Development and internal validation of a nomogram for predicting bacterial meningitis in preterm infants with late-onset sepsis

Late-onset sepsis (LOS) is a common and severe condition in preterm infants and may progress to bacterial meningitis (BM), which is associated with high mortality and adverse neurological outcomes. Cerebrospinal fluid examination relies on LP, an invasive procedure that can be challenging in preterm infants. Therefore, an early and practical tool for identifying infants at high risk of BM is needed. This study aimed to develop and internally validate an exploratory nomogram for estimating the risk of BM among preterm infants with LOS who underwent LP. This retrospective single-center study included 275 preterm infants with LOS who underwent LP after LOS onset, including 53 infants with BM and 222 without BM. Candidate predictors were derived from maternal, neonatal, clinical, and laboratory variables. Univariable analysis and univariable logistic regression were initially performed for variable screening. Penalized regression methods, including LASSO, Elastic Net, and Ridge regression, were compared, and LASSO regression was used for predictor selection. Predictors with non-zero coefficients at the lambda.1se value were entered into a multivariable logistic regression model to construct the nomogram. Model performance was assessed by discrimination, calibration, and complete-pipeline bootstrap internal validation. Five predictors were retained in the final model: gestational hypertension, maternal age, gestational age, time from initial symptom onset to initiation of antibiotic therapy, and albumin. Gestational hypertension (OR 2.71, 95% CI 1.23–5.94, P = 0.013), maternal age (OR 1.10, 95% CI 1.02–1.17, P = 0.008), and time from initial symptom onset to initiation of antibiotic therapy (OR 1.04 per hour, 95% CI 1.02–1.06, P < 0.001) were positively associated with BM, whereas gestational age (OR 0.85 per 7-day increase, 95% CI 0.75–0.97, P = 0.019) and albumin (OR 0.90 per 1-g/L increase, 95% CI 0.84–0.97, P = 0.007) were inversely associated with BM. The apparent AUC was 0.805 (95% CI 0.75–0.86). After complete-pipeline bootstrap validation, the optimism-corrected C-index was 0.739 (95% CI 0.71–0.84) and the calibration slope was 0.734. We developed and internally validated a nomogram for estimating the risk of BM among preterm infants with LOS who underwent LP. The model retained moderate discrimination after optimism correction but showed evidence of overfitting. It may support preliminary risk stratification before definitive CSF results are available; however, it should not be used as a stand-alone tool to determine the need for LP or to provide definitive individual risk estimates. Prospective external validation and potential recalibration in a broader, consecutively enrolled LOS population are required before clinical implementation.