CATCHING CHOLESTEROL CULPRITS: MACHINE-LEARNING ALGORITHM (MLA)-BASED APPROACH TO DETECTION OF UNDIAGNOSED FAMILIAL HYPERCHOLESTEROLEMIA (FH)

Abstract

Therapeutic Area

ASCVD /CVD Risk Reduction

Background

FH is a genetic disorder marked by elevated LDL-C and risk for premature ASCVD. Most patients remain undiagnosed. MLAs offer scalable approaches for identifying high-risk individuals using electronic health record data. The Flag-Identify-Network-Deliver™ Initiative (FIND-FH®) uses an MLA from the Family Heart Foundation to flag potential cases, followed by chart review and patient outreach. This study evaluates the impact of MLA-driven identification on FH diagnosis and management.

Methods

Patients flagged by the FIND-FH® MLA from January 2017-June 2022 at a single academic medical center (n=471) were reviewed. After excluding individuals with known FH, those with LDL-C >190mg/dL and family history of ASCVD were deemed “likely FH” (n=115). In August 2024, patients were contacted via MyChart messages and phone calls recommending FH evaluation; primary care physicians (PCP) were also notified. The primary outcome was new documentation of FH post-contact; secondary outcomes included response rates, changes in lipid-lowering therapy, and specialty of managing clinician, tracked via chart review.

Results

Of 115 identified patients, 113 were contacted; 2 had died. Forty-one (36.3%) patients responded; 43 (38.1%) viewed messages. PCP response rate was 53.2%. Seventeen (15%) patients received a new diagnosis of FH post-contact; 2 (1.8%) had a family history of FH. Post-outreach, 16 (14.2%) had new notes discussing FH, and 22 (19.5%) had changes (escalation or reduction) to lipid-lowering therapy, 6 were newly diagnosed with FH. Twenty-two (19.5%) remained untreated, 74 (65.5%) on 1 therapy, 14 (12.4%) on 2, and 3 (2.7%) on 3; most (78.8%) were on statins. Lipids were managed by PCPs (48.7%), university-based cardiologists (38.1%), outside/community cardiologists (6.2%), or others (7.1%). Patients with a new diagnosis of FH, clinical notes discussing FH, or changes to lipid-lowering therapy were deemed as having “action taken.” Mean LDL-C among patients with action taken (n=39) was 130.5mg/dL compared to 108.1mg/dL for patients without (n=69, p=0.047). Lp(a) testing occurred in 11.5%; CAC scoring in 17.7%.

Conclusions

MLA-driven outreach led to new FH documentation and enhanced care coordination for previously undiagnosed individuals. Results suggest that MLA-enhanced case finding may help address gaps in FH diagnosis and management. Future efforts include optimizing outreach and embedding realtime MLA tools into clinical workflows.