Background:
Estimation of disease incidence from administrative data requires an adequate look-back (prevalence) period to exclude pre-existing conditions from the incidence risk set. We characterized optimal lengths of the prevalence period to minimize misclassification of incident heart failure (HF) hospitalization, a proxy for incident HF.
Methods:
Data for participants of the ARIC Study (a prospective longitudinal cohort of 15,792 individuals sampled from 4 US communities) were linked with CMS Medicare claims from the years 2000-2012. We included only participants with >36 months of continuous CMS Medicare fee for service (FFS) enrollment. Each participant’s time-in-observation was divided into two phases. The first 36 months were the prevalence period. Observation time after an index date 36 months following the date of enrollment was the incidence period. HF hospitalizations were identified from CMS MedPAR records using ICD-9 code 428.xx in any position. Patients were classified as having a HF hospitalization in (a) both the prevalence and incidence periods, (b) in the prevalence period only, (c) in the incidence period only, or (d) neither. Incident HF was defined as the first HF hospitalization in the incidence period not preceded by a HF hospitalization in the prevalence period. The proportion of events misclassified as incident HF hospitalization was estimated from incremental reductions of the prevalence period to start 36, 30, 24, 18, 12, or 6 months before the index date. The impact of misclassification was estimated as differences in incidence per 1,000 patients at risk.
Results:
Of 11,054 ARIC participants enrolled in Medicare FFS, 9,568 met the study inclusion criteria. A total of 1,129 incident HF hospitalizations were identified based on the 36 month prevalence period, considered as the referent (incidence rate 118 HF hospitalizations per 1,000 patients at risk). Shortening the prevalence period to 24 months increased the HF incidence rate to 123 per 1,000, overestimating the number of incident HF hospitalizations by 4.2% while retaining over 90% of the sample. A 12 month prevalence period yielded an overestimation of the number of incident HF hospitalizations by 11% (incidence rate 129 per 1,000 patients at risk) while retaining 95% of the sample.
Conclusions:
Selection of too short of a prevalence period to define incident hospitalized HF from CMS Medicare claims data can introduce substantial misclassification. Consideration of several prevalence periods indicates that a 24 month prevalence period reduces the potential for bias in the estimation of incident hospitalized HF while retaining most observations.
Associations Between Metabolomic Compounds and Incident Heart Failure Among African Americans: The ARIC Study
