Extraction and summarization of evidence-based medicine

Project Summary/Abstract Evidence-based medicine (EBM) requires collecting and ranking relevant evidence according to its epistemological strength, aiming to identify the most suitable evidence to inform guidelines and policies. This process generally prioritizes strong evidence from randomized clinical trials (RCTs), cohort studies, systematic reviews, and meta-analyses. However, the volume of scientific literature being published in peer-reviewed journals is increasing exponentially. This makes it extraordinarily financial and societal costly for scientists to obtain a reasonably precise, cumulative overview of the conceptual structure of the literature about a certain scientific topic, and to identify potential new research opportunities. It also presents a challenge for practitioners and policymakers in assimilating and understanding the prevailing findings, further slowing down the translation from biomedical discoveries to improved health outcomes. It is imperative to develop scalable methods for efficient clinical evidence extraction and summarization. Responding to PA-23-034, this project will develop and validate a novel informatics framework to extract computable Population, Intervention, Comparison, and Outcome (PICO) elements and summarize them from publications of interventionaland observational studies. While the framework can be applied to any type of condition, the focus will be on 23 clinical domains, such as neurological, respiratory, digestive system, cardiovascular, and mental health conditions, because they are extremely common and affect people of all ages, races, and socioeconomic statuses. Our extensive preliminary data show that we can effectively employ deep learning (DL) and natural language processing (NLP) methods to extract PICO elements from the clinical trial registries and RCT publications in PubMed. We can also perform zero-shot clinical evidence summarization across six clinical domains. These studies lend further support to the feasibility of the proposed research. Based on our preliminary data and our experience with an interdisciplinary team, our specific aims are: (1) Systematically design and data-driven evaluate a computable PICO ontology; (2) Extract multi-granular PICO elements from study publications; (3) Generate rational-based evidence summaries; and (4) Develop and validate the EBM toolkit via a user-centered design. The research proposed in this project is novel and innovative because it will produce and rigorously test new solutions to improve computable EBM. Successful completion of these aims will empower the general public, stakeholders in EBM, and public health policymakers to efficiently discover and summarize pertinent medical evidence, thereby facilitating well-informed healthcare decision-making.