Bulletin of the World Health Organization

Presumptive diagnosis of severe HIV infection to determine the need for antiretroviral therapy in children less than 18 months of age

Nicolas Grundmann a, Peter Iliff b, Jeff Stringer c & Catherine Wilfert d

a. Stanford University School of Medicine, Medical School Office Building (Room 323), 251 Campus Drive, Stanford, CA, 94305-5404, United States of America (USA).
b. ZVITAMBO Project, Harare, Zimbabwe.
c. Centre for Infectious Disease Research in Zambia, Lusaka, Zambia.
d. Elizabeth Glaser Pediatric AIDS Foundation, Washington, USA.

Correspondence to Nicolas Grundmann (e-mail: nico.grundmann@gmail.com).

(Submitted: 09 January 2011 – Revised version received: 24 March 2011 – Accepted: 08 April 2011 – Published online: 28 April 2011.)

Bulletin of the World Health Organization 2011;89:513-520. doi: 10.2471/BLT.11.085977


Important strides have been made in recent years in the prevention of mother-to-child transmission (PMTCT) of the human immunodeficiency virus (HIV). Donor investment in PMTCT programmes has been substantial and governments have made a commitment to roll out the programmes. Yet despite these advances, in developing countries approximately 15% of all new cases of HIV infection are diagnosed in children.1 Between one half and two thirds of the children who become infected with HIV will die before their second birthday2 unless they are diagnosed and placed on antiretroviral therapy (ART).3 In sub-Saharan Africa, only around 15% of HIV-exposed infants are tested for HIV in the first two months of life, and the majority of infected children less than 2 years of age die without having their HIV status definitively confirmed through virologic tests.4,5

The World Health Organization (WHO) recommends giving ART to all infants less than 24 months of age with a confirmed diagnosis of HIV infection (i.e. with positive virologic tests results), even if they have no symptoms.4 Antibody tests cannot provide the definitive diagnosis of HIV infection in very young children because maternal IgG antibodies cross the placenta into the fetal circulation during pregnancy and remain detectable for up to 18 months after birth.6 Thus, a virologic method, typically nucleic acid testing to detect viral desoxyribonucleic acid (DNA) or ribonucleic acid (RNA), is required. Such tests are much more complex and costly than more common HIV laboratory assays and have proven difficult to implement in national treatment programmes worldwide.710

In resource-constrained settings, lack of access to virologic tests leads to reliance on the presumptive diagnosis of HIV infection for case management. In fact, in its revised 2010 guidelines WHO recommends initiating ART in all children less than 18 months of age with a presumptive clinical diagnosis of severe HIV disease, established on the basis of a defined set of clinical symptoms in infants with antibodies against HIV.4 Historically, however, health-care providers have been reluctant to treat children with a presumptive diagnosis due to various concerns, including but not limited to: initiating HIV negative infants on potentially toxic antiretrovirals, stigma for infants with unclear HIV status, or physician comfort with moderately predictive diagnostic tools. Clinical screening parameters alone, without antibody testing, have demonstrated low sensitivity (20–23%), although their specificity is reasonably high (92–94%).1115 Thus, efforts to improve the management of children with a presumptive diagnosis of severe HIV disease have focused on two goals: (i) identifying a set of clinical criteria with high enough sensitivity to detect severe HIV disease so that more affected children receive ART, and (ii) identifying HIV-positive children at high risk of death15 and in need of ART without delay.

In our model we have used a set of modified clinical criteria developed by Iliff et al. to define severe HIV disease. The criteria include the following: pneumonia, current diarrhoea, ear discharge, very low weight, generalized lymphadenopathy, oral thrush, and tuberculosis in either the infant or in a sibling or parent of the infant. These criteria, based on the results of a clinical trial of approximately 14 000 infants in Zimbabwe, have shown a sensitivity and specificity ranging from 19.8% to 67.7% and from 94.0% to 90.4%, respectively, depending on the child’s age. They are “modified” in the sense that they resemble WHO’s criteria for the Integrated Management of Childhood Illness15 but exclude persistent diarrhoea (> 14 days) and parotid swelling and include underweight (i.e. weight below the median for age). By adding underweight to the algorithm, Iliff et al. achieved a sensitivity of 65% at 6 weeks, 79% at 6 months and 86% at 12 months, respectively. This innovation is important, as HIV-positive children lose weight early and consistently.16 According to some, the modified clinical criteria are diagnostically superior to the WHO criteria.15

