recent JAMA article which contrasts evidence-based medicine with “personalized” medicine (see evidencebased medicine jama 2013 doi below). basically contrasts EBM and its focus on randomized controlled trials with targeted therapies (not just genomics, but trying to contour therapy to the specific patient being treated). one of the obvious problems with EBM is through the basic reductionism of medical studies: enrolling many patients who are x% male of x average age and x% white who have a specific disease but with specified inclusion and exclusion criteria and get a medication vs placebo. so, it is often a bit of a leap to apply the conclusion of the study to a young mostly healthy capeverdean woman (not part of the cohort of the study) or an older latino man with chronic kidney disease and lung cancer in remission (excluded by ethnicity and chronic diseases). and one of the issues is that guidelines are based on these studies and providers are evaluated/held accountable according to adherence to these guidelines, without clear evidence they really apply to the individual patient being treated. one other issue is that many of the RCTs end up having secondary/subgroup analyses done afterwards, which do give some insight into more specific and patient-based questions (more pesonalized), but these conclusions (correctly) are not considered to be statistically rigorous and are not generally incorporated into guidelines. this JAMA article gives the example of the MADIT-II study, which when it came out created some consternation since it seemed to me that several of my patients should have an implantable defibrillator (ICD). in brief,
MADIT-II enrolled 1200 pts s/p MI and EF<30% into getting ICD vs placebo and found a 31% decreased mortality in those with ICDs. however, a subsequent analysis (see cad madit2 risk stratification jacc 2008 doi below) did a detailed analysis of many variables to assess outcomes, boiling it down to 5 which were statistically associated with mortality (age>70, BUN>26, NYHA class >2, afib, and QRS>0.12). then they looked at the number of these risk factors and efficacy of ICDs. they found:
–those at very high risk (5% of cohort, basically those with many of the above risk factors) had a 48% mortality over about 2 years, independent of whether they got an ICD
–those with no risk factors (1/3 of their group) also had no better results with ICD.
–but those with 1-2 risk factors had a 49% mortality benefit
–although there was low mortality and short followup of the low risk group, there are other studies (DINAMIT, IRIS) which do not show ICD benefit and (again) posthoc analysis shows most of the benefit occurs early on (eg in first 18 months, with no clear benefit after 1-2 years)
so, this type of posthoc analysis added huge clinical information about an expensive and risky intervention, though is not statistically rigorous and is not included in the current guidelines. but the design of the original trial concluded (and is incorporated into guidelines) that pts s/p MI and EF<30% should have ICDs. not so helpful to us in primary care. should we really refer all of these patients? what if they have significant renal failure (not included in MADIT-II study, and we know they have much higher mortality rates?) what if we decide not to treat either a very low or high risk patient in spite of the guidelines — what about the malpractice risk if there is an adverse event? the ICD manufacturer was a major sponsor for the study – what is their role in designing the study to create the most favorable, generalizable conclusion (ie, not in their direct interest to have a very narrow, limited cohort of patients)?
geoff
cad madit2 risk stratification jacc 2008 – doi:10.1016/j.jacc.2007.08.058
evidencebased medicine jama 2013 – doi: 10.1001/jama.2013.6629
(search the doi to locate the article)