UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Risk-benefit tradeoffs for NSAIDs for the management of rheumatoid arthritis (RA)- a discrete choice… Chen, Belinda Shin-Chieh 2010

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

Item Metadata

Download

Media
24-ubc_2011_spring_chen_belinda.pdf [ 3.02MB ]
Metadata
JSON: 24-1.0071558.json
JSON-LD: 24-1.0071558-ld.json
RDF/XML (Pretty): 24-1.0071558-rdf.xml
RDF/JSON: 24-1.0071558-rdf.json
Turtle: 24-1.0071558-turtle.txt
N-Triples: 24-1.0071558-rdf-ntriples.txt
Original Record: 24-1.0071558-source.json
Full Text
24-1.0071558-fulltext.txt
Citation
24-1.0071558.ris

Full Text

Risk-benefit tradeoffs for NSAIDs for the management of rheumatoid arthritis (RA)- A discrete choice experiment (DCE) by Belinda Shin-Chieh Chen  BSc., University of British Columbia, 2007  A THESIS SUBMITTED IN PARTIAL FUFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES (Pharmaceutical Sciences)  The University of British Columbia (Vancouver)  December 2010 © Belinda Shin-Chieh Chen, 2010  Abstract  Background Nonsteroidal anti-inflammatory drugs (NSAIDs) are drugs that are used to reduce pain and inflammation in those with rheumatoid arthritis (RA). There are two types of NSAID drugs available for treating those with RA, nonselective NSAIDs and coxibs. Rofecoxib, is a coxib that was taken off the market due to evidence of an increased risk of cardiovascular events from the use of this drug relative to placebo. However, there are those who benefited from this drug and were willing to take the risk. Objective The primary objective of this study is to use a DCE to quantify the risk preferences of those with RA, i.e., how much potential risk they are willing to accept in order to gain a specified potential benefit. It is hypothesized that given greater potential benefit, those with RA will be willing to accept greater risk of AEs, in particular myocardial infarction and stroke, than what they are currently exposed to in their NSAIDs. Methodology The DCE requires the creation of a questionnaire-based survey that was administered online to a sample of RA patients. Discrete choice experimentation allows the elicitation of patients’ preference for different treatment attributes through paired choice scenarios where respondents are required to choose between treatment options based on the risks of adverse events and benefits associated with each treatment. Conclusions For those with RA, the benefits outweigh the risks. In particular, given greater chance of benefit with improvement in pain and function, RA patients are willing to accept a greater risk of ulcers, dyspepsia, myocardial infarction, and stroke than what is present in current and past NSAIDs. Increased knowledge of the risk preferences of those with RA may aid decision-makers in better-informed drug approval and withdrawal decisions, thereby preventing patients from losing access to potentially beneficial drugs.  ii  Preface  The work presented in this thesis was conducted and disseminated by the Master’s candidate. The co-authors of the manuscript that comprise part of this thesis made contributions only as is commensurate with a thesis committee or as experts in the specific area as it pertains to the work. The co-authors provided direction and support. The co-authors will review the manuscript prior to submission for publication and offer critical evaluations, however, the candidate will be responsible for the writing and the final content of the manuscript. Versions of Chapters 2 (Chen B, Lynd LD, Marra CA, Lacaille D, Kopec JA. Methods for evaluating patients’ preferences) and 5 (Chen B, Lynd LD, Marra CA, Lacaille D, Kopec JA. Benefit-risk tradeoffs for NSAIDs for the management of rheumatoid arthritis: a discrete choice experiment) will be submitted for publication by Belinda Chen. Ethics for this study was obtained from the UBC Behavioural Research Ethics Board (BREB) # H08-00952 and VCHRI # V090278.  iii  Table of contents  Abstract .......................................................................................................................... ii Preface........................................................................................................................... iii Table of contents .......................................................................................................... iv List of tables ................................................................................................................. vi List of figures ............................................................................................................... vii List of nomenclature .................................................................................................. viii Acknowledgements ...................................................................................................... ix Chapter 1: Rheumatoid arthritis and NSAIDs ............................................................. 1 1.1 Introduction ............................................................................................................ 1 1.2 Management of RA ................................................................................................ 3 1.3 Research needs and justification ........................................................................... 5 1.4 Study objective, hypothesis, and study aims ......................................................... 7 1.5 Expected outcomes and implications of study ....................................................... 8 Chapter 2: Methods for evaluating patients’ preferences ....................................... 10 2.1 Introduction .......................................................................................................... 10 2. 2 Qualitative methods ............................................................................................ 10 2.3 Quantitative methods ........................................................................................... 13 2.4 Conclusion ........................................................................................................... 17 Chapter 3: General literature review of RA patients’ preferences .......................... 23 3.1 Introduction .......................................................................................................... 23 3.2 Methods ............................................................................................................... 24 3.3 Results ................................................................................................................. 24 3.4 Discussion and conclusion................................................................................... 28 Chapter 4: DCE overview and methodology ............................................................. 31 4.1 Introduction .......................................................................................................... 31 4.2 DCE ..................................................................................................................... 31 4.3 Theoretical background ....................................................................................... 32 4.4 DCE questionnaire design ................................................................................... 36 4.4.1 Attribute selection .......................................................................................... 36 4.4.2 Level and level-range selection..................................................................... 37 4.5 Experimental design ............................................................................................ 39 4.6 Inclusion of an opt-out option............................................................................... 40 4.7 Data collection ..................................................................................................... 42 4.7.1 Pilot testing .................................................................................................... 42 4.7.2 Methods of data collection............................................................................. 43 4.8 DCE analysis ....................................................................................................... 45 4.8.1 Coding of variables........................................................................................ 45 4.8.2 Choice models............................................................................................... 46 4.9 Interpretation of results ........................................................................................ 50 iv  4.10 Conclusion ......................................................................................................... 52 Chapter 5- Benefit-risk tradeoffs for NSAIDs for the management of rheumatoid arthritis: a discrete choice experiment...................................................................... 58 5.1 Introduction .......................................................................................................... 58 5.2 Methods ............................................................................................................... 60 5.2.1 Recruitment strategy ..................................................................................... 60 5.2.2 Discrete choice experiment ........................................................................... 61 5.3 Analysis................................................................................................................ 64 5.3.1 Descriptive and bivariate analysis ................................................................. 64 5.3.2 Choice models............................................................................................... 65 5.4 Results ................................................................................................................. 70 5.4.1 Sample characteristics .................................................................................. 70 5.4.2 Model 1.......................................................................................................... 70 5.4.3 Model 2.......................................................................................................... 71 5.4.4 Maximum acceptable risk .............................................................................. 73 5.4.5 Comparison of preferences for different NSAIDs .......................................... 74 5.5 Discussion............................................................................................................ 75 5.6 Conclusion ........................................................................................................... 77 Chapter 6: Summary and conclusions of thesis ..................................................... 92 6.1 Introduction .......................................................................................................... 92 6.2 Discussion of key findings from thesis ................................................................. 93 6.3 Conclusion ........................................................................................................... 97 References ................................................................................................................. 100 Appendix 1: The discrete choice experiment questionnaire................................. 108 Appendix 2: The complete list of definitions for treatment characteristics....... 141  v  List of tables Table 2.1 Example TT choice-set question................................................................ 19	
   Table 4.1 Example of one choice-set question .......................................................... 56	
   Table 4.2 Illustration comparing dummy vs. effect coding for 3-level attribute .......... 57	
   Table 5.1: Summary of attributes and levels in DCE questionnaire........................... 78	
   Table 5.2: Summary of demographic and clinical characteristics of respondents ..... 79	
   Table 5.3: Mean relative preference weights for MNL model .................................... 80	
   Table 5.4: Mean relative preference weights for MXL model..................................... 81	
   Table 5.5 Comparison of the utility gain given a benefit versus the disutility given greater risk ................................................................................................................. 82	
   Table 5.6: MAR for increased chance of benefit........................................................ 83	
   Table 5.7: MAR for greater improvement in function ................................................. 83	
   Table 5.8: Comparison of preference profiles of NSAIDs with similar benefits and different risks.............................................................................................................. 84	
    vi  List of figures Figure 2.1: Visual analog scale (VAS) for a chronic health state preferred to death . 20	
   Figure 2.2: Standard gamble for a chronic health state preferred to death ............... 21	
   Figure 2.3: Time trade-off (TTO) for a chronic health state preferred to death.......... 22 Figure 5.1: Flowchart of respondent recruitment results............................................ 85	
   Figure 5.2: Preference distribution for a risk of dyspepsia......................................... 86	
   Figure 5.3: Preference distribution for a risk of MI ..................................................... 87	
   Figure 5.4: Preference distribution for a risk of stroke ............................................... 88	
   Figure 5.5: Relative mean preference weights by number of tender joints................ 89	
   Figure 5.6: Relative mean preference weights by SF6D score.................................. 90	
   Figure 5.7: Relative mean preference weights by HAQ score ................................... 91	
    vii  List of nomenclature  o AE(s)=Adverse event(s) o ACE= Arthritis consumer experts o Attributes= Treatment characteristics o DCE= Discrete choice experiment o DMARD(s)= Disease modifying anti-rheumatic drugs(s) o GI = Gastrointestinal o HAQ= Health assessment questionnaire o Levels = Ranges of plausible estimates for an attribute o MAR = Maximum acceptable risk o MI=myocardial infarction o MNL= Multinomial logit model o MPAP = Mary pack arthritis program o MXL = Mixed logit model o NSAID(s)=Non-steroidal anti-inflammatory drug(s) o OA=Osteoarthritis o Preference weights=magnitude of importance that a respondent places on a level of an attribute o RA=Rheumatoid arthritis o RUT=Random utility theory o SG=Standard gamble technique o TT=Probabilistic threshold technique o TTO=Time trade-off technique o VAS=Visual analogue scale o Utility=Preference  viii  Acknowledgements  I would like to thank my supervisors, Dr. Larry Lynd and Dr. Carlo Marra for their unconditional support and guidance throughout my masters program. In addition I would also like to thank my committee members, Dr. Diane Lacaille and Dr. Jacek Kopec for always being willing to provide some helpful suggestions and insightful comments into my research project, making it the best it can be. Finally last but not least thank-you to all the faculty and staff from the Faculty of Pharmaceutical Sciences because without you the program would not have been the same.  ix  Chapter 1: Rheumatoid arthritis and NSAIDs  1.1 Introduction Rheumatoid Arthritis (RA) is a chronic autoimmune disease where the body’s immune system attacks its own healthy joints (1). For RA patients, the joints most commonly affected are those in the hands, feet, and knees (1). An estimated 1 in 100 Canadians will develop RA during their lifetime. Although RA can occur in both men and women, the incidence in women is approximately two-fold greater than in men (13). The effect of RA on those afflicted is enormous as progressive destruction of joints can lead to severe disability, fatigue, and chronic pain (2-4). As a result of joint damage and RArelated symptoms, the ability of those with RA to perform in normal daily activities may be limited (6).  Those with RA do not die specifically from this disease. Instead, they tend to die from increased susceptibility to co-morbidities such as infection, cancer, respiratory, cerebrovascular (CVA) and cardiovascular (CVD) diseases. In particular, CVD is found to be the major contributing factor of death among those with RA (17-19). Given an increased occurrence of co-morbidities, those with RA also tend to live a shorter life compared to the rest of the general population. In large Canadian and American mortality studies, the RA Standard Mortality Ratio (SMR) is estimated to range from 1.28-2.26 (16,19). A SMR gives the ratio of the observed number of deaths in the diseased population, e.g., RA, over the expected number of deaths in a demographically similar population with no disease (16,19). This means that there is an  1  estimated 1.28 to 2.26 times more cases of death in people with RA compared to a similar population with no RA.  Out of all musculoskeletal disorders (MSDs), RA is considered to pose the greatest economic burden on society (20). Direct costs, i.e., costs associated with medical expenditure for treatments or inpatient stays, is a substantial area of economic burden for those with RA. The estimated average direct cost of RA is $US9,519 per person per year; (4,16,20-23). In addition to direct costs, there are indirect costs associated with having RA.  The estimated average indirect cost of RA varies from $US1,082 to $US33,000 per person per year (4,16,20-23). The major contributing factor to indirect costs is productivity losses, whereas the large variation cost estimates are attributed to differences in the estimation of other contributors such as the economic burden on caregivers, the time spent on taking care of those with RA, etc (4,16,20-23). Productivity losses in people with RA occur because of absenteeism or presenteeism (4,16,20-23). Absenteeism is when people with RA do not go to work because of painful joints and limited function. In contrast, for those with RA that still work, despite suffering the same degree of pain and limited function as those who do not work, there may be reduced work efficiency. This is presenteesim and can also result in productivity losses. Despite the great economic burden of RA, treatments exist that can improve patient outcomes and quality of life, thereby reducing the burden of disease for those afflicted.  Current treatment strategies for RA management include taking a combination of different types of drugs, each with their own varying levels of efficacy and adverse 2  events (AEs). Thus for those with RA, decisions about which drug to take can often be based on their personal preferences for the benefits and AEs of the drug(s). However, drug approval or withdrawal decisions are typically based on the AEs of treatments, without consideration of patients’ preferences for the benefits. A disregard of patients’ preferences can be problematic for patients who are willing to accept a greater risk of specific AEs given the benefit they may receive from a treatment. This was the case for RA patients when the non-steroidal anti-inflammatory drug (NSAID), rofecoxib (Vioxx®), was voluntarily withdrawn due to an increased risk of cardiovascular (CV) events.  Discrete choice experimentation (DCE) has been used to empirically and quantitatively demonstrate whether the potential benefits of NSAIDs outweigh their potential risks from the perspective of those with RA.  1.2 Management of RA There are many treatments available for RA management. Some treatments serve to alleviate RA symptoms, e.g., analgesics, NSAIDs, or corticosteroids, while others help to slow disease progression, e.g., disease modifying anti-rheumatic drugs (DMARDs), or biologic DMARDs (24). Given the types of treatments available, the most common method of RA management is to take a combination of NSAIDs, analgesics, corticosteroids and DMARDs (24). To give time for drugs such as DMARDs to take effect and slow long-term disease progression, NSAIDs can be taken in the beginning to reduce pain and inflammation in the joints (24). Non-steroidal anti-inflammatory drugs decrease joint pain and inflammation through their inhibition of the cyclo-oxygenase (COX) enzyme (25,26). The COX 3  enzyme promotes the production of prostaglandins, which are important for the development of tissues in many areas of the body such as the stomach and brain (25,26). There are two isoenzymes, or forms of the COX enzyme that are important in the context of NSAIDs: COX-1 and COX-2 (26). The COX- 1 form is constitutively produced in the gastric tissues, while COX-2 is produced in specialized tissues such as in the brain, ovaries, and kidneys (26). In addition, COX-2 expression can be induced in response to stressors such as injury or inflammation (25,26). Therefore by inhibiting the inflammation promoting COX enzymes, NSAIDs provide pain relief and improved functional ability in those with RA.  There are two major types of NSAIDS taken by those with RA: non-selective NSAIDS (NSNSAIDs) and selective NSAIDs, called COX-2 inhibitors or coxibs. Nonselective NSAIDS, e.g. aspirin or diclofenac, inhibit both COX-1 and COX-2 isoforms to the same degree. However, inhibition of the COX-1 enzyme may result in a substantial decrease in the amount of gastrointestinal (GI) prostaglandins that are produced to protect the stomach tissues. Therefore, those taking NSNSAIDs are at an increased risk of adverse GI events such as ulcers and dyspepsia (25,26,30,31). Because of the additional burden of adverse GI outcomes, e.g., stomach bleeds or death, as well as the additional treatments that need to be taken to treat them, patients may choose to take a coxib instead (32).  Coxib drugs are selective COX-2 inhibiting NSAIDs, i.e., they inhibit the COX-2 isoform at a greater rate than the COX-1 isoform (27,33). Thus, those taking coxibs may be at a reduced risk of GI events compared to those taking NSNSAIDs (26,34). In particular, one study found that the relative risk (RR) of adverse GI events, such as 4  perforation, obstruction and stomach bleeds, in those taking coxibs relative to NSNSAIDs was 0.5 (25). This means that those taking a coxib are 50% less likely to experience an adverse GI event compared to those taking NSNSAIDs (25,35,36). Examples of coxib drugs that have been available in Canada are rofecoxib, celecoxib, and valdecoxib (26).  However, the reduced risk of adverse GI events from a coxib relative to a NSNSAID may be dose dependent. Therefore with greater exposure to coxibs, there is a potential for more inhibition of the COX-1 enzymes and an increased risk of GI events. In particular, Mamdani and colleagues found that the risk of upper GI hemorrhage with celecoxib was the same as taking no NSAID at all (37). Consequently if the benefit from a coxib and NSNSAID is the same, RA patients may need to consider their risk preferences for the other AEs that are associated with coxibs and NSNSAIDs before deciding whether or not they would prefer taking a coxib to a NSNSAID.  1.3 Research needs and justification The RA risk tolerance for NSAID-related AEs was called into question when Merck voluntarily withdrew rofecoxib. This withdrawal was due to evidence of an increased CV risk from use of rofecoxib compared to placebo. In particular, results from the APPROVE trial estimated that those taking rofecoxib had approximately twice the risk of experiencing CV AEs such as MI, sudden cardiac death, unstable angina pectoris, ischemic insult, transitory ischemic attacks, or peripheral vascular occlusion, relative to placebo (33,38). However, this proved to be problematic for patients who had  5  been on rofecoxib that were willing to accept the increased CV risk in order to receive the benefit that they were receiving from the drug (7-9).  