Browsing by Author "Mhlanga, Laurette"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Optimisation and benchmarking of analytical approaches to estimation of population level HIV incidence from survey data
    (Stellenbosch : Stellenbosch University, 2022-04) Mhlanga, Laurette; Welte, Alex; Grebe, Eduard; Stellenbosch University. Faculty of Medicine and Health Sciences. Dept. of Global Health. Epidemiology and Biostatistics.
    ENGLISH SUMMARY: Disease prevalence (the proportion of a population with a condition of interest) is conceptually and procedurally much more straightforward to estimate than disease incidence (the rate of occurrence of new cases - for example, infections). For long-lasting conditions, incidence is fundamentally more difficult to estimate than prevalence, but also more interesting, as it sheds light on current epidemiological trends such as the emerging burden on health systems and the impact of recent policy interventions. Progress towards reducing reliance on questionable assumptions in the analysis of large population based surveys (for the estimation of HIV incidence) has been slow. The work of Kassanjee et al and the work of Mahiane et al, in particular, provide rigorous ways of estimating incidence by using 1) markers of ‘recent infection’, 2) the ‘gradient’ of prevalence, and 3) ‘excess mortality’ associated with HIV infection, without the need for simplifying assumptions to the effect that any particular parameters are constant over ranges of time and/or age. To date, the use of these methods has largely ignored 1) the rich details of the age and time structure of survey data, and 2) the opportunities for combining the two methods. The primary objective of this work was to find stable approaches to applying the Mahiane and Kassanjee methods to large age/time structured population survey data sets which include HIV status, and optionally, ‘recent infection’ status. In order to evaluate proposed methods, a sophisticated simulation platform was created to simulate HIV epidemics and generate survey data sets that are structured like real population survey data, with the underlying incidence, prevalence, and mortality explicitly known. The first non-trivial step in the analysis of survey data amounts essentially to performing a smoothing procedure from which the (age/time specific) prevalence of HIV infection, the prevalence of ‘recent infection’, and the gradient of prevalence of infection can be inferred without recourse to ‘epidemiological’ assumptions. The second step involves the correct accounting for uncertainty in a context-specific weighted mean of the Mahiane and Kassanjee estimators. These two steps are approached incrementally, as there are numerous details which have not previously been systematically elucidated. The investigation culminates in a proposed generic ‘once size fits most’ algorithm based on: 1) fitting survey data to generalised linear models defined by simple link functions and high order polynomials in age and time; 2) the use of a ‘moving window’ rule for data inclusion into a separate analysis for each age/time point for which incidence is to be estimated; 3) a ‘variance optimal’ weighting scheme for the combination of the Mahiane and Kassanjee estimators (when both are applicable); 4) flexible use of a delta method expansion or bootstrapping to estimate confidence intervals and p values. We find it is relatively easy to obtain estimates with practically negligible bias, but samplesizes/ sampling-density requirements are always considerable. We also make numerous observations on survey design and the inherent challenges faced by all attempts to estimate HIV incidence using surveys of reasonable size.