UBC Faculty Research and Publications

How confidence intervals become confusion intervals McCormack, James; Vandermeer, Ben; Allan, G M Oct 31, 2013

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

Item Metadata

Download

Media
52383-12874_2013_Article_1007.pdf [ 220.3kB ]
Metadata
JSON: 52383-1.0223852.json
JSON-LD: 52383-1.0223852-ld.json
RDF/XML (Pretty): 52383-1.0223852-rdf.xml
RDF/JSON: 52383-1.0223852-rdf.json
Turtle: 52383-1.0223852-turtle.txt
N-Triples: 52383-1.0223852-rdf-ntriples.txt
Original Record: 52383-1.0223852-source.json
Full Text
52383-1.0223852-fulltext.txt
Citation
52383-1.0223852.ris

Full Text

DEBATE Open AccessHow confidence intervals become confusionintervalsJames McCormack1, Ben Vandermeer2 and G Michael Allan3*AbstractBackground: Controversies are common in medicine. Some arise when the conclusions of research publicationsdirectly contradict each other, creating uncertainty for frontline clinicians.Discussion: In this paper, we review how researchers can look at very similar data yet have completely differentconclusions based purely on an over-reliance of statistical significance and an unclear understanding of confidenceintervals. The dogmatic adherence to statistical significant thresholds can lead authors to write dichotomized abso-lute conclusions while ignoring the broader interpretations of very consistent findings. We describe three examplesof controversy around the potential benefit of a medication, a comparison between new medications, and a medi-cation with a potential harm. The examples include the highest levels of evidence, both meta-analyses and random-ized controlled trials. We will show how in each case the confidence intervals and point estimates were very similar.The only identifiable differences to account for the contrasting conclusions arise from the serendipitous finding ofconfidence intervals that either marginally cross or just fail to cross the line of statistical significance.Summary: These opposing conclusions are false disagreements that create unnecessary clinical uncertainty. Weprovide helpful recommendations in approaching conflicting conclusions when they are associated with remarkablysimilar results.Keywords: Confidence intervals, Evidence based medicine, Statistical analysis, Statistical significanceBackgroundMost published reports of clinical studies begin with anabstract – likely the first and perhaps only thing manyclinicians, the media and patients will read. Within thatabstract, authors/investigators typically provide a briefsummary of the results and a 1–2 sentence conclusion.At times, the conclusion of one study will be different,even diametrically opposed, to another despite the au-thors looking at similar data. In these cases, readers mayassume that these individual authors somehow founddramatically different results. While these reported differ-ences may be true some of the time, radically diverse con-clusions and ensuing controversies may simply be due totiny differences in confidence intervals combined with anover-reliance and misunderstanding of a “statistically sig-nificant difference.” Unfortunately, this misunderstandingcan lead to therapeutic uncertainty for front-line clinicianswhen in fact the overall data on a particular issue is re-markably consistent.A key concept of science is the formulation of hypoth-eses and the testing these hypotheses by observing a set ofdata. Typically in medicine one starts with an idea that atherapy will have an effect. A statistical test assumes thatan intervention has no effect and this is called the null hy-pothesis. A statistical evaluation simply provides informa-tion as to how likely that the finding of a particulardifference could be due to chance and if there really wasno difference between the treatment groups.We can NEVER prove a null hypothesis, meaning theintervention has absolutely no effect. However, we de-sign clinical studies with the hope they will provide in-formation to help decide if we should reject or fail toreject the null hypothesis. Rejecting the null hypothesisis often interpreted to mean the intervention has an ef-fect; failing to reject the null hypothesis is interpreted tomean the intervention does not have an effect. Thesesimplistic interpretations ignore important factors such* Correspondence: michael.allan@ualberta.ca3Evidence-Based Medicine, Department of Family Medicine, University of Alberta,Room 1706 College Plaza, 8215 - 112 Street NW, Edmonton, AB, CanadaFull list of author information is available at the end of the article© 2013 McCormack et al.; licensee BioMed Central Ltd. This is an open access article distributed under the terms of theCreative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use,distribution, and reproduction in any medium, provided the original