Big Data/Machine Learning/AI
Estimating Heterogeneous Effects of Spousal Cardiovascular Event on Depression Toshiaki Komura* Toshiaki Komura Kosuke Inoue
Background:
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. We thus aim to examine the heterogeneity in the relationship between a spouse’s CVD and subsequent mental health of individuals.
Method:
Using a nationwide claims and medical database which covers approximately 40% of the working-age population in Japan, we applied target trial emulation framework to synthesize a cohort of index individuals whose spouses received first CVD diagnosis and those without the spousal CVD event between 2016 and 2019. Using 85,424 matched index individuals, we examined effect heterogeneities of spousal CVD events on depression risks of index individuals within 2 years via a novel machine learning approach, accelerated Bayesian causal forest (XBCF), adjusting for major sociodemographic characteristics, comorbidities of index individuals and spouses, health behaviors, and objectively measured physical health conditions. We built XBCF model using randomly selected 50% training sample, and evaluated the heterogeneity using the remaining 50% test sample.
Result:
During the 2 years of follow-up, a new onset of depression was observed in 1,296 index individuals (1.52%). When we applied our XBCF algorithm to the test sample, we observed a consistent increase in risk differences and odds ratios for depression due to spouse’s CVD according to the ranking of estimated conditional average treatment effect (CATE; Q1 [most resilient], RD [95% CI] = -0.11pp [-0.35, 0.57]; Q2, RD [95% CI] = +0.30pp [-0.17, 0.76]; Q3, RD [95% CI] = +0.42pp [-0.07, 0.91]; and Q4 [most vulnerable], RD [95% CI] = +0.69pp [0.22, 1.17]). Individuals with high CATE were more likely to be younger, female, have fewer disease histories, and have more frequent unhealthy behaviors than those with lower CATE.
Conclusion:
The expected increases in depression risk due to spouse’s CVD were heterogeneous across individuals. Vulnerable group was characterized by younger age, female, and unhealthy despite fewer prevalence of comorbidities. Future studies should elaborate on mechanisms of the spillover effect within households and targeted interventions for family members as well as a CVD patient.