Answers are grounded in this paper and cite their sources.
Feature attribution methods are routinely used to justify readmission-risk predictions to clinicians and auditors, yet their stability under insurance coding revision remains unquantified. the present analysis treats attribution stability as a reliability property of the explanation pipeline and estimates it through a replication-based evaluation of the before-after-control-impact (BACI) estimator applied to a declared variance structure. No observational data are used or claimed: the analysis characterises the operating characteristics of the design under a between-cluster standard deviation of 6.0, a residual standard deviation of 4.5 and an intraclass correlation of 0.640. Power was estimated across 2000 design cells spanning cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design required 54 clusters to detect a 5-percentage-point shift at 80% power. The results indicate that attribution-stability audits built on the BACI estimator are feasible at realistic cluster counts only for effects of 8 percentage points or larger, and that the minimum detectable effect falls steeply as clusters increase from 8 to 32. Coding-revision audits should be pre-registered against a minimum detectable effect rather than a nominal significance threshold, and it specifies the reliability bound that such audits can defensibly claim.
No takes yet. Share an insight, caveat, or question.
Aïssa et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: