Answers are grounded in this paper and cite their sources.
Host-based diagnostics coupled with machine learning may support faster discrimination of infectious states and reduce unnecessary antimicrobial treatment. We reviewed recent studies that applied machine-learning methods to clinical patient data for pathogen detection, infection classification, or closely related host-response diagnosis. Eligible infection targets included bacterial and viral infections, bacterial-vs.-viral classification, and syndrome-based presentations such as sepsis. Malaria was included when evaluated within a mixed febrile cohort. Four literature sources were searched for articles published from January 2020 through August 2025. Of 144 retrieved records, 51 passed an initial host-diagnostics screen and 21 articles met the final eligibility criteria. The included articles covered structured clinical variables, gene or molecular-expression measurements, medical imaging, physiological signals, and microscopy. For the quantitative model-class comparison, we extracted 54 evaluations belonging to four focal model classes: random forest (23), neural network (17), support vector machine (9), and logistic regression (5). Less frequently evaluated approaches, including gradient boosting, decision trees, naive Bayes, and k-nearest neighbors, were summarized descriptively. Differences in input modalities, cohorts, preprocessing, validation, hyperparameter selection, and endpoints prevented direct performance comparisons. Among the 54 focal-class evaluations, 14 reported accuracy alone, 20 reported area under the receiver operating characteristic curve (AUC) alone, and 20 reported both. Only 16 evaluations reported additional measures such as sensitivity, specificity, F1 score, or Matthews correlation coefficient. Imaging studies generally favored neural-network approaches, whereas molecular-expression studies frequently used random forests. However, the evidence did not support a universally optimal model.
No takes yet. Share an insight, caveat, or question.
Samuel Harrington (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: