Effectiveness of pediatric antiretroviral therapy in resource-limited settings: a systematic review and meta-analysis.

BACKGROUND
Responses to antiretroviral therapy (ART) among human immunodeficiency virus (HIV)-infected children in resource-limited settings have recently been reported, but outcomes vary. We sought to derive pooled estimates of the 12-month rate of virologic suppression (HIV RNA, <400 copies/mL) and gain in CD4 cell percentage (DeltaCD4%) for children initiating ART in resource-limited settings.


METHODS
We conducted a systematic review and meta-analysis of published reports of HIV RNA and CD4 outcomes for treatment-naive children aged 0-17 years old by means of the Medline, EMBASE (Excerpta Medica Database), and LILACS (Latin American and Caribbean Health Sciences Literature) electronic databases and the Cochrane Clinical Trials Register. Pooled estimates of the reported proportion with HIV RNA <400 copies/mL and DeltaCD4% after 12 months of ART were derived using patient-level estimates and fixed- and random-effects models. To approximate intention-to-treat analyses, in sensitivity analyses children with missing 12-month data were assumed to have HIV RNA>400 copies/mL or DeltaCD4% of zero.


RESULTS
In patient-level estimates after 12 months of ART, the pooled proportion with virologic suppression was 70% (95% confidence interval [CI], 67%-73%); the pooled DeltaCD4% was 13.7% (95% CI, 11.8%-15.7%). Results from the fixed- and random-effects models were similar. In approximated intention-to-treat analyses, the pooled estimates decreased to 53% with virologic suppression (95% CI, 50%-55%) and to a DeltaCD4% of 8.5% (95% CI, 5.5%-11.4%).


CONCLUSIONS
Pooled estimates of reported virologic and immunologic benefits after 12 months of ART among HIV-infected children in resource-limited settings are comparable with those observed among children in developed settings. Consistency in reporting on reasons for missing data will aid in the evaluation of ART outcomes in resource-limited settings.

Manuscript review-Publications were included in the final analysis if they reported the proportion of patients with HIV RNA level below assay limit of detection (virologic suppression) or the change in CD4 percentage (ΔCD4%) at 12 months after ART initiation, or if they reported sufficient information to perform these calculations.
Citations were limited to reports in which ≥95% of children were treatment-naïve (having received no prior antiretroviral drugs, except for prevention of mother-to-child transmission (PMTCT)). Cohorts including both adults and children were included if pediatric outcomes were reported separately. When more than one publication reported outcomes from the same cohort, the most recent publication (or the largest study cohort, if two publications occurred within a 1-year period) was used. Citations were not included if they reported only on outcomes among critically ill children.
Titles and abstracts were independently reviewed by three authors (AB, AC, AM). If disagreements between authors were encountered, eligibility for inclusion was determined by consensus.

Data extraction and outcomes definitions
Data were extracted independently by two authors (pairs of AB, AC, AM), and discrepancies in data extraction were resolved by repeat manuscript review and consensus. Baseline data, collected at the time of ART initiation (or, if not available, at the time of enrollment), are outlined in Table 1.

Primary outcomes
Virologic suppression: Virologic suppression was defined as the proportion of children reported to have RNA <400 copies/ml after 12 months of ART. When virologic suppression thresholds were reported as RNA <50, <100, <250, or <300 copies/ml, we conservatively analyzed these results as <400 copies/ml. ΔCD4%: When the mean or median ΔCD4% was reported for all children with 12-month data, this value was used in the meta-analysis. When ΔCD4% was not reported, or was reported for only a subset of children with baseline and 12-month values, 8 the ΔCD4% was calculated by subtracting the mean (or median) CD4% at baseline from the mean (or median) 12-month value. Most of the reported interquartile ranges (IQRs) of CD4% showed symmetry around the reported medians, supporting the assumption that CD4% or ΔCD4% might be nearly normally distributed. 15,16 Mean and median values were therefore analyzed together.
Secondary outcomes-Secondary 12-month outcomes included growth parameters (weight-for-age, height-for-age, or weight-for-height Z-scores); 17,18 mortality and loss-tofollow up (LTFU) rates; and number of children with 12-month RNA and CD4 data. Secondary outcomes were not aggregated into pooled estimates, but were collected to describe the effect of ART on growth and mortality and to evaluate the impact of missing data.