WHO’s 2010 guidelines for presumptively diagnosing severe HIV disease call for the use a rapid HIV antibody test in addition to clinical symptoms. In facilities that practice PMTCT or provide HIV care and treatment, rapid HIV antibody tests are uniformly available; CD4 count can usually be performed on site or is accessible to maternal and child health facilities, but not as widely as antibody tests. When compared with nucleic acid testing (gold standard), CD4 count has shown a sensitivity of 87% and a specificity of 78% for the presumptive diagnosis of severe HIV disease in 3-month-old children.17 However, the problem with using only antibody tests or CD4 count to presumptively diagnose severe disease is that in very young children their individual predictive values are unacceptably low. Besides, not all facilities can perform a CD4 count.15

No published reports until now have explored the combined use of rapid HIV antibody testing and the modified clinical criteria, with or without CD4 count, for the presumptive diagnosis of severe HIV disease in very small children. We propose a screening algorithm based on a combination of these different parameters and hypothesize that such an algorithm can effectively identify HIV-infected children less than 18 months of age who need ART. We quantify how the algorithm improves on the test characteristics of the criteria described in WHO’s 2010 guidelines and suggest additional modifications to further improve test performance. In short, the objective of this study is to develop, for the benefit of clinicians, a means of effectively identifying children under 18 months of age who are in need of ART in settings where virologic tests are not available.


We constructed a conditional probability model to compare the test performance characteristics of three algorithms based on non-virologic parameters to presumptively diagnose severe HIV disease (Fig. 1). The algorithms were as follows: (i) modified clinical criteria, as described earlier, alone; (ii) modified clinical criteria plus HIV antibody testing; (iii) modified clinical criteria plus HIV antibody testing plus CD4 count. Each algorithm was evaluated for diagnostic accuracy against polymerase chain reaction (PCR), the WHO gold standard. The model considered ART eligibility at 6 weeks, 6 months and 12 months of age. A hypothetical cohort of 1000 children was analysed at each of these ages to generate the test characteristics (sensitivity, specificity and positive and negative predictive values). Model outcomes are based on conditional probabilities and hence are independent of the number of individuals simulated. Whenever possible the model was populated with data specific to sub-Saharan Africa. If the literature provided no suitable model input values, we used data collected by the Centre for Infectious Disease Research in Zambia (CIDRZ), whose ART and PMTCT programmes are described in detail in another paper.18 We constructed the model using the Tree Age Pro Suite software package (TreeAge Software, Inc., Williamstown, United States of America) and used Excel (Microsoft Corp., Redmond, USA) to aggregate the data. Table 1 provides a list of model inputs and sources.

Fig. 1. Screening algorithms for identifying infants exposed to the human immunodeficiency virus (HIV) who are eligible for antiretroviral therapy
Fig. 1. Screening algorithms for identifying infants exposed to the human immunodeficiency virus (HIV) who are eligible for antiretroviral therapy
Ab, HIV antibody testing; CD4, CD4+ T lymphocyte.

Henceforth throughout this paper, “antibody-positive” will designate positivity to an HIV antibody test; “clinically-positive” will designate less than the median weight for age and two or more of the signs or symptoms of the modified clinical criteria; and “CD4-positive” will apply to children with a CD4 count below the 2006 WHO threshold for initiating ART (CD4 ≤ 25% of total lymphocytes at 6 weeks and 6 months of age and CD4 ≤ 20% of total lymphocytes at 12 months of age).10

Some subpopulations of HIV-infected infants are at a particularly high risk of dying. According to data from Zimbabwe, of HIV-positive children whose diagnosis is confirmed by PCR at 6 weeks, those who are clinically positive are 50% more likely to die before reaching 6 months of age than those who are clinically negative.15 Furthermore, young CD4-positive infants who are not started on ART have a risk of dying by 6 months of age 50% higher than CD4-negative infants.19 Clinically positive or CD4-positive infants therefore constitute a high-risk subgroup among HIV-positive infants. We explored the usefulness of the screening algorithms for identifying these high-risk infants.

Model inputs

Prevalence of HIV infection

We assumed a prevalence of HIV infection of 11.8% among HIV-exposed infants 6 to 8 weeks old. This is the reported perinatal rate of mother-to-child HIV transmission in settings where single-dose nevirapine is routinely administered to mothers during labour and to children within 72 hours of birth.20,21 Thus, we used this prevalence on the assumption that minimal antiretroviral prophylaxis services are available to mothers and infants. It is an underestimate of the prevalence of HIV infection in places where ART is not available or not optimally implemented, and it is an overestimate of the prevalence where more complex prophylactic regimens or ART are routinely administered.

