Forecasting heart failure: Seasonal alignment of heart failure outcomes in New York.

backgroundSeasonal variations have been observed in heart failure (HF) hospitalization. Numerous explanatory mechanisms have been proposed, but no prior studies have examined potential contributors directly. Our objective was to identify specific factors that could contribute to seasonal variability using a large longitudinal dataset of HF hospitalizations.

methodsHospital discharge data were obtained for all hospitals in the state of New York from 1994 to 2007. Records with a primary diagnosis of HF by the International Classification of Diseases-9 Clinical Modification (ICD-9-CM) code (428.xx and 425.xx) were included. Year and month of admission were used as predictors to evaluate outcomes of in-hospital mortality, population-adjusted daily rate of hospital admissions and length of stay (LOS) using univariable regression including a sinusoidal model to assess the seasonality of HF outcomes. Observations were then adjusted for multiple medical covariables as well as the average local monthly temperature and humidity at each hospital using data from the Global Historical Climate Network to identify potential modifiers of seasonal variability.

resultsAmong 949 907 records, the median age was 76 [inter-quartile range (IQR) 65-84 years old], and the cohort was 54% female (510 945 records). The population-adjusted rate of HF admissions per day increased by 1.1 admissions/day/year between 1994 and 2007, whereas in-hospital mortality and LOS decreased by -0.3%/year and -0.3 days/year, respectively (P < 0.001 for all). Seasonal trends were identified for daily HF admissions (February peak, P < 0.0001), mortality (January peak, P < 0.001) and LOS (January peak, P < 0.01). Cosinor analysis revealed significant periodicity for HF admission rate (amplitude = ±0.9 admissions/day/100 000 people, P < 0.001), in-hospital mortality (amplitude = ±0.47%, P < 0.001) and LOS (amplitude = ±0.23 days, P < 0.01). No other patient characteristics were significant modifiers of seasonality. Odds of mortality were highest in July rather than January when adjusted for average local temperature but not humidity.

conclusionsAdverse outcomes in patients hospitalized with HF were significantly worse in the winter months even when adjusted for patient characteristics and concurrent acute diagnoses such as pneumonia. Local ambient temperature was the strongest modifier of the observed seasonality. Given the increasing frequency of extreme weather events, additional work to determine the mechanisms of this and other environmental risk factors is urgently needed.