjudication:  Pregnancy? (circle)  YES  NO  Induced abortion? (circle)  YES  NO  44  Expert adjudication of compatibility In order to achieve a high degree of accuracy in defining pregnancy-related outcomes, a panel of three experts was assembled to adjudicate the outcomes of the women in the cohort based on medically relevant administrative codes on the time profiles. The experts were asked to adjudicate independently whether a compatible pregnancy (i.e., a pregnancy that was compatible with the index UPI for which EC was requested) was present and whether a compatible induced abortion (an indicator of unintended pregnancy) was present. It was agreed among the experts that compatibility was based on the nature and timing of the codes. Three pilot phases were conducted. In each phase, the experts were given 20 time profiles to adjudicate and they met afterwards to discuss cases with discordant adjudication. After the pilot phases, the experts were asked to adjudicate all time profiles. The profiles were divided into three approximately equal sets. The experts adjudicated one set at a time and, in order to minimize bias related to experience, they received the sets in staggered order. After all time profiles had been adjudicated, the experts met to discuss cases with discordant adjudication. The experts were blinded to the EC regimen type throughout the adjudication and discussion procedures. Inter-rater agreement was estimated with Fleiss’ kappa statistic using SPSS® version 15.0 (Chicago, IL) and David Nichols’ macro.122 The expert-adjudicated outcome data were used to investigate the four aims of this research programme.  Power estimation The power of the study to detect decreases in the risk of pregnancy with the LNG regimen was estimated over a range of plausible pregnancy rates in the YZP group, based on sample sizes of 4943 and 3323 in the LNG and YZP groups, respectively (Table 4). For example, given a 2% pregnancy rate in the YZP group, the study would have approximately 82% 45  power to detect a relative risk of 0.6 for LNG versus YZP (a 40% risk reduction). The method used to estimate the power was based on the method developed by Fleiss et al.123 to estimate the power of a two-sided test for the difference in two proportions.  Table 4 Power estimation over a range of plausible pregnancy rates in the Yuzpe group Proportion of pregnancies in the Yuzpe group  Relative risk of pregnancy with the levonorgestrel regimen 0.3  0.4  0.5  0.6  0.7  0.01  97.33%  89.98%  74.76%  53.50%  32.26%  0.02  99.98%  99.50%  95.71%  81.82%  55.58%  0.03  100.00%  99.98%  99.44%  93.91%  72.86%  0.04  100.00%  100.00%  99.94%  98.18%  84.34%  46  Aim 1: To estimate the observed pregnancy rate (emergency contraceptive failure rate) associated with the levonorgestrel or Yuzpe regimen under conditions of routine use in the community The methods for data linkage and expert adjudication of outcomes have been described in the sections above. Characteristics of the women were analyzed in the two EC groups: LNG and YZP. Descriptive statistics were performed; differences and confidence intervals were estimated with the YZP group as the comparator. The mean cycle day on which UPI occurred was determined for the subset of women whose LMP date was known to be between Day 1 (the first day of the menstrual cycle) and Day 40. The number of pregnancies expected in the absence of EC was calculated based on conception probabilities estimated by Wilcox et al.;101 this calculation is explained in detail under Aim 3. Neighbourhood income data were derived from the 2001 Census data at the dissemination area level, appended to the postal code of residence of the women in the cohort, and then divided into income quintiles.116,117 Pregnancy was defined as an event deemed by at least two of the three experts to be a pregnancy that was compatible with the index UPI (the “majority vote”). The pregnancy rate observed after each regimen was enumerated and reported as separate EC failure rates. The pregnancy rates in the LNG and YZP groups were compared using the Chi-square test (two-sided alpha level, 0.05). The rate of induced abortion was also evaluated as a specific type of pregnancy outcome and as an estimate of the minimum number of unintended pregnancies. The crude (i.e., unadjusted) odds ratio (and 95% CI) of pregnancy was estimated for the LNG regimen against the YZP regimen. A series of assumptions was made about missed pregnancies, as a sensitivity analysis of potential bias by differential outcome misclassification.  47  Aim 2: To compare the effect of the levonorgestrel and Yuzpe regimens on pregnancy under conditions of routine use, with adjustment for potential confounding Hypothesis The hypothesis was that receiving the LNG regimen for emergency contraception is associated with a lower risk of pregnancy than receiving the YZP regimen, after adjusting for patient- and treatment-related characteristics.  Assessing and controlling for confounding Confounding is a potential issue in this study because EC regimen use was based on selfselection and not randomized allocation. It is not known whether patient- or treatment-related variables systematically biased the selection of regimen or independently contributed to part or all of the observed pregnancy-related outcomes. Therefore, a series of procedures was undertaken to assess and control for potential confounding. Assessment began with examination of the variable of interest (EC regimen type) and potential confounders by means of graphical diagnostics. Potential confounders were evaluated in exploratory analyses. They were then assessed and controlled for with stratification and multivariate analyses.124-126  Potential confounders in this study A confounder in this study is a variable that is associated with an EC type selection and with pregnancy. However, for any given variable to be considered a confounder, it cannot be an effect of the EC type selection.124,127  48  Potential confounders in this study include variables that were thought to be related to fertility and/or sexual behaviour. They included age,128-131 time to receive EC after UPI,52 income,128 menstrual cycle day of the UPI,100,101,129 1-year history of pregnancy,128,132 1-year history of any EC dispensation (any prescriber), 5-year history of relevant gynaecological conditions, 1-year history of hormonal contraceptive (HC) use, and concurrent HC use. None of these variables was an effect of EC type selection. Dissemination area-level income data from Statistics Canada’s 2001 Census were used as ecological indicators of socioeconomic status.116 Postal codes were used to derive neighbourhood income at the dissemination area level.118,119 Dissemination area was the smallest geographical unit for which census profile data were available.116 Each dissemination area was composed of one or more neighbouring blocks, with a population of 400-700 individuals.133 The neighbourhood income variable (hereafter referred to as the income variable) was analyzed in quintiles.116,117 Past pregnancy and/or gynaecological conditions were identified by screening for relevant codes in MSP and Hospital Separation records. Relevant gynaecological conditions included pelvic inflammatory disease,134-136 endometriosis,134,137 ovarian dysfunction,134 ectopic pregnancy,138 infertility, and sterilization. An expert was consulted to verify the list of relevant conditions and the ICD-9, fee items, and CCP codes to screen for. Hormonal contraceptive dispensation was identified by reviewing PharmaNet records of oral contraceptive medications, medroxyprogesterone 150 mg/ml vials, and levonorgestrelreleasing intrauterine systems. This evaluation excluded all Plan B® prescriptions, all Preven® prescriptions, and Ovral® prescriptions in quantities not normally used for regular contraception.  49  Concurrent use of HC was defined as having an HC prescription that overlapped the index EC dispensing date. Potential confounders are hereafter referred to as covariates.  Exploratory analyses The distributions of the variables were examined, initially to check for apparent accuracy, and also to plan for control of confounding. The variables in this study are listed in Table 5. The explanatory variable of interest, EC regimen type, was dichotomous: LNG and YZP. The outcome variable was pregnancy, which was also dichotomous (pregnant or not pregnant). Potential confounders were either categorical or continuous variables. In the early stages of data analysis, it was determined that stratification and multivariate logistic regression would be appropriate for eliciting the association between EC type and pregnancy, with adjustment for potential confounding.124,139 Therefore, it was important to examine the data in ways that would inform the stratified and multivariate regression analyses. First, the range of data of each continuous variable would need to be categorized into discrete categories (strata), and then the association between EC type and pregnancy would be estimated within each category (stratum). In this study, categorization of continuous variables made use of the natural ordering of the data. In addition, the variables were analyzed, at least initially, with the same category cutpoints as have been used in other studies or reports, enabling potential comparison of the study results with those published elsewhere. The age variable was categorized as it has been categorized in pregnancy outcome reports published by Statistics Canada.140 The time variable was categorized as it has been categorized in a re-analysis of the WHO study data.52 Income data  50  were divided into quintiles as they have been in other studies that used Canada census data.116,117 The cycle day variable was not defined according to any specific method of categorization except that the natural ordering of the data was retained, and within-category uniformity of effect was taken into consideration when the initial set of categories was created. To simplify the stratified analysis and to avoid “sparse stratification”127 (i.e., small cells), observed pregnancy rates were examined to identify neighbouring strata that could be combined, if applicable. For logistic regression, the multiplicative nature of the modeling technique implied an exponential relationship between a continuous variable and pregnancy.124,141 Using the observed data, log odds of pregnancy were plotted against each continuous variable. The initial analyses helped to identify the appropriate form of entry of continuous variables into multivariate logistic models.  51  Table 5 Variables in the study and their data type Variable  Data type  Pregnancy (the outcome variable)  Categorical  Emergency contraceptive (EC) regimen type  Categorical  Age  Continuous  Menstrual cycle day on which the index unprotected intercourse reportedly occurred  Continuous  Time between unprotected intercourse and EC dispensing  Continuous  Neighbourhood incomea  Continuous  History of pregnancy within 1 year prior to the index EC (reference = no history of pregnancy)  Categorical  History of gynaecological conditions that may reflect fertility and/or behaviour within 5 years prior to the index EC (reference = no history of conditions)b  Categorical  History of physician- or pharmacist-prescribed EC within 1 year prior to the index EC (reference = no history of EC prescription)  Categorical  Concurrent hormonal contraceptive (HC)c prescription overlapping the index EC dispensing date (reference = no concurrent HC prescription)  Categorical  History of HCc prescription within 1 year prior to the index EC (excluding current use) (reference = no history of HC prescription)  Categorical  a  Neighbourhood income at the dissemination area-level, from the 2001 Canada Census data Gynaecological conditions included pelvic inflammatory disease, endometriosis, ovarian dysfunction, ectopic pregnancy, infertility, and sterilization. Medical Services Plan and hospital separation records were examined to capture relevant diagnostic, procedural, and billing codes. c Hormonal contraceptives included PharmaNet records of oral contraceptive pills, medroxyprogesterone 150 mg/ml vials, and levonorgesterone-releasing intrauterine devices. Plan B prescriptions, Preven prescriptions, and Ovral prescriptions in quantities not normally used for regular contraception were excluded. b  52  Stratified analyses The association between EC regimen type and pregnancy was stratified on (i.e., adjusted by) one covariate at a time to avoid sparse stratification, which has been defined as less than five subjects per cell.127 For each categorical variable, stratification was a cross-tabulation of the data to yield stratum-specific risk ratios and odds ratios for the effect of the LNG regimen compared to the YZP regimen. For continuous variables, the analyses required that each variable be categorized into an array of strata. The strata needed to be sufficiently broad to avoid sparse stratification, and at the same time, be not so broad that within-stratum uniformity of effect could not be ensured.127 Neighbouring strata that were found to have seemingly homogenous odds ratio of pregnancy (comparing LNG against YZP) in the initial exploratory analyses were combined into a single stratum.125 Then, stratum-specific risk ratios for pregnancy were pooled using the Mantel-Haenszel method to derive a weighted summary effect estimate.124  Multivariate logistic regression analyses Multivariate regression modeling was used to describe the association between EC type and pregnancy while controlling or adjusting for other variables.126,139 In this study, the approach was to find a parsimonious model that satisfactorily described the observed data.125 There were two main considerations in model building in this study: (1) which variables to include in the model, and (2) in what form the continuous variables should be included in the model. Many methods have been proposed for variable selection.124,127,142-144 In this study, variable selection was guided by prior information about the covariates as well as the data.127 Variables that were identified as potential confounders (covariates) in this study were selected based on background knowledge of their clinical relevance to pregnancy (see “Potential  53  Confounders” section above).145 To estimate the strength of association between each covariate and pregnancy, univariate regression analyses were conducted, with a covariate as the only explanatory variable, and pregnancy the outcome variable. A series of multivariate models was fitted iteratively, beginning with continuous covariates that appeared to have the greatest effect on the observed log odds of pregnancy and categorical variables that were statistically significant in univariate analyses. At each step, one additional variable was entered into the model, and the model fit was assessed using the Akaike Information Criterion (AIC).146 When comparing models, lower AIC indicates better model fit. Using this approach, a parsimonious multivariate model was selected as the main model, and the adjusted odds ratio of pregnancy for the LNG regimen relative to the YZP regimen was noted. In sensitivity analyses, other variables were added, and the impact of these variations on the odds ratio of interest, the width of the 95% CI, and the AIC were evaluated. The categorical variables in this study were dichotomous and so they were coded: 1 for presence of the variable, and 0 for absence of the variable. On the other hand, the continuous variables required more intense scrutiny to determine the appropriate form in which they should be entered into the model. The problem is complicated in a logistic model because the effect of incremental increases in explanatory variables is multiplicative.124,141 Therefore, the shape of the dose-response for each continuous variable was examined, with consideration for biological plausibility, and alternatives were explored to find a form that was well suited to the shape of the observed data.127,147 The observed log odds of pregnancy (rather than the observed pregnancy rate) for each continuous variable informed the form of entry of these variables, because the predicted log odds of pregnancy was to be estimated in subsequent logistic regression analysis. In the primary  54  analysis, continuous variables with a monotonic relationship with pregnancy on a log-odds plot were included as a linear term, and for variables that displayed a curvilinear relationship with pregnancy, a quadratic term was included along with the linear term.127,147-149 The quadratic term was created by taking the difference between the value of a continuous variable and the mean of the variable (i.e., centered at the mean), and then squaring the difference.149 For example, a logistic regression model that related age to pregnancy had the following formula: Ln odds(pregnancy) = ß0 + ß1*Age + ß2*(Age – Mean age)2 Interaction was evaluated by incorporating the product of selected variables, in addition to the individual variable terms,127 and then assessing the significance of the interaction term. Multicollinearity was assessed by examining the Tolerance and Variance Inflation Factor (VIF) for each variable;150 for this assessment, continuous variables were considered in their original, continuous form. In sensitivity analyses, alternative modeling strategies for entering continuous variables were explored.127 In one iteration, each continuous variable was modeled as an array of dichotomous categorical variables with values of 1 or 0 to denote category membership. In another, the curve-smoothing technique of logistic B-spline regression was performed, using a macro by Gregory et al.151 The degree and knot specifications were determined on principles of parsimony and AIC-minimization.151 These regression specifications were compared against the simplest model in which continuous variables were included as single linear terms (i.e., in their original continuous form). Regression modeling in this study was conducted using SAS® version 9.2 (SAS Institute Inc., Cary, NC). The possible effect of measured and unmeasured confounding on the observed EC regimen type-pregnancy association was assessed jointly over an array of assumptions about the  55  strength of confounding on pregnancy, and imbalance in the prevalence of confounder(s) between the two EC regimen groups.152,153 The objective was to determine the extent of confounding necessary to fully explain the observed association between EC regimen type and pregnancy.153  56  Aim 3: To estimate an expected pregnancy rate and use both relative and absolute effect measures to report the apparent effectiveness of the levonorgestrel and Yuzpe regimens under conditions of routine use The magnitude by which ECs prevent pregnancy was measured in both relative and absolute terms. The relative reduction in pregnancy risk (“relative risk reduction”) was estimated using the formula:67,71 1 - (observed pregnancy rate / expected pregnancy rate) where the observed pregnancy rate was the percentage of women deemed by the majority vote of the experts to have had a compatible pregnancy among all women in the treatment group; the expected pregnancy rate was estimated as follows: (1)  The consent records included the dates of the women’s first day of the last menstrual cycle (Day 1 of the cycle) and the day on which UPI reportedly occurred. Using this information, the cycle day of UPI was determined.  (2)  Wilcox et al.101 have estimated the probability of clinical pregnancy for women who had UPI on Day 1 to 40 of the cycle. Three sets of probabilities have been developed, for women with regular cycles, irregular cycles, and for the combined cohort, respectively. For example, in the combined cohort, the probability of pregnancy for UPI on Day 6 was estimated to be 0.9%; whereas, the mean probability on Day 13 was 8.6%.101  (3)  In the present study, the set of probabilities for the combined cohort was used. The number of expected pregnancies on a specific cycle day was the product of the day’s probability for pregnancy and the number of women who reportedly had UPI on that cycle day.  57  (4)  The expected pregnancy rate was the sum of cycle day-specific numbers of expected pregnancies divided by the number of women in whom cycle day information was available.  (5)  The expected pregnancy rate in the cohort was extrapolated from the expected pregnancy rate among women with cycle day information (i.e., assumed to be the same rate). For each regimen, the absolute risk reduction was the absolute difference between the  expected pregnancy rate and the observed pregnancy rate. The NNT (to prevent one pregnancy) was the reciprocal of the absolute risk reduction. For each of the two EC regimens, 95% CIs were estimated for the absolute risk reduction and NNT in the subgroup of women in whom cycle day information was available (i.e., among whom the expected and observed pregnancy rates had the same denominator).  58  Aim 4: To investigate the effect of timing of emergency contraceptives on pregnancy under conditions of routine use To estimate the effect of the timing of ECs on pregnancy, the time variable was first converted into an array of seven categories as it has been categorized elsewhere.52 The observed pregnancy rate was calculated within each category. The expected pregnancy rate was estimated using the method described above under Aim 3. Within each category, the absolute risk reduction was the absolute difference between the expected pregnancy rate and the observed pregnancy rate. Logistic regression models were built to estimate the effect of time on the odds of pregnancy, controlling for potential confounders (covariates). Continuous variables were modeled using the same approach described under Aim 2. In brief, variables with a monotonic relationship with pregnancy were included as a linear term. Variables with curvilinear relationship with pregnancy were modeled with a quadratic term in addition to the linear term.127,147-149  59  RESULTS Results common to all four aims A total of 14110 and 13579 pharmacist-provided EC prescriptions were identified in the PharmaNet and consent form databases, respectively. After applying the algorithms (Figures 3 and 4 above), 11014 PharmaNet records and 12482 consent records remained and were included in the matching process. A total of 8622 matched ECs were found; they were prescribed for 7493 unique women. The first prescription for each unique woman was included in the study cohort. The matching procedure is illustrated in Figure 6. Characteristics of records in the study cohort and those excluded are shown in Table 6. The records of 467 of the 7493 women screened positive for pregnancy-related codes. The three experts’ adjudications on pregnancy as an outcome were initially concordant on 413 (88.4%) of the 467 cases. After the experts met to discuss discordant cases, a further 44 cases had concordant adjudication (i.e., 97.9% overall agreement), and Fleiss’ kappa was found to be 0.97 (p < 0.001).  60  Figure 6 The procedure for matching PharmaNet records of emergency contraceptive (EC) prescriptions and consent forms  61  Table 6 Characteristics of cohort records and excluded records Characteristic  Cohort records (n = 7493)  Excluded PharmaNet records (n = 6617)  Excluded consent records among ECs for immediate use (n = 5937)  25.7 (7.8)  25.9 (7.3)  25.2 (7.0)  Levonorgestrel  59.7  60.9  61.0  Yuzpe  40.3  39.1  39.0  Mean time to receive EC after UPI (SD) (hours)a  26.2 (19.3)  N/A  26.7 (20.1)  Mean menstrual cycle day of UPI (SD) (years)b  16.3 (7.5)  N/A  16.3 (7.5)  Mean age (SD) (years) EC regimen type (%)  a  For this estimation, n = 7200 among cohort records and n = 5761 among excluded consent records with available information b For this estimation, n = 6683 among cohort records and n = 5415 among excluded consent records with known menstrual cycle day of the index act of UPI, ranging from Day 1 to Day 40 EC N/A SD UPI  Emergency contraceptive Not applicable because PharmaNet records did not contain the information Standard deviation Unprotected intercourse  62  Aim 1 The cohort included 4470 (59.7%) and 3023 (40.3%) women in the LNG and YZP groups, respectively. Characteristics of the women are shown in Table 7. Differences were computed with YZP as the comparison group. A total of 193 pregnancies were deemed by at least two of the three experts as being compatible with the index act of UPI; the observed pregnancy rate in the cohort of 7493 women was 2.6%. There were 99 (2.2%) and 94 (3.1%) compatible pregnancies in the LNG and YZP groups, respectively (p = 0.02; risk difference -0.9%, 95% CI -1.7% to -0.1%). The rate of induced abortion was 1.7% and 2.3%, respectively (p = 0.06). The odds ratio of pregnancy for the LNG regimen against the YZP regimen was 0.71 (95% CI: 0.53 to 0.94), based on the observed numbers of pregnancy, 99 and 94 in the LNG and YZP groups, respectively. The sensitivity of the odds ratio to the possibility of missed pregnancies was tested, as shown in Table 8. In addition to the 99 pregnancies identified in the LNG group, seven or more pregnancies would need to have been missed in the LNG group, compared to the YZP group, in order for an odds ratio of 1 to be included in the 95% CI. For example, the odds ratio would be 0.757 (95% CI: 0.571-1.003) if there were 7 missing pregnancies in the LNG group (beyond the 99 pregnancies already identified) and no missing pregnancies in the YZP group.  63  Table 7 Characteristics of the women in the study cohort Characteristics  Levonorges- Yuzpe (YZP) Difference trel (LNG) regimen (95% CI lower, regimen (n = (n = 3023) upper limits) 4470)  Combined (n = 7493)  Mean age (SD) (years)  26.4 (7.9)  24.7 (7.6)  1.7 (1.3, 2.0)  25.7 (7.8)  Mean menstrual cycle day of unprotected intercourse (SD) (day number)a,b  16.3 (7.3)  16.5 (7.7)  -0.2 (-0.6, 0.2)  16.3 (7.5)  Mean time to receive emergency contraceptive (EC) after unprotected intercourse (SD) (hours)c  25.0 (19.4)  27.9 (19.2)  -2.9 (-3.8, -2.0)  26.2 (19.3)  Mean dissemination area-level neighbourhood income (SD) (dollars)d  31140 (11843)  29275 (8880)  1866 (1342, 2390)  30396 (10797)  Number (%) of cases with a history of pregnancy within one year prior to the index EC  440 (9.8%)  340 (11.2%)  -1.4% (-2.8, 0.0%)  780 (10.4%)  Number (%) of cases with a history of specified gynaecological conditions within five years prior to the index ECe  228 (5.1%)  130 (4.3%)  0.8% (-0.2, 1.8%)  358 (4.8%)  Number (%) of cases with a history of any EC prescription within one year prior to the index EC  362 (8.1%)  313 (10.4%)  -2.3% (-3.6, 0.9%)  675 (9.0%)  Number (%) of cases with concurrent use f of hormonal contraceptiveg  308 (6.9%)  214 (7.1%)  -0.2% (-1.4, 1.0%)  522 (7.0%)  64  Characteristics  Levonorges- Yuzpe (YZP) Difference trel (LNG) regimen (95% CI lower, regimen (n = (n = 3023) upper limits) 4470)  Number (%) of cases with a history of hormonal contraceptiveg prescription (excluding concurrent usef) within one year prior to the index EC  703 (15.7%)  534 (17.7%)  -2.0% (-3.7, -0.2%)  Combined (n = 7493)  1237 (16.5%)  a  Day number 1 refers to the first day of the current menstrual cycle. For this estimation, n = 4020 (LNG), n = 2663 (YZP) with available information on the menstrual cycle day of unprotected intercourse ranging from Day 1 to Day 40. c For this estimation, n = 4312 (LNG), n = 2888 (YZP) with available information. d For this estimation, n = 4061 (LNG), n = 2695 (YZP) with available information on dissemination area-level, average population neighbourhood income. e Gynaecological conditions included pelvic inflammatory disease, endometriosis, ovarian dysfunction, ectopic pregnancy, infertility, and sterilization. Medical Services Plan and Hospital Separation records were examined to capture relevant diagnostic, procedural, and billing code. f Theoretical concurrent use based on having hormonal contraceptive prescription that overlapped the index EC dispensing date. g Hormonal contraceptives included PharmaNet records of oral contraceptive pills, medroxyprogesterone 150 mg/ml vials, and levonorgestrel-releasing intrauterine systems. Ovral® prescriptions in quantities not normally used for regular contraception were excluded. b  65  Table 8 Sensitivity of the odds ratio (and 95% CI) of pregnancy for the levonorgestrel regimen group against the Yuzpe regimen group to the number of pregnancies in each group (the observed number of pregnancy was 99 and 94 in the two groups, respectively) Number of pregnancies in Number of pregnancies in the levonorgestrel (LNG) the Yuzpe (YZP) regimen regimen group (n = 4470) group (n = 3023)  Odds ratio of pregnancy (LNG vs. YZP) (95% CI)  99  94  0.706 (0.530-0.940)*  100  94  0.713 (0.536-0.949)  101  94  0.720 (0.542-0.958)  102  94  0.728 (0.548-0.967)  103  94  0.735 (0.554-0.976)  104  94  0.742 (0.559-0.985)  105  94  0.750 (0.565-0.994)  106  94  0.757 (0.571-1.003)  106  95  0.749 (0.565-0.991)  107  95  0.756 (0.571-1.000)  * Odds ratio based on the observed data in this study  66  Aim 2 The association between EC regimen type and pregnancy was evaluated with adjustment for potential confounding. First, the observed, unadjusted pregnancy outcome data are summarized in Table 9.  Table 9 Pregnancy outcome data based on the majority vote of the three experts Levonorgestrel (LNG) regimen 99  Yuzpe (YZP) regimen  Number of women not pregnant  4371  2929  Total  4470  3023  Risk  0.022  0.031  Number of women pregnant  Risk difference(LNG vs. YZP) (95% CI lower, upper limits)  -0.009 (-0.017, -0.001)  Risk ratio(LNG vs. YZP) (95% CI)  0.712 (0.539, 0.941)  Odds ratio(LNG vs. YZP) (95% CI)  0.706 (0.530, 0.940)  94  67  Exploratory analyses As an initial step to assessing and controlling for confounding, the data relationship between each continuous variable and pregnancy was examined. This section is a report of the observed data relationships. The effects of the initial findings on subsequent data analysis are explained in the Stratified Analyses and Multivariate Regression Modeling sections below. The Y-axes of univariate data plots were kept in the same scale so that the strength of effect among covariates might be compared. The age variable was found to have a concave curvilinear relationship with the observed pregnancy rate, as illustrated in Figure 7. Category-specific odds and log odds of pregnancy based on the observed data are shown in Table 10 and Figure 8. The cycle day variable was initially analyzed in 9 naturally ordered categories. The observed pregnancy rates over the 9 categories (Figure 9) suggested that the variable could be redefined using fewer categories, which would help to simplify the stratified analysis and avoid sparse stratification. Therefore, the variable was reorganized into quintiles. The observed pregnancy rates over quintiles followed, roughly, a similar concave curvilinear shape (Figure 10). Observed odds and log odds of pregnancy were estimated over the quintiles, as shown in Table 11 and Figure 11. The time variable was assessed in 7 chronological categories. Observed pregnancy rates across the 7 categories, and the observed odds and log odds of pregnancy are shown in Figure 12 and Table 12. It was noted that the last category of time (>72 hours) had only 4 pregnancies. As shown in Figure 13, there was an approximately linear increase in the log odds of pregnancy with increasing time category.  68  The income variable was analyzed in quintiles. The results on the main analysis, using average population neighbourhood income data, are shown in Figure 14, Table 13, and Figure 15. In addition, the log odds of pregnancy was analyzed against median household neighbourhood income quintiles (Figure 16). The two income measures produced similar results; average population neighbourhood income was available for 90% of the women in the study cohort, but only 83% of women on the median household neighbourhood income measure. Therefore, average population neighbourhood income was designated as the income variable in subsequent analyses.  69  Figure 7 Observed pregnancy rate by age categories  70  Table 10 Observed odds and log odds of pregnancy by age categories Age (years)  N  N pregnant  N not pregnant  Odds of pregnancy  Log odds of pregnancy  ≤19  1930  25  1905  0.01312  -4.3334  20-24  1934  48  1886  0.02545  -3.6710  25-29  1407  57  1350  0.04222  -3.1648  30-34  1083  43  1040  0.04135  -3.1858  ≥35  1139  20  1119  0.01787  -4.0245  71  Figure 8 Log odds of pregnancy by age categories  72  Figure 9 Observed pregnancy rate by nine cycle day categories  73  Figure 10 Observed pregnancy rate by cycle day quintiles  74  Table 11 Observed odds and log odds of pregnancy by cycle day quintiles Cycle day  N  N pregnant  N not pregnant  Odds of pregnancy  Log odds of pregnancy  Quintile 1  1267  13  1254  0.010367  -4.56914  Quintile 2  1286  39  1247  0.031275  -3.46493  Quintile 3  1541  59  1482  0.039811  -3.22361  Quintile 4  1273  34  1239  0.027441  -3.59570  Quintile 5  1316  26  1290  0.020155  -3.90430  75  Figure 11 Log odds of pregnancy by cycle day quintiles  76  Figure 12 Observed pregnancy rate by time categories  77  Table 12 Observed odds and log odds of pregnancy by time categories Time (hours)  N  N pregnant  N not pregnant  Odds of pregnancy  Log odds of pregnancy  0-12  1944  41  1903  0.021545  -3.83761  13-24  2171  51  2120  0.024057  -3.72735  25-36  1019  25  994  0.025151  -3.68286  37-48  1155  33  1122  0.029412  -3.52636  49-60  404  12  392  0.030612  -3.48636  61-72  398  15  383  0.039164  -3.23998  >72  109  4  105  0.038095  -3.26767  78  Figure 13 Log odds of pregnancy by time categories  79  Figure 14 Observed pregnancy rate by income quintile  80  Table 13 Observed odds and log odds of pregnancy by income quintiles Income  N  N pregnant  N not pregnant  Odds of pregnancy  Log odds of pregnancy  Quintile 1  1352  38  1314  0.02892  -3.5432  Quintile 2  1350  37  1313  0.02818  -3.5692  Quintile 3  1351  31  1320  0.02348  -3.7514  Quintile 4  1353  37  1316  0.02812  -3.5714  Quintile 5  1350  25  1325  0.01887  -3.9703  81  Figure 15 Log odds of pregnancy by average population neighbourhood income quintiles  82  Figure 16 Log odds of pregnancy by median household neighbourhood income quintiles  83  Stratified analyses The association between EC type and pregnancy was estimated with stratification on one covariate at a time. The age variable was initially converted into five categories: ≤19, 20-24, 2529, 30-34, and ≥35 years). The observed pregnancy rate within each category was 1.3%, 2.5%, 4.1%, 4.0%, and 1.8%, respectively (Figure 7 above). Given the seemingly homogenous effect in the 25-29 and the 30-34 categories, the two were combined, so that the age variable was analyzed in four strata. As shown in Table 14, the stratum-specific risk ratios (LNG vs. YZP) were similar at 0.50, 0.69, 0.72, and 0.62 in the four strata, respectively. The pooled risk ratio (RRMH), adjusting for age, was 0.67 (95% CI: 0.51-0.89). The cycle day variable was stratified in quintiles, as shown in Table 15. The five strata had similar stratum-specific risk ratios except in the first stratum, which included small cells of only 4 and 9 pregnancies in the LNG and YZP groups, respectively. The time variable was initially converted into seven categories. Based on the observed pregnancy rates, the variable was redefined and analyzed in three strata (Table 16). The income variable was initially converted into quintiles; however, when two of the quintiles were observed to have seemingly homogenous pregnancy rates (Figure 14 above), they were combined for a simpler analysis with only four strata (Table 17). The results of stratified analyses for the categorical covariates are shown in Tables 18, 19, 20, and 21. Stratified analysis was not performed on the Current HC variable because the data were very sparse. For several of the analyzed covariates, the results of stratification suggested possible interaction, because the risk ratios across strata of the variables were dissimilar. Small event rates were noted in subgroups of most variables, and this finding has resulted in wide confidence  84  intervals in some of the stratum-specific risk ratios. Nevertheless, the possibility of interaction was evaluated in logistic regression models, as explained in the next section. The unadjusted risk ratio for EC type (LNG vs. YZP) and the pooled risk ratio of pregnancy adjusted by each covariate are summarized in Table 22. The pooled risk ratios were found to be similar to, and statistically significant in the same direction as, the unadjusted risk ratio of 0.71 (95% CI: 0.54-0.94).  85  Table 14 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the age variable Age category (years) ≤19  20-24  ≥35  25-34  LNG  YZP  LNG  YZP  LNG  YZP  LNG  YZP  Preg.  9  16  23  25  56  44  11  9  Not preg.  1010  895  1084  802  1533  857  744  375  Total  1019  911  1107  827  1589  901  755  384  Risk  0.009  0.018  0.021  0.030  0.035  0.049  0.015  0.023  RR  0.503 (0.223-1.132)  0.687 (0.393-1.202)  0.722 (0.490-1.062)  0.622 (0.260-1.487)  OR  0.498 (0.219-1.134)  0.681 (0.384-1.208)  0.711 (0.475-1.065)  0.616 (0.253-1.500)  LNG OR RR YZP  Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.670 (95% CI: 0.506-0.886)  86  Table 15 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the cycle day variable Cycle day Quintile 1  Quintile 2  Quintile 3  Quintile 4  Quintile 5  LNG  YZP  LNG  YZP  LNG  YZP  LNG  YZP  LNG  YZP  Preg.  4  9  22  17  33  26  18  16  11  15  Not preg.  741  513  750  497  926  556  757  482  758  532  Total  745  522  772  514  959  582  775  498  769  547  Risk  0.005  0.017  0.028  0.033  0.034  0.045  0.023  0.032  0.014  0.027  RR  0.311 (0.096-1.006)  0.862 (0.462-1.607)  0.770 (0.466-1.275)  0.723 (0.372-1.404)  0.522 (0.242-1.127)  OR  0.308 (0.094-1.005)  0.858 (0.451-1.631)  0.762 (0.451-1.288)  0.716 (0.362-1.418)  0.515 (0.235-1.129)  LNG OR RR YZP  Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.688 (95% CI: 0.512-0.925)  87  Table 16 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the time variable Time category (hours) 0-36  37-60  >60  LNG  YZP  LNG  YZP  LNG  YZP  Preg.  65  52  19  26  7  12  Not preg.  3092  1925  871  643  258  230  Total  3157  1977  890  669  265  242  Risk  0.021  0.026  0.021  0.039  0.026  0.050  RR  0.783 (0.546-1.122)  0.549 (0.307-0.984)  0.533 (0.213-1.331)  OR  0.778 (0.538-1.126)  0.539 (0.296-0.983)  0.520 (0.201-1.343)  LNG OR RR YZP  Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.688 (95% CI: 0.515-0.919)  88  Table 17 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the income variable Income Quintiles 1 & 2  Quintile 3  Quintile 4  Quintile 5  LNG  YZP  LNG  YZP  LNG  YZP  LNG  YZP  Preg.  43  32  16  15  18  19  10  15  Not preg.  1478  1149  787  533  815  501  894  431  Total  1521  1181  803  548  833  520  904  446  Risk  0.028  0.027  0.020  0.027  0.022  0.037  0.011  0.034  RR  1.043 (0.665-1.638)  0.728 (0.363-1.460)  0.591 (0.313-1.116)  0.329 (0.149-0.726)  OR  1.045 (0.657-1.661)  0.722 (0.354-1.474)  0.582 (0.303-1.120)  0.321 (0.143-0.721)  LNG OR RR YZP  Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.730 (95% CI: 0.543-0.980)  89  Table 18 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of pregnancy variable History Prior pregnancy present  Prior pregnancy absent  LNG  YZP  LNG  YZP  Pregnant  14  12  85  82  Not pregnant  426  328  3945  2601  Total  440  340  4030  2683  Risk  0.032  0.035  0.021  0.031  RR (95% CI)  0.902 (0.423-1.924)  0.690 (0.512-0.931)  OR (95% CI)  0.898 (0.410-1.968)  0.683 (0.503-0.929)  LNG OR RR YZP  Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.716 (95% CI: 0.542-0.945)  90  Table 19 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of gynaecological condition variable History Gynaecological condition present  Gynaecological condition absent  LNG  YZP  LNG  YZP  7  5  92  89  Not pregnant  221  125  4150  2804  Total  228  130  4242  2893  Risk  0.031  0.038  0.022  0.031  Pregnant  RR (95% CI)  0.798 (0.259-2.464)  0.705 (0.529-0.940)  OR (95% CI)  0.792 (0.246-2.547)  0.698 (0.520-0.938)  LNG OR RR YZP  Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.710 (95% CI: 0.538-0.939)  91  Table 20 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of emergency contraceptive variable History Prior EC present  Prior EC absent  LNG  YZP  LNG  YZP  5  15  94  79  Not pregnant  357  298  4014  2631  Total  362  313  4108  2710  Risk  0.014  0.048  0.023  0.029  Pregnant  RR (95% CI)  0.288 (0.106-0.784)  0.785 (0.584-1.055)  OR (95% CI)  0.278 (0.100-0.775)  0.780 (0.576-1.056)  EC LNG OR RR YZP  Emergency contraceptive Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.713 (95% CI: 0.539-0.944)  92  Table 21 The association between emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pregnancy stratified by the history of hormonal contraceptive (excluding concurrent use) variable History Hormonal contraceptive present  Hormonal contraceptive absent  LNG  YZP  LNG  YZP  Pregnant  22  23  77  71  Not pregnant  681  511  3690  2418  Total  703  534  3767  2489  Risk  0.031  0.043  0.020  0.029  RR (95% CI)  0.727 (0.409-1.289)  0.717 (0.521-0.985)  OR (95% CI)  0.718 (0.396-1.302)  0.711 (0.513-0.985)  LNG OR RR YZP  Levonorgestrel regimen Odds ratio Risk ratio Yuzpe regimen  Summary estimate of risk ratios across strata of the variable, pooled using the Mantel-Haenzel method: RRMH = 0.719 (95% CI: 0.544-0.950)  93  Table 22 The unadjusted risk ratio for emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) and pooled risk ratios stratified on one covariate at a time Stratification  RRMH (95% CI)  EC type (unadjusted)  0.712 (0.539-0.941)  EC type + age  0.670 (0.506-0.886)  EC type + cycle day  0.688 (0.512-0.925)  EC type + time  0.688 (0.515-0.919)  EC type + income  0.730 (0.543-0.980)  EC type + history of pregnancy within 1 year  0.716 (0.542-0.945)  EC type + history of gynaecological condition within 5 years  0.710 (0.538-0.939)  EC type + history of EC within 1 year  0.713 (0.539-0.944)  EC type + history of HC (excluding concurrent use) within 1 year  0.719 (0.544-0.950)  EC Emergency contraceptive HC Hormonal contraceptive RRMH Summary estimate of risk ratios across strata of the variable, pooled using the MantelHaenzel method  94  Multivariate logistic regression analyses The results of the univariate logistic regression analyses are shown in Table 23. Continuous variables were categorized in the univariate analyses so that the effect across categories of the variables could be examined. The odds of pregnancy was estimated relative to the first category (the reference category). The unadjusted odds ratio for EC type as a predictor was 0.706 (95% CI: 0.530-0.940). The formula of the unadjusted model was: Ln odds(pregnancy) = ß0 + ß1*EC type where ß0 was the coefficient for the intercept, ß1 was the coefficient for EC type. The unadjusted model had an AIC of 1792.  