Epidemiology and Comparative Effectiveness – How can epidemiology best contribute?
Two seminal reports, the first by the Institute of Medicine and the second by the Patient-Centered Outcomes Research Institute, focus on the priorities for comparative effectiveness research (CER). An important question is the role that observational, rather than interventional, studies can play in addressing the priorities raised in these two reports. It is likely that both financial and feasibility issues will limit the number of trials that can be conducted. Some questions that we’d like to address include:
· What areas of CER can be addressed by observational studies, particularly where trials may not be feasible?
· Can there be better analytical methods to control for indication bias in evaluating treatments?
· What can be contributed by existing cohorts, registries, medical data bases, and surveillance populations?
We welcome your comments and other questions that you would like to discuss.
Posted by the Epidemiology Branch, NHLBI
We can all probably agree on 2 things:
1) Some important CER questions will not be answered by trials. This is especially true for interventions that need to be sustained over time and/or whose effects are only apparent after a long period.
2) Someone, somewhere will use observational data to try and answer those important CER questions
As a consequence, it is critical to establish standards for the design, analysis, and reporting of CER observational studies. Recently the journal EPIDEMIOLOGY asked some prominent experts to give their perspective on these issues. I introduced these commentaries by proposing that CER observational studies be evaluated using a modified version of the CONSORT criteria. See the link
http://journals.lww.com/epidem/Fulltext/2011/05000/With_Great_Data_Comes_Great_Responsibility_.3.aspx
The basic idea is that CER observational studies need to emulate hypothetical RCTs as closely as possible.
I think the rules of order suggested in Epidemiology are quite reasonable. Standardization, and ultimately access to the original data, is a necessity for this field to move forward.
I would just like to point out an interesting example from the Canadian experience which further highlights the power of population data sets for CER -” Physicians’ warning for unfit drivers and the risk of trauma from road crashes”. NEJM Sept 27, pp 1228. Basically, they had billing data which designated patients who had been told not to drive by their physician, and they then linked those data in a time series to ER visits. This is similar to the excellent work earlier in Canada on the efficacy of universal flu vaccination, where they did similar linkages. Unless I am missing something, the outcomes in both cases were unequivocal (ie, could not have been substantially improved through an RCT – which would have been expensive and likely impossible to carry out).
In a population-based health system with sufficient data the opportunities to evaluate CER are virtually limitless. Which is not to say, of course, that there is not at the same time an entire other universe of questions that require trial evidence. How one apportions their relative contribution, and cost effectiveness, I don’t know, but if I were forced to make a wild guess I would say it is something like 50-50. Data quality and bias of course are what give administrative and observational data limited credibility, but there are many very important questions where the outcome is – shall we say – crude enough to overcome the first objection, and if the data extend to the whole population then bias is reduced.