Of course, the question will arise as to who should be completing the analysis of these data. i suggest that this should be the true role of the proposed Institute for Comparative Effectiveness Research. It is only through the analysis of millions of records, examined with true longitudinal integrity, that we will be able to arrive at reliable results for the multitude of relevant subsets. In effect everyone becomes an observed trial participant without having to undergo any modification to their treatment, location of care or anything else. It also allows for proper analysis of different doctor behavior without creating a false environment within a trial.

The only problem is that those proposing the ICER are already salivating at the thought of the huge budgets that will be involved. The alternative idea is to take the proposals already around for the use of EMRs as standard, feed the anonymized data into an analytical capability and derive real knowledge from real behavior instead of pseudo-knowledge from “controlled” trials.

For example, in clinical practice patients are placed on antidepressants for 6 months, even with a first episode. This is not considered long term treatment for prevention of a second episode. Yet when there was a proposal for FDA to extend the requirement for antidepressant studies from 3 months to 6 months drug companies and the Key Opinion Leaders who consult for them got the proposed requirement killed.

Safety is another big issue. Current ICH requirements are 500 subjects at ANY clinically used dose for 6 months and 100 subjects for 1 year and was implemented around the time Phen-Fen and Zyprexa was under development. This could definitely be increased. In addition there is no reason why these subjects could not continue to be studied as it takes approximately 18 months from the completion of phase III studies until a drug is approved.

Thus even if you have 100 subjects at 1 year at filing you might have several hundred at 1 year by the time of approval and another couple of hundred on the drug for up to 2.5 - 3 years. It’s not perfect but it’s a lot better than we have now.

Salmon

You have to keep in mind that when you attempt to generate increased external validity (a “real world” approach that is applicable to the general population), you do so by sacrificing internal validity (the ability for an experiment to accurately measure what you’re attempting to measure).

This is a problem with science, generally, not just the medical sciences. The more you control an experiment, the less “real world” that experiment becomes and the more detached and artificial the results. The less you control an experiment, the less you know what you’re really measuring.

As a counter point to your statement, some argue that medical studies are messy enough as they are currently carried out. Perhaps what we need is a better general understanding of how to control medical experiments–and incentives for conducting unlikely experiments (e.g. head to head studies, non-biased pharmacoeconomic analyses, long-term studies in properly powered sub-populations, etc.).