Key result
Federated hypertension risk models require ~52 clusters for adequate power under differential consent withdrawal.
Why the study?
Federated learning privacy guarantees for community clinic hypertension risk models usually assume a static cohort, whereas in practice patient consent withdrawal is non-random and clusters by clinic, age, and prior blood-pressure trajectory, inducing bias in the global risk model.
Provides a withdrawal-aware inference protocol to account for non-random patient consent withdrawal when training federated hypertension risk models across community clinics, showing that 52 clusters are required for 80% power at a 5-percentage-point effect.
Differential consent withdrawal biases federated hypertension models; leaves open whether static-cohort privacy guarantees require attrition adjustments before use.
Federated learning promises to train hypertension risk models across community clinics without centralising patient records, but its privacy guarantee is usually stated for a static cohort. In practice patients withdraw consent, and withdrawal is not random: it clusters by clinic, by age, and by prior blood-pressure trajectory. the present analysis formalises consent withdrawal as a differential attrition process operating on the federated estimator and derives the bias it induces in the global risk model. The analysis is a design-based evaluation of the before-after-control-impact estimator under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64), with power estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The results show that 52 clusters are required for 80% power at a 5-percentage-point effect, and that the minimum detectable effect rises steeply below 24 clusters. The contribution is a withdrawal-aware inference protocol that reports the consent-attrition sensitivity of every federated hypertension estimate alongside its confidence interval.
No takes yet. Share an insight, caveat, or question.
Waiss et al. (2026) studied Hypertension. Withdrawal-aware federated inference protocol under differential consent withdrawal was evaluated on Statistical power and minimum detectable effect under differential consent attrition. Federated hypertension risk modeling under differential consent withdrawal requires 52 clusters for 80% power at a 5-percentage-point effect, with minimum detectable effects rising below 24 clusters.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: