Swaziland has the most severe HIV epidemic in the world, with a measured HIV prevalence of 26% among adults aged 15–49 years in 2006–07.¹ To combat this epidemic, in 2009, the Government of the Kingdom of Swaziland (GKOS) initiated support for scale-up of national HIV prevention and treatment programmes, including a voluntary medical male circumcision campaign.² The Swaziland HIV Incidence Measurement Survey (SHIMS) was designed to assess the effect of these programmes on HIV incidence by prospectively measuring HIV seroconversions in a household-based, nationally representative sample of adults before and after programme expansion.

The approach of “know your epidemic, know your response”³ is crucial to effective HIV programmes, and having accurate and detailed HIV incidence estimates is fundamental to this approach. Incidence estimates allow identification of groups at the highest risk of new infections and, when repeated, determine the effect of programmes over time. Although incidence estimates of large populations are often modelled from trends in HIV prevalence,⁴ such modelled estimates provide little demographic detail. HIV incidence laboratory assays^5,6 are intended for use in cross-sectional surveys but have yet to achieve optimum performance.⁷ HIV incidence estimates based on the gold standard of observed HIV seroconversions have not been available at a national level although they have been available from randomised clinical trials or observational cohorts restricted to subpopulations at increased risk of HIV.^8,9 We report the national estimate, before programme expansion, of HIV incidence based on population-level, prospectively observed seroconversions.

Between Dec 10, 2010, and June 25, 2011, study staff approached 14 891 selected households from the four administrative regions of Swaziland (figure 1). Among the 13 335 occupied households (figure 2), the head of household for 12 571 (94%) households provided information about 54 655 household members, of whom 24 630 (45%) were eligible for inclusion in this study on the basis of age and residence. Of these, 18 177 (74%) adults agreed to participate in the survey and undergo HIV testing; survey participation rates were higher in women (11 044 [81%] of 13 582) than in men (7133 [64%] of 11 048). Of the 18 172 participants with available HIV test results, 5803 (32%) were HIV seropositive and 13 were NAAT positive, as previously reported.^10,13 Among the remaining 12 369 potentially eligible HIV-seronegative individuals, 11 897 (96%) enrolled in the cohort and, of these, 11 232 (94%) were successfully retained and retested for HIV at a 6 month follow-up visit which occurred between Aug 23, 2011, and Feb 4, 2012; the mean duration of follow-up was 6·5 (IQR 6·0–6·7) months.

Survey-weighted estimates show the demographic and behavioural profile of the population of HIV-seronegative adults in Swaziland in 2011 (table 1). The mean age was 28·3 years (27·4 years in men; 29·3 years in women), with about half of participants aged between 20–29 years. Most participants had completed either primary or secondary education, were living in rural areas, and were unemployed. About a third of the seronegative population reported no previous HIV testing.

Among those who reported ever having had sex, most reported at the follow-up visit having one sexual partner within the past 6 months and 967 (10%) of 9855 participants: 874 (18%) of 4788 men and 93 (2%) of 5067 women reported having two or more partners in the past 6 months. Among individuals reporting one or more partners in the past 6 months, most reported all partners as HIV negative, 917 (12%) of 8048 reported one or more HIV-positive partner, and 920 (11%) of 8048 reported having any partner with unknown HIV status. Among individuals reporting one or more partners in the past 6 months, only 24 (<1%) of 8048 reported anal sex, and most opted not to answer this question (data not shown). Potentially eligible HIV-seronegative adults who did not participate in the incidence cohort were less likely to be married (p=0·003), more likely to have two or more partners (p=0·002), and were less likely to have previously tested for HIV (p=0·015).

After applying survey weighting, 145 HIV seroconversions occurred during 6086 person-years of observation, corresponding to an annualised population incidence estimate of 2·4 per 100 person-years (95% CI 2·1–2·8). Incidence was nearly twice as high in women (3·1 per 100 person-years, 95% CI 2·6–3·7) as in men (1·7 per 100 person-years, 95% CI 1·3–2·1; p<0·0001) and patterns of incidence by demographic and behavioural characteristics varied by sex (table 2, figure 3).