This led to expert panels from both the Food and Drug (FDA) and Health Canada to convene and weigh the evidence on rofecoxib. After much discussion and analysis of the risks and benefits of this drug, each panel concluded that from the perspective of the patient, the benefits of rofecoxib outweigh its risks (10-12). Therefore it was felt that if no alternative beneficial treatment were available, patients who were receiving a desired benefit should be given the option to take rofecoxib. The panels’ conclusions were later supported by an incremental net-benefit analysis from the societal perspective that also found that rofecoxib’s benefits outweigh its risks relative to naproxen; the evidence presented during the FDA and Health Canada panel discussions was used in this analysis (40).  Consequently when only a drug’s safety concerns are considered during drug approval or withdrawal decision, there is a potential for false positive or negative decisions. When false positive decisions are made, a drug is approved that patients may not take or adhere to because of the risks of AEs (6). This is a problem in that money and time is wasted in packaging and marketing a drug that patients wouldn’t want. In addition, patients may be unnecessarily placed at a greater risk of AEs than they are willing to accept. Contrastly, it is also a problem when treatments that patients would want to take are not approved or withdrawn due to safety concerns (6); this is a false negative decision. False negative decision should be avoided as patients will lose access to treatments that work for them. To potentially avoid false positive and  6  negative decisions, methods of benefit-risk analysis from the perspective of the patient need to be conducted.  The US Institute of Medicine (IOM) and the Committee for Medicinal Products for Human Use (CHMP), of the European Medicines Agency (EMA), recommend that proper methods are needed to effectively measure, quantify, and apply benefit-risk analysis during the pre- and post-approval processes of drug regulatory decisions (44,45). One such method that is being evaluated for use in these processes is the stated choice technique. The discrete choice experiment is a stated preference technique to elicit patients’ preferences for healthcare attributes through the administration of a questionnaire-based survey.  1.4 Study objective, hypothesis, and study aims The primary objective of this study is to use a DCE to quantify the risk preferences of those with RA, i.e., how much risk of ulcers, dyspepsia, myocardial infarction (MI) and stroke they are willing to accept in order to gain a greater chance of experiencing pain relief and functional improvement in their most valued daily activities.  RA patients generally have to endure many types of treatments before finding one that gives them the benefit they want. Therefore it is hypothesized that given a greater potential benefit, those with RA will be willing to accept a greater risk of AEs, in particular MI and stroke, than what they are currently exposed to in their NSAIDs.  Subsequent study aims are to: 7  1. empirically and quantitatively evaluate if from the RA patient perspective, the potential benefits of coxibs outweigh the potential risks compared to NSNSAIDs; and  2. to compare RA patients’ preference for coxibs to their preference for NSNSAIDs  1.5 Expected outcomes and implications of study In patients with chronic debilitating diseases, the mere existence of a potentially effective treatment may be enough for patients to accept the potential risk of serious treatment-related AEs. This is because there may be no other alternatives available that would give them the benefit they want. In particular for those with life-threatening conditions with no cure, a treatment that may work would be particularly valued. For example, the expedited process of the treatment azidothymidine (AZT), from drug development to approval for marketing, was partly due to the anxious demand of those infected with Human Immunodeficiency Virus (HIV) as there was no other available treatment at the time. Thus given no other beneficial treatments, patients with chronic illnesses such as RA, may be willing to accept a greater risk of AEs given a greater potential benefit.  The DCE is one method being evaluated by the IOM and the CHMP for incorporating into pre and post drug regulatory decisions. Thus, this study will demonstrate the usefulness of the DCE as method of benefit-risk analysis from the patients’ perspective. At the conclusion of this project, it is anticipated that the results would demonstrate that from the perspective of those with RA, the benefits of NSAIDs 8  outweigh their risks, i.e., on average, most people with RA would be willing to tolerate greater risk of AEs than what they are currently exposed to in their NSAIDs. Furthermore, although the results may not be able to change previous decisions on drugs previously withdrawn from the market, such as rofecoxib, it may substantially impact decisions made on the approval of newer drugs. In addition, it is hoped that by putting RA patients’ risk preferences into perspective, regulators and clinical decisionmakers will be able to make better-informed decisions.  9  Chapter 2: Methods for evaluating patients’ preferences  2.1 Introduction In the context of healthcare, the term preference describe the features or attributes of healthcare options that are most valued by patients (46). It is important to understand patients’ preferences for healthcare attributes as variation in these attributes may influence patients’ healthcare choices. For example, if rheumatoid arthritis (RA) patients are strongly averse to treatments with a risk of dyspepsia, then they may choose to take less beneficial treatments to avoid experiencing this adverse event (AE).  There are many preference elicitation methods available. Qualitative methods, such as questionnaires or interviews, collect descriptive information about patients’ preferences. However in comparison to quantitative methods such as standard gamble (SG) or probabilistic threshold technique (TT), qualitative methods do not allow the estimation of the strength of patients’ preferences for healthcare attributes. In this chapter, the currently available qualitative and quantitative preference evaluation methods are presented. These methods were critically appraised within the context of their usefulness to elicit patients’ preferences for treatments.  2. 2 Qualitative methods Qualitative methods for preference elicitation provide descriptive information about patients’ preferences such as, preferred modes of treatment administration or healthcare features that are important to patients. There are three types of qualitative 10  methods generally used to gather descriptive preference information: (1) face to face or telephone interviews, (2) focus groups, and (3) questionnaires and surveys (47). Face to face or telephone interviews consists of a setting with one patient and interviewer (47). In comparison, focus groups are interviews in a group setting where patients can discuss the answer to the question amongst each other (47).  For either one-one interviews or focus groups, the interviewer only speaks to ask a question and then sits quietly to record the answers given. Respondents’ answers are transcribed so that they may be referred to when the interviewer reviews the answers to extract data that is relevant to the study objective(s). Once the interviewer has gone over the data collected, the next goal is to analyze and categorize the information in accordance to their ability to address the goals of the study. For example, if the objective of the interview was to identify the benefits of treatment that are most important to rheumatoid arthritis (RA) patients, the interviewer will review the transcripts for the benefits that the majority of patients mention that they want.  One advantage of using the interview format is that interviewers are present to clarify any questions participants may have. This may be particularly important for older study participants or those where English is not the first language. Thus by having an interviewer present, missing data is potentially avoided. In addition, the presence of interviewers permits the opportunity to gain clarification about respondents’ answers that may be confusing or interesting. For example, imagine that RA patients mention that the benefits they most value from treatments are pain reduction and functional improvement. However, when those with RA described the pain relief they wanted, they described it in terms of their ability to function in daily activities. Therefore with 11  interviewers present, it could be clarified if patients view these two treatment attributes to be the same thing.  Although interviewers are carefully trained to interview all participants in a similar manner, there is still the possibility of human error, i.e., interviewer bias. This bias occurs when the interviewer incorrectly records the data heard or unintentionally (or intentionally) influences the responses of the respondents (47). Another limitation of the interview method is that it can be costly in terms of resources, i.e., manpower, time, and money. With regards to time, the process of arranging a time and day for interviews can be particularly difficult for patients with limited functional ability, such as those with RA.  In comparison, questionnaires/surveys provide the advantage that patients don’t feel pressured to answer the question right away. This may allow participants to take time to think about their answer, potentially increasing the quality of the data. Questionnaires or surveys can be mailed, given to respondents to take home and mail back, or presented to them online. Thus depending on the method of sample recruitment, this may allow a better sampling of patients in different geographical locations compared to the interview format. Furthermore, if resources are scarce for the study, the use of questionnaires/survey can reduce study costs and the time needed for data collection.  Although the use of questionnaires/surveys eliminates interviewer bias, it increases the chance of non-response. Non-response is when patients are mailed or presented with the questionnaire but they do not fill it out because they forget, or they don’t want to answer the questions, or they don’t understand the questions. However, a 12  potential solution for minimizing non-response is to follow-up with participants to make sure they are able to complete the questions they are given.  In summary, qualitative preference elicitation methods are advantageous in that they are not cognitively tasking for patients. However, these methods do not allow the quantitative evaluation of patients’ preferences. For example it can’t be determined whether those with rheumatoid arthritis (RA) are more averse to a risk of dyspepsia compared to a risk of stroke.  2.3 Quantitative methods There are two general categories of methods for quantitative preference elicitation: scaling and probably trade-off techniques (48-50-51). Examples of scaling techniques are standard gamble (SG), time trade-off (TTO), or rating scales (49,52-55). The most common rating scale is the visual analogue scale (VAS). Variations of probability trade-off techniques are probabilistic threshold technique (TT), and stated preference (SP) techniques.  The VAS is the simplest example of a rating scale that can be used to estimate patients’ preferences for health outcomes. The VAS design is relatively easy to comprehend, thereby minimizing cognitive burden for participants (Figure 2.1) (49,5255). In studies using a VAS, respondents are asked to rate or assign their value for a specific health state on a line, the scale, which can be anchored between two health outcomes, e.g., death and optimal health. The ranges for the scale can be from 0 to 10 or 100, where 0 represents death and 10 or 100 is optimal or perfect health. 13  One disadvantage to rating scales is that patients’ risk preferences may be underestimated for less serious AEs and overestimated for more serious ones (49,52,53). This is because patients rate their risk tolerance of an AE separately, i.e., patients trade-off between one benefit and risk attribute. Another disadvantage of rating scales is that they are not based on random utility theory. Therefore, patients’ utilities can’t be elicited.  Utilities describe the value of importance of an attribute to a patient and are useful in that they allow for the comparison of patients’ preferences across a variety of disease states (46). For example, preferences for pain reduction between those with cancer and arthritis can be compared. Nonetheless, because of the simplicity of the VAS, the use of this technique has been thought to be an adequate measure of patients’ quality of life (49,52).  Two scaling methods that allow the evaluation of patients’ utilities are standard gamble (SG) and time trade-off (TTO) (49,52-55). In studies using SG, participants are required to choose between two choices, Choice A (the sure outcome) or Choice B (the gamble) (49,52-55). If patients choose B, they are faced with a probability p of a specified health benefit, e.g., perfect health, and a probability 1-p of a specified adverse outcome, e.g., death (34,56-59) (Figure 2.2). In contrast if A is chosen, they will remain in described health state i (34,56-59). By varying the probabilities of p, the probability at which respondents are indifferent between taking the gamble (Choice A), or remaining in the current health state (Choice B), can be estimated. This indifference probability, p, represents respondents’ utility for health state i (34,56-59). Consequently, when a respondent’s utility for the health state is low, the indifference probability will be low, and 14  vice versa, if utility for the health state is high, the indifference probability will be high (49,52-55).  Compared to rating scales, patients may find SG harder to understand (49,5255). In addition, there is evidence that there may be a ceiling effect, i.e., most health states may have similar utilities. This is because most people are risk averse and may only accept the gamble at smaller probabilities. (49,52-55). Therefore, one alternative scaling method that is suggested to be more useful in eliciting patients’ utilities for similar health states is TTO (49,52-55).  In TTO the respondent is also shown two choices, A and B. With Choice B, respondents would remain in a described health state i with a life expectancy of t years (49,52-55). In comparison, if respondents choose A they would be at optimal health but have a shorter life expectancy of x years, i.e., x<t (49,52-55). Therefore, respondents must indicate how many years of their remaining life expectancy t they would be willing to trade off for a state of optimal health (Figure 2.3) (49,52-55). The value of t at which a respondent is indifferent between Choice A and B is generally divided by the number of years of remaining life expectancy, i.e., x/t, to give a respondent’s utility for health state i (49,52-55).  In real treatment decisions, patients may be faced with trading off between many different choice options, each varying in their risk-benefit profile. Thus, although SG and TTO are valid ways of estimating patients’ utilities for healthcare attributes, they do not adequately model healthcare scenarios where patients are faced with the task of trading off between multiple attributes at a time. Methods that allow for the variation in patients’ 15  preferences as a result of the presence of multiple healthcare attributes are probability trade-off techniques.  Probabilistic threshold technique (TT) is one example of a probability trade-off technique. This method allows the estimation of patients’ maximum acceptable risk (MAR) of an AE given a specified benefit (51). In TT, respondents are first presented with a choice-set question with two hypothetical treatment options, the target and reference option, with the same treatment attributes (51). The risk of side effects are the same in each option, however, the target option has greater benefit than the reference option (51). Therefore in the first choice-set, patients should prefer the target as it has the same risks as the reference but has greater benefit (51). To elicit patients’ MAR for an AE, e.g., myocardial infarction (MI), the risk of the AE in the target option is incrementally increased in subsequent choice-sets until patients indicate that would choose to take the reference option instead. The magnitude of risk at which patients switch their preference of the target to the reference is a patient’s MAR for the AE (Table 2.1) (51).  Similar to scaling techniques, the preferences elicited with TT do not incorporate for the influence that other treatment attributes may have on patients’ MAR as patients are only trading off between two attributes at a time, i.e., a risk of AE for a benefit. Therefore, MARs may be overestimated for more serious AEs and underestimated for less serious AEs. For example, when given a specified benefit, patients may be willing to accept up to a 7% risk of ulcer. However if the treatment has other risks of AEs, their MAR for a risk of ulcer may change. Thus to properly model treatment scenarios where  16  patients may need to simultaneous trade-off between more than two attributes, stated preference techniques should be used.  Choice experiments, such as conjoint analysis (CA), are stated preference techniques used to evaluate patients’ preferences (3). It is a well-established method that has been used for market research in the area of transport and environmental economics (3). More recently, CA has been applied in the area of healthcare to elicit patients’ preferences for healthcare attributes (67).  Conjoint analysis elicits preferences by requiring respondents to rank, rate, or make pair-wise choices between services/goods (60). A form of CA that requires respondents to make pair-wise choices between two hypothetical healthcare options is discrete choice experimentation (DCE). Unlike earlier CA ranking methods, DCEs allow the determination of how patients’ healthcare decisions vary as a function of changes in the magnitudes of their healthcare attributes (60). This method is useful as it attempts to mimic real healthcare scenarios that patients may be faced with. Discrete choice experimentation has been successfully used to elicit the preferences of patients with a wide range of illnesses including: cancer, irritable bowel syndrome, diabetes, and asthma (42,71-73).  2.4 Conclusion There are many preference elicitation methods for healthcare services. Qualitative methods have the advantage of being simple and straightforward, i.e., a question is asked, and the answer is given. However, if interviewers are present there is 17  potential for interviewer bias, whereas if questionnaire/surveys are used, there is a greater chance of missing data due to non-response. In addition unlike quantitative methods, qualitative preference evaluation methods do not allow the estimation of the strength of patients’ preferences for specific healthcare attributes.  Scaling techniques are preference elicitation methods that offer quantitative information about patients’ preferences. However, the major limitation with these methods is that they assume that patients’ preferences are not affected by the presence of multiple healthcare attributes. This may not be realistic as some healthcare decisions may require patients to trade-off between different healthcare options that vary in their risk-benefit profile. One preference elicitation method that overcomes this limitation is DCE. Discrete choice experimentation evaluates patients’ preferences by requiring patients to choose between different healthcare options with varying benefits and risks of AEs. Because DCEs are administered through a questionnaire-based survey, this method may also be less difficult to understand compared to SG and TTO. An overview of DCE methodology will be provided in Chapter 4.  18  2.5 Tables  Pain  Option A (Reference)  Option B (Target)  5  3  2%  2% (Incrementally  Pain experienced while walking after taking pills daily on a 0-10 scale is… Stomach bleed  raised to identify MAR) Heart attack  1%  1%  Table 2.1 Example TT choice-set question  19  2.6 Figures  Death  Optimal Health Health State i  Figure 2.1: Visual analog scale (VAS) for a chronic health state preferred to death  20  Health State i  Choice A p  Choice B 1-p 1p  Specified Health Benefit E.g. Perfect Health Specified Adverse Outcome E.g. Death  Figure 2.2: Standard gamble for a chronic health state preferred to death  21  Optimal Health  Health State i  0 Death 0  x  t  Time  Figure 2.3: Time trade-off (TTO) for a chronic health state preferred to death  22  Chapter 3: General literature review of RA patients’ preferences  3.1 Introduction At present, pre and post drug approval or withdrawal decisions are primarily based on the risks of adverse events (AEs) associated with the treatment. However, for those with a chronic debilitating disease such as rheumatoid arthritis (RA), patients’ risk tolerance of AEs may increase given greater benefit. Thus, without consideration of patients’ preferences for treatment attributes, those with decision-making powers may withhold drugs with risks that patients are willing to accept given the benefit they may receive. This was the case when the drug rofecoxib (Vioxx®) was voluntarily withdrawn due to an increased risk of cardiovascular (CV) events in those taking the drug compared to placebo. However, when the evidence was actually reviewed from the patients’ perspective, expert panels from the Food and Drug Administration (FDA) and Health Canada concluded that rofecoxib’s potential benefits outweigh its potential risks (1-3). Thus, methods of benefit-risk analysis from the patients’ perspective may be useful in helping decision-makers make better- informed drug approval and withdrawal decisions.  There are many qualitative, e.g. questionnaires, and quantitative methods, e.g., standard gamble, to elicit patients’ preferences for healthcare attributes. A general overview of past studies conducted to evaluate RA patients’ preferences is presented. The purpose of this chapter is to highlight potential gaps in knowledge on RA patients’ preferences for their healthcare services by critically appraising the conclusions drawn in these studies from a methodological perspective. 23  3.2 Methods An initial literature search of MEDLINE OvidSP (1950 to date) and EMBASE OvidSP (1980 to present) using the following subject topic headings, treatment preferences, patient preferences, and rheumatoid arthritis, was undertaken to identify other potentially important synonyms of these subject headings. Alternative key search terms found include: arthritis, risk attitudes, patient attitudes, rheumatic disease, and drugs. After identification of the key search terms, MEDLINE and EMBASE were used by to find studies that used qualitative and quantitative methods to elicit RA patients’ preferences for healthcare by combining keyword and MeSH subject heading searches. In addition, any studies included in this review were used as topic searches in Google scholar to decrease the chance of missing any studies that pertained to RA patients’ preferences. Finally, the reference lists of all studies meeting the criteria for inclusion were also searched to make sure no important RA patient preference studies were missed. The criteria for inclusion of studies in this review were that all studies must present information on RA patients’ preferences for the area of healthcare. The reviewer then critically appraised the studies for general and methodological quality.  