Evaluating the BED Capture Enzyme Immunoassay to Estimate HIV Incidence Among Adults in Three Countries in Sub-Saharan Africa

Abstract

Serological assays for estimating HIV-1 incidence are prone to misclassification, limiting the accuracy of the incidence estimate. Adjustment factors have been developed and recommended for estimating assay-based HIV-1 incidence in cross-sectional settings. We evaluated the performance of the recommended adjustment factors for estimating incidence in national HIV surveys in three countries in sub-Saharan Africa. The BED-capture enzyme immunoassay was applied to stored blood specimens from (1) pregnant women aged 15–49 years attending antenatal clinics in Côte d'Ivoire (1998–2004), (2) adults aged 15–49 years participating in a demographic health survey in Kenya (2003), and (3) adults aged 15–49 years participating in a national household serosurvey in South Africa (2005). Assay-derived incidence estimates were corrected for misclassification using recommended adjustment factors and, where possible, were compared to mathematically modeled incidence in the same populations. Trends in HIV prevalence were compared to trends in assay-derived incidence to assess plausibility in the assay-derived trends. Unadjusted incidence was 3.8% [95% confidence interval (CI) 3.3–4.5] in Côte d'Ivoire, 3.5% (2.7–4.3) in Kenya, and 4.4% (CI 2.3–6.5]) in South Africa. Adjusted incidence was 2.9% (CI 2.1–3.7) in Côte d'Ivoire, 2.6% (CI 2.0–3.2) in Kenya, and 2.4% (CI 1.7–3.1) in South Africa. After adjustment, peak incidence shifted from older to younger age groups in Côte d'Ivoire and South Africa. Modeled HIV incidence was 1.0% (CI 1.02–1.08) in Kenya and 2.0% (CI 1.7–2.4) in South Africa. After applying the recommended adjustments factors, adjusted assay-derived estimates remained implausibly high in two of three populations evaluated. For more accurate measures of assay-derived population incidence, adjustment factors must be locally derived and validated. Until improved assays are available, caution should be applied in the use and interpretation of data from incidence assays.

Introduction

I n countries with generalized HIV epidemics, where >1% of population are HIV positive, HIV epidemic patterns are monitored using HIV prevalence surveillance data collected from anonymous seroprevalence surveys in public antenatal clinics and from general population-based serologic surveys.^1
–3 While HIV prevalence is a measure of the relative burden of disease in a population, prevalence data are poor measures of more acute changes in an HIV epidemic, including the emergence of new subepidemics or true declines in HIV transmission. Trends in HIV incidence or the measure of new HIV infections in a population over time can provide critical information needed for targeted HIV prevention planning and to measure the impact of HIV prevention programs. In addition, several large-scale global initiatives have been implemented to treat, provide care for, and prevent new HIV infections for millions of persons living in resource-limited countries. Given these goals, monitoring trends in HIV incidence is vital to assessing the impact of these initiatives.

The gold standard for estimating HIV incidence is through prospective cohort studies of HIV-negative persons followed over time. Because of the high complexity and costs to implement such studies, HIV incidence is alternatively estimated using mathematical models that infer adult incidence based on HIV prevalence data from cross-sectional serologic surveys and mortality assumptions.^4

–11

HIV-1 incidence can also be estimated using serologic assays that directly measure recent HIV-1 infection in cross-sectional specimens.^{12

–21} The key advantages of the laboratory-based method for measuring incidence are that laboratory assays do not require longitudinal follow-up of persons over time, the collection of serial blood specimens to estimate HIV-1 incidence, or assumptions about prevalence and mortality needed for mathematical models.

At the time this article was written, the BED-capture enzyme immunoassay (henceforth, for brevity, the BED assay) was the only commercially available HIV-1 incidence assay. The assay uses a multisubtype synthetic peptide to measure the increasing proportions of HIV-1-specific IgG to total IgG. This assay was calibrated and validated in various prospective cohorts comprised of differing HIV-1 subtypes, including the United Staes (subtype B), Thailand (subtypes B and E), The Netherlands (subtype B), Kenya (subtype A and C), Ethiopia (subtype C), Zimbabwe (subtype C), and India (subtype C) (B. Parekh, personal communication). The BED assay is currently used to estimate national HIV-1 incidence and monitor incidence trends in the population by applying the assay to cross-sectional specimens from national HIV surveillance surveys.^22

–27 Since 2005, the BED assay has been integrated into the U.S. HIV case-based surveillance system for HIV-1 incidence surveillance.²²

The BED assay has been found to overestimate HIV-1 incidence by misclassifying a proportion of long-term infections as recent infection on the assay.^28

–31 To correct for this misclassification, two post hoc adjustment factors derived from separate validations of the BED assay in longitudinal cohorts with follow-up of HIV-negative and HIV-positive persons were developed. These data originated from a cohort of postpartum women in Harare, Zimbabwe (N = 2749) and from the AIDSVAX B/B vaccine trial, which utilized data for men who have sex with men largely from the United States and the Netherlands (N = 120).^28,30 The adjustment factors in the two studies, which ranged from 0.052 in Zimbabwe to 0.056 in the United States and The Netherlands, differed by less than 5% from each other (p > 0.05 for the difference), and the adjusted BED estimates in both studies corresponded closely with the prospective follow-up estimates of HIV-1 incidence. These two studies represented the first application of BED adjustment factors to produce robust measures of assay-derived HIV-1 incidence in cross-sectional settings and due to these findings, the use of these “default” adjustment factors was subsequently recommended by the U.S. Office of the Global AIDS Coordinator for widespread application in settings where local BED adjustment factors were not available.³²

In this study, we evaluated the performance of the two recommended adjustment factors for estimating assay-derived HIV-1 incidence in three countries in sub-Saharan Africa without locally established BED adjustment factors. Unadjusted and adjusted BED estimates were compared to HIV-1 incidence estimates from mathematical models in the same population. In addition, trends in HIV-1 prevalence and adjusted HIV-1 incidence were compared across 10-year age bands and years to determine the plausibility of the adjusted HIV-1 incidence estimates.

Materials and Methods

Data sources

South Africa National Household Survey

South Africa conducted a national household survey of HIV-1 prevalence, behavior, and communication in 2005.³³ In brief, this survey was a probability-based sample of all persons aged ≥2 years in South Africa and residing in households. All participants provided verbal consent for data collection, including HIV testing using dried blood spots and a detailed questionnaire on demographics, behavior, and knowledge and attitudes. Caretakers of children aged <12 years provided proxy consent for the child. HIV test results were anonymous and linked to a behavioral questionnaire using a unique study number. A detailed description of the methods and results of this survey are described elsewhere.^26,33

The 2003 Kenya Demographic Health Survey

The 2003 Kenya Demographic Health Survey (KDHS), described in detail elsewhere,³⁴ was a probability-based population and health household survey of women aged 15–49 years and men aged 15–54 years in Kenya that has provided nationally representative estimates of demographic, behavioral, and social factors and HIV prevalence in the country. The HIV testing protocol for the 2003 KDHS was anonymous and conducted on dried blood spots identified by a unique study number that was linked to the survey questionnaire.

