Cardiometabolic Risk Estimation Using Exposome Data and Machine Learning

• DOI: 10.2139/ssrn.4367352
• Publication Date: 2023-03-08T19:30:15Z
• Authors: Angélica María At...; Polyxenia Gkontra; Marina Camacho; Oliver Diaz; Maria Bulgheroni; Valentina Simonetti; Marc Chadeau-Hyam; Janine F. Felix; Sylvain P. Sébert; Karim Lekadir
• Keywords: Exposome

The human exposome is nowadays recognized as a significant contributor to the overall risk of developing major pathologies, such cardiovascular disease (CVD) and diabetes. Therefore, personalized early assessment based on attributes might be promising tool for identifying high-risk individuals improving prevention. In this work, we present novel, fair machine learning (ML) model CVD type 2 diabetes (T2D) prediction set readily available factors. We evaluated our using internal external validation groups from multi-center cohort. From UK Biobank, identified 5,348 1,534 participants who within 13 years baseline visit were diagnosed with T2D, respectively. An equal number did not develop these pathologies randomly selected control group. 109 exposure variables six different categories (physical measures, environmental, lifestyle, mental health events, sociodemographics, early-life factors) participant's considered. adopted XGBoost ensemble predict at diseases. model's performance was compared that an integrative ML which biological, clinical, physical sociodemographic variables, and, additionally CVD, Framingham score. Moreover, assessed proposed potential bias related sex, ethnicity age. Lastly, interpreted results SHAP, state-of-the-art explainability method.The presents comparable despite solely information, achieving ROC-AUC $0.78\pm0.01$ $0.77\pm0.01$ Additionally, prediction, exposome-based improved over traditional No in terms key sensitive identified. factors play important role patients naps during day, age completed full-time education, past tobacco smoking, frequency tiredness/unenthusiasm, current work status. Overall, demonstrates T2D tool.