Credible Ecological Inference for Personalized Medicine: Formalizing Clinical Judgment (WP-16-19)


Charles F. Manski

This paper studies an identification problem that arises when clinicians seek to personalize patient care by making health risk assessments conditional on observed patient attributes. Let y be a patient outcome of interest and let (x = k, w = j) be patient attributes that a clinician observes. The clinician may want to choose a care option that maximizes the patient's expected utility conditional on the observed attributes. To accomplish this, the clinician needs to know the conditional probability distribution P(y|x = k, w = j). It is common to have a trustworthy risk assessment that predicts y conditional on a subset of the observed attributes, say x, but not conditional on (x, w). Then the clinician knows P(y|x = k) but not P(y|x = k, w = j). Partial conclusions about P(y*x = k, w = j) may be drawn if the clinician also knows P(w = j|x = k). Tighter conclusions may be possible if he combines knowledge of P(y|x) and P(w|x) with credible structural assumptions embodying some a priori knowledge of P(y|x, w). This is the ecological inference problem studied here. A substantial psychological literature comparing actuarial predictions and informal clinical judgments has concluded that clinicians should not attempt to subjectively predict patient outcomes conditional on attributes such as w that are not utilized in evidence-based risk assessments. The analysis in this paper suggests that formalizing clinical judgment through analysis of the inferential problem may enable clinicians to make more informative personalized risk assessments.

Charles F. Manski, Board of Trustees Professor in Economics and IPR Fellow, Northwestern University

PDF icon Download working paper PDF