A Proposal to Correct Mortality Rates due to Diabetes Mellitus for Colombian Cities with Underreporting Problems, 2005 - 2017

Abstract

The mortality count for low-prevalence diseases such as Diabetes Mellitus (DM), demands serious problems of underreporting due to lack of data, generally represented as zeros in vital statistics systems, which makes it difficult to identify patterns, especially with diseases such as hypertension and diabetes, which are the main silent killer diseases. This paper proposes a correction to the mortality rates for small populations with high inflation in zeros, such as the annual series of Diabetes Mellitus (DM) for Colombian cities from 20 to 80 years old, between 2005 and 2017. A variation to the hierarchical Bayesian model of Alexander et al. (2017) is performed, replacing the original Poisson with a Negative Binomial distribution, but using the DM mortality patterns of other countries in the region in recent years. Therefore, DM mortality rates were corrected for age, sex, and city. The results showed a satisfactory correction for ages that officially report zeros, maintaining the natural gender gap, and a better correction for smaller cities, compared to larger ones. Because of this methodology, it will be possible to help the epidemiological surveillance of low-prevalence diseases such as hypertension, coronary heart disease and the like, with important levels of underreporting; but that require precise measurements at the subnational level for public health decision-making. Wrong data only leads to wrong decisions, so generating these types of corrections is vitally important.