3.3 Results In total, fourteen studies were found that evaluated the healthcare preferences of those with RA. Nine of these studies used qualitative techniques, i.e., focus groups, interviews, and questionnaires (4-12) to evaluate RA patients’ preferences. In comparison, five studies used quantitative techniques, i.e., the standard gamble (13-15) or rating scales (16,17) to evaluate RA patients’ risk tolerance given a specified benefit.  24  The interview method was used to elicit the amount of information and participation in their treatment decisions that those with RA wanted when administered their healthcare (5-7). Preference studies found that those with RA wanted to be fully informed about their treatments, particularly in the context of the potential risks of side effects (5-7). In one study, women with RA were suggested to have a greater preference for full disclosure compared to men (5).  Because these studies used the interview format, there is the potential for interviewer bias. For example, one of the questions was if patients feel that even if the news is bad, they should be well informed (18). This question may imply that patients are not well informed, which may influence the responses that were obtained. Nonetheless, methods such as questionnaires, which do not have interviewers present, have also found that RA patients’ prefer greater disclosure of their treatment information (7). On explanation for why RA patients may prefer for full disclosure is to feel more in control over their disease management. Research suggests that feelings of self-efficacy may be important determinants of good physical and psychological health status in those with RA (7). Thus, RA patients who feel more in control of their health outcomes may be healthier than those who do not feel self-efficacious.  The need for those with RA to be able to control and manage their own illness could potentially influence their choice of treatment. In a study that administered a questionnaire to those with RA to assess their willingness to switch to newer treatments, it was found that RA patients’ fear of a loss of control over their symptoms was a major factor behind their resistance to changing therapies (4). Thus, if RA patients find a treatment that gives them a benefit, this may stop them from wanting to 25  switch to other potentially beneficial treatments. These conclusions can be supported by real-life case examples such as patients’ willingness to accept the increased CV risks of rofecoxib because they had found a drug that worked for them and did not want to lose control of their symptoms.  Another major determinant behind RA patients’ potential resistance to changing therapies may be their fear of side effects (4,9) These conclusions are supported by Frankel and colleagues who found that RA patients were unwilling to risk AEs commonly associated with RA treatments; in this study a rating scale was used to assess RA patients willingness to accept a risk of a specific AE of non-steroidal antiinflammatory drugs (NSAIDs), prednisone, or disease-modifying anti-rheumatic drugs DMARDs), given a specified benefit (16). However, the results of this study are limited by their use of a rating scale as it may overestimate the importance of more serious AEs and underestimate the importance of less serious AEs (19). In addition, the benefits in this study are traded off for risk attributes of different RA treatments, i.e., NSAIDs, DMARDs, and prednisone. Thus, as NSAIDs and DMARDs have different side effects and benefits, patients were not trading off between attributes of treatments that they would actually be exposed to in real life. Furthermore, as it should be expected that patients would be averse to side effects, what is more important is if patients risk tolerance will increase given greater benefit.  In a questionnaire administered to those with RA, it was found that RA patients were willing to accept the risks of NSAID-related gastrointestinal events (GI) given the benefits they were receiving from NSAIDs (8). These conclusions are supported by Pullar and Ho where given greater benefit, those with RA were willing to accept a 26  greater risk of AEs (15,17); rating scales were used to assess RA patients’ willingness to risk rash, headache, peptic ulcer, death or serious given a specified potential benefit (15,17). The potential benefit in these studies included a chance of a cure, e.g., 90% chance of cure, or improvement in symptoms, e.g., 30% improvement in symptoms (15,17). However when rating scales are used, participants are only trading off one attribute level for another, e.g., a risk of MI for a chance of cure. Thus the preferences elicited may not accurately reflect patients’ real clinical decisions when they have to simultaneously trade off between multiple treatment risks and benefits. Another potential limitation of the preference studies is if the benefits for trade-off are not meaningful to the participants. For example, a benefit included was a chance in improvement of RA symptoms (17). This could be ambiguous as the meaning of symptom improvement could vary among patients, e.g., some patients may believe it to be pain relief whereas others stiffness relief. Therefore, because the type of symptom relief could be interpreted differently, this may affect RA patients’ willingness to accept greater risk if the benefit is not one that is important to them. Consequently, prior to eliciting patients’ preferences for attributes of treatment, it is important to make sure that the attributes included in preference studies are meaningful enough to patients to influence their treatment decisions.  Focus groups, questionnaires, or interviews have been used to ask RA patients what benefits of their treatments are important to them (15-17). All three studies identified that those with RA most valued pain relief and improved function from their treatments (10). These attributes were identified to be important because they would allow for increased ability to participate in normal daily activities (10,11). Although, the use of interviews or questionnaires helps to identify important attributes of treatment for 27  those with RA, these methods do not provide insight into how patients’ preferences for the benefits would change given the potential for side effects.  To determine the magnitude of importance of a specific benefit of treatment given a risk of an AE, e.g., MI, patients’ maximum acceptable risk (MAR) can be determined; MAR is the maximum acceptable probability of experiencing a treatmentrelated adverse outcome, e.g., a risk of ulcer, given a specified potential benefit (10,11). The SG was used to evaluate RA patients’ maximum acceptable risk (MAR) for death (20-24). In these studies, those with RA were asked whether they would stay at their current health status or take a treatment with a potential for perfect health as well as a risk of death (13-15). It was found that given a treatment with a potential cure, RA patients’ MAR for death was approximately 27% (20-24). Although the SG method is a valid method to evaluate patients’ utility for a specific health state, like rating scales it only requires patients to trade-off between two attributes at a time. Thus, it may not adequately model real clinical scenarios faced by those with RA.  3.4 Discussion and conclusion RA patients’ preference studies have identified those with RA to be averse to risks of AEs. Therefore, given, their averseness to the risks of RA treatments, patients would prefer full disclosure of information about their healthcare options, e.g., detailed information on the risks and benefits (13,14). However, it is to be expected that all patients would be averse to risks of AEs. Therefore, given greater potential benefit from their treatments, those with RA may be willing to accept greater risks of AEs (5-7).  28  In summary, although the previous studies on RA patients’ preferences are useful, they should be considered with a couple limitations in mind. Firstly, the attributes included for trade-off were not appraised for their meaningfulness to patients prior to preference elicitation. If attributes and levels for trade-off are not meaningful to patients, the preferences elicited may be incorrect as patients are not trading off between attributes that are important in influencing their treatment decisions.  Qualitative preference elicitation methods such as focus groups and questionnaires can be used to determine the healthcare attributes that are important to those with RA, but unlike quantitative methods such as standard gamble, they are unable to evaluate patients’ utilities. Patients’ utilities are important to indicate which treatment attributes influence patients’ treatment decisions more than others. However, quantitative preference evaluation methods such as TTO and SG do not incorporate for variation in patients’ preferences given multiple treatment attributes for trade-off. In addition, these methods may be more cognitively challenging for patients compared to qualitative preference elicitation methods such as a survey.  The discrete choice experiment (DCE) is a method to elicit patients’ preferences that mimics real treatment decision-making scenarios by requiring patients to simultaneously trade off between multiple attributes of different healthcare options. The DCE is administered to patients through the form of a questionnaire-based survey. Therefore, an advantage of this method compared to quantitative methods such as SG is that it may be less difficult for patients to understand how to complete. Furthermore, an important and recommended step to constructing DCE questionnaires is to hold focus groups or interviews to try to include all key attributes and levels that influence 29  patients’ treatment decisions. In the next chapter, an overview of the DCE methodology is presented.  30  Chapter 4: DCE overview and methodology  4.1 Introduction In the past, methods to evaluate patients’ preferences for health outcomes have primarily focused on the use of techniques such as standard gamble (SG) or time-trade off (TTO). However, these methods are limited in that they do not reflect real healthcare decision-making scenarios where patients may be required to simultaneously trade-off between different healthcare options with multiple attributes. Therefore, attention has shifted to other methods such as the discrete choice experiment (DCE).  Discrete choice experiments provide insight into how patients’ preferences may change in response to changes in the variables or attributes that affect their preference choices (46,65,69,93,94). These attributes are features of a healthcare service or good, e.g., method of treatment administration, frequency of dosage, etc. This method has been used to evaluate the healthcare preferences of a variety of patient populations including those with asthma, diabetes, irritable bowel syndrome, and cancer (56-59,6061). The purpose of this chapter is to provide an overview of DCE theory and methodology within the context of using this method to elicit patients’ preferences for treatments.  4.2 DCE Discrete choice experimentation is a stated preference technique used to elicit patients’ preferences for healthcare attributes through the administration of a 31  questionnaire-based survey (42,95-97). As a form of choice experiment, the DCE technique evolved from market research where researchers used conjoint analysis (CA), a form of choice experiment, to identify the factors that influence the individuals’ demand for goods and services (95-97). However unlike traditional CA, the DCE method requires patients to choose between at least two hypothetical choice options that vary in the magnitude of the attributes.  In a DCE to elicit patients’ preferences for treatment, respondents are presented with a series of choice-set questions, each with a pair of hypothetical treatment options. Each treatment option has the same treatment attributes, but vary in the levels of these attributes (42,95-98). Therefore, based on the attribute levels of each treatment option, respondents are asked to indicate which of the two treatment options they prefer (Table 4.1) (42,95-98). The underlying assumption is that respondents’ choices in the DCE will reflect their preferences for different attribute levels (97,99).  4.3 Theoretical background Adapted from Lancaster’s approach to consumer theory and random utility theory, DCEs have two main assumptions. Firstly, it is assumed that each choice option can be characterized by its relevant characteristics or more specifically the attributes of the choice options (65). In the context of treatment options the attributes would be the treatment benefits and risks of adverse events (AEs). In a DCE, each attribute can be defined by two or more levels, which are a range of plausible estimates of values for an attribute (65,97). Thus, in a DCE a respondent trades off between the attribute levels of different treatments before deciding which option they prefer. 32  The second assumption of the DCE is that each individual has their own preference, i.e., utility, for each attribute level. The utility represents the relative value that an individual places on an attribute level (42,65,96,97). Thus, the greater the utility, the greater the importance of the attribute level to the individual. True respondent utility cannot be observed, it can only be estimated.  An individual’s true utility Ui can be broken down into two components: (1) the observable component Vi, and (2) the unobservable component unobservable component  i  i(42,65,96,97).  The  is included as a random error term as it incorporates for  factors that influence an individuals’ preference that cannot be explained (42,71,72,92,98,100).  The equation for the utility function that describes respondent i’s utility for choice option j is shown below (4.1); the number of choice options in a DCE is 1 to J.  (4.1)  The standard practice of DCEs is to assume that this utility function is linearly additive (65). Thus, Vij can be further broken down into a vector of coefficients,  ,  representing respondents’ mean utility for attributes, as well a vector of variables,  ,  that represents each attribute presented to respondents (Equation 4.2).  33  (4.2)  This equation shows that by asking respondents to choose between treatments based on each treatments’ attribute levels, it is possible to estimate their utility for these levels. Furthermore given that the utility function is additive, respondents’ mean utility for choice option j can be estimated by summing respondents’ mean utilities for the attribute levels of the choice option. Consequently, when  is incorporated  into Equation 4.1, it can be seen that respondents’ true utility can also be broken down into a vector of coefficients representing respondents’ mean preferences for each attribute or attribute level, as well as a vector of variables that represent the attributes that are presented to respondents (Equation 4.3).  (4.3)  Discrete choice experimentation assumes that respondents are rational and utility-maximizing. Given this, they will choose the treatment option in the DCE choiceset for which they have the highest utility (65,96,97). Because the unobservable component  i  is not observed in the DCE, a joint probability distribution for the non-  explainable random component is assumed (Equation 4.4).  (4.4)  Pi (Yi=1) =Prob  ,  34  where Yi represents the choice outcome for respondent i. A value of 1 means that choice option A is chosen. Thus, the probability of a respondent choosing option A is 100% when their utility for A is greater than their utility for option B (Equation 4.4) (13). Replacing  (4.5)  by  gives Equation 4.5 below:  Pi (Yi=1) =Prob  The rearrangement of equation 4.5 gives equation 4.6.  (4.6)  Pi (Yi=1) =Prob  Discrete choice data are modeled based on the joint probability distribution for the non-explainable random component. The type of model used to fit DCE data is chosen based on what is known about the unobserved factors. For example, if the distribution of  ib -  ia  is assumed to be independently and identically distributed, a  multinomial logit (MNL) model applies (60,65,97). The MNL model is the most basic choice model used to model discrete choice data. Other variations in modeling DCE data, such as the mixed logit (MXL) model, are based on the MNL (60,65,97). The advantages and disadvantages of different choice models will be later described.  35  4.4 DCE questionnaire design  4.4.1 Attribute selection Attributes are the features of healthcare options. Thus, they can be quantitative, e.g., the risk of treatment-related myocardial infarction (MI), or qualitative, e.g., mode of treatment administration (65,97). Ways to identify attributes that are important for inclusion in a DCE can be through literature reviews and/or consultations with content experts in the field, as well as through qualitative methods, e.g., interviews, questionnaires, or focus groups. Questionnaires or interviews help to identify the healthcare attributes that are important enough to patients to influence their healthcare choices (65,97). There are a number of factors that are important to consider when determining the key attributes for inclusion in a DCE study (65,97).  Firstly, attributes should be comprehensive (65,97). For example, all the attributes that would influence rheumatoid arthritis (RA) patients’ choice of their nonsteroidal anti-inflammatory drugs (NSAIDs) need to be included in a DCE for trade-off. This is important because if they don’t then utility estimates for attributes may be incorrect as patients are not trading off between all attributes that may influence their healthcare decisions. Another important quality of DCE attributes is that they have to be operational, i.e., meaningful to respondents (60,65,97). For example, RA patients may value treatments that result in relief of joint pain whereas those with chronic headaches may value treatments that reduce the pain from headaches. Thus, when evaluating the preferences of those with RA, it would not make sense to include headache pain relief as an attribute. One way to ensure that all the most meaningful attributes are included is 36  to use focus groups or questionnaires to ask patients what attributes, e.g., risks or benefits of treatment, may influence their healthcare decisions.  In addition to being aware of whether or not the attributes are operational, it is also important to avoid attribute redundancy (65,97). For example, if patients consider a benefit of reduction in pain to be analogous to a benefit of improved function, it would be necessary to only include one of these benefits as a DCE attribute. Prevention of redundancy is important as it stops the researcher from double estimating the utility for the same attribute. In addition, keeping the number of attributes to a minimum simplifies the DCE design, and reduces cognitive burden for patients by ensuring that they are not trading off between more attributes than they need to (60,65,97).  4.4.2 Level and level-range selection In addition to identifying the key attributes for a DCE, the attribute levels and their ranges need to be chosen appropriately. Similar to attributes, levels can be either quantitative or qualitative (65,97). There is no set number for the number of levels, but it is also advised to keep the number to a minimum. Keeping the number of levels small decreases response variability due to task complexity (65,97). In addition, increasing the number of levels increases the number of choice-sets that need to be presented to a respondent. This can be cognitively burdening for respondents. There are two factors important in level and level-range selection.  Firstly, levels must be plausible to respondents. Therefore, the range of levels can exceed current possible healthcare options but must still be feasible. For example, 37  it would not make sense to present a treatment option that has a complete cure to those with Crohn’s disease as this does not exist and the preferences elicited would aid in understanding patients’ real life treatment choices. In addition, the level ranges either capture the estimates/values that are commonly associated with each particular attribute or at least be hypothetically feasible and reasonable (65,97). For example, the estimated risk of a heart attack associated with taking NSAIDs is less than 1% for those with RA. Therefore, the range of levels for this attribute should capture and include 1%, e.g., 0%, 1%, 2%.  Another important factor to consider is to attempt to make sure that both levels and their ranges are meaningful enough to respondents that they would be willing to make trade-offs between choice options. As an example, levels of 20%, 21%, and 22%, may be meaningful risk levels to the respondent, but the range between each level may be too small that the respondent is unlikely to differentiate between them. This may result in preference estimates being similar for all three attribute levels (97). On the other hand, when patients’ have similar preferences for levels within an attribute, it could actually mean that they are indifferent to these levels of risk. For example, if one finds that respondents’ preferences are the same for a 0%, 1%, and 2% risk of MI, this may suggest patients’ risk tolerance level is potentially higher than assumed given greater potential benefit. Consequently, one way to ensure that levels and ranges are meaningful to patients is to pilot test the DCE prior to conducting the full survey.  38  4.5 Experimental design In a DCE, the total number of possible choice options in a DCE with ‘X’ number of attributes and ‘L’ levels is Lx (97). A full factorial design is where all choice options are presented to respondents. Full fractional designs allow for main effects estimation, i.e., the estimation of the effect of each attribute on influencing patients’ preference for a choice option, as well as the estimation of the interaction effects between two or more attributes on patients’ preferences (65,97). However, a full factorial design can be too large to be used in practice. For example, the total number of possible treatment options that would need to be presented to respondents in a DCE with 6 attributes, each with 3 levels, is 729 (i.e. 36)! Therefore, a fractional factorial design is more practical.  A fractional factorial design is a variation of the full factorial design (65,97). This design presents a sample of the total Lx number of possible choice options to respondents while still allowing for efficient estimation of preferences for attributes (65,97). However, this design may be more limited in measuring interaction effects (97).  To ensure that treatment preferences are estimated with precision, the design efficiency or D-efficiency of the DCE experimental design should be evaluated (65,97); a higher D-efficiency indicates that the parameter estimates should be more precise. The D-efficiency of a design is typically evaluated by comparing the design strength of one DCE choice design relative to that of an alternate DCE choice design. Optimal Defficiency is when four factors are achieved: level balance, orthogonality, minimal overlap, and utility balance (60,65,97).  