HIV incidence in men peaked in those aged 30–34 years (3·1 per 100 person-years, 95% CI 1·9–5·1). HIV incidence was high in those reporting some condom use in the past 6 months (3·1 per 100 person-years, 2·1–4·3), two or more partners (3·8 per 100 person-years, 2·5–5·6), or a partner with unknown HIV status (7·0 per 100 person-years, 4·6–10·5). Age, employment status, sexual activity in the past 6 months, number of partners in the past 6 months, partner’s HIV status, and condom use were included in the multivariable model for the men, and partner’s HIV status in the past 6 months emerged as the only significant independent predictor of seroconversion for men (table 3). In analyses adjusted for sexual activity, number of partners and condom use, all in the past 6 months, as well as age and employment, the risk of seroconversion was markedly higher in those men reporting a partner who was either HIV positive (HR 2·67, 95% CI 1·06–6·82) or of unknown HIV status (4·64, 2·32–9·7, p<0·0001), compared with reporting only HIV-negative partners. Reporting an HIV-positive partner accounted for 16% of HIV incidence in men and a partner of unknown status 30%, on the basis of the calculated population attributable risk (data not shown).

HIV incidence peaked in women in two age strata: 4·3 per 100 person-years (95% CI 3·3–5·6) in 20–24 year olds and 4·0 per 100 person-years (2·2–7·3) in 35–39 year olds (table 2, figure 3). HIV incidence was 4·1 per 100 person-years (3·3–5·0) in unmarried women, 3·7 per 100 person-years (2·4–5·5) in married women not living with their partner, and 1·4 per 100 person-years (0·9–2·2) in married women living with their partner. Among women reporting sexual risk behaviours or partner characteristics associated with risk, incidence was increased: for example, two or more sexual partners in the past 6 months (HIV incidence was 10·0 per 100 person-years, 5·0–19·2) or not knowing the HIV status of a partner from the past 6 months, (8·0 per 100 person-years, 5·3–12·0).

Age, marital status, sexual activity, condom use, and partner’s HIV status were included in the multivariable model for women, and marital status, sexual activity in the past 6 months, and partner’s HIV status remained the significant independent predictors of seroconversion (table 3). The risk of HIV seroconversion was nearly three times higher in women who reported not being married than those who were married (HR 2·90, 95% CI 1·44–5·84, p=0·0030) or having a marital partner staying elsewhere than in those who live with their partner (2·66, 1·29–5·45, p=0·0078). Not being married or being married with a partner living elsewhere accounted for 50% and 25% of HIV incidence among women, respectively (data not shown). Reporting no sexual activity at either baseline or follow-up was protective, with an HR of 0·22 (95% CI 0·05–0·99, p=0·048). Reporting a partner in the past 6 months with HIV-positive status was associated with an increased risk of seroconversion (HR 1·78, 0·97–3·27, p=0·063); by contrast, reporting a partner with unknown HIV status in the past 6 months predicted nearly three times the risk of seroconversion compared with reporting only HIV-negative partners (HR 2·87, 1·44–5·84, p=0·0030). Reporting any partner with unknown HIV status accounted for 16% of HIV incidence (data not shown).

The national HIV incidence estimate from SHIMS, a prospective survey of adults in Swaziland, on the basis of observed seroconversions was 2·4% in 2011. Modelled estimates of HIV seroincidence have ranked Swaziland’s HIV incidence as the highest globally for more than a decade, at 4·07% in 2001 and 2·66% in 2009.^17,18 Taking into account important differences in methods, the 2011 SHIMS estimate is consistent with previous modelled estimates, confirming that HIV incidence in Swaziland is alarmingly high but might be stable. Key subsets of the population, including women aged 20–24 years and 35–39 years and men aged 30–34 years, had incidence rates substantially higher than the national rate.

Few characteristics were independent predictors of HIV seroconversion in our study, emphasising the homogeneous nature of this generalised epidemic. For women, reporting no sexual activity at baseline or follow-up was protective, as might be expected; this predictor, however, had only marginal statistical significance. For both women and men, a significant predictor of HIV acquisition was reporting a sex partner with unknown HIV status in the past 6 months. Although most (89%) of the sexually active, seronegative population of adults in Swaziland reported knowing the HIV status of partners they have been with in the past 6 months, only 65% of seronegative adults and 71% of the overall adult population in Swaziland¹⁰ reported any HIV testing before participation in SHIMS. This discrepancy between knowledge of partner’s status and the prevalence of reporting prior testing suggests it will be crucial to expand HIV testing through a wide range of approaches while encouraging HIV disclosure to partners.¹⁹ The high population attributable risk observed with reporting partners in the past 6 months with unknown HIV status, and for women, being unmarried, also suggests the need to consider the use of pre-exposure prophylaxis,²⁰ expansion of voluntary medical male circumcision, and continued scale-up of treatment as prevention, with use of antiretroviral therapy at higher CD4 thresholds.^21–23