Data analysis
Methods for combining estimates-Analyses were performed using SAS v9.1 (Cary, NC). Because all studies included in the analysis were cohort studies without control arms, we used a straight-forward pooling method of weighting each study by the number of children with 12-month RNA or CD4 data (patient-level analysis). For comparison, we also calculated pooled estimates using two traditional meta-analytic methodologies: 1) a fixed-effects model approach (weighted by the inverse of the variance from each study), and 2) a random-effects model approach (based on the DerSimonian-Laird method; weighted by the inverse of the sum of between-and within-study variances). 19 Methods for examining heterogeneity, bias, and study quality-Statistical heterogeneity was assessed using Q statistics with Chi-square tests 19 and was summarized by the I 2 statistic, 20 which reflects the proportion of total variation across studies that is due to heterogeneity rather than to chance. The presence of publication bias was assessed using Begg's and Egger's tests. 19 We also examined the relationship between several clinical and programmatic factors and the primary outcomes to better understand sources of anticipated statistical heterogeneity. Information regarding study quality and comparability was collected according to published guidelines 21,22 Sensitivity analyses --effects of missing data-The majority of reported viral suppression and ΔCD4% results were derived from "on-treatment" analyses; children who initiated ART, but for whom 12-month RNA and CD4% data were missing, were excluded. To examine the effects of missing data on the patient-level estimates, we conducted two sensitivity analyses.
First, we excluded studies in which 12-month data were available for <50% of children initiating ART, in which data to calculate this proportion were not reported, or in which missing data could not be assessed because cohorts were limited to children with complete follow-up data. Second, we calculated a proxy for "intention-to-treat" outcomes for each study. For the viral suppression outcome, we used as the denominator all children who began ART >12 months prior to the date of analysis, or all children in the cohort when entry dates were not specified. The numerator remained the number of children known to have RNA<400/ml. This analysis assumed that children who died or lacked RNA data at 12 months in fact had RNA levels >400/ml. For the ΔCD4% outcome, we assumed that all patients who initiated ART but lacked 12-month follow-up data had zero CD4% change.
Supplementary information-Additional details regarding the literature search, data extraction, and data analysis are available from the authors upon request.

Literature search
Primary search and abstract review-436 citations were retrieved from Medline, 168 from EMBASE, 16 from the Cochrane Clinical Trials registry, 52 from LILACS, and 10 from expert discussion and bibliography review. After duplicate citations were eliminated, 591 citations remained; 546 abstracts were excluded for the reasons outlined in Figure 1.
Manuscript review-Forty-five published abstracts were selected for full manuscript review. Thirty manuscripts were excluded (Figure 1), leaving 15 papers 8-10,23-34 eligible for at least one of the primary analyses ( Table 1). Table 1 describes characteristics of the included cohorts, from 15 countries in Asia, Africa and the Caribbean. Numbers of children initiating ART in each program ranged widely (16-2,938; total: 5,928), as did age at ART initiation (mean/median, 0.1-10.0 years; range, 0.0-15.2 years). Overall, children initiated ART with low immune function: mean/median baseline CD4% ranged from 3.8-30.0% (mean 8.1%). First-line ART was NNRTI-based in 81% of children for whom regimens were described. All but one of the studies were observational. The single trial randomized infants to initiate ART before three months of age or to defer ART until WHO 2006 criteria for ART initiation were met. 23 To remain consistent with other included studies, reflecting ART initiation in accordance with pre-2008 guidelines, 35,36 we included only data from the deferred ART arm of this study in the pooled analyses. Eleven studies 8,10,[24][25][26][27][28][29][30][31]34 reported baseline growth parameters. In five [25][26][27]29,30 of these studies, mean or median values indicated at least moderate underweight (weight-for-age), stunting (height-for-age), or wasting (weight-for-height), defined as Z-scores <-2. Only three manuscripts 8,23,32 specifically reported on prior receipt of antiretroviral drugs for PMTCT.
Heterogeneity and bias-There was no statistically significant heterogeneity in either the RNA (p=0.26) or ΔCD4% outcome (p=0.99). The percentage of variation due to heterogeneity (I 2 ) was 20.4% for the RNA outcome and 0% for the ΔCD4% outcome. There was no evidence of publication bias for the RNA (Begg's test: p=0.40, Egger's test: p= 0.53) or ΔCD4% outcome (Begg's test: p=0.78, Egger's test: p=0.12).
Graphical visualization of scatter plots did not reveal any association between the primary outcomes and geographic region, study size, year of program initiation, type of ART (PI-vs. NNRTI-based), or stage of disease or age at ART initiation. (Table 2 and Table 3, Sections IIB) -Exclusion of studies with high proportion of missing data. In three studies, >50% of children initiating ART lacked 12-month data; 8,10,31 in two studies, data to calculate the proportion of children with missing data were not reported; 29,34 and in two studies, cohorts were limited to children with complete data 28,33 (Table 4). When these studies were excluded, the pooled estimate of viral suppression was 72% (95%CI: 69-75), and the pooled estimate of ΔCD4% was 14.0% (95%CI: 8.9-19.1).

