Learning a Health Knowledge Graph from Electronic Medical Records

Abstract

Demand for clinical decision support systems in medicine and self-diagnostic symptom checkers has substantially increased in recent years. Existing platforms rely on knowledge bases manually compiled through a labor-intensive process or automatically derived using simple pairwise statistics. This study explored an automated process to learn high quality knowledge bases linking diseases and symptoms directly from electronic medical records. Medical concepts were extracted from 273,174 de-identified patient records and maximum likelihood estimation of three probabilistic models was used to automatically construct knowledge graphs: logistic regression, naive Bayes classifier and a Bayesian network using noisy OR gates. A graph of disease-symptom relationships was elicited from the learned parameters and the constructed knowledge graphs were evaluated and validated, with permission, against Google’s manually-constructed knowledge graph and against expert physician opinions. Our study shows that direct and automated construction of high quality health knowledge graphs from medical records using rudimentary concept extraction is feasible. The noisy OR model produces a high quality knowledge graph reaching precision of 0.85 for a recall of 0.6 in the clinical evaluation. Noisy OR significantly outperforms all tested models across evaluation frameworks (p < 0.01).

Robustly Extracting Medical Knowledge from EHRs: A Case Study of Learning a Health Knowledge Graph

“…In contrast to clinical trials which are conducted on restricted subpopulations, [6][7][8][9] the breadth of electronic health records (EHRs) allows for the inclusion of all patients who enter the healthcare system. 10 With this data, researchers can build models to extract general medical knowledge [11][12][13] and diagnose patients. [14][15][16] Although diagnostic knowledge exists in medical textbooks 17 or online repositories like Mayo Clinic, 18 inferring that medical knowledge automatically from EHRs provides different strengths.…”

“…Open questions include error analysis on the learned health knowledge graph and the inclusion of non-linear models. Similar to prior work, 12 we use a manually curated health knowledge graph from Google for an automated evaluation. 22 We aim to provide guidance to researchers about the relationship between dataset size, model specification, and performance for constructing health knowledge graphs.…”

Using indication embeddings to represent patient health for drug safety studies

“…Recent studies have used the EHR to evaluate variation in drug treatment decisions. [1,2] Researchers aiming to create a complete knowledge base of health and medicine have harnessed EHR data to identify symptoms related to diseases, [3,4] and to annotate the medical conditions for which drugs are prescribed, known as the indications for a drug. [5,6] Another area of research uncovers adverse effects among people taking a drug, as a way to complement randomized trials.…”

Processing of Electronic Medical Records for Health Services Research in an Academic Medical Center: Methods and Validation (Preprint)

“…However, dealing with EMR data is often labor-intensive [5] and challenging due to the lack of standardization in data entry, changes in coding procedures over time, and the impact of missing information [6,7]. Processing EMR data for analysis is a critical step in health services research requiring significant time and effort.…”

“…However, most have focused only on partial aspects of data processing [11][12][13][15][16][17][18] or processing related to a specific disease [6,11,19]. Designing an efficient and structured way to standardize records, process features, link data, and select cohorts for analysis is urgently needed given the increasing emphasis on big data and analytics to improve patient care and reduce healthcare expenditure [20,21].…”