Blood biomarkers and penalized logistic regression for the prediction of complicated appendicitis in children: a retrospective cohort study
Abstract
Background: Early preoperative identification of complicated appendicitis in children is important for risk stratification. Several blood biomarkers have been proposed for this purpose, but their performance has rarely been compared directly within the same cohort. We aimed to clarify their relative diagnostic performance and to determine whether combining laboratory parameters using an elastic-net model provides incremental diagnostic value. Materials and methods: This single-center, retrospective observational study included children younger than 16 years who underwent emergency appendectomy between January 2015 and December 2024. Preoperative laboratory parameters and composite biomarkers were evaluated using receiver operating characteristic analyses, and pooled area under the curve (AUC) estimates were calculated after multiple imputation using Rubin’s rules. AUC values were interpreted as measures of discriminative performance, with higher values indicating better discrimination. Pairwise AUC comparisons were performed using DeLong’s test with Holm adjustment. All analyses were performed using R version 4.5.2. Results: Among 308 patients, 109 (35.4%) had complicated appendicitis. The C-reactive protein (CRP)-to-albumin ratio (CAR) showed the numerically highest AUC among the evaluated biomarkers [0.852; 95% confidence interval (CI), 0.800–0.892], followed by CRP (0.848; 95% CI, 0.795–0.889), chloride (0.817; 95% CI, 0.763–0.860), sodium (0.765; 95% CI, 0.703–0.818), and albumin (0.715; 95% CI, 0.649–0.773). DeLong comparisons showed no significant difference between CAR and either CRP or chloride (P = 0.19 for both), whereas CAR significantly outperformed sodium and albumin (P = 0.02 and P < 0.001, respectively). The elastic-net model achieved a pooled AUC of 0.871 (95% CI, 0.824–0.906) and provided only modest additional discrimination over CAR. Conclusion: CAR showed robust diagnostic performance and may provide a practical balance between accuracy and simplicity for early risk stratification of complicated appendicitis in children. Laboratory-only multivariable modeling provided limited additional value, suggesting that further improvement may require integration of clinical and ultrasonographic findings.