PORTRAIT: a calibrated patient ‘Passport’ with built-in refusal - describing individuals against a reference population

Background Average-based summaries serve individual patients poorly. PORTRAIT is a calibrated, abstention-aware tool that describes where one patient sits relative to a reference population across 12 cardiometabolic markers, how confident that placement is, and which features drive it. PORTRAIT describes; it does not diagnose or predict. Abstention is a designed feature given the known limits of conditional coverage.

Methods Conformal calibration was combined with distribution-free coverage bounds, quantile-regression coordinates, and copula-based joint structure. A frozen reference cohort (n=9421) supplied fixed calibration; a held-out cohort (n=2247) tested transportability across six strata. A release gate required the minimum per-slice coverage to hold across 4 of 5 seeds. Coverage was re-tested under survey weighting to the US adult population. Coherence was reported as a descriptive joint coordinate. Discrimination was summarised with Harrell’s C and multiplicity controlled by BH-FDR. Interface conformance was assessed against defined requirements, Nielsen heuristics, and WCAG 2.2 AA, with attention to automation bias and risk-graph design.

Results Estimating the conformal quantile once on the deep frozen reference held all six strata within band (0.864-0.903) at abstention 0.113, whereas a local re-split - calibrating each stratum on far fewer rows - under-covered to 0.71 at abstention 0.227; the difference is a calibration-depth effect, and coverage survived survey weighting. The release gate passed on 4 of 5 seeds at abstention 0.101 against a nominal 0.90, and in the frozen-reference configuration that ships, all six strata held inside the calibrated band (0.864-0.903). Coherence showed orthogonality 0.444 to raw extremity and correlated 0.892 with a copula-Mahalanobis distance while remaining deliberately non-identical, so it adds per-feature information. Two transfer tests returned negatives; the ocular transfer did not hold coverage at thin-n. Adding coherence changed mortality discrimination by deltaC 0.0047. Interface requirements moved from 14/27/18 to 38/14/7 (MET/PARTIAL/UNMET); Nielsen severity resolved 7 of 10 issues; WCAG 2.2 AA text criteria passed.

Conclusions PORTRAIT situates a patient against a frozen reference, holds coverage under survey weighting to the US adult population, and abstains when calibration cannot be supported. The headline result is a calibration-depth effect: estimating the conformal quantile once on the deep frozen reference held per-stratum coverage on a disjoint cohort.