The basic message is "everyone does not react the same to a given treatment".
This should be obvious but in the frenzy of jumping on the Evidence Based Band wagon, it may have been forgotten or overlooked.
An excellent article that should be part of medical school teaching was published in a journal that no one would have ever read in the pre-internet era and even now probably languishes on the shelves of unread papers. ( I would not have found it had it not been mentioned in a December 2004 blog on Health Care Renewal).
The Journal is Milbank Quarterly and the article is " Evidence-based Medicine;Heterogeneity of treatment effects and the trouble with averages" by Kravitz,RL,Duan N, and Braslow J.
Here is a brief outline of the main points but the entire article is worth reading and should be handed out to medical students as part of their course in Statistics or EBM or something.
Heterogeneity of treatment effects (they abbreviate it as HTE) simply means
that the basic fact of life in a clinical trial is that everyone does not respond the same to the treatment.If , for example the average blood pressure decrease is x, there will be patients whose BP change is greater than that, less than that and in some patients some type of harm actually occurs.
The clinical problem is how to apply this population data to the individual patients.
The corresponding statistical problem is heterogeneity of treatments effects.
This variation in turn is due to the following three factors, acting alone or interacting with each other: 1) the degree to which an individual disease is likely to progress (i.e.the risk without treatment) 2) the responsiveness of the patients to treatment (this sounds a bit circular,but refers to how well does the drug favorably work in a given patient), 3) vulnerability to side effects.In the future to a greater degree than now we may be aware of various genetic polymorphisms that might explain some that variation and allow us to predict better who will respond to what.Additonally-in deciding should a particular treatment be recommended- there is the issue of utilities for different outcomes, meaning various patients value the same outcome differently.
All of this means you cannot expect the "average effect" to occur in the given patient you are treating. This is the basic clinical conundrum.
On the one hand the RCT may be the best information the doctor has since he lacks specific information on the patients' risk without treatment and susceptibility to positive or negative effects on treatment.
So what do you do? 1.Realize that the RCT is better guidance that just "winging it" completely, 2. As best you can try and determine to what degree your patient is like the average of the group treated and/or like or dissimilar from any subgroup that was analyzed in the study ( yes, we know subgroup analysis can be fallacious but..) and remember that your patient who is at greatest risk from the disease is likely to get the greatest net benefit from treatment. Also ask would your patient even been eligible for the trial and if not was the reason likely to be clinically important.It is a clinical judgment issue and for that there is no meta-algorithm.The physician's ability to predict how a given patient will respond to treatment X depends on our knowledge of those 3 factors.
So the student should remember that the often modest benefit that occurs on average in a RCT typically reflects a mixture of significant benefit for some, slight change for many and harm for some. There is a lot of clinical context and patient preference to factor in and blind adherence to guidelines - be it for pay-for performance or some sort of quality ranking-is not the way to go.
There are obvious implications for guideline makers and the pay- for- performance issues and even FDA approval for drugs,to all of which the authors address comments.