The development of an inexpensive, simple and widely accessible test for diagnosing human immunodeficiency virus (HIV) infection in infants could significantly reduce paediatric acquired immunodeficiency syndrome (AIDS) deaths in sub-Saharan Africa.
The numbers of acquired immunodeficiency syndrome (AIDS)-related deaths among infants in developing countries are exceptionally high, largely because human immuno-deficiency virus (HIV) infection remains undiagnosed in many cases. Current HIV testing methods are either impractical for developing-country settings or inaccurate for use in infants. There is an urgent need to develop and deploy a new, easy-to-use HIV test, which could transform the management of paediatric HIV/AIDS in developing countries and avert millions of infant deaths.
In 2005, there were an estimated 2.3 million children living with HIV/AIDS, 2 million of whom resided in sub-Saharan Africa, and 570,000 AIDS-related deaths occurred in children aged <15 years1,2,3. HIV disease progresses rapidly in children, and AIDS-related mortality among infants is exceptionally high. Roughly 33% of untreated HIV-infected infants in the developing world die during the first year of life, and >50% die by the age of 2 years4.
The clinical diagnosis and management of HIV disease in infants in resource-limited settings is severely hindered by the lack of inexpensive, simple and widely accessible diagnostic tests. This is further complicated by the fact that infants retain maternal antibodies for up to 12–18 months after birth. The enzyme-linked immunosorbent assay (ELISA) immunoglobulin G (IgG) antibody HIV tests currently used for adult diagnosis in resource-limited settings do not distinguish between maternal and infant antibodies. Hence, these relatively low-cost tests cannot provide a definitive diagnosis of HIV infection in infants aged <12 months5. Furthermore, assays to detect the virus or its components, such as the qualitative HIV DNA and RNA polymerase chain reaction (PCR) or the more recent ultrasensitive p24 immunoassays6,7,8,9,10, are too expensive and technically sophisticated to be optimally useful in resource-limited settings11.
Consequently, in some parts of sub-Saharan Africa, practical diagnosis is based on the Integrated Management of Childhood Illness (IMCI) algorithm developed by the World Health Organization (WHO) and the United Nations Children's Fund (UNICEF), which allows trained providers at the primary care level to recognize symptoms consistent with HIV infection in infants. The IMCI protocol has been adopted as a worldwide strategy for improving paediatric care in resource-limited settings since its introduction in 1995; however, the guidelines only recently incorporated a specific algorithm for the identification of children with HIV12. The IMCI/HIV algorithm trains providers to identify symptomatic HIV infection using screening questions and by looking for independent predictors, such as enlarged lymph nodes, oral thrush and parotid swelling12. HIV-infected infants without symptoms are generally not identified for HIV-specific care.
Antiretroviral therapy (ART) is the current state-of-the-art treatment for HIV disease in children. However, in resource-limited areas, the availability and coverage of ART is limited: current treatment guidelines restrict its use to children diagnosed with the virus who are also classified as HIV clinical stage III or IV13. Given the limited availability of ART, health-care providers focus instead on the prevention and early treatment of opportunistic infections such as pneumonia caused by Pneumocystis jiroveci (formerly known as Pneumocystis carinii pneumonia or PCP), which is a common and life-threatening AIDS-defining illness in infants4,14. Co-trimoxazole (CTX) prophylaxis is currently recommended for all infants born to HIV-infected mothers and those diagnosed with HIV infection15,16,17,18.
The lack of HIV diagnostic tests that can be used in resource-limited areas is a major barrier to the provision of effective care and treatment of HIV-infected children. Consequently, there is an urgent need for alternative strategies and biomarkers for early diagnosis of HIV infection in infants. In light of the enormous burden of HIV/AIDS on children in developing countries, as well as the need for better diagnostic tests, the HIV working group of the Bill & Melinda Gates Foundation Global Health Diagnostics Forum explored the potential health benefits associated with a hypothetical new test for the early diagnosis of HIV infection in infants aged <12 months. We focused on sub-Saharan Africa, where ∼90% of all children with HIV reside19. The analysis considers test-performance characteristics and access requirements associated with a new diagnostic, as well as the availability of ART. In addition to the problem of HIV diagnosis in infants, we recognize the need for tests that can better guide and monitor ART treatment in HIV-infected adults. However, as point-of-care CD4 assays for adults have already been clearly identified as a global priority, and funding has been made available for their development, we elected to focus solely on the problem of HIV diagnosis in infants, which so far has not received adequate attention.