Côte d'Ivoire antenatal clinic surveillance

Temporal trends in HIV prevalence among pregnant women aged 15–49 years in Côte d'Ivoire have been monitored using data from antenatal (ANC) surveillance in selected clinic sites since 1997.³⁵ Unlinked and anonymous HIV testing is conducted on residual sera routinely collected for prenatal syphilis testing from pregnant women attending their first ANC visit. Blood specimens are linked by a unique study identification number to demographic and behavioral data collected during the prenatal visit and tested for HIV-1 antibodies using the national HIV testing algorithm.

Laboratory methods and HIV-1 incidence estimation

Using methods described elsewhere,¹⁷ the BED assay was applied to stored frozen HIV-1-positive dried blood spot samples from the 2003 KDHS, stored HIV-1-positive serum samples from 10 urban ANC sites in Côte d'Ivoire in 1998, 2000, 2002, and 2004, and fresh HIV-1-positive dried blood spots collected for the 2005 South African National Household Survey. In the ANC sample, annual BED HIV-1 incidence and 95% confidence intervals (CI) were calculated using the U.S. Centers for Disease Control and Prevention's consensus formula for calculating HIV-1 incidence and standard errors using cross-sectional samples.^15,36 HIV-1 incidence was expressed as the cumulative incidence or risk of incidence infections in a 12-month period, which was a close approximation to the instantaneous incidence rate. \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6} \begin{document} \begin{align*} I_ { 0 } = \frac { ( 365 / w ) N_ { inc } } { N_ { neg } + ( 365 / w ) N_ { inc } / 2 } \times 100 \end{align*} \end{document} \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6} \begin{document} \begin{align*} 95 \% \ { \rm CI} : \ I_{0} \pm 1.96 \times \big( I_0 / { \sqrt N}_{inc}\big) \end{align*} \end{document}

Calculation of unadjusted BED HIV-1 incidence for complex survey data with probability sampling (i.e., the 2003 KDHS and the South Africa 2005 National Household Survey) required individual sampling weights to be included in the calculation of BED HIV-1 incidence. The weighted formula for BED HIV-1 incidence and 95% CI was \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6} \begin{document} \begin{align*}{ \hat I } = \frac{\omega {\hat N } _ { inc }} {{ \hat N } _{neg} + \omega { \hat N } _ { inc } / 2 } \end{align*} \end{document} \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6} \begin{document} \begin{align*} 95 \% \ { \rm CI } : \ { \hat I } _ { 0 } \pm 1.96^ { \ast } Deft^{ \ast } \frac { { \hat I } _ { 0 } ( 2 - { \hat I } _ { 0 } ) } { 2 } \sqrt { \left( \frac { 1 } { { \hat N } _ { inc } } + \frac { 1 } { { \hat N } _ { neg } } \right) } \end{align*} \end{document}

where I ₀ = the unadjusted HIV-1 incidence estimate, Î₀ = weighted unadjusted HIV-1 incidence estimate, N_inc = number of persons testing recent on the assay, N_inc = weighted number of persons testing recent on the assay, N_neg = number of HIV-1-negative persons, N_neg = weighted number of HIV-1-negative persons, w=365 days/window period, and Deft = design effect.

The BED assay consensus mean window period of 155 days for all HIV-1 subtypes was used for Côte d'Ivoire and Kenya. A window period of 180 days for HIV-1 subtype C infection was used in South Africa. Default values for survey design effects, used for calculating sample sizes for ANC surveillance and national household surveys, were assumed for Côte d'Ivoire (Deft = 1.0), Kenya (Deft = 2.0), and South Africa (Deft = 2.0). To maximize comparability between countries, the analysis was restricted to persons aged 15–49 years in each survey.

Posttest adjustments

Adjusted HIV-1 incidence estimates (referred to below as I_I and I_II ) with 95% CI were calculated using methods described elsewhere.^28,30 In brief, I_I adjusts for the assay's short-term sensitivity [1 – α, where α is the proportion of false long-term cases among true recent infections occurring during the initial window period (w)], short-term specificity (1 – β, where β is the proportion of false recent cases among true long-term infections greater than w but less than 2w), and long-term specificity (1 – ɛ, where ɛ is the proportion of false-recent cases among true long-term infections ≥2w) on the BED.³⁰ The published values of α, β, and ɛ used for this analysis were 0.7682, 0.7231, and 0.0557, respectively. In contrast, I_II corrects for long-term specificity only and assumes α and β do not differ significantly and can therefore be assumed identical.²⁸ Because the I_I and I_II estimates never differed significantly in this analysis, for brevity, all adjusted estimates are presented for I_I only.

Mathematical and plausibility models for HIV-1 incidence

The Estimation and Projection Package (EPP) and Spectrum software package, developed by UNAIDS,^10,11,37 is a standardized modeling package that has been widely used to create national estimates of HIV-1 incidence and prevalence. The EPP and Spectrum model infer HIV-1 incidence in the adult population using HIV-1 prevalence data from ANC surveillance and an expected 11-year survival rate from HIV-1 infection to death. These data are calibrated using national estimates of HIV-1 prevalence from national population-based household surveys. Plausibility bounds (as opposed to 95% CI) for the EPP and Spectrum modeled HIV-1 incidence in the overall population under observation (e.g., males and females aged 15–49 years) are calculated using the methods developed by UNAIDS for the 2005 HIV/AIDS estimates.³⁶ The outputted number of new infections by age in the model can be used in combination with the estimated population size to estimate HIV-1 incidence by age groups. To date, plausibility bounds for HIV-1 incidence by age groups (e.g., 15–24 years) are not available in the EPP and Spectrum model.