39  Level balance is when the levels within an attribute occur with equal frequency in all included choice options of the design; level balance is not considered when using full factorial designs as all alternatives are given to respondents (60,65,97). For example, for an attribute with three levels, each level should appear in one third of all shown choice options. In addition to appearing in equal frequency with the levels of the same attribute, an attribute level must occur in equal frequency with all other attribute levels the DCE. This is orthogonality and is important to ensure that there is minimal overlap of attribute levels, i.e., a level does not repeat itself in a choice set, as well as no colinearity between attributes (65,97). Without level balance and orthogonality, the preferences elicited for each attribute level may be biased as patients would not be trading equally as often between this attribute level and other attribute levels.  The final important factor of DCE experimental designs is utility balance. Utility balance helps to prevent dominance of one choice option over another. An example of dominance is a choice option that has greater benefit and lower risk than the other option in a DCE choice-set. If dominance occurs, respondents don’t need to make tradeoff between attributes levels of different choice options and there is no meaningful preference data acquired for that choice set.  4.6 Inclusion of an opt-out option In each DCE choice-set, there is also the option of including a status quo or optout option as a third choice option. This may be the recommended method to conducting DCEs as it is more congruent with situations where patients may not be forced to choose between two potentially unappealing alternatives (60,65,97,99,101). 40  By forcing respondents to choose between two alternatives researchers may overestimate the probability that respondents may actually choose one choice option over another in real life. Therefore, inclusion of an opt-out option may reduce potential bias in the estimation of respondent utilities for attributes levels in comparison to a DCE design with no opt-out option (60,65,97,99,101).  However, one potential limitation to having an opt-out option is that patients may have different viewpoints on what the opt-out option entails (99). For example, does it mean that they would be taking no drug treatments and therefore would not experience any benefits and side effect, or that they would be at their current health status? However, this limitation may be avoided if the meaning of the opt-out choice option, including the relevant attribute levels such as 0% risk and 0% benefit, are clearly explained to the respondents prior to completion of the DCE choice-sets.  Another problem that may occur with including an opt-out option is that some patients could choose the opt-out option to avoid making difficult choices (99,101). Therefore, it may be difficult to differentiate between patients choosing who are indifferent to the attribute levels and those who do not like making difficult decisions. However, one potential solution is to assume that those who do not like having to choose between options would consistently choose the opt-out for all the choice-sets. Therefore, the consistent opt-out respondents could be excluded from the analysis. Additionally, testing of a significant difference between preference estimates from this analysis compared to one that includes those that consistently chose to opt-out could be conducted. In addition, the socio-demographic and clinical characteristics between these two sub-groups could also be compared. 41  The final problem when including an opt-out option is when choosing what model to use to fit discrete choice data for analysis. In the beginning, the most basic model to analyze DCE data is the MNL model. However, the use of this model is only appropriate when all choice options are close substitutes. This is because the MNL assumes proportional substitution patterns across all choice options (101). Thus, given choice options A, B, and the opt-out option, the MNL model assumes respondents would equally prefer each of these choice options. Given this assumption, an increase in the probability of option A would result in a proportional decrease in the probability of choosing option B and the opt-out. However, this may not be true, as changes in the magnitude of the attributes in option A may affect the probability of respondents choosing A and B, but not the opt-out (101). Consequently, alternative models such as the generalized extreme-value, probit, and mixed logit model (which will be described later) can be used when modeling DCEs that have included an opt-out option.  4.7 Data collection  4.7.1 Pilot testing Prior to conducting a full-scale survey, it is important to pilot the survey for a number of reasons. The first is to test if the survey is understandable. Another reason for pilot tests is to identify any potential problems in the study methodology or DCE questionnaire that need to be addressed before running the full study. For example, if the patients in the pilot study are taking too long to answer the DCE questionnaire, this may mean a problem with the administration of the DCE questionnaire itself, i.e., it is too long, there are too many choice-sets being presented, or that there are too many 42  attributes to trade-off. By knowing these problems in a small sub-set of the study population, researchers may be able to fix these problems and prevent incorrect data collection before the questionnaire is sent out to the entire study population. In addition, pilot testing also validates that the attributes and descriptions of the attributes are comprehensive, e.g., respondents understand what having an ulcer would entail (65,97).  Pilot tests also test if the attributes, levels, and ranges are meaningful to the patients. For example if researchers find that preference estimates are the same for all risk and benefit attributes, this may mean that respondents are indifferent between the levels of risks and benefits. One potential reason could then be that the level ranges are too close, e.g., 1%, 2%, 3.  4.7.2 Methods of data collection There are many methods to collect respondent data in a DCE study. Previously used methods include interviews or mailed questionnaires/surveys. Recently, due to advances in technology programs such as Sawtooth® CBC/SSI Web V6.4.2 (Sawtooth Software, Inc. Sequim, WA, USA), DCE questionnaires can be administered online. This allows respondents to complete DCE surveys in the comfort of their own homes, or wherever they can gain internet access.  The primary advantage of using an interview format for administering a DCE is that it allows the respondent the opportunity to ask the interviewer questions, e.g., how to fill out the DCE questionnaire. Although questions such as the aforementioned 43  should be anticipated and detailed instructions should be provided in DCE studies without interviewers present, the presence of the interviewer increases the probability that the participants are filling out the DCE questionnaire correctly. However, because the interviewer is present there are a couple potential disadvantages.  The first is the cost in time and money to train interviewers and have them administer the questionnaire to patients. Secondly, interviewers need to be strictly trained to ensure that they administer the questionnaire in the same format to all patients. However, there is potential for human error. For example, interviewer bias may occur if interviewers unintentionally (or intentionally) influence the respondents to give the answers that they want  Mailed and online questionnaires prevent interviewer bias. They are also less costly in time and resources in comparison to DCEs given in an interview-format. In addition, because respondents complete the questionnaire on their own time, they are given more time to think about their answers. This may increase the quality of the data compared to DCE questions that have to be filled out right away in interviews. In addition, depending on the strategy of respondent recruitment, mailed/online questionnaires may allow a more diverse sample of respondents in a wider range of geographic locations.  However, in DCEs that are online respondents that don’t own a computer or are computer illiterate may be missed. If this occurs, the preferences of the sample population may not be representative of the entire patient population. In addition, because interviewers are not present in online or mailed DCEs, there detailed 44  instructions on how to complete the DCE questionnaire as well as the meaning of each attribute and their outcomes, must be provided for respondents. If the instructions are not adequate, respondents may not fill out the questionnaire completely and missing data can occur. For example, if patients do not understand the seriousness of having an ulcer, then they may choose treatment options with a higher risk of ulcers than they would if they realized the seriousness of this adverse event. Consequently, all forms of DCE administration have their strengths and limitations. Therefore, the method of data collection is up to the researcher based on the type of resources available as well as the types of biases, associated with each form of data collection that can be accepted as limitations of the study.  4.8 DCE analysis  4.8.1 Coding of variables Every attribute included in the DCE questionnaire is considered as a separate variable in the model used to fit DCE data. Attributes which have levels that are quantitative, e.g., 0%, 50%, 100%, can be considered continuous variables if respondents’ preferences are linear across the levels within each attribute (65,97). In comparison, attribute levels that are considered qualitative or categorical, e.g., low, moderate, or high, can be coded as a separate variable (65,97); it is important to note that levels which are numerical, e.g., 1%, 2%, 3% can also be considered categorical. Levels that are considered categorical can be coded in two ways: dummy or effect coding.  45  When variables are dummy coded, L-1 dummy variables are created, where L is the number of levels for the attribute (65,97,102). These dummy variables are assigned values of ‘1’ and ‘0’ to reflect the presence or absence of an attribute level (65,97,102); the reference level ‘L’ is assigned a value of ‘0’. In effect coding, L-1 variables are also created, however, the reference level ‘L’ is assigned a value of ‘-1’ instead (Table 4.2) (65,97,102).  Because the reference level is assigned a value of ‘0’ in dummy coding, the preference weight for the reference level is incorporated into the regression intercept (65,97,102). This may cause misinterpretation of parameter estimates, as the intercept would include the preference weights for the reference levels of all attributes(65,97,102). In comparison, in effect coding the reference level is the negative sum of the other levels (65,97). Thus, the reference level of each attribute is not part of the intercept. Consequently effect coding is the recommended coding method for all variables whose levels are considered categorical in discrete choice analysis (65,97,102).  4.8.2 Choice models Because multiple observations are obtained from each respondent in a DCE, random effects choice models are used to fit discrete choice data (7-73,92,98,102-103). Different ways to model and analyze DCE data are based on the different assumptions about the distribution of the unobserved portion, i.e., (  ), of respondents’ true  utility. As mentioned above, the most basic choice model is the MNL model. Other  46  choice models include: generalized extreme-value (GEV), probit, and multinomial or mixed logit (96).  The MNL model is the most widely used to model discrete choice data. It is based on the independence of irrelevant alternatives (IIA) assumption, i.e., the unobserved factors that influence respondents’ preferences are independent of each other over choice options and time (96). In the context of treatment options, this means that the unobserved factors that may influence a patient’s choice of one treatment option are unrelated to the unobserved factors influencing the same patient’s choice of other treatment options. However, this can be unrealistic. For example, if a patient does not like drugs that give them gastrointestinal (GI) problems, then it would be assumed that this factor would make them averse to a treatment with increased risk of peptic ulcers in one DCE choice-set, as well as another treatment with an increased risk of dyspepsia in another choice-set.  Another assumption with the IIA is that the factors influencing choice options are uncorrelated over time, i.e., the choice of a treatment in one choice-set is unrelated to preferred treatment options in future choice-sets (96). This assumption can also be problematic. Going back to the example of a patient who is strongly averse to GI AEs, their preference of a treatment with a reduced risk of ulcer is most likely going to affect their preference of a treatment in a subsequent choice-set that also has a reduced ulcer risk. Thus, it is probable that patients’ answers to previous choice-set questions may be correlated to their choice of preferred treatment options over time.  47  By assuming independence of unobserved factors over alternatives and time the MNL model is also restricted to proportional substitution rates among treatment alternatives (96). This means that an increase in the probability of choosing one treatment alternative results in a proportional decrease in probability of choosing the other available treatment alternative. As an example, consider two treatment options, A and B, with the same probability of being chosen, i.e., 0.5. If the probability of option A being chosen increases by 0.25 because it has greater benefit and lower risks, than it would be assumed that the probability of B being chosen would decrease by 0.25. However it may not be appropriate to assume proportional substitutions in situations where an opt-out option is included in the DCE. This is because changes in the treatment benefits and risks of option A and B may not influence or change the probability of the opt-out being chosen to the same degree. Consequently, there are other models that relax the MNL IIA assumptions.  One such model is the generalized extreme-value (GEV) model. The GEV model is not as restrictive as MNL as it allows for the correlation of unobserved factors over treatment options and becomes the MNL model when the correlation between them is zero (96). However, correlation can only exist between unobserved factors for groupings of treatment options called nests. These nests are assumed a priori, with no correlation between nests. An example of a nest may be a group of treatments with risk of GI events. Therefore, the limitation of the GEV is that it requires that the correct assumptions be made on the number of nests needed, as well as the choice options that belong to each nest.  48  A choice model that does not require that choice options be grouped into nests is the probit model. Similar to GEV, the probit model also allows for correlation of unobserved factors over time and choice options; however instead of creating nests, the probit model assumes the unobserved factors follow a normal distribution (96). This may not be realistic for all attributes. For example, if cost is one of the attributes included in a DCE study, it is expected that patients would only be averse to increases in treatment costs. Therefore it would not be logical to assume that this attribute would be normally distributed, as that would imply that some patients would prefer increasing cost. Consequently, if attributes in the DCE cannot be specified with a normal distribution an alternative choice model to use is the mixed logit model.  An advantage of the mixed logit (MXL) model over a probit model is that it allows the unobserved factors to follow any type of distribution. In addition to allowing for the correlation of the unobserved factors over choice options and time, the MXL model also incorporates for variation in preferences between individuals (96). Thus, it is does not assume that all respondents have the same preferences for attribute levels.  When using the MXL model to fit DCE data, the type of distribution for each attribute is assumed a priori. Therefore, a potential limitation is if the assumed distributions are incorrect. However, one way to avoid an error in the specified distribution for attributes is to try different types of distributions and compare the model fit between them before staying with a particular distribution for an attribute.  49  4.9 Interpretation of results Choice models used to fit DCE data allow the evaluation of the study sample’s mean preferences for attribute levels. These preferences are given in the form of regression coefficients. Levels within an attribute that are treated as categorical have their own regression coefficient, whereas attributes with levels that are assumed to run on a linear scale have a regression coefficient for the attribute itself.  For attribute levels that are categorical, the coefficient reflects respondents’ mean utility for that attribute level relative to the other levels within the attribute. In comparison when regression coefficients are evaluated for an attribute itself, a positive coefficient would indicate that respondents prefer more of this attribute, whereas a negative coefficient indicates that respondents prefer less of this attribute.  As mentioned previously, the preference parameters or regression coefficients elicited with a choice model are assumed to be additive. Thus, an estimation of patients’ preference for a treatment with specific benefits and risks is possible. For example, if a patients’ utility was 0.5 for a 20% chance of benefit, and -0.2 for a 2% risk of ulcer, the overall preference for a treatment with these attribute levels would be 0.3 (0.5 + -0.2). Consequently, by estimating patients’ preferences for treatments with different benefitrisk profiles, it can be hypothesized what treatments patients may prefer over others.  Regression coefficients for attribute levels can be used to determine patients’ maximum acceptable risk (MAR) of an AE given a specific benefit. The MAR is the risk of an AE that patients are willing to accept for a given benefit. (23,24). Johnson and  50  colleagues have previously used MAR to quantify the risk tolerance of those with multiple sclerosis, Crohn’s disease, and irritable bowel syndrome (IBS) (60,61,93).  If all risk attribute levels are specified to be categorical variables, respondents’ MAR of AE i is estimated by first determining the utility gain given a benefit (Equation 4.7).  (4.7)  Let  represent respondents’ mean relative utility for an initial benefit level, e.g.,  20%, whereas  represents respondents’ mean relative utility for a final benefit  level, e.g., 100%. Thus,  would represent the utility gain going from a 20%  to a 100% chance of benefit. In DCEs with multiple benefit attributes, j is the number of benefit attributes. For example, if j = 2 the benefit could be a greater chance of benefit in functional improvement.  The disutility going from 0% to the MAR level must offset the utility gain given a benefit improvement. Thus, assuming respondents’ MAR of AE i, lies between risk attribute level k-1 and k , e.g., the MAR lies between risk attribute level 1 and 3, the MAR of AE i, can be estimated by the following (Equation 4.8) :  (4.8) 51  where  and  represents the risk at k and k-1 respectively, e.g., at k=2,  = 2%.  The above equation (4.8) is only valid if k is one of the 1 to M risk attribute levels included in the DCE. For example in a DCE with a 3-level risk attribute, k must be 1, 2, or 3. Fraction,  , is determined through piecewise linear interpolation between  respondents’ mean relative utility at two discrete risk levels using the following equation (4.9):  (4.9)  where  and  represent respondents’ mean relative utility at k and k-1 respectively.  is respondents’ mean relative utility given at the reference risk level, i.e., 0% risk. However, If the utility gain given a benefit is greater than the disutility going from risk attribute levels 1 to M, i.e.,  , then respondents’ MAR of AE i is  estimated to be greater than M. For example, if the utility gain given a benefit is greater than the diutility going from a 0% to a 4% risk of MI, then the MAR of MI is greater than 4%. This is because it may not be appropriate to assume linearity between levels of risk beyond the data points that are collected in the DCE. Thus, the exact value of respondents’ MAR of AE i is not estimated.  4.10 Conclusion It is hoped that through the use of DCE methodology, patients’ preferences for healthcare attributes can be estimated. However, the DCE methodology has several 52  potential limitations. The first is that regression coefficients for each level within an attribute are centered on zero, i.e., the sum of the regression coefficients for all levels within an attribute must be zero. Therefore, given a three-level attribute, with two of the levels having positive coefficients, the last remaining level has to be negative. For example, if preference estimates is -0.2 for both 4% and 8% risk of stroke, then the preference estimate for a 2% risk of stroke is the negative sum of the estimates for the other two levels, i.e., +0.4 (-1*(-0.2+-0.2)). As mentioned previously, a positive coefficient would indicate that patients prefer a 2% risk of stroke. However, this preference is only relative to the fact that patients’ are averse to 4% and 8% risk of stroke.  Another potential limitation of DCEs is if the potential external validity of the preferences elicited are valid for the entire patient population. Given that DCEs are only administered to a sample of patients, this may not be true depending on how the study sample was recruited, e.g., if the patients recruited are all from the same clinic. If patients were all recruited from the same area, they may have socio-demographic and clinical characteristics that are different from the rest of the general patient population. For instance, if the patients sampled have more severe disease, results may indicate a greater risk tolerance than that of the general patient population.  Another factor that may affect the generalizability of DCE results is the method of DCE administration, i.e., if the DCE is given in an interview format, mailed out, or administered online. This may affect the representativeness of the study sample as there may be different types of patients responding to the different forms of administration. As an example when administering a DCE online, you may miss those 53  of the population who do not own a computer or are computer illiterate. Furthermore, even if participants recruited into a DCE study are representative of the entire patient population, there is also the issue of if the choices that respondents make in the study reflect their real choices in their daily lives.  The preferences elicited may not be representative of the entire population because attributes of treatments that are important to patients may be missed because of error or just simply not included as it is unclear what levels to include in the DCE for these attributes. Example of attributes that would be difficult to specify levels for would be attributes that have to do with how patients feel about the treatment outcomes, e.g., feeling sad or better. However, it is hoped that if qualitative methods such as focus groups were used prior to constructing the DCE questionnaire, all the key attributes have been included in the DCE.  Regardless of the potential limitations of the DCE methodology, evidence supports the DCE technique to be an internally valid and consistent tool to elicit patients’ preferences (91). Discrete choice experimentation has been found to encompass all three key axioms of a good preference measurement tool: completeness, stability and rationality (92). In addition, the results of DCEs have been found to be similar to patients’ preferences using previously validated preference measurement tools, such as SG (91,92).  