We developed a decision-tree model that depicts current and potential new diagnostic-screening strategies for HIV infection in infants aged <12 months in sub-Saharan Africa. We focused on this age group as most uninfected children have lost passively acquired maternal antibodies by the age of 12 months, and so a positive ELISA IgG antibody result usually indicates HIV infection13. The model was used to calculate annual health benefits in terms of life years saved, disability-adjusted life years (DALYs) saved — a measure of the potential lives lost due to premature mortality and the productive life years lost due to disability — and the proportion of the total disease burden averted with alternative new diagnostic tests. These scenarios varied according to the following parameters: diagnostic infrastructure requirements (moderate, minimal or none) and the corresponding level of health-care access, test-performance characteristics (sensitivity and specificity) and ART availability (5, 50 or 100%). These scenarios were also used to assess the impact of a new diagnostic on ART utilization by calculating the number of regimens saved. All outcomes were adjusted to account for potential harm associated with treatment.
The comparative health benefits of new HIV diagnostic scenarios were measured according to the differences in outcomes provided by the new test compared with the current standard of care (referred to as the status quo). Specifically, we calculated the incremental number of true-positive and true-negative cases of HIV infection relative to the status quo, and translated them into health outcomes using data from the published literature on the risk of HIV-related mortality in ART-treated and untreated children. We explored the impact of potential diagnostics for infant HIV infection both when ART was conditional on clinical staging following the WHO treatment guidelines, and when diagnostic testing replaced clinical staging for the initiation of treatment and ART was administered to all infants with confirmed infection, irrespective of the clinical stage. We varied assumptions about the model input parameters extensively in our sensitivity analyses to test the robustness of the results.
Modelling HIV paediatric diagnosis and management
The decision tree shown in Fig. 1a is a generalized characterization of the current status quo HIV infant diagnostic practice in sub-Saharan Africa. In general, only children who present with symptoms (such as pneumonia, diarrhoea or history of ear discharge) to an IMCI-trained provider have an opportunity to be clinically evaluated for HIV infection12. To this end, the decision tree initially divides the population into infants with symptoms who present to an IMCI-trained provider, and those who have no symptoms or have symptoms but no access to a trained provider. Following Horwood and colleagues12, we assume that symptomatic infants presenting to an IMCI-trained provider might be diagnosed with either HIV infection (IMCI/HIV algorithm positive) or another illness (IMCI/HIV algorithm negative). In accordance with the recent WHO treatment guidelines, we assume that infants who present with symptoms and are diagnosed with HIV infection are then classified as HIV clinical stage III/IV or I/II.
In the model, the probability of receiving appropriate treatment for HIV infection depends on the following factors: the sensitivity and specificity of the IMCI/HIV algorithm, the distribution of infants by HIV disease stage and the availability of ART. Consistent with the WHO treatment guidelines, we assume that only stage III/IV infants are eligible to receive ART13. As our model does not account for maternal HIV status, we assume that CTX is given to all infants who are classified as IMCI/HIV algorithm positive, irrespective of the stage of HIV disease or exposure status. HIV-infected infants who receive ART and/or CTX have reduced mortality and increased life expectancy relative to those who are not treated. For simplicity in the status quo, we additionally assume that infants without symptoms are not identified as HIV exposed, and that asymptomatic infants without access to an IMCI-trained provider do not receive either ART or CTX prophylaxis, even though we recognize that many well-child clinic staff can administer CTX. Finally, we assume that HIV-infected infants are twice as likely to be symptomatic as asymptomatic.
The decision tree shown in Fig. 1b illustrates the introduction of a hypothetical new test for early diagnosis of HIV infection in infants, with some proportion having access to the diagnostic depending on its technological and infrastructure requirements. Specifically, as the technological requirements of a test decrease, it becomes more accessible. We assume that the new diagnostic is available initially to infants with the best access to care, and that availability decreases as access to care worsens. Conditional on access to the new diagnostic, infants are again divided into those with and without symptoms who present to an IMCI-trained provider. In this scenario, infants who present to an IMCI-trained provider might be diagnosed using either the IMCI/HIV algorithm or the new test. Infants receive the new test only if it is associated with improved outcomes (in terms of adjusted life years saved or DALYs) relative to the status quo.
