Artificial Intelligence Distributionally Robust Prediction of Maternal Sepsis Deterioration Under Variable Vital-Sign Charting Interval Irregularity: A Conformal Evaluation of Healthcare Escalation Equity

Maternal sepsis deterioration models are typically trained and evaluated on vital-sign series recorded at near-uniform intervals, yet in routine obstetric care charting intervals vary with staffing, ward acuity and time of day. the present analysis treats interval irregularity not as a missing-data nuisance but as a distributional shift that can systematically distort escalation decisions across patient groups. We formalise irregular charting as a covariate shift over observation-time densities, define a distributionally robust prediction rule over an ambiguity set of charting-interval distributions, and wrap the rule in a split-conformal calibration layer that yields group-conditional escalation guarantees. The framework is methodological: no observational cohort is analysed. We characterise the operating characteristics of the proposed estimator under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64) using a before-after-control-impact design with 2000 replications per cell across cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points.52 clusters are required for 80% power at a 5-percentage-point effect, and that conformal coverage degrades predictably as charting-interval heterogeneity widens. The contribution is a formal bridge between distributional robustness, conformal prediction and escalation equity that specifies when a deterioration model's guarantees transfer across charting regimes.