Academic journal article Bulletin of the World Health Organization

Prediction of Community Prevalence of Human Onchocerciasis in the Amazonian Onchocerciasis Focus: Bayesian Approach. (Research)

Academic journal article Bulletin of the World Health Organization

Prediction of Community Prevalence of Human Onchocerciasis in the Amazonian Onchocerciasis Focus: Bayesian Approach. (Research)

Article excerpt


Data produced from public health research often are organized in a hierarchical structure, with clustering within units. For example, individuals "cluster" in the same community, and communities cluster within regions. Individuals who belong to the same "unit" may share common genetic, behavioural, or social risk factors of disease. They may also have similar exposures to environmental factors or, in the case of communicable diseases, infectious agents. The health outcomes of two individuals within the same unit, therefore, will correlate more highly than those of two individuals from different units. This correlation structure must be accounted for irrespective of whether data on chronic diseases or on communicable diseases are being analysed.

Hierarchical or random effect models acknowledge the nested form of such data and allow for appropriate modelling of the correlation structure (1-4). The advantages of hierarchical models are not exclusive to the Bayesian framework; nevertheless, Bayesian hierarchical models are unique in that they provide a single coherent framework that allows the incorporation of multiple sources of variability (including variability that arises from missing outcomes or exposures) and subsequent quantification of within- and between-unit variability in outcome through the investigation of potential risk factors at each "level" of the model. The appropriate pooling of information across units means that hierarchical Bayesian models also overcome problems associated with small sample sizes and thus produce more reliable estimates (or predictions) of individual- and unit-specific parameters.

A fully Bayesian approach to inference requires the specification of a full probability (likelihood) model for the data, together with a prior distribution for all the unknown parameters. Once data are available, inference is made on the basis of the posterior distribution. The posterior represents what is known currently, including the prior information and that contained in the data. By Bayes' theorem, this joint posterior distribution is proportional to the product of the likelihood function and the prior distribution.

In practice, interest lies typically with the marginal posterior distributions of a subset of parameters. In realistically complex applications, evaluation of these marginal posterior distributions requires high-dimensional integration and rarely is possible analytically. One powerful technique (and the approach taken in this paper) is the implementation of a Markov chain Monte Carlo algorithm, such as the Gibbs sampler, to obtain samples from the marginal posteriors. These sampled values are then used to describe the complete distributions for the parameters of interest or to provide summaries, such as point and interval estimates. It also is possible to estimate the posterior distribution of any arbitrary function of the parameters; this is particularly useful when estimating quantities needed to inform decision making. For example, several WHO guidelines for the control of parasitic infections use threshold prevalence values to guide priority interventions (5-6). Often, definitive diagnosis at an individual level is difficult to acquire, in which case, interest lies with the probability that, conditional on easily observable characteristics, the (unknown) prevalence of infection in a given community is above a pre-defined threshold.

This study aimed to show the use of Bayesian methods in the analysis of the type of clustered data often encountered in public health research. In particular, we developed a Bayesian hierarchical model for human onchocerciasis to explore a variety of factors thought to influence the prevalence of infection. Onchocerciasis is caused by the parasitic nematode Onchocerca volvulus and is transmitted from person to person by the bite of river-breeding blackfly vectors of the genus Simulium (7). Onchocerciasis is the second most common worldwide cause of infectious blindness, and it also causes severe and incapacitating skin disease. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.