External Validation of a Logistic Regression Model to Predict Cesarean Section Risk in Indian Women

BACKGROUND: Clinical prediction models for lower segment cesarean section (LSCS) require external validation before they can be reliably used across diverse obstetric populations. This study assessed the performance of an existing LSCS risk prediction model in an independent cohort and examined whether recalibration improved model fit. METHODS: The previously developed logistic regression model was applied to a separate external dataset without refitting. Predicted probabilities were generated using the original coefficients, and model performance was evaluated through discrimination (area under the curve [AUC]), calibration (Hosmer-Lemeshow [HL] test), and classification metrics at both the default probability threshold (0.50) and the optimal cutoff was identified using Youden's Index. Intercept-only recalibration was undertaken to correct calibration drift. RESULTS: In the external cohort, the model demonstrated good discrimination (AUC: 0.82) and satisfactory calibration (HL χ2 = 4.26, P = 0.3719). At the 0.50 threshold, accuracy was 81.7%, sensitivity 90.0%, specificity 73.3%, and kappa 0.63. The optimal cutoff of 0.919 maximized sensitivity (96.7%) with acceptable specificity (70%). Recalibration yielded an updated intercept of -3.029 and improved alignment between predicted and observed risks. CONCLUSION: The LSCS prediction model retained strong performance in an external population, and recalibration further enhanced calibration. These findings support the model's applicability across clinical settings, with potential utility for antenatal risk stratification.