JMIR Publications

ABSTRACT

Background:

Cesarean section (C-section) is the most common surgical procedure in the United States, yet its use varies widely across regions and institutions. While clinical risk factors are important, growing evidence suggests that delivery decisions are also shaped by geographic context and health system characteristics. Understanding these systemic drivers is critical for improving obstetric equity and aligning C-section use with clinical need.

Objective:

To examine the extent of geographic and institutional variation in C-section use across the United States and to evaluate the performance and implications of machine learning–based risk prediction across clinically defined risk groups.

Methods:

This population-based study analyzed 38·1 million U.S. births from the 2013–2022 National Vital Statistics System Natality Detailed Files. Interpretable logistic regression models with county fixed effects and supervised machine learning approaches were used to examine factors associated with C-section delivery and to predict C-section risk in the full population and in low-risk and high-risk subgroups. County-level health system capacity and socioeconomic characteristics were linked from national administrative sources. Model performance was assessed using accuracy, recall, and area under the receiver operating characteristic curve (AUC), with probability thresholds optimized to balance overall performance and clinical relevance.

Results:

C-section rates exhibited substantial and persistent geographic variation, with consistently higher utilization in the U.S. South despite a modest national decline over time. After adjustment for detailed maternal risk factors, county fixed effects explained a large share of residual variation. Machine learning models demonstrated good predictive performance (AUC 0·81–0·83), but performance varied markedly across risk groups, with substantially lower accuracy in low-risk pregnancies. Threshold optimization revealed clinically meaningful trade-offs between sensitivity and specificity.

Conclusions:

C-section use in the United States is strongly influenced by geographic and institutional context in addition to maternal risk. Policies narrowly focused on low-risk populations may fail to address—and may inadvertently exacerbate—system-level drivers of variation. Risk-adjusted, context-aware prediction tools may support more equitable and clinically appropriate obstetric decision-making.