Interpretable machine learning identifies patients at risk for persistent influenza A nucleic-acid positivity

A considerable proportion of patients with influenza A remain nucleic-acid positive after a standard course of antiviral therapy, but tools for identifying them at admission are lacking. We developed an interpretable machine-learning model for early risk prediction. This retrospective study included 432 hospitalised adults with laboratory-confirmed influenza A, randomly allocated by stratified random sampling on the outcome to a training cohort ( n = 346) and a validation cohort ( n = 86). Persistent positivity was defined as a positive respiratory nucleic-acid test within 72 h after completion of a standard course of antiviral therapy. Predictors were selected by elastic-net penalised logistic regression. Seven algorithms were evaluated using nested cross-validation with stratified ten-fold cross-validation repeated five times and stratified ten-fold randomised search for tuning. Discrimination, calibration, decision-curve analysis and SHAP-based interpretability were assessed. Persistent positivity occurred in 29.5% and 29.1% of the training and validation cohorts, respectively. XGBoost achieved the highest discrimination, with nested cross-validated ROC-AUC 0.793 (95% CI 0.748–0.835) and PR-AUC 0.600 (95% CI 0.530–0.676) in the training cohort, ROC-AUC 0.835 (95% CI 0.734–0.914) and PR-AUC 0.714 (95% CI 0.573–0.851) in the validation cohort. At a pre-specified decision threshold, validation sensitivity was 0.800, specificity 0.656 and negative predictive value 0.889. SHAP analysis identified C-reactive protein as the dominant predictor, followed by lymphocyte count, age, vomit, aspartate aminotransferase, urea, and kidney disease. The interpretable XGBoost model showed promising predictive performance and may support early risk stratification and targeted follow-up. External validation and prospective evaluation are needed before clinical implementation.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-14
DOI
https://doi.org/10.1038/s41598-026-80585-z
Primary Topic
Influenza Virus Research Studies
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Interpretable machine learning identifies patients at risk for persistent influenza A nucleic-acid positivity

Hanzhen Zhao, 景荣先, Jinhui Xu, Zhiwei Zhuang et al.
Scientific Reports
Influenza Virus Research Studies
article

Interpretable machine learning identifies patients at risk for persistent influenza A nucleic-acid positivity

Hanzhen Zhao, 景荣先, Jinhui Xu, Zhiwei Zhuang, Yanxia Yu, Lian Tang, Xiaoying Li, Qi Ding, Jingjing Li, Xiaowen Xu
article en

Abstract

A considerable proportion of patients with influenza A remain nucleic-acid positive after a standard course of antiviral therapy, but tools for identifying them at admission are lacking. We developed an interpretable machine-learning model for early risk prediction. This retrospective study included 432 hospitalised adults with laboratory-confirmed influenza A, randomly allocated by stratified random sampling on the outcome to a training cohort ( n = 346) and a validation cohort ( n = 86). Persistent positivity was defined as a positive respiratory nucleic-acid test within 72 h after completion of a standard course of antiviral therapy. Predictors were selected by elastic-net penalised logistic regression. Seven algorithms were evaluated using nested cross-validation with stratified ten-fold cross-validation repeated five times and stratified ten-fold randomised search for tuning. Discrimination, calibration, decision-curve analysis and SHAP-based interpretability were assessed. Persistent positivity occurred in 29.5% and 29.1% of the training and validation cohorts, respectively. XGBoost achieved the highest discrimination, with nested cross-validated ROC-AUC 0.793 (95% CI 0.748–0.835) and PR-AUC 0.600 (95% CI 0.530–0.676) in the training cohort, ROC-AUC 0.835 (95% CI 0.734–0.914) and PR-AUC 0.714 (95% CI 0.573–0.851) in the validation cohort. At a pre-specified decision threshold, validation sensitivity was 0.800, specificity 0.656 and negative predictive value 0.889. SHAP analysis identified C-reactive protein as the dominant predictor, followed by lymphocyte count, age, vomit, aspartate aminotransferase, urea, and kidney disease. The interpretable XGBoost model showed promising predictive performance and may support early risk stratification and targeted follow-up. External validation and prospective evaluation are needed before clinical implementation.

Scientific Reports
Suzhou Municipal Hospital (CN), Nanjing Medical University (CN)
Nanjing Medical University
Good health and well-being
Influenza Virus Research Studies
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