Derivation and Validation of a Prediction Model for Adverse Perinatal Outcomes Among Women of Advanced Maternal Age: A Population-Based Study Using Florida Birth Records (2005–2023)

Abstract

Background: Advanced maternal age (AMA; ≥35 years at delivery) is associated with elevated risk of adverse perinatal outcomes (APO), including preterm birth, abnormal birth weight, gestational hypertensive disorders, gestational diabetes, and cesarean delivery. Despite the growing proportion of AMA births and Florida's perinatal mortality rate exceeding the national average, a comprehensive statewide assessment and population-level prediction model for APO among AMA women remained lacking.Methods: This population-based retrospective study used Florida Bureau of Vital Statistics birth records from January 1, 2005, to December 31, 2023. All singleton and multiple births to women aged ≥35 years were included. After excluding records with missing outcome data and biologically implausible values, the analytic sample comprised 706,511 births. The primary outcome was a composite APO defined by the presence of at least one fetal, neonatal, or maternal morbidity event from 20 weeks of gestation through 28 days after delivery. The study has three aims: (1) prevalence trends analysis over the 19-year period; (2) identification of risk and protective factors and assessing the effect modification by major social determinants of health such as race/ethnicity, nativity, and insurance status; and (3) development and validation of a prediction model for APO. Missing data were addressed using Multiple Imputation by Chained Equations (MICE). Covariate selection was guided by a directed acyclic graph, and a multivariable logistic regression was used for Aim 2. For Aim 3, LASSO-penalized regression with 10-fold cross-validation was used as the primary variable-selection approach. Prediction model performance was assessed using discrimination (AUC), calibration, and decision curve analysis. Reporting followed STROBE and TRIPOD guidelines. Results: Of 706,511 births analyzed, 224,187 (31.7%) experienced at least one APO. A clear dose-response pattern was observed, with APO prevalence of 31% among women aged 35–39 years, 35% among 40–44 years, and 46% among ≥45 years (p < 0.001). Annual APO prevalence increased from 28.7% in 2005 to 36.4% in 2023 (annual percentage change 1.37%), with a structural breakpoint identified around 2013–2014. In fully adjusted logistic regression, women aged 40–44 years had 1.16 times the odds of APO (aOR = 1.16; 95% CI: 1.14–1.17), and those aged ≥45 years had 1.64 times the odds (aOR = 1.64; 95% CI: 1.57–1.72), compared to women aged 35–39 years. Black race (aOR 1.17–1.22) and Asian race (aOR 1.54–1.56) were associated with higher odds relative to White race. Foreign-born nativity (aOR ≈ 0.87–0.88) and higher educational status were protective. Pre-pregnancy obesity (aOR 1.81; 95% CI: 1.78–1.84), previous preterm birth (aOR ≈ 2.10–2.23), and use of fertility treatment (aOR 1.8–2.0) were prominent predictors. No meaningful effect modification was detected for age-by-race, age-by-nativity, or age-by-insurance interactions. The final prediction model included 21 LASSO-selected predictors. In the random-split validation cohort (n = 211,986), the model achieved an AUC of 0.693 (95% CI: 0.690–0.696), excellent calibration (slope = 1.002; intercept = −0.001), a Brier score of 0.1895, and a negative predictive value of 0.790. Prespecified transportability criteria were fully satisfied (ΔAUC = 0.000; ΔBrier = −0.0002). Decision curve analysis demonstrated positive net benefit across threshold probabilities of approximately 10–50%. Conclusions: Adverse perinatal outcomes were highly prevalent and rising among AMA women in Florida from 2005 to 2023. Approximately one-third of this population had experienced at least one APO, and this proportion has increased with advancing maternal age subgroups, and the prevalence of APO has increased over the study period. Several potentially modifiable predictors, including pre-pregnancy obesity, fertility treatment exposure, inadequate gestational weight gain, and tobacco use, represent actionable targets for preconception optimization and tailored prenatal care. The final parsimonious prediction model demonstrated acceptable discrimination, excellent calibration, favorable clinical utility, and stable transportability across time periods. The finding supports its potential use for prenatal risk stratification among AMA women pending external validation. Keywords: advanced maternal age, adverse perinatal outcomes, prediction model, LASSO, Florida vital statistics, social determinants of health