A classification model for the telephonic triage of low acuity abdominal complaints in South Africa.
African journal of emergency medicine · 2026-08-01 · Method & validation
Abstract
INTRODUCTION: Emergency medical services (EMS) dispatch emergency resources according to the appropriate patient acuity, ensuring those in need of urgent care receive prompt attention. Abdominal complaints have been highlighted as an incident type that is over-triaged in EMS. The development of a telephonic classification tool could improve emergency medical dispatch efficiencies by reducing the inappropriate prioritisation of responses to these cases. The study aimed to identify variables that classified abdominal complaints as low acuity at dispatch. METHODS: A retrospective cross-sectional study was performed on EMS data. Chi-square Automatic Interaction Detector (CHAID) was the classification model used in this study. Nodes split at p-values of 0.05, with the Bonferroni method used for correcting multiple comparisons. Classification tree validation used a 50/50 split and depth limits were applied to mitigate overfitting for training and testing samples. RESULTS: Of the 14 inputs, four were significant predictors, producing a decision tree with 11 nodes. Mobility status was a strong predictor, with 93.3% of walking patients classified as low acuity. Apyrexia was more strongly associated with low acuity, and patients with no bleeding were more likely to be classified as low acuity than those with controlled bleeding. Specific age groups demonstrated a greater probability of being classified as low acuity relative to others. CHAID test model provided a predictive accuracy of 90.4%, a specificity of 50.00%, sensitivity of 92.10%, positive predictive value 94.55% and negative predictive value 40.18%. CONCLUSION: This study is the first of its type in a low- to middle-income country. It provides a potential classification system for low acuity abdominal complaints that may otherwise demand an emergent response from EMS. Applying these classification variables have the potential to inform and strengthen current emergency medical dispatch policies and processes. The dispatch algorithm can also be used to divert a low acuity case to more appropriate services.