Antimicrobial resistance (AMR) poses a severe threat, particularly in low- and middle-income countries (LMICs), where limited diagnostics and empirical antibiotic prescribing fuel multidrug resistance (MDR). MDR, the ability of microbes to resist multiple antibiotics, including critical classes like cephalosporins and carbapenems, is exacerbated by socioeconomic barriers and inadequate surveillance. By 2050, AMR could cause 1.91 million direct deaths and be linked to 8.22 million deaths globally, with LMICs facing the highest mortality rates.
MDRisk Stratify is an innovative machine learning (ML) platform designed to predict individualized MDR risks and antibiotic resistance profiles to guide tailored antibiotic prescribing in LMICs, reducing MDR spread. It leverages datasets from Vivli’s AMR Register, including the LMIC-focused SOAR dataset and ATLAS global data on pathogens like K. pneumoniae and E. coli. The methodology involves preprocessing data by encoding categorical variables (e.g., gender, infection type), normalizing continuous variables (e.g., age), and using SMOTE to address class imbalance. ML models, including Random Forest, Gradient Boosting, and Logistic Regression, will be trained on 80% of the data to predict MDR and resistance to key antibiotics. Performance will be evaluated using F1-score, AUROC, and G-mean, with SHAP analyses ensuring interpretability. A 20% dataset reserved for LMIC contexts, such as Uganda, will validate generalizability.
The platform prioritizes accessibility through a low-bandwidth Streamlit web application and a USSD interface, enabling use in remote areas with limited internet. MDRisk Stratify aims to reduce inappropriate antibiotic prescribing by 20%, achieve over 85% prediction accuracy, enhance clinician accessibility, and inform WHO’s AMR policies. Building on CAMO-Net’s ongoing work, this initiative addresses diagnostic gaps, promotes equitable healthcare, and supports global AMR management.