work is properly cited.McCormack et al. BMC Medical Research Methodology 2013, 13:134http://www.biomedcentral.com/1471-2288/13/134as clinical importance, precision of the estimate, andstatistical power.A well-designed randomized controlled trial (RCT) isusually the least biased way to evaluate the differencebetween different therapeutic interventions. Unless anRCT has studied the entire population of interest, theobserved difference or ratio that is found is called apoint estimate because only a small sample of the entirepopulation has been evaluated. A point estimate is ty-pically presented with a 95% (or less commonly 99%)confidence interval (CI). A CI, while it has other inter-pretations, can be thought of as a range of numeric out-comes that we are reasonably confident includes theactual result.The choice of a specific CI typically comes from theconvention of a p-value of 0.05 representing a statisticalsignificance. This threshold has been discussed as beingarbitrary but has also been suggested to represent a rea-sonable threshold [1]. Any statistical threshold can bedebated because a threshold depends on how comfort-able one is that the results of a particular study may bedue to chance. However, what cannot be debated is thatthis threshold was never developed to allow researchersor clinicians to make dichotomous conclusions that, if ap-value is greater than 0.05, the intervention has no ef-fect and, if a p-value is less than 0.05, the interventionhas an effect. When using CIs to assess statistical signifi-cance, the “no effect” cut-off occurs when the CI touchesthe line of 1 for relative risks or odds ratios, and 0 forabsolute risks and weighted mean differences.We have chosen three examples of this problem – a po-tential benefit of a medication, a comparison between newmedications, and a medication with a potential harm. Wewill show this problem occurs with the highest-level evi-dence – randomized controlled trials and meta-analyses.We have framed these into three clinical questions:1) In patients without a history of cardiovasculardisease, do statins reduce mortality?2) In patients with atrial fibrillation, when compared towarfarin, is apixaban more effective than dabigatranat reducing mortality?3) In patients who smoke, does the use of vareniclineincrease the risk of serious cardiovascular adverseevents?StatinsExample 1. In patients without a history of cardiovasculardisease, do statins reduce mortality?Statins are widely used in patients with and without estab-lished cardiovascular disease. An important clinically rele-vant question is: do statins have an effect on overallmortality in patients who have not experienced a cardio-vascular event? Because of the relatively low baseline 5-year risk of mortality in this population (roughly 5% over5 years), no single study has been powered sufficiently toprovide a clear answer. For that reason, at least five differ-ent meta-analyses examining this question have been pub-lished [2-6].The authors of these meta-analyses concluded thefollowing: Studer et al.: “statins and n-3 fatty acids are themost favorable lipid-lowering interventions with re-duced risks of overall and cardiac mortality.” [2] Thavendiranathan et al.: “statin therapy decreasesthe incidence of major coronary and cerebrovascularevents and revascularizations but not coronary heartdisease or overall mortality.” [3] Mills et al.: “We examined the impact of statintherapy on major events and found an importantrole in preventing all-cause mortality” [4] althoughthis quote was not found in the abstract but ratherthe first line of the discussion. Brugts et al.: “statin use was associated withsignificantly improved survival.” [5] Ray et al.: “this literature-based meta-analysis didnot find evidence for the benefit of statin therapy onall-cause mortality.” [6]Three groups of investigators felt they had found thepooled clinical trial evidence sufficient to state that sta-tins reduce overall mortality; yet two others felt theirevidence did not support statins reducing overall mortal-ity. Figure 1 shows the characteristics and overall mor-tality relative risks of the five meta-analyses. Althoughthe different meta-analyses included some different stud-ies, overall the investigators used similar data and, notsurprisingly, found similar results. The range of point es-timates was 0.86 to 0.93 with an average point estimateof 0.90. The lower limits of the CIs ranged from 0.76 to0.83 and the upper limits ranged from 0.96-1.01. TheCIs overlap considerably and there is little meaningfuldifference in the results. The only “differences” lie inthree meta-analyses [2,4,5] in which the upper limits ofthe CI fell just below 1.0 and just above 1.0 in the othertwo [3,6]. It appears the differing conclusions were duesolely to a