Abstract MP90: Validity of Hospital Discharge Diagnosis Codes for Stroke: Evaluation of New American Heart Association/American Stroke Association Definitions

Abstract

Background: Characterizing International Classification of Disease (ICD-9) code validity is essential given widespread use of hospital discharge and claims databases in research. Estimates for acute stroke vary depending on the codes investigated. We sought to estimate the validity of ICD-9 codes grouped according to the 2013 American Heart Association/ American Stroke Association (AHA/ASA) updated stoke definition and to explore differences by patient characteristics and study site. Methods: Medical records (N=4,260) containing ICD-9 codes 430-438 or stroke keywords in the discharge summary were abstracted for hospitalizations of Atherosclerosis Risk in Communities (ARIC) Study cohort members from 1987-2010. A computer algorithm and physician reviewer identified definite and probable ischemic stroke, intracerebral hemorrhage (ICH) and subarachnoid hemorrhage (SAH) with differences adjudicated by a second physician. Using ARIC diagnosis as a gold standard, we calculated the positive predictive value (PPV) and sensitivity of groups of ICD-9 codes matched to stroke subtypes by the AHA/ASA (ischemic stroke: 433.01, 433.11, 433.21, 433.31, 433.81, 433.91, 434.01, 434.11, 434.91; ICH: 431; SAH: 430). We excluded codes for spinal and retinal infarcts (336.1, 362.31, 362.32), which were not validated in ARIC (N=3 events). Results: Thirty-three percent of 4,260 hospitalizations were validated as definite or probable strokes (1,251 ischemic, 120 ICH, 46 SAH), and 30% (1,276 of 4,260) of hospitalizations included ICD-9 codes identified by the AHA/ASA. The AHA/ASA code groups had PPV 76% and 68% sensitivity, compared to PPV 40% and 95% sensitivity for ICD-9 codes 430-438 (not 435) traditionally used to identify stroke. For ischemic stroke, AHA/ASA identified ICD-9 codes were present for 1,043 hospitalizations. Among these, PPV was 76% overall and was slightly higher for blacks (80%, N=400) compared to whites (74%, N=643; p=0.04). However, differences by race diminished conditional undergoing a CT scan or MRI (blacks 81%, N=390; whites 78%, N=601). Among whites, PPV for ischemic stroke ranged from 60-79%, and sensitivity ranged from 60-70% across study sites. ICD-9 codes 430-431 for ICH and SAH were present for 225 hospitalizations and had PPV 61%. PPV was higher among blacks (73%, N=89) compared to whites (53%, N=136; p=0.003), and differences by race were not diminished conditional on undergoing a CT scan or MRI. Among whites, PPV for ICH and SAH ranged from 38-60%, and sensitivity ranged from 78-89% across study sites. Conclusion: New AHA/ASA code groups had higher PPV but lower sensitivity for identifying acute stroke than a traditional code group. PPV was higher among blacks compared to whites and both PPV and sensitivity varied by study site. These data may be useful for calibrating estimates of stroke incidence generated from administrative claims data and in sensitivity analyses.