Machine Learning Insights into Mental Health Among Venezuelan Migrants in Colombia: Pre- and Post-Migration Stressors in Predicting Major Depressive Episodes

Background and Purpose: Mass migration from Venezuela has been motivated both by the deteriorating political and economic situation, which has accelerated since the United States imposed sanctions on Venezuela in 2015. Declining relations in civil society and material deprivation are important pre-migration factors that can contribute to mental health inequities in migrants. Post-migration experiences of cultural-related stressors may also contribute to poor behavioral health outcomes. This study uses machine learning to evaluate the importance of pre-migration deprivation and post-migration behavioral health and cultural stress-related factors in classifying probable major depressive episodes (MDE) among members of the Venezuelan diaspora living in Colombia (N = 519).

Methods: Random forest (RF), a supervised machine learning method, was used to predict MDE. RF is a tree-based ensemble method that has previously been found useful for identifying the importance of cultural stress-related factors in predicting depression and post-traumatic stress disorder among other crisis migration populations. First, the data were randomly partitioned into training (80%; n = 423) and test (20%; n = 96) data sets. Cases of probable depression had the following distribution in the total (54.9%), train (54.4%), and test (57.3%) datasets. Predictors included demographic, behavioral health, and migration-related cultural stress variables such as perceived discrimination, negative context of reception, and pressure to acculturate. We inspected the area under the receiver operating characteristic curve (AUC), F1 score, accuracy, recall, and specificity to evaluate model performance. We evaluated the importance of each attribute by inspecting variable importance measures (mean decrease accuracy).

Results: For Venezuelan migrants in Colombia, probable MDE was a common occurrence, as indicated by a CESD-10 score greater or equal to 10. Top predictors of depression included both behavioral health and cultural stress-related variables, including perceived negative psychological adjustment, anxiety, post-traumatic stress symptoms, negative context of reception, and pressure to acculturate. Demographic and pre-migration factors, such as unmet needs or perceived safety in Venezuela, were ranked lower. When evaluated on the test set, the model performed adequately on the following metrics: accuracy (0.7812 [95%CI: 0.6853, 0.8592]), recall (0.8364), specificity (0.7073), F1 (0.8142), and AUC (0.8361 [95%CI: 0.7550, 0.9107]).

Conclusions and Implications: Various migration-related stressors were predictive of probable MDE, even when accounting for the relationship of commonly co-occurring behavioral health conditions. One important limitation is that probable MDE was inferred from the results of a commonly used depression screening tool, rather than a comprehensive diagnostic evaluation. In addition, findings may not generalize to other crisis migration populations who originate from or have migrated to different contexts. Nevertheless, crisis migrants face a range of pressures to acculturate, discrimination, and poor perceived adjustment to their new contexts, which underscores the need for culturally-informed practices to be implemented in receiving countries to reduce the strain on systems of care and promote the health and well-being of migrant populations.