The new diagnostic offers particular advantages for infants who were previously without access to an IMCI provider and/or who do not have symptoms. Specifically, we allow for the possibility that a new diagnostic with minimal infrastructure requirements will be utilized in well-baby settings (that is, immunization sites) for screening HIV infection in infants without symptoms, and for detecting HIV infection in sick infants without access to an IMCI provider.
Table 1 summarizes the values and plausible ranges used in the sensitivity analyses. We employed estimates of the number of births in 2005 to derive our model population of infants aged <12 months20. As a direct reflection of the lack of a widely available test for the early diagnosis of HIV infection in infants, the true number of cases is unknown. We therefore derived estimates of infant prevalence of infection based on maternal seroprevalence data from countries in sub-Saharan Africa and from estimates of mother-to-child (MTC) transmission, which incorporates in utero, perinatal and breastfeeding transmission.
The probability that a sick infant will present to an IMCI provider in sub-Saharan Africa and the proportion of all providers who are IMCI trained are not precisely known; we therefore relied heavily on expert opinion. Two experts who were not members of the HIV working group of the Global Health Diagnostics Forum suggested that ∼40% of all infants in Africa present to a provider for fever and cough at least once during the first year of life. This is generally consistent with data from the WHO on the estimated coverage of IMCI training in these regions (http://www.who.int/child-adolescent-health/overview/child_health map12_04.jpg). For simplicity, and in the absence of available data, we assumed that all providers caring for infants who were sick enough to present with symptoms were IMCI trained. We did not make any assumptions about the level of training among providers at well-child clinics.
Access to different levels of health care in the model were estimated from a multinomial logistic-regression model informed by data obtained from the Demographic and Health Surveys (DHS) from 17 African countries for 2000–2005 (ref. 21). Moderate infrastructure implies consistent access to running water and electricity, and minimal laboratory equipment as well as a trained provider (such as a nurse or laboratory technician). Minimal infrastructure implies limited access to water and electricity, a physical location with no laboratory equipment and a minimally trained health provider (such as a pharmacist or village health worker). No infrastructure refers to settings without reliable water or electricity, and makes no assumptions about the training or literacy of the caregiver.
Several studies have indicated that, in the absence of ART, most HIV-infected children will not reach the age of 5 years4,22,23,24. We estimated that untreated HIV-infected infants would live on average 3.75 years based on a study by Marston and colleagues25, which used statistical methods and indirect data sources to project net survival for HIV-infected children in Africa in the absence of ART. This treatment has not been available in Africa for long enough to compile data on the life expectancy of individuals who have been properly treated since infancy. Based on the experience of children treated with ART in the USA, our experts predicted that an infant who was appropriately treated for HIV infection in sub-Saharan Africa could be expected to achieve at least 50% of the normal life expectancy (that is, 23 years).
There is considerable debate and uncertainty about the contribution of CTX prophylaxis to the overall survival of HIV-infected infants26. One randomized clinical trial in HIV-infected children in Zambia found that CTX prophylaxis reduced mortality by 43%; however, the study focused on the role of CTX prophylaxis after infancy26,27,28. Given the difficulty of estimating the incremental life expectancy gains of HIV-infected infants treated with CTX compared with untreated HIV-infected infants, we relied on data from a modelling study of CTX prophylaxis in adults, which found that individuals treated with CTX had a life-expectancy gain of 5.2 months (0.45 years) compared with those without prophylaxis18. We therefore estimated the life expectancy of HIV-infected infants treated with CTX alone to be 3.75 + 0.45 = 4.2 years, and we varied this estimate extensively in the sensitivity analyses. We relied on published data for estimates of the prevalence of infants receiving ART. A recent WHO report stated that 660,000 children aged <15 years were in need of immediate ART treatment in 2005, 90% of whom lived in sub-Saharan Africa29. Another report from UNICEF indicated that <5% of young children in need of AIDS treatment received it30. Conservatively, we used 5% as an estimate of the upper limit of the proportion of infants currently receiving ART during the first year of life. However, in light of the notable rise in funding for HIV/AIDS, which, coupled with reduced drug costs, has facilitated the rapid scale-up of ART in the developing world, we allowed for the possibility that it might become available to more infants in the future. Therefore, all of our analyses considered three levels of ART availability: 5, 50 and 100%.
