Cohort design and natural language processing to reduce bias in electronic health records research

Abstract

Electronic health record (EHR) datasets are statistically powerful but are subject to ascertainment bias and missingness. Using the Mass General Brigham multi-institutional EHR, we approximated a community-based cohort by sampling patients receiving longitudinal primary care between 2001-2018 (Community Care Cohort Project [C3PO], n = 520,868). We utilized natural language processing (NLP) to recover vital signs from unstructured notes. We assessed the validity of C3PO by deploying established risk models for myocardial infarction/stroke and atrial fibrillation. We then compared C3PO to Convenience Samples including all individuals from the same EHR with complete data, but without a longitudinal primary care requirement. NLP reduced the missingness of vital signs by 31%. NLP-recovered vital signs were highly correlated with values derived from structured fields (Pearson r range 0.95–0.99). Atrial fibrillation and myocardial infarction/stroke incidence were lower and risk models were better calibrated in C3PO as opposed to the Convenience Samples (calibration error range for myocardial infarction/stroke: 0.012–0.030 in C3PO vs. 0.028–0.046 in Convenience Samples; calibration error for atrial fibrillation 0.028 in C3PO vs. 0.036 in Convenience Samples). Sampling patients receiving regular primary care and using NLP to recover missing data may reduce bias and maximize generalizability of EHR research.

One Clinician Is All You Need–Cardiac Magnetic Resonance Imaging Measurement Extraction: Deep Learning Algorithm Development

“…To our knowledge, this is the first example of using a transformer-based model (without pretraining from scratch) fine-tuned on clinician labels to extract numerical measurements from diagnostic text. We previously demonstrated the value of extracting 4 vital sign measurements from clinical text based on a large number of weak labels that were generated using a rule-based approach [ 16 ]. Our previous approach was based on the assumption that it would be impractical to accrue a sufficient quantity of gold-standard annotations in order to fine-tune a transformer-based approach.…”

“…EWOC comprises 99,252 adults aged 18 years or older with ≥2 cardiology clinic visits within 1 to 3 years between 2000 and 2019. A broad range of EHR data are available for each individual in the cohort, including demographics, anthropometrics, vital signs, narrative notes, laboratory results, medication lists, radiology and cardiology diagnostic test results, pathology reports, and procedural and diagnostic administrative billing codes [ 16 ]. These data were processed using the JEDI Extractive Data Infrastructure [ 17 ].…”

“…Baseline characteristics were defined using previously published groupings of International Classification of Diseases, 9th and 10th revision diagnosis codes [ 16 ]. Definitions for clinical features used in the analysis are provided in Table S1 in Multimedia Appendix 1 .…”

“…BERT was originally pretrained on general domain text such as Wikipedia but has since been adapted for use in clinical applications by pretraining on domain-specific text [11][12][13][14]. Although transformer-based models have shown efficacy in extracting nonnumerical entities such as anatomical terms and disease states from clinical text [14], their application to extracting numerical quantities from clinical text has been limited [15,16].…”

Establishment of a Simple Pediatric Lower Respiratory Tract Infections Database Based on the Structured Electronic Medical Records

“…Electronic medical records (EMRs) are widely used in hospitals, and a large amount of clinical data are generated by EMRs in daily medical activities. ( 3 ) In recent years, the Surveillance, Epidemiology, and End Results (SEER) and Medical Information Mart for Intensive Care (MIMIC) databases have been known worldwide ( 4 , 5 ), which provided clinicians with numerous clinical data to deal with. Clinicians and researchers have realized that the clinical data generated by the actual clinical practices in their own centers may help to provide more appropriate solutions for the medical problems in their daily work.…”