Journal of Biometrics & Biostatistics

Time Series Analysis on Diabetes Mortality in the United States, 1999- 2015 by Kolmogorov-Zurbenko Filter

Abstract

Stella Arndorfer and Igor Zurbenko

Kolmogorov-Zurbenko filters can be utilized in the public health context analyzing mortality data. This paper aims to expand upon the robust methodology of the KZ filters and their many applications. As a low-pass filter the KZ filters are proven to be the optimal means of analysis for non-stationary data such as mortality data which usually contains various underlying signals: seasonality, long-term trend, and short-term fluctuations. As diabetes incidence and prevalence increases, the burden of health care cost increases, thus prompting the need to understand patterns underlying adverse events related to diabetes, such as mortality. Increasing incidence and prevalence of diabetes prompts the need for preventative measures and understanding what environmental factors are related to adverse events as a result of diabetes. Diabetes mortality across time analyzed with non-parametric models has not previously been studied, thus this extension to the KZ filters is utilized as a preliminary analysis to address the gap in knowledge of diabetes mortality in the United States. Non-parametric time series analysis methods identify an 8.5-year long-term trend as well as annual seasonality of diabetes mortality. Spectral and time analysis of diabetes mortality introduces the relationship between solar activity and diabetes mortality, which is quantified utilizing the cross-correlation between diabetes mortality and total solar irradiation. The strong correlation between solar activity and diabetes mortality confirms the environmental role related specifically to diabetes mortality.