In the context of eliciting patients’ treatment preferences, the DCE methodology allows researchers to examine the factors that influence patients’ treatment decisions using hypothetical treatment scenarios. Thus, one advantage to the use of the DCE is 54  that it allows the presentation of treatment options with attributes and magnitudes of these attributes that are similar to real treatment options currently available to patients. In addition, because treatment scenarios are hypothetical, choice-sets can also consist of attributes with treatment benefits and risks that feasible but outside of current treatment options. This allows the extrapolation of patients maximum risk tolerance and may be useful for more informed drug regulatory and clinical decisions.  55  4.11 Tables Which of these two options do you prefer? Attributes  Option A  Option B  Risk of Heart attack  1%  2%  Pain reduction  50%  70%  Table 4.1 Example of one choice-set question * Note checked box indicated option A is chosen  56  Level  Dummy  Dummy  Effect  Effect  code 1  Code 2  Code 1  Code 2  A  0  1  0  1  B  1  0  1  0  C  0  0  -1  -1  Table 4.2 Illustration comparing dummy vs. effect coding for 3-level attribute  57  Chapter 5- Benefit-risk tradeoffs for NSAIDs for the management of rheumatoid arthritis: a discrete choice experiment  5.1 Introduction Rheumatoid arthritis (RA) is a chronic and debilitating disease where the body’s immune system attacks its own healthy joints (1). Those afflicted with RA suffer from joint inflammation, which results in functional disability, fatigue and chronic pain (2-4). Although there is no cure, there are effective treatments that can reduce RA symptoms, thereby improving quality of life and reducing the burden of this disease on society.  Treatment strategies for RA typically include a combination of drugs such as non-steroidal anti-inflammatory drugs (NSAIDs), corticosteroids, disease-modifying antirheumatic drugs (DMARDs), or biologic DMARDs. Some of these drugs are taken to reduce joint pain and inflammation, e.g., NSAIDs or corticosteroids, while others are taken to slow RA disease progression, e.g., DMARDs. In addition to varying in the potential benefits, these drugs also differ in their risks of adverse events (AEs). Thus, for those with RA, treatment decisions can often be based on their personal preferences for the benefits and risks of AEs. However, in general drug approval or withdrawal decisions are based primarily on the risks and not on patients’ preferences. This can be problematic for patients who are willing to accept the risks given the potential benefit they might experience.  Rofecoxib (Vioxx®) was voluntarily withdrawn by Merck due to concern of a potential increased risk of cardiovascular (CV) events, such as myocardial infarction, 58  sudden cardiac death, unstable angina pectoris, ischemic insult, transitory ischemic attacks, and peripheral vascular occlusion, relative to placebo (5,6). However, the withdrawal occurred despite the fact that some RA patients who had been on the drug may be willing to accept the risk of CV events given the benefits that they had experienced from rofecoxib (7,8). Furthermore, when expert panels from the Food and Drug Administration (FDA) and Health Canada convened to weigh the evidence on rofecoxib, they concluded that its benefits outweigh the potential safety concerns in patients with RA (9-11). These conclusions were later supported by an incremental netbenefit analysis from the societal perspective, using the same data presented during the expert panels, which found that the benefits of rofecoxib outweigh its risks relative to naproxen (12). Although it has not been empirically demonstrated that the RA risk tolerance is greater than what they are currently exposed to in their treatments, there is evidence to suggest this may be true.  When drug approval or withdrawal decisions are solely based on safety concerns, patients’ access to potentially beneficial drugs may be hindered. Thus, the US Institute of Medicine (IOM) and the Committee for Medicinal Products for Human Use (CHMP), of the European Medicines Agency (EMA), recommend that methods are needed to effectively measure, quantify, and apply benefit-risk analysis during the preand post-approval processes of drug regulatory decisions (13,14). One such method that is being evaluated for use in these processes is the stated choice technique. The discrete choice experiment (DCE) is a stated choice technique that can be used to evaluate patients’ willingness to trade-off risks for potential benefits. The FDA has previously considered patients’ preferences elicited by a DCE during their deliberation on the re-approval of the drug natalizumab (Tysabri) (15). 59  Discrete choice experimentation allows the elicitation of patients’ preferences for treatment through the administration of a questionnaire-based survey. In a DCE survey to elicit patients’ treatment preferences, respondents must choose between different hypothetical treatments based on the different magnitudes of the positive and negatives features, i.e., attributes, of each treatment option (16). Through a DCE, the maximal acceptable risk that patients are willing to accept given a specified benefit can be estimated. The DCE was used to evaluate the treatment preferences of those with RA with the primary objective of estimating RA patients’ MAR of ulcers, dyspepsia, myocardial infarction (MI), and stroke given a chance of improvement in pain and function. Subsequent study aims were to: (a) determine if from the perspective of patients with RA, the potential benefit of coxibs outweigh their risks compared to NSNSAIDs, and (b) to compare RA patients’ preference for coxibs to their preference for NSNSAIDs. Respondents’ mean relative preferences for treatment attributes were then used to estimate and compare RA patients’ preference for coxibs to their preference for non-selective NSAIDs (NSNSAIDs). It was hypothesized that given greater potential benefit, RA patients’ MAR for myocardial infarction (MI) and stroke is greater than the risks of these AEs that patients are exposed to in current and past NSAIDs.  5.2 Methods  5.2.1 Recruitment strategy Participants were recruited from the Mary Pack Arthritis Program (MPAP) and Arthritis Consumer Expert (ACE) databases. Those with RA in these databases were 60  mailed invitations to participate in the study. Participants were eligible for enrollment if they were 19 years or older, had physician diagnosed RA, and were fluent in both reading and writing English. Potential interested participants, who met the inclusion criteria, were emailed a link to complete the DCE questionnaire online. As this questionnaire was administered online, participants needed to be computer literate and have access to a computer.  In addition to completing the DCE questionnaire, patients’ socio-demographic and RA-related characteristics were also collected. Clinical characteristics were obtained using questions from the health assessment questionnaire (HAQ), short form 6-D (SF-6D), as well as a validated instrument of measuring patients’ self-reported number of tender joints. The University of British Columbia Behavioral Research Ethics Board (BREB) and the Vancouver Coastal Health research ethics boards approved this study. Each participant received a $20 honorarium.  5.2.2 Discrete choice experiment In a DCE to elicit patients’ preferences for treatment, respondents are presented with a series of choice-sets, each with a pair of hypothetical treatment options. Each treatment option has the same treatment attributes, but varies in the levels of these attributes (16-20). Therefore, based on the attribute levels of each treatment, respondents are asked to indicate which of the two options they prefer (16-20). For this DCE a third alternative, the opt-out option, was included in each choice-set as it is more congruent with the actual RA clinical decision-making scenario as patients could choose  61  to remain untreated and thus are not forced to choose between two potentially unappealing alternatives.  Appropriate attributes for inclusion in the DCE were determined through the conduction of 5 focus groups, each with 3-4 RA patients recruited from MPAP and ACE. The primary objective of the focus groups was to identify key attributes and levels for inclusion in the DCE. Participants of the focus groups were asked to identify the attributes of NSAIDs, i.e., benefits and AEs, and magnitudes of these attributes that would influence their treatment decisions.  In the focus groups, participants with RA mentioned that they valued treatments with pain reduction and functional improvement. However, when asked to describe the magnitude of benefit they preferred, they described that they wanted enough pain relief to perform their most valued daily activities. Therefore, to prevent redundancy of attributes in the DCE questionnaire, the benefit included in the DCE was levels of functional improvement.  Participants were told that treatment options which gave  greater function improvement in their most valued daily activities (MVDAs) also had greater pain reduction; prior to completing the DCE choice-sets RA patients were asked to think of a minimum of one MVDA that they wanted improved function to perform.  In the focus groups, RA patients also identified that they were averse to side effects commonly associated with NSAIDs, i.e., ulcers, dyspepsia, MI, and stroke. However, they described these AEs in layman’s terms, e.g., ulcers were stomach bleeds, dyspepsia was heartburn, and MI was heart attack. Thus, because patients may not understand what it means to have an ulcer, dyspepsia or MI, these attributes were 62  presented to DCE participants in layman’s terms so that patients would better understand the DCE choice-sets. Similar to past RA preference literature, those with RA in the focus groups differed in the magnitude of their risk tolerance. Thus, levels for risk attributes were determined by making sure they captured the estimates/values that are commonly associated with each risk attribute. For example, the NSAID-related risk of MI for those with RA is less than 1%. Therefore, the levels for MI in this DCE were 0%, 2%, and 4%.  Thus through focus groups and a review of literature on RA patients’ preferences, six attributes were included in this DCE: (1) chance of benefit (2) level of improved function, and risks of (3) ulcers (stomach bleeds), (4) dyspepsia (heartburn and upset stomach), (5) MI (heart attack) and (6) stroke. Each attribute had three possible levels (Table 5.1). Each attribute was accompanied with a detailed description of the outcomes associated with it. For example, the seriousness of a stomach bleed was described (Appendix 2- The complete list of definitions for treatment characteristics).  The levels for the attribute ‘level of function improvement’ were based on levels from the SF-6D health index for the pain domain. The attribute ‘chance of benefit’ was included as it is more realistic of treatment scenarios where patients are not guaranteed that their treatment will work to give them a benefit. The level range of this attribute (20%, 50%, 100%) was chosen to create a wide range of benefit. However, it was felt that treatment options with 0% chance of benefit would not be realistic given that the least amount of function improvement is a little bit, i.e., 0% of a little bit of functional improvement would not make sense. In addition, pilot tests of the DCE questionnaire 63  were conducted in a sample of RA patients to assess the appropriateness of the attributes and levels for evaluating RA patients’ preferences. Pilot tests also tested the clarity of attribute descriptions for those with RA.  Following the selection of attributes and levels, twenty versions of the DCE questionnaire were created. This was done using a commonly used algorithm to construct a statistically efficient fractional factorial experimental design using Sawtooth® CBC/SSI Web V6.4.2 (Sawtooth Software, Inc. Sequim, WA, USA). To ensure optimal design efficiency, the experimental design was tested for: (1) orthogonality, i.e., minimal correlation between attributes, (2) level balance, i.e., levels occur at equal frequencies, and (iii), minimum overlap, i.e., no attribute level repeats itself in a choice-set (16). Each version of the DCE questionnaire consisted of 10 hypothetical choice-sets. Furthermore, two additional choice-sets were included in each version to test respondents’ comprehension of the DCE question and task as well as to evaluate respondents’ rationality by fixing one choice alternative as the ‘logical’ or dominant option (i.e. greater benefit and less risk). Those who answered either one or both dominant questions incorrectly were excluded from the final analysis.  5.3 Analysis  5.3.1 Descriptive and bivariate analysis The study sample’s socio-demographic and clinical characteristics are reported as means or medians if they were continuous variables; if categorical, frequencies of respondents belonging to each category were reported. Significant demographic 64  differences, setting p=0.05, was tested between respondents recruited through MPAP versus ACE, as well as between those who answered the dominant questions correctly and incorrectly. Categorical variables were compared using Chi-square tests while continuous variables were compared using two-sample t-tests.  5.3.2 Choice models Two separate choice models were specified: (1) a multinomial logit (MNL) model, and (2) a mixed logit (MXL) model. Discrete choice data was analyzed using the MNL in SAS 9.1 (SAS Institute, Inc., Cary, NC, USA; www.sas.com), whereas the MXL model was used to fit discrete choice data using a Matlab code created by Train (available from: http://elsa.berkeeley.edu/~train). Compared to the MNL, the MXL model allows for the existence of variation in respondents’ preferences.  In the MXL model, all attribute levels, except for levels with a 0% risk, were specified to have a normal distribution; a normal distribution is assumed as some respondents may differ in how averse they are to a risk level. Levels of 0% risk were assumed to have no respondent heterogeneity in preference and were specified with a fixed distribution. By specifying the distributions of each attribute level, an estimate of the mean and standard deviation for each level was elicited. Attribute levels that had standard deviations that were statistically significant from zero (p<0.05, 2 tailed) suggested that preference heterogeneity existed among respondents. To explain for preference heterogeneity, respondents were stratified based on their sociodemographic and clinical characteristics; appropriate strata groups were determined based on the frequency distribution of respondents. The Wald test was used to 65  determine if there is a significant difference in mean preference estimates between different strata.  In both choice models, all six attributes were effect coded to allow for the estimation of respondents’ mean relative utility for each attribute level. To do this, L-1 variables were created, where L is the number of levels for an attribute (16,21,22). These variables were assigned values of ‘1’ and ‘0’ to reflect the presence or absence of an attribute level; the reference level ‘L’ was assigned a value of ‘-1’(16,21,22). Thus RA patients’ mean true utility U for a treatment option j was estimated using the following model:  (5.1)  where in a choice-set t, respondent i chooses a treatment option based on the following treatment attributes and levels: chance of benefit ( improvement (  ), risk of ulcer (  ), level of function  ), risk of dyspepsia (  ), risk of MI 66  (  ), and risk of stroke (  ). In addition, two-way interaction terms were  incorporated into the MNL model to examine if the interaction between the attributes chance of benefit and level of improved function would significantly influence patients’ treatment choices; the code for the MXL model did not allow for incorporation of interaction terms. If model fit was not significantly better with the inclusion of interaction terms, these terms were left out of the final model.  Parameter estimates or respondents’ mean relative preference weights for each attribute level were given in the form of regression coefficients. Positive coefficients indicated more of this attribute was preferred, whereas negative coefficients indicating that patients are averse to increases in this attribute. It was assumed in this study that the signs of the coefficients would be negative for greater risk, and positive for greater benefit. The values of the preference weights indicated the relative strength of patients’ preference for levels within each attribute.  Rheumatoid arthritis patients’ maximal acceptable risk (MAR) of ulcers, dyspepsia, MI, and stroke was estimated using respondents’ mean relative preference weights for each attribute level from the MNL model. Johnson and colleagues previously used this method to assess the risk tolerance of those with irritable bowel syndrome (IBS) (23). The MAR represents the risk of an AE that RA patients would be willing to accept for a given benefit. (23,24). In this DCE , the RA MAR of ulcers, dyspepsia, MI and stroke given a 20% to a 50% or a 20% to a 100% chance of benefit, as well as a little bit to moderate or a little bit to a lot of functional improvement could be estimated. This equates to 16 MAR estimates that were evaluated, i.e., 4 MARs for each benefit. 67  If all attribute levels are specified to be categorical variables, respondents’ MAR of AE i is estimated by firstly evaluating the utility gain given a benefit improvement, i.e., . Let  represent respondents’ mean relative utility for an initial benefit  level, e.g., 20%, and let level, e.g., 100%. Thus,  represent respondents’ mean relative utility for a final benefit is the utility gain going from a 20% to a 100%  chance of benefit. In DCEs with multiple benefit attributes, j is the number of benefit attributes. For example, if j = 2 the benefit could be a chance of benefit in functional improvement.  The disutility going from 0%, to the MAR level must offset the utility given a benefit improvement. Thus, assuming the MAR of AE i lies between attribute level k-1 and k ,e.g., the MAR of MI lies between risk attribute level 2 and 3 , RA patients’ MAR of AE i, can be estimated by the following (Equation 5.2) :  (5.2)  where  and  represents the risk at level k and k-1 respectively. For example, the  risk of MI in this DCE has 3 levels, and at k = 2, the risk is 2%. Fraction,  , is  determined through piecewise linear interpolation between respondents’ mean relative utility at two discrete risk levels. The above equation (5.2) is only valid if k is one of the 1 to M risk attribute levels included in the DCE. Thus, since each risk attribute in this DCE has 3 levels, i.e., M=3 for each risk attribute, k must be 1, 2, or 3 for all the risk attributes. However, if the utility given a benefit improvement is greater than the disutility 68  going from 1 to M levels of a risk attribute, than the MAR of AE i is estimated to be greater than the risk level M. For this DCE, the risk of ulcers, dyspepsia, MI and stroke at M are 5%, 20%, 4%, and 4% respectively. Thus if the utility gian for a benefit is greater than the disutility difference going from a 0% to a 5% risk of ulcer, a 0% to a 20% risk of dyspepsia, and a 0% to a 4% risk of MI and stroke, then RA patients’ MAR of ulcers, dyspepsia, MI and stroke is estimated to be greater than their risk at M.  In addition to estimating RA patients’ MAR, patients’ overall preference for treatments with specific benefit-risk profiles can be estimated. The log-likelihoods of the MNL and MXL model were compared to determine the best choice model to estimate patients’ preferences for different NSAIDs. To estimated patients’ overall preference for a treatment, the mean relative preference weights for each attribute level are assumed to be additive. For example, patients’ preference for a treatment with a 50% chance of benefit and a 2% risk of ulcer can be determined by summing patients’ mean relative utility at that benefit and risk level. Rheumatoid arthritis patients’ overall preference for a coxib and a NSNSAID were estimated by summing respondents’ relative mean utilities for levels of benefit and risks associated with a coxib and a NSNSAID. Because there is no substantial evidence to prove that coxibs give those with RA greater benefit than NSNSAIDs, the benefits of coxibs and NSNSAIDs were assumed to be the same for overall preference estimation. Thus, coxibs and NSNSAIDs were arbitrarily given a 50% chance of benefit with moderate levels of functional improvement. Congruent with past literature on the risks of ulcers, dyspepsia, MI and stroke in those with RA, coxibs were assumed to have greater risk of MI and stroke and less risk of GI events than NSNSAIDs.  69  5.4 Results  5.4.1 Sample characteristics Of the 181 respondents who completed the DCE questionnaire, 129 (66.5%) were recruited through the MPAP, and 65 (33.5%) through ACE (Figure 5.1). In addition, 13 (6.7%) respondents answered the two dominant strategy questions incorrectly and were excluded from analysis; this error rate is consistent with previous DCE studies (19,25,26). There were no significant differences in socio-demographic and clinical characteristics between those recruited from MPAP versus ACE or between those who answered the dominant questions correctly and incorrectly. A summary of socio-demographic and clinical characteristics is given in Table 5.2. The mean age of respondents was 57 (12.8) years, with a mean age of diagnosis at 45 (14.7). Most of these respondents were well educated, with more than 50% having completed college or received an undergraduate university degree. In addition, compared to the general RA population, a vast majority of respondents had better control of their RA symptoms as HAQ scores were 1 (0.75), and 69% of respondents had been on biologic DMARDS in the past 6 months.  5.4.2 Model 1 Respondents’ mean relative preference weights elicited from the MNL model are summarized in Table 5.3. All the signs of the regression coefficients are in the a priori hypothesized direction, i.e., negative for greater risk and positive for greater benefit. Thus, negative coefficients indicate that relative to the other levels of each risk attribute, respondents are averse to a 5% risk of ulcer (-0.07), a 10% and 20% risk of dyspepsia 70  (-0.03 and -0.16 respectively), and a 4% risk of myocardial infarction (MI) and stroke (0.38 and -0.52 respectively). Relative to a 0% risk, respondents were most averse to a 4% risk of MI and stroke and least averse to a 20% risk of dyspepsia.  For benefit attributes, positive coefficients indicated that relative to a 20% chance of benefit, respondents preferred a treatment that would work at least 50% (0.02) of the time. In addition, a treatment that offered moderate (0.27) or a lot of improvement in function (0.53) was preferred relative to one that offered a little bit of functional improvement. Respondents’ preferences of treatments were significantly affected by the interaction of the benefit attributes, ‘chance of benefit’ and ‘level of improved function’. However, the inclusion of interaction terms into the MNL model did not significantly improve the fit of model, and therefore these terms were excluded from the final analysis.  5.4.3 Model 2 The estimated standard deviations (SD) in the MXL model are statistically significant for a risk of dyspepsia, MI and stroke that is greater than 0%. This indicated that there is heterogeneity in respondent’s preferences at these risk levels. In contrast, no significant variation in respondents’ preferences was found for levels of a risk of an ulcer, chance of benefit, or level of improved function. Therefore, keeping benefit attributes and risk of ulcer as fixed covariates while specifying normal distributions for risk of dyspepsia, MI and stroke greater than 0% , respondents’ mean relative preference weights for each attribute level using the MXL model are illustrated in Table 5.4. 