proclivity for a p-value of 0.05 and adherenceto statistical significance as a dichotomous outcome.Novel oral anticoagulantsExample 2. In patients with atrial fibrillation, whencompared to warfarin, is apixaban more effective thandabigatran at reducing mortality?A new class of oral anticoagulants (OACs) has recentlybeen released on the market. An important clinical ques-tion is which one of these new agents is the most effective;does either agent reduce mortality more than the “gold-McCormack et al. BMC Medical Research Methodology 2013, 13:134 Page 2 of 6http://www.biomedcentral.com/1471-2288/13/134standard” warfarin? Two separate studies compare two ofthe new OACs and warfarin [7,8].Connolly et al. state “The mortality rate was 4.13% peryear in the warfarin group, as compared with 3.75%per year with 110 mg of dabigatran (P = 0.13) and 3.64%per year with 150 mg of dabigatran (P = 0.051)” [7]. Theauthors do not make any specific conclusions on mortal-ity differences between warfarin and dabigatran.In contrast, Granger et al. (ARISTOTLE) concludedthat “apixaban was superior to warfarin in preventingstroke or systemic embolism, caused less bleeding, andresulted in lower mortality” [8]. Interestingly, the pressstated: “ARISTOTLE: A major win for apixaban inAF”, “the most positive yet” and “first of the three neworal anticoagulants to show a clearly significant reduc-tion in all-cause mortality.” [9]Figure 2 shows the relative risk for mortality with eachdrug (and dose) compared to warfarin. For all intentpurposes the results are basically identical, particularlywhen comparing dabigatran 150 mg and apixaban.The Granger paper [8] illustrates the importance thatauthors attach whether or not results cross the magicalline of 1.0 and statistically significance. In Table Two ofthe article, the authors present 10 outcome results –for9 they include the upper limit of the CIs with numbersto two decimal points. Yet for the mortality data, theyshow the upper limit number to 3 decimal points (0.80-0.998) [8]. Interestingly, for the mortality CIs in theStudy Point Estimate(95% Confidence Intervals)Relative RiskOverall MortalityMills 2008 19 (63,899) 0.93 (0.87 – 0.99)Ray 2010 11 (65,229) 0.91 (0.83 – 1.01)0.65 0.7 0.75 0.8 0.85 0.90 0.95 1 1.05 1.10.6Number of Trials in Meta-analysis(Number of Patients)Brugts 2009 9 (67,476) 0.88 (0.81 – 0.96)Thavendiranathan 2006 6 (39,937) 0.92 (0.84 – 1.01)Studer 2005 9 (26,641) 0.86 (0.76 – 0.99)Figure 1 Comparison of 5 meta-analyses examining relative risk of overall mortality with statin use in primary prevention. Footnote:Brugts 2009 point estimate and confidence intervals are odds ratios (not relative risks).Study Point Estimate(95% Confidence Intervals)Relative RiskOverall Mortality0.7 0.8 0.9 1 1.1MedicationGranger 2011 Apixaban 0.89 (0.80-0.998)110 mg dabigatran 0.91 (0.80-1.03)150 mg dabigatran 0.88 (0.77-1.00)Connolly 2009Figure 2 Comparison of 2 randomized controlled trials examining the relative risk of overall mortality with 2 novel oral anticoagulantsversus warfarin in atrial fibrillation.McCormack et al. BMC Medical Research Methodology 2013, 13:134 Page 3 of 6http://www.biomedcentral.com/1471-2288/13/134abstract and the body of the paper, this 0.998 result wasrounded down to 0.99 rather than the more correctrounding up to 1.00.VareniclineExample 3. In patients who smoke, does the use ofvarenicline increase the risk of serious cardiovascularadverse events?Varenicline is a new smoking cessation medicationthat is widely used. As with most new medications, thelong-term or rare side effects are unknown. For thatreason investigators have conducted meta-analyses ofmany small trials to try to identify any previously un-known adverse effects. Two meta-analyses have exam-ined a possible risk of serious cardiovascular eventswith varenicline [10,11].Singh et al. reported that “Our meta-analysis raisessafety concerns about the potential for an increased riskof serious adverse cardiovascular events associated withthe use of varenicline.” [10]In contrast a later meta-analysis by Prochaska et al.reported that varenicline used for smoking cessationproduced “no significant increase in cardiovascularserious adverse events associated with vareniclineuse” [11]. These authors go on to state “The conse-quence of inflated risk estimates, such as those fromSingh and colleagues’ meta-analysis concerning theeffect of varenicline on serious adverse events relatedto cardiovascular disease, can be unnecessary publicalarm and real harm, since patients may discontinuetheir drug treatment out of fear of adverse effects andclinicians may recommend cessation treatments of re-duced efficacy or discourage use of the drug treat-ment altogether.” [11]Figure 3 shows the characteristics and serious car-diovascular event peto odds