(288) Evaluating The Impact of Electronic Health Record (EHR) Loss to Follow-Up Data in Patients with Atherosclerotic Cardiovascular Disease (ASCVD)

Publication Number: 962

Background: When conducting studies using EHR, loss to follow-up is always a concern and might leads to incomplete results thus introduce information bias.

Objectives: To quantify the impact of loss to follow-up data on measurements of outcome event and follow-up duration using ASCVD patients as example.

Methods: We identified patients aged 20 or older with incident ASCVD event during 2016 to 2018 from the National Health Insurance Database (NHID) in Taiwan. NHI is a nation-wide insurance program that covers more than 99% of all citizens’ health service reimbursement, including outpatient visits, hospitalizations and medication prescriptions. The discharge date of the first ASCVD hospitalization (=the index hospitalization) during the study period was defined as the index date. In this example, we would like to describe the follow-up of ASCVD recurrence in these patients under different data conditions: (1) full-data-link; (2) index-facility-only. The former used all medical records in NHID for each patient while the later utilized records only from the hospital that was the same with the index hospitalization, thus mimic the single-institution scenario. We followed the patients from index date to occurrence of any ASCVD event, death, lost to follow-up or end of study. Lost to follow-up was defined as failure to return to any healthcare facility for 180 days or longer from the last medical visit. Other outcomes included median follow-up of any medical visits, lipid-lowering drug use, and lipid exam.

Results: We included a total of 185,730 incident ASCVD patients. For full-data-link vs. index-facility-only scenarios respectively, there were 23,310 and 17,988 recurrent ASCVD events, corresponded to incidence rates of 95.8 and 72.0 per thousand patient-years, respectively. Under index-facility-only scenario, the incidence rate was under-estimated by approximately 25%. Median follow-up days for any medical visit, lipid-lowering drug use, and lipid exam were 516 vs. 379, 432 vs. 377, and 296 vs. 293 days for full-data-link vs. index-facility-only scenarios respectively, corresponded to percentage of under-estimation by 27%, 13% and 1%. When restricted to records from medical centers, the extent of under-estimation in index-facility-only scenario decreased by 3%.

Conclusions: Loss to follow-up due to lack of medical records from other institutions contributed to 25% under-estimation of ASCVD recurrence during a 3-year observation period among patients with ASCVD. Follow-up durations of any medical service and medication treatment but not lipid exam were under-estimated as well. Patients who were initially cared at medical centers were slightly less susceptible to the influence of loss to follow-up data.