Childhood living conditions as predictors of self-rated health status in middle-aged and older adults: Evidence from a machine learning analysis in 27 high-income countries.

Purpose: Health inequalities in high-income countries may have roots in childhood. We aimed to develop machine learning (ML) algorithms to assess the impact of childhood living conditions on self-rated health before and during the COVID-19 pandemic. Methods: We analyzed data from 45,570 individuals aged 50+ across 26 European countries and Israel. Information on 39 childhood living conditions and self-rated health was collected. Seven ML algorithms were used to predict health outcomes. Results: All algorithms performed well, with AUC scores ranging from 0.699 to 0.774 and accuracy between 0.621 and 0.724, with CatBoost as the best performer. Self-rated health before age 15 was the most critical predictor. Other key predictors varied by domain such as religion importance, exposure to World War II, physical harm from others and the relationship with the father. We revealed heterogeneity across genders, regions of Europe and levels of COVID-19 lockdown stringency. For instance, mother education was a more prominent predictor of late-life health for females than for males. Conclusion: This study is one of the first to predict health based on childhood conditions and ML. Our results demonstrate the usefulness of ML in identifying early-life conditions that influence health and provide insights for addressing health inequalities.