Explainable machine learning integrating virological and immune-inflammatory biomarkers for in-hospital mortality prediction in severe fever with thrombocytopenia syndrome

Severe fever with thrombocytopenia syndrome (SFTS), characterized by an excessive inflammatory response and impaired adaptive immunity, is associated with a high risk of mortality. This study aims to establish and validate a machine learning (ML)-based predictive model integrating clinical and immune-inflammatory biomarkers for in-hospital mortality in patients with SFTS. A total of 238 patients, including 177 survivors and 61 non-survivors, were divided by stratified random sampling into a training cohort ( n = 167) and an internal validation cohort ( n = 71) at a 7:3 ratio. Features were selected from the intersection of Boruta, Least Absolute Shrinkage and Selection Operator (LASSO), and random forest recursive feature elimination (RF-RFE). Model hyperparameters were optimized exclusively within the training cohort using repeated stratified five-fold cross-validation, with the internal validation cohort reserved for final evaluation. Model performance was then assessed using the area under the receiver operating characteristic curve (AUC), calibration plots, and decision curve analysis (DCA). SHAP (SHapley Additive exPlanations) was employed to improve model interpretability. To facilitate individualized risk estimation, the selected model was presented as a nomogram and incorporated into an interactive web-based calculator. The final logistic regression model incorporated four predictors: SFTSV RNA load, CD3+CD4+ T-cell count, IL-6, and PCT. In the internal validation cohort, the area under the ROC curve was 0.841 (95% CI: 0.736–0.930). Based on the maximum Youden index in the training cohort, the optimal cutoff probability was established at 0.2053. When this cutoff was applied to the validation cohort, the model yielded a sensitivity of 0.944 (95% CI: 0.742–0.990) and specificity of 0.698 (95% CI: 0.565–0.805). The model also demonstrated a Brier score of 0.139 (95% CI: 0.094–0.189) and potential net clinical benefit. We present an interpretable predictive model based on viral, immune, and inflammatory biomarkers for predicting in-hospital mortality in patients with SFTS. A multicenter prospective evaluation is highly recommended prior to its clinical use.

Authors

Institutions

Publication Details

Journal
BMC Infectious Diseases
Published
2026-11-14
DOI
https://doi.org/10.1186/s12879-026-14844-9
Primary Topic
Viral Infections and Vectors
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Explainable machine learning integrating virological and immune-inflammatory biomarkers for in-hospital mortality prediction in severe fever with thrombocytopenia syndrome

Ruize Ma, Zhihai Chen, Shuwen Ding, Yanli Xu et al.
BMC Infectious Diseases
Viral Infections and Vectors
article

Explainable machine learning integrating virological and immune-inflammatory biomarkers for in-hospital mortality prediction in severe fever with thrombocytopenia syndrome

Ruize Ma, Zhihai Chen, Shuwen Ding, Yanli Xu, Ling Lin
article en

Abstract

Severe fever with thrombocytopenia syndrome (SFTS), characterized by an excessive inflammatory response and impaired adaptive immunity, is associated with a high risk of mortality. This study aims to establish and validate a machine learning (ML)-based predictive model integrating clinical and immune-inflammatory biomarkers for in-hospital mortality in patients with SFTS. A total of 238 patients, including 177 survivors and 61 non-survivors, were divided by stratified random sampling into a training cohort ( n = 167) and an internal validation cohort ( n = 71) at a 7:3 ratio. Features were selected from the intersection of Boruta, Least Absolute Shrinkage and Selection Operator (LASSO), and random forest recursive feature elimination (RF-RFE). Model hyperparameters were optimized exclusively within the training cohort using repeated stratified five-fold cross-validation, with the internal validation cohort reserved for final evaluation. Model performance was then assessed using the area under the receiver operating characteristic curve (AUC), calibration plots, and decision curve analysis (DCA). SHAP (SHapley Additive exPlanations) was employed to improve model interpretability. To facilitate individualized risk estimation, the selected model was presented as a nomogram and incorporated into an interactive web-based calculator. The final logistic regression model incorporated four predictors: SFTSV RNA load, CD3+CD4+ T-cell count, IL-6, and PCT. In the internal validation cohort, the area under the ROC curve was 0.841 (95% CI: 0.736–0.930). Based on the maximum Youden index in the training cohort, the optimal cutoff probability was established at 0.2053. When this cutoff was applied to the validation cohort, the model yielded a sensitivity of 0.944 (95% CI: 0.742–0.990) and specificity of 0.698 (95% CI: 0.565–0.805). The model also demonstrated a Brier score of 0.139 (95% CI: 0.094–0.189) and potential net clinical benefit. We present an interpretable predictive model based on viral, immune, and inflammatory biomarkers for predicting in-hospital mortality in patients with SFTS. A multicenter prospective evaluation is highly recommended prior to its clinical use.

BMC Infectious Diseases
Capital Medical University (CN), Yantai Infectious Diseases Hospital (CN), Yantaishan Hospital (CN), Beijing Ditan Hospital (CN)
Viral Infections and Vectors
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon