Small Area Estimation With An Application To Bivariate Spatial Modelling Of Hypertension And Diabetes Prevalence In Kenya

Abstract

Background Comorbidity of hypertension and diabetes leads to signi cant risks of mortality and other non-communicable diseases (NCDs) such as heart attacks and strokes. Kenya, like many low and middle-income countries (LMICs), faces a rapid increase in NCDs burden. However, sub-national burden pro les to inform health policy at the county level; the current health planning units are implausible due to small sample sizes from the existing NCDs data sources in Kenya. The main objective of this study was to determine the distribution of hypertension and diabetes disease prevalence at county units in Kenya using small area estimation methods. Methods Data from a nationally representative Kenya STEPwise survey for NCDs risk factors (STEPs-2015) was used. The survey collected health information (physical and biochemical measurements), risky behaviour and demographic indicators related to NCDs for 4,500 persons aged 18-69 years. Multivariate conditional autoregressive models that account for spatial autocorrelation and dependence between diseases (latent e ects) were t to estimate the county-speci c prevalence of hypertension and diabetes. Simple multivariate improper CAR, improper multivariate CAR, proper multivariate CAR and M-model latent e ects were explored. A mixed-e ects multinomial logistic regression model was t to identify the macro-risk factors of hypertension and diabetes in Kenya. Results The M-model was selected as the best t based on DIC. Substantial geographical variation in the prevalence of hypertension ranging from 9% in Wajir county and 54% in Nyeri county while diabetes ranged from 0.1% in Narok to 8.1% in Makueni were observed. Overall, 47% (22 counties) and 36% (17 counties) had hypertension and diabetes prevalence estimates above the national burden, 26.4% and 2.7% respectively. Notably, Mombasa, Kiambu, Embu and Nyeri had a substantial burden of both hypertension and diabetes. High cholesterol, central obesity, age, BMI, harmful alcohol intake and high sugar intake were signi cantly associated with hypertension and diabetes. Conclusion The county-speci c prevalence estimates provide the first evaluation of hypertension and diabetes burden that policymakers can use to inform interventions aimed at prevention and treatment of NCDs in Kenya. Implementation of comprehensive screening programs and awareness building for NCDs control are crucial in reducing hypertension and diabetes burden in Kenya.