ICRH Kenya

Modelling the impact of microbicides on HIV incidence and HIV prevalence

Team members:
Wim Delva; Patricia Claeys; Stanley Luchters; Marleen Temmerman

OBJECTIVES

General objective:

  • The overall objective of this project is to develop and apply dynamic epidemiological microsimulation models to assess the expected impact of microbicides on HIV incidence and HIV prevalence. These models will be applied in a population of female sex workers and their clients in Mombasa, Kenya. The expected impact of microbicides on HIV incidence and HIV prevalence is a function of the intrinsic efficacy of the modeled microbicide and the behavioral, demographical and epidemiological dynamics of the population in which the microbicides are introduced.

Specific objective:

  • Measurement of the demographic, behavioural and epidemiological factors that are expected to be associated with HIV incidence in female sex workers in Mombasa. Statistical analysis will investigate significant associations. Factors under study include: (a) age structure, (b) migration patterns to and from Mombasa, (c) sexual behaviour, (d) vaginal practices, (e) sexually transmitted infections and bacterial vaginosis, (f) alcohol and drug use, and (g) sexual violence.
  • Adjusting the intrinsic microbicide effects (efficacy) for bias originating from non-random selection, loss to follow-up, and indirect effects of microbicides via changes in sexual behaviour.
  • Development and projection of model simulations for different scenarios of efficacy, uptake and adherence of microbicides in the study population. These simulations will display for each scenario what the expected impact of microbicides on HIV incidence and HIV prevalence is within the given socio-demografic / epidemiological setting.

METHODS

  • The data collected in the Mombasa Cohort Study (female sex workers) for Estimation of HIV-1 Incidence allow for longitudinal analysis of key determinants of sexual risk behavior:

a) number of casual partners, regular partners, and husbands/boyfriends over the last week
b) condom use in each of these types of partnerships over the last 3 months
c) anal sex acts (ever or never) and condom use during last anal sex act in each of these partnerships
d) suspected STIs at the time of the study visit

  • First, a number of discrete probability distributions, such as the Poisson distribution, the negative binomial distribution and the zeta distribution (a power-law distribution), will be fitted to the empirical data for the number of casual partners over the last week at each study visit separately. Pearson Chi² tests, Akaike and Bayesian information criteria will be provided to compare the goodness of fit of these distributions with the empirical distribution. Similar comparisons and parameterizations will be provided for the distribution of the number of regular partners and the number of husbands/boyfriends that one had sexual contact with over the last week. Subsequently, loglinear random effects models (fitted via maximum likelihood estimation and assuming the probability distribution which fitted the data best in the previous analysis) will be used to analyze time trends in the number of partners (for each of the 3 types of partnerships) of participants of the Mombasa Cohort Study for Estimation of HIV-1 Incidence.
  • For the longitudinal analysis of condom use in each of the three types of partnerships over the last 3 months, logistic random effects models (fitted via maximum likelihood estimation) will be used. The occurrence of anal sex acts and condom use during such sex acts will be modeled via similar models.
  • Additional questions in the 12-month questionnaire of the Cohort Study for Estimation of HIV-1 Incidence will allow for cross-sectional analysis of additional determinants of sexual risk behavior:

a) number of sex acts per partner over the last week
b) proportion of protected sex acts per partner over the last week
c) age (group) of partners

  • A number of discrete probability distributions, such as the Poisson distribution, the negative binomial distribution and the zeta distribution (a power-law distribution), will be fitted to the empirical data for the number of sex acts per casual partner over the last week. Pearson Chi² tests, Akaike and Bayesian information criteria will be provided to compare the goodness of fit of these distributions with the empirical distribution. Similar comparisons and parameterizations will be provided for the distribution of the number sex acts per regular partner and the number of sex acts per husband/boyfriend that one had sexual contact with over the last week. Binomial distributions will be fitted for the parameterization of the number of protected sex acts per partner, conditional on the total number of sex acts.
  • Log-linear models will be applied to determine the pattern of dependence between the number of partners, number of sex acts per partner, proportion of protected sex acts per partner, age of the participant the age group of the partner.
  • Additional model inputs will be derived from the socio-demographical section of the questionnaires of the Mombasa Cohort Study for Estimation of HIV-1 Incidence and from other relevant data bases, such as the 2003 Kenyan Demographic and Health Survey. In this way, the age distribution, probability of male circumcision among clients of participants of the Mombasa Cohort Study for Estimation of HIV-1 Incidence, and demographic determinants of the general population will be parameterized.
  • Finally, microsimulation models will be developed based upon the parameterized distributions as obtained from the above described statistical analysis. These models will be used to make projections about:

a) the expected HIV incidence in the Mombasa Preparedness Study
b) the expected impact of microbicides on HIV incidence and HIV prevalence

  • Microsimulation models for modeling of the HIV epidemic typically model the population over time by allowing interaction between individuals such that the transition of an individual from one state into another (e.g. from HIV negative to HIV positive) is governed by stochastic processes. The stochastic processes arise from the modeled probability that a discrete event occurs over the next (small) time step. Thus, the formation of a new sexual partnership, the use of a condom during a sex act, and the infection of a sexual partner with HIV and/or another STI, is said to follow a predefined probability distribution.

EXPECTED RESULTS

By altering the parameter estimates for these probability distributions (e.g. reducing the transmission probability of HIV if a microbicide is used), one can investigate the impact of the introduction of microbicides on the incidence and prevalence of HIV in the modeled population.