71  All mean relative preference weights from the MXL model were statistically significant. This indicated that respondents had a significant preference for each attribute level. The signs and gradients of respondents’ relative preferences were similar to MNL. Thus relative to a 0% risk, respondents were still most averse to a 4% risk of MI and stroke (-0.61 and -0.83 respectively). In addition, treatments with at least 50% chance of giving a benefit relative to 20% (0.03) and moderate or a lot of functional improvement (0.33 and 0.83 respectively) were preferred relative to the other levels within each benefit attribute. The magnitudes of the mean relative preference weights for each attribute level from the MXL model are larger than those from the MNL model. However, this is expected as the MXL model incorporates an added error term for variation in respondents’ preferences due to unobservable factors.  Figures illustrating the distribution of preference heterogeneity for a 10% and 20% risk of dyspepsia as well as a 2% and 4% risk of MI and stroke are presented in 5.2, 5.3, and 5.4, respectively. There is greater preference heterogeneity in respondents for a greater risk of dyspepsia, MI, and stroke. Past and current treatment medication, i.e., whether RA patients had been on coxibs, non-selective NSAIDs, DMARDs, or biologics was not identified by segmentation analysis to be significant determinants of this heterogeneity. However, respondents differed in their preference for a 4% risk of MI based on their number of tender joints (Figure 5.5) and HAQ score (Figure 5.7). In addition, the number of tender joints and respondents’ SF-6D score (Figure 5.6) was also found to significantly influence respondents’ preferences for a 20% risk of dyspepsia. Those with greater joint pain tended to place less importance on a 20% risk of dyspepsia and a 4% risk of MI relative to those with less joint pain. Given a greater number of tender joints and poorer self-reported health status, respondents were also 72  less averse to a 20% risk of dyspepsia relative to those with a less number of tender joints and better health status respectively. When respondents had given greater disability, indicated by their HAQ scores, they were more tolerant of a 4% risk of MI compared to those with less disability.  5.4.4 Maximum acceptable risk Maximal acceptable risk of AE i is the risk level at which the utility difference from a 0% risk to the MAR offsets respondents utility for a benefit. If respondents’ utility for a benefit is greater than the utility difference from a 0% to a 5% risk of ulcer, a 0% to a 20% risk of dyspepsia, and a 0% to a 4% risk of MI and stroke, then RA patients’ MAR of ulcers, dyspepsia, MI and stroke is estimated to be greater than 5%, 20%, 4% and 4% respectively. For this DCE, the differences in utility given a benefit or an increase in risk are presented in Table 5.5.  As can be seen respondents’ utility for a 20% to 100% (0.98) chance of benefit as well as an increase in functional ability from a little bit to moderate (1.06) or a little bit to a lot (1.33), is greater than the utility difference from a 0% to a 5% risk of ulcer (0.09), a 0% to a 20% risk of dyspepsia (0.34), and a 0% to a 4% risk of MI ( 0.71) and stroke (0.90). Thus, respondents’ MAR of ulcers, dyspepsia, MI and stroke is greater than their risk at level M. In comparison, the utility difference of a 20% to a 50% chance of benefit (0.52) can be offset by the difference in utility given a 0% risk to a 4% risk of MI (0.71) and stroke (0.90). Because respondents’ utility for a chance of benefit and level of functional improvement was generally greater than their utility going from risk levels 1 to M, RA patients’ MAR of ulcers, dyspepsia, MI and stroke was only estimated for each 73  benefit attribute and not a combination of benefit attributes, e.g., patients’ MAR given a 20% chance of benefit and a little bit of improved function was not estimated. This is because the risk of MAR would need to be substantially greater than the risk at level M to offset respondents’ utility for a benefit.  Tables 5.6 presents the estimated MAR of ulcers, dyspepsia, MI and stroke given a greater chance of benefit. When the chance of experiencing a benefit increases from 20% to 50%, respondents were willing to accept a risk of ulcer and dyspepsia greater than 5% and 20% respectively. In comparison, respondents were less willing to accept a risk of MI and stroke, given a 20% to a 50% chance of benefit, as MAR of MI is 3% (1%) and 2.8% (0.8%) for stroke. The MAR of MI and stroke increases to greater than 4% when the chance of benefit increases from 20% to 100%. When the benefit is a little bit to moderate or a lot of functional improvement, respondents’ are also willing to accept a risk of MI and stroke that is greater than 4% (Table 5.7). Given a greater chance of benefit or increase in functional ability, respondents had the greatest tolerance for a risk of dyspepsia in comparison to a risk of MI and stroke.  5.4.5 Comparison of preferences for different NSAIDs From the log-likelihoods, it was determined that the MXL model preference estimates are better to estimate patients’ preferences for treatment. Using relative preference weights estimates from the MNL model, an estimate of respondents’ overall preference for coxibs as well as their preference for NSNAIDS was determined. The difference in efficacy between different NSAIDs is not clear. However, even assuming that the benefit from a coxib is the same to that of a NSNSAID (at 50% chance of 74  benefit and moderate functional improvement), there respondents’ preference for coxibs (1.61) is greater than their preference for NSNSAIDs (1.23) (Table 5.8).  5.5 Discussion In this DCE, it was found that RA patients are willing to trade-off the potential risks of their NSAIDS for its potential benefits. This contrast with findings by Frankel and colleagues (30), whom found that despite a potential benefit, RA patients are generally unwilling to accept a risk of side effects (30). However, the results from this study were drawn from a rating scale, which may overestimate how averse patients are to more serious AEs. Furthermore, unlike DCEs, rating scales do not appropriately model real clinical scenarios where patients may need to simultaneously trade off between multiple attributes of different choice options. The external validity of this DCE to elicit RA patients’ preferences is increased by the inclusion of an opt-out option and the use of focus groups to identify key attributes and levels for the DCE.  It is expected that all treatments will have a risk of AEs. Thus, patients with chronic illnesses with no cure may have a greater risk tolerance given a treatment that would provide a benefit they would want. This is true for those with RA, as many have to try different treatments before finding one that gives them a preferred benefit. Additionally, the RA preference for a potential benefit may outweigh their concern for a risk of dyspepsia, MI and stroke if they have a greater number of tender joints and disability as well as poorer health status. This is realistic of real clinical decision-making scenarios where those who are sicker may be more willing to accept an increased risk of specific AEs given a benefit. 75  Thus, given a potential benefit from their coxib, RA patients may prefer taking a coxib relative to a NSNSAID. This is because given a greater chance of benefit with improvement in pain and function, those with RA are willing to accept a risk of ulcer, dyspepsia, MI, and stroke that is greater than the risks that they are exposed to in past and current NSAIDs; the MAR estimates of ulcers and MI and stroke are similar to those elicited by Kopec and colleagues in osteoarthritis (OA) patients using probabilistic threshold technique (TT) (33). In the context of specific AES, RA patients are more accepting of treatments with the less serious risk of dyspepsia than treatments with the more serious risk of ulcers, MI and stroke. This is consistent with past studies that found RA patients’ risk tolerance to differ based on the relative seriousness of the AEs (31,32).  The use of a DCE to elicit patients’ preferences assumes that patients’ choices in the DCE will reflect their real-life treatment decisions. Representativeness can be affected if the DCE is difficult to understand, there are missing key attributes are included, or the sample population is not representative of the true patient population. On average, the sample population in this DCE were better educated and had better disease control than the general RA population. Thus, the preferences elicited may not be entirely correct. However, it is felt this potential limitation is trivial as the estimated MARs are felt to be accurate of real healthcare environments where patients’ risk tolerance for AEs may depend on the relative seriousness of these AEs or their disease severity. Furthermore, as the benefits significantly outweighed the risks with this study population, even if the DCE was re-administered to more patients with poorer health status, it would only be more evident that RA patients’ preference for a benefit overshadows their concern for the risks. 76  5.6 Conclusion It is important to differentiate between patients’ preferences for treatment versus the preferences of other decision-makers, e.g., societal preferences and/or regulators’ preferences. This is because patients are generally more willing to take the risks of adverse events than those who are not ill. From the perspective of RA patients the benefits of NSAIDs outweigh its risks. In particular the MARs of ulcers, dyspepsia, MI, and stroke in those with RA may be substantially greater than the risks of current and past NSAIDs. If true, this may affect their choices of treatments with different benefitrisk profiles, e.g., a coxib may be preferred over a NSNSAID with the same benefit.  This DCE further supports the existing evidence that the voluntarily withdrawal of rofecoxib was a false-negative decision as patients would be potentially willing to accept the CV risks given a benefit (9-12). Thus, given the risk preferences of those with RA it is hoped that regulators may be better able to make better-informed decisions when evaluating new or current treatments. In addition, an insight into RA patients’ preferences for the potential risks and benefits of treatment can aid physicians and patients during clinical decisions among multiple treatment options with varying AEs and benefits. This study has demonstrated the usefulness of DCEs in providing insight into how patients’ preferences for treatment can be influenced in response to changes in important attributes of their treatments. Future directions may include comparing the results of this study with other methods that are being evaluated for use in regulatory decisions by the IOM and CHMP.  77  5.7 Tables Attributes Chance of benefit  Levels 20 out of 100 people will benefit (20%) 50 out of 100 people will benefit (50%) 100 out of 100 people will benefit (100%)  Level of improved function  A little bit Moderate A lot  Risk of ulcers None out of 100 (0%) 2 out of 100 (2%) 5 out of 100 (5%) Risk of dyspepsia None out of 100 (0%) 10 out of 100 (10%) 20 out of 100 (20%) Risk of heart attack  None out of 100 (0%) 2 out of 100 (2%) 4 out of 100 (4%)  Risk of stroke None out of 100(0%) 2 out of 100 (2%) 4 out of 100 (4%) Table 5.1: Summary of attributes and levels in DCE questionnaire  78  Characteristics Age Gender Male Female Smoking Status Never Currently Quit Marital Status Single Married Living with partner Divorced/Separated Widowed Prefer not to answer Ethnicity Caucasian Hispanic First Nations, Aboriginal Black East Indian Asian Education Some high school Technical/Trade/Vocational College Community College Some university University degree obtained Graduate or Post-graduate training Age of Diagnosis Age of 1st Symptoms RA-related hospitalizations VAS Pain VAS Global Well Being Tender Joint Count SF6D index HAQ Score Past 6 Month Medication History NSAIDs DMARDs Biologics  Mean (SD) N (%) N (%) N (%) N (%) N (%)  57 28 28 138  N= 181 (12.8) (17%) (17%) (83%)  71 (43%) 18 (18%) 76 (46%)  N (%) N (%) N (%) N (%) N (%) N (%)  16 103 11 22 12 2  (10%) (62%) (7%) (13%) (7%) (1%)  N (%) N (%) N (%) N (%) N (%) N (%)  147 1 6 3 10 5  (89%) (1%) (4%) (2%) (6%) (3%)  N (%) N (%) N (%) N (%) N (%) N(%) Mean(SD) Mean(SD) Median (Range) Median (Range) Median (Range) Median (Range) Mean(SD) Mean(SD)  10 24 21 30 25 36 45 42 0 31.5 31 25 0.6 1  (6%) (14%) (13%) (18%) (15%) (22%) (14.7) (14.7) (0-2) (0-90) (0-99) (0-48) (0.37) (0.75)  Frequency(%) Frequency(%) Frequency(%)  39% 85% 69%  Table 5.2: Summary of demographic and clinical characteristics of respondents  79  Parameters Chance of Benefit 20% 50% 100% Level of Improved Function A little bit Moderate Complete Ulcer Risk 0% 2% 5% Dyspepsia Risk 0% 10% 20% MI Risk 0% 2% 4% Stroke Risk 0% 2% 4%  Estimate  Std Err -0.50 (0.06)* 0.02 (0.05) 0.48 (0.06)* -0.80 (0.07)* 0.27 (0.05)* 0.53 (0.06)* 0.03 (0.07) 0.04 (0.05) -0.07 (0.06) 0.18 (0.06)* -0.02 (0.05) -0.16 (0.06)* 0.34 (0.06)* 0.03 (0.05) -0.37 (0.06)* 0.39 (0.06)* 0.13 (0.05)* -0.52 (0.06)*  Table 5.3: Mean relative preference weights for MNL model * Mean relative preference weights are statistically significant from zero  80  Parameters Chance of Benefit 20% 50% 100% Level of Improved Function A little bit Moderate Complete Ulcer Risk 0% 2% 5% Dyspepsia Risk 0% 10% 20% MI Risk 0% 2% 4% Stroke Risk 0% 2% 4%  Estimate  Std Err -0.80 (0.08)* 0.03 (0.06) 0.77 (0.08)* -1.15 (0.09)* 0.32 (0.06)* 0.83 (0.08)* 0.15 (0.08)* 0.06 (0.06) -0.21 (0.08)* 0.31 (0.09)* -0.01 (0.06) -0.30 (0.08)* 0.55 (0.08)* 0.06 (0.06) -0.61 (0.08)* 0.66 (0.08)* 0.17 (0.06)* -0.83 (0.09)*  Table 5.4: Mean relative preference weights for MXL model *Mean relative preference weights are statistically significant from zero  81  Benefit Attribute  Benefit Improvement  Utility Gain  Risk Attribute  Disutility going from 1 to M levels of a risk attribute  Chance of benefit  20%- 50%  0.52  0.09  20%-100%  0.98  Risk of ulcer (0%-5%) Risk of dyspepsia (0%-20%)  A little bitmoderate  1.06  Risk of MI (0%-4%)  0.71  A little bit- a lot  1.33  Risk of stroke (0%4%)  0.90  Level of improved function  0.34  Table 5.5 Comparison of the utility gain given a benefit versus the disutility given greater risk  82  Initial Health State  Final Health State  Ulcer Risk  Dyspepsia Risk  MI Risk (95% CI)  Stroke Risk (95% CI)  20%  50%  >5%  >20%  3% (1.0%5.0%)  2.8% (1.2%4.4%)  20%  100%  >5%  >20%  >4%  >4%  Table 5.6: MAR for increased chance of benefit  Initial Health State  Final Health State  Ulcer Risk (SE)  Dyspepsia Risk (SE)  MI Risk (SE)  Stroke Risk (SE)  A little bit  Moderate  >5%  >20%  >4%  >4%  A little bit  A lot  >5%  >20%  >4%  >4%  Table 5.7: MAR for greater improvement in function  83  Coxib  Non-selective NSAID (NSNSAID)  Treatment Attributes Benefit  SAME BENEFIT 50% chance of benefit with moderate functional improvement  Ulcer Risk  LOWER  HIGHER  Dyspepsia Risk  LOWER  HIGHER  MI Risk  HIGHER  LOWER  Stroke Risk  HIGHER  LOWER  1.61  1.30  Mean Overall Utility for Treatment  Table 5.8: Comparison of preference profiles of NSAIDs with similar benefits and different risks  84  5.8 Figures  N=167  N=96  MPAP  ACE  N=263 Total  Dropout N=69 (MPAP=38) (ACE=31)  Failed Dominance N=13 (MPAP=7) (ACE=6)  N=194  Total For Analysis N=181  Answered Neither To All N=6 (MPAP=5) (ACE=1)  Figure 5.1: Flowchart of respondent recruitment results 85  Figure 5.2: Preference distribution for a risk of dyspepsia  86  Figure 5.3: Preference distribution for a risk of MI  87  Figure 5.4: Preference distribution for a risk of stroke  88  Figure 5.5: Relative mean preference weights by number of tender joints * Indicates significant preference heterogeneity  89  Figure 5.6: Relative mean preference weights by SF6D score * Indicates significant preference heterogeneity  90  Figure 5.7: Relative mean preference weights by HAQ score *Indicates significant preference heterogeneity  91  Chapter 6: Summary and conclusions of thesis  6.1 Introduction Traditionally drug approval or withdrawal decisions are based on the risks of adverse events (AEs) and not on benefits or patients’ preferences for the benefits. However, a disregard of patients’ preferences can result in false negative decisions where treatments that patients are willing to take are withdrawn from the market due to safety concerns. A false-negative decision was demonstrated when the drug alosetron (Lotronex) was re-approved after consideration of patients’ preferences for the benefits and risks. To prevent patients from losing access to potentially beneficial treatments, the US Institute of Medicine (IOM) and the Committee for Medicinal Products for Human Use (CHMP), of the European Medicines Agency (EMA), are evaluating different methods of benefit-risk analysis for aiding in pre- and post-approval processes of drug regulatory decisions (1,2). One of these methods is the stated choice technique. The discrete choice experiment (DCE) is a stated choice technique to evaluate patients’ preferences for their healthcare attributes. This method has been used to evaluate the risk preferences of those with rheumatoid arthritis (RA).  Rheumatoid arthritis is a chronic autoimmune disease. Those afflicted with RA may experience severe disability, fatigue, and chronic pain (3-5). Current treatment strategies for RA management include taking a combination of different types of drugs such as analgesics, corticosteroids, disease modifying anti-rheumatic drugs (DMARDs), biologic DMARDs, and non-steroidal anti-inflammatory drugs (NSAIDs) (6). Nonsteroidal anti-inflammatory drugs are often taken in the beginning of RA management to 92  reduce pain and inflammation in the joint, while giving time for drugs such as DMARDs to slow long-term disease progression (6). The anti-inflammatory properties of NSAIDs are due to their inhibition of the cyclo-oxygenase (COX) enzyme (7,8). There are two isoenzymes, or forms of the COX enzyme that are important in the context of NSAIDs: COX-1 and COX-2 (8). The NSAIDs that non-selectivity inhibits both forms of the COX enzyme are called non-selective NSAIDS (NSNSAIDs), whereas NSAIDs that selectively inhibits the COX-2 isoenzyme to a greater degree are called COX-2 inhibitors or coxibs. An example of a coxib that has been available for those with RA is rofecoxib (Vioxx®).  Rofecoxib was voluntarily withdrawn due to evidence of a potential increased risk of cardiovascular events (CVs) relative to placebo (9,10). Although there is evidence that from the patients’ perspective the benefits of rofecoxib outweigh its risks (11-15), this had not been empirically demonstrated. Therefore, the purpose of this thesis was to quantify RA patients’ risk preferences for ulcers, dyspepsia, myocardial infarction (MI), and stroke given a chance of experiencing a level of functional improvement. This chapter discusses the key findings and potential implications of the thesis.  6.2 Discussion of key findings from thesis Past studies on RA patients’ preferences have identified those with RA to be a risk averse population that would like full disclosure of any information regarding their treatment risks and benefits (13,14). However, there is evidence to suggest that given greater potential benefit, those with RA may be willing to accept greater risk of specific AEs(5-7). Past RA preference literature are potentially limited in that the attributes 93  included for trade-off may not be meaningful to patients. For example, the importance of an improvement in RA symptoms could vary between patients. Thus, prior to conducting preference studies it is important to try to include attribute and levels that are meaningful to patients.  Another potential limitation of past RA preference studies are the methods used to elicit patients’ preferences. Previous methods for eliciting RA patients’ preferences have included qualitative methods, such as focus groups or questionnaires, or quantitative methods such as standard gamble (SG) or probability trade-off techniques. Qualitative methods are useful for determining the healthcare attributes that are most important to patients but they are not based on random utility theory. Thus, they do not evaluate patients’ utilities for healthcare attributes. In comparison, although SG and TTO allow the valuation of the importance of healthcare attributes to patients, they do so by only requiring that patients trade-off between two attributes at a time, i.e., a risk of AE for a benefit. Thus, they may not appropriately mimic real clinical scenarios where RA patients may need to trade-off between different treatments with multiple treatment attributes. Discrete choice experimentation is a stated preference technique that elicits patients’ preferences by requiring them to choose between healthcare options that may vary in multiple attributes at a time.  An online version of the DCE questionnaire was administered to those with RA to determine patients’ preferences for the attributes, i.e., risks and benefits, of NSAIDs. To be more congruent with a real-market environment where patients are not forced to choose between two potentially unappealing alternatives, an opt-out option was included as a third choice alternative in each DCE choice-set. Appropriate attributes for 94  inclusion in the DCE were determined through focus groups that asked RA patients what risks and benefits of treatment were important in influencing their treatment decisions.  In the focus groups, patients identified that they were averse to treatments with a risk of ulcer, dyspepsia, MI and stroke. In terms of benefits of treatment, RA patients wanted pain reduction and function improvement from their NSAIDs. However, an interesting finding was that those with RA might view benefits of pain reduction and function improvement to be analogous. For example if told a treatment gave them pain reduction, they may see it as improved function in valued daily activities. Thus, to prevent redundancy of attributes in the DCE questionnaire, a benefit of pain reduction and improved function was combined into one attribute called the level of function improvement.  From focus groups, six attributes were included in this DCE: (1) chance of benefit or chance that the drug will work, (2) level of improved function, and risks of (3) ulcers, (4) dyspepsia, (5) myocardial infarction (MI) and (6) stroke. The attribute ‘chance of benefit’ was included as it is more realistic of treatment scenarios where patients are not guaranteed that their treatment will work to give them a benefit. To potentially avoid bias in the preferences elicited, each attribute was accompanied with a detailed description of the outcomes associated with it. For example, what it means to have an ulcer was described. Each attribute level was created so that the level ranges captured the estimates/values that are commonly associated with that attribute. The DCE questionnaire also was piloted tested in those with RA to potentially prevent any missing key attributes as well as assess the meaningfulness of the levels to those with 95  RA. Pilot tests also helped to identify if patients had any problems understanding what each attribute entailed given the descriptions that accompanied each attribute.  