DISCUSSION
We performed a systematic review and meta-analysis of 12-month virologic and immunologic outcomes for treatment-naïve, HIV-infected children initiating antiretroviral therapy in resource limited settings. Data from nine studies, representing 1,097 children with complete follow-up data, contributed to a pooled estimate of 70% virologic suppression (RNA<400/ml); data from 12 studies and 1,839 children with complete data contributed to a pooled estimate of 13.7% absolute ΔCD4%. These findings are similar to reported ART outcomes for treatmentnaïve children in the US and Europe, which include 12-month virologic suppression rates (<400/ml) of 53-84% 2,5,11,37-43 and median ΔCD4% of 10-13%. 5,[38][39][40] As has been reported for adults, 44 our study highlights that comparable outcomes in children are observed in resource-limited and developed settings, despite advanced stages of disease at ART initiation, predominantly NNRTI-based ART, and substantial barriers to ART delivery in resourcelimited settings. In addition, clinically significant improvements in growth parameters are noted after 12 months on ART. 45 However, missing data remain an important concern, and a proxy for an "intention-to-treat" analysis generates much lower estimates: 53% virologic suppression and 8.5% ΔCD4% at 12 months.
Two recent systematic reviews without meta-analysis summarized responses to pediatric ART in Africa 11 and in a variety of resource-limited settings. 12 A pooled patient-level analysis from 16 African sites also provided mortality and LTFU estimates for children on ART. 13 The current study reinforces the findings of these analyses, including similar ranges of 12-month viral suppression rates 11 , ΔCD4% 11 and mortality; 12 advanced stage of disease and large proportion of children aged >5 years at ART initiation, [11][12][13] wide variation in study size, 11 and notable inconsistency in data reporting. 11 Our analyses suggest that the pooled estimates were not significantly affected by statistical heterogeneity. However, we anticipated that variation in clinical and programmatic factors (clinical heterogeneity) would contribute to differences in the primary outcomes. Although we found no association between RNA or ΔCD4% outcomes and many such factors, including age at ART initiation and PI-vs. NNRTI-based ART, our ability to formally assess such associations was limited due to the relatively small number of included studies. Additionally, incomplete data precluded examination of the effects of receipt of medications for PMTCT, nutritional status, resource-related factors (pharmacy stockouts; free provision of medications), and prevalence or incidence of tuberculosis, malaria, anemia, and diarrheal disease. This analysis has several limitations. First, because conference proceedings may be of more variable quality than published reports and did not change study results when added to published reports in a previous review, 11 we included only published reports. This may have omitted very recent reports of ART outcomes. Second, we included only 12-month treatment outcomes, because 12-month data were provided in the greatest number of reports. Third, programs made use of HIV RNA assays with varying limits of detection. We conservatively designated all children with RNA <50, <250, or <300/ml as having RNA levels <400/ml, thus underestimating true rates of suppression to <400/ml. Finally, consistent with most reports from both developed and resource-limited settings, 9,46,47 the included studies did not report clinical correlations between ΔCD4% or RNA suppression and risks of AIDS-related morbidity. However, ΔCD4% and RNA suppression are likely to serve as reliable surrogate outcomes for these events. 48 Analyses using individual patient data from a large pediatric cohort are anticipated, and these will avoid many of the limitations of meta-analysis. 13,49 Our study also has several notable strengths. First, we expand upon the previous literature by reviewing studies from resource-limited settings outside Africa and by analyzing 10 new reports not included in the prior published review. 11 Next, in addition to narrative review, we use recommended techniques for meta-analyses of observational studies 21 to provide, to our knowledge, the first pooled estimates of the virologic suppression rate and ΔCD4% for children on ART in resource-limited settings. Finally, we conduct two sensitivity analyses demonstrating the impact of missing data.
Data were incomplete for many children who initiated ART. Of 5,928 children initiating ART, 81% lacked 12-month RNA data and 69% lacked 12-month CD4% data. Children who lack follow-up data may be more likely to have died than those who were followed, as has been observed in adults, 50,51 and thus may be assumed to have inferior virologic and immunologic outcomes. If true, this would lead our pooled estimates to overestimate the benefit of ART. However, lack of CD4% or RNA data may not reflect true loss to follow-up to programs. Instead, ART initiation may have occurred <12 months before data reporting, absolute CD4 cell count may have been obtained in preference to CD4% for children <5 years old, or laboratory testing may have been unavailable on the occasions on which children were seen (in the single study reporting this outcome, 25% of children still in care at 12 months lacked CD4% data 27 ).
In sensitivity analyses, virologic suppression rates and ΔCD4% did not change substantially when we excluded studies with a high proportion of missing data. However, when we assumed that all children who died or lacked 12-month data had HIV RNA>400/ml, the pooled estimate of viral suppression fell to 53%. This estimate likely comprises the lower bound of expected 12-month viral suppression rates. Similarly, when we assumed that children with missing CD4% data had true ΔCD4% of zero, the pooled estimate of ΔCD4% fell to 8.5%. Because children who die or are lost to follow-up may have a decline, rather than zero change, in CD4%, it is possible that the true ΔCD4% may be even lower than 8.5%. However, due to some expected decline in CD4% with increasing age, 16,35 small gains in CD4% may represent true improvements in immune function.
Definition and reporting of loss to follow-up (LTFU) and clear descriptions of reasons for missing data are therefore important considerations in interpreting reports of pediatric ART effectiveness and the success of ART programs. Given the high mortality in the first 3-6 months after ART initiation observed in children well as adults, 9,11,27,52 programmatic efforts to retain children in care will be crucial to improving clinical outcomes during the first year on ART.

