| dc.description.abstract | Introduction: Tuberculosis is second only to HIV as the greater killer worldwide due to a single infectious agent. Improving the treatment outcome of tuberculosis is part of the Millennium Development Goals. Given the infectious nature of tuberculosis, its distribution and treatment outcomes should consider spatial patterning. Information on the distribution of tuberculosis treatment outcomes in Kenya is scarce, yet treatment outcome is an important indicator of tuberculosis management. Spatial analysis tools can be used to characterize spatial patterns of these treatment outcomes, thereby identifying areas at risk of the given outcomes. This study examined the spatial distribution of the tuberculosis treatment outcomes across the counties in Kenya.
Objective: To model the spatial distribution of tuberculosis treatment outcomes using Bayesian techniques.
Methods: Study area was Kenya, a country in the East Africa region. Secondary data was obtained from the national tuberculosis registers from January 2014 to March 2014 with incorporation of data from the Kenya Demographic and Health Survey 2014 and Census 2009. Treatment outcomes were categorized as cured, dead, defaulted, failure and treatment complete. Exploratory data analysis was done to estimate the proportions of the various covariates, and tests for global and local spatial auto correlation done to assess the relationship of the various outcomes per county. Covariates were selected using purposeful selection of variables, and variables with a significant univariate test were selected as candidates for the multivariate analysis. Augmentation of the linear predictors with a set of spatially correlated random effects was done, using conditional autoregressive prior distributions, specified by a set univariate full conditional distributions. Inference was based on obtaining the posterior distribution, of the different TB treatment outcomes, using the Integrated Nested Laplace Approximation Methodology (INLA) as a way of approximating the posterior marginals as proposed by Besag et al.
Results: A total number of 23,488 records were analysed comprising of 60.38% male, 70.32% patients between the age of 15-45, patients with pulmonary tuberculosis at 82.30%, and HIV positive patients were 59.03%. There was significant global spatial autocorrelation seen for the patients who were cured, failed treatment, died and those who completed treatment. However, most of the covariates did not show significant
vii
spatial dependence. The fitted data also showed uniform distribution of the outcomes of tuberculosis treatment across the counties with occasional high risk spots.
Conclusion: The spatial effect of the tuberculosis treatment outcomes appeared weak across the various counties. This may imply that appropriate risk factors were adjusted for in the model such that the spatial random effect became less important.
Recommendation: Future related studies involving various TB outcomes should be traced at household level to minimize mismatch between risk factors and the TB outcomes.
Keywords: Bayesian statistics, Conditional Autoregressive Models, Tuberculosis treatment outcome | en_US |
| dc.description.department | a
Department of Psychiatry, University of Nairobi, ; bDepartment of Mental Health, School of Medicine,
Moi University, Eldoret, Kenya | |