In this evaluation mathematically modeled HIV-1 incidence was estimated using the 2009 version of the EPP and Spectrum model. A second mathematical model of HIV-1 incidence, developed by the Actuarial Society of South Africa (ASSA), was also applied to the South Africa data.^5,7,33 The assumptions, inputs, and outputs of the ASSA model are comparable to those of the EPP and Spectrum model though plausibility bounds were not available for the ASSA modeled estimates. The level of unadjusted and adjusted BED estimates for the population-based surveys in Kenya and South Africa were compared to mathematical models of HIV-1 incidence among adults aged 15–49 years. Additionally, “plausibility” models comparing BED HIV-1 incidence to corresponding trends in HIV-1 prevalence across time and 10-year age bands in the same population were assessed for plausibility with respect to expected epidemic trends, namely (1) changes in HIV-1 prevalence are expected to occur after changes in HIV-1 incidence, and (2) HIV-1 incidence is expected to be higher in younger age groups (that is, in adolescents and adults aged 15–24 years) and decline with increasing age.^38
–40

The influence of pregnancy on HIV incidence is not clearly established. Recent evidence suggests that acquisition of HIV infection may be significantly higher during pregnancy compared to the postpartum period; however, a prospective study of hormonal contraception and HIV-1 incidence found no difference in HIV acquisition by pregnancy and lactation status.^41,42 Due to uncertainty at this time about the effect of pregnancy on the risk of HIV-1 acquisition in the general population, we did not compare the level of assay-derived incidence in pregnant women with modeled incidence in the general adult population for Côte d'Ivoire, but rather focused the analysis on a comparison of trends in prevalence and incidence.

Results

South Africa 2005 National Household Survey

In previously reported data for the South Africa 2005 National Household Survey,⁴³ unadjusted BED HIV-1 incidence (I ₀) in adults aged 15–49 years was 4.4% (95% CI 2.3–6.5). The adjusted estimate (I_I ) for adults aged 15–49 years was 2.4% (95% CI 1.7–3.1),²⁶ 1.0% (95% CI 0.4–1.6) in males, and 3.8% (95% CI 2.6–5.0) in females. HIV-1 prevalence peaked in males aged 35–44 years at 20.9% and females aged 25–34 years at 29.5% (Fig. 1). In comparison, the I_I estimate was highest in males aged 25–34 years [2.6% (95% CI 0–5.4)] and females aged 15–24 years [4.6% (95% CI 3.2–6.0)] and 25–34 years [5.2% (95% CI 1.8–8.6)].

FIG. 1.

HIV prevalence, I ₀ unadjusted estimates, I_I adjusted estimates, and modeled estimates (I_MA and I_MS ) among males (A) and females (B) in the South Africa 2005 National Household Survey, by age group.

The EPP and Spectrum modeled incidence estimates (I_MS ) for South African males and females aged 15–49 years was 2.0% [plausibility bounds (PB) 1.7–2.4] overall, 1.7% (PB 1.4–2.0) in males, and 2.4% (PB 2.0–3.0) in females. In comparison, ASSA modeled incidence estimates (I_MA ) were 2.2% overall, 1.9% in males, and 2.5% in females. The modeled estimates (I_MS and I_MA ) were approximately two times lower than the unadjusted BED estimates (I ₀), but when compared to the adjusted BED estimates (I_I ), the modeled estimates fell within the range of the 95% CI of the adjusted BED estimate. For youth aged 15–24 years, both modeled estimates of HIV-1 incidence (I_MS and I_MA ) fell within the range of the 95% confidence intervals for the I_I estimate.

Kenya Demographic Health Survey

In the KDHS, 386 (91.5%) of 422 HIV-1-positive specimens from respondents aged 15–49 years were available for BED testing; 71 (18.4%) were classified as recent infections on the BED assay. Unadjusted BED HIV-1 incidence (I ₀) for survey participants aged 15–49 years was 3.5% (95% CI 2.7–4.3) and was higher in females [4.3% (95% CI 3.0–5.5)] than in males [2.6% (95% CI 1.6–3.6)]. The adjusted estimates (I_I ) were 2.6% (95% CI 2.0–3.2) overall, 3.1% (95% CI 2.2–4.0) in females, and 2.1% (95% CI 1.3–2.8) in males.

HIV-1 prevalence was highest in women aged 25–34 years (12.7%) and in males aged 35–49 years (7.8%) (Fig. 2). The I_I estimate was highest among women aged 25–34 years [3.4% (95% CI 1.8–5.0)] and men aged 35–49 years [4.2% (95% CI 2.1–6.3). EPP and Spectrum modeled incidence for Kenya in 2003 (I_MS ) was 1.0% (PB 1.02–1.08) for all adults aged 15–49 years, 0.9% (PB 0.86–0.92) for males aged 15–49 years, and 1.2% (PB 1.17–1.24) for females aged 15–49 years. The I_MS estimate was 3.5 times lower than the I ₀ estimate and approximately 2.5 times lower than the I_I estimate. For female youth aged 15–24 years, the I_MS estimate (1.5%) was nearly two times lower than the I_I estimate [2.9% (95% CI 1.3–4.5)], but the estimates for I_MS and I_I for male youth were similar.

FIG. 2.
HIV prevalence, I ₀ unadjusted estimates, I_I adjusted estimates, and modeled estimates (I_MS ) among males (A) and females (B) in the 2003 KDHS, by age group.

Côte d'Ivoire ANC Surveillance

From 1998 to 2004, 277 (9.2%) of 3016 pregnant women aged 15–49 years attending 10 urban ANC sites in Côte d'Ivoire tested positive for HIV-1 infection in 1998, followed by 257 (8.3%) of 3097 pregnant women in 2000, 176 (8.9%) of 1985 pregnant women in 2002, and 270 (7.7%) of 3498 pregnant women in 2004. Recent HIV-1 infections were detected by the BED assay in 55, 36, 35, and 53 HIV-positive specimens, respectively.

The unadjusted BED incidence estimate (I ₀) was 4.6% (95% CI 3.4–5.8) in 1998, 2.9% (95% CI 2.0–3.9) in 2000, 4.5% (95% CI 3.0–5.9) in 2002, and 3.8% (95% CI 3.3–4.5) in 2004 (Fig. 3). Over the same years the I_I estimate was 3.6% (95% CI 2.6–4.5), 1.9% (95% CI 1.3–2.5), 3.5% (95% CI 2.3–4.6), and 2.9% (95% CI 2.1–3.7), respectively.

FIG. 3.
HIV prevalence, I ₀ , unadjusted estimates, and I_I adjusted estimates among pregnant women attending urban antenatal clinics in Côte d'Ivoire from 1998 to 2004 (A) and by age group (B).

From 1998 to 2004, HIV-1 prevalence was highest among women aged 25–34 years at 11.2%. The I ₀ estimate was highest in women aged 15-24 years at 4.2% (95% CI 3.3 – 5.0) and 25–34 years at 4.4% (95% CI 3.3–5.4). The I_I estimate was highest in youth women 15–24 years at 3.4% (95% CI 2.7–.0) and lowest in the oldest age group (aged 34–49 years) at 1.5% (95% CI 0.6–2.4).