95  Table 23 Odds ratio of pregnancy predicted by univariate regression modeling Variable  N  Predicted odds ratio (95% CI) compared to the reference group marked with an asterisk  p-value  Emergency contraceptive regimen type Yuzpe regimen*  3023  1  Levonorgestrel regimen  4470  0.706 (0.530-0.940)  ≤19*  1930  1  20-24  1934  1.939 (1.191-3.158)  0.008  25-29  1407  3.217 (2.000-5.175)  <0.001  30-34  1083  3.151 (1.913-5.188)  <0.001  ≥35  1139  1.362 (0.753-2.463)  0.307  0.017  Age (years)  Menstrual cycle day on which the act of unprotected intercourse occurred Quintile 1*  1267  1  Quintile 2  1286  3.017 (1.603-5.679)  0.001  Quintile 3  1541  3.840 (2.097-7.034)  <0.001  Quintile 4  1273  2.647 (1.390-5.040)  0.003  Quintile 5  1316  1.944 (0.995-3.800)  0.052  96  Variable  N  Predicted odds ratio (95% CI) compared to the reference group marked with an asterisk  p-value  Time between the act of unprotected intercourse and emergency contraceptive dispensing (hours) 0-12*  1944  1  13-24  2171  1.117 (0.737-1.692)  0.603  25-36  1019  1.167 (0.706-1.931)  0.547  37-48  1155  1.365 (0.858-2.172)  0.189  49-60  404  1.421 (0.740-2.728)  0.291  61-72  398  1.818 (0.996-3.317)  0.052  >72  109  1.768 (0.622-5.029)  0.285  Quintile 1*  1352  1  Quintile 2  1350  0.974 (0.616-1.542)  0.912  Quintile 3  1351  0.812 (0.502-1.313)  0.396  Quintile 4  1353  0.972 (0.614-1.539)  0.904  Quintile 5  1350  0.652 (0.392-1.087)  0.101  Neighbourhood income  History of pregnancy within 1 year Absent*  6713  1  Present  780  1.352 (0.888-2.057)  0.160  History of gynaecological condition within 5 years Absent*  7135  1  Present  358  1.332 (0.736-2.414)  0.344 97  Variable  N  Predicted odds ratio (95% CI) compared to the reference group marked with an asterisk  p-value  History of emergency contraceptive within 1 year Absent*  6818  1  Present  675  1.173 (0.733-1.876)  0.506  Concurrent hormonal contraceptive prescription Absent*  6971  1  Present  522  0.496 (0.232-1.060)  0.070  History of hormonal contraceptive prescription (excluding concurrent use) within 1 year Absent*  6256  1  Present  1237  1.558 (1.110-2.188)  0.010  * Reference category.  98  Covariates were added iteratively to build a series of multivariate models. Age and cycle day were the first two covariates to be entered into the model because they were most strongly associated with pregnancy. These two variables had non-monotonic dose-response curves, as shown in Figures 8 and 11 in the Exploratory Analyses section above. The shapes suggested that the addition of polynomial terms could improve model fit. To examine the effect of polynomial terms, a new univariate regression model was built for each of these variables, with the addition of a quadratic term that was centered at the mean. The univariate model with age was: Ln odds(pregnancy) = ß0 + ß1*Age + ß2*(Age – Mean age)2 The univariate model with cycle day was: Ln odds(pregnancy) = ß0 + ß1*Cycle day + ß2*(Cycle day – Mean cycle day)2 The predicted values of the variables and the log odds of pregnancy are shown in Figures 17 and 18 for the age variable, and Figures 19 and 20 for the cycle day variable. Based on these results, quadratic terms were included in multivariate models whenever the age and cycle day variables were included, as follows. The model adjusting for age was: Ln odds(pregnancy) = ß0 + ß1*EC type + ß2*Age + ß3*(Age – Mean age)2 Addition of age to the model improved the fit (AIC was lower at 1748). The model adjusting for age and cycle day was: Ln odds(pregnancy) = ß0 + ß1*EC type + ß2*Age + ß3*(Age – Mean age)2 + ß4*Cycle day + ß5*(Cycle day – Mean cycle day)2 Addition of both age and cycle day further improved the fit, as demonstrated by a substantial reduction of the AIC to 1534. This was deemed to be the best model on balance of parsimony  99  and model fit. The Hosmer-Lemeshow Goodness-of-Fit test statistic had a p-value of 0.5725. The regression equation associated with this model was: Ln odds(pregnancy) = - 12.0795 - 0.4466*EC type + 0.4863*Age - 0.00826*(Age – Mean age)2 + 0.2582*Cycle day – 0.00713*(Cycle day – Mean cycle day)2 The odds ratio for EC type, adjusted for age and cycle day, was 0.640 (95% CI: 0.471-0.870) for the LNG group relative to the YZP group. As shown in Table 24, addition of other variables did not change the odds ratio or AIC substantially. In all iterations, EC type was significant and the odds ratio was approximately 0.65. These findings suggested that the model adjusting for age and cycle day adequately described the association between EC regimen type and pregnancy, while controlling for the most influential amongst measured covariates. It should be noted that the time variable was modeled as a single linear term (i.e., in its original continuous form) because log odds of pregnancy was approximately linearly related to time (see Figure 13 in the Exploratory Analyses section above). The income variable was excluded because it was not a predictor of pregnancy, given the available data in this study. When interaction terms were incorporated as the product of selected variables, these terms were not statistically significant. In the assessment of multicollinearity, none of the variables were found to have a Tolerance below 1.0 or a VIF value above 10. In fact, the variables had VIF values well below the commonly used threshold of 10.150  100  Figure 17 Probability of pregnancy predicted by incorporating age in univariate regression as a linear term as well as a quadratic term  101  Figure 18 Log odds of pregnancy predicted by incorporating age in univariate regression as a linear term as well as a quadratic term  102  Figure 19 Probability of pregnancy predicted by incorporating cycle day in univariate regression as a linear term as well as a quadratic term  103  Figure 20 Log odds of pregnancy predicted by incorporating cycle day in univariate regression as a linear term as well as a quadratic term  104  Table 24 Multivariate regression model-building to estimate the odds ratio of pregnancy for emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) as the explanatory variable of interest, adjusted for potential confounders Variable(s) in the Model  ß  SE  OR  95% CI  Wald Chi-sq p-value  AIC  HL pvalue  EC type (unadjusted)  -0.3486 0.1460 0.706  0.530-0.939  0.0169  1792  EC type + age  -0.4184 0.1471 0.658  0.493-0.878  0.0044  1748 0.6131  EC type + age + cycle day  -0.4466 0.1566 0.640  0.471-0.870  0.0043  1534 0.5725  EC type + age + cycle day + history of HC (excluding concurrent use)  -0.4365 0.1567 0.646  0.475-0.879  0.0054  1532 0.7666  EC type + age + cycle day + history of HC (excluding concurrent use) + time  -0.4001 0.1573 0.670  0.492-0.912  0.0110  1526 0.4612  EC type + age + cycle day + history of HC (excluding concurrent use) + time + history of pregnancy  -0.3958 0.1575 0.673  0.494-0.917  0.0120  1527 0.3313  EC type + age + cycle day + time  -0.4103 0.1571 0.663  0.488-0.903  0.0090  1528 0.6401  ß AIC EC HL HC OR SE  Beta-coefficient for emergency contraceptive type as a predictor Akaike information criterion emergency contraceptive Hosmer-Lemeshow Goodness-of-Fit test statistic Hormonal contraceptive Odds ratio Standard error  105  Sensitivity analyses In the primary analysis, the odds ratio for EC type, adjusted for age and cycle day using polynomial terms of these covariates, was 0.64 (95% CI: 0.47-0.87). Alternative model specifications were explored to determine the sensitivity of this finding. A new logistic regression model was built with each of age and cycle day as an array of dichotomous categorical variables. The odds ratio for EC type as a predictor was 0.651 (95% CI: 0.478-0.885), and the AIC was 1541. In yet another model specification, an AIC-minimizing combination of spline expansions of age and cycle day was identified, and then logistic regression was performed on a second degree spline expansion with one knot of the age variable and a second degree spline expansion with three knots of the cycle day variable. The odds ratio for EC type in this model was 0.639 (95% CI: 0.470-0.869), and the AIC was 1525. The results from the various model specifications are summarized in Table 25. The effect of measured and unmeasured confounding was assessed using an array of assumptions152,153 about the prevalence of confounder(s) in the two EC regimen groups, and the confounder-pregnancy association (Table 26). The joint magnitude of association of confounders would have to be strong with the EC regimen type and with pregnancy to change the results substantially. For example, an odds ratio of 0.21 for the EC type-confounder association and an odds ratio of 5 for the confounder-pregnancy relationship would be necessary to move the point estimate of the EC type-pregnancy association to the null.  106  Table 25 Adjusted odds ratio for emergency contraceptive regimen type (levonorgestrel vs. Yuzpe) under various logistic regression model specifications for controlling the age and cycle day variables Method  ß  SE  OR  95% CI  Wald Chi-sq p-value  AIC  (1) Single linear terms  -0.3968  0.1560  0.672  0.495-0.913  0.0110  1589  (2) Categorization  -0.4299  0.1567  0.651  0.478-0.885  0.0061  1541  (3) Polynomial terms  -0.4466  0.1566  0.640  0.471-0.870  0.0043  1534  (4) Spline expansions  -0.4482  0.1569  0.639  0.470-0.869  0.0043  1525  (1) The age and cycle day variables were each modeled as a single linear term (i.e., in the original continuous form). (2) The age variable was converted into an array of 4 discrete categories (i.e., 3 indicator variables) and the cycle day variable was converted into an array of 9 discrete categories (i.e., 8 indicator variables). (3) The age and cycle day variables were each modeled as a quadratic term in addition to the linear term. The quadratic term was created by taking the difference between the value of a continuous variable with the mean of the variable, and then squaring the difference. (4) The analysis was performed on a second degree spline expansion with one knot of the age variable and a second degree spline expansion with three knots of the cycle day variable. ß AIC OR SE  Beta-coefficient for emergency contraceptive regimen type as a predictor Akaike Information Criterion Odds ratio Standard error  107  Table 26 Effect of confounding on the emergency contraceptive type-pregnancy odds ratio based on an array of assumptions (adjusted odds ratios for the emergency contraceptive type-pregnancy association are shown inside the thick border) Prevalence of the confounder(s) among levonorestrel users  Prevalence of the confounder(s) among Yuzpe users  Emergency contraceptive typeconfounder odds ratio  Confounderpregnancy odds ratio assumed to be 2  Confounderpregnancy odds ratio assumed to be 5  Confounderpregnancy odds ratio assumed to be 10  0.02  0.30  0.048  0.899  1.438  2.213  0.05  0.30  0.123  0.874  1.294  1.801  0.02  0.20  0.082  0.830  1.176  1.675  0.05  0.20  0.211  0.807  1.059  1.363  0.02  0.10  0.184  0.761  0.915  1.136  0.05  0.10  0.474  0.739  0.823  0.925  0.10  0.10  1.000  0.706  0.706  0.706  0.10  0.05  2.111  0.674  0.605  0.539  108  Aim 3 The observed pregnancy rates in the LNG and YZP groups were 2.2% and 3.1%, respectively. The cycle day of UPI was available for 6683 (89.2%) of the women in the study cohort. Figure 21 illustrates the distribution of expected pregnancies in the cohort over a range of menstrual cycle days. The expected pregnancy rate was estimated to be 4.2% and 4.1% for the LNG and YZP groups, respectively. The magnitude of effect is expressed in various ways in Table 27. Against the estimated expected pregnancy rate of 4.2% in the LNG group, EC effectiveness (as the relative risk reduction of pregnancy) was 47.6%, the absolute risk reduction was 2.0%, and accordingly, the NNT was 50 women to prevent one pregnancy. For the YZP regimen, the relative risk reduction was 24.4%, the absolute risk reduction was 1.0%, and the NNT was 100. In the subgroup of women with cycle day information (n = 6683), 4020 (60.2%) received the LNG regimen and 2663 (39.8%) received the YZP regimen. Expected and observed pregnancy rates in the subgroup were consistent with the results in the entire cohort. The absolute risk reduction was 2.0% (95% CI: 1.5-2.5%) and 0.9% (95% CI: 0.3-1.6%) in the LNG and YZP groups, respectively. The NNT was 50 (95% CI: 41-65) and 107 (95% CI: 63-370), respectively.  109  Figure 21 The number of pregnancies expected after unprotected intercourse on menstrual cycle Day 1 to Day 40 among women in the cohort  110  Table 27 The magnitude of the treatment effect of emergency contraceptives expressed in a number of ways Effect measure  Levonorgestrel regimen (n = 4470)  Yuzpe regimen (n = 3023)  Expected pregnancy rate (risk)*  4.2%  4.1%  Observed pregnancy rate (risk)  2.2%  3.1%  Relative risk reduction  47.6%  24.4%  Absolute risk reduction  2.0%  1.0%  50  100  Number of women needed to treat to prevent one pregnancy  * The expected pregnancy rate in the absence of emergency contraceptive was estimated based on the set of conception probabilities developed by Wilcox et al. (Contraception 2001;63:211-5.) For this estimation, 6683 (89.2%) of women in the cohort reported that the index act of intercourse occurred between cycle Day 1 and Day 40. The expected pregnancy rate among these women was extrapolated to the entire cohort for calculations of relative and absolute risk reductions and the number needed to treat.  111  Aim 4 Women in the two EC regimen groups were combined for analysis of the effect of timing of postcoital EC dispensing on the pregnancy rate and odds of pregnancy. Data on the time variable were available for 7200 (96.1%) women in the study cohort. The expected pregnancy rate was approximately 4.1% over seven consecutive time categories; whereas, the observed pregnancy rate increased with time category. As shown in Table 28 and Figure 22, the absolute risk reduction decreased with time from ~2% for EC dispensed within 24 hours after UPI to ~1% at 60 hours.  Table 28 Expected and observed pregnancy rates by the time category (between unprotected intercourse and emergency contraceptive dispensation) and the absolute risk reduction Time Category (hours)  N  Expected Number of Pregnancies*  Expected Pregnancy Rate  Observed Number of Pregnancies  Observed Pregnancy Rate  Absolute Risk Reduction  0-12  1944  74.977/1819  4.12%  41  2.1%  2.02%  13-24  2171  84.189/2014  4.18%  51  2.3%  1.88%  25-36  1019  38.833/949  4.09%  25  2.5%  1.59%  37-48  1155  43.706/1069  4.09%  33  2.9%  1.19%  49-60  404  15.047/369  4.08%  12  3.0%  1.08%  61-72  398  14.668/355  4.13%  15  3.8%  0.33%  >72  109  4.402/106  4.15%  4  3.7%  0.45%  Total  7200  275.822/6681  4.13%  181  2.5%  1.63%  *Denominator is the number of women with data on the last menstrual period date  112  Figure 22 Expected and observed pregnancy rates by the time category (between unprotected intercourse and emergency contraceptive dispensing)  113  The effect of the time variable was estimated with adjustment for potential confounders using logistic regression modeling. The time variable was initially modeled in units of days: day 1, day 2, day 3, and beyond day 3. After adjusting for EC type, age, and cycle day, the time variable was significantly associated with pregnancy, with an odds ratio of 1.255 (95% CI: 1.037-1.518) for each day-increase in time (Table 29). Women in the latest time category (beyond day 3 or 72 hours) were observed to have only 4 pregnancies (Table 28 above). Therefore, the time variable was re-categorized as day 1, day 2, and beyond day 2. After controlling for the same potential confounders, the time variable remained significantly associated with pregnancy, with an odds ratio of 1.274 (95% CI: 1.038-1.565) for each additional day after UPI (Table 30). Relative to day 1, EC dispensing beyond day 2 had an odds ratio of 1.624 (95% CI: 1.054-2.502) (Figure 23).  