Working Paper: NBER ID: w22643
Authors: Charles F. Manski
Abstract: This paper studies the ecological inference 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). In practice, it is common to have a trustworthy evidence-based 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.
Keywords: personalized medicine; ecological inference; clinical judgment; risk assessment
JEL Codes: C18; I10
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| observed attributes (C90) | conditional probabilities (pyxk) (D80) |
| conditional probabilities (pyxk) (D80) | patient outcomes (pyxk w j) (I14) |
| prevalence of observed attributes (pw jxk) (C46) | patient outcomes (pyxk w j) (I14) |
| knowledge of conditional probabilities (pyxk) and prevalence of observed attributes (pw jxk) (D80) | tighter conclusions about patient outcomes (pyxk w j) (I11) |
| structural assumptions (P16) | credibility of predictions (C53) |
| formalizing clinical judgment (D91) | improved personalized risk assessments (D91) |
| actuarial methods (C58) | accuracy of outcomes (C52) |