Answers are grounded in this paper and cite their sources.
Transformer-based malaria diagnosis models are increasingly proposed as substitutes for rapid diagnostic test (RDT) readers in low-resource clinics, yet the interpretability audits used to certify them assume a stable reader population. the present analysis formalises the calibration fairness problem that arises when an RDT reader is substituted mid-deployment and the audit must decide whether the model's attention maps and calibration curves remain valid for the new reader. The contribution is a design-theoretic analysis of the before-after-control-impact (BAC I) audit estimator under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64). Power was estimated across 2000 design cells for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points.54 clusters are required for 80% power at a 5-percentage-point effect, and that the minimum detectable effect at 80% power falls steeply as cluster count rises. Interpretability audits of substituted-reader deployments are underpowered at the cluster counts typically available in district malaria programmes, and it derives the audit-design conditions under which a calibration-fairness claim can be sustained.
No takes yet. Share an insight, caveat, or question.
Ssentongo et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: