Pretest Probability in Ophthalmology: Preserving Diagnostic Calibration While Moving Beyond Race

Using a Bayesian framework with emphasis on pretest probability, to evaluate the consequences of removing race-based proxies from clinical reasoning without replacing the information they encode with more precise causal predictors.

Narrative review and conceptual analysis of Bayesian diagnostic reasoning as applied to current recommendations regarding the use of race and ethnicity in medicine and ophthalmology. Examples are presented in which suppression of population-level proxies alters pretest probability estimation and diagnostic calibration. Representative conditions discussed include sarcoidosis-associated uveitis, giant cell arteritis, and hemoglobinopathy-associated vitreous hemorrhage. A review is presented of the French experience with restrictions on race and ethnicity data collection in healthcare and the ensuing problems encountered and costs incurred.

Bayesian diagnostic theory predicts that test performance depends on pretest probability. Uniform diagnostic thresholds applied across populations with differing disease prevalence may increase low-yield testing in low-prevalence groups while delaying diagnosis in higher-risk populations. Race is a socially constructed category rather than a biological determinant; however, in current clinical practice it may function as a proxy for genetic susceptibility, environmental exposure, geographic ancestry, access to care, referral patterns, and diagnostic delay. Suppressing such proxies without substituting more precise causal variables may degrade diagnostic calibration. Examples from ophthalmology demonstrate how inaccurate estimation of baseline disease probability can lead to both unnecessary testing and avoidable morbidity. Eliminating population descriptors has historically forced clinicians toward less transparent and less precise surrogate markers.

The goal should not be preservation of race-based medicine, but preservation of accurate and equitable diagnostic calibration while transitioning toward mechanism-based predictors. More precise variables, including genetic markers, environmental exposures, and structural determinants of health, should replace imprecise proxies whenever possible. Until such replacements are routinely available, eliminating variables that contribute to estimation of disease probability risks impairing Bayesian clinical reasoning and increasing both over-testing and under-diagnosis.