114  Table 29 Multivariate regression model-building to estimate the odds ratio of pregnancy for the time between unprotected intercourse and emergency contraceptive dispensing as the explanatory variable of interest (in units of days: day 1, day 2, day 3, and beyond day 3), adjusted for potential confounders Variable(s) in the Model  ß  SE  Time (unadjusted)  0.1960  0.0938  Time + EC type  0.1804  Time + EC type + age Time + EC type + age + cycle day ß AIC EC HL OR SE  OR  95% CI  Wald Chi-sq p-value  AIC  HL pvalue  1.216 1.012-1.462  0.0366  1691 0.8933  0.0944  1.198 0.996-1.441  0.0558  1686 0.7572  0.1796  0.0948  1.197 0.994-1.441  0.0582  1648 0.7767  0.2268  0.0972  1.255 1.037-1.518  0.0196  1531 0.5689  Beta-coefficient for time as a predictor Akaike Information Criterion Emergency contraceptive Hosmer-Lemeshow Goodness-of-Fit test statistic Odds ratio Standard error  115  Table 30 Multivariate regression model-building to estimate the odds ratio of pregnancy for the time between unprotected intercourse and emergency contraceptive dispensing as the explanatory variable of interest (in units of days: day 1, day 2, and beyond day 2), adjusted for potential confounders Variable(s) in the Model  ß  SE  Time (unadjusted)  0.2094  0.1008  Time + EC type  0.1919  Time + EC type + age Time + EC type + age + cycle day ß AIC EC HL OR SE  OR  95% CI  Wald Chi-sq p-value  AIC  HL pvalue  1.233 1.012-1.502  0.0379  1691 0.8373  0.1014  1.211 0.993-1.478  0.0584  1686 0.7062  0.1900  0.1017  1.209 0.991-1.476  0.0617  1648 0.5774  0.2423  0.1048  1.274 1.038-1.565  0.0208  1531 0.5049  Beta-coefficient for time as a predictor Akaike Information Criterion Emergency contraceptive Hosmer-Lemeshow Goodness-of-Fit test statistic Odds ratio Standard error  116  Figure 23 Odds ratios (and 95% CI) of pregnancy for the time between unprotected intercourse and emergency contraceptive dispensing as the explanatory variable of interest (in categories of days: day 1, day 2, and beyond day 2), adjusted for emergency contraceptive regimen type, age, and cycle day.  117  DISCUSSION Discussion common to all four aims The present programme of research was a population-based investigation of EC effectiveness under conditions of usual care. Linkable health data on the prescription and medical histories of women who received ECs in the province were used in the analyses. Secondary analysis of administrative data, including claims data, is an established method of pharmacoepidemiologic research.154,155 In this study, the magnitude of effect of ECs under real world usage conditions was evaluated on a large scale by linking data within three provincial databases (PharmaNet, MSP, and Hospital Separation) and with the cohort women’s menstrual cycle information on de-identified consent forms. This was the first study to use linkable health data to evaluate EC effectiveness. Linkable health data in British Columbia have been used by researchers to study a variety of health outcomes.156,157 In this study, prescription data were linked to medical records to elicit the association between an exposure (EC prescription) and an outcome (pregnancy). In addition, the timing of the women’s LMP and index act of UPI were discernible from consent form data, so that the information could be used to describe the characteristics of the women. Data linkage played two major roles in this study. First, a connection was made between individual instances of EC exposure and subsequent pregnancy-related events of the woman, so that an observed pregnancy rate could be estimated in the assembled cohort of women. Second, a connection was made between individual EC exposure and the woman’s prescription history and medical history, so that covariate information could be incorporated into statistical procedures in the comparative effectiveness study. The dataset generated from data linkages incorporated patient-specific demographic and reproductive health information from consent forms,  118  corresponding prescription data from PharmaNet, and medical diagnoses from health service databases. The linked data were then used to investigate the four main aims. The present study was conducted on data that were collected in the most data-rich period between December 2000 and December 2002. In fact, the chosen study period is the most datarich period in the history of EC provision in British Columbia, and may continue to be so because the LNG regimen was deregulated to non-prescription status in May 2007, and information on usage and the users has not been routinely collected since then. Within this 25month period, approximately 14000 ECs were prescribed by pharmacists in the province. A major challenge in data linkage was the lack of a personal identification code that was common between PharmaNet and consent records, so it was not possible to make direct links between the two databases. Unambiguous matching of PharmaNet and consent records was made possible by the algorithms (Figures 3 and 4), which selected for records with unique combinations of the four matching criteria. Records of EC prescriptions for 7493 women were matched unambiguously on the four criteria. The characteristics of women with matched records (the study cohort) were found to be similar to those related to excluded records (Table 6). The exposure variable of interest was the EC regimen type, which was discernible from PharmaNet and consent data. In both databases, the accuracy of this information depended on the accuracy of pharmacists’ data entry. Despite a recent report on the inaccuracy of PharmaNet records of some types of medications,158 other researchers have suggested minimal misclassification or underreporting of prescription information in PharmaNet.57,159 Certified pharmacists in British Columbia were trained to counsel women on the indications, treatment options, possible mechanisms of action, and common adverse effects of ECs. It was expected that the choice between the YZP and LNG regimens would be decided upon by the women, in  119  discussion with the pharmacists. As the data revealed, approximately 40% of women selected the YZP regimen, despite being informed by pharmacists of possible lower effectiveness and increased incidence of nausea and vomiting compared to the LNG regimen.160 The outcome variable of interest was pregnancy. It was not classifiable without ambiguity using a priori criteria, because there is no standard definition of pregnancy with the MSP and Hospital Separation databases. In this study, the outcome was ascertained by a process of expert adjudication, which is an approach taken by some observational study investigators and clinical trialists.161,162 A list of potential pregnancy indicators was developed in consultation with experts, who were experienced in billing for related services during the study period. The screening periods for relevant pregnancy (including abortion-specific) codes were selected based on local clinical practice during the study period. In British Columbia, post-term pregnancy is defined as pregnancy beyond 42 weeks gestation from the LMP, and only 0.1% of pregnancies were post-term during the years 2000-2002.121 In theory, the 42-week screening period used in this study captured almost all pregnancy events. Even if labour and delivery occurred beyond 42 weeks, prenatal care services would have been identified, and the subsequent pregnancy outcome would have been manually captured and presented to the experts for outcome adjudication. Furthermore, the reference point of screening was the EC dispensing date, which occurred after the LMP. Thus, the actual screening period exceeded 42 weeks’ gestation, making the screening method a conservative approach that favoured inclusion of potential events. Abortion-related codes were screened for up to 20 weeks—the oldest gestational age at which any clinic in British Columbia performed surgical abortions at that time, according to the experts interviewed in this study.  120  For the expert adjudication procedure, time profiles were created to present the administrative codes, definitions, and dates in a standardized fashion. The presence of candidate pregnancy-related codes was not sufficient to define the outcome. The codes also had to be (apparently) time-compatible with the index UPI. Preliminary data analysis suggested that most women in the cohort who became pregnant had an abortion. Therefore, local experts were interviewed for clinic- or hospital-specific guidelines on the earliest and latest gestational age at which abortions were performed circa 2001-2002. For each case with abortion-related service code(s) on the timeline, the specific clinic or hospital facility guidelines were documented on the time profiles, without identifying the facility. The three experts invited to perform outcome adjudication were among the group of experts who advised on the list of codes to screen for. They were insightful in the likely timing of abortion and other pregnancy-related events in relation to the billing and clinical codes, and they were blinded to the EC regimen type throughout the adjudication process. If the pregnancy status was misclassified in some cases, the outcome misclassification would likely have been non-differential between the two EC regimen groups, and the bias produced by the misclassification would therefore be towards the null, assuming absence of bias from other sources.152  121  Aim 1 The characteristics of women who requested ECs from community pharmacists in British Columbia are described in Table 7. Between the two oral ECs, the YZP regimen was (and continues to be) substantially less expensive to the uninsured public than the LNG regimen. In this study, the mean age of women who received the lower-cost regimen was younger, but only modestly so. The women in the YZP group also appeared to have delayed EC access compared to the LNG group, with a mean difference of 3 hours. The two regimen groups were otherwise similar in characteristics. In the study cohort of 7493 women, 99 and 94 pregnancies were observed in the LNG and YZP groups, respectively, and the odds ratio was 0.71 (95% CI: 0.53 to 0.94). The outcome would have been misclassified if pregnancy occurred but was not documented in the MSP and Hospital Separation databases (e.g., due to coding error, unrecognized early pregnancy loss, or lack of MSP registration and the associated PHN), or documented with uncommonly used codes. Misclassification is also possible if some of the cohort women left the province and sought pregnancy-related services elsewhere. Other studies of EC users have noted subsets of women lost to follow-up,6,13,51,53-55,72,74-95,97 to the extent of 33% of enrolled women.79 In the present study, the extent of missed pregnancies would have to be differential between the two regimens, with at least seven additional pregnancies in the LNG group (compared to the YZP group) to nullify the significant result in the odds ratio of pregnancy (Table 8). The observed pregnancy rate was 2.2% and 3.1% in the LNG and YZP groups, respectively. However, in a previous review of published studies, the pooled pregnancy rate was only 1.7% and 2.0% among women who received the LNG and YZP regimens, respectively.67 The data suggest that the observed pregnancy rates in the present study might have been  122  overestimated. The use of administrative data for outcome ascertainment potentially biased the pregnancy rates upward, by including pregnancies that were not the result of the index UPI. As mentioned above, code screening was conducted in a manner that favoured inclusion of potential events. In comparison, some clinical studies have also reported pregnancy rates that included women who were found to have become pregnant from acts of UPI before or after the index act.6,51 Unless women are systematically tested at the time of EC use and study enrollment to exclude existing pregnancies, and they abstain from UPI until the next menses, the pregnancy rate cannot be accurately determined. Overestimated observed pregnancy rates (in this study and elsewhere in the literature) bias EC effectiveness estimates computed from these rates; the effect will be discussed in the section on Aim 3.  123  Aim 2 Post-marketing comparative effectiveness research like the present study is important for evaluating medications under usual clinical circumstances.163 Although the LNG and YZP regimens have been directly compared in randomized trials,51,85,90 the inclusion criteria of these trials have restricted enrollment to women with regular menstrual cycles and reportedly only one act of intercourse within 48 or 72 hours of requesting EC. Thus, the results could not be generalized to the routine clinical setting. In this first investigation of the comparative effectiveness of the two regimens, data that were collected during routine clinical practice were linked and analyzed. The LNG regimen was associated with a lower pregnancy rate than the YZP regimen (2.2% vs. 3.1%, p = 0.02). The unadjusted odds ratio was 0.71 (95% CI: 0.530.94), favouring the LNG regimen as the more effective regimen. The apparent difference in effectiveness between the two regimens was evaluated in stratified and multivariate regression analyses, with adjustment for potential confounding. In this study, the following covariates were measured and analyzed: age, cycle day (on which the index UPI reportedly occurred), time (between UPI and EC dispensing), neighbourhood income, and history of pregnancy, specified gynaecological conditions, previous EC dispensation, and past or concurrent HC use. Before the stratified and multivariate regression analyses could be carried out, the covariates were described in a series of exploratory analyses. It was particularly important to examine the shapes of dose-response for the covariates measured on continuous scales, because the results implicate the stratified analyses, and the form in which the variables should be fitted in regression models.127,147 Age and cycle day were important predictors of pregnancy in this study. Complete data were available on the women’s age, as documented in PharmaNet and consent databases. Cycle  124  day information was derived from self-reported LMP and UPI dates on consent forms, which were available for 89% of the women. While the index UPI was a recent event for women at the time of requesting EC, the accuracy of the LMP date was limited by the women’s recall, which would have been especially inaccurate for women whose length of recall was 3 weeks or longer.164 Exploratory analyses revealed biologically plausible concave relationships between the observed log odds of pregnancy and each of age and cycle day (Figures 8 and 11), as anticipated for these variables.101,129,131 The observed covariate-pregnancy association suggested that the age and cycle day variables could be fitted in polynomial regression. The methods used to fit these variables are discussed in detail below. The time variable was also dependent on the accuracy of information on consent forms. It was calculated using the date and time of dispensing documented by the pharmacists, and also on the UPI date and time reported by the women on the consent forms. Information was available for 96% of the prescriptions included in the cohort. The observed pregnancy rate increased with time (between UPI and EC dispensing), as has been reported elsewhere.52,56 There was a monotonic relationship between time and the observed log odds of pregnancy (Figure 13), and so the linear term alone was deemed to be appropriate when time was included in logistic models. Socioeconomic status is associated with the health outcomes of a variety of medical conditions.165,166 Studies have identified an association between socioeconomic status and pregnancy among adolescents,167,168 but data are lacking in older segments of the population. In a study comparing individual-level and aggregate measures of socioeconomic status, correlation between the two measures was found to depend on the medical condition in question and the income level of study subjects.165 Self-reported, individual-level measures of income or  125  education are infrequently available.166 Hence, many Canadian studies have used Census-derived neighbourhood income, an aggregate measure, as a socioeconomic status indicator.165,166 In the present study, only aggregate measures were available, including two Censusderived neighbourhood income measures: average population neighbourhood income and median household neighbourhood income. The latter has been used in the main analysis in a number of studies.118,119,165 In the present study cohort, average population neighbourhood income was available for 90% of women, and so it was used in the main analysis. The observed log odds of pregnancy of the cohort women was not sensitive to levels of either neighbourhood income measure (Figures 15 and 16), and so the use of median household neighbourhood income data would have had minimal effect on the results. Ecological measures such as neighbourhood-level income may be insufficiently sensitive to socioeconomic effects in some studies.169 The present study was thus limited in its ability to control for confounding related to socioeconomic status. The possible effect of residual confounding is discussed below. Other covariates of clinical interest were derived from the women’s medical and prescription histories. History of pregnancy (within 1 year) and history of relevant gynaecological conditions (within 5 years) were screened for in the MSP and Hospital Separation databases, and so the data were subject to possible inaccuracies. History of EC or HC use was based on the presence of EC or HC dispensations (within 1 year) in the PharmaNet database. The data were also potentially limited, because there is conflicting opinion on the accuracy of PharmaNet data, and dispensation records do not accurately reflect actual use.57,158,159 Although it was feasible to create levels of these variable to make fuller use of the available data, the covariates were categorized as dichotomous variables (i.e., presence or absence) to simplify the stratified and multivariate regression analyses.  126  Stratified analyses were used to inspect the distribution of women and their pregnancy rates in subgroups of each measured covariate. It was also a method of controlling confounding, to the extent that the confounding variable has been stratified on.125 The method controlled only inter-stratum confounding, and not within-stratum confounding.124,127 To reduce residual confounding, the covariates could not be categorized into strata that were too broad.126,127 However, the strata could not be so narrow that cell sizes became too small. In the present study, the data were sparse in several strata (Tables 14-21). The association between EC type and pregnancy was adjusted by one covariate at a time. The pooled risk ratio across strata of each measured covariate (RRMH) was found to be similar to, and statistically significant in the same direction as, the unadjusted risk ratio of 0.71 (95% CI: 0.54-0.94) (Table 22). Stratification has limited ability to assess and control for multiple covariates, and so the covariates were analyzed simultaneously using regression modeling.139 A variety of methods have been proposed for selecting covariates to include in multivariate models and the form in which continuous covariates are to be included.124,127,142-144 Multivariate regression has a number of applications; the appropriate modeling approach to take depends on the application and goal of the modeling process.125 When building a predictive model, accurate prediction of the expected value of the outcome variable outweighs the goal of parsimony.125 In the present comparative effectiveness study, the goal of modeling was to control for the effect of covariates that were imbalanced between the two regimen groups, in order to estimate an adjusted association between EC regimen type and pregnancy. In doing so, a parsimonious model that fitted the data reasonably well was sought.125 Sensitivity analyses were performed to evaluate how the odds ratio for EC type changed with the addition of excluded variables or alternative model specifications.127  127  The multivariate regression analyses included various stages of model building. In this study, covariate selection was guided by the clinical relevance of covariates to pregnancy, as well as observed associations in the data.127 Among the measured covariates, age, cycle day, and history of HC were significantly associated with pregnancy in exploratory, univariate analyses. The univariate results informed the order of entry of covariates in multivariate analyses. The primary model was constructed by manually incorporating one covariate at a time, beginning with stronger predictors of pregnancy. Model fit was assessed at each step; the best model was selected on balance of model fit and parsimony. As mentioned above, the age and cycle day variables should be fitted with flexibility to account for the observed trends in dose-response. Logistic regression modeling with only the linear terms of these variables would fit these trends suboptimally, because the model inherently assumes a linear relation between levels of the variables and the log odds of pregnancy.148 To enhance the flexibility of the fitted curve, alternative model forms were specified. Polynomial terms have been recommended when continuous variables vary nonlinearly with the outcome variable.127,147 Based on the concave shape of the observed dose-response (with one directional change in the log odds of