ratios of the two meta-analyses. As with the previous examples, the tworesults are quite consistent with each other. The onlyapparent reason for the contradictory conclusionsabout risk was that the lower limit of the CI fell justbelow 1.0 in one meta-analysis, and was just above 1.0in the other.Pragmatic interpretation of the included studiesBased on the evidence presented, we believe the fol-lowing represents a reasonable and pragmatic interpretationof the results and how a clinician might use the information.Statins’ effect on overall mortality in primary preventionIf you were a betting person, you should bet that statinslikely reduce mortality in primary prevention. The aver-age point estimate in these meta-analyses was around0.90 or a 10% relative reduction [2-6]. The relative re-duction may vary from 0 to 20% but 10% is a reasonableapproximation [2-6]. The baseline risk of mortality wasapproximately 5% [2-6], so using the 10% risk reductiongives a 0.5% absolute reduction. Therefore, we need totreat 200 primary prevention patients for one extra per-son to avoid death over 5 years. Conversely, 199 will notget a benefit in changing their risk of mortality. The CIalso suggests the absolute reduction in risk could be ashigh as 1% or as little as 0%. Given the results we canconfidently say it is very unlikely that statins increasemortality. Bottom-line, statins seem to have roughly a 1in 200 effect on overall mortality in primary prevention.Novel anti-coagulants effect on overall mortality in atrialfibrillationThe evidence does not suggest any differences betweenapixaban and dabigatran and their effect on mortalitycompared to warfarin. As with the statins, dabigatranand apixaban likely do reduce mortality, approximately10% with a CI of 0% to 20% [7,8]. The baseline deathrate on warfarin was roughly 4% per year [7,8] so the re-duction in risk would be 0.4% although it could be ashigh as 0.8% or low as 0%. Therefore, we need to treat250 atrial fibrillation patients with apixaban or dabiga-tran instead of warfarin for one extra person to avoiddeath over 1 year. Conversely, 249 will not get a benefitin changing their risk of mortality. (Note: baseline deathrates may vary considerably based on a specific CHADSscore but we are using the average from the clinicaltrials included [7,8].) Based on the evidence, it is veryunlikely that dabigatran or apixaban increase mortalitycompared to warfarin. Bottom-line, compared to warfarin,Study Point Estimate(95% Confidence Intervals)Peto Odds RatioSerious Cardiovascular Events 0 0.5 1 1.5 2 2.5 3Number of Trials in Meta-analysis(Number of Patients)Prochaska 2012 22 (9232) 1.58 (0.90-2.76)Singh 2011 14 (8216) 1.72 (1.09-2.71)Figure 3 Comparison of 2 meta-analyses examining peto odds ratio of serious cardiovascular events with varenicline use insmoking cessation.McCormack et al. BMC Medical Research Methodology 2013, 13:134 Page 4 of 6http://www.biomedcentral.com/1471-2288/13/134dabigatran and apixaban seem to have a 1 in 250 benefiton overall mortality in one year in atrial fibrillationpatients.Varenicline effect on serious cardiovascular outcomesIf you were a betting person, you should bet that vareni-cline likely does increase the risk of cardiovascularevents. The point estimate was roughly 60-70% but the ef-fect could be as high as a 176% increase or an actual 10%reduction in cardiovascular events [10,11]. Importantly,baseline cardiovascular risk can vary dramatically. A 30-year-old female with no risk factors except smoking mayhave 0.5% 5-year risk or a 0.1% one year risk of a cardio-vascular event. If varenincline increased risk by 60%, oneextra cardiovascular event would occur for every 1667 pa-tients treated. In patients immediately post-MI, whoseCVD risk may approach 5% in the first year, vareniclinemay lead to one extra cardiovascular event for every 34patients treated (7 of whom would quit smoking). In thesepatients, the benefits may still outweigh risks but a discus-sion around potential risks and options would be appropri-ate. Bottom-line, varenicline likely produces an increasedrisk of cardiovascular events but this risk may be <1/1000for low risk patients to as much as in 1/34 for the highestrisk patients. However, the benefits of smoking cessationare considerable.DiscussionIt appears that medical authors feel the need to makeblack and white conclusions when their data almostnever allows for such dichotomous statements. This isparticularly true when comparing results to similar stud-ies with largely overlapping CIs. Virtually all of the con-clusion confusion discussed in this paper can be linkedto slavish adherence to an arbitrary threshold for statis-tical significance. Even if the threshold is reasonable, itstill cannot be used to make dichotomous conclusions.Although we have