CONCLUSIONS
This systematic review and meta-analysis demonstrates that the pooled 12-month HIV RNA suppression rate (70%) and ΔCD4% (13.7%) for children initiating ART in resource-limited settings are comparable to those seen in developed countries. This work also highlights important inconsistencies in the reporting of data which may guide the interpretation of clinical and programmatic outcomes, such as definitions of LTFU and descriptions of patient disposition. As pediatric ART programs are expanded worldwide, clear and comprehensive reporting of these data will be crucial to interpreting and comparing the effectiveness of ART in resource-limited settings.

Figure 1. Selection of publications for a systematic review and meta-analysis of pediatric ART effectiveness
Flowchart describing the results of the literature search, including the source of each publication reviewed and reasons for exclusion of studies from the final analysis. After the initial search, 546 abstracts were excluded because: they reported only on ART-experienced children (1), critically ill children (1), animals (1), or only adults (58); they provided only crosssectional data (2) or case reports (3); they did not report on HIV (7); they reported on nonresource-limited settings (12); they described diagnostic tests (20) or cohorts not treated with ART (22); the format was commentary or review, rather than original research (119); they primarily addressed prevention of mother-to-child transmission (146); or they did not report outcomes related to viral load or CD4 cells (154). Abbreviations: ART: combination antiretroviral therapy, RLS: resource-limited setting, MTCT: mother-to-child transmission of HIV.

