The Methodological Pitfalls Complicating the Analysis of Early Term Births

ABSTRACT Background Evaluating the success of policies aiming to reduce early term birth (37 +0 –38 +6 weeks' gestation) is analytically challenging. Objectives Determine the best method of reporting temporal changes of early term birth rates. Methods We analysed 54,646 singleton births at Western Australia's main tertiary hospital (2009–2019). Using interrupted time series, segmented and nominal regression, we assessed how varying gestational age outcomes, denominator, or statistical method affected conclusions regarding temporal changes (pre and post‐Initiative implementation) in the timing of birth. Results Interrupted time series showed yearly rates of birth at 37 +0 –38 +6 weeks increased pre‐Initiative, with no instant level change or change in the slope post‐Initiative, regardless of denominator: all births (slope change −0.0, 95% CI −0.5, 0.5); births > 36 +6 weeks (slope change: −0.4, 95% CI −1.0, 0.2), and no change to the number of early term births averted. Conversely, when examining births < 39 +0 weeks, pre‐Initiative rates increased by 0.9% (95% CI 0.5, 1.3) per year, then stabilised post‐Initiative (slope change: −0.8, 95% CI −1.4, −0.3), with no change to the number of births < 39 +0 weeks averted. Nominal logistic regression showed that pre‐Initiative, the rise in the early term birth rate was driven by a reduction in births at ≥ 39 +0 weeks, and post‐Initiative by a reduction in preterm birth. Using segmented quantile regression, lengthening of gestational age at birth in days was observed more so at the lower percentiles, compared to the estimated counterfactual. Conclusion Examining early term birth rates in isolation can mask changes in birth timing in the presence of a changing preterm or ≥ 39 +0 birth rate. Researchers should examine all births < 39 +0 weeks rather than just early term births, or use nominal logistic regression. For gestational age in days as the outcome, segmented quantile regression can be used.