Global temporal changes in the proportion of children with advanced disease at the start of combination antiretroviral therapy in an era of changing criteria for treatment initiation

Abstract Introduction The CD4 cell count and percent at initiation of combination antiretroviral therapy (cART) are measures of advanced HIV disease and thus are important indicators of programme performance for children living with HIV. In particular, World Health Organization (WHO) 2017 guidelines on advanced HIV disease noted that >80% of children aged <5 years started cART with WHO Stage 3 or 4 disease or severe immune suppression. We compared temporal trends in CD4 measures at cART start in children from low‐, middle‐ and high‐income countries, and examined the effect of WHO treatment initiation guidelines on reducing the proportion of children initiating cART with advanced disease. Methods We included children aged <16 years from the International Epidemiology Databases to Evaluate acquired immunodeficiency syndrome (AIDS) (IeDEA) Collaboration (Caribbean, Central and South America, Asia‐Pacific, and West, Central, East and Southern Africa), the Collaboration of Observational HIV Epidemiological Research in Europe (COHERE), the North American Pediatric HIV/AIDS Cohort Study (PHACS) and International Maternal Pediatric Adolescent AIDS Clinical Trials (IMPAACT) 219C study. Severe immunodeficiency was defined using WHO guidelines. We used generalized weighted additive mixed effect models to analyse temporal trends in CD4 measurements and piecewise regression to examine the impact of 2006 and 2010 WHO cART initiation guidelines. Results We included 52,153 children from fourteen low‐, eight lower middle‐, five upper middle‐ and five high‐income countries. From 2004 to 2013, the estimated percentage of children starting cART with severe immunodeficiency declined from 70% to 42% (low‐income), 67% to 64% (lower middle‐income) and 61% to 43% (upper middle‐income countries). In high‐income countries, severe immunodeficiency at cART initiation declined from 45% (1996) to 14% (2012). There were annual decreases in the percentage of children with severe immunodeficiency at cART initiation after the WHO guidelines revisions in 2006 (low‐, lower middle‐ and upper middle‐income countries) and 2010 (all countries). Conclusions By 2013, less than half of children initiating cART had severe immunodeficiency worldwide. WHO treatment initiation guidelines have contributed to reducing the proportion of children and adolescents starting cART with advanced disease. However, considerable global inequity remains, in 2013, >40% of children in low‐ and middle‐income countries started cART with severe immunodeficiency compared to <20% in high‐income countries.


| INTRODUCTION
World Health Organization (WHO) guidelines on initiating combination antiretroviral therapy (cART) in children have expanded paediatric cART eligibility. The initial WHO guidelines (2002, revised in 2006) established CD4 thresholds for cART initiation across all paediatric age groups [1]. In 2010, immediate cART initiation regardless of immunological or clinical thresholds was recommended for children aged <2 years [2] and expanded in 2013 to all children <5 years [3] and in 2016 to all adults and children living with HIV [4]. These modifications were largely predicated by the rapid HIV disease progression and high morbidity and mortality in infants and children [5][6][7], together with evidence of lower mortality (particularly in infants <3 months of age) [8] as well as better growth, immunological and morbidity outcomes and less chronic organ system disease associated with early cART initiation [9][10][11][12][13][14][15][16][17][18][19]. An additional motivation for recommending earlier cART in children was to address the lag in paediatric cART coverage [20].
Serial monitoring of CD4 at cART initiation is important for programme evaluation, indicating the extent to which guideline changes are successfully implemented so that children initiate cART before onset of advanced HIV disease. An analysis from sub-Saharan Africa, Asia, North America and Latin America from 2004 to 2010 demonstrated reductions in the proportions of children initiating cART with severe immunodeficiency, but even in 2010 the vast majority of children in low-and middle-income countries continued to start cART late [21]. The ongoing high burden of advanced HIV disease was a key motivation for the recent WHO 2017 guidelines on managing advanced disease [22]. We analysed CD4 measures at cART initiation from an international collaboration of treatment programmes in sub-Saharan Africa, Asia, Europe, Central and South America, and North America from 1996 to 2013 in order to assess changes in the proportion of children starting cART with advanced HIV disease and the impact of WHO treatment initiation guidelines on this proportion.

| Data sources
Data were collated from four major cohort research networks: the International Epidemiology Databases to Evaluate AIDS (IeDEA); the Collaboration of Observational HIV Epidemiological Research Europe (COHERE); the Adolescent Master Protocol (AMP) study of the Pediatric HIV/AIDS Cohort Study (PHACS) network; and the 219C Long-Term Follow-Up Study of the International Maternal Pediatric Adolescent AIDS Clinical Trials (IMPAACT) network. IeDEA is a global consortium with regional centres that pool clinical and epidemiological data on individuals living with HIV; regions collecting paediatric data are Caribbean/Central and South America, Asia-Pacific, East Africa, West Africa, Central Africa and Southern Africa [23]. COHERE is a collaboration of European HIV cohorts that conducts epidemiological research on the prognosis of people living with HIV across Europe [24]. IMPAACT 219C and PHACS AMP are US-based prospective cohort studies designed to evaluate the impact of HIV infection and cART on youth. New enrolment in recent years is limited [25][26][27]. Pooling of data and their use in collaborative analyses were approved by local institutional review boards. For the present study, regional centres sent de-identified data to the University of Bern, Switzerland, for cleaning and analysis.

| Inclusion criteria and definitions
All patients living with HIV aged <16 years at cART start (irrespective of likely mode of infection) were included in descriptive analyses if they had documented sex and cART start date after 1994, were treatment-na€ ıve at cohort entry (except for exposure to antiretrovirals for prevention of mother-to-child transmission (PMTCT)) or had CD4 measured at first cART start if not na€ ıve at cohort entry. For further analyses, we excluded data from low-or middle-income countries before widespread cART rollout began in 2002 and from countries that contributed <50 patients with a CD4 measurement at cART start. We also excluded data on children who started therapy in a year and country for which <10 children with CD4 measures were reported as well as data from the last calendar year in countries where no CD4 measures were available after May of that calendar year. cART was typically defined by participating cohorts as a regimen of ≥3 antiretroviral drugs from ≥2 drug classes. The baseline CD4 value was defined as the value nearest the cART start date within À6 to +1 months of start. Countries were grouped according to the World Bank classification of annual Gross National Income per capita 2013 as low-income (LIC, ≤US$1045), lower middleincome (LMIC, US$1046-4125), upper middle-income (UMIC, US$4126 to 12,745) and high-income (HIC, ≥US$12,746). We used the 2013 classification as this was the most recent year for which data was included [28]. Age groups at cART initiation were less than twelve months, twelve to thirty-five months, thirty-six to fifty-nine months, five to eleven years and twelve to fifteen years. Severe immunodeficiency was defined according to WHO as CD4% <25% (age <12 months), <20% (12 to 35 months), <15% (36 to 59 months) and CD4 count <200 cells/lL or CD4% <15% (≥5 years) [29].

| Multiple imputation of missing CD4 measurements
We imputed missing CD4 measures at cART initiation from 1996 (HIC) and 2002 (LIC, LMIC and UMIC) onwards for countries and calendar years. We imputed the arcsine