ClinEX - Clinical Evidence Extraction, Representation, and Appraisal

SUMMARY Evidence-based medicine faces increasingly mounting challenges. With the explosively growing scientific literature, it will be harder than ever to identify the best evidence available, especially given the large volume of non-traditional and emerging sources of evidence: e.g., evidence derived from trial registries and data repositories; observational datasets; publications without peer review; and scientific blogging. Individual studies using conventional methods for evidence generation, especially randomized controlled trials, may be significantly flawed in their planning, conduct, analysis, or reporting, resulting in ethical violations, wasted scientific resources, and dissemination of misinformation with subsequent health harm. Furthermore, a new randomized controlled trial should be initiated or interpreted in the context of the existing evidence. However, clinical evidence extraction, appraisal, and aggregation remain laborious human tasks given its free-text format. To support evidence-based research so that new research hypothesis selection and testing can be well-grounded on the existing scientific literature and existing evidence can be easily accessible and computable to researchers, patients, or clinicians, we will develop novel, scalable, and generalizable methods for clinical evidence extraction and appraisal so that we can help the public identify reliable evidence easily. We will contribute computable evidence representations and accompanying natural language processing pipelines, achieving symbiosis between the two to support core tasks for evidence-based medicine, such as faceted evidence retrieval (e.g., “retrieve all the randomized controlled trials publications about the efficacy of HCQ on severe COVID-19 patients, with each study having a sample size over 200”), extraction and representation of clinical findings (e.g., “HCQ for people infected with COVID-19 has little or no effect on the risk of death, and probably no effect on progression to mechanical ventilation”), and evidence quality ranking and biases detection. Therefore, we propose four specific aims: Aim 1. — Represent and extract Population, Intervention, Comparison, and Outcome (PICO) information. Aim 2. — Represent and extract clinical findings and their metadata relevant for evidence quality ranking and study biases detection. Aim 3. — Develop and validate an extensible living clinical evidence knowledge graph based on the FAIR principles. Aim 4. — Develop and validate an Augmented Intelligence (AI) system for evidence appraisal. INNOVATION There is no scalable and generalizable informatics solution for literature-based, fine-grained clinical evidence extraction and representation, evidence quality ranking, evidence biases detection, and user- augmented clinical evidence aggregation and appraisal. ClinEX will be the first solution to achieve these goals.