Machine Learning Model Predicts Complicated Appendicitis in Adults Across Multicenter Cohorts

A multicenter retrospective cohort study of 1,573 adult patients across four hospitals has demonstrated that a machine learning support vector machine model can predict complicated appendicitis before surgery, although researchers found that local recalibration is mandatory prior to clinical deployment.

Accurate differentiation between complicated appendicitis and uncomplicated appendicitis is clinically important in the nonoperative management of acute appendicitis in adults. However, multicenter evidence regarding risk stratification for complicated appendicitis has remained limited.

The study sought to develop and externally validate a machine learning model integrating routine clinical, laboratory, and computed tomography information for preoperative risk stratification. Across the multicenter cohort, 652 patients from center 1 were used for model development and randomly assigned to a training cohort and an internal validation cohort at a 7:3 ratio. Model evaluation further incorporated 438 patients from center 2 as external validation cohort 1, along with 483 combined patients from centers 3 and 4 as external validation cohort 2.

Researchers trained candidate models using 10 machine learning algorithms and optimized them through cross-validation. Discrimination, calibration, and clinical net benefit were evaluated using the area under the receiver operating characteristic curve, Brier score, and decision curve analysis, respectively. Based solely on cross-validation results from the training cohort and a prespecified model selection strategy, the support vector machine model was selected as the final architecture.

In internal validation, the support vector machine model achieved an area under the receiver operating characteristic curve of 0.910 with a 95 percent confidence interval of 0.840 to 0.968, exhibiting good calibration and clinical net benefit. In external testing, the model reached an area under the curve of 0.797 (95 percent confidence interval, 0.743 to 0.846) in external validation cohort 1 and 0.829 (95 percent confidence interval, 0.776 to 0.879) in external validation cohort 2. Both external validation cohorts displayed substantial calibration drift; after intercept-only recalibration, Brier scores decreased from 0.16 to 0.14 in cohort 1 and from 0.14 to 0.12 in cohort 2.

Model interpretation using SHapley Additive exPlanations indicated that fibrinogen, the neutrophil-to-lymphocyte ratio, and pain duration were important predictors of complicated disease. A web-based online prediction platform was established to provide individualized risk estimates. The authors concluded that while the model may support preoperative risk stratification for complicated appendicitis in adults, the original unrecalibrated model must not be applied directly to local populations, requiring local recalibration, local validation, and prospective evaluation of clinical impact.

In evaluating the broader clinical picture, the multicenter framework reflects the challenges clinicians face when attempting nonoperative pathways for acute appendicitis in adult patients. The investigation established that routine clinical features, basic laboratory indices, and computed tomography findings can be systematically unified into an algorithmic framework to assess whether an acute presentation is complicated or uncomplicated prior to surgical intervention. By structuring model development across distinct clinical centers rather than relying on a single-institution population, the investigators created an explicit basis for testing generalizability across separate healthcare settings.

The study protocol relied on 10 distinct machine learning algorithms, which were iteratively screened using cross-validation within the initial 652-patient developmental cohort. Cross-validation results alone governed the selection strategy that ultimately favored the support vector machine algorithm over competing candidates. In assessing the resulting predictive outputs, the investigators prioritized multiple complementary validation metrics—coupling the area under the receiver operating characteristic curve for discrimination with the Brier score for calibration accuracy and decision curve analysis to gauge net clinical utility across varied decision thresholds.

A critical technical finding centered on how the algorithm behaved when transferred from its development environment to external facilities. While the support vector machine maintained competitive discrimination in external cohort 1 and external cohort 2, the occurrence of calibration drift revealed that raw risk probabilities shifted systematically when encountering new patient populations. Implementing an intercept-only recalibration procedure corrected this baseline mismatch, bringing the Brier scores down across both external groups. This adjustment demonstrated that while the internal weighting of predictive variables remained robust, baseline risk alignments must be tuned to each local hospital environment before clinicians can safely interpret individual estimates.

Interpretability assessments via SHapley Additive exPlanations further isolated the key clinical and biological signals driving model predictions. Specifically, elevations in circulating fibrinogen, shifts in the neutrophil-to-lymphocyte ratio, and the reported duration of symptoms before evaluation were identified as primary determinants of complicated disease status. To translate these mathematical relationships into practical workflows, the authors established a dedicated web-based prediction platform designed to calculate individualized probability scores. Nevertheless, the investigators emphasized that prospective validation and dedicated assessments of direct clinical impact must occur alongside mandatory local recalibration before routine deployment in emergency care.

Subject of Research: Machine learning prediction and external validation for complicated appendicitis in adults

Article Title: Development and validation of a machine learning prediction model for complicated appendicitis in adults: a multicenter retrospective cohort study

Article References: Yuan, Y., Feng, H., Sun, X., Feng, M., Zheng, Z., Peng, H., Zhang, Q., Shen, Y., Song, K., & Yu, Q. (2026). Development and validation of a machine learning prediction model for complicated appendicitis in adults: a multicenter retrospective cohort study. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03972-6

Image Credits: AI Generated

DOI: 10.1186/s12911-026-03972-6

Keywords: complicated appendicitis, acute appendicitis, machine learning, support vector machine, preoperative risk stratification, multicenter cohort study, computed tomography, internal validation, external validation, calibration drift, intercept recalibration, Brier score, area under the curve, SHAP analysis, fibrinogen, neutrophil to lymphocyte ratio, pain duration, nonoperative management