Given the attribute levels for trade-off, it was found that with greater chance of improvement in function and pain those with RA are willing to accept greater risks of treatment-related ulcers, dyspepsia, MI, and stroke. Thus, RA patients are willing to accept greater risk of AEs given greater benefit. Concurrent with past RA patients’ preference literature (16,17), the RA risk tolerance depends on the seriousness of individual AEs. Thus, those with RA are most willing to accept a risk of dyspepsia compared to a risk of MI and stroke. In addition, RA patients that have a greater number of tender joints, RA-related disability, and poorer self-reported health status may be more risk tolerant than those who are healthier.  Rheumatoid arthritis patients’ preference for greater benefit may outweighs their concern for treatment-related AEs. Thus, their concern for the current CV risks of coxibs may not be enough to offset their utility for the benefits of coxibs. In particular, despite receiving the same benefit from a coxib and a NSNSAID, RA patients may still prefer taking a coxib to a NSNSAID. This may because RA patients’ estimated maximum acceptable risk (MAR) of ulcers, dyspepsia, MI and stroke is greater than the risks of these AES in current and past NSAIDs.  These results should be interpreted with two main issues in mind. Firstly, respondents’ mean relative preference weights for each attribute level are centered on zero. Therefore if two attribute levels have positive signs, the reference level has to be negative. However, this is probably not a large limitation as all the signs of benefit and 96  risk attributes occur in the a priori hypothesized direction, i.e., positive signs for greater benefit and negative for greater risk.  Another potential limitation is if the preferences elicited are not representative of the true patient population as on average, respondents tended to be better educated and have better disease control than those of the general RA population. However, the variation in RA patients’ estimated MAR of ulcers, dyspepsia, MI and stroke is accurate of real healthcare environments where patients’ are more willing to accept risk of less serious AEs compared to more serious ones.  6.3 Conclusion The use of coxibs is not limited to those with RA. However, RA patients may be more willing to accept a risk of specific AEs given the benefit. This is because those with RA may have to try many different treatments before finding one that gives them a preferred benefit. This was true in the case of the voluntary withdrawal of rofecoxib, where some RA patients lost access to a treatment that worked for them. This study shows that from RA perspective, the benefits of NSAIDs outweigh their risks. Thus, the decision to voluntarily withdraw rofecoxib is a false-negative one as RA patients’ MAR of ulcers, dyspepsia, MI and stroke is greater than the risks that they are exposed to in current and past NSAIDs.  The use of a DCE to evaluate the RA patients’ preferences is advantageous in that real clinical decision scenarios can be replicated by including attributes that influence patients’ treatment decisions. In addition, the levels of risks and benefits 97  included in a DCE can be outside what patients may be exposed to in their current treatments. Thus, although this study is not able to change past decisions, such as the voluntary withdrawal of rofecoxib, it is hoped that by quantifying RA patients’ risk tolerance of AEs given a benefit, the occurrence of future false negative decisions could be potentially avoided. As a result, regulators may be better able to make betterinformed decisions on newer treatments with potential safety concerns.  An understanding of RA patients’ risk preferences could also be important for at the clinical decision-making level. Going back to the rofecoxib example, there is evidence to suggest that patients are not at an increased CV risk for the first 18 months of taking rofecoxib relative to placebo (10). Therefore, given RA patients’ MAR of a MI is greater than the current risks of rofecoxib, physicians may choose to let some RA patients try rofecoxib for a short period of time to see if it gives them a benefit they want. If patients experience a preferred benefit, then they may be allowed to continue taking this drug as research suggests RA patients’ risk tolerance is greater than the current CV risks of rofecoxib. Alternatively, if patients do not experience a preferred benefit, they may then be taken off the treatment before they are put at a substantial risk of MI. Thus, given insight into patients’ risk preferences of patients can be helpful for physicians and patients when they are making clinical decisions among multiple treatment strategies.  There is clearly a need for methods of benefit-risk analysis for more informed drug approval and withdrawal decisions. Discrete choice experimentation is only one of the methods being evaluated by the IOM and the CHMP of the EMA. Other methods being evaluated for use are the incremental net-benefit analysis and multi criteria 98  decision analysis (MCDA). Thus, future directions could include repeating this study using other methods for risk-benefit analysis to compare their uses in eliciting RA patients’ preferences for their healthcare options.  99  References (1) Smolen J, Steiner G. Therapeutic strategies for rheumatoid arthritis. Nat Rev Drug Discov 2003;2(6):473-88. (2) Hewlett S, Hehir M, Kirwan JR. Measuring fatigue in rheumatoid arthritis: a systematic review of scales in use. Arthritis Rheum 2007;57(3):429-39. (3) Pollard LC, Choy EH, Gonzalez J, Khoshaba B, Scott DL. Fatigue in rheumatoid arthritis reflects pain, not disease activity. Rheumatology 2006;45(7):885-9. (4) Markenson JA. Worldwide trends in the socioeconomic impact and long-term prognosis of rheumatoid arthritis. Semin Arthritis Rheum 1991;2(1):4-12. (5) Goldring S,. Pathogenesis of bone and cartilage destruction in rheumatoid arthritis. Rheumatology 2003;42(2):ii11-6. (6) WHO. Global burden of rheumatoid arthritis in the year 2000. [Online]. 2000 [cited 2010 Jan 20]. Available from: http:/ / www3.who.int/ whosis/ menu.cfm?path=evidence,burden,burde n_gbd2000docs&language=english. (7) Carr A, Hewlett S, Hughes R, Mitchell H, Ryan S, Carr M, et al. Rheumatology outcomes: the patient's perspective. J Rheumatol 2003;30(4):880-3. (8) Kvien T. Epidemiology and burden of illness of rheumatoid arthritis. Pharmacoeconomics 2004;22(1):1-12. (9) Solomon DH, Karlson EW, Rimm EB, Cannuscio CC, Mandl LA, Manson JE, et al. Cardiovascular morbidity and mortality in women diagnosed with rheumatoid arthritis. Circulation 2003;107(9):1303-7. (10) Naz S. Mortality in established rheumatoid arthritis. Best Practice & Research Clinical Rheumatology 2007;21(5):871-83. (11) Wolfe F, Mitchell D. The mortality of rheumatoid arthritis. Arthritis Rheum 1994;37(4):481-94. (12) Pugner, K., Scott, D., Holmes, J., Hieke, K. The costs of rheumatoid arthritis: an international long-term view. Semin Arthritis Rheum 2000;29(5):305-20. (13) Cooper NJ. Economic burden of rheumatoid arthritis: a systematic review. Rheumatology 2000;39(1):28-33. (14) Kavanaugh A. Health economics: implications for novel antirheumatic therapies. Ann Rheum Dis 2005;64(4):iv65-9. (15) Clarke AE, Zowall H, Levinton C, Assimakopoulos H, Sibley JT, Haga M, et al. Direct and indirect medical costs incurred by Canadian patients with rheumatoid arthritis: a 12 year study. J Rheumatol 1997;24(6):1051-60. 100  (16) Bansback NJ, Regier DA, Ara R, Brennan A, Shojania K, Esdaile JM, et al. An overview of economic evaluations for drugs used in rheumatoid arthritis: focus on tumour necrosis factor-alpha antagonists. Drugs 2005;65(4):473-96. (17) Bombardier C, Laine L, Reicin A, Shapiro D, Burgos-Vargas R, Davis B, et al. Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. N Engl J Med 2000;343(21):1520-8. (18) Wright JM. The double-edged sword of COX-2 selective NSAIDs. CMAJ 2002 ;167(10):1131-7. (19) Silverstein FE, Faich G, Goldstein JL, Simon LS, Pincus T, Whelton A, et al. Gastrointestinal toxicity with celecoxib vs nonsteroidal anti-inflammatory drugs for osteoarthritis and rheumatoid arthritis: the CLASS study a randomized controlled trial. JAMA 2000;284(10):1247-55. (20) Hawkey CJ, Langman MJS. Non-steroidal anti-inflammatory drugs: overall risks and management. Complementary roles for COX-2 inhibitors and proton pump inhibitors. Gut 2003;52(4):600-8. (21) Hunt RH, Choquette D, Craig BN, De Angelis C, Habal F, Fulthorpe G, et al. Approach to managing musculoskeletal pain: acetaminophen, cyclooxygenase-2 inhibitors, or traditional NSAIDs? Can Fam Physician 2007;53(7):1177-84. (22) Jaksch W, Dejaco C, Schirmer M. 4 Years after withdrawal of rofecoxib: where do we stand today? Rheumatol Int 2008;28(12):1187-95. (23) Rahme E, Nedjar H. Risks and benefits of COX-2 inhibitors vs non-selective NSAIDs: does their cardiovascular risk exceed their gastrointestinal benefit? A retrospective cohort study. Rheumatology 2007;46(3):435-8. (24) Wright V. Historical overview of non-steroidal anti-inflammatory drugs. Br J Rheumatol 1995;34(1):2-4. (25) Cannon CP, Curtis SP, FitzGerald GA, Krum H, Kaur A, Bolognese JA, et al. Cardiovascular outcomes with etoricoxib and diclofenac in patients with osteoarthritis and rheumatoid arthritis in the multinational etoricoxib and diclofenac arthritis long-term (MEDAL) programme: a randomised comparison. Lancet 2006;368(9549):1771-81. (26) Laine L, Bombardier C, Hawkey CJ, Davis B, Shapiro D, Brett C, et al. Stratifying the risk of NSAID-related upper gastrointestinal clinical events: results of a double-blind outcomes study in patients with rheumatoid arthritis. Gastroenterology 2002;123(4):1006-12. (27) Mamdani M, Rochon R, Juurlink DN, Kopp A, Anderson GM, Naglie G, et al. Observational study of upper gastrointestinal hemmorhage in elderly patients given cyclo-oxygenase-2 inhibitors or conventional non-steroidal anti-inflammatory drugs. Br Med J 2002;324(7365):1-6. 101  (28) Bresalier RS, Sandler RS, Quan H, Bolognese JA, Oxenius B, Horgan K, et al. Cardiovascular Events Associated with Rofecoxib in a Colorectal Adenoma Chemoprevention Trial. N Engl J Med 2005;352(11):1092-102. (29) Chustecka Z. The CV risks with coxibs: is it really proven, how big is it, and are traditional NSAIDS any safter in this regard? [Online]. 2005 [cited 2010 March 3]. Available from: http://www.jointandbone.org. (30) Hughes S. CV risks of anti-inflammatories discussed at cardiology meeting. [Online]. 2005. [cited 2010 March 3]. Available from: http://wwww.jointandbone.org/ printArticle.do?primaryKey=404911. (31) Tannenbaum H. Patient empowerment: rofecoxib revisited. J Rheumatol 2006;33(1):1033-66. (32) Heim HK, Broich K. Selective cox-2 inhibitors and risk of thromboembolic eventsregulatory aspects. J Thromb Haemost 2006;96:423-32. (33) Health Canada. Report on the cardiovascular risks associated with cox-2 inhibitors selective non-steroidal anti-inflammatory drugs. [Online]. 2006 [cited 2010 June 5]. Available from: http://www.hs-sc.gc.ca/dhp-mps/prodpharma/activit/sciconsult/cox2/rofecoxib_benefits-avantages-eng.php. (34) Shiel WC. Cox-2 inhibitors-what's next. [Online]. 2005 [cited 2010 June 5]. Available from: http://www.medicinenet.com/script/main/art.asp?articlekey=46145. (35) Lynd LD, Marra CA, Najafadeh M, Sadatsafavi M. A quantitative evaluation of the regulatory assessment of the benefits and risks of rofecoxib relative to naproxen: an application of the incremental net-benefit framework. Pharmacoepidemiol Drug Saf [Online]. 2010 [cited 2010 August 10]. Available from: doi: 10.1002/pds.1994. (36) Blair A, Saracci R, Vineis P, Cocco P, Forastiere F, Grandjean P, et al. Epidemiology, public health, and the rhetoric of false positives. Environ Health Perspect 2009;117(12):1809-13. (37) Garrison LP, Towse A, Bresnahan B. Assessing a structured quantitative health outcomes approach to drug risk-benefit analysis. Tools Incentives 2007;26(3):684-95. (38) European medicines agency. Report on benefit-risk assessment models and methods. [Online]. 2008 [cited 2010 Sept 10]. Available from: http://www.ema.europa.eu/docs/ en_GB/document_library/Regulatory_and_procedural_guideline/2010/01/WC50006966 8.pdf. (39) Bowling A, Ebrahim S. Measuring patients' preferences for treatment and perceptions of risk. Qual Health Care 2001;10(90001):i2-8. (40) Streiner DL, Norman GR. Health measurements scale-a practical guide to their development and use. 4th ed. New York: Oxford University Press; 2008. 102  (41) Stiggelbout AM, de Haes JCJM. Patient preference for cancer therapy: an overview of measurement approaches. J of Clin Oncol 2001;19(1):220-30. (42) Lane DA, Hutchinson TA. The notion of 'acceptable risk': the role of utility in drug management. J Chronic Dis 1987;40(6):621-5. (43) Bridges J. Stated preference methods in health care evaluation: an emerging methodological paradigm in health economics. Appl Health Econ Health 2003;2(4):21324. (44) Kopec JA, Richardson CG, Llewellyn-Thomas H, Klinkhoff A, Carswell A, Chalmers A. Probabilistic threshold technique showed that patients' preferences for specific tradeoffs between pain relief and each side effect of treatment in osteoarthritis varied. J Clin Epidemiol 2007;60(9):929-38. (45) Petrou S, Henderson J. Preference-based approaches to measuring the benefits of perinatal care. Birth 2003;30(4):217-25. (46) Torrance GW. Measurement of health state utilities for economic appraisal. J Health Econ 1986;5(1):1-30. (47) Llewellyn-Thomas H, Sutherland HJ, Tibshirani R, Ciampi A, Till JE, Boyd NF. The measurement of patients' values in medicine. Medical Decis Making 1982;2(4):449-62. (48) Read JL, Quinn RJ, Berwick DM, Fineberg HV, Weinstein MC. Preferences for health outcomes-comparison of assessment methods. Medical Decis Making 1984;4(3):315-29. (49) Louviere J. Conjoint analysis modeling of stated preferences. A review of theory, methods, recent developments, and external validity. J Transport Econo Policy 1998;22(1):93-119. (50) Maddala T, Phillips KA, Johnson FR. An experiment on simplifying conjoint analysis designs for measuring preferences. Health Econ 2003;12(12):1035-47. (51) Phillips KA, Maddala T, Johnson R. Measuring preferences for health care interventions using conjoint analysis: An application to HIV testing. Health Serv Res 2002;37(6):1681-705. (52) Ryan M, Farrar S. Using conjoint analysis to elicit preferences for health care. Br Med J 2000;320(7248):1530-3. (53) Ryan M,. Using conjoint analysis to take account of patient preferences and go beyond health outcomes: an application to in vitro fertilization. Soc Sci Med 1999;48(1):535-46. (54) Kjaer T. A review of the discrete choice experiment - with emphasis on its application in health care [discussion paper]. University of Southern Denmark. 2005. 103  (55) Huber J, Zwerina KB. The importance of utility balance in efficient choice designs. J Mark Res 1996;33(1):307-17. (56) Guimaraes C, Marra CA, Gill S, Meneilly G, Simpson S, Godoy A, et al. Exploring patients' perceptions for insulin therapy in type 2 diabetes: a brazilian and canadian qualitative study. Patient Prefer Adherence 2010;4:171-9. (57) McTaggart-Cowan HM, Shi P, Fitzgerald JM, Anis AH, Kopec JA, Bai TR, et al. An evaluation of patients' willingness to trade symptom-free days for asthma-related treatment risks: a discrete choice experiment. J Asthma 2008;45(8):630-8. (58) Ryan M, Netten A, Skatun D, Smith P. Using discrete choice experiments to estimate a preference-based measure of outcome-An application to social care for older people. J Health Econ 2006;25(5):927-44. (59) Sculpher M, Bryan S, Fry P, de Winter P, Payne H, Emberton M. Patients' preferences for the management of non-metastatic prostate cancer: discrete choice experiment. Br Med J 2004;328(7436):382-90. (60) Johnson FR, Hauber B, Ozdemir S, Lynd L. Quantifying women's stated benefitrisk trade-off preferences for IBS treatment outcomes. Value Health 2010; 13(4):1-6. (61) Johnson FR, Ozdemir S, Mansfield C, Hass S, Miller DW, Siegel CA, et al. Crohn's disease patients' risk-benefit preferences: Serious adverse event risks versus treatment efficacy. Gastroenterology 2007;133(3):769-79. (62) Wolfe F, Michaud K. Resistance of rheumatoid arthritis patients to changing therapy. Arthritis Rheum 2007;56(7):2135-42. (63) Fraenkel L, Bogardus S, Concato J, Felson D. Preference for disclosure of information among patients with rheumatoid arthritis. Arthritis Rheum 2001;45(2):136-9. (64) Schildmann J, Grunke M, Kalden JR, Vollmann J. Information and participation in decision-making about treatment: a qualitative study of the perceptions and preferences of patients with rheumatoid arthritis. J Med Ethics 2008;34(11):775-9. (65) Barlow JH, Cullen LA, Rowe LF. Educational preferences, psychological well-being and self-efficacy among people with rheumatoid arthritis. Patient Educ Couns 2002;46(1):11-9. (66) Wolfe F, Zhao S, Lane N. Preference for nonsteroidal antiinflammatory drugs over acetaminophen by rheumatic disease patients. Arthritis Rheum 2000;43(2):378-85. (67) Goekoop-Ruiterman YP, de Vries-Bouwstra JK, Allaart CF, Kerstens PJ, Grillet BA, de Jager MH, et al. Patient preferences for treatment: report from a randomized comparison of treatment strategies in early rheumatoid arthritis (BeSt trial). Ann Rheum Dis 2007;66(9):1227-32.  104  (68) Bagge E, Traub M, Crotty M, Conaghan PG, Oh E, Brooks PM. Are rheumatoid arthritis patients more willing to accept non-steroidal anti-inflammatory drug treatment risks than osteoarthritis patients? Br J Rheumatol 1997;36(470):2-6. (69) Heiberg T, Kvien TK. Preferences for improved health examined in 1,024 patients with rheumatoid arthritis: pain has highest priority. Arthritis Rheum 2002;47(4):391-7. (70) Ahlmen M, Nordenskiold U, Archenholtz B, Thyberg I, Ronnqvist R, Linden L, et al. Rheumatology outcomes: the patient's perspective. A multicentre focus group interview study of swedish rheumatoid arthritis patients. Rheumatology 2005;44(1):105-10. (71) O'Brien B, Elswood J, Calin A. Willingness to accept risk in the treatment of rheumatic disease. J Epidemiol Community Health 1990;44(3):249-52. (72) Thompson MS. Willingness to pay and accept risks to cure chronic disease. Am J Public Health 1986;76(4):392-6. (73) Ho M, Lavery B, Pullar T. The risk of treatment. A study of rheumatoid arthritis patients' attitudes. Rheumatology 1998;37(4):459-60. (74) Fraenkel L, Bogardus S, Concato J, Felson D. Unwillingness of rheumatoid arthritis patients to risk adverse effects. Rheumatology 2002;41(3):253-61. (75) Pullar T, Wright V, Feely M. What do patients and rheumatologists regard as an 'acceptable' risk in the treatment of rheumatic disease? Br J Rheumatol 1990;29(3):2158. (76) Maximum acceptable risk and healthy-year equivalents in AD. 5th International Pharmacoeconomic conference on Alzheimer's disease; 2008 March 28. New York. (77) Johnson FR. Measuring stated preferences for pharmaceutical risk management: A brief introduction. Research Triangle Park NC: RTI Health Solutions, 2005: p. 1-12. (78) Lynd LD, Najafzadeh M, Colley L, Byrne MF, Willan AR, Sculpher MJ, et al. Using the incremetal benefit framework for quantitative benefit-risk analysis in regulatory decision-making - a case study of alosetron in irritable bowel syndrome. Value Health 2010;13(4): 411-7. (79) Wensing M. Evidence-based patient empowerment. Qual Health Care 2000;9(4):200-1. (80) Wensing M, Elwyn G. Methods for incorporating patients' views in health care. Br Med J 2003;326(7394):877-9. (81) Wensing, M., Jung, P., Mainz, J., Olesen, F., Grol, R. A systematic review of the literature on patient priorities for general practice care. Part 1: description of the research domain. Soc Sci Med 1998;47(10):1573-88.  105  (82) Ryan M, Gerard K. Using discrete choice experiments to value health care programmes: current practice and future research reflections. Appl Health Econ Health. 2003;2(1):55-64. (83) Train K. Discrete Choice Methods with Simulation. Cambridge: Cambridge University Press; 2003. (84) Lancsar E, Louviere J. Conducting discrete choice experiments to inform healthcare decision making: a user's guide. Pharmacoeconomics 2008;26(8):661-77. (85) Ryan M, Major K, Skatun D. Using discrete choice experiments to go beyond clinical outcomes when evaluating clinical practice. J Eval Clin Pract 2005;11(4):328-38. (86) Farrar S, Ryan M, Ross D, Ludbrook A. Using discrete choice modelling in priority setting: an application to clinical service developments. Soc Sci Med 2000;50(1):63-75. (87) Amaya M. None matters: alternative modeling approaches of 'no choice' options in stated choice experiments and application [discussion paper].2003;1-24. (88) Ubach C, Scott A, French F, Awramenko M, Needham G. What do hospital consultants value about their jobs? A discrete choice experiment. Br Med J 2003 28;326(7404):1432-5. (89) Ryan M, Eastmond CJ, Ludbrook A. Use of discrete choice experiments to elicit preferences. Br Med J 2001;10(1):55-60. (90) Ryan M, Skatun D. Modelling non-demanders in choice experiments. Health Econ 2004;13(4):397-402. (91) Bech MG, D. Effects coding in discrete choice experiments. Health Econ 2005 2005;14(1):1079-83. (92) McIntosh E, Ryan M. Using discrete choice experiments to derive welfare estimates for the provision of elective surgery: Implications of discontinuous preferences. J Econ Psychol 2002;23(3):367-82. (93) Johnson RF, Van Houtven G, Ozdemir S, Hass S, White J, Francis G, et al. Multiple sclerosis patients' benefit-risk preferences: serious adverse event risks versus treatment efficacy. J Neurol 2009;256(4):554-62. (94) Ryan M. Discrete choice experiments in health care: NICE should consider using them for patient centered evaluations of technologies [editorial]. Br Med J 2004;328:360-1. (95) Grindrod KA, Marra CA, Colley LA, Tsuyuki RT, Lynd L. Pharmacists' preferences for providing patient-centered services: a discrete choice experiment to guide health policy. Ann Pharmacother 2010; 44(10):1554-64.  106  (96) Lagakos SW. Time-to-event analyses for long-term treatments--the APPROVe trial. N Engl J Med 2006;355(2):113-7. (97) Fogoros R. What the FDA panel really said about cox-2 drugs. [Online]. 2005 [cited 2010 June 15]. Available from: http://heartdisease.about.com/od/ otherriskfactors/a/cox22.htm.  107  Appendix 1: The discrete choice experiment questionnaire  The following pages contain the entire paper-based version of the online DCE questionnaire that those with RA completed. Included are 12 DCE choice-sets. As described in Chapter 4, with the number of attributes and attribute levels included in the final DCE we could have used 729 possible choice-sets. To reduce this number, a fractional factorial design was employed that used only 200 choice-sets divided into 20 different versions (10 choice-sets per version). In addition, we included two fixed choice-set questions created with the dominant strategy technique (i.e., questions 10 and 10) to test respondent’s understanding of the questionnaire. Therefore, the DCE questionnaire-survey included in this appendix includes only one version of the DCE  108  BACKGROUND INFORMATION The main goal of this study is to see how much potential risk those with Rheumatoid Arthritis (RA) are willing to accept in response to levels of potential benefit in their RA treatment. This will be achieved using a questionnaire-based survey. There are 12 choice-set questions in this survey. In each choice-set, you are presented with two treatment options (called Option A and Option B). Each treatment option is made up of 6 different treatment characteristics: (1)number of people who will benefit, (2)level of improved function in most valued daily activities, and number of people who will have (3) stomach bleeds, (4)heartburn, (5)heart attacks, or (6)stroke. Treatment options differ in the levels of these treatment characteristics(e.g., 1%,2%,etc.). Based on the levels of the treatment characteristics for each treatment option, you will be asked to choose which treatment option you most prefer. (Note: Definitions have been provided for each treatment characteristic in the appendix section of this survey package.)  