Here we report outcomes in terms of annual adjusted life years and DALYs saved by a new diagnostic. Specifically, when an infant receives either appropriate or inappropriate ART, we adjust the life years saved to allow for the possibility that some harm to society follows (for example, the development of antiretroviral drug resistance and/or stigma associated with treatment). Moreover, each time an infant is treated inappropriately with ART, scarce resources are utilized that might have otherwise been used to treat infected infants. In the absence of empirical or other data that would permit a direct estimation of the number of life years lost due to specific negative consequences of treatment, we quantified the harm of treatment using a single measure, C, which represents the proportion of life years lost as a result of treating one infant. Conceptually, if an HIV diagnostic is widely accepted and adopted by the medical community, then using the test must be better than the alternatives of treating every infant or treating no infants. According to neoclassical economics, the preferences of rational people are revealed by the choices they make; therefore, we believe that these preferences provide adequate information with which to estimate how much weight the medical community assigns to specificity relative to sensitivity. We then translate this revealed preference into an estimate for C. Further details of this approach and calculation are presented elsewhere21.
To account for a high degree of uncertainty in the model, we conducted one-way and two-way sensitivity analyses, and a probabilistic multivariate uncertainty analysis. A Monte-Carlo simulation was used to estimate the probable range of expected outcomes. All of the input parameters were allowed to vary independently over a defined range unless they were explicitly linked in the model31. The ranges were derived from the published literature and expert opinion. The Monte-Carlo simulation was run for 1,000 iterations, for each of which the values of the input variables were chosen at random from the defined probability distributions and fed into the model to produce outcomes. The standard deviation of the distribution of outcomes was then used as a measure of the outcome uncertainty.
Before evaluating the potential impact of a new test for the early diagnosis of HIV infection in infants, we assessed the ability of our model to predict reasonable estimates of HIV/AIDS-related mortality in the target population. Initially, we obtained estimates of the number of AIDS-related deaths for children aged 0–5 years from the WHO World Health Report, which represented the best available data. We then populated the model with estimates of 2–5 year cumulative-mortality rates for treated and untreated HIV-infected infants from the published literature18,25,26. Our model predicted 435,649 deaths over a 1 year period due to HIV/AIDS in sub-Saharan Africa, which was ∼1.5 times higher than the WHO estimate (http://www.who.int/whr/2005/annex/en/index.html). However, the WHO estimates are model projections based on a number of cross-sectional data sources, rather than empirically derived results, and they are also relatively dated. Our experts therefore felt that the model estimate, although high, was within the plausible range.
Improved outcomes as ART eligibility and availability increase
Table 2 presents the attributable health outcomes for the scenario in which all infants who test positive by the IMCI/HIV algorithm or the new diagnostic are eligible to receive ART, irrespective of clinical stage. Each numbered row in Table 2 corresponds to a potential new diagnostic defined by its sensitivity, specificity, infrastructure requirements and ART availability. The level of infrastructure required determines the probability that an infant has access to a new test21. A test requiring no health-care infrastructure is assumed to be available to all infants, and a test requiring moderate infrastructure is assumed to be available only to those with symptoms who can access a hospital or clinic with an IMCI-trained provider and laboratory services.
The results indicate that a test that is more specific than sensitive compared with the status quo is associated with better outcomes across all infrastructure and ART levels. For example, an improvement in specificity from 80 to 90% over the status quo test results in more adjusted life years saved than an increase in sensitivity from 70 to 90%. Specifically, a new test that is 70% sensitive, 90% specific and requires a minimal level of infrastructure (Table 2, test 13) saves an additional 83,895 adjusted life years compared with the status quo IMCI/HIV algorithm, which is 70% sensitive and 80% specific. By contrast, at the same level of minimal infrastructure and ART availability, a test that is 90% sensitive and 80% specific (Table 2, test 14) saves 73,836 additional adjusted life years compared with the status quo. The relative importance of specificity compared with sensitivity holds true across all levels of access and ART availability. The greater impact of test specificity is partially explained by the relatively high harm associated with ART, as higher specificity indicates fewer false-positive test results and, consequently, fewer ART regimens being used inappropriately. As specificity increases, ART regimens are saved compared with the status quo IMCI/HIV algorithm, and are then translated into additional life years saved. By contrast, the sensitivity of the test has no bearing on the number of ART regimens saved.
