Abstract:
Diabetes is a significant public health issue in developing countries, with an increasing burden on
the healthcare system. However, accurate reporting of diabetes cases is often hindered by under reporting, particularly in rural areas where access to healthcare is limited. When dealing with count
data, both under-reported and over-reported cases are encountered. If it is assumed that the count
data obtained from the field is always accurate, then modelling it with other count-data models will
be erroneous. This study aimed to improve the existing Poisson-Binomial mixture model by
factoring in covariates to make it suitable to estimate the number of under-reported diabetes cases
in each county of Kenya and map the distribution of these cases. The covariates used in the model
include the education level, poverty index, and access to healthcare in respective counties, making
the probability of reporting vary from one county to another. The data was obtained from the Kenya
Diabetes Management Information Centre and Kenya National Bureau of Statistics. The results
revealed that at least each of the 47 counties had under-reported the diabetes data, with the
probability of reporting ranging from 0.9002423 for Migori County and 0.7164098 for Mombasa
County. Nairobi and Mombasa counties reported the highest underreporting rate with 16,708 and
11,784 cases, respectively underreported, while Lamu had 1269 underreported cases, the least in
all the 47 counties. The resulting maps identified high-risk areas for under-reporting, and the
prevalence which provides valuable information for policymakers and public health practitioners
to target resources towards improving diabetes prevention and management in Kenya
