Finding Causal Impacts of Social Determinants of Mental Health on Opioid Use Disorder from Clinical Notes

Opioid Use Disorder (OUD) affects physical health, mental well-being, and socio-economic stability, underscoring the need to understand the causal effects of social determinants on OUD outcomes. Despite extensive correlation studies between social determinants of mental health (SDMHs) and OUD, specific causal determinants remain unidentified due to the lack of robust causal models. This paper proposes a two-step causal effects iden-tification framework to detect and identify the effects of SDMHs on OUD progression. Firstly, we developed a Multitask Multilabel Clinical-Longformer(MMCL) model to detect social determi-nants of Mental health from clinical notes, effectively processing and identifying relevant SDMHs within unstructured text data. Secondly, we employed a novel Siamese Neural Network(SNN)-based subgroup discovery technique to ascertain the causal effects of these social determinants on OUD. This technique leverages the Siamese architecture's capability to handle complex relationships and identify homogeneous subgroups within the data, enhancing the precision of causal inference. To support this research, we collaborated with experts to create a new dataset, SDMH-OUD-Clinic, comprising social determinants of Mental health-annotated clinical notes and OUD annotations, sub-sampled from the MIMIC-IV dataset. We evaluated the proposed models using the Infant Health and Development Program (IHDP) dataset and applied them to our newly created SDMH-OUD-Clinic dataset. The results demonstrate the model's effectiveness and provide detailed explanations of the identified causal relationships.