The role of acute and early HIV infection in the spread of HIV and implications for transmission prevention strategies in Lilongwe, Malawi: a modelling study

Abstract

Background: HIV transmission risk is higher during acute and early HIV infection than it is during chronic infection, but the contribution of early infection to the spread of HIV is controversial. We estimated the contribution of early infection to HIV incidence in Lilongwe, Malawi, and predict the future effect of hypothetical prevention interventions targeted at early infection only, chronic infection only, or both stages.

Methods: We developed a deterministic mathematical model describing heterosexual HIV transmission, informed by detailed behavioural and viral-load data collected in Lilongwe. We included sexual contact within and outside of steady pairs and divided the infectious period into intervals to allow for changes in transmissibility by infection stage. We used a Bayesian melding approach to fit the model to HIV prevalence data collected between 1987 and 2005 at Lilongwe antenatal clinics. We assessed interventions that reduced the per-contact transmission probability to 0.00003 in people receiving them, and varied the proportion of individuals receiving the intervention in each stage.

Findings: We estimated that 38.4% (95% credible interval 18.6-52.3) of HIV transmissions in Lilongwe are attributable to sexual contact with individuals with early infection. Interventions targeted at only early infection substantially reduced HIV prevalence, but did not lead to elimination, even with 100% coverage. Interventions targeted at only chronic infections also reduced HIV prevalence, but coverage levels of 95-99% were needed for the elimination of HIV. In scenarios with less than 95% coverage of interventions targeted at chronic infections, additional interventions reaching 25-75% of individuals with early infection reduced HIV prevalence substantially.

Interpretation: Our results suggest that early infection plays an important part in HIV transmission in this sub-Saharan African setting. Without near-complete coverage, interventions during chronic infection will probably have incomplete effectiveness unless complemented by strategies targeting individuals with early HIV infection.

Funding: National Institutes of Health, University of North Carolina Center for AIDS Research.

Figure 1. Flow diagram of modelling methods

Analyses began with model development (Step 1), followed first by a Bayesian melding procedure (Steps 2-4) to identify model parameter values most compatible with observed epidemic dynamics in Lilongwe, and next by estimation of the contribution of EHI and prediction of intervention effects (Step 5). * See Figure 2 for schematic of basic model structure and Web Appendix for corresponding equations. † See Table 1 for complete listing of model input parameters and corresponding prior distributions. ‡ See Figure 3 for plot of ANC data and posterior distribution of model simulations § See Figure 4 for estimated EHI contribution ** See Figure 5 for predicted intervention effects

Figure 2. Simplified diagram of model structure

Unshaded boxes represent single (unpaired) individuals; shaded boxes represent steady partnerships. As detailed by the accompanying labels, arrows represent flows from one compartment to another via demographic processes (entering & exiting the population), partnership formation and dissolution, or HIV transmission. For ease of illustration, the diagram does not illustrate the two separate risk groups or the multiple stages of infection.

Figure 3. HIV ANC prevalence data and posterior distribution of output prevalence curves

HIV prevalence data from the sentinel surveillance site in a Lilongwe antenatal clinic are shown as points, with the corresponding 95% confidence intervals as bracketed vertical lines. HIV prevalence output generated from the mode (i.e., best-fitting) set of input parameters is shown as the solid black curve. The dashed curves were generated from the 2·5^th and 97·5^th percentile values from the entire posterior set of model-produced prevalence predictions at each time point. The red curve represents the median value at each time point.

Figure 4. Estimated proportion of incident HIV infections attributable to contact with EHI index case

The solid curve represents the annual proportion of incident HIV infections attributable to contact with an EHI index case predicted by the mode set of input parameters. The dashed curves correspond to the simulations producing the 2·5^th and 97·5^th percentile values in the year 2010.

Figure 5. Predicted effects of interventions during EHI only, CHI only, or both periods on HIV prevalence and incidence

HIV prevalence (left panel) and incidence (right panel) in Lilongwe is shown for scenarios with no intervention (solid black curve) and for interventions initiated in 2010 with various levels of coverage in early HIV infection (EHI) and/or chronic HIV infection (CHI). The figures in the top row (5A, 5B) compare the “no-intervention” scenario with “EHI-only” interventions suppressing transmission in 25%, 50%, 75%, 85%, 90%, 95%, and 100% of those with EHI. Figures 5C-5F compare the “no-intervention” scenario with “CHI-only” interventions suppressing transmission in 25%, 50%, 75%, 85%, 90%, 95%, and 100% of those with CHI; 5C and 5D assume no increase in life expectancy associated with the intervention, while 5E and 5F assume increased life expectancy associated with the intervention (see Methods). Figures 5G-5J and 5K-5N compare the “no-intervention” scenario with four different strategies in which 75% of CHI cases (5G-5J) or 85% of CHI cases (5K-5N) are reached: one that suppresses transmission only in CHI cases, and three that also suppress transmission in 25%, 50%, and 75% of EHI cases, respectively. In these last two sets of figures, 5G, 5H, 5K, and 5L assume no increase in life expectancy associated with the CHI intervention; 5I, 5J, 5M, and 5N assume increased life expectancy associated with the CHI intervention (see Methods). All figures are based on input parameters from the modal simulation.

Figure 5. Predicted effects of interventions during EHI only, CHI only, or both periods on HIV prevalence and incidence

HIV prevalence (left panel) and incidence (right panel) in Lilongwe is shown for scenarios with no intervention (solid black curve) and for interventions initiated in 2010 with various levels of coverage in early HIV infection (EHI) and/or chronic HIV infection (CHI). The figures in the top row (5A, 5B) compare the “no-intervention” scenario with “EHI-only” interventions suppressing transmission in 25%, 50%, 75%, 85%, 90%, 95%, and 100% of those with EHI. Figures 5C-5F compare the “no-intervention” scenario with “CHI-only” interventions suppressing transmission in 25%, 50%, 75%, 85%, 90%, 95%, and 100% of those with CHI; 5C and 5D assume no increase in life expectancy associated with the intervention, while 5E and 5F assume increased life expectancy associated with the intervention (see Methods). Figures 5G-5J and 5K-5N compare the “no-intervention” scenario with four different strategies in which 75% of CHI cases (5G-5J) or 85% of CHI cases (5K-5N) are reached: one that suppresses transmission only in CHI cases, and three that also suppress transmission in 25%, 50%, and 75% of EHI cases, respectively. In these last two sets of figures, 5G, 5H, 5K, and 5L assume no increase in life expectancy associated with the CHI intervention; 5I, 5J, 5M, and 5N assume increased life expectancy associated with the CHI intervention (see Methods). All figures are based on input parameters from the modal simulation.

Figure 6. Intervention Coverage Levels Required for HIV Elimination in Lilongwe

Predicted levels of coverage required in EHI and/or CHI to result in HIV elimination within 30 years of intervention implementation, assuming (A) no increase in life expectancy associated with CHI interventions or (B) increased life expectancy associated with CHI interventions (see Methods). The “Granich definition” requires a reduction in annual HIV incidence to 1 case per 1000 persons. The “Dahlem definition” requires a reduction in annual incident cases to 0.