Natural Language Processing to Assess Atherosclerotic Cardiovascular Disease, Diabetes, and Familial Hypercholesterolemia Lipid Management in Canada

Background: Elevated low-density lipoprotein cholesterol (LDL-C) leads to atherosclerotic cardiovascular disease (ASCVD). This study assessed treatment patterns & achievement of guideline-recommended LDL-C levels in Canadian patients with ASCVD, diabetes mellitus (DM), or familial hypercholesterolemia (FH).Methods: Natural language processing (NLP) was utilized to extract demographic, clinical characteristics, and lipid lowering treatment (LLT) information from de-identified electronic health records of patients from cardiology or internal medicine settings in 4 provinces. The study period spanned from 1-January-2016 to 30-November-2020, and included identification, baseline, and 12-month follow-up periods. Results: A total of 10,992 patients were identified; ASCVD (n=9,415), DM (n=1,132), and FH (n=445). Failure to achieve recommended LDL-C levels was common at baseline (38% ASCVD, 38% DM, and 75% FH) and at follow-up for patients with uncontrolled baseline LDL-C (43% ASCVD, 55% DM, and 52% FH). There was no documented LLT in 33-49% of patients with uncontrolled baseline LDL-C. LDL-C was not documented in 45%, 59%, and 23% of patients with ASVCD, DM, and FH, respectively. LDL-C levels decreased over time in all patients, with the largest decrease in patients receiving PCSK9 monoclonal antibodies, ezetimibe, or high intensity statins.Conclusions: The present study revealed that over a third of patients with uncontrolled baseline LDL-C lacked documented LLT, almost 50% of patients did not attain recommended LDL-C levels, and that treatment modification in patients with uncontrolled LDL-C could have been more intensive. Our findings were consistent with studies using traditional administrative datasets, suggesting a promising role for NLP in future quality improvement initiatives and research.