A Vision of Accessible Epidemiology

I may have been invited to comment on a proposal to register observational epidemiologic studies1 because my views are strong, perhaps even radical. I believe recognition is inexorably growing that access to epidemiologic research, like medical care access, is a right and not a privilege. Although experimental studies of intended effects of medical interventions are on the vanguard of this transparency revolution,2,3 I do not share the view that this development reflects a fundamental distinction of subject matter or research design.4 Most epidemiologic research—observational and experimental—has the potential to affect many lives and improve scientific understanding of pathogenesis. STUDY REGISTRATION I would support a requirement to make all epidemiologic studies known to the public. An online forum along the lines of the International Agency for Research on Cancer's sadly defunct "Directory of Ongoing Research in Cancer Epidemiology"5 would suffice to provide the basic design structures and data elements. DATA ACCESS In my view, all epidemiologic study data should be as available as the public-release data from the National Health Interview Study6 and the National Health and Examination Survey.7 Access should be provided as a nonprofit, public service. Participant privacy should be protected, but not used as an excuse for impeding access. Investigators should be given a few years to conduct analyses of data they collect. Thereafter, all epidemiologic study data, analyzed or not, should be made available to the public. Anyone should have access to the data. No protocol, demonstration of competence, declaration of present or absent conflicts of interest, or approval of aims or methods8 should be required. To clarify, I hold that even incompetent and biased members of the public should have cheap and easy access to all epidemiologic study data for any reason. This access is everyone's natural right, not the privilege of a deserving elite. For example, suppose there is a conflict between results from industry-funded and other research. Full access would facilitate investigation of the discrepancy by enabling the researchers, and others, to replicate each others' methods and reanalyze each others' data. REPORTING OF ANALYSES I believe that the methods and results of all analyses of epidemiologic data should be made available to the public. If not published in journals or books, they should be made available in other ways (eg, online). PREDESIGNATION OF HYPOTHESES I would not support a requirement for hypotheses to be designated before an epidemiologic study is conducted. I believe any problems such a requirement would be intended to ameliorate would be more effectively reduced, if not eliminated, by complete disclosure of data collection, full access to data, and comprehensive reporting of analyses. As my views on this point clash most obviously with those in the workshop report,1 I offer illustrative examples. In reading a paper on childhood cancers and traffic density,9 should we care that the study was designed to examine electric and magnetic fields?10 Should our interpretation of associations between breast cancer and variants of telomere pathway genes11 be affected by knowing that a study's founding purpose was to investigate environmental pollutants?12 Should it matter that a cohort study's original motivation, to investigate long-term effects of oral contraceptives,13 appears fifth on a list of its key findings?14 I believe the only defensible answer to such questions is "No." Does carvedilol extend the lives of heart failure patients? The answer should depend on all relevant theory and evidence at the time the question is asked, not on the state of mind of investigators before they collected relevant data. The drug's beneficial effect, widely accepted today,15 was clearly evident in the mid-1990s (Fig.)16—while a regulatory agency's advisory committee dithered for many months, obsessing over which, and how many, "primary end points" the original trialists had predesignated.17 Any school of statistical thought in which such information would seem relevant, let alone crucial,18 should have been tossed onto the scrap heap of misbegotten notions long ago.FIGURE.: Study-specific16 and fixed-effect summary mortality relative risks (RRs) and 95% confidence intervals (CIs) from randomized controlled trials of carvedilol in heart failure. Do you know in which studies mortality was a predesignated "primary end point"?18 Should we care?Do rofecoxib and naproxen have cardiovascular effects? A systematic-review team chose acute myocardial infarction as their primary outcome and compiled 29 relevant studies.19 Hypothesis predesignation by the original investigators was of no concern to the reviewers. Good for them. SUMMARY Observational and experimental epidemiologic studies are public goods. They should not be kept secret. Hypotheses should not have to be designated in advance of conducting such studies. The collected data should be made available for anyone to analyze. All analysis results should be made public. Space limitations have prohibited detailed explications and defenses of these views. An obvious criticism is the infeasibility of my admittedly idealistic dreams of transparent data collection, universal data access, and full reporting of results. However unattainable these ideals might be, I firmly believe that steps toward them would be steps in right directions.