pregnancy), age and cycle day were fitted with the addition of quadratic terms, which were second-degree transformations (i.e., x2) of these covariates. For each of these covariates, the resulting log odds of pregnancy took on a concave shape that followed the data more closely than modeling the linear term alone (Figures 18 and 20). Compared to the logistic model with only the linear terms of the age and cycle day variables, the model that incorporated quadratic terms of these variables had a lower AIC (1589 and 1534 for the two methods, respectively) (Table 25). The tradeoff for improved curve fit was a more complicated model. For a continuous explanatory variable (e.g., the time variable) that  128  was linearly associated with the outcome variable, the beta-coefficient associated with the linear term of the explanatory variable described the unit-increase in the log odds of the outcome, and the odds ratio was multiplied by a constant factor.141,170 In the polynomial models, however, the log odds was a function of the linear and quadratic terms, with a beta-coefficient associated with each term. In addition to polynomial regression, other feasible strategies for fitting age and cycle day included categorization and spline regression. While all three methods have been recommended for flexible modeling of continuous covariates,141,147,148,151 polynomial regression was chosen as the primary approach, because it required fewer terms in the model than the other two methods, and it was therefore more parsimonious. In a simulation study,148 the authors observed similar magnitudes of residual confounding, when a covariate-outcome association with only one directional change was modeled using polynomial terms, categorization, or spline regression. This finding was extended to the present study, in which the results of the primary analysis were robust against those obtained using alternative modeling techniques for incorporating continuous variables. As shown in Table 25, categorization and curve-smoothing with spline regression changed the main odds ratio of interest and model fit minimally. Compared to the simplest model with only linear terms, the three flexible methods resulted in lower AIC and slightly lower point estimate in the odds ratios (Table 25). Regardless of the method used to fit the age and cycle day variables, EC regimen type was significantly associated with pregnancy, and the LNG regimen was consistently superior to the YZP regimen. Curve fitting with flexible methods of modeling helped to improve the precision of the main estimate of interest. After adjusting for age and cycle day by incorporating linear and quadratic terms of these variables, the odds ratio was 0.64 (95% CI: 0.47-0.87) for the LNG regimen compared to the  129  YZP regimen. This was considered to be the best model on balance of model fit and parsimony. In a series of analyses, covariates that were excluded in the primary analysis (including those that were nonsignificant in univariate models) were considered, and they did not change the resulting odds ratio substantially. As in other cohort studies, the multivariate analyses included only measured covariates. Unmeasured characteristics that were potential confounders included gravidity,128 coital frequency,171 and under-dosing secondary to vomiting (which is more common with the YZP regimen),51,85 among others. As illustrated in Table 26, the joint magnitude of association of confounders would have to be strong with the EC regimen type and with pregnancy by a clinically implausible magnitude for the point estimate of interest to move to the null. Therefore, the observed findings are unlikely to have been explained by confounding. Furthermore, unmeasured characteristics could be correlated with a measured covariate and thus have been partially adjusted for in the multivariate models.153 In view of the apparent strength of association between EC type and pregnancy after adjusting for imbalances in the measured covariates, the LNG regimen was deemed to be superior to the YZP regimen for preventing pregnancy in this study cohort. The superiority of the LNG regimen in this cohort study was consistent with the conclusion in the largest randomized trial to date. In the WHO Task Force trial,51 the risk ratio of pregnancy in the primary analysis was 0.36 (95% CI: 0.18-0.70) for the LNG regimen, compared to the YZP regimen. Instead of conducting an intent-to-treat analysis, the authors excluded from analysis the 2.2% of women for whom the outcome was unknown. In theory, the result was potentially biased because the apparent equivalence in group attributes (about both measured and unmeasured characteristics) produced by random assignment was not preserved. Also, the authors noted that one woman in the LNG group and three women in the YZP group were  130  already pregnant at trial enrolment. The risk ratio was modified to 0.36 (95% CI: 0.17-0.73) after excluding those four women.51 In the present study, the adjusted odds ratio was 0.64 (95% CI: 0.47-0.87). As an approximation of the risk ratio, the odds ratio underestimated the risk ratio, but only slightly, because the baseline pregnancy risks in the two regimen groups were small (far less than 20%).172 The odds ratio estimate in this study was more precise (the 95% CI was narrower) than in the WHO trial, because the study cohort was 3.8 times the size of the analyzed sample in the WHO trial. As in the WHO trial,51 the LNG group was found to be superior to the YZP group in this study, and the magnitude of difference was less pronounced than in the randomized trial. These findings were possibly related to methodological limitations of this study. First, the administrative data used in this study could not be verified, and so there was a possibility of information bias (i.e., exposure and/or outcome misclassification). However, as discussed under the section on Aim 1, if any misclassification occurred, it would likely have been inconsequential. Second, although potential confounding was assessed and controlled to the extent allowed by the available data, including several clinically important covariates, adjustments were possibly incomplete. Residual confounding stemming from unmeasured confounding and/or inadequate control of confounding with the measured covariates could, in theory, have contributed to the observed difference between the two EC regimen groups. As discussed above, a substantial confounding effect would be required to nullify the effect of LNG as measured by the odds ratio. Selection bias could also have affected the findings. The study cohort comprised the first pharmacist-initiated EC prescription of each woman whose prescription record was matched to a consent form. Selection bias would result if the women had different probabilities of being  131  included in the study cohort, because of the systematic influence of selection factor(s).125 Theoretically, when a covariate is found to be imbalanced in the study cohort, it could be the result of either selection bias or confounding.125 The two types of bias would be indistinguishable unless the distribution of the selection or confounding factor(s) in the reference population (from which observations were taken) could be examined.125 In this study, the theoretical reference population could not be described in detail, because analysis was limited to characteristics that were measurable without data linkage. If the effectiveness of the LNG and YZP regimens is truly (i.e., after excluding random and systematic errors) different between the two EC regimens, then the findings would suggest that the LNG regimen is superior to the YZP regimen in the routine clinical setting. The results of this pharmacoepidemiologic study support the LNG regimen as the more effective of the two.  132  Aim 3 In the previous section of analyses the LNG regimen was found to be more effective than the YZP regimen for preventing pregnancy under conditions of usual care. The LNG regimen has had a decade-long reputation as the more effective regimen, since the publication of the WHO Task Force trial.51 In the WHO trial and in the present cohort study, the observed pregnancy rate in the LNG group was approximately 1% and 2%, respectively. In early studies of EC, the observed pregnancy rate was the only information communicated about the outcome.173 However, it is obvious that in the remaining 98% or 99% of women who did not become pregnant, most of the women would not have become pregnant even if they had not received ECs.173 Therefore, to determine the magnitude of effect attributable to EC use, the baseline or expected pregnancy rate needs to be estimated, and compared against the observed rate. Except in two publications,19,67 EC effectiveness has been estimated and reported exclusively as a relative risk reduction, which is the proportional reduction of pregnancy risk from an expected rate to the observed. The expected pregnancy rate should ideally be estimated in a placebo study arm. However, a placebo has never been used in the study of ECs, and it is no longer feasible to implement a placebo-controlled trial for ethical reasons.67,70 Instead, the counterfactual baseline or expected pregnancy rate has been estimated by extrapolating pregnancy probability data from women who had not received EC. A number of methods have been proposed for determining the expected pregnancy rate.99-102 Using an older method, expected pregnancy rates in two studies of the YZP regimen51,89 were estimated to be 7.4% and 6.2%.69 This method was limited in that it relied on an estimated day of ovulation, which was not  133  necessarily accurate19,37,41,42,105 or available,67 and it was not adjusted for natural variation in the cycle day of ovulation.101 Subsequently, Wilcox et al.101 developed a set of probability estimates relative to UPI on a given cycle day counting from the onset of LMP, and adjusted for variation in the day of ovulation.67 When the Wilcox method was applied to the same two studies of the YZP regimen mentioned above,51,89 the expected pregnancy rates were lower at 5.2% and 5.4%, and the EC effectiveness computed using these expected rates was lower at 46.8% and 53.0%, respectively.69 In other words, EC effectiveness estimated using the older method has been overstated. One recent report, for instance, cited EC effectiveness in the range of 52-94%.174 If the expected pregnancy rate among women seeking ECs is indeed in the neighbourhood of 4%, as it was estimated to be in the present study and another large cohort of EC seekers using the Wilcox method,69 then actual EC effectiveness will be at the lower end of the cited range. The expected pregnancy rate in the present study cohort was estimated to be 4.1%, which is comparable to the rate of 3.9% among 20437 EC seekers in another study.69 These estimates are limited due to the possibility that the assumptions in the estimates were incorrect. First, the timing of the index act of UPI in the menstrual cycle was based on the women’s self-reported LMP and UPI dates. It was therefore subject to error in the women’s recall.164 Second, the pregnancy probabilities developed by Wilcox et al. were derived from a cohort of women who planned to conceive and had discontinued use of any contraceptive.101 Compared to the women (EC users) in the present study, the women enrolled in Wilcox’s study were likely to have a higher probability of pregnancy, because of their intent and behaviour. Hence, the Wilcox method—though it is likely more accurate than older methods—still potentially overestimates expected pregnancy rates (among EC users) and EC effectiveness by a small, indefinable margin.  134  The clinical community has, for many years, advocated for reporting the treatment effect of interventions using informative absolute effect measures (the absolute risk reduction and NNT), because they are more clinically meaningful and less susceptible to misinterpretation than the relative risk reduction.109,110,175 In the present study, the absolute risk reduction for the LNG regimen was 2%, and the NNT to prevent one pregnancy was 50 women (Table 27). Data from the YZP study arm of the WHO Task Force trial were re-analyzed using the Wilcox method to derive an expected pregnancy rate of 5.2%.69 Assuming that the women in the LNG group in the WHO trial had an equivalent expected pregnancy rate, the absolute risk reduction for the LNG regimen would be 4.1%, and the NNT would be 24 women. Nevertheless, the clinical literature has repeatedly referred to the LNG regimen as having an effectiveness of 85%10,176,177—the relative risk reduction estimated using an older method for estimating expected pregnancy rates. It is worth noting that none of relative risk reduction, absolute risk reduction, or NNT communicates any information about the baseline risk,67 which is important for meaningful clinical interpretation of the treatment effect. One recommended method for reporting EC effectiveness is to present it in absolute terms and accompanied by the expected pregnancy rate in the computation.67 This method of reporting was adopted in the present study. As explained above, the Wilcox method potentially overestimates expected pregnancy rates and EC effectiveness. While overestimation is a limitation for both relative and absolute measures of effect, the degree of overestimation would appear to be exaggerated in a relative risk reduction estimate. The absolute reduction of pregnancy risk was found to be only 1-2% in the present study cohort (Table 27). This magnitude of effect may be surprising to some people who have anticipated a greater effect from EC use, especially when ECs are said to reduce pregnancy by  135  50-60% (some information sources still report up to 88%) in relative terms.67 Researchers in the field are becoming increasingly aware that these medications may only be “weakly efficacious.”60,61 There is growing evidence that the LNG regimen acts primarily, if not entirely, through a preovulatory mechanism.32,49 If the LNG and YZP regimens act only by delaying or suppressing ovulation, they would only be potentially effective in the subset of EC seekers who have not yet ovulated at the time of use. Women with an ongoing need for contraception should be encouraged to consider regular contraceptive methods. In North America, many methods of regular contraception are available, with varying effectiveness under typical clinical use conditions.178 The most effective methods include long-acting reversible contraceptives (i.e., sterilization, IUDs, and implants), which have been collectively termed “forgettable contraception.”179 Other contraceptive methods are not as effective in typical use, because their effectiveness is contingent on the users’ ongoing adherence and correct usage. For women who choose to use contraceptive methods other than the “forgettable” ones, there will be situations of missed doses, usage error, or method failure, in which women become at risk of unintended pregnancy. In these situations, ECs are indicated. Given the limited effectiveness of the LNG and YZP regimens, alternative forms of emergency contraception should be considered. The alternatives include copper IUD, ulipristal acetate, mifepristone, and other products under development. A copper IUD is an effective but neglected EC option.12 In a systematic review of 42 studies, copper IUDs were associated with a very low failure rate (observed pregnancy rate) of 0.09%, after excluding one study with unusual results.180 Although IUDs are more invasive and less convenient to obtain than oral ECs, they can be left in place to provide ongoing contraception for 10 years or more.180 In recent years, ulipristal acetate, a selective progesterone receptor modulator, has become available in some  136  countries. In a randomized trial, the observed pregnancy rates among women assigned to the ulipristal and LNG regimen groups were 1.8% (95% CI: 1.0-3.0%) and 2.6% (95% CI: 1.73.9%), respectively.13 Compared to the LNG and YZP regimens, copper IUDs appear to be far more effective and, along with ulipristal, these alternatives have been shown to be effective among women seeking emergency contraception beyond 72 hours after UPI. The timing of EC is discussed in detail in the next section.  137  Aim 4 The present investigation is the largest study to have analyzed the timing of EC administration after an act of UPI. The effect of time was evaluated in 7200 women (96% of the study cohort) whose self-reported timing of UPI was abstracted from the consent forms. In all studies of ECs (including randomized trials), the time after UPI at which women presented to the clinic with a request for EC is not specifiable in the study protocol. Therefore, the timing of EC administration is an uncontrolled characteristic of treatment in both experimental and observational studies. Early studies of the LNG and YZP regimens have restricted enrollment to women who presented within 72 hours after UPI.5,51,53 Subsequent analyses in small samples of latepresenters suggested that the EC regimens might be effective between 72 and 120 hours after UPI.54,55,92 Hence, the Society of Obstetricians and Gynaecologists of Canada and the American College of Obstetricians and Gynecologists have recommended that the LNG and YZP EC regimens be offered up to 5 days after UPI.11,181 Recently, the results of four WHO-sponsored studies of the LNG regimen (total n = 6794) were combined for analysis by Piaggio et al.56 In this analysis, the observed pregnancy rate was 1.0%, 0.7%, 1.6%, 0.8%, and 5.2%, respectively, over the first 5 days after UPI. However, the expected pregnancy rate of the women in this analysis was not reported, and so the magnitude of effect attributable to EC use on each day after UPI could not be evaluated. The observed pregnancy rate (i.e., EC failure rate) was found to increase over seven chronological categories of time in the present study (Figure 12 and Table 12). It appeared to increase steadily within the first 60 hours, and then to increase by a slightly greater magnitude in the final two categories of time (61-72 and >72 hours). Consequently, the time variable was  138  approximately linearly related to pregnancy on a logarithmic scale (Figure 13). A similar trend has been identified in an analysis of 1950 women who received EC within 72 hours in the WHO Task Force Trial.52 In both the WHO trial and the present study, the data in later time categories (>48 hours) were sparse, and so the observed trends were possibly less accurate at the upper end. The present investigation is the first study in which the timing of EC is contextualized with absolute risk reduction estimates. The expected pregnancy rate was found to be approximately 4.1%, regardless of the time category of EC dispensing (Table 28 and Figure 22). With a trend of increasing observed pregnancy rate, the absolute risk reduction was found to decrease with time. The available data suggested an absolute risk reduction of approximately 2% for EC dispensed within 24 hours, and less than a 1% reduction beyond 60 hours (Table 28). As mentioned above, the paucity of data in the later categories limited the interpretation of findings beyond 48 hours after UPI. It is important to note that the expected pregnancy rate was derived from external data, the limitations of which have been discussed in the section on Aim 3. The effect of EC dispensing with each additional day after UPI was evaluated by analyzing the timing of EC in units of days. In two unadjusted analyses (based on two methods of categorizing time), the odds of pregnancy was shown to increase by 1.2 (95% CI: 1.0-1.5) for each day of delay (Tables 29 and 30). After adjusting for EC regimen type, the women’s age, and cycle day of UPI, the association between time and pregnancy remained statistically significant. As shown in Figure 23, EC dispensing beyond the second day was associated with an odds ratio of 1.6 (95% CI: 1.1-2.5), relative to the first day after UPI. This finding is comparable to the aforementioned analysis of the LNG regimen by Piaggio et al., in which the pooled odds ratio on the third day was 1.7 (95% CI: 0.9-3.2), relative to the first day.56 Beyond 72 hours after UPI, the observed pregnancy rate was 3.7% among 109 women in the present  139  study (Table 28), 2.4% among 616 women in the analysis by Piaggio et al.,56 and 3.6% among 111 women in a study of the YZP regimen.55 Interpretation of these findings requires the expected pregnancy rate, which was not reported, except in the present study. The results of the present study suggest very limited EC effectiveness (<1% absolute risk reduction) beyond 72 hours after UPI. Women and clinicians need to be aware of this information, because more effective alternatives are available. Copper IUDs have been recommended for use as an EC up to 5 days after UPI in one set of clinical guidelines181 and up to 7 days in another.11 The results of small studies have suggested that the copper IUD may be effective when used up to 10 or more days after UPI.180 Compared to the LNG and YZP EC regimens, copper IUDs are less frequently offered to women.182 However, this strategy is far more effective than oral ECs and can provide ongoing contraception.12,180 Ulipristal acetate is another potential option for late-presenting EC seekers;13 however, it is currently not available in Canada. For women who present with a request for EC within 72 hours after UPI, all available options, including the highly effective copper IUD, should be considered.12 Between 72 and 120 hours after UPI, the LNG and YZP regimens appear to have very limited effectiveness; in this situation, the LNG or YZP regimen is reasonable only if a copper IUD is declined or contraindicated. Whenever women desire to prevent unintended pregnancy after an act of UPI, EC should be used as soon as is feasible.51 Timely access to ECs is important for achieving their optimal effect. Public programmes aimed at reducing barriers to access (e.g., pharmacist provision of ECs) can profoundly impact EC use on a population level.19,160  140  Future directions Novel findings in this research programme include the relative effectiveness of the LNG and YZP regimens, and the effectiveness of these ECs in relation to time after UPI. Knowledge translation efforts need to be made to disseminate these findings to health providers, women of childbearing age (especially adolescents), school teachers, and counselors. Public education should aim at providing women and their partners with the tools to make personal decisions about family planning. Also, the public would benefit from health policies that enable timely and affordable access to ECs and regular forms of contraception. The administrative data used to address the four main aims of this study can potentially be used to explore additional research questions. For example, an expanded dataset containing both pharmacist- and physician-prescribed ECs can be used to investigate the long-term impact of ECs as a strategy to reduce the unintended pregnancy rate in the population over time. To date, ECs have not been shown to reduce the unintended pregnancy rate of a population.61 The expanded dataset mentioned above may help to elucidate the reasons for the apparent lack of population effect. Further research on the science and practice of ECs can include qualitative measurement of women’s preferences among the available types of contraceptives, and their willingness to use IUDs as ECs. In addition, programmes aimed at improving adherence to regular contraceptives (e.g., oral contraceptive pills) will likely help to reduce women’s need for emergency contraception.  141  CONCLUSIONS Emergency contraceptives are a unique form of birth control because they are usually administered after an act of UPI. In Canada and the United States, the LNG and YZP regimens have been the most commonly used ECs. The last decade has seen intense lobbying for increasing women’s access to these regimens. British Columbia was the first Canadian province in which specially trained and certified pharmacists were granted independent authority to prescribe ECs. This policy change led to a substantial increase in EC prescriptions in the province in subsequent years. Administrative data collected during the first 25 months of this policy formed the basis of this programme of research, in which de-identified data from multiple databases were linked at the individual woman level. The linked dataset was used to measure the observed and expected pregnancy rates among EC users under conditions of usual care, compare the LNG and YZP regimens, and estimate EC effectiveness in both relative and absolute terms. In addition, the effect of the timing of EC administration after an act of UPI was evaluated. The study cohort comprised 7493 women for whom de-identified EC prescription records and treatment consent forms were uniquely matched, and medical records were appended. The pregnancy status of each woman was ascertained by a process of expert adjudication, based on the coding and timing of potential pregnancy indicators. In this study, the LNG and YZP regimens had observed pregnancy rates (i.e., EC failure rates) of 2.2% and 3.1%, respectively. In multivariate analysis, the LNG regimen was found to be superior to the YZP regimen. After adjusting for imbalances in selected covariates, the LNG regimen was associated with a 35% lower odds of pregnancy compared to the YZP regimen. Over the past decade, the LNG regimen has become favoured in North America as the oral EC of choice, due in part to the observation of superior effectiveness in a randomized trial, and also to its tolerability and convenience. This is  142  the first comparative effectiveness study of the two regimens under conditions of routine clinical use; the findings provide further evidence that the LNG regimen should continue to be preferred when a choice needs to be made between the two EC regimens. Clinical decision-making is aided by clearly presented research findings. The recommended approach for reporting EC effectiveness is to present it as the absolute risk reduction and NNT, along with the expected pregnancy rate in the computation. In this study, the absolute risk reduction with the LNG regimen was 2.0% (the NNT was 50 women to prevent one pregnancy), based on a 4.2% expected pregnancy rate. For the YZP regimen, the absolute risk reduction was 1.0% (the NNT was 100 women), based on a 4.1% expected pregnancy rate. Compared to the traditional method of reporting EC effectiveness (as a relative risk reduction), absolute effect measures are less susceptible to misinterpretation. The expected pregnancy rate should be reported as well, to contextualize the treatment effect. While it is helpful to identify the LNG regimen as the more effective regimen, it is important to note that both the LNG and YZP regimens are far less effective than copper IUD for emergency contraception. Until more effective oral ECs become available, the LNG regimen remains a convenient option with minimal adverse effects. For each woman seeking EC, the choice of EC should be personalized according to the woman’s preferences, and a regular contraceptive method should be considered, if the woman has an ongoing need for contraception. The effectiveness of a given EC treatment is dependent on the timing of administration after UPI. It was noted in this study that women seeking oral ECs need to present to a clinic or pharmacy within 60 hours after UPI to benefit from a 1-2% absolute reduction in pregnancy risk. For all forms of ECs, knowledge dissemination efforts need to emphasize timely access.  143  Post-marketing research is important for evaluating various aspects of drug use in routine care. With the use of administrative data, four questions related to EC effectiveness were investigated on a large scale in this project. The findings will contribute to the growing body of EC research, and help to inform women and clinicians about this unique form of contraception.  144  REFERENCES 1.  Grimes DA. Emergency contraception and fire extinguishers: a prevention paradox. Am J  Obstet Gynecol 2002;187:1536-8. 2.  Ellertson C. History and efficacy of emergency contraception: beyond Coca-Cola. Fam  Plann Perspect 1996;28:44-8. 3.  Croxatto HB, Devoto L, Durand M, et al. Mechanism of action of hormonal preparations  used for emergency contraception: a review of the literature. Contraception 2001;63:111-21. 4.  Yuzpe AA, Thurlow HJ, Ramzy I, Leyshon JI. Post coital contraception--A pilot study. J  Reprod Med 1974;13:53-8. 5.  Yuzpe AA, Lancee WJ. Ethinylestradiol and dl-norgestrel as a postcoital contraceptive.  Fertil Steril 1977;28:932-6. 6.  Webb AM, Russell J, Elstein M. Comparison of Yuzpe regimen, danazol, and  mifepristone (RU486) in oral postcoital contraception. BMJ (Clinical research ed) 1992;305:92731. 7.  Webb AM. Gestagens, danazol and antiprogestogen in emergency contraception. Eur J  Contracept Reprod Health Care 1997;2:127-9. 8.  Zuliani G, Colombo UF, Molla R. Hormonal postcoital contraception with an  ethinylestradiol-norgestrel combination and two danazol regimens. Eur J Obstet Gynecol Reprod Biol 1990;37:253-60. 9.  Hansen LB, Saseen JJ, Teal SB. Levonorgestrel-only dosing strategies for emergency  contraception. Pharmacotherapy 2007;27:278-84. 10.  Dunn S, Guilbert E, Lefebvre G, et al. Emergency contraception. J Obstet Gynaecol Can  2003;25:673-9.  145  11.  Black A, Francoeur D, Rowe T, et al. Canadian contraception consensus. J Obstet  Gynaecol Can 2004;26:143-56. 12.  Belden P, Harper CC, Speidel JJ. The copper IUD for emergency contraception, a  neglected option. Contraception 2012;85:338-9. 13.  Glasier AF, Cameron ST, Fine PM, et al. Ulipristal acetate versus levonorgestrel for  emergency contraception: a randomised non-inferiority trial and meta-analysis. Lancet 2010;375:555-62. 14.  Croxatto HB, Brache V, Massai R, et al. Feasibility study of Nestorone-ethinylestradiol  vaginal contraceptive ring for emergency contraception. Contraception 2006;73:46-52. 15.  Jesam C, Salvatierra AM, Schwartz JL, Croxatto HB. Suppression of follicular rupture  with meloxicam, a cyclooxygenase-2 inhibitor: potential for emergency contraception. Hum Reprod 2010;25:368-73. 16.  The International Consortium for Emergency Contraception. Welcome to ICEC:  International Consortium for Emergency Contraception, 2012. Available from http://www.cecinfo.org. Accessed April 9, 2012. 17.  Glasier A. Emergency postcoital contraception. N Engl J Med 1997;337:1058-64.  18.  Jones DA, Stammers T. Why emergency contraception remains controversial. South Med  J 2009;102:5-7. 19.  Leung VWY, Soon JA, Levine M. Emergency contraception update: a Canadian  perspective. Clin Pharmacol Ther 2008;83:177-80. 20.  Trussell J, Jordan B. Mechanism of action of emergency contraceptive pills.  Contraception 2006;74:87-9.  146  21.  Larimore WL, Stanford JB, Kahlenborn C. Does pregnancy begin at fertilization? Fam  Med 2004;36:690-1. 22.  Grimes DA, Raymond EG, Scott Jones B. Emergency contraception over-the-counter: the  medical and legal imperatives. Obstet Gynecol 2001;98:151-5. 23.  Trussell J, Stewart F, Guest F, Hatcher RA. Emergency contraceptive pills: a simple  proposal to reduce unintended pregnancies. Fam Plann Perspect 1992;24:269-73. 24.  Shoveller J, Chabot C, Soon JA, Levine M. Identifying barriers to emergency  contraception use among young women from various sociocultural groups in British Columbia, Canada. Perspect Sex Reprod Health 2007;39:13-20. 25.  Gee RE, Shacter HE, Kaufman EJ, Long JA. Behind-the-counter status and availability of  emergency contraception. Am J Obstet Gynecol 2008;199:478.e1-.e5. 26.  Shacter HE, Gee RE, Long JA. Variation in availability of emergency contraception in  pharmacies. Contraception 2007;75:214-7. 27.  Fielder JH. Issues in ethics. Pharmacists refuse to fill emergency contraception  prescriptions. IEEE Eng Med Biol Mag 2005;24:88-91. 28.  Cantor J, Baum K. The limits of conscientious objection--may pharmacists refuse to fill  prescriptions for emergency contraception? N Engl J Med 2004;351:2008-12. 29.  Cohen SA. Objections, confusion among pharmacists threaten access to emergency  contraception. The Guttmacher report on public policy 1999;2:1-3. 30.  de Irala J, Del Burgo CL, de Fez CM, Arredondo J, Mikolajczyk RT, Stanford JB.  Women's attitudes towards mechanisms of action of family planning methods: survey in primary health centres in Pamplona, Spain. BMC Women's Health 2007;7:10. 31.  Grimes DA, Raymond EG. Emergency contraception. Ann Intern Med 2002;137:180-9.  147  32.  Leung VWY, Levine M, Soon JA. Mechanisms of action of hormonal emergency  contraceptives. Pharmacotherapy 2010;30:158-68. 33.  Dye HM, Stanford JB, Alder SC, Kim HS, Murphy PA. Women and postfertilization  effects of birth control: consistency of beliefs, intentions and reported use. BMC Women's Health 2005;5:11. 34.  Trussell J, Raymond EG. Statistical evidence about the mechanism of action of the  Yuzpe regimen of emergency contraception. Obstet Gynecol 1999;93:872-6. 35.  Trussell J, Ellertson C, Dorflinger L. Effectiveness of the Yuzpe regimen of emergency  contraception by cycle day of intercourse: implications for mechanism of action. Contraception 2003;67:167-71. 36.  Mikolajczyk RT, Stanford JB. Levonorgestrel emergency contraception: a joint analysis  of effectiveness and mechanism of action. Fertil Steril 2007;88:565-71. 37.  Novikova N, Weisberg E, Stanczyk FZ, Croxatto HB, Fraser IS. Effectiveness of  levonorgestrel emergency contraception given before or after ovulation--a pilot study. Contraception 2007;75:112-8. 38.  Noe G, Croxatto HB, Salvatierra AM, et al. Contraceptive efficacy of emergency  contraception with levonorgestrel given before or after ovulation. Contraception 2011;84:486-92. 39.  Greene MF. Emergency contraception: a reasonable personal choice or a destructive  societal influence? Clin Pharmacol Ther 2008;83:17-9. 40.  Valenzuela CY. Postovulatory effects of levonorgestrel in emergency contraception.  Contraception 2007;75:401-2. 41.  Stirling A, Glasier A. Estimating the efficacy of emergency contraception--how reliable  are the data? Contraception 2002;66:19-22.  148  42.  Espinos JJ, Rodriguez-Espinosa J, Senosiain R, et al. The role of matching menstrual data  with hormonal measurements in evaluating effectiveness of postcoital contraception. Contraception 1999;60:243-7. 43.  Ling WY, Robichaud A, Zayid I, Wrixon W, MacLeod SC. Mode of action of DL-  norgestrel and ethinylestradiol combination in postcoital contraception. Fertil Steril 1979;32:297302. 44.  Swahn ML, Westlund P, Johannisson E, Bygdeman M. Effect of post-coital contraceptive  methods on the endometrium and the menstrual cycle. Acta Obstet Gynecol Scand 1996;75:73844. 45.  Croxatto HB, Fuentealba B, Brache V, et al. Effects of the Yuzpe regimen, given during  the follicular phase, on ovarian function. Contraception 2002;65:121-8. 46.  Durand M, del Carmen Cravioto M, Raymond EG, et al. On the mechanisms of action of  short-term levonorgestrel administration in emergency contraception. Contraception 2001;64:227-34. 47.  Marions L, Hultenby K, Lindell I, Sun X, Stabi B, Gemzell Danielsson K. Emergency  contraception with mifepristone and levonorgestrel: mechanism of action. Obstet Gynecol 2002;100:65-71. 48.  Croxatto HB, Brache V, Pavez M, et al. Pituitary-ovarian function following the standard  levonorgestrel emergency contraceptive dose or a single 0.75-mg dose given on the days preceding ovulation. Contraception 2004;70:442-50. 49.  Leung VWY, Levine M, Soon JA. Evaluating emergency contraceptives using the  preovulatory/postovulatory method. Pharmacotherapy 2011;31:496e-7e.  149  50.  Trussell J, Ellertson C, Rodriguez G. The Yuzpe regimen of emergency contraception:  how long after the morning after? Obstet Gynecol 1996;88:150-4. 51.  Task Force on Postovulatory Methods of Fertility Regulation. Randomised controlled  trial of levonorgestrel versus the Yuzpe regimen of combined oral contraceptives for emergency contraception. Lancet 1998;352:428-33. 52.  Piaggio G, von Hertzen H, Grimes DA, Van Look PF, Task Force on Postovulatory  Methods of Fertility Regulation. Timing of emergency contraception with levonorgestrel or the Yuzpe regimen. Lancet 1999;353:721. 53.  Yuzpe AA, Smith RP, Rademaker AW. A multicenter clinical investigation employing  ethinyl estradiol combined with dl-norgestrel as postcoital contraceptive agent. Fertil Steril 1982;37:508-13. 54.  Rodrigues I, Grou F, Joly J. Effectiveness of emergency contraceptive pills between 72  and 120 hours after unprotected sexual intercourse. Am J Obstet Gynecol 2001;184:531-7. 55.  Ellertson C, Evans M, Ferden S, et al. Extending the time limit for starting the Yuzpe  regimen of emergency contraception to 120 hours. Obstet Gynecol 2003;101:1168-71. 56.  Piaggio G, Kapp N, von Hertzen H. Effect on pregnancy rates of the delay in the  administration of levonorgestrel for emergency contraception: a combined analysis of four WHO trials. Contraception 2011;84:35-9. 57.  Shalansky SJ, Levy AR, Ignaszewski AP. Self-reported Morisky score for identifying  nonadherence with cardiovascular medications. Ann Pharmacother 2004;38:1363-8. 58.  Shrader SP, Hall LN, Ragucci KR, Rafie S. Updates in hormonal emergency  contraception. Pharmacotherapy 2011;31:887-95.  150  59.  Office of Population Research, Princeton University and Association of Reproductive  Health Professionals. Types of emergency contraception: Which daily birth control pills can be used for emergency contraception worldwide? Available from http://ec.princeton.edu/worldwide/default.asp. Accessed April 7, 2012. 60.  Ho PC. Emergency contraception: Does improved access reduce the pregnancy rate?  Gynecol Endocrinol 2007;23:497-8. 61.  Raymond EG, Trussell J, Polis CB. Population effect of increased access to emergency  contraceptive pills: a systematic review. Obstet Gynecol 2007;109:181-8. 62.  Turok DK, Simonsen SE, Marshall N. Trends in levonorgestrel emergency contraception  use, births, and abortions: the Utah experience. Medscape J Med 2009;11:30. 63.  Goulard H, Moreau C, Gilbert F, Job-Spira N, Bajos N. Contraceptive failures and  determinants of emergency contraception use. Contraception 2006;74:208-13. 64.  Moreau C, Bouyer J, Goulard H, Bajos N. The remaining barriers to the use of  emergency contraception: perception of pregnancy risk by women undergoing induced abortions. Contraception 2005;71:202-7. 65.  Sorensen MB, Pedersen BL, Nyrnberg LE. Differences between users and non-users of  emergency contraception after a recognized unprotected intercourse. Contraception 2000;62:1-3. 66.  Baecher L, Weaver MA, Raymond EG. Increased access to emergency contraception:  why it may fail. Hum Reprod 2009;24:815-9. 67.  Leung VWY, Soon JA, Levine M. Measuring and reporting of the treatment effect of  hormonal emergency contraceptives. Pharmacotherapy 2012;32:210-21. 68.  Raymond E, Taylor D, Trussell J, Steiner MJ. Minimum effectiveness of the  levonorgestrel regimen of emergency contraception. Contraception 2004;69:79-81.  151  69.  Trussell J, Ellertson C, von Hertzen H, et al. Estimating the effectiveness of emergency  contraceptive pills. Contraception 2003;67:259-65. 70.  Raymond EG, Liku J, Schwarz EB. Feasibility of recruitment for an efficacy trial of  emergency contraceptive pills. Contraception 2008;77:118-21. 71.  Trussell J, Rodriguez G, Ellertson C. Updated estimates of the effectiveness of the Yuzpe  regimen of emergency contraception. Contraception 1999;59:147-51. 72.  Percival Smith R, Ross A. Post-coital contraception using dl-norgestrel/ethinyl estradiol  combination. Contraception 1978;17:247-52. 73.  Schilling LH. An alternative to the use of high-dose estrogens for postcoital  contraception. J Am Coll Health Assoc 1979;27:247-9. 74.  Guillebaud J, Kubba A, Rowlands S. Post-coital contraception with danazol, compared  with an ethinylestradiol-norgestrel combination or insertion of an intra-uterine device. J Obstet Gynaecol 1983;3:S64-8. 75.  Tully B. Post coital contraception - a study. Br J Fam Plann 1983;8:119-24.  76.  Van Santen MR, Haspels AA. A comparison of high-dose estrogens versus low-dose  ethinylestradiol and norgestrel combination in postcoital interception: a study in 493 women. Fertil Steril 1985;43:206-13. 