Discussion

BED and modeled estimates of HIV incidence

In South Africa, assay-based estimates of HIV-1 incidence, adjusted for test sensitivity and short- and long-term specificity, fell within a reasonable range of HIV-1 incidence estimated by two mathematical models. However, in Kenya the overall adjusted BED estimate was over two times higher than the mathematical modeled estimates of incidence (Table 1). Moreover, in Côte d'Ivoire, the adjusted incidence estimates over the survey period (2.9–3.6%) were not plausible for the level of prevalence observed during the same time period (7.7–9.2%).

Table 1.
Comparison of Adjusted BED HIV Incidence Estimates with Mathematical Models of HIV Incidence, South Africa National Household Survey 2005 and Kenya Demographic Health Survey 2003

South Africa 2005 Kenya 2003

Age group (years) Adjusted BED HIV incidence^a % per year (95% CI)^b EPP and Spectrum model % per year^c ASSA model BED to EPP and Spectrum ratio BED to ASSA ratio Adjusted BED HIV incidence^a % per year (95% CI)^b EPP and Spectrum model % per year^c BED to EPP and Spectrum ratio

Adults (15–49 years) 2.4 (1.7–3.1) 2.0 (1.7–2.4) 2.2 1.2 1.1 2.6 (2.0–3.2) 1.0 (1.02–1.08) 2.5

Males 1.0 (0.4–1.6) 1.7 (1.4–2.0) 1.9 0.6 0.5 2.1 (1.3–2.8) 0.9 (0.86–0.92) 2.4

Females 3.8 (2.6–5.0) 2.4 (2.0–3.0) 2.5 1.6 1.5 3.1 (2.2–4.0) 1.2 (1.17–1.24) 2.6

Youth (15–24 years) 2.2 (1.3–3.1) 2.5^d 2.9 0.9 0.8 1.8 (0.9–2.6) 1.1 1.6

Males 0.3 (0.03–0.6) 2.0 1.8 0.2 0.2 0.6 (0–1.3) 0.7 0.9

Females 4.6 (3.2–6.0) 3.0 4.1 1.5 1.1 2.9 (1.3–4.5) 1.5 1.9

a
For illustration, I_I is shown.

b
95% confidence intervals calculated for I_I.

c
Plausibility bounds calculated for EPP and Spectrum estimates.

d
Plausibility bounds for EPP and Spectrum estimates were not available for 10-year age bands.

For South Africa and Côte d'Ivoire, the adjusted estimates appeared to produce more reasonable age patterns in HIV-1 incidence compared to unadjusted estimates, with peak incidence shifted toward younger age groups (that is, adolescents and adults aged 15–24 years); however, these observations were based on only qualitative assessments.^38
–40 Notably, in Kenya, plausible levels of incidence were not observed even after adjustment. Rather, the adjusted incidence increased with increasing age, with highest incidence (adjusted) observed in the oldest age group (aged 35–49 years).

The difference between the adjusted estimates and the mathematically modeled estimates for Kenya may be explained, in some part, by the limitations of mathematical models in estimating HIV-1 incidence. In particular, though modeled estimates were used as the gold standard estimate for HIV-1 incidence in this analysis, at the time of writing the two models used for this analysis had not yet been validated against other measures of HIV-1 incidence in the same study population. Moreover, estimates in mathematical models of HIV-1 incidence are only as good as the quality of HIV prevalence and demographic data that are inputted into the model. In this regard the adjusted BED estimates and the modeled estimates were similar for South Africa, a country with fairly complete reporting of HIV-1 prevalence and demographic data. In contrast, at least some of the difference observed between the BED and modeled estimates in Kenya may be due to less complete input data for the model. Nonetheless, data have suggested that in relatively stable and mature epidemics, such as in the three countries under analysis, HIV-1 incidence in the general population should not be much higher than 10% of HIV prevalence,⁴⁴ indicating that the adjusted BED estimates for Kenya and Côte d'Ivoire likely overestimated the true incidence in these settings.

Other sources of error in the BED estimates

There were several other potential sources of error that may have contributed to the overestimation of the adjusted incidence estimates in Kenya. First, timely and proper storage of specimens prior to BED testing is critical to the performance and accuracy of the BED assay. Antibody optical density values for enzyme immunoassays have been found to progressively decline under harsh conditions.⁴⁵ Under this assumption, false recent classifications may be possible on antibody-based incidence assays, such as the BED assay (B. Parekh, personal communication). Indicators of improper storage conditions, such as mold on dried blood spot specimens (observed in a sample of the 2003 Kenya DHS specimens), can negatively impact the stability of the antibody and the accuracy of the test. Second, inappropriate values may have been used to adjust for long-term specificity of the assay, defined as the proportion (ɛ) of treatment-naive patients who continue to classify as recent by the BED assay even after infection for more than a year.^28,30,46

Recent unpublished estimates of ɛ, which relate to East Africa, do suggest that ɛ may in fact be substantially higher in East Africa than the default ɛ used in this analysis. Two separate studies in Uganda, combined for reasonable sample sizes, have estimated that ɛ is approximately 15.0% in treatment-naive patients with known long-term HIV-1 infections.^37,47 Application of the “local” ɛ to the Kenya data would have lowered the overall BED HIV-1 estimate by 4-fold, placing the adjusted estimate within the plausibility bounds of the modeled estimate for Kenya. Karita and colleagues also reported substantially higher levels of ɛ in East Africa²⁹; however, because these measurements were based on very small sample sizes (N = 14 and 23) this local ɛ could not be used reliably to adjust the BED incidence in this analysis. Additionally, recent data from KwaZulu Natal, South Africa reported a lower ɛ of 1.7%, providing evidence that ɛ may vary by region within a country.⁴⁸ Application of an ɛ of 1.7% to the South Africa data would have increased the adjusted assay-based incidence estimate among adults aged 15–49 years from 2.4% to 3.7%, a level nearly two times higher than the modeled estimates of incidence.