We assumed that the prevalence of HIV infection was higher among infants between 6 and 12 month of age than among younger infants because of potential transmission during breastfeeding. This increase in HIV prevalence in older infants is offset to some degree by higher early mortality among HIV-positive infants, which was accounted for in the prevalence estimates.20,22

Seroreversion in HIV-negative children

In HIV-negative (HIV−) children maternal antibodies against HIV decay and disappear over time. By 6 weeks of age 1.5% of infants have seroreverted. Because seroreversion takes an average of 11.6 months,23,24 a positive antibody test in a child younger than 18 months is not indicative of HIV infection (potential false positive). On the other hand, a negative antibody test indicates that the child is not infected (likely a true negative). We used seroreversion rates from a cohort of South African children to approximate the age-dependent test performance of rapid HIV testing.24,25 While nearly all seroreversion studies date from the early 1990s and measured the rate of decay of maternal IgG antibodies with enzyme-linked immunosorbent assays or Western blots, the values found in those studies closely match more recent rapid HIV antibody test results in infants (Table 2).27 In this study we assumed that infants were given a rapid HIV antibody test.

Sensitivity and specificity of the modified clinical criteria

The modified clinical criteria consist of nine HIV-associated clinical signs and symptoms. A child with poor weight gain in the presence of any two symptoms included among the modified clinical criteria is considered clinically positive, as indicated earlier. The sensitivity of the modified clinical criteria increases with age. It increases from 19.8% at 6 weeks to 67.7% at 1 year, whereas specificity drops slightly, from 94% to 90.4%. The positive predictive value increases from 51% to 54% and the negative predictive value, from 78.8% to 85.5%.15

Sensitivity and specificity of CD4 cell count

We obtained CD4 counts for HIV-infected children from CIDRZ data, which are specific to infants enrolled in CIDRZ programmes in Zambia from 2004–2007.16 The values in this cohort closely match the values given in the literature for other sub-Saharan African cohorts and values used in previously published modelling analyses.28 For our model, CIDRZ data were stratified by age into 0–3 months, 4–9 months and 10–15 months to compose the 6-week, 6-month and 12-month age groups, respectively. More than 99% of HIV-negative children have a CD4 count above the recommended threshold for treatment eligibility. This translates into a specificity of 99% for CD4 count in our model.25,26 The 1% of HIV uninfected infants whose CD4 count is below WHO thresholds for treatment eligibility are categorized as false positives.

High-risk infant and sensitivity analyses

Several of the clinical signs and symptoms in the modified clinical criteria are also symptoms of stage III/IV disease in HIV-infected infants and could therefore be used to identify the infants at greatest risk of dying within 6 months. We assumed a correlation coefficient of 0.5 between a diagnosis of severe HIV disease according to the modified clinical criteria and a diagnosis of stage III/IV HIV disease according to WHO criteria. In sensitivity analyses this correlation coefficient ranged from 0.5 to 1.0.


The test characteristics of the three screening algorithms used to presumptively diagnose severe HIV disease are summarized in Table 3. The performance of each algorithm depends on the model inputs. The percentage of children correctly treated (true positives plus true negatives divided by total tested) varies from 79% in our worst-performing algorithm to 96% in our best-performing one, respectively.

Table 4 shows the performance of each screening algorithm when used to determine which infants are at high risk of dying and eligible for ART. Sensitivity analyses in Table 4 represent the correlation between the results obtained with the modified clinical criteria and WHO staging. As shown, by using the best-performing screening combination (modified clinical criteria + CD4 count + HIV antibody testing) infants at high risk of dying can be correctly treated without virologic testing 93% of the time at 6 weeks, 91% at 6 months and 98% at 12 months, respectively.


International guidelines do not recommend using clinical algorithms without concurrent HIV antibody testing to presumptively diagnose severe HIV disease in children less than 18 months of age. We included this among our algorithms to determine how much of an improvement in test performance can be expected when routine rapid HIV antibody testing is added to the clinical criteria. In children 6 weeks of age test performance remains essentially unchanged when rapid HIV testing is added to routine clinical screening. This is not surprising because infants in this age group still have high titres of circulating maternal antibodies. In children 12 months of age test specificity can improve by nearly 10% and positive predictive value by nearly 30% by adding rapid HIV antibody testing to routine clinical screening as recommended in WHO’s 2010 guidelines for the presumptive diagnosis of severe HIV infection in infants.

