External validation of 5A score model for predicting in-hospital mortality among the accidental hypothermia patients: JAAM-Hypothermia study 2018-2019 secondary analysis.

Background: The 5A score including five components "Age, Activities of daily living, Arrest, Acidemia and Albumin" was developed as an easy-to-use screening tool for predicting in-hospital mortality among patients with accidental hypothermia. However, the external validity of the 5A score has not yet been evaluated. We aimed to perform an external validation of the 5A score model.

Method: This secondary analysis of the multicenter, prospective cohort Japanese Association for Acute Medicine-Hypothermia Study (2018-2019), which was conducted at 87 and 89 institutions throughout Japan, collected data from December 2018 to February 2019 and from December 2019 to February 2020. Adult accidental hypothermia patients whose body temperature was 35 °C or less were included in this analysis. The probability of in-hospital mortality was calculated using a logistic regression model of the 5A score. The albumin was not recorded in this database; thus, it was imputed by estimation. Predictions were compared with actual observations to evaluate the calibration of the model. Furthermore, decision-curve analysis was used to evaluate the clinical usefulness.

Results: Of the 1363 patients registered in the database, data of 1139 accidental hypothermia patients were included for analysis. The median [interquartile range] age was 79 [68-87] years, and there were 625 men (54.9%) in the study cohort. The predicted probability and actual observation by risk groups produced the following results: low 7% (5.4-8.6), mild 19.1% (17.4-20.8), moderate 33.2% (29.9-36.5), and high 61.9% (55.9-67.9) predicted risks, and the low 12.4% (60/483), mild 17.7% (59/334), moderate 32.6% (63/193), and high 69% (89/129) observed mortality. These results indicated that the model was well calibrated. Decision-curve analysis visually indicated the clinical utility of the 5A score model.

Conclusion: This study indicated that the 5A score model using estimated albumin value has external validity in a completely different dataset from that used for the 5A model development. The 5A score is potentially helpful to predict the mortality risk and may be one of the valuable information for discussing the treatment strategy with patients and their family members.