The results of our analysis also indicate a marked improvement in health outcomes for a new test that allows the detection and treatment of symptomatic and asymptomatic HIV-infected infants when ART treatment eligibility and availability increase. Table 3 shows that when treatment is conditional on clinical stage III/IV disease and ART is 100% available, a perfectly sensitive and specific test requiring moderate, minimal or no infrastructure offsets ∼16, 32 or 38% of the disease burden, respectively. However, when clinical staging is removed and all infants who test positive for HIV are eligible for ART, the same test offsets ∼27,53 or 63% of the disease burden (see ref. 32 for a full presentation of the health outcomes when ART is conditional on clinical staging). Figure 2 indicates that within and across infrastructure requirements, health outcomes for a test with more-realistic performance characteristics (that is, 90% sensitivity and 90% specificity) improve with increased ART availability. As this test outperforms the IMCI/HIV algorithm, it is likely to be adopted into clinical practice by both IMCI-trained and non-IMCI-trained providers.
We did not find any associated improvement in health outcomes with a new diagnostic with worse test characteristics than the status quo IMCI/HIV algorithm, even when infrastructure and ART availability levels surpassed the status quo (data not shown). This phenomenon can be partially explained by the high harm of treatment. For example, the model indicates that the costs (for example, from the treatment of false positives) in terms of life years associated with deployment of a less-specific test outweigh the benefits of a new test with minimal or no infrastructure requirements that can reach more infants with limited access to treatment. Similarly, with a test that is less sensitive than the status quo IMCI/HIV algorithm, enough HIV-infected infants are missed that the loss of life years from the lack of ART outweighs the potential benefit of wider test access.
Effects of assumptions about the new diagnostic
We explored the effects of the model parameter estimates on the outcomes of test 15 in Table 2 through a series of one-way sensitivity analyses, in which we varied each parameter independently. We used elasticity to quantify the effect of a percentage change in any one of the model parameters on adjusted life years saved by the new diagnostic compared with the status quo. Figure 3 shows that for a test that performs better than the current IMCI/HIV algorithm and requires minimal infrastructure, the greatest impact on outcomes is due to uncertainties about life expectancy, and about the sensitivity and specificity of the new diagnostic and the IMCI/HIV algorithm. For this particular new test, the harm of treatment — one of the most uncertain estimates — has little impact on health outcomes.
Although one-way sensitivity analysis can be a useful tool, its results are dependent on the particular point in the parameter space around which the analysis is performed. Therefore, we also performed a series of two-way sensitivity analyses to explore continuous trade-offs among key variables, such as access to care, and the sensitivity and specificity of the new diagnostic. Consistent with the base case results discussed above, the isocurves in Fig. 4 clearly show that even universal access to a highly sensitive and specific diagnostic test would have limited impact on adjusted life years saved in the current environment of 5% ART availability.
We quantified the potential effects of new diagnostics for HIV/AIDS in a population of infants aged <12 months in sub-Saharan Africa. By focusing on infrastructure requirements, and levels of current and hypothetical ART availability, we assessed the impact of a new diagnostic under the assumption that infants with geographic access to it would receive it. We qualitatively controlled for current socioeconomic, cultural and behavioural barriers to care in our assessment.
Our analyses highlight the interplay between access to a hypothetical new diagnostic (as it relates to infrastructure requirements), availability of ART, and the trade-off between sensitivity and specificity. Although the interactions of these factors make it difficult to pinpoint optimal test sensitivity and specificity, the results provide important insights that could inform the development of new diagnostic tests, and highlight the conditions that will influence their potential health impact.
Our results showed that the benefit of a new test for early diagnosis of HIV infection in infants is influenced by the availability of ART in resource-limited countries. At the current level of 5% ART availability, even an ideal test (that is, one with perfect sensitivity and specificity, and universal accessibility) confers only a small incremental benefit compared with the status quo. However, if ART availability is increased to 50%, the incremental benefit of the new diagnostic is significantly enhanced.
