Concordance between clinical outcomes in the Systolic Blood Pressure Intervention Trial and in the electronic health record
Chu CD, Lenoir KM, Rai NK, Soman S, Dwyer JP, Rocco MV, Agarwal AK, Beddhu S, Powell JR, Suarez MM, Lash JP, McWilliams A, Whelton PK, Drawz PE, Pajewski NM, Ishani A, and Tuot DS. Concordance between clinical outcomes in the systolic blood pressure intervention trial and in the electronic health record. Contemp Clin Trials 2023; 128:107172.
Contemp Clin Trials
BACKGROUND: Randomized trials are the gold standard for generating clinical practice evidence, but follow-up and outcome ascertainment are resource-intensive. Electronic health record (EHR) data from routine care can be a cost-effective means of follow-up, but concordance with trial-ascertained outcomes is less well-studied.
METHODS: We linked EHR and trial data for participants of the Systolic Blood Pressure Intervention Trial (SPRINT), a randomized trial comparing intensive and standard blood pressure targets. Among participants with available EHR data concurrent to trial-ascertained outcomes, we calculated sensitivity, specificity, positive predictive value, and negative predictive value for EHR-recorded cardiovascular disease (CVD) events, using the gold standard of SPRINT-adjudicated outcomes (myocardial infarction (MI)/acute coronary syndrome (ACS), heart failure, stroke, and composite CVD events). We additionally compared the incidence of non-CVD adverse events (hyponatremia, hypernatremia, hypokalemia, hyperkalemia, bradycardia, and hypotension) in trial versus EHR data.
RESULTS: 2468 SPRINT participants were included (mean age 68 (SD 9) years; 26% female). EHR data demonstrated ≥80% sensitivity and specificity, and ≥ 99% negative predictive value for MI/ACS, heart failure, stroke, and composite CVD events. Positive predictive value ranged from 26% (95% CI; 16%, 38%) for heart failure to 52% (95% CI; 37%, 67%) for MI/ACS. EHR data uniformly identified more non-CVD adverse events and higher incidence rates compared with trial ascertainment.
CONCLUSIONS: These results support a role for EHR data collection in clinical trials, particularly for capturing laboratory-based adverse events. EHR data may be an efficient source for CVD outcome ascertainment, though there is clear benefit from adjudication to avoid false positives.
Medical Subject Headings
Aged; Female; Humans; Male; Acute Coronary Syndrome; Antihypertensive Agents; Blood Pressure; Cardiovascular Diseases; Electronic Health Records; Heart Failure; Hypertension; Myocardial Infarction; Stroke; Treatment Outcome