Development and validation of risk prediction models for high-risk patients with non-traumatic acute abdominal pain: a prospective observational study.

Purpose: This study develops and validates a machine learning-based model to help triage nurses identify high-risk patients with non-traumatic acute abdominal pain, enhancing accuracy and safety. Patients and methods: Utilizing information technology, a data collection form was embedded into the electronic pre-triage systems of the emergency departments in two tertiary general hospitals (Shanghai Tongren Hospital and the First Affiliated People's Hospital of Soochow University). Data from 3,090 patients were prospectively collected and preprocessed. Predictive factors for non-traumatic acute abdominal pain were screened through univariate analysis, LASSO regression, and multivariate analysis. Risk early warning models were constructed using seven methods based on R software and externally validated at different time points. Results: The incidence of high-risk patients with non-traumatic acute abdominal pain was 14.49%. Ten predictive factors were identified: (1) age, (2) mode of admission, (3) history of heart disease, (4) history of tumor, (5) MEWS score ≥5, (6) trigger being post-coital, (7) knife-like pain, (8) accompanied by abdominal distension and fullness, (9) tenderness, and (10) muscle tension. All seven predictive models demonstrated good predictive performance, among which the random forest model (AUC = 0.786) showed the best overall predictive performance. External validation results indicated that the logistic regression model had good extrapolation and generalization ability. In this study, the logistic regression model was visualized using a nomogram. Conclusion: Machine learning models were developed for early risk prediction in non-traumatic acute abdominal pain; random forest showed the best discrimination, while logistic regression with a nomogram offered superior clinical applicability.