selected three examples here, theseare certainly not the only ones. In another example weconsidered, the authors of two meta-analyses of primaryprevention with aspirin report the exact same point esti-mate and confidence interval 0.94 (0.88-1.00) but haddiffering conclusions [12,13]. We tried to select a smallbut representative group of examples that would be fa-miliar to most readers.We are not the first authors to write about the misin-terpretation of CIs and statistical significance. About 60years ago, RA Fischer introduced the p-value for hypoth-esis and significance testing [14]. Although 0.05 was sug-gested as a reasonable indicator for significance, he didassert the interpretation was open [14]. Over 30 yearsago, a number of articles were published encouraging me-dical researchers to report their results with CIs [15-18].CIs provide an estimation reflecting the potential range ofeffect rather than simply stating if results are statisticallysignificant or not [15-17]. Unfortunately, this goal fell shortas CIs are frequently used to define whether a result is or isnot statistically significant. In the 60 plus year history, arti-cles on application of statistical reporting continue to en-courage authors to present their findings [17,19,20], allowreaders to interpret results [14,17,19,20] and not use CIsstrictly for reporting statistical significance [17,20].We encourage authors to avoid statements like “X hasno effect on mortality” as they are likely to be both un-true and misleading. This is especially true as results get“close” to being statistically significant. Results shouldspeak for themselves. For that to happen, readers (clini-cians and science reporters) need to understand the lan-guage of statistics and approach authors’ conclusionswith a critical eye. We are not trying to say that thereader should not review the abstract but when authors’conclusions differ from others, readers must examineand compare the actual results. In fact, all but one of themeta-analyses provided point estimates and CIs in theabstracts. This facilitates quick comparisons to otherstudies reported to be “completely different,” and to de-termine if the CIs demonstrate clinically important dif-ferences. The problem lies in the authors’ conclusions,which often have little to do with their results but ratherwhat they want the results to show. We encourage jour-nal editors to challenge authors’ conclusions, particularlywhen they argue they have found something unique ordifferent than other researchers but the difference isbased solely on tiny variations in CIs or p-value (statisti-cally significant or not).We are not suggesting the elimination of statistical test-ing or statistical significance, but rather that all people(authors, publishers, regulators etc.) who write aboutmedical interventions use common sense and good judg-ment when presenting results that differ from others andnot be so beholden to the “magical” statistical significancelevel of 0.05. We urge them to consider the degree towhich the results of the “differing” study overlap withtheir own, the true difference in the point estimates andrange of possible effects, where the preponderance of theeffect lies and how clinicians might apply the evidence.It appears that readers of the papers discussed herewould be better served by reviewing the actual resultsthan reading the authors’ conclusions. To do that, clini-cians need to be able to interpret the meaning of CIsand statistical significance.SummaryDogmatic adherence to statistical significance thresholdscan lead authors to write dichotomized absolute conclu-sions while ignoring the broader interpretations of veryconsistent findings. These opposing conclusions are falsedisagreements that create unnecessary clinical uncertainty.McCormack et al. BMC Medical Research Methodology 2013, 13:134 Page 5 of 6http://www.biomedcentral.com/1471-2288/13/134Authors are encouraged to report data more pragmatic-ally. Readers and clinicians are encouraged to comparethe actual data and precision of the results rather thanrely on the conclusions of the authors.AbbreviationsCI: Confidence intervals; RCT: Randomized controlled trial.Competing interestsWe do not have any competing interest related to this article.Authors’ contributionsGMA conceived of the paper, collected the examples, completed the firstdraft of the figures, managed the manuscript, edited the article substantially,and is the guarantor. JM helped refine the idea, completed the first draft,edited the figures and edited the article. BV contributed to early discussionsand crafting of the article and edited the article. All authors read andapproved the final manuscript.Authors’ informationGMA is general practitioner and academic physician with focus in evidence-based medicine and knowledge translation. JM is doctor of pharmacy andan academic with a focus on evidence-based medicine and knowledgetranslation. BV is biostatistician and has written on statistical methodologyand interpretation.AcknowledgmentsWe would like the acknowledge Lynda Cranston, a medical writer, whoreviewed a near-final draft of the article for grammatical errors and typos.Author details1Faculty of Pharmaceutical Sciences, University of British Columbia,Vancouver, BC, Canada. 