Rethinking the Evidence in Evidence-Based

Evidence-based medicine is conventionally defined as the conscientious, explicit, and judicious use of current best evidence in making decisions about the care of individual patients, integrating individual clinical expertise with the best available external evidence and the patient's values.

The definition has always contained the tension this chapter has been describing. It requires evidence to be external and best available, while requiring the decision to be about an individual. When the individual becomes specified enough, those two requirements begin to pull apart.

Consider what evidence is being asked to mean. The founding formulation was already a compound of three things, external evidence, clinical expertise, and the patient's values, and the movement's subsequent history is largely the story of the first crowding out the other two.

Two decades in, several of the discipline's own architects argued it had become a movement in crisis: the evidentiary brand captured by vested interests; the volume of guidance unmanageable; statistically significant benefits too small to matter promoted as though they were not; and rule-based management crowding out the judgment the original definition required.²² Their remedy was not less evidence but usable evidence, combined with context and expertise so that a particular patient gets an optimal decision rather than an average one.

That critique conceals a harder problem. Evidence in medicine is nearly always evidence about a class, while a decision is always about a member. A randomized trial licenses a claim about the average effect in the population sampled; transferring it to this patient requires a further judgment, that they resemble that population in the respects that matter, which the trial cannot supply and no hierarchy of evidence ranks.

The regnant approach ranks study designs, not their fit to the person in the room. Hierarchies answer how securely an average was estimated, not whether it applies here. Add the ordinary complications, among them publication bias, surrogate endpoints standing in for outcomes patients care about, trial populations unlike the patients later treated, and the near-total absence of the multimorbid, and the gap between what the evidence establishes and what the physician must decide is wide. More studies of the same kind will not close it.

For personalized medicine this is the central difficulty rather than a caveat. As specification increases, the reference class shrinks until it holds too few members to support an average, and the traditional hierarchy inverts. Mechanism, the patient's own prior responses, and a well-conducted N-of-1 experiment become better guides than a large trial run on people who differ in the relevant way. The N-of-1 trial therefore belongs at the top of a personalized hierarchy while sitting near the bottom of the conventional one.

For integrative medicine the difficulty runs the other way. Complex, multi-component, individually tailored interventions, such as a program of exercise, sleep, diet, and psychological work assembled for one person, are what the randomized trial handles worst. Blinding is often impossible, the intervention is not a fixed dose, and the effect depends on the recipient's engagement.

This has been used to argue for exempting such approaches from evidentiary scrutiny, and the argument should be refused. The correct response is different methods, not lower standards: N-of-1 and single-case designs; pragmatic and stepped-wedge trials; and outcomes capturing function and experience alongside biomarkers. Personalization and integration make the evidentiary problem harder for the same reason, since both insist on the particular person, and both will be judged by whether they answer that problem or evade it.

Inverting the hierarchy carries a consequence no individual physician can manage alone. Regulators approve on the conventional hierarchy, payers reimburse on it, guideline bodies write to it, and courts test the standard of care against it. A physician who reasons correctly under the personalized hierarchy, treating mechanism and a well-conducted N-of-1 result as better evidence for this patient than a large trial run on people who differ in the relevant way, is exposed under the old one and knows it.

The rational response to that asymmetry is to follow the guideline and document conformity, which is the behavior the inversion exists to correct. The remedy is institutional. Guideline bodies would have to state explicitly where their recommendations cease to apply, certifying boards would have to examine reasoning under a shrinking reference class, and the standard of care would have to recognize a defensible departure. Chapter XII returns to who could actually do this.