Antenatal corticosteroid (ACS) administration reduces neonatal mortality in preterm birth, but its effectiveness depends on accurate gestational dating. In community clinics without ultrasound, gestational age is estimated from last menstrual period or symphysis-fundal height, both of which carry substantial error. The framework treats clinic-level variation in ACS availability as an instrument, establishes identification conditions under which the local average treatment effect (LATE) is recoverable despite misclassification of gestational age, and derives bias bounds for the case where the exclusion restriction is partially violated. A replication-based sensitivity analysis, calibrated to published estimates of dating error variance, quantifies the conditions under which IV estimates remain informative. IV inference can recover the direction and approximate magnitude of the ACS timing effect even when dating error is severe, provided that the instrument is strongly predictive of treatment and that the proportion of always-takers is modest. The framework is positioned as a methodological contribution to causal inference in low-resource maternal health settings, with implications for the design of clinic-level ACS programmes and the interpretation of observational safety data.
No takes yet. Share an insight, caveat, or question.
Andrianarisoa et al. (2026) studied this question.