Key strengths of this study included the systematic sampling of a household-based population, a high rate of participation in those sampled, and a high cohort retention rate. Although the study did not directly assess the role of migration and migrant labour on the epidemic, transient individuals who spent the previous night in someone’s home were included in the sampled population. Additionally, having a partner who lives elsewhere, a characteristic which might indicate a partner involved in migrant labour, accounted for 25% of the HIV incidence in women. A rigorous HIV testing algorithm excluded seronegative individuals with acute HIV detected by NAAT testing at enrolment and determined the primary endpoint of HIV seroconversion. The number of seroconverters permitted detailed assessment of predictors of HIV acquisition. Moreover, Swaziland’s geographic location near other high prevalence countries in southern Africa make the results of this study highly relevant to the southern Africa region, the centre of the global HIV epidemic.

Study limitations include some imbalance in survey participation, with fewer men and fewer younger individuals taking part. The sex bias reflects the greater number of women than men in the overall Swaziland population (53% vs 47%¹) and higher participation rates by women than men in the SHIMS household survey (81% vs 64%¹⁰). These sex and age biases, however, were reduced by the weighting of the data. Because we did not include injection drug use and men who have sex with men behaviours in the model, we were unable to quantify the attributable risk of these well known risk behaviours. The survey did not collect data for injection drug behaviours, but did collect data on anal sex; however, the non-response rate for this question was too high to permit inclusion in the proportional hazards model. The survey also collected data about the number of partners in the past 6 months, a behaviour associated with sex work; this variable was included in the model but was not a significant predictor. Because institutionalised individuals whose HIV incidence might be higher or lower than found in SHIMS were not included, the SHIMS estimates of HIV incidence might under-represent or over-represent actual HIV incidence.

HIV incidence was defined solely by seroconversion and NAAT testing was not conducted at follow-up; this approach might have led to a potential underestimate of overall incidence of almost 9% (ie, 2·4% vs 2·6%), assuming a similar number of acute cases at enrolment and follow-up. The study was not powered to determine independent predictors of HIV seroconversion, and this might explain why some factors previously linked with HIV protection or acquisition were not significant predictors of HIV acquisition, such as age in women and circumcision in men. Finally, although this was an observational study, risk-reduction counselling and the provision of condoms during the enrolment phase might have altered participants’ risk behaviours and reduced the observed incidence rates below that of the rest of the national population.

SHIMS2, a cross-sectional, population-based HIV survey in Swaziland, has recently begun (Aug 30, 2016) and will assess the national estimate of HIV incidence after the 2009 programme expansion. As with other population-based HIV impact assessments,²⁴ the cross-sectional design of SHIMS2 will assess HIV incidence by taking advantage of advances in HIV incidence assays, namely the limiting antigen avidity assay combined with HIV RNA (viral load)⁷ rather than prospectively observed seroconversions. Results should be available by late 2017 and, when available, will show whether the epidemic in Swaziland has improved or is much the same. SHIMS2 and population-based HIV impact assessments will allow us to understand whether incidence remains the most suitable measure of a national HIV epidemic or whether other indicators, such as viral load suppression, will emerge as a more informative indicator.

Overall, this study is the first to report HIV incidence at a national level with the use of prospectively observed HIV seroconversion. The national HIV incidence in Swaziland of 2·4% is the highest national rate known. We found alarmingly high rates in men and women in specific age strata. In the context of Swaziland’s treatment and male circumcision coverage, these high incidence rates warrant further expansion of treatment initiation criteria and male circumcision scale-up in Swaziland. Our findings show the value of detailed incidence measures in characterising those at highest risk of new infections and emphasise the need in high-prevalence countries for evidence-based HIV programmes that include frequent HIV testing, support for disclosure of HIV status, expansion of HIV treatment, and consideration of pre-exposure prophylaxis as public health priorities.