The need to measure ɛ under a variety of conditions

The results of this analysis strongly suggest that ɛ may vary by site and therefore is applicable only in settings from which the local ɛ was quantified. Given this, local validations of ɛ, optimally using large samples (N ≥ 1000) of known long-term HIV-1 infections, infected for at least 12 months, are required to estimate ɛ in populations under a variety of conditions before an incidence assay can be widely applied in surveillance settings in that population. The validation population should be representative of the populations to which the incidence assay will be applied. If ɛ will be used to calibrate national level HIV-1 incidence, for example, the local ɛ should be derived from a nationally representative sample. At a minimum data on gender, age, geography, approximate duration since first HIV-1 positive test result based on previous HIV testing history, opportunistic and coinfections, CD4⁺ T cell count, and antiretroviral therapy (ART) regimen must be collected to establish the variation in ɛ with these variables.⁴⁹

Effects of low CD4⁺ T cell counts and treatment regimen on estimates of ɛ

Evidence suggests that ART use and low CD4⁺ T cell counts may be particularly important contributors to errors in BED estimates. A high proportion of patients (>20%) on ART for greater than 12 months will misclassify as recent infection on the BED assay due to a drop in viral levels and a corresponding decrease in antibody response. This error does not appear to be constant but varies between populations and increases by time on ART.^37,47,50 Similarly, persons with very low CD4 cell counts may experience a drop in antibody response as the host immune response becomes more compromised by disease and will subsequently falsely classify a proportion of these as recent infection on the BED assay.⁵¹

These results have two important consequences. First, because the error from ART and low CD4 cell counts may be considerable, the sample in which the BED is planned to be applied for incidence estimation should measure current ART use and CD4 cell counts to correctly identify and exclude patients on ART and with low CD4 cell counts from the incidence analysis. Second, due to the inconsistency of error from ART and low CD4 cell count <50 cells/mm³ and to ensure comparability with the population in which the incidence assay will be applied, when estimating ɛ it is recommended that persons on ART and with CD4⁺ T cell counts <50 cells/mm³ are excluded from the validation sample. Failure to comply with these requirements will undoubtedly result in overestimation of HIV-1 incidence.

Increased complexity: suggested solutions

It is recognized that these requirements introduce a substantial level of complexity to the wide successful use of the BED assay in the field, particularly when integrated into national HIV surveillance systems. Currently unlinked anonymous testing (UAT) in ANC surveillance sites can utilize only routinely collected clinic data for the purposes of HIV surveillance, most of which do not include ART status or CD4 T cell count measurements on HIV-1-positive patients. Similarly, among all national population-based surveys with HIV testing that have been completed to date, only very few have collected data on ART use or measured CD4 T cell counts. Additionally, HIV testing in national population-based surveys is typically conducted on dried blood spots, thereby precluding any possibility for CD4 T cell count testing. As international efforts move toward the use of program data for estimating ANC HIV-1 prevalence, the World Health Organization (WHO) now recommends that all ANC sites with prevention of mother-to-child transmission (PMTCT) programs begin to record routinely collected PMTCT variables on the ANC surveillance form (such as current ART use) to monitor the PMTCT program and evaluate its future utility in estimating HIV prevalence in the ANC population.⁵² Furthermore, population-based surveys that collect venous blood specimens have the ability to obtain information on ART status, self-reported HIV status, in combination with previous HIV testing history to identify known long-term HIV infection, and test for CD4⁺ T cell counts among HIV positives, provided there are appropriate financial, logistical, and laboratory support for doing so.⁵³

If ART data cannot be obtained in the survey setting ART use can alternatively be measured by directly testing HIV-1-positive samples for the presence of ART. This option is a feasible option if well-trained regional laboratories can be used and resources are available. Additionally, costs are assumed to be minimal (approximately $20 U.S. dollars per sample tested) if testing is conducted only on those specimens testing as a recent infection on the incidence assay.

Limitations of the study

This evaluation was subject to the following limitations. Due to the rarity of recent infection, large sample sizes were needed to detect statistically significant differences between unadjusted, adjusted, and modeled estimates of HIV-1 incidence. Whereas the South Africa 2005 National Household Survey was powered for incidence analysis, the surveys in Côte d'Ivoire and Kenya were post hoc evaluations using stored specimens from completed surveillance studies powered for national-level prevalence analysis. Additionally, appropriate statistical methods for detecting differences between independent HIV-1 incidence measures are currently under development; therefore differences between BED and modeled-based incidence could only be qualitatively assessed. Prior to considering incidence assay testing, countries should work with statisticians to consider sample sizes requirements needed for incidence analysis. If the minimum required sample sizes cannot be reached in the survey setting, incidence testing activities should be reconsidered. Standard errors were not available for the ASSA model and age-specific incidence in the EPP and Spectrum model. Therefore, a reasonable degree of uncertainty exists for these estimates. We were unable to exclude persons on ART from the three surveys in this analysis as ART use was not directly measured. It is possible, therefore, that persons on ART or with very low CD4⁺ T cell counts were included in the analysis, thereby contributing to overestimation of BED HIV-1 incidence, even after adjustment. The size of this error is assumed to be low given that these surveys were implemented during the early stages of ART expansion in these countries. According to reports from country Ministry of Health programs and the WHO, approximately 0.4% of HIV-infected persons in Côte d'Ivoire and 1% of HIV-infected persons in Kenya and South Africa may have been on ART during the survey periods.^54
–56 If we assume the same distribution of ART uptake in our three samples and further assume that an estimated 20% of HIV-infected persons on ART will misclassify as recent on the assay,³⁷ this could have resulted in one misclassified case in Kenya and Cote d'Ivoire's data and up to three misclassified cases in South Africa's data due to ART use. Though the effect of ART is presumed to be low in these samples, given the very absence of availability of treatment, however, the proportion of persons with very low CD4 counts may have been correspondingly higher.

The way forward

Although it is anticipated that all antibody-based serologic assays that distinguish recent from long-term HIV-1 infection will continue to have some degree of misclassification, improved incidence assays that may overcome some of these limitations are currently under development.^{20,57

–63} This includes examination of a serial testing algorithm, using a highly sensitive assay as the screening assay and a more specific assay based on a different biological principle as a confirmatory assay.^14,47 As new assays are developed and existing assays improved, appropriate caution should be applied in countries that currently use or plan to use the BED assay for HIV-1 incidence estimation. First, for countries to appropriately use these assays for national HIV-1 incidence estimates, ART use and CD4 cell counts in HIV-1-positive persons should be measured in national surveillance settings. Second, prior to estimating national-level HIV-1 incidence in a population, countries should plan to implement local incidence assay validation activities in country to measure ɛ in the populations in which the assay will be applied for estimating HIV-1 incidence. Third, HIV incidence surveillance settings should meet all requirements for proper specimen storage and appropriate sample sizes for incidence analysis. If the requirements listed above are not met, it is advised that the incidence assay not be applied. Finally, assay-based HIV-1 incidence should be compared to other measures of HIV-1 incidence in the same population to evaluate the assay's ability to accurately measure HIV-1 incidence. Similarly, mathematical models of HIV-1 incidence should be validated against independent measures of HIV-1 incidence, including directly observed HIV-1 incidence in longitudinal cohorts. To date, one such model has been validated in this manner and has been used to produce national estimates of HIV incidence in countries with at least one nationally representative HIV survey.^64,65 A validated incidence model such as this can be interpreted in combination with assay-based incidence data to help assess the plausibility of the assay-derived estimates.

