Modelling Spatio-Temporal Patterns Of Disease Risk For Data With Misalignment And Measurement Errors: An Application On Measles And Hiv Prevalence Data In Namibia

Abstract

Disease mapping has important applications in public health because it enables the

identication of areas which are at high risk of particular health problems. It helps

visualising the spatial pattern of the disease distribution, which is of interest to

the health sector as it enables the sector to plan, evaluate and redesign prevention

and control strategies, and also make important policy decisions particularly for

geographically targeted intervention in resource poor settings. Analyses of spatial

disease patterns are generally based on data of a single disease and they are often

fraught with challenges that include lack of a representative sample, often incomplete

and most of which may have measurement errors, and may be spatially and

temporally misaligned. This thesis focused on the development and extension of

statistical models with particular interest to dealing with misalignment, measurement

errors and jointly modeling of data from multiple sources.

The rst objective was to estimate and map the risk of measles at a sub-region level

(i.e. constituency level) in Namibia using data obtained at the regional level. Direct

inferences at constituency level made on basis of the original level of aggregation

may lead to an inferential problem known as a misalignment in the statistical literature.

Using measles data from Namibia for the period 2005-2014, both multi-step

and direct approaches were applied to correct the misalignment. The multi-step

approach model provided a relatively better model.

The second objective was to t a spatio-temporal model while dealing with misalignment

and measurement error, again applied to measles data aggregated at regional

level over the period 2005 to 2014. Again this leads to a spatial misalignment problem

if the purpose is to make decisions at constituency level. Moreover, data on

risk factors of measles were not available each year between 2005 and 2014. Thus,

assuming that covariates were constant through the study period would induce measurement

errors which might have eects on the analysis results. The multi-step

approach was further extended to include temporal eects and account for measurement

errors. Consequently, spatio-temporal models, which included Bernardinelli

and Knorr-Held approaches, and classical measurement error models were adopted.

Comparison of the results obtained from the nave method (i.e. modelling that

i

ignored errors in covariates) and those from the approach that accounts for measurement

errors showed that the latter modelling approach performed better than

the former. The study showed a spatio-temporal variation of the measles risk over

the 2009-2014 period.

The third objective of this study was to develop a joint spatial model for HIV prevalence,

using two sources (i.e. 2014 National HIV Sentinel survey (NHSS) among

pregnant women aged 15-49 years attending antenatal care (ANC) and the 2013

Namibia Demographic and Health Surveys (NDHS)), which would enable the estimation

at any location of the constituency or district level while dealing with misalignment

in data. The shared component modelling approach was adopted through

the use of stochastic partial dierential equations (SPDE). The bivariate modelling

approach developed allowed to combine two data sources that are available at different

spatial levels in a single model and it catered for a specication of dierent

spatial processes through the link function. Findings revealed that health districts

and constituencies in the northern part of Namibia were highly associated with HIV

infection. Also, the study showed that the place of residence, gender, gravida, marital

status, number of kids dead, wealth index, education, and condom use were

signicantly associated with HIV infection in Namibia. Finally, it was shown that

the prediction of HIV prevalence using the NDHS data source can be enhanced by

jointly modelling other HIV data such as NHSS data.

In conclusion, results showed that the multi-step approach may be used to deal with

misalignment. Moreover, introducing the error model proved to be a useful approach

to correct for measurement errors in data and improve inferences in situations where

mismeasured values in covariates are encountered instead of nave analyses that

ignore the presence of errors in measurements. Lastly, the thesis showed that the

prediction of HIV prevalence using the NDHS data source can be enhanced by jointly

modelling other HIV data such as NHSS data.

ii