# Machine learning models for early mortality prediction in severe fever with thrombocytopenia syndrome

## TL;DR

This study developed an AI model to predict early mortality in a severe viral disease, using data from over 1,600 patients and deploying a web tool for real-time risk assessment.

## Contribution

A novel interpretable machine learning model for early mortality prediction in SFTS, validated across multiple hospitals and deployed as an open-access web tool.

## Key findings

- The XGBoost model achieved high accuracy (AUC of 0.916 in training and 0.905 in validation).
- SHAP analysis identified six key predictors of mortality in SFTS patients.
- A real-time web-based tool was developed to provide individualized risk predictions with visual explanations.

## Abstract

Severe fever with thrombocytopenia syndrome (SFTS) is a high-fatality viral disease where early mortality risk prediction is vital for clinical management. This retrospective multicenter cohort study enrolled 1,690 hospitalized SFTS patients from five Chinese hospitals (2014–2023) to develop, validate, and deploy an interpretable machine learning (ML) model for early mortality risk assessment. Using LASSO regression for feature selection then comparing eight ML algorithms, the XGBoost model achieved an AUC of 0.916 in the training cohort and 0.905 in the temporal validation cohort. SHapley Additive exPlanations (SHAP) analysis identified six key predictors, which were used to deploy a real-time, open-access web-based tool (https://sftsprognosis.com) that provides individualized risk predictions with visual explanations. The XGBoost model has the potential to enhance timely clinical decision-making, facilitate efficient allocation of critical care resources, and provide a generalizable framework for applying machine learning in the management of emerging infectious diseases.

•This retrospective multicentre cohort study collected 10 years from five hospitals in China•Eight machine learning models were trained, validated, and interpreted using SHAP analysis•An XGBoost model was developed to predict early mortality risk in patients with SFTS•A publicly accessible, real-time web tool was deployed to facilitate timely decision-making

This retrospective multicentre cohort study collected 10 years from five hospitals in China

Eight machine learning models were trained, validated, and interpreted using SHAP analysis

An XGBoost model was developed to predict early mortality risk in patients with SFTS

A publicly accessible, real-time web tool was deployed to facilitate timely decision-making

Public health; Artificial intelligence applications