Artificial Intelligence replication of District Hospital Postpartum Haemorrhage Risk Scoring Under Unmeasured Blood Loss Visual Estimation Error: A Healthcare Calibration-Fairness Evaluation

Visual estimation of postpartum blood loss in district hospitals is known to be biased, yet risk scores for postpartum haemorrhage are routinely calibrated on measurements that already contain this error. the present analysis treats the calibration of such scores as a problem of estimation under a mismeasured covariate and asks what happens to bias, interval coverage and threshold-based fairness when the measurement error is left unmodelled. A replication-based evaluation was conducted: 600 data sets of N = 500 were drawn from a declared data-generating process with intercept 12.0, exposure 0.42, group -3.1, severity 1.35 and error SD 4.2, and the ordinary least squares estimator was characterised across replications. The estimated bias for the intercept was 0.03992 and for the exposure coefficient -0.00125; the root mean squared errors were 0.90134 and 0.0166 respectively, and empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The results indicate that under the stated process the estimator is essentially unbiased and its intervals are conservative, so the calibration problem in this setting is not one of point estimation but of the fairness consequences that follow when a score is thresholded. Calibration-fairness evaluation of postpartum haemorrhage risk scores must separate the behaviour of the estimator from the behaviour of the decision rule, and it derives the conditions under which unmeasured visual estimation error becomes consequential for group-specific sensitivity.