Abstract: Clinical decision support systems (CDSS) provide intelligently filtered knowledge and patient-specific information to the clinicians, nursing staff and healthcare professionals. CDSS can significantly improve the quality, safety, efficiency and effectiveness of health care. Over the last decade, American hospitals have adopted electronic health records (EHRs) widely resulting in a massive collection of clinical notes such as admission notes, physician notes, nursing notes and discharge summaries. For the past couple decades, most of the work in CDSS has been focused on developing knowledge-based systems using structured data such as medications and ICD codes. In contrast, the EHR notes incorporate rich information such as adverse drug events, social determinants of health and suicidal behaviors which are substantially under-represented in the structured data. This presents a unique opportunity for natural language processing (NLP), with its ability to process massive data beyond the scope of human capability, to provide new clinical insights and frameworks for critical clinical applications previously missed out by any CDSS systems.
In this thesis, we contribute to the NLP and clinical community by developing a robust multi-task learning framework to automate different novel clinical decision support systems. First, we focussed on identifying causality between medication and its adverse drug reactions (ADRs) using a clinically standardized assessment technique called Naranjo Scale. Our multi-task learning framework takes a question, from Naranjo Scale, along with a patient's note to identify relevant evidence sentences and paragraphs in the note and predicts the final answer for the question. Second, we worked on extracting suicide attempt (SA) and suicide ideation (SI) events from patients' clinical notes. We created the first publicly available suicide attempt and ideation events (ScAN) dataset. Leveraging the framework used for CDSS of Naranjo scale, we built a strong multi-task learning model ScANER (Suicide Attempt and Ideation Attempts Retriever) to extract the relevant suicidal behavior evidences from all the clinical notes of a hospital stay and identify the type of SA or SI concluded by the clinical professionals. Next, we conducted a suicidal behavior study on at-risk veterans (~7 million) to study the associations between traumatic brain injury (TBI) and post-traumatic stress disorder (PTSD) with suicidal behavior. We found that veterans with TBI and/or PTSD are more than twice as likely to have suicidal behavior as compared to the control population (veteran without TBI or PTSD diagnosis). We also realized that risk factors related to social determinants of health (SDOH) for these veterans are mainly documented in different sections of patient EHRs but there is no effective method for extracting these risk factors. Hence, we developed a publicly available tool to extract SDOH entities from EHRs (EASE) using the knowledge of the existing 156 medical vocabularies. To conclude the thesis, we are working on developing adaptive learning techniques which can help in fine-tuning our publicly available ScANER model for different hospitals while only using a fraction of the annotations. Our preliminary results show that by tuning <10% of the parameters, ScANER can be effectively used for extracting SA and SI evidences from EHRs of different private hospitals.