Figure 2. Forest plot of viral suppression rates (proportion of children with HIV RNA<400 copies/ ml) 12 months after ART initiation for treatment-naïve children in resource-limited settings
Studies included in the pooled analysis of the proportion of children with HIV RNA <400 copies/ml 12 months after initiating ART are shown on the left of the figure. Reported proportions of children with RNA <400 copies/ml are depicted as circles, with the size of each circle proportional to the number of children included in each study; reported estimates and confidence intervals are shown to the right. The dashed vertical line indicates the value of the patient-level pooled estimate (70%). The last column indicates the percentage of total included children contributed by each included study. The last column indicates the percentage of total included children contributed by each included study. CD4%, defined as the percent of total lymphocytes that are CD4+ cells, is a more useful representation of immune function in children <5 years of age than absolute CD4 cell count (total number of CD4 cells/µL of serum), because CD4% maintains a more constant value during normal immune system maturation. 35 As a result, CD4% is often reported for children of any age, while absolute CD4 is usually reported only for adults and children >5 years old. Table 1 Studies included in a meta-analysis of the effectiveness of pediatric ART in resource-limited settings I: A total of 63 infants were randomized (43 to immediate, 20 to deferred); 3 infants from the immediate ART arm were lost to the study before initiating ART, and were not incuded in the cohort reported here. Of 20 randomized to the deferred arm, 3 infants were lost to the study before initiating ART, and 4 did not start ART. To be consistent with other cohorts reporting only on ART initiators, these 7 were not included in the cohort reported here (leaving 13 in the deferred arm and 53 total ART initiators). Data not in parentheses are rates of receipt of any antiretroviral drug for PMTCT, as reported in each study. Data in parentheses are World Health Organization estimates of receipt of any antiretroviral drug for PMTCT among antenatal clinic attendees, based on the each country and the fiscal year closest to the program dates. WHO data on type of antiretroviral drug used for PMTCT are not available; however, in most cases where rates of PMTCT coverage are low, the regimen is likely to be single-dose. 55 Table 2 Meta-analysis of viral suppression rate (proportion of children with HIV RNA<400 copies/ml) 12 months after ART initiation for treatment-naïve children in resource-limited settings.  Secondary outcomes: mortality and loss to follow-up 12 months after ART initiation for treatment-naïve children in resource-limited settings.  Notes:

I. Included Studies
A: Cohort defined as children with 12-month follow-up data B: 1147 children started ART more than 12 months prior to analysis date (and therefore were eligible to have 12-month data). C: 117 children were started on ART; 22 died in the first 6 months. 95 children contributed baseline data; 68 children had follow-up data at 12 months. D: Data are reported only on children with complete 18-month follow-up, not all ART initiators. E: A total of 63 infants were randomized (43 to immediate, 20 to deferred); 3 infants from the immediate ART arm were lost to the study before initiating ART, and were not included in the cohort reported here. Three infants from the deferred ART arm were lost to the study before initiating ART, and 4 did not start ART; these 7 were not included in the cohort reported here (leaving 53 ART initiators: 13 in the deferred arm and 40 in the immediate arm).
F: Over entire duration of follow-up (median follow-up 26 months) G: Over entire duration of follow-up (median follow-up 6 months, range 2-12 months). In addition to 7.5% reported lost to follow-up, 3% had "unknown outcomes." H: Over entire duration of follow-up (median follow-up 378 days) I: Over entire duration of follow-up (median follow-up 20 months, range 0-36) J: 171 children were "in follow-up > 1year." Over median 20 months of follow-up, 24 were reported lost to follow-up and 21 (9%) deceased. Not reported if those remaining (not dead or lost to followup at 12 months) had initiated ART <12 months prior and therefore were ineligible for 12 month data. It is reported that 98% of those in care at 12 months had absolute CD4 cell count data (not included in meta-analysis) and 58% had RNA data. K: Over entire duration of follow-up (median follow-up 16.8 months) L: Not reported if both children who died and 1 child who was lost to follow-up were among the 51 ART-naïve children included in this analysis; the possible range is reported here.