Prognostic markers should display high sensitivity for identifying a specific condition as well as enough specificity to exclude individuals without the condition. Of our modelled screening algorithms, the optimal non-virologic diagnostic combination for determining ART eligibility in infants appears to consist of the modified clinical criteria plus rapid HIV testing plus CD4 count. With the addition of CD4 count to the currently recommended rapid HIV antibody test plus clinical screening, test sensitivity at 6 weeks of age can improve from 19.5% to 71.2%, test specificity from 94.1% to 93.9%, positive predictive value from 30.7% to 60.9% and negative predictive value from 89.7% to 96.1%. The benefits of including CD4 count to the current recommended screening parameters at 12 months of age are less dramatic but still noteworthy. The proposed modified clinical criteria plus antibody testing with or without CD4 count satisfy established thresholds for low-cost diagnostic tests for use in children under 18 months of age (> 70% test sensitivity and > 90% test specificity)28 and respond to WHO’s request for novel tools for use in settings where nucleic acid tests are not available.4

The predictive power of this optimal diagnostic algorithm in children 6 weeks of age is primarily determined by the high sensitivity of CD4 testing and the poor sensitivity of rapid HIV antibody testing. For this age group, the algorithm’s predictive power derives from the rapid HIV antibody test. In settings where CD4 count and nucleic acid testing are not available, applying the modified clinical criteria plus rapid antibody tests is superior to using only the current WHO clinical criteria for the Integrated Management of Childhood Illness.15 Adding rapid HIV antibody testing to the modified clinical criteria is particularly beneficial when diagnosing older infants (6- and 12-month-olds), as the positive predictive value increases from 34.5% to 51.1% and from 61.6% to 96.9%, respectively.

Modelled test characteristics are most heavily influenced by three variables: the prevalence of HIV infection in each age group, the time to seroreversion in HIV-negative children and immunological status (CD4 count) in HIV-positive infants. Uncertainty around the input for the prevalence of HIV infection is low. Modelled prevalence data are specific to sub-Saharan African, concur with internal CIDRZ analyses and closely match the inputs used in previous analyses.28 The rates of maternal antibody decay and the performance characteristics of rapid HIV antibody tests are not well documented in African infants. Data from South African and Ugandan cohorts of children born to HIV-infected mothers suggest that rapid HIV test characteristics are similar to the well studied outcomes from enzyme-linked immunosorbent assays and Western blots.23,24,27 Large uncertainty does surround the distribution of CD4 counts among HIV-infected infants. Decreasing the specificity of CD4 testing would increase the number of infants with false-positive results. For model inputs we used CIDRZ data for infants enrolled in care and treatment programmes in Zambia. These data closely agree with information from outside experts and with published data on the subject.16,25,26,28

Benefits to high-risk children

According to data from South Africa, HIV-related mortality among HIV-exposed infants peaks at 2 to 3 months postpartum.29 By training health-care staff to apply the modified clinical criteria and rapid HIV antibody test, more children at high risk of dying can be identified and treated. These tools are currently readily accessible and can be applied to assess all children. The tools allow staff to quickly identify infants at high risk of dying whose mothers’ HIV status is unknown or undocumented. They can then try to obtain a definitive diagnosis in these children.

According to WHO guidelines, any infant born to an HIV-positive mother or any HIV-positive infant should be placed on co-trimoxazole prophylaxis at 6 weeks of age.30 The rapid HIV antibody test administered at 6 weeks can identify all HIV-exposed infants and ensure that co-trimoxazole prophylaxis is started as early as possible in children who need it. In older HIV-exposed children who are no longer breastfeeding, a negative rapid HIV antibody test is basis enough to discontinue co-trimoxazole prophylaxis.

In countries that are scaling up the use of nucleic acid testing, screening algorithms will allow health-care providers to start therapy on infants with presumptively severe HIV disease while they await the results of nucleic acid tests. In high-volume clinics and hospitals, combined use of rapid HIV testing and modified clinical criteria can reduce the number of nucleic acid tests needed and identify most HIV-positive children who are eligible for ART.15

Study limitations

The test characteristics of the modelled diagnostic algorithms were obtained by using a combination of Zambia-specific data and the best assumptions that could be made from the literature regarding unknown parameters. Hence, actual test performance could vary depending on factors such as seroreversion rate, rate of HIV transmission from mother to child, and non-HIV-related immune parameters (e.g. CD4 counts in HIV-uninfected, immune-competent infants). Iliff et al. have presented a large body of data substantiating the utility of the modified clinical criteria. Operational research is currently under way to determine the exact test efficacy of this model in Zambia and further studies are needed to assure the generalized applicability of the diagnostic algorithm herein discussed.


Health staff in countries lacking access to nucleic acid tests need to rely on a presumptive diagnosis of severe HIV disease to initiate ART in young children. In practice, therapy is often delayed until the age of 18 months, when antibody tests can provide a definitive diagnosis. The optimal algorithm discussed in this paper has better sensitivity and specificity than other algorithms for the presumptive diagnosis of severe HIV disease and, if implemented, will increase the number of HIV-infected infants placed on ART in settings where nucleic acid tests are not available.


Funding was provided by Stanford University School of Medicine, the Elizabeth Glaser Pediatric AIDS Foundation, the Centre for Infectious Disease Research in Zambia and the Zimbabwe Vitamin A for Mothers and Babies Project.

Competing interests:

None declared.