Comprehensive Summary
This study, presented by Pinto et al., examines how machine learning (ML) can be ethically integrated into population and public health. By compiling a panel of academics across a variety of fields including computer science, epidemiology, and public health, while also conducting a literature review of existing guidelines, the authors use a modified Delphi approach to develop five main recommendations for implementing ML into population and public health. Specifically, the five recommendations were to (1) prioritize partnerships and interventions that are disadvantaged by social and economic policies, (2) use ML in public health emergencies and other dynamic situations by collecting and analyzing population-wide deidentified data, (3) ensure fairness across populations and mitigate bias, (4) ensure public availability of data sources, model methodologies, and technical details, along with bias mitigation strategies to promote trust and transparency, and (5) facilitate regular, multidisciplinary discussions to identify biases and ensure fair implementation. With these recommendations, the authors emphasize the importance of equitable design and delivery so that ML would not reinforce existing health disparities and instead contribute to improving health outcomes.
Outcomes and Implications
With the increase in data availability and the rapid advancements in ML-based innovations, there is high potential for ML to improve health outcomes through areas including the surveillance of infectious diseases and predicting the burden of noncommunicable diseases. However, if the ML models are trained on incomplete or biased data, this poses a significant risk of producing inaccurate risk assessments and misguided interventions, ultimately worsening existing health disparities. Although the framework provides a strong foundation for healthcare stakeholders, Pinto et al. note that future research should pilot these proposed guidelines in public health settings such as LMICs to assess their effectiveness, as well as ensuring ongoing assessment and adaptation of these guidelines before full-scale implementation.