Predicting gram negative bacteria resistance to third generation cephalosporins among hospitalized patients at a tertiary facility in the southern highlands, tanzania 2021-2025: machine learning approach
1
School of Public Health, Kilimanjaro Christian Medical Centre University, Kilimanjaro, Tanzania, United Republic of
Popul. Med. 2026;8(Supplement Supplement 1):
ABSTRACT
INTRODUCTION:
Antimicrobial resistance (AMR) threatens health coverage in low- and middle-income countries like Tanzania, where third-generation cephalosporins (3GC) resistance in gram-negative bacteria exceeds 50%, driven by empirical prescribing and limited diagnostics, increasing mortality and costs1-3. Machine learning (ML) predicts antibiotic resistance, enhancing AI-driven clinical decision-making for targeted therapy and antimicrobial stewardship4-9. This study developed ML models using Tanzania’s AMR data to forecast 3GC resistance, addressing sub–Saharan Africa’s diagnostic gaps and promoting AI-driven self-reliance in health security. Objectives included model development, performance evaluation, and predictor identification to guide empirical prescribing and mitigate AMR.
METHODS:
Cross-sectional analysis of deidentified AMR surveillance data from hospitalized patients’ clinical isolates of WHO priority pathogens at Mbeya Zonal Referral Hospital (2021-2025). Inclusion: First isolate per hospitalized patient per pathogen per sample type within 30 days; exclusion: non-Tanzanian residents or follow-up cases. Models (Gradient boosting Decision tree, logistic regression, random forest, Multilayer perceptron, and k-nearest neighbours) trained via 10-fold validation with SMOTE balancing, optimized by grid search in R 4.4.1. Performance assessed on 30% test set using AUC-ROC, accuracy, sensitivity/specificity, PPV/NPV, and SHAP values for feature importance.
RESULTS:
Preliminary analysis indicates that of 4523 isolates, 2487(55%) showed 3GC resistance. Random forest achieved the highest AUC-ROC 0.88 (95%CI: 0.85-0.91), accuracy 85%, sensitivity 82%, specificity 87%, PPV 84%, NPV 86%; followed by a multilayer perceptron (AUC-ROC 0.84) outperformed logistic regression (AUC-ROC 0.76). Key predictors: hospital-associated infections (SHAP 0.32) and age >65 years (SHAP 0.15). Models reduced inappropriate 3GC use by 25% in simulation versus empirical guidelines.
CONCLUSIONS:
Integrating the Random Forest model into stewardship tools will optimize empirical prescribing, reduce broad-spectrum antibiotic use by 25%, and inform national guidelines, advancing AI-driven precision health for AMR control in Tanzania. Not only will this foster equity in precision health, but also support Africa’s UHC goals through scalable digital tools.
Antimicrobial resistance (AMR) threatens health coverage in low- and middle-income countries like Tanzania, where third-generation cephalosporins (3GC) resistance in gram-negative bacteria exceeds 50%, driven by empirical prescribing and limited diagnostics, increasing mortality and costs1-3. Machine learning (ML) predicts antibiotic resistance, enhancing AI-driven clinical decision-making for targeted therapy and antimicrobial stewardship4-9. This study developed ML models using Tanzania’s AMR data to forecast 3GC resistance, addressing sub–Saharan Africa’s diagnostic gaps and promoting AI-driven self-reliance in health security. Objectives included model development, performance evaluation, and predictor identification to guide empirical prescribing and mitigate AMR.
METHODS:
Cross-sectional analysis of deidentified AMR surveillance data from hospitalized patients’ clinical isolates of WHO priority pathogens at Mbeya Zonal Referral Hospital (2021-2025). Inclusion: First isolate per hospitalized patient per pathogen per sample type within 30 days; exclusion: non-Tanzanian residents or follow-up cases. Models (Gradient boosting Decision tree, logistic regression, random forest, Multilayer perceptron, and k-nearest neighbours) trained via 10-fold validation with SMOTE balancing, optimized by grid search in R 4.4.1. Performance assessed on 30% test set using AUC-ROC, accuracy, sensitivity/specificity, PPV/NPV, and SHAP values for feature importance.
RESULTS:
Preliminary analysis indicates that of 4523 isolates, 2487(55%) showed 3GC resistance. Random forest achieved the highest AUC-ROC 0.88 (95%CI: 0.85-0.91), accuracy 85%, sensitivity 82%, specificity 87%, PPV 84%, NPV 86%; followed by a multilayer perceptron (AUC-ROC 0.84) outperformed logistic regression (AUC-ROC 0.76). Key predictors: hospital-associated infections (SHAP 0.32) and age >65 years (SHAP 0.15). Models reduced inappropriate 3GC use by 25% in simulation versus empirical guidelines.
CONCLUSIONS:
Integrating the Random Forest model into stewardship tools will optimize empirical prescribing, reduce broad-spectrum antibiotic use by 25%, and inform national guidelines, advancing AI-driven precision health for AMR control in Tanzania. Not only will this foster equity in precision health, but also support Africa’s UHC goals through scalable digital tools.
| eISSN: | 2654-1459 |
We process personal data collected when visiting the website. The function of obtaining information about users and their behavior is carried out by voluntarily entered information in forms and saving cookies in end devices. Data, including cookies, are used to provide services, improve the user experience and to analyze the traffic in accordance with the Privacy policy. Data are also collected and processed by Google Analytics tool (more).
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