77.  Van Santen MR, Haspels AA. Interception II: postcoital low-dose estrogens and  norgestrel combination in 633 women. Contraception 1985;31:275-93. 78.  Luerti M, Tonta A, Ferla P, Molla R, Santini F. Post-coital contraception by  estrogen/progestagen combination or IUD insertion. Contraception 1986;33:61-8. 79.  Wright DW, Thompson PM. Monitoring a post-coital contraception service. Br J Fam  Plann 1986;12:88-91.  152  80.  Percival-Smith RK, Abercrombie B. Postcoital contraception with dl-norgestrel/ethinyl  estradiol combination: six years experience in a student medical clinic. Contraception 1987;36:287-93. 81.  Bagshaw SN, Edwards D, Tucker AK. Ethinyl oestradiol and D-norgestrel is an effective  emergency postcoital contraceptive: a report of its use in 1,200 patients in a family planning clinic. Aust N Z J Obstet Gynaecol 1988;28:137-40. 82.  Friedman EHI, Rowley DEM. Post-coital contraception - a two year evaluation of a  service. Br J Fam Plann 1988;13:139-44. 83.  Kane LA, Sparrow MJ. Postcoital contraception: a family planning study. N Z Med J  1989;102:151-3. 84.  Glasier A, Thong KJ, Dewar M, Mackie M, Baird DT. Mifepristone (RU 486) compared  with high-dose estrogen and progestogen for emergency postcoital contraception. N Engl J Med 1992;327:1041-4. 85.  Ho PC, Kwan MS. A prospective randomized comparison of levonorgestrel with the  Yuzpe regimen in post-coital contraception. Hum Reprod 1993;8:389-92. 86.  Espinos JJ, Senosiain R, Aura M, et al. Safety and effectiveness of hormonal postcoital  contraception: a prospective study. Eur J Contracept Reprod Health Care 1999;4:27-33. 87.  Falk G, Falk L, Hanson U, Milsom I. Young women requesting emergency contraception  are, despite contraceptive counseling, a high risk group for new unintended pregnancies. Contraception 2001;64:23-7. 88.  Ashok PW, Stalder C, Wagaarachchi PT, Flett GM, Melvin L, Templeton A. A  randomised study comparing a low dose of mifepristone and the Yuzpe regimen for emergency contraception. BJOG 2002;109:553-60.  153  89.  Ellertson C, Webb A, Blanchard K, et al. Modifying the Yuzpe regimen of emergency  contraception: a multicenter randomized controlled trial. Obstet Gynecol 2003;101:1160-7. 90.  Farajkhoda T, Khoshbin A, Enjezab B, Bokaei M, Karimi Zarchi M. Assessment of two  emergency contraceptive regimens in Iran: levonorgestrel versus the Yuzpe. Niger J Clin Pract 2009;12:450-2. 91.  Arowojolu AO, Okewole IA, Adekunle AO. Comparative evaluation of the effectiveness  and safety of two regimens of levonorgestrel for emergency contraception in Nigerians. Contraception 2002;66:269-73. 92.  von Hertzen H, Piaggio G, Ding J, et al. Low dose mifepristone and two regimens of  levonorgestrel for emergency contraception: a WHO multicentre randomised trial. Lancet 2002;360:1803-10. 93.  Hamoda H, Ashok PW, Stalder C, Flett GM, Kennedy E, Templeton A. A randomized  trial of mifepristone (10 mg) and levonorgestrel for emergency contraception. Obstet Gynecol 2004;104:1307-13. 94.  Ngai SW, Fan S, Li S, et al. A randomized trial to compare 24 h versus 12 h double dose  regimen of levonorgestrel for emergency contraception. Hum Reprod 2005;20:307-11. 95.  Creinin MD, Schlaff W, Archer DF, et al. Progesterone receptor modulator for  emergency contraception: a randomized controlled trial. Obstet Gynecol 2006;108:1089-97. 96.  Mittal S, Sehgal R, Jindal VI, et al. Single dose levonorgestrel and two regimens of  centchroman for emergency contraception. J Turkish German Gynecol Assoc 2008;9:132-7. 97.  Dada OA, Godfrey EM, Piaggio G, von Hertzen H, Nigerian Network for Reproductive  Health Research and Training. A randomized, double-blind, noninferiority study to compare two  154  regimens of levonorgestrel for emergency contraception in Nigeria. Contraception 2010;82:3738. 98.  Young-Xu Y, Chan KA. Pooling overdispersed binomial data to estimate event rate.  BMC Medical Research Methodology 2008;8:58. 99.  Barrett JC, Marshall J. The risk of conception on different days of the menstrual cycle.  Popul Stud 1969;23:455-61. 100.  Wilcox AJ, Weinberg CR, Baird DD. Timing of sexual intercourse in relation to  ovulation. Effects on the probability of conception, survival of the pregnancy, and sex of the baby. N Engl J Med 1995;333:1517-21. 101.  Wilcox AJ, Dunson DB, Weinberg CR, Trussell J, Baird DD. Likelihood of conception  with a single act of intercourse: providing benchmark rates for assessment of post-coital contraceptives. Contraception 2001;63:211-5. 102.  Dixon GW, Schlesselman JJ, Ory HW, Blye RP. Ethinyl estradiol and conjugated  estrogens as postcoital contraceptives. JAMA 1980;244:1336-9. 103.  Trussell J, Rodriguez G, Ellertson C. New estimates of the effectiveness of the Yuzpe  regimen of emergency contraception. Contraception 1998;57:363-9. 104.  Stanford JB, Mikolajczyk RT. Methodological review of the effectiveness of emergency  contraception. Curr Womens Health Rev 2005;1:119-29. 105.  Stanford JB. Emergency contraception: overestimated effectiveness and questionable  expectations. Clin Pharmacol Ther 2008;83:19-21. 106.  Wilcox AJ, Dunson D, Baird DD. The timing of the "fertile window" in the menstrual  cycle: day specific estimates from a prospective study. BMJ (Clinical research ed) 2000;321:1259-62.  155  107.  Mikolajczyk RT, Stanford JB. A new method for estimating the effectiveness of  emergency contraception that accounts for variation in timing of ovulation and previous cycle length. Fertil Steril 2005;83:1764-70. 108.  Levine M, Soon JA. Risk of pregnancy among women seeking emergency contraceptives  from pharmacists in British Columbia. J Obstet Gynaecol Can 2006;28:879-83. 109.  Barratt A, Wyer PC, Hatala R, et al. Tips for learners of evidence-based medicine: 1.  Relative risk reduction, absolute risk reduction and number needed to treat. CMAJ 2004;171:353-8. 110.  Cook RJ, Sackett DL. The number needed to treat: a clinically useful measure of  treatment effect. BMJ (Clinical research ed) 1995;310:452-4. 111.  Ho PC. Emergency contraception: methods and efficacy. Curr Opin Obstet Gynecol  2000;12:175-9. 112.  Soon JA, Levine M, Ensom MH, Gardner JS, Edmondson HM, Fielding DW. The  developing role of pharmacists in patient access to emergency contraception. Dis Manage Health Outcomes 2002;10:601-11. 113.  Soon JA, Levine M, Osmond BL, Ensom MH, Fielding DW. Provision of emergency  contraceptives by pharmacists. Can Pharm J 2004;137:23-9. 114.  Hutchings J, Winkler JL, Fuller TS, et al. When the morning after is Sunday: pharmacist  prescribing of emergency contraceptive pills. J Am Med Womens Assoc 1998;53:230-2. 115.  Westley E, Bigrigg A, Webb A, et al. Risk of pregnancy and external validity in clinical  trials of emergency contraception. J Fam Plann Reprod Health Care 2006;32:165-9.  156  116.  Stephenson A, Hux J, Tullis E, Austin PC, Corey M, Ray J. Socioeconomic status and  risk of hospitalization among individuals with cystic fibrosis in Ontario, Canada. Pediatr Pulmonol 2011;46:376-84. 117.  Luo ZC, Wilkins R, Kramer MS. Effect of neighbourhood income and maternal  education on birth outcomes: a population-based study. CMAJ 2006;174:1415-20. 118.  Rabi DM, Edwards AL, Southern DA, et al. Association of socio-economic status with  diabetes prevalence and utilization of diabetes care services. BMC Health Serv Res 2006;6:124. 119.  Rabi DM, Edwards AL, Svenson LW, Graham MM, Knudtson ML, Ghali WA.  Association of median household income with burden of coronary artery disease among individuals with diabetes. Circ Cardiovasc Qual Outcomes 2010;3:48-53. 120.  Raj P, Soon JA, Shoveller J. Using GIS maps to interpret patterns of contraceptive use  among youth across British Columbia. Can J Clin Pharmacol 2010;17:e99. 121.  British Columbia Reproductive Care Program. Obstetric guideline 7: postterm pregnancy,  2005. Available from http://www.perinatalservicesbc.ca/NR/rdonlyres/504450DA-46F1-431EAF06-E0D364AD08E2/0/OBGuidelinesPostTerm7.pdf. Accessed April 15, 2008. 122.  King JE. Software solutions for obtaining a kappa-type statistic for use with multiple  raters, 2004. Available from http://www.ccitonline.org/jking/homepage/genkappa.doc. Accessed March 17, 2010. 123.  Fleiss JL, Tytun A, Ury HK. A simple approximation for calculating sample sizes for  comparing independent proportions. Biometrics 1980;36:343-6. 124.  Rothman KJ. Epidemiology: an introduction. New York: Oxford University Press; 2002.  125.  Szklo M, Nieto FJ. Epidemiology: beyond the basics. 2nd ed. Sudbury: Jones and Bartlett  Publishers; 2007.  157  126.  McNamee R. Regression modelling and other methods to control confounding. Occup  Environ Med 2005;62:500-6. 127.  Greenland S. Modeling and variable selection in epidemiologic analysis. Am J Public  Health 1989;79:340-9. 128.  Spira A. The use of fecundability in epidemiological surveys. Hum Reprod  1998;13:1753-6. 129.  Dunson DB, Colombo B, Baird DD. Changes with age in the level and duration of  fertility in the menstrual cycle. Hum Reprod 2002;17:1399-403. 130.  Speroff L. The effect of aging on fertility. Curr Opin Obstet Gynecol 1994;6:115-20.  131.  Leridon H. Can assisted reproduction technology compensate for the natural decline in  fertility with age? A model assessment. Hum Reprod 2004;19:1548-53. 132.  Taylor GM, Faragher EB, Chantler E, Seif MW. Fecundity in the modern city: a  comparison of couples attending antenatal clinics in Manchester (UK) and Melbourne (Australia). J Obstet Gynaecol 1999;19:489-95. 133.  Statistics Canada. 2001 Census geographic units: dissemination area. Available from  http://www12.statcan.ca/english/census01/Products/Reference/dict/geo021.htm. Accessed April 12, 2012. 134.  Healy DL, Trounson AO, Andersen AN. Female infertility: causes and treatment. Lancet  1994;343:1539-44. 135.  Westrom L, Joesoef R, Reynolds G, Hagdu A, Thompson SE. Pelvic inflammatory  disease and fertility. A cohort study of 1,844 women with laparoscopically verified disease and 657 control women with normal laparoscopic results. Sex Transm Dis 1992;19:185-92.  158  136.  Pavletic AJ, Wolner-Hanssen P, Paavonen J, Hawes SE, Eschenbach DA. Infertility  following pelvic inflammatory disease. Infect Dis Obstet Gynecol 1999;7:145-52. 137.  Ozkan S, Murk W, Arici A. Endometriosis and infertility: epidemiology and evidence-  based treatments. Ann N Y Acad Sci 2008;1127:92-100. 138.  Bernoux A, Job-Spira N, Germain E, Coste J, Bouyer J. Fertility outcome after ectopic  pregnancy and use of an intrauterine device at the time of the index ectopic pregnancy. Hum Reprod 2000;15:1173-7. 139.  Wunsch H, Linde-Zwirble WT, Angus DC. Methods to adjust for bias and confounding  in critical care health services research involving observational data. J Crit Care 2006;21:1-7. 140.  Statistics Canada. Pregnancy outcomes 2005. Catalogue no. 82-224-X. Available from  http://www.statcan.ca/english/freepub/82-224-XIE/82-224-XIE2005000.pdf. Accessed November 8, 2008. 141.  Rothman KJ. Modern epidemiology. 1st ed. Boston, Toronto: Little, Brown and  Company; 1986. 142.  Bursac Z, Gauss CH, Williams DK, Hosmer DW. Purposeful selection of variables in  logistic regression. Source Code Biol Med 2008;3:17. 143.  Austin PC, Tu JV. Automated variable selection methods for logistic regression produced  unstable models for predicting acute myocardial infarction mortality. J Clin Epidemiol 2004;57:1138-46. 144.  Sun GW, Shook TL, Kay GL. Inappropriate use of bivariable analysis to screen risk  factors for use in multivariable analysis. J Clin Epidemiol 1996;49:907-16. 145.  Greenland S. Invited commentary: variable selection versus shrinkage in the control of  multiple confounders. Am J Epidemiol 2008;167:523-9.  159  146.  Mazerolle MJ. Improving data analysis in herpetology: using Akaike's Information  Criterion (AIC) to assess the strength of biological hypotheses. Amphibia-Reptilia 2006;27:16980. 147.  Greenland S. Dose-response and trend analysis in epidemiology: alternatives to  categorical analysis. Epidemiology 1995;6:356-65. 148.  Brenner H, Blettner M. Controlling for continuous confounders in epidemiologic  research. Epidemiology 1997;8:429-34. 149.  Katz MH. Multivariable analysis: a practical guide for clinicians. 2nd ed. New York:  Cambridge University Press; 2006. 150.  O'Brien RM. A caution regarding rules of thumb for variance inflation factors. Quality &  Quantity 2007;41:673-90. 151.  Gregory M, Ulmer H, Pfeiffer KP, Lang S, Strasak AM. A set of SAS macros for  calculating and displaying adjusted odds ratios (with confidence intervals) for continuous covariates in logistic B-spline regression models. Comput Methods Programs Biomed 2008;92:109-14. 152.  Greenland S. Basic methods for sensitivity analysis of biases. Int J Epidemiol  1996;25:1107-16. 153.  Schneeweiss S. Sensitivity analysis and external adjustment for unmeasured confounders  in epidemiologic database studies of therapeutics. Pharmacoepidemiol Drug Saf 2006;15:291303. 154.  Sorensen HT, Sabroe S, Olsen J. A framework for evaluation of secondary data sources  for epidemiological research. Int J Epidemiol 1996;25:435-42.  160  155.  Schneeweiss S, Seeger JD, Maclure M, Wang PS, Avorn J, Glynn RJ. Performance of  comorbidity scores to control for confounding in epidemiologic studies using claims data. Am J Epidemiol 2001;154:854-64. 156.  Etminan M, Carleton BC, Samii A. Non-steroidal anti-inflammatory drug use and the risk  of Parkinson disease: a retrospective cohort study. J Clin Neurosci 2008;15:576-7. 157.  Morgan SG, Yan L. Persistence with hypertension treatment among community-dwelling  BC seniors. Can J Clin Pharmacol 2004;11:e267-73. 158.  Price M, Bowen M, Lau F, Kitson N, Bardal S. Assessing accuracy of an electronic  provincial medication repository. BMC Med Inform Decis Mak 2012;12:42. 159.  Marra F, Patrick DM, Chong M, Bowie WR. Antibiotic use among children in British  Columbia, Canada. J Antimicrob Chemother 2006;58:830-9. 160.  Soon JA, Levine M, Osmond BL, Ensom MH, Fielding DW. Effects of making  emergency contraception available without a physician's prescription: a population-based study. CMAJ 2005;172:878-83. 161.  Fischer JE, Seifarth FG, Baenziger O, Fanconi S, Nadal D. Hindsight judgement on  ambiguous episodes of suspected infection in critically ill children: poor consensus amongst experts? Eur J Pediatr 2003;162:840-3. 162.  Granger CB, Vogel V, Cummings SR, et al. Do we need to adjudicate major clinical  events? Clin Trials 2008;5:56-60. 163.  Schneeweiss S. Developments in post-marketing comparative effectiveness research. Clin  Pharmacol Ther 2007;82:143-56. 164.  Wegienka G, Baird DD. A comparison of recalled date of last menstrual period with  prospectively recorded dates. J Womens Health (Larchmt) 2005;14:248-52.  161  165.  Marra CA, Lynd LD, Harvard SS, Grubisic M. Agreement between aggregate and  individual-level measures of income and education: a comparison across three patient groups. BMC Health Serv Res 2011;11:69. 166.  Alter DA, Brandes S, Irvine J, Iron K. Impact of socioeconomic status on cardiovascular  outcomes in Canada. Expert Rev Pharmacoecon Outcomes Res 2003;3:691-702. 167.  Hayward MD, Grady WR, Billy JOG. The influence of socioeconomic status on  adolescent pregnancy. Social Science Quarterly 1992;73:750-72. 168.  Cromer BA, Brown RT. Update on pregnancy, condom use, and prevalence of selected  sexually transmitted diseases in adolescents. Curr Opin Obstet Gynecol 1992;4:855-9. 169.  McGregor MJ, Reid RJ, Schulzer M, Fitzgerald JM, Levy AR, Cox MB. Socioeconomic  status and hospital utilization among younger adult pneumonia admissions at a Canadian hospital. BMC Health Serv Res 2006;6:152. 170.  Katz MH. Multivariable analysis: a primer for readers of medical research. Ann Intern  Med 2003;138:644-50. 171.  Weinstein M, Stark M. Behavioral and biological determinants of fecundability. Ann N Y  Acad Sci 1994;709:128-44. 172.  Davies HT, Crombie IK, Tavakoli M. When can odds ratios mislead? BMJ  1998;316:989-91. 173.  Creinin MD. A reassessment of efficacy of the Yuzpe regimen of emergency  contraception. Hum Reprod 1997;12:496-8. 174.  Association of Reproductive Health Professionals. Update on emergency contraception.  ARHP Clinical Proceedings Feb 2011. Available from http://www.arhp.org/uploadDocs/CPECUpdate.pdf. Accessed January 28, 2012.  162  175.  McAlister FA. The "number needed to treat" turns 20--and continues to be used and  misused. CMAJ 2008;179:549-53. 176.  Black KI. Developments and challenges in emergency contraception. Best Pract Res Clin  Obstet Gynaecol 2009;23:221-31. 177.  Marions L, Cekan SZ, Bygdeman M, Gemzell-Danielsson K. Effect of emergency  contraception with levonorgestrel or mifepristone on ovarian function. Contraception 2004;69:373-7. 178.  Trussell J. Contraceptive failure in the United States. Contraception 2011;83:397-404.  179.  Grimes DA. Forgettable contraception. Contraception 2009;80:497-9.  180.  Cleland K, Zhu H, Goldstuck N, Cheng L, Trussell J. The efficacy of intrauterine devices  for emergency contraception: a systematic review of 35 years of experience. Hum Reprod 2012;27:1994-2000. 181.  The American College of Obstetrics and Gynecologists. ACOG Practice Bulletin No.  112: Emergency contraception. Obstet Gynecol 2010;115:1100-9. 182.  Karasahin KE, Keskin U. Copper IUD and emergency contraception. Contraception  2011;84:205.  163  APPENDICES Appendix A Pregnancy outcome data in clinical studies of the Yuzpe regimen for emergency contraception, containing a combination of ethinyl estradiol and either dl-norgestrel or levonorgestrel (Reproduced with permission of John Wiley and Sons)67 First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  608  Observed pregnancy rate among women with known outcome (%) 0.2  Yuzpe5  72  1  Percival Smith72  72  Schilling73  72  Yuzpe53  72  Women were instructed to contact the clinic at onset of menstrual bleeding or after 21 days if no bleeding ensued. Women were instructed to return to the clinic after onset of menstrual bleeding or after 21 days if no bleeding ensued. Women were asked to return to clinic in 3 weeks or after their next period, whichever occurred first. Clinic visit (not described in detail).  0  A pregnancy test was performed at enrolment whenever the possibility of pre-treatment pregnancy existed.  0  184  0.0  5  0  115  0.0  Not stated by the authors.  11  647  1.7  45  164  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  491  Observed pregnancy rate among women with known outcome (%) 1.8  Guillebaud74  Not stated by the authors  Clinic visit at 3 weeks after treatment.  9  25  395  2.8  116  We assumed that 25 women were lost to follow-up in the Yuzpe regimen arm (75% of the total number of women lost to follow-up in the study). We used 395 as the denominator of the pregnancy rate. This number included 298 women who kept their specific follow-up appointment and 97 seen later and about whom the authors said, “from the notes it seems obvious that post coital contraception had worked for these patients.”  Tully75  72  Appointmen t at 3-4 weeks after treatment.  11  165  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  225  Observed pregnancy rate among women with known outcome (%) 0.4  van Santen76,77  72  Mail-in questionnair e or telephone.  1  14  4  332  1.2  48  We assumed that 14 women were lost to follow-up based on the authors’ statement that the follow-up rate was 94.3% (239 x 5.7% = 14). We used 225 as the denominator of the pregnancy rate (239 – 14 = 225). Although 5 women became pregnant, only 4 were counted (excluded as the authors had) because 1 woman took half the prescribed treatment. We excluded this woman from the numerator and denominator.  van Santen77  72  Mail-in questionnair e or telephone.  Luerti78  72  Telephone and mail.  8  436  1.8  81  166  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  165  Observed pregnancy rate among women with known outcome (%) 3.6  Wright79  130  Clinic visit at one month after treatment. Women who did not return to clinic were contacted by telephone.  