Temporal trends in validated ischaemic stroke hospitalizations in the USA.

BACKGROUND: Accurate assessment of the burden of stroke, a major cause of disability and death, is crucial. We aimed to estimate rates of validated ischaemic stroke hospitalizations in the USA during 1998-2011. METHODS: We used the Atherosclerosis Risk in Communities (ARIC) study cohort's adjudicated stroke data for participants aged ≥55 years, to construct validation models for each International Classification of Diseases (ICD)-code group and patient covariates. These models were applied to the Nationwide Inpatient Sample (NIS) data to estimate the probability of validated ischaemic stroke for each eligible hospitalization. Rates and trends in NIS using ICD codes vs estimates of validated ischaemic stroke were compared. RESULTS: After applying validation models, the estimated annual average rate of validated ischaemic stroke hospitalizations in the USA during 1998-2011 was 3.37 [95% confidence interval (CI): 3.31, 3.43) per 1000 person-years. Validated rates declined during 1998-2011 from 4.7/1000 to 2.9/1000; however, the decline was limited to 1998-2007, with no further decline subsequently through 2011. Validation models showed that the false-positive (∼23% of strokes) and false-negative rates of ICD-9-CM codes in primary position for ischaemic stroke approximately cancel. Therefore, estimates of ischaemic stroke hospitalizations did not substantially change after applying validation models. CONCLUSIONS: Overall, ischaemic stroke hospitalization rates in the USA have declined during 1998-2007, but no further decline was observed from 2007 to 2011. Validated ischaemic stroke hospitalizations estimates were similar to published estimates of hospitalizations with ischaemic stroke ICD codes in primary position. Validation of national discharge data using prospective chart review data is important to estimate the accuracy of reported burden of stroke.

Duke Scholars

Author Lisa Wruck Biostatistics & Bioinformatics, Division of Biostatistics