2Alberta Research Centre for Health Evidence,University of Alberta, Edmonton Alberta, Canada. 3Evidence-Based Medicine,Department of Family Medicine, University of Alberta, Room 1706 College Plaza,8215 - 112 Street NW, Edmonton, AB, Canada.Received: 8 August 2013 Accepted: 29 October 2013Published: 31 October 2013References1. Cowles M, Davis C: On the origins of the.05 level of statisticalsignificance. Am Psychol 1982, 37:553–558.2. Studer M, Briel M, Leimenstoll B, Glass TR, Bucher HC: Effect of differentantilipidemic agents and diets on mortality: A Systematic Review. ArchIntern Med 2005, 165:725–730.3. Thavendiranathan P, Bagai A, Brookhart MA, Choudhry NK: Primaryprevention of cardiovascular diseases with statin therapy: a meta-analysisof randomized controlled trials. Arch Intern Med 2006, 166:2307–2313.4. Mills EJ, Rachlis B, Wu P, Devereaux PJ, Arora P, Perri D: Primary preventionof cardiovascular mortality and events with statin treatments: a networkmeta-analysis involving more than 65,000 patients. J Am Coll Cardiol2008, 52:1769–1781.5. Brugts JJ, Yetgin T, Hoeks SE, Gotto AM, Shepherd J, Westendorp RGJ, et al:The benefits of statins in people without established cardiovasculardisease but with cardiovascular risk factors: meta-analysis of randomisedcontrolled trials. BMJ 2009, 338:b2376.6. Ray KK, Seshasai SRK, Erqou S, Sever P, Jukema JW, Ford I, et al: Statins andall-cause mortality in high-risk primary prevention: a meta-analysis of 11randomized controlled trials involving 65,229 participants. Arch InternMed 2010, 170:1024–1031.7. Connolly SJ, Ezekowitz MD, Yusuf S, Eikelboom J, Oldgren J, Parekh A, et al:Dabigatran versus warfarin in patients with atrial fibrillation. N Engl J Med2009, 361:1139–1151.8. Granger CB, Alexander JH, McMurray JJV, Lopes RD, Hylek EM, Hanna M, etal: Apixaban versus warfarin in patients with atrial fibrillation. N Engl JMed 2011, 365:981–992.9. Hughes S: ARISTOTLE: A major win for apixaban in AF. Heartwire; 2011. http://www.medscape.com/viewarticle/748682 (accessed November 5, 2013).10. Singh S, Loke YK, Spangler JG, Furberg CD: Risk of serious adversecardiovascular events associated with varenicline: a systematic reviewand meta-analysis. CMAJ 2011, 183:1359–1366.11. Prochaska JJ, Hilton JF: Risk of cardiovascular serious adverse eventsassociated with varenicline use for tobacco cessation: systematic reviewand meta-analysis. BMJ 2012, 344:e2856.12. Seshasai SR, Wijesuriya S, Sivakumaran R, Nethercott S, Erqou S, Sattar N, RayKK: Effect of aspirin on vascular and nonvascular outcomes: meta-analysisof randomized controlled trials. Arch Intern Med 2012, 172:209–216.13. Raju N, Sobieraj-Teague M, Hirsh J, O’Donnell M, Eikelboom J: Effect of as-pirin on mortality in the primary prevention of cardiovascular disease.Am J Med 2011, 124:621–629.14. Sterne JA, Davey Smith G: Sifting the evidence-what’s wrong with signifi-cance tests? BMJ 2001, 322:226–231.15. Rothman KJ: A show of confidence. N Engl J Med 1978, 299:1362–1363.16. Gardner MJ, Altman DG: Confidence intervals rather than P values:estimation rather than hypothesis testing. Br Med J (Clin Res Ed) 1986,292:746–750.17. Poole C: Beyond the confidence interval. Am J Public Health 1987, 77:195–199.18. Bulpitt CJ: Confidence intervals. Lancet 1987, 329(8531):494–497.19. Cooper RJ, Wears RL, Schriger DL: Reporting research results:recommendations for improving communication. Ann Emerg Med 2003,41:561–564.20. Cummings P, Koepsell TD: P values vs estimates of association withconfidence intervals. Arch Pediatr Adolesc Med 2010, 164:193–196.doi:10.1186/1471-2288-13-134Cite this article as: McCormack et al.: How confidence intervals becomeconfusion intervals. BMC Medical Research Methodology 2013 13:134.Submit your next manuscript to BioMed Centraland take full advantage of: • Convenient online submission• Thorough peer review• No space constraints or color figure charges• Immediate publication on acceptance• Inclusion in PubMed, CAS, Scopus and Google Scholar• Research which is freely available for redistributionSubmit your manuscript at www.biomedcentral.com/submitMcCormack et al. BMC Medical Research Methodology 2013, 13:134 Page 6 of 6http://www.biomedcentral.com/1471-2288/13/134

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.52383.1-0223852/manifest

Comment

Related Items