B-136 Implementation of DUKE ML-DQA Framework in LMIC Healthcare Institution

Background Data quality faces critical challenges across all regions in its impact on machine learning (ML) modelling, yet these challenges become particularly severe and distinct in low-middle income countries (LMICs) where resources are scarce and disease burdens are high. This impacts the efficacy of ML models in these settings, potentially compromising patient care and research. Methods We conducted a comprehensive evaluation of data quality using the ML-DQA framework, previously validated in high-income countries with robust electronic health record (EHR) systems, in a setting lacking such infrastructure. The ML-DQA framework, which comprises four phases—pre-DQA, data processing, ML-DQA checks, and ML-DQA adjudication—was implemented to assess the quality of the data elements for both ML models. Results The ML-DQA framework was effective in identifying adequate data elements that passed checks based on plausibility, conformance, and completeness. It was helpful in uncovering issues within the data and determining the applicability of ML models with the available information. The study highlights the need to adapt the ML-DQA framework to accommodate specific data quality challenges in the studied LMIC tertiary care hospital (Aga Khan University Hospital) environment. Conclusion The study suggests that data quality assurance processes require adaption to LMIC settings to support the effective implementation of ML algorithms in healthcare. Enhanced data management practices, aligned with the realities of low-resource settings, are essential to ensure ML benefits experienced in high-resource settings are shared by LMICs.