	South Africa 2005	Kenya 2003
Adults (15–49 years)	2.4 (1.7–3.1)	2.0 (1.7–2.4)	2.2	1.2	1.1	2.6 (2.0–3.2)	1.0 (1.02–1.08)	2.5
Males	1.0 (0.4–1.6)	1.7 (1.4–2.0)	1.9	0.6	0.5	2.1 (1.3–2.8)	0.9 (0.86–0.92)	2.4
Females	3.8 (2.6–5.0)	2.4 (2.0–3.0)	2.5	1.6	1.5	3.1 (2.2–4.0)	1.2 (1.17–1.24)	2.6
Youth (15–24 years)	2.2 (1.3–3.1)	2.5^d	2.9	0.9	0.8	1.8 (0.9–2.6)	1.1	1.6
Males	0.3 (0.03–0.6)	2.0	1.8	0.2	0.2	0.6 (0–1.3)	0.7	0.9
Females	4.6 (3.2–6.0)	3.0	4.1	1.5	1.1	2.9 (1.3–4.5)	1.5	1.9

Footnotes

Acknowledgments

We are grateful to John Stover and Robert Dorrington for their inputs on the mathematical models used for this analysis and to Meade Morgan for critical comments on data analysis for this manuscript. This research was sponsored in part by the U.S. President's Emergency Plan for AIDS Relief, Office of the Global AIDS Coordinator, the Nelson Mandela Foundation, the Swiss Agency for Development and Cooperation, and the Human Sciences Research Council. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the U.S. Centers for Disease Control and Prevention.

Author Disclosure Statement

No competing financial interests exist.

References

1.
Ghys PD , Kufa E , George MVC . Measuring trends in prevalence and incidence of HIV infection in countries with generalised epidemics. Sex Transm Infect, 2006; 82,Suppl 1:i52–56.

2.
UNAIDS/WHO. Guidelines for conducting HIV sentinel serosurveys among pregnant women and other groups. Edited by UNAIDS/WHO working group on Global HIV/AIDS and STI Surveillance: Geneva, 2003.

3.
UNAIDS/WHO. Guidelines for measuring national HIV prevalence in population-based surveys. Edited by UNAIDS/WHO working group on Global HIV/AIDS and STI Surveillance: Switzerland, 2003Geneva, 2006.

4.
Cleghorn FR , Jack N , Murphy JR , Edwards J , Mahabir B , Paul R et al. Direct and indirect estimates of HIV-1 incidence in a high-prevalence population. Am J Epidemiol, 1998; 147:834–839.

5.
Dorrington RE . The ASSA 2000 suite of models. Modeling the Demographic Impact of HIV/AIDS in South Africa and the Likely Impact of Interventions. Somerset West: South Africa, 2000.

6.
Gouws E , Williams BG , Sheppard HW , Enge B , Karim SA . High incidence of HIV-1 in South Africa using a standardized algorithm for recent HIV seroconversion. J Acquir Immune Defic Syndr, 2002; 29:531–535.

7.
Johnson LF , Dorrington RE . Modeling the demographic impact of HIV/AIDS in South Africa and the likely impact of interventions. Demograph Res, 2006; 14:541–574.

8.
Saidel T , Sokal D , Rice J , Buzingo T , Hassig S . Validation of a method to estimate age-specific human immunodeficiency virus (HIV) incidence rates in developing countries using population-based seroprevalence data. Am J Epidemiol, 1996; 144:214–223.

9.
Sakarovitch C , Alioum A , Ekouevi DK , Msellati P , Leroy V , Dabis F . Estimating incidence of HIV infection in childbearing age African women using serial prevalence data from antenatal clinics. Stat Med, 2007; 26:320–335.

10.
Stover J . Projecting the demographic consequences of adult HIV prevalence trends: The Spectrum Projection Package. Sex Transm Infect, 2004; 80,Suppl 1:i14–18.

11.
Stover J , Walker N , Grassly NC , Marston M . Projecting the demographic impact of AIDS and the number of people in need of treatment: Updates to the Spectrum projection package. Sex Transm Infect, 2006; 82,Suppl 3:iii45–50.

12.
Barin F , Meyer L , Lancar R , Deveau C , Gharib M , Laporte A et al. Development and validation of an immunoassay for identification of recent human immunodeficiency virus type 1 infections and its use on dried serum spots. J Clin Microbiol, 2005; 43:4441–4447.

13.
Chawla A , Murphy G , Donnelly C , Booth CL , Johnson M , Parry JV et al. Human immunodeficiency virus (HIV) antibody avidity testing to identify recent infection in newly diagnosed HIV type 1 (HIV-1)-seropositive persons infected with diverse HIV-1 subtypes. J Clin Microbiol, 2007; 45:415–420.

14.
Constantine NT , Sill AM , Jack N , Kreisel K , Edwards J , Cafarella T et al. Improved classification of recent HIV-1 infection by employing a two-stage sensitive/less-sensitive test strategy. J Acquir Immune Defic Syndr, 2003; 32:94–103.

15.
Janssen RS , Satten GA , Stramer SL , Rawal BD , O'Brien TR , Weiblen BJ et al. New testing strategy to detect early HIV-1 infection for use in incidence estimates and for clinical and prevention purposes. JAMA, 1998; 280:42–48.

16.
Li H , Ketema F , Sill AM , Kreisel KM , Cleghorn FR , Constantine NT . A simple and inexpensive particle agglutination test to distinguish recent from established HIV-1 infection. Int J Infect Dis, 2007; 11:459–465.

17.
Parekh BS , Kennedy MS , Dobbs T , Pau CP , Byers R , Green T et al. Quantitative detection of increasing HIV type 1 antibodies after seroconversion: A simple assay for detecting recent HIV infection and estimating incidence. AIDS Res Hum Retroviruses, 2002; 18:295–307.

18.
Schupbach J , Gebhardt MD , Tomasik Z , Niederhauser C , Yerly S , Burgisser P et al. Assessment of recent HIV-1 infection by a line immunoassay for HIV-1/2 confirmation. PLoS Med, 2007; 4:e343.

19.
Soroka SD , Granade TC , Candal D , Parekh BS . Modification of rapid human immunodeficiency virus (HIV) antibody assay protocols for detecting recent HIV seroconversion. Clin Diagn Lab Immunol, 2005; 12:918–921.

20.
Suligoi B , Massi M , Galli C , Sciandra M , Di Sora F , Pezzotti P et al. Identifying recent HIV infections using the avidity index and an automated enzyme immunoassay. J Acquir Immune Defic Syndr, 2003; 32:424–428.

21.
Wilson KM , Johnson EI , Croom HA , Richards KM , Doughty L , Cunningham PH et al. Incidence immunoassay for distinguishing recent from established HIV-1 infection in therapy-naive populations. AIDS, 2004; 18:2253–2259.

