Rethinking preventive healthcare through artificial intelligence

Abstract

Preventive healthcare is a key strategy in reducing the global disease burden. Among millions of deaths worldwide, two-third of them were attributed to preventable chronic diseases like cardiovascular disease (CVD), diabetes, and cancer. Recent advancements of artificial intelligence (AI) in healthcare offer promise to significantly improve the design of preventive strategies. Even a modest reduction in disease risk has the potential to save millions of lives annually around the world. Preventive healthcare is a comprehensive, long-term process that can be broadly segmented into four key phases: risk factor discovery, prognostic screening, treatment prioritization, and dynamic treatment planning. Each phase presents unique challenges for AI solutions, necessitating a balance between cost efficiency and precision, handling heterogeneous patient profiles, and addressing the complexity of treatment selection and adjustment. This thesis addresses these challenges with innovative AI solutions. Duet, inspired by decision trees, effectively identifies genomic structural variants, enhancing the cost-effective detection of crucial biomarkers. P-CARDIAC, a scalable risk prediction model, combines machine learning and statistical methods to provide accurate risk estimations based on varying levels of input factors. VISTA, a pipeline merging survival analysis with feature selection, forecasts personalized treatment effects, aiding clinicians in prescribing the most effective drugs. Moreover, Duramax introduces a data-driven framework leveraging reinforcement learning to optimize dynamic preventive strategies. By learning from vast treatment decisions, Duramax outperforms individual clinicians in disease prevention, demonstrating a reduction in CVD risk through aligned treatment decisions. Looking ahead, the thesis suggests future directions for enhancing preventive healthcare and medical AI. A proposed digital twin system could monitor multiple comorbidities simultaneously, offering a holistic view for chronic disease prevention. By integrating structured and unstructured data sources, such as electronic health records and imaging, the field could benefit from increased predictive accuracy and interpretability. Additionally, fostering human-AI interaction in the preventive loop could enhance clinician acceptance and refinement of AI recommendations, potentially revolutionizing clinical decision-making processes. In conclusion, the integration of AI into preventive healthcare holds immense promise for improving disease prevention strategies and personalized treatment approaches. By addressing current challenges and exploring future opportunities, the field stands poised for significant advancements in healthcare outcomes and patient wellbeing. The thesis showcases how AI can be integrated seamlessly into disease prevention strategy design and provides theoretical foundations for the forementioned future directions.