This survey should take approximately 20 minutes. Go to the next page for the consent form.  109  STUDY CONSENT TO PARTICIPATE Principal Investigator:  CoInvestigator:  Larry Lynd, PhD  Carlo Marra, PhD  Associate Professor Faculty of Pharmaceutical Sciences University of British Columbia  Associate Professor Faculty of Pharmaceutical Sciences University of British Columbia  604-827-3397  Co-Investigator: Co-Investigator: Jacek Kopec, MD,PhD  Diane Lacaille, MD, FRCPC  Associate Professor  Associate Professor  School of Population and Public Health  Medicine  University of British Columbia  University of British Columbia 604-871-4589  604-871-4588 604-827-3398  Co-Investigator:  Study Coordinator:  Andrew Chalmers, MD, FRCPC  Belinda Chen  Chair of Mary Pack Arthritis Program Research Committee  MSc. Candidate Faculty of Pharmaceutical Sciences  Mary Pack Arthritis Center University of British Columbia 604-875-5353 604-682-2344 ext. 62851  Consent to Participate: Your consent to participate in this study only requires that you: (1) answer the questions in this questionnaire-based survey and (2) mail it back in the provided postage-paid envelope.  110  Should you agree to participate, please print and sign your name at the end of this consent form and mail it back with your completed survey. Once you sign the consent form, you may start the survey. Your identity will be kept confidential. No information that discloses your identity will be released or published. Only the principal investigator and the study coordinator will have access to this information. Remuneration: You will not have any direct costs for participating in this study. All that is asked is that you volunteer your time to fill out the survey. To offset the amount of time and inconvenience required to participate in this study, each participant will receive a $20 honourarium.  Contact information to Address Questions about Rights as a Subject during the Study: If you have any concerns or questions about your treatment or rights as a research participant, you may contact the Research Subject Information Line in the University of British Columbia Office Of Research Services at 604-822-8598.  Subject Consent to Participate I have had sufficient time to consider the information provided. I have had the opportunity to have my questions answered. I understand that all of the information collected will be kept private and that the results will be used for scientific objectives. Full Name (Print):_______________________ Signature:_____________________________ Date: (YY)  _  / (MM)  / (DD)  __  111  MOST VALUED DAILY ACTIVITIES In the numbered spaces below, write 1-5 daily activities (either in your work or personal life or both) that you feel is most important to you (i.e., what activities are most affected by your RA). You must list a minimum of 1 activity. These activities will be referred to as your ‘Most Valued Daily Activities’(MVDA(s)). Your MVDA(s) will be the activities that you would gain improved function in when you take the treatment options in this survey. (E.g., Riding my bike, walking the dog, etc.)  1.  _________________________________  2.  _________________________________  3.  _________________________________  4.  __________________________________  5.  __________________________________  112  For the following 12 choice-set questions, choose the treatment option you MOST PREFER based on the levels of the treatment characteristics. If you do NOT want to choose either of these two treatment options, you can choose the option of none. Choose by checking the box below the treatment option that you prefer.  113  Choice-Set Question 1 Treatment Characteristics  Option A  Option B  Number of people who will benefit  10 out of 10 people will benefit  2 out of 10 people will benefit  Level of improved function in most valued daily activities  A little  Moderate  Number of people who will have stomach bleeds  0 out of 100 (0%)  5 out of 100 (5%)  Number of people who will have heartburn and stomach upset  10 out of 100 (10%)  20 out of 100 (20%)  Number of people who will have a heart attack  2 out of 100 (2%)  0 out of 100 (0%)  Number of people who will have a stroke  2 out of 100 (2%)  None out of 100(0%)  None  I wouldn't choose ANY of these options  114  Choice-Set Question 2 Treatment Characteristics  Option A  Option B  Number of people who will benefit  5 out of 10 people will benefit  10 out of 10 people will benefit  Level of improved function in most valued daily activities  A lot  A little  Number of people who will have stomach bleeds  2 out of 100 (2%)  None out of 100 (0%)  Number of people who will have heartburn and stomach upset  None out of 100 (0%)  10 out of 100 (10%)  Number of people who will have a heart attack  2 out of 100 (2%)  4 out of 100 (4%)  Number of people who will have a stroke  4 out of 100 (4%)  None out of 100(0%)  None  I wouldn't choose ANY of these options  115  Choice-Set Question 3 Treatment Characteristics  Option A  Option B  Number of people who will benefit  5 out of 10 people will benefit  2 out of 10 people will benefit  Level of improved function in most valued daily activities  Moderate  A lot  Number of people who will have stomach bleeds  5 out of 100 (5%)  2 out of 100 (2%)  Number of people who will have heartburn and stomach upset  None out of 100 (0%)  20 out of 100 (20%)  Number of people who will have a heart attack  2 out of 100 (2%)  None out of 100(0%)  Number of people who will have a stroke  2 out of 100 (2%)  4 out of 100 (4%)  None  I wouldn't choose ANY of these options  116  Choice-Set Question 4 Treatment Characteristics  Option A  Option B  Number of people who will benefit  2 out of 10 people will benefit  10 out of 10 people will benefit  Level of improved function in most valued daily activities  Moderate  A lot  Number of people who will have stomach bleeds  None out of 100 (0%)  5 out of 100 (5%)  Number of people who will have heartburn and stomach upset  10 out of 100 (10%)  None out of 100 (0%)  Number of people who will have a heart attack  None out of 100 (0%)  2 out of 100 (2%)  Number of people who will have a stroke  None out of 100(0%)  4 out of 100 (4%)  None  I wouldn't choose ANY of these options  117  Choice-Set Question 5 Treatment Characteristics  Option A  Option B  Number of people who will benefit  10 out of 10 people will benefit  10 out of 10 people will benefit  Level of improved function in most valued daily activities  A lot  A little  Number of people who will have stomach bleeds  None out of 100 (0%)  5 out of 100 (0%)  Number of people who will have heartburn and stomach upset  None out of 100 (0%)  10 out of 100 (10%)  Number of people who will have a heart attack  None out of 100 (0%)  2 out of 100 (4%)  Number of people who will have a stroke  None out of 100 (0%)  4 out of 100(4%)  None  I wouldn't choose ANY of these options  118  Choice-Set Question 6 Treatment Characteristics  Option A  Option B  Number of people who will benefit  5 out of 10 people will benefit  10 out of 10 people will benefit  Level of improved function in most valued daily activities  Moderate  A little  Number of people who will have stomach bleeds  2 out of 100 (2%)  None out of 100 (0%)  Number of people who will have heartburn and stomach upset  None out of 100 (0%)  20 out of 100 (20%)  Number of people who will have a heart attack  4 out of 100 (4%)  None out of 100(0%)  Number of people who will have a stroke  2 out of 100 (2%)  4 out of 100 (4%)  None  I wouldn't choose ANY of these options  119  Choice-Set Question 7 Treatment Characteristics  Option A  Option B  Number of people who will benefit  2 out of 10 people will benefit  5 out of 10 people will benefit  Level of improved function in most valued daily activities  A lot  A little  Number of people who will have stomach bleeds  5 out of 100 (5%)  2 out of 100 (2%)  Number of people who will have heartburn and stomach upset  10 out of 100 (10%)  20 out of 100 (20%)  Number of people who will have a heart attack  4 out of 100 (4%)  2 out of 100 (2%)  Number of people who will have a stroke  None out of 100(0%)  2 out of 100 (2%)  None  I wouldn't choose ANY of these options  120  Choice-Set Question 8 Treatment Characteristics  Option A  Option B  Number of people who will benefit  5 out of 10 people will benefit  10 out of 10 people will benefit  Level of improved function in most valued daily activities  A lot  Moderate  Number of people who will have stomach bleeds  None out of 100 (0%)  2 out of 100 (2%)  Number of people who will have heartburn and stomach upset  20 out of 100 (20%)  10 out of 100 (10%)  Number of people who will have a heart attack  2 out of 100 (2%)  None out of 100 (0%)  Number of people who will have a stroke  None out of 100 (0%)  2 out of 100(2%)  None  I wouldn't choose ANY of these options  121  Choice-Set Question 9 Treatment Characteristics  Option A  Option B  Number of people who will benefit  2 out of 10 people will benefit  10 out of 10 people will benefit  Level of improved function in most valued daily activities  A lot  A little  Number of people who will have stomach bleeds  5 out of 100 (5%)  None out of 100 (0%)  Number of people who will have heartburn and stomach upset  None out of 100 (0%)  20 out of 100 (20%)  Number of people who will have a heart attack  4 out of 100 (4%)  2 out of 100(2%)  Number of people who will have a stroke  2 out of 100 (2%)  4 out of 100 (4%)  None  I wouldn't choose ANY of these options  122  Choice-Set Question 10 Treatment Characteristics  Option A  Option B  Number of people who will benefit  5 out of 10 people will benefit  5 out of 10 people will benefit  Level of improved function in most valued daily activities  A little  A lot  Number of people who will have stomach bleeds  5 out of 100 (5%)  None out of 100 (0%)  Number of people who will have heartburn and stomach upset  20 out of 100 (20%)  None out of 100 (0%)  Number of people who will have a heart attack  4 out of 100 (4%)  None out of 100 (0%)  Number of people who will have a stroke  4 out of 100(4%)  None out of 100 (0%)  None  I wouldn't choose ANY of these options  123  Choice-Set Question 11 Treatment Characteristics  Option A  Option B  Number of people who will benefit  5 out of 10 people will benefit  2 out of 10 people will benefit  Level of improved function in most valued daily activities  A little  Moderate  Number of people who will have stomach bleeds  5 out of 100 (5%)  2 out of 100 (2%)  Number of people who will have heartburn and stomach upset  10 out of 100 (10%)  None out of 100 (0%)  Number of people who will have a heart attack  None out of 100 (0%)  4 out of 100 (4%)  Number of people who will have a stroke  None out of 100 (0%)  4 out of 100 (4%)  None  I wouldn't choose ANY of these options  124  Choice-Set Question 12 Treatment Characteristics  Option A  Option B  Number of people who will benefit  10 out of 10 people will benefit  5 out of 10 people will benefit  Level of improved function in most valued daily activities  A little  Moderate  Number of people who will have stomach bleeds  2 out of 100 (2%)  None out of 100 (0%)  Number of people who will have heartburn and stomach upset  None out of 100 (0%)  10 out of 100 (10%)  Number of people who will have a heart attack  4 out of 100 (4%)  None out of 100 (0%)  Number of people who will have a stroke  None out of 100 (0%)  4 out of 100(4%)  None  I wouldn't choose ANY of these options  125  Thankyou for completing the choice-set questions! In order for us to properly analyze the information we received from you, it is necessary to get some descriptive information about you.  Go to the next page.  126  General Information  1.  Age:  2. Gender:  Years  Male  Female  3. What is your approximate height?(Please enter in feet and inches, with decimal places. Eg. 5'4 = 5.4 - Note: 1 meter = 3.3 feet)  4. What is your approximate weight? (Please enter enter in pounds(lbs) - Note: 2.2lbs = 1 kg )  5. Do you or have you smoked cigarettes, cigars, and/or pipes? Never Smoked Currently Smoke Quit Smoking  6.  What is your marital status? Single Married Living with partner Divorced/Separated Widowed I prefer not to answer this question  127  7. How would you best describe your ethnic group? Please check all that apply. Caucasian Hispanic First Nations, Aboriginal Black East Indian Asian Other  8. What is your highest level of education completed? Elementary school Some high school Technical/Trade/Vocational College Community College Some university University degree obtained Graduate or Post-graduate training  9. What type of health insurance coverage do you have? (Please check all that apply) I don't currently have medical insurance Plan C (income assistance) Plan E (Basic MSP) - self paid Plan E (Basic MSP) - employer paid Extended medical - self paid Extended medical - employer paid Prescription drug plan (3rd party coverage) Other  128  Information About Your RA 1.  At what age was your RA first diagnosed?  ______ Years  2. At what age did the symptoms of your RA first begin? _______ Years  3. Over the past year, how many times have you been admitted to the hospital due to your RA (e.g., joint surgeries)? Please put the answer in numbers. _______Times  4. How much pain have you had because of your illness IN THE PAST WEEK: Place a vertical (|) mark on the line to indicate the severity of the pain  129  5. Considering all the ways that your arthritis affects you, rate how you are doing on the following scale by placing a vertical mark (|) on the line.  130  6. Please indicate, by checking the circle(s) marking the joint(s), the joints that you feel are PAINFUL or TENDER at present.  131  Medical History 1. Have YOU ever been diagnosed/experienced any of the following conditions below? Please check all that apply. Stomach bleeds Heartburn and stomach upset Heart attack(s) Stroke(s) Diabetes High Blood Pressure High Cholesterol Kidney and/or bladder problems Lung Problems Intestinal/bowel disorder Liver Problems Anemia or other blood disorders Fibromyalgia Osteoporosis Back Pain Depression  2. Has anyone in your FAMILY ever been diagnosed/experienced any of the following conditions below. Please check all that apply. Stomach bleeds Heartburn and stomach upset Heart attack(s) Stroke(s) Diabetes High Blood Pressure High Cholesterol Kidney and/or bladder problems Lung Problems Intestinal/bowel disorder Liver Problems Anemia or other blood disorders Fibromyalgia Osteoporosis Back Pain Depression  132  3. Have you ever taken any of the following non-steroidal antiinflammatory drugs (NSAIDS)? Please check all that apply. If your NSAID is not listed, please check other and write the name of the drug. Aspirin Ibuprofen (Advil, Motrin) Acetaminophen (Tylenol) Celecoxib (Celebrex) Sulindac (Clinoril) Indomethacin (Indocid) Meloxicam (Mobicox) Naproxen (Naprosyn) Nabumetone (Relafen) Diclofenac (Voltaren) Other: _______________________________  4. Have you taken any of the following non-steroidal anti-inflammatory drugs (NSAIDS) in the PAST 6 MONTHS? Please check all that apply. If your NSAID is not listed, please check other and write the name of the drug. Aspirin Ibuprofen (Advil, Motrin) Acetaminophen (Tylenol) Celecoxib (Celebrex) Sulindac (Clinoril) Indomethacin (Indocid) Meloxicam (Mobicox) Naproxen (Naprosyn) Nabumetone (Relafen) Diclofenac (Voltaren) Other: _______________________________  133  5. Have you ever taken any of the following non-biologic disease – modifying anti-rheumatic drugs (DMARDs)? Please check all that apply. If your DMARD is not listed, please check other and write the name of the drug. Leflunomide (Arava) Azathioprine (Imuran) Methotrexate (Methotrexat) Hydroxychloroquine (Plaquenil) Sulfasalazine (Salazopyrin) Auranofin (Gold Therapy) Other: _________________________________  6. Have you taken any of the following non-biologic disease –modifying anti-rheumatic drugs (DMARDs) in the PAST 6 MONTHS? Please check all that apply. If your DMARD is not listed, please check other and write the name of the drug. Leflunomide (Arava) Azathioprine (Imuran) Methotrexate (Methotrexat) Hydroxychloroquine (Plaquenil) Sulfasalazine (Salazopyrin) Auranofin (Gold Therapy) Other: _________________________________  7. Have you ever taken any of the following biologics or steroid drugs? Please check all that apply. If your drug is not listed, please check other and write the name. Etanercept (Enbrel) Abatacept (Orencia) Adalimumab (Humira) Anakinra (Kineret) Infliximab (Remicade) Rituximab (Rituxan) Prednisone (Prednicot) Cortisone shot Other:____________________________________  134  8. Have you taken any of the following biologics or steroid drugs in the PAST 6 MONTHS? Please check all that apply. If your drug is not listed, please check other and write the name. Etanercept (Enbrel) Abatacept (Orencia) Adalimumab (Humira) Anakinra (Kineret) Infliximab (Remicade) Rituximab (Rituxan) Prednisone (Prednicot) Cortisone shot Other:____________________________________  135  136  137  Please check the level that you feel represents your health in the following categories. Physical Functioning Your health does not limit you in vigorous activities Your health limits you a little in vigorous activities Your health limits you a little in moderate activities Your health limits you a lot in moderate activities Your health limits you a little in bathing and dressing Your health limits you a lot in bathing and dressing Role Limitations You have no problems with your work or other regular daily activities as a result of your physical health or any emotional problems You are limited in the kind of work or other activities as a result of your physical health You accomplish less than you would like as a result of emotional problems You are limited in the kind of work or other activities as a result of your physical health and accomplish less than you would like as a result of emotional problems Social Functioning Your health limits your social activities none of the time Your health limits your social activities a little of the time Your health limits your social activities some of the time Your health limits your social activities most of the time Your health limits your social activities all of the time Pain You have no pain You have pain but it does not interfere with your normal work (both outside the home and housework) You have pain that interferes with your normal work (both outside the home and housework) a little bit You have pain that interferes with your normal work (both outside the home and housework) quite a bit You have pain that interferes with your normal work (both outside the home and housework) moderately You have pain that interferes with your normal work (both outside the home and housework) extremely  138  Mental health You feel tense or downhearted and low none of the time You feel tense or downhearted and low a little of the time You feel tense or downhearted and low some of the time You feel tense or downhearted and low most of the time You feel tense or downhearted and low all of the time Vitality You have a lot of energy all of the time You have a lot of energy most of the time You have a lot of energy some of the time You have a lot of energy a little of the time You have a lot of energy none of the time  139  Thankyou for your participation! You are now done the survey.  140  Appendix 2: The complete list of definitions for treatment characteristics  (A) Benefits associated with treatment 1) Number of people who will benefit There is no guarantee that the treatment will work for you. Some people will experience a benefit in pain reduction and improved function, while others will not. Taking a treatment option, either 20, 50, or 100 out of 100 people will experience benefit in pain reduction and improved function (see levels below). In those that do not benefit, they will not experience any effect from the treatment, i.e., benefit. The greater the number of people that benefit, the higher the chance that this treatment option will work for you. 20 out of 100 people will benefit (20%) 50 out of 100 people will benefit (50%) 100 out of 100 people will benefit (100%)  2) Level of improved function in most valued daily activities The benefit gained from treatment options in this survey is pain reduction. With pain reduction, you will gain improved function in your most valued daily activities (MVDA(s)). There are three different levels of improved function (see levels below) that you may experience taking a treatment option. Each treatment option will not offer complete pain reduction and therefore, you will have remaining pain that will interfere with your function in your MVDA(s). For example, you will have greater remaining pain interfering with your function in your MVDA(s) from the treatment option giving you a little bit of improved function versus a lot. A little bit Moderate  141  A lot Adverse events associated with treatment 3) Number of people who will have stomach bleeds With stomach bleeds you will feel unwell and vomit blood. Treatment may include surgery or taking additional drugs to prevent bleeds from happening. However, these drugs don’t work for 100% of the people taking it. Left untreated, the long-term outcome may be death. Taking a treatment option, either 0, 2, or 5 out of 100 people will get stomach bleeds (see levels below). The greater the number of people that have stomach bleeds, the greater chance it will be for you to have stomach bleeds from taking this treatment option. None out of 100 (0%) 2 out of 100 (2%) 5 out of 100 (5%)  4) Number of people who will have heartburn and stomach upset With NSAID therapy, you may experience, nausea, stomach pain, and heartburn (burning feeling in the chest). Sometimes this problem can be treated by changing the dosage of the drug you are taking, or by taking other drugs with your NSAIDs. Taking a treatment option, either 0, 10, or 20 out of 100 people will have heartburn, nausea, and stomach pain (see levels below). The greater the number of people that have heartburn and stomach upset, the greater chance it will be for you when taking this treatment option. None out of 100 (0%) 10 out of 100 (10%)  142  20 out of 100 (20%) 5) Number of people who will have a heart attack With a heart attack, you will experience extreme pain in your chest and may stop breathing. Procedures for a heart attack can include hospitalization, drug treatments, or surgery. There is a chance of death due to a heart attack, however, if the heart muscle is not too badly damaged, some can go on to live a normal life. Taking a treatment option, either 0, 2, or 4 out of 100 people will have a heart attack(see levels below). The greater the number of people that have a heart attack, the greater chance it will be for you when taking this treatment option. None out of 100 (0%) 2 out of 100 (2%) 4 out of 100 (4%)  6) Number of people who will have a stroke A stroke is where there is bleeding in the brain that may lead to a loss of brain function. This may cause permanent disability or paralysis in major body parts. Death may also occur. Those who survive may go on to live a normal life. Taking a treatment option, either 0, 2, or 4 out of 100 people will have a stroke(see levels below). The greater the number of people that have a stroke, the greater chance it will be for you when taking this treatment option. None out of 100(0%) 2 out of 100 (2%) 4 out of 100 (4%)  143  

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.24.1-0071558/manifest

Comment

Related Items