Heterogeneity in the depression risk associated with spousal cardiovascular events

Although previous studies have reported interpersonal associations between cardiovascular disease (CVD) and depression, evidence is lacking as to whether this association varies by individual characteristics. Using a nationwide database, we applied a sequential matching framework to construct a cohort of 59,900 index individuals (primary insured) whose spouses (dependent) received first CVD diagnosis and those without spouses’ CVD event between 2016 and 2018. We examined heterogeneity in the association between spousal CVD events and subsequent depression in index individuals within 3 years via a novel machine learning approach, Bayesian causal forest (BCF), adjusting for sociodemographic characteristics, comorbidities, health behaviors, and physical health conditions. We trained the BCF model using randomly selected 50 % sample and evaluated the heterogeneity by estimating conditional average treatment effect (CATE) using the remaining 50 % sample. Our BCF algorithm identified heterogeneity in the association between spouse's CVD and depression (the lowest quartile of CATE [most resilient group], risk difference [95 %CI] = −0.54 percentage point [-1.24, 0.16] and odds ratio [95 %CI] = 0.78 [0.57, 1.07]; and the highest quartile of CATE [most vulnerable group], risk difference [95 %CI] = +0.73 percentage point [0.02, 1.44] and odds ratio [95 %CI] = 1.38 [1.01, 1.88]). Individuals with higher CATEs were more likely to have a higher proportion of females, have fewer disease histories, and have more frequent unhealthy behaviors. Most individuals were male (95 %), potentially limiting the generalizability. The depression risk associated with spouse's CVD varied across individuals. Future studies should elaborate on building targeted interventions to promote better health security of family members as well as patients with CVD. • Depression risks after spousal CVD events were heterogeneous across individuals. • Vulnerable groups had more females, fewer diseases, and more unhealthy behaviors. • Machine learning can uncover complex interactions between lifestyle and diseases. • Future studies should build targeted interventions for both patients and families.