22.
Hall HI , Song R , Rhodes P , Prejean J , An Q , Lee LM et al. Estimation of HIV incidence in the United States. JAMA, 2008; 300:520–529.

23.
Jiang Y , Wang M , Ni M , Duan S , Wang Y , Feng J et al. HIV-1 incidence estimates using IgG-capture BED-enzyme immunoassay from surveillance sites of injection drug users in three cities of China. AIDS, 2007; 21,Suppl 8:S47–51.

24.
Saphonn V , Parekh BS , Dobbs T , Mean C , Bun LH , Ly SP et al. Trends of HIV-1 seroincidence among HIV-1 sentinel surveillance groups in Cambodia, 1999–2002. J Acquir Immune Defic Syndr, 2005; 39:587–592.

25.
Wolday D , Meles H , Hailu E , Messele T , Mengistu Y , Fekadu M et al. Temporal trends in the incidence of HIV infection in antenatal clinic attendees in Addis Ababa, Ethiopia, 1995–2003. J Intern Med, 2007; 261:132–137.

26.
Rehle T , Shisana O , Pillay V , Zuma K , Puren A , Parker W . National HIV incidence measures—new insights into the South African epidemic. S Afr Med J, 2007; 97:194–199.

27.
Mermin J , Musinguzi J , Opio A , Kirungi W , Ekwaru JP , Hladik W et al. Risk factors for recent HIV infection in Uganda. JAMA, 2008; 300:540–549.

28.
Hargrove JW , Humphrey JH , Mutasa K , Parekh BS , McDougal JS , Ntozini R et al. Improved HIV-1 incidence estimates using the BED capture enzyme immunoassay. AIDS, 2008; 22:511–518.

29.
Karita E , Price M , Hunter E , Chomba E , Allen S , Fei L et al. Investigating the utility of the HIV-1 BED capture enzyme immunoassay using cross-sectional and longitudinal seroconverter specimens from Africa. AIDS, 2007; 21:403–408.

30.
McDougal JS , Parekh BS , Peterson ML , Branson BM , Dobbs T , Ackers M , Gurwith M . Comparison of HIV type 1 incidence observed during longitudinal follow-up with incidence estimated by cross-sectional analysis using the BED capture enzyme immunoassay. AIDS Res Hum Retroviruses, 2006; 22:945–952.

31.
UNAIDS: Statement on the use of the BED-assay for the estimation of HIV-1 incidence for surveillance or epidemic monitoring. Meeting of the UNAIDS/WHO Reference Group on Estimates, Modelling and Projections. Athens, Greece, 2005.

32.
Office of Global AIDS Coordinator (OGAC). Interim Recommendations for the Use of the BED Capture Enzyme Immunoassay for Incidence Estimation and Surveillance— Statement from the Surveillance and Survey and the Laboratory Working Groups to the Office of the Global AIDS Coordinator. OGAC: Washington, DC, 2006.

33.
Rehle T , Dorrington R , Shisana O , Pilay V , Puren A . National HIV incidence estimates: Direct measures compared with mathematical modeling. South African AIDS Conference. Durban: South Africa, 2007.

34.
Central Bureau of Statistics (CBS). Kenya Ministry of Health, ORC Macro: Kenya Demographic Health Survey. ORC Macro: Calverton, MD, 2003.

35.
Ministry of Health. Côte d'Ivoire: Surveillance of HIV/AIDS and syphilis in Côte d'Ivoire. Abidjan: Côte d'Ivoire, 2004.

36.
Morgan M , Walker N , Gouws E , Stanecki KA , Stover J . Improved plausibility bounds about the 2005 HIV and AIDS estimates. Sexually Trans Infect, 2006; 82:iii71–77.

37.
Hladik W , Olara D , Were W , Mermin JRD . The effect of antiretroviral treatment on the specificity of the serological ED HIV-1 incidence assay. HIV Implementers Meeting, June 16–19, 2007. Kigali: Rwanda, 2007[Abstract 988].

38.
Kumwenda N , Hoffman I , Chirenje M et al. HIV incidence among women of reproductive age in Malawi and Zimbabwe. Sexually Trans Dis, 2006; 33:646–651.

39.
Taha T , Dallabetta G , Hoover DR et al. Trends of HIV-1 and sexually transmitted disease among pregnant and postpartum women in urban Malawi. AIDS, 1998; 12:197–203.

40.
Zaba B , Boerma T , White R . Monitoring the AIDS epidemic using HIV prevalence data among young women attending antenatal clinics: prospects and problems. AIDS, 2000; 14:1633–1645.

41.
Gray R , Li X , Kigozi G et al. Increased risk of incidence HIV during pregnancy in Rakai, Uganda: A prospective study. Lancet, 2005; 366:1182–1188.

42.
Morrison CS , Wang J , Van Der Pol B , Padian N , Salata RA , Richardson BA . Pregnancy and the risk of HIV-1 acquisition among women in Uganda and Zimbabwe. AIDS, 2007; 21:1027–1034.

43.
Shisana O , Rehle T , Simbayi LC et al. South African National HIV Prevalence, HIV Incidence, Behavior and Communication Survey. Human Sciences Research Council: Cape Town, 2005.

44.
Garnett GP , Gregson S , Stanecki KA et al. Criteria for detecting and understanding changes in the risk of HIV infection at a national level in generalized epidemics. Sexually Trans Infect, 2006; 82:i48–51.

45.
Behets F , Kashamuka M , Pappaioanou M et al. Stability of human immunodeficiency virus type 1 antibodies in whole blood dried on filter paper and stored under various tropical conditions in Kinshasa, Zaire. J Clin Microbiol, 1992; 30:1179–1182.

46.
Welte TMT . Methodological progress in estimating incidence with a test for recent infection. The 2nd Global HIV/AIDS Surveillance Meeting. Bangkok: Thailand, 2009.

47.
Laeyendecker O . BED+ avidity testing algorithm for incidence estimates in Uganda (PowerPoint presentation) The 2nd Global HIV/AIDS Surveillance Meeting. Bangkok: Thailand, 2009.

48.
Barnighausen T , Wallrauch C , Welte A , McWalter TA , Mbizana N , Viljoen J et al. HIV incidence in rural South Africa: Comparison of estimates from longitudinal surveillance and cross-sectional cBED assay testing. PLoS One, 2008; 3:e3640.

49.
Hallett TB , Ghys P , Barnighausen T , Yan P , Garnett GP . Errors in 'BED'-derived estimates of HIV incidence will vary by place, time and age. PLoS One, 2009; 4:e5720.