6  82  We assumed that 82 women were lost to follow-up in the Yuzpe regimen arm (89% of the total number of women lost to follow-up in the study were in the Yuzpe arm).  PercivalSmith80  72  Bagshaw81  72  Women were instructed to contact clinic at onset of menstrual bleeding or after 21 days if no bleeding ensued. Women were instructed to return to clinic for a pregnancy test within 5 weeks after treatment. Women who did not return for follow-up were contacted by telephone or by mail.  18  774  2.3  93  6  525  1.1  75  167  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  437  Observed pregnancy rate among women with known outcome (%) 3.4  Friedman82  72  15  Kane83  72  Zuliani8  72  Women were asked to notify the clinic when she menstruated and to return to the clinic 3 weeks after treatment. Follow-up letters were sent to women who did not return to the clinic. Clinic visit at 4 weeks after treatment. Women were instructed to return to clinic for follow-up 4 weeks after treatment. Women who did not return for follow-up provided information by telephone.  20  We assumed that all 20 women lost to follow-up were in the Yuzpe arm.  21  836  2.5  73  9  407  2.2  Not stated by the authors.  All women underwent pelvic examination at enrolment. The authors excluded from analysis women who were lost to follow-up.  168  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  385  Observed pregnancy rate among women with known outcome (%) 1.0  Glasier84  72  4  Webb6  72  Women were instructed to keep a diary and return to clinic within 5 days of the expected onset of next menses. If menstrual bleeding had not occurred, further follow-up was pursued. Women were instructed to keep a diary and return to clinic after the expected onset of next menses.  13  We assumed that 13 women were lost to follow-up in the Yuzpe regimen arm (proportional to the number of women in that arm).  5  191  2.6  9  We assumed that 9 women were lost to follow-up in the Yuzpe regimen arm (~33% of the total number of women in the study). The authors included 191 women in their analysis. One woman was already pregnant at the time of enrolment, and so we excluded this subject from the numerator and denominator.  169  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  424  Observed pregnancy rate among women with known outcome (%) 3.5  Ho85  48  15  WHO51*  72  Women were instructed to keep a diary and return to clinic at 3 and 6 weeks after treatment. Women were instructed to keep a diary and return for followup about 1 week after the expected onset of next menses.  16  976  2.9  18  6 of the 15 pregnancies occurred among women with further acts of intercourse within the study cycle. A blood or urine sample was collected at enrolment to rule out pretreatment pregnancy. 3 women were already pregnant at enrolment and so we excluded them from the numerator and denominator. 583 of the 976 women had no further acts of intercourse within the study cycle (among other criteria for “correct use”); there were 11 pregnancies in this subgroup.  28  170  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  355  Observed pregnancy rate among women with known outcome (%) 0.6  Espinos86  72  2  Rodrigues54*  72 or 120  Women were instructed to contact the centre by telephone after menstruatio n or within 10 days after the expected onset of menses. Women were instructed to return to clinic for follow-up 3 weeks after treatment. Women who did not return to clinic were contacted by telephone about their outcome.  55  The authors excluded from analysis women with irregular menstrual cycles.  4  300  1.3  11  1  99  1.0  35  131 and 169 women were treated within 72 hours and between 72 and 120 hours, respectively. Urinary hCG was tested at enrolment to rule out pretreatment pregnancy. The authors excluded from analysis women with irregular menstrual cycles or missing menstrual data. A pregnancy test was performed at enrolment.  Falk87*  72  Clinic visit at 3 weeks after treatment.  171  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  471  Observed pregnancy rate among women with known outcome (%) 3.6  Ashok88  72  17  Ellertson55,89  120  Women were instructed to return to clinic for follow-up within 1 week of the expected onset of next menses, or further follow-up until menstrual bleeding. If they were unable to return for follow-up visit, they were asked to return a questionnair e. Women were instructed to return to clinic for follow-up 1 week after the expected onset of next menses. Attempts were made to contact women by telephone or by mail if they missed scheduled follow-up visits.  29  A pregnancy test was performed at enrolment whenever the possibility of pre-treatment pregnancy existed.  21  786  2.7  26  Women who had a urine pregnancy test at enrolment and tested positive were excluded. Of the 786 women with follow-up data, 693 had no further acts of intercourse within the study cycle.  172  First author  Postcoital time (hours)  Farajkhoda90*  72  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  Observed pregnancy rate among women with known outcome (%) 8.3  No. of women with no outcome data  Comments  Women 5 60 2 A pregnancy were test was instructed to performed at keep a diary enrolment. The and return authors for followexcluded up visit 1 women with week after irregular the expected menstrual onset of next cycles. menses. * Pre-treatment pregnancy was systematically assessed, and women who were already pregnant at enrolment were excluded from the observed pregnancy rate (in the numerator and denominator), where applicable.  173  Appendix B Pregnancy outcome data in clinical studies of the levonorgestrel regimen for emergency contraception (Reproduced with permission of John Wiley and Sons)67 First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  410  Observed pregnancy rate among women with known outcome (%) 2.9  Ho85  48  12  WHO51*  72  Women were instructed to keep a diary and return to clinic at 3 and 6 weeks after treatment. Women were instructed to keep a diary and return for followup about 1 week after the expected onset of next menses.  30  975  1.0  25  4 of the 12 pregnancies occurred among women with further acts of intercourse within the study cycle. A blood or urine sample was collected at enrolment to rule out pretreatment pregnancy. One woman was already pregnant at enrolment and so we excluded her from the numerator and denominator. 574 of the 975 women had no further acts of intercourse within the study cycle (among other criteria for “correct use”); there were 5 pregnancies in this subgroup.  10  174  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  1118  Observed pregnancy rate among women with known outcome (%) 1.0  Arowojolu91  72  11  von Hertzen92*  120  Hamoda93  72 or 120  Women were instructed to keep a diary and followed up at the clinic or by home visit until onset of menses. Women were instructed to keep a diary until the next menses or the scheduled follow-up visit at 1 week after the expected onset of next menses. Women were instructed to return to clinic for follow-up within 7-10 days of the expected date of next menses. Women who did not return were contacted by telephone or through subsequent attendance to the clinic.  Comments  44  2715  1.6  41  A pregnancy test was performed at enrolment to rule out pretreatment pregnancy (inclusion criterion). Results from the two LNG arms were combined.  20  858  2.3  163  Women received treatment within 72 hours after intercourse (or within 120 hours if they declined to receive an intrauterine device).  42  175  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  No. of women with no outcome data  Comments  2018  Observed pregnancy rate among women with known outcome (%) 2.0  Ngai94  120  Women were instructed to keep a diary and return to clinic for follow-up ~1 week after the expected onset of next menses.  40  53  13  774  1.7  54  Results from the two LNG arms were combined. 26 pregnancies occurred among 1566 women who reported no further acts of intercourse, and 14 pregnancies among 446 women with further acts of intercourse during the study cycle. A urine pregnancy test was performed at enrolment to rule out pretreatment pregnancy.  Creinin95*  72  Novikova37  120  Women were instructed to keep a diary and return for a followup visit 5-7 days after the expected onset of next menses. Telephone at 4-6 weeks after treatment.  3  99  3.0  Not stated by the authors.  Mittal96  120  Clinic visit at 7 days after the expected onset of next menses.  2  50  4.0  0  11 women were excluded due to insufficient blood samples for complete hormonal analysis.  176  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  Farajkhoda90*  72  Women were instructed to keep a diary and return for followup visit 1 week after the expected onset of next menses.  0  62  Observed pregnancy rate among women with known outcome (%) 0  No. of women with no outcome data  Comments  0  A pregnancy test was performed at enrolment. The authors excluded women with irregular menstrual cycles.  177  First author  Postcoital time (hours)  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  Glasier13*  120  Women were instructed to return for follow-up 57 days after the expected onset of next menses.  25  958  Observed pregnancy rate among women with known outcome (%) 2.6  No. of women with no outcome data  Comments  57  A urine pregnancy test was performed at enrolment to rule out pretreatment pregnancy. The authors excluded 159 women from analysis for a number of reasons, e.g., >35 years of age, lost to follow-up, pregnancy status unknown, pregnant before treatment or became pregnant ≥10 days after treatment. 22 pregnancies occurred among women who received the LNG regimen within 72 hours after intercourse and 3 pregnancies among women who received it between 72 and 120 hours.  178  First author  Postcoital time (hours)  Dada*  120  Follow-up procedure for pregnancy outcome  No. of pregnancies among women with outcome data  No. of women with outcome data  Observed pregnancy rate among women with known outcome (%) 0.6  No. of women with no outcome data  Comments  Clinic visit 17 2823 199 The authors at 1 week excluded after the women who expected had a positive onset of next pregnancy test menses. or whose Women who pregnancy test did not results were return were unknown at contacted by enrolment. telephone or home visit. * Pre-treatment pregnancy was systematically assessed, and women who were already pregnant at enrolment were excluded from the observed pregnancy rate (in the numerator and denominator), where applicable.  179  Appendix C Reproduction of the standardized consent form completed by the women who received emergency contraceptives and by the prescribing pharmacists (two pages)  180  181  Appendix D Methodology for correctly utilizing the age estimated from year of birth information in the matching procedure The exact age was unknown in a subset of PharmaNet records because the date of birth (clntbrdt) was missing in the Medical Services Plan (MSP) database. In these situations, it was not necessary to exclude all accompanying rows with the same dispensing date, EC type, and pharmacy local health area (LHA). The method for handling these situations is illustrated in this example (hypothetical data). Suppose that multiple rows of PharmaNet data were available. For the first four StudyIDs, the exact age could be derived from the clntbrdt in the MSP database. However, the exact age was unknown in the last six rows because clntbrdt was missing in the MSP database.  StudyID  Dispensing date  Age derived from clntbrdt  EC type (coded)  Pharmacy local health area  Estimated age from the woman’s year of birth in the PharmaNet data (age_ecp)  000001  01-JAN-01  17  2  035  18  000002  01-JAN-01  18  2  035  18  000003  01-JAN-01  18  2  035  19  000004  01-JAN-01  19  2  035  20  000005  01-JAN-01  Unknown  2  035  18 (therefore, real age is either 17 or 18)  000006  01-JAN-01  Unknown  2  035  18 (therefore, real age is either 17 or 18)  000007  01-JAN-01  Unknown  2  035  17 (therefore, real age is either 16 or 17)  000008  01-JAN-01  Unknown  2  035  19 (therefore, real age is either 18 or 19)  182  StudyID  Dispensing date  Age derived from clntbrdt  EC type (coded)  Pharmacy local health area  Estimated age from the woman’s year of birth in the PharmaNet data (age_ecp)  000009  01-JAN-01  Unknown  2  035  16 (therefore, real age is either 15 or 16)  000010  01-JAN-01  Unknown  2  035  40 (therefore, real age is either 39 or 40)  On any given row of data, one of the following was true: (a) age = age_ecp (b) age = age_ecp minus 1 where age was derived from clntbrdt (the women’s real age), and age_ecp was estimated using birth year in PharmaNet data.  For StudyID 000005, remove: (a) 000005 (the row with missing clntbrdt). (b) 000001 (because age = 17 and 000005 potentially has an identical real age of 17). (c) 000002 (because age = 18 and 000005 potentially has an identical real age of 18). (d) 000003 (because age = 18 and 000005 potentially has an identical real age of 18, regardless of the observation that 000003 has an age_ecp of 19). (e) 000006 (because age is also unknown, and age_ecp indicates a potential identical real age of 17 or 18). (f) 000007 (because age is also unknown, and age_ecp indicates a potential identical real age of 17).  183  (g) 000008 (because age is also unknown, and age_ecp indicates a potential identical real age of 18).  The only rows to be retained in this example were 000004, 000009, and 000010 because their: Age (if known) ≠ 17 or 18 Real age based on age_ecp (if clntbrdt was missing) ≠ 17 or 18 Therefore: For any given row with unknown age (due to missing clntbrdt) "Row A": (a) Remove Row A. Also, remove any accompanying row "Row B" (with same dispensing date, EC type, and pharmacy LHA): (b) If Row B has valid clntbrdt, and Row B's age = Row A's age_ecp minus 1. (c) If Row B has valid clntbrdt, and Row B's age = Row A's age_ecp. (d) Same logic as (c). (e) If Row B is missing clntbrdt, and Row B’s age_ecp = Row A’s age_ecp. (f) If Row B is missing clntbrdt, and Row B’s age_ecp = Row A’s age_ecp minus 1. (g) If Row B is missing clntbrdt, and Row B’s age_ecp = Row A’s age_ecp plus 1.  Summary of the method: (1) Remove the index row that is missing clntbrdt (therefore, age is unknown). (2) Remove accompanying row if age is known, and age = index row’s age_ecp or age_ecp minus 1. (3) Remove accompanying row if missing clntbrdt, and age_ecp = index row’s age_ecp +/- 1.  184  Appendix E Pregnancy-, prenatal care- and delivery-related administrative codes (excluding pregnancies with abortive outcomes) Pregnancy and delivery-related ICDs 640-648: Complications mainly related to pregnancy 650-659: Normal delivery and other indications for care in pregnancy labour and delivery 660-669: Complications occurring mainly in the course of labour and delivery 670-676: Complications of the puerperium ICD Definition 640 Haemorrhage in early pregnancy 641 Antepartum haemorrhage, abruptio placentae, and placenta praevia 642 Hypertension complicating pregnancy, childbirth and the puerperium 643 Excessive vomiting in pregnancy 644 Early or threatened labour 645 Prolonged pregnancy 646 Other complications of pregnancy, not elsewhere classified 647 Infective and parasitic conditions in the mother classifiable elsewhere Other current conditions in the mother classifiable elsewhere but complicating 648 pregnancy, childbirth and the puerperium 650 Delivery in a completely normal case 651 Multiple gestation 652 Malposition and malpresentation of fetus 653 Disproportion 654 Abnormality of organs and soft tissues of pelvis 655 Known or suspected fetal abnormality affecting management of mother 656 Other fetal and placental problems affecting management of mother  185  ICD Definition 657 Polyhydramnios 658 Other problems associated with amniotic cavity and membranes 659 Other indications for care or intervention related to labour and delivery 660 Obstructed labour 661 Abnormality of forces of labour 662 Long labour 663 Umbilical cord complications 664 Trauma to perineum and vulva during delivery 665 Other obstetrical trauma 666 Postpartum haemorrhage 667 Retained placenta or membranes, without haemorrhage Complications of the administration of anaesthetic or other sedation in labour 668 and delivery 669 Other complications of labour and delivery, not elsewhere classified 670 Major puerperal infection 671 Venous complications in pregnancy and the puerperium 672 Pyrexia of unknown origin during the puerperium 673 Obstetrical pulmonary embolism 674 Other and unspecified complications of the puerperium, not elsewhere classified 675 Infections of the breast and nipple associated with childbirth Other disorders of the breast associated with childbirth, and disorders of 676 lactation  186  Prenatal care Fee items Fee item Definition 14090 Prenatal visit, complete examination 14091 Prenatal visit, subsequent examination  Delivery-related Fee items Fee item Definition 4000 Complicated delivery - includes shoulder dystocia, premature delivery 4014 Complicated delivery - midcavity surgical delivery (operation only) 4015 Confinement terminating in a caesarean section 4016 Confinement terminating in an elective caesarean 4017 Midcavity rotation - surgical delivery (operation only) 4018 Breech vaginal birth (operation only) 4021 Obstetrical delivery, complicated 4050 Caesarean section - elective 4052 Caesarean section - emergency 4092 Multiple natural births 4093 Caesarean section, multiple births 4104 Post-natal care and delivery 4105 Caesarean section 4107 Supervision of labour and vaginal delivery 4109 Delivery, attendance 14104 Delivery and post-natal care 187  Fee item Definition 14108 Post-natal care after elective caesarean section 14109 Attendance at delivery and post-natal care 14199 Management of prolonged second stage of labour  Delivery-related CCP codes CCP code Definition 84.0 Low forceps delivery (without episiotomy) 84.1 Low forceps delivery with episiotomy 84.2 Mid forceps delivery 84.3 High forceps delivery 84.4 Forceps rotation of fetal head 84.5 Breech extraction 84.6 Forceps application to aftercoming head 84.7 Vacuum extraction 84.8 Other specified instrumental delivery 84.9 Unspecified instrumental delivery 85.0 Artificial rupture of membranes 85.1 Other surgical induction of labour 85.2 Internal version and extraction 85.3 Failed forceps 85.4 Operations on fetus to facilitate delivery 85.5 Medical induction of labour 188  CCP code Definition 85.6 Manually assisted delivery 85.7 Episiotomy 85.9 Other operations assisting delivery 86.0 Classical cesarean section 86.1 Cervical cesarean section 86.2 Extraperitoneal caesarean section 86.3 Removal of intraperitoneal embryo 86.4 Other removal of embryo 86.8 Caesarean section of other specified type 86.9 Caesarean section of unspecified type 87.3 Amniocentesis 87.4 Intrauterine transfusion 87.5 Other intrauterine operations on fetus and amnion 87.6 Removal of retained placenta 87.7 Repair of obstetric laceration of uterus 87.8 Repair of other obstetric lacerations 87.9 Other obstetric operations  189  Appendix F Abortion-related administrative codes (not necessarily induced abortion codes) Abortion-related ICD codes ICD Definition 630 Hydatidiform mole 631 Other abnormal product of conception 632 Missed abortion 633 Ectopic pregnancy 634 Spontaneous abortion 635 Legally induced abortion 636 Illegally induced abortion 637 Unspecified abortion 638 Failed attempted abortion 639 Complications following abortion and ectopic and molar pregnancies  Abortion-related Fee items Fee item Definition Abortion, therapeutic (vaginal), by whatever means: - less than 14 weeks 4111 gestation (operation only) Abortion, therapeutic (vaginal), by whatever mean: - 14 to 18 weeks (operation 4110 only) Abortion, therapeutic by D&E or amniocentesis – 18 weeks and over (operation 4027 only) 14545 Medical abortion  190  Abortion-related CCP codes CCP code Definition 87.0 Intra-amniotic injection for termination of pregnancy 87.1 Vacuum aspiration for termination of pregnancy 87.2 Other termination of pregnancy  191  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0073326/manifest

Comment

Related Items