EHRs may underestimate quality of care

There is good news and bad news for advocates of using EHRs to examine the quality of cholesterol management care. Researchers reported that it is feasible to measure the performance of providers who treat patients with coronary artery disease (CAD) based on EHR data, but the accuracy of the assessment is suspect.

EHRs increasingly play a role in efforts to track adherence to guideline-based management of lipids in patients at risk of cardiovascular events. Healthcare systems may use them internally to inform quality improvement initiatives and externally to report quality measures.

Christopher P. Danford, MD, of the Duke Clinical Research Institute in Durham, N.C., and colleagues assessed the feasibility and accuracy of lipid performance measures using EHRs in a retrospective cohort study published online Sept. 17 in the American Heart Journal. For their study, they obtained data on 3,779 patients with CAD using their medical center’s EHR. They also manually reviewed 300 randomly selected charts. To be eligible, patients needed to have visited a Duke Cardiology provider at least twice in 2009 to 2010.

They examined three endorsed performance measures: obtaining low-density lipoprotein (LDL) cholesterol measurements; determining if they met the threshold of 100 mg/dL or less; and statin prescription. The 300 chart reviews included 100 patients who did not have an LDL measurement; 100 who were not prescribed a statin; and 100 who were prescribed a statin.

Danford et al reported that according to the EHR, 73 percent of patients had an LDL measurement within the past three years. Of those patients, about a third had LDL cholesterol levels of 100 mg/dL or less and 88 percent received a statin prescription.

But the manual review found that 15 percent of the 100 patients with CAD who, according to the EHR, had not been prescribed a statin actually had been prescribed a statin in the clinic note. Of the remaining 85 patients, 44.7 percent had an electronically documented allergy or intolerance to statins.

“Compared with manual chart review, our results suggest that while medication use and laboratory data were well captured in EHRs, there were discrepancies in CAD patient identification data, limitations in capturing indications and contraindications to lipid measurement and statin prescription, and a lack of integration with outside data sources such as pharmacy dispensing records or outside laboratory data,” they wrote. “These limitations may lead to underestimates of guideline adherence.”

Based on their manual review, Danford et al calculated that the EHR data had an accuracy of 93 percent. They also noted that claims-based data were sensitive but not specific for CAD. They recommended that future data repositories tap resources beyond administrative data and strive for more interconnectivity between health systems.

“These limitations need to be carefully considered prior to the widespread implementation of automated care quality assessment,” they concluded.