Population Averages Can Miss the Patient in Front of You
Evidence-based medicine changed clinical practice by replacing authority with appraisal, but its own methodologists have long pointed to a persistent tension: the average result in a trial does not always map cleanly onto the person sitting in the examination room. In this conceptual analysis, the central problem is framed as an ergodicity issue. Population-level estimates are treated as if they describe each patient, even though treatment response often differs across people and can shift within the same person over time.
The practical consequence is familiar at the bedside. A guideline can be statistically sound and still fit poorly for a single patient with unusual biology, unstable symptoms, or strong time-varying fluctuations in response. The paper argues that this mismatch is not a minor inconvenience; it is a structural feature of how group data are often applied in routine care.
Why the Mean Does Not Automatically Apply to One Person
Ergodicity holds only when a process is homogeneous across individuals and stationary over time. Medicine rarely enjoys either condition for long. Treatment effects vary between patients because genetics, comorbidity, adherence, physiology, environment, and disease stage all shape response. Even within one patient, the effect of a therapy can drift as symptoms evolve, competing risks emerge, or the underlying disorder changes its pattern.
In that setting, a group average can be clinically elegant but personally incomplete. Relative effects are especially difficult to translate into absolute benefit for an individual, and the usual trial summary gives no built-in stop rule for a therapy that is not helping. The paper treats those gaps as consequences of assuming the wrong level of analysis rather than merely using the wrong statistic.
A treatment can be “effective” for a population and still be a poor fit for a given patient.
Individual Variability Must Be Measured, Not Assumed Away
The argument turns on a simple clinical fact: when response is reproducible within an individual, individual-level inference becomes both necessary and feasible. Where within-person variability is large, every inference weakens unless that variability is directly measured. The paper resists the habit of smoothing away instability and instead places instability itself at the center of decision-making.
This is where the reference class problem becomes clinically relevant. Two patients can belong to the same diagnostic category and still behave differently enough that identical treatment assumptions become misleading. Population evidence remains essential, but it cannot by itself solve the question of whether this specific patient benefits, harms, or remains unchanged.
The manuscript’s methodologic stance is conservative rather than iconoclastic. It does not reject controlled evidence. It argues instead that controlled evidence should be divided into two tasks: establishing what tends to work in groups, and determining what works for an individual over time.
N-of-1 Trials Return to the Center of Decision-Making
The proposed corrective is the N-of-1 randomized trial, a design that estimates treatment effect in a single patient. The paper identifies this as an existing proof of concept rather than a speculative invention. Guyatt and Sackett published the approach in 1986, and the Oxford 2011 Levels of Evidence rank it at Level 1. Even so, a systematic review found only 108 published trials over twenty-five years, a number that exposes a major gap between methodological promise and clinical uptake.
That gap matters because N-of-1 trials answer a question ordinary trials cannot answer with the same precision: does this intervention help this person, now, under these conditions? In chronic disease care, pain management, and cardiology, such within-person measurement already appears in practice. The paper uses those examples as existence proofs that individual measurement and closed-loop intervention are not theoretical luxuries but working clinical tools.
Individual-Dynamic Medicine as a Biocybernetic Framework
The proposed framework, called Individual-Dynamic Medicine, consists of six principles. Each principle is anchored to an existence proof in mainstream medicine and is intended to define how population evidence and within-person trials divide responsibility in a single clinical decision. The framework is described as biocybernetic, meaning it treats medicine as a feedback system in which the patient’s response informs the next therapeutic move.
That structure is especially relevant in disorders where treatment effects are unstable, delayed, or sensitive to context. A static prescription model assumes that one decision can hold across time. A dynamic model accepts that the decision may need revision as the patient’s own data accumulate. In other words, the patient becomes both the recipient of care and the source of the most relevant evidence for continued care.
This is not a call to discard guidelines. It is a call to stop pretending that guidelines alone can settle every individual case. Population estimates can identify candidate treatments, expected harms, and likely benefit ranges. Individual trials can then confirm whether the expected pattern actually appears in one patient’s lived trajectory.
What Would Make the Proposal Fail
The paper does something many theoretical essays avoid: it states the assumptions, limitations, and predictions under which the proposal could fail. That is a useful discipline because any framework that claims to improve precision must also specify when precision will not be possible. If responses are not reproducible within a patient, if within-person variability overwhelms signal, or if the clinical state changes too quickly for sequential measurement to be meaningful, then the value of the model narrows.
Still, the conceptual message is sharp. The average patient does not exist as a bedside decision-making unit. Clinicians care for one person at a time, and therapeutic success or failure is experienced one patient at a time. A system that measures only the group average risks missing the moment when a treatment is helping no one, helping only a subset, or helping a patient for a limited interval before the effect disappears.
The paper places evidence-based medicine on firmer philosophical ground by separating population inference from individual inference rather than conflating the two. Its main challenge to current practice is straightforward: when the clinical question is about one person, the evidence strategy should be designed to answer a one-person question.