Use of electronic health records for ascertainment of disease outcomes in large population-based studies holds much promise due to low costs, diminished study participant burden, and reduced selection bias. However, the validity of cardiovascular disease endpoints derived from electronic records is unclear.Methods:
Participants were 7860 study members of the UK Whitehall II cohort study. We compared cardiovascular disease ascertainment using linkage to the National Health Service’s Hospital Episode Statistics database records (hereafter, “HES-ascertainment”) against repeated biomedical examinations—our gold standard ascertainment method (Whitehall-ascertainment). Follow-up for both methods was from 1997 to 2013 for coronary heart disease and from 1997 to 2009 for stroke.Results:
We identified 950 prevalent or incident nonfatal coronary heart disease cases and 118 prevalent or incident nonfatal stroke cases using Whitehall-ascertainment. The corresponding figures for HES-ascertainment were 926 and 107. For coronary heart disease, the sensitivity of HES-ascertainment was 70%, positive predictive value 72%, specificity 96%, and the negative predictive value 96%. The pattern of results for stroke was similar. These statistics did not differ in analyses stratified by age, sex, baseline risk factor status, or after exclusion of prevalent cases. Estimates of risk factor–disease associations were similar between the two ascertainment methods. Including fatal cardiovascular disease in the outcomes improved the agreement between the methods.Conclusion:
Our analyses support the validity of cardiovascular disease ascertainment using linkage to the UK Hospital Episode Statistics database records by showing agreement with high resolution disease data collected in the Whitehall II cohort.