Machine learning algorithm identifies FH in health care system
Click Here to Manage Email Alerts
A machine learning algorithm to identify patients with familial hypercholesterolemia was successfully implemented in a large health care system, according to a presentation.
However, according to the study presented at the virtual American Society for Preventive Cardiology Congress on CVD Prevention, among the patients identified as having suspected familial hypercholesterolemia (FH), a low percentage of them came into the clinic for further evaluation.
The study was the Early Career Presentation First Place Winner at the meeting.
Samip Sheth, BA, who was a student at the University of Pennsylvania during the conduct of the study and is now a second-year medical student at Georgetown University School of Medicine, and colleagues implemented the FIND FH algorithm into health care encounter data of 1,607,606 patients (mean age, 47; 58% women) from the Penn Medicine system.
As Healio previously reported, this machine learning model was able to analyze large health care encounter databases to identify patients with FH.
“We not only have a consensus definition of familial hypercholesterolemia, but we know that FH is very underdiagnosed in the United States,” Sheth said during the presentation. “Approximately 90% of FH patients in the United States are undiagnosed. Even more than that, for those who are diagnosed, many of them are undertreated. This creates a very large problem in the United States today.”
This algorithm uses health care encounter data to identify patients with suspected FH.
“One interesting thing about the algorithm is that ... it does not require patients to have specific LDL results in their [electronic health record],” Sheth said during the presentation. “Approximately half of the patients who were flagged did not have LDL results.”
The algorithm flagged 8,614 individuals (mean age, 62 years; 57% women) as having suspected FH.
After the preventive cardiology clinic reviewed the list of flagged patients, 442 providers granted permission for the researchers to contact 1,607 of their patients about scheduling a preventive cardiology visit.
From that group, 153 flagged patients underwent workup in a preventive clinic. This cohort had mean total cholesterol 234 mg/dL and mean LDL 151 mg/dL at time of presentation to the clinic. In addition, 44% had a family history of premature atherosclerotic CVD, 14% met Dutch Lipid Criteria and 3% met MEDPED criteria. A clinical or genetic diagnosis of FH occurred in 30% of patients who came in for workup.
The algorithm was able to benefit both patients who were diagnosed with FH and those who were not, according to the presentation.
“For patients that were flagged by the algorithm and did not ultimately have FH after not being diagnosed in the clinic, we were still able to clinically manage them in a better way, whether this was intensifying therapy, referring to in-house nutritional counseling, ordering specific labs and imaging to improve their risk profile,” Sheth said.
This study also found that not many physicians and patients were aware of FH.
“We did see a low awareness of FH among physicians, cardiologists and primary care physicians alike and patients in terms of distinguishing genetically high cholesterol from environmentally high cholesterol,” Sheth said.
Collapse
Sheth S. Early Career Presentation First Place. Presented at: American Society for Preventive Cardiology Congress on CVD Prevention; July 25-26, 2020 (virtual).
Click Here to Manage Email Alerts