External Validation of a Prognostic Model for Predicting 24-Hour Mortality in Patients with Severe Septic Shock Presenting to the Emergency Department of Phra Nakhon Si Ayutthaya Hospital

Abstract

Objective: To validate the accuracy of a 10-variable prediction model for 24-hour mortality in patients diagnosed with septic shock presenting to the ED of Phra Nakhon Si Ayutthaya Hospital. Methods: This was a prognostic prediction study in the external validation phase, utilizing a retrospective cohort design. Medical records from 2023–2024 were reviewed for patients aged over 18 years who were diagnosed with septic shock (ICD-10: R57.2) upon presentation to the ED. Inclusion required clinical documentation indicating shock due to suspected or confirmed infection in any organ system. Patient outcomes were compared between those who died within 24 hours and those who survived beyond 24 hours. Data were analyzed in two cohorts: derivation data (2018–2022, n=602) and validation data (2023–2024, n=216). Group comparisons were conducted using exact probability tests and Wilcoxon rank-sum tests for categorical and continuous variables, respectively. Model discrimination was evaluated using the c-statistic in both datasets. Mortality risk was stratified based on total scores into three categories: low (0–27.5 points), moderate (28.0–31.0 points), and high (31.5–37.5 points) according to the original model. The 24-hour mortality rate in each risk group was assessed in both datasets. Results: C-statistics were 0.79 for the derivation cohort and 0.69 for the validation cohort. Four of the ten model variables showed statistically significant differences between the two datasets: diastolic blood pressure, pH, phosphorus, and albumin. The observed 24-hour mortality rates by risk category were as follows: https://shorturl.asia/YfM4C Conclusion: The model demonstrated reduced predictive performance in the 2023–2024 validation dataset (c-statistic < 0.7). These findings suggest the need for periodic recalibration or revision of the model coefficients and/or risk threshold scores to maintain accuracy in predicting 24-hour mortality among septic shock patients. Finally, if the model can no longer maintain adequate predictive performance, it should be recalibrated or redeveloped by re-evaluating and incorporating the relevant input variables.