50.
Marinda ET , Hargrove J , Preiser W , Slabbert H , van Zyl G , Levin J et al. Significantly diminished long-term specificity of the BED capture enzyme immunoassay among patients with HIV-1 with very low CD4 counts and those on antiretroviral therapy. J Acquir Immune Defic Syndr, 2010; 53:496–499.

51.
Xiao Y , Jiang Y , Feng J , Xu W , Wang M , Funkhouser E et al. Seroincidence of recent human immunodeficiency virus type 1 infections in China. Clin Vaccine Immunol, 2007; 14:1384–1386.

52.
WHO. Final Report of the Fifth Meeting of the WHO/AFRO Technical Network on HIV/AIDS and STI Surveillance, Harare, 2006. WHO Africa Regional Office: Brazzaville, Congo.

53.
NASCOP. 2007 Kenya AIDS Indicator Survey: Final Report. National AIDS and STI Control Programme. Nairobi: Kenya, 2009.

54.
UNAIDS/WHO. Epidemiological Fact Sheet on HIV/AIDS and Sexually Transmitted Infections, Cote d'Ivoire, 2004. Edited by UNAIDS/WHO working group on Global HIV/AIDS and STI Surveillance: Geneva, 2004.

55.
UNAIDS/WHO. Epidemiological Fact Sheets on HIV/AIDS And Sexaully Transmitted Infections, Kenya, 2004 Update. Edited by UNAIDS/WHO working group on Global HIV/AIDS and STI Surveillance: Geneva, 2004.

56.
Rehle T , Shisana O , Simbayi L , Pillay-Van Wyk V , Jooste S , Zuma K , Puren A , Parker W , Mbelle N , Van Zyi J . Trends in HIV prevalence, incidence, and risk behaviors among children, youth, and adults in South Africa, 2002 to 2008. 17th Conference on Retroviruses and Opportunistic Infections, San Francisco, CA, 2010.

57.
Kim J . Is it antibody titre or avidity? (Maybe it's both) STARHS Workshop in association with the 16th International Conference on AIDS;, August 13–18, 2006. Toronto: Canada, 2006.

58.
Loschen S , Batzing-Feigenbaum J , Poggensee G , Cordes C , Hintsche B , Rausch M et al. Comparison of the human immunodeficiency virus (HIV) type 1-specific immunoglobulin G capture enzyme-linked immunosorbent assay and the avidity index method for identification of recent HIV infections. J Clin Microbiol, 2008; 46:341–345.

59.
Martro E , Suligoi B , Gonzalez V , Bossi V , Esteve A , Mei J , Ausina V . Comparison of the avidity index method and the serologic testing algorithm for recent human immunodeficiency virus (HIV) seroconversion, two methods using a single serum sample for identification of recent HIV infections. J Clin Microbiol, 2005; 43:6197–6199.

60.
Murphy G , Charlett A , Dean GL , Fisher M , McElborough D , Parry JV . Measuring antibody avidity improves the utility of STARHS in identifying recent HIV infection. 15th International Conference on AIDS, Bangkok, Thailand, 2004.

61.
Sakarovitch C , Rouet F , Murphy G , Minga AK , Alioum A , Dabis F et al. Do tests devised to detect recent HIV-1 infection provide reliable estimates of incidence in Africa? J Acquir Immune Defic Syndr, 2007; 45:115–122.

62.
Suligoi B , Butto S , Galli C , Bernasconi D , Salata RA , Tavoschi L et al. Detection of recent HIV infections in African individuals infected by HIV-1 non-B subtypes using HIV antibody avidity. J Clin Virol, 2008; 41:288–292.

63.
Wei XLX , Dobbs T , Kuehl D , Nkengasong JN , Hu DJ , Parekh BS . Development of two avidity-based assays to detect recent HIV Type 1 seroconversion using a multisubtype gp41 recombinant protein. AIDS Res Hum Retroviruses, 2010; 26,1:61–71.

64.
Hallett TB , Mishra V , Ghys PD , Gregson S , Boerma T . Estimates of HIV incidence from household-based prevalence surveys. AIDS, 2010; 24:147–152.

65.
Hallett TB , Zaba B , Todd J , Lopman B , Mwita W , Biraro S et al. Estimating incidence from prevalence in generalised HIV epidemics: Methods and validation. PLoS Med, 2008; 5:e80.

	South Africa 2005					Kenya 2003
Age group (years)	Adjusted BED HIV incidence^a % per year (95% CI)^b	EPP and Spectrum model % per year^c	ASSA model	BED to EPP and Spectrum ratio	BED to ASSA ratio	Adjusted BED HIV incidence^a % per year (95% CI)^b	EPP and Spectrum model % per year^c	BED to EPP and Spectrum ratio
Adults (15–49 years)	2.4 (1.7–3.1)	2.0 (1.7–2.4)	2.2	1.2	1.1	2.6 (2.0–3.2)	1.0 (1.02–1.08)	2.5
Males	1.0 (0.4–1.6)	1.7 (1.4–2.0)	1.9	0.6	0.5	2.1 (1.3–2.8)	0.9 (0.86–0.92)	2.4
Females	3.8 (2.6–5.0)	2.4 (2.0–3.0)	2.5	1.6	1.5	3.1 (2.2–4.0)	1.2 (1.17–1.24)	2.6
Youth (15–24 years)	2.2 (1.3–3.1)	2.5^d	2.9	0.9	0.8	1.8 (0.9–2.6)	1.1	1.6
Males	0.3 (0.03–0.6)	2.0	1.8	0.2	0.2	0.6 (0–1.3)	0.7	0.9
Females	4.6 (3.2–6.0)	3.0	4.1	1.5	1.1	2.9 (1.3–4.5)	1.5	1.9

Evaluating the BED Capture Enzyme Immunoassay to Estimate HIV Incidence Among Adults in Three Countries in Sub-Saharan Africa

Abstract

Introduction

Materials and Methods

Data sources

South Africa National Household Survey

The 2003 Kenya Demographic Health Survey

Côte d'Ivoire antenatal clinic surveillance

Laboratory methods and HIV-1 incidence estimation

Posttest adjustments

Mathematical and plausibility models for HIV-1 incidence

Results

South Africa 2005 National Household Survey

Kenya Demographic Health Survey

Côte d'Ivoire ANC Surveillance

Discussion

BED and modeled estimates of HIV incidence

Other sources of error in the BED estimates

The need to measure ɛ under a variety of conditions

Effects of low CD4+ T cell counts and treatment regimen on estimates of ɛ

Increased complexity: suggested solutions

Limitations of the study

The way forward

Footnotes

Acknowledgments

Author Disclosure Statement

References

Effects of low CD4⁺ T cell counts and treatment regimen on estimates of ɛ