A nationwide registry study on heart failure in Norway from 2008–2018: variations in lookback period affect incidence estimates

Abstract

Abstract Background/Introduction Incidence estimations of heart failure from registry-based studies may vary because they depend on a retrospective search in the database to exclude previous events (prevalent cases), termed the lookback period. Purpose The aim of this study was to assess to what extent different lookback periods affect temporal trends in heart failure incidence utilizing national registry data. Methods We identified all heart failure hospital contacts (ICD-10 codes I11.0, I13.0, I13.2, I42.x and I50.x) in adult Norwegian individuals in the Norwegian Patient Registry (NPR) during 2008–2018. To calculate the influence of varying lookback period on incident cases, we defined 2018 with 10 years of lookback as a reference and calculated the relative difference by using one through nine years of lookback. Temporal trends in age-adjusted incidence rates were estimated with sensitivity analyses using fixed and varying lookback periods (including all available data). Results Using a lookback period of 10 years, we identified 14 862 incident patients in 2018 (6 842 women, 8 020 men) with a diagnosis of heart failure. Compared to a 10-year lookback period, application of four, six, and eight years resulted in an overestimation of incident cases by 13.5%, 6.2% and 2.3%, respectively. This corresponds to incidence rates of 5.40, 5.04 and 4.85 per 1000 person-years, respectively. Figure 1 shows that the overestimation of incident cases declined with increasing number of years included in the lookback period. The overestimation was largest in the beginning of the observational period. When assessing temporal trends in incidence rate using a fixed lookback period, the incidence rates were lower with additional years in the lookback period. However, incidence rates increased regardless of whether four, six or eight years were applied. In contrast, incidence rates were lower and declined during the period when including all available data and thereby increasing the lookback period with time. Fig. 2 shows that a relatively shorter lookback period provided higher incidence rate estimates and that the direction of the curves were similar when using a fixed lookback period. Moreover, it shows that including all available data instead of using a fixed lookback period resulted in the misleading conclusion of declining incidence rates. Conclusions The length of the lookback period affects incidence estimates when calculating incidence rates from longitudinal health registry data. Our results suggest that one to five years of lookback is too short since incident cases are overestimated by 64% - 9%. A fixed lookback period of six year or more seems beneficial with less overestimation (≤6%). Funding Acknowledgement Type of funding sources: Other. Main funding source(s): This work was supported by the Research Council of Norway and Novartis Norway AS. KMO is a PhD-student at the University of Oslo and an employee of Novartis Norway AS. Overestimation of incident cases in 2018Incidence rates