Irrespective of ART availability, we found that improved health outcomes are realized when a new HIV/AIDS diagnostic reaches not only symptomatic infants but also otherwise healthy infants and those who currently lack access to an IMCI provider. Such a diagnostic tool could have a substantial impact on the ability of the health-care system to identify children who need treatment, and could fuel more-directed CTX therapy. However, wider access to the current IMCI/HIV algorithm is not sufficient to generate better health outcomes; improved outcomes require wider access to a diagnostic with better sensitivity and specificity than the current IMCI/HIV algorithm.
Finally, we demonstrated that when ART is severely limited or clinical staging dictates treatment-eligibility decisions, the full benefit of a new diagnostic will not be realized. Specifically, as long as ART is conditional on clinical stage III/IV, and infants without symptoms or access to an IMCI-trained provider are ineligible to receive it, the potential for a new diagnostic to reduce the disease burden is limited. Notably, the effectiveness of ART is constrained by the availability and distribution of health-care workers who can prescribe it. If a new diagnostic is integrated into well-child clinics, for example, the caregivers must be able to provide ART to HIV-infected infants or refer them to an HIV clinic.
There are important limitations to our analysis. First, we did not consider maternal HIV status in the model. Although services for the prevention of MTC transmission are increasingly available in resource-limited countries, the prevention-service coverage in 2003 was only 5% in the 30 African countries with the highest HIV prevalence1. Moreover, although more women are receiving tests to determine their HIV status, the current IMCI/HIV algorithm does not include screening questions about maternal status. In addition, linking known maternal HIV status to a sick infant presenting at a clinic has proven difficult. In practice, providers are currently able to link <5% of infants with HIV disease to maternal status20,33. Although we did not model infant exposure via maternal status, we recognize that maternal antibody tests might become an important part of infant diagnosis and care. A second limitation was that we did not consider the possibility that HIV might continue to be transmitted from HIV-infected women to their infants through breastfeeding after initial testing. Third, although we assumed that all infants who were IMCI/HIV algorithm positive received CTX, we appreciate that the coverage of CTX prophylaxis in HIV-exposed infants is severely limited because the infection status of most HIV-infected women is not known.
Conditional on ART availability, significant health benefits in resource-limited settings might be realized by a diagnostic that can definitively identify HIV infection in infants irrespective of outward signs or symptoms, and can be administered in health-care settings such as immunization sites. Hence, there is an urgent need to develop new strategies and biomarkers for the early diagnosis of infant HIV infection that can be deployed in resource-limited areas. Currently, immune biomarkers such as CD4/CD8 T-cell ratios are under evaluation34, and new detection methods for p24 antigen are in development to reduce the cost of clinical screening and the probability of false-negative results35.
Widely available, accurate and reliable tests for the early detection of HIV/AIDS have the potential to facilitate the scale-up of ART and cause a paradigm shift in the treatment of infants. The availability of a higher performance test (that is, >90% sensitivity and >90% specificity) that requires minimal infrastructure could transform current treatment standards by identifying more infants in need of care, thereby allowing providers to expand beyond treating only the most severely ill and paving the way for earlier treatment. We believe that such a diagnostic should optimally be conducted using a heel stick, blood spot or saliva, and should require <2 h to produce results, no refrigeration, electricity or external reagents, and minimal training of providers. Diagnostics requiring minimal or no infrastructure might also improve care-seeking behaviour if their availability does not increase current cost, time and transportation barriers. To this end, resources should be garnered to improve on the currently available serologic biomarkers and assays, which are ineffective in infants who carry maternal HIV antibodies, and to develop new and improved methods for the direct measurement of HIV infection in infants.
The authors thank T. Denny (UMDNJ–New Jersey Medical School, New Jersey, USA), S. A. Bozzette (RAND Corporation, California, USA), M. Urdea (Halteres Associates LLC, California, USA), P. Musoke (Makerere University, Kampala, Uganda), T. M. Meyers (Chris Hani Baragwanath Hospital, Soweto, South Africa) and D. M. Gibb (Medical Research Council Clinical Trials Unit, London, UK).
About this article
This article has not been written or reviewed by the Nature editorial team and Nature takes no responsibility for the accuracy or otherwise of the information provided.