Gweru Provincial Hospital Tests Five Machine-Learning Models to Flag Drug-Resistant Infections

Zimbabwe's national AMR surveillance has been strengthened in recent years with support from the UK's Fleming Fund, but hospital-level predictive tools remain rare, and many diagnostic laboratories still lack electronic data systems. Researchers working with Gweru Provincial Hospital, a general referral hospital in Midlands Province, set out to test whether routinely collected laboratory data could power a predictive early-warning model rather than only retrospective reporting.
The retrospective cross-sectional study drew on 4,054 clinical bacterial isolates from 874 patient records collected between 2022 and 2024. The team trained and compared five machine-learning models — support vector machine (SVM), random forest, logistic regression, gradient boosting and k-nearest neighbours — to predict resistance. SVM performed best, reaching 72.08% accuracy, 73.25% precision, 79.78% recall, an F1 score of 0.76 and an AUC-ROC of 0.79. The authors highlight the approach's potential to direct antimicrobial-stewardship resources toward the highest-risk hospital wards and pathogens.
The study is a single-hospital retrospective analysis, posted as a preprint in September 2025 and not yet through independent peer review; no prospective validation or clinical deployment has been reported. Evidoria includes it as a credible, well-documented proof-of-concept from a public provincial hospital — real diagnostic performance figures from real records, with the normal caveats of an unreviewed, single-site retrospective study.

Read the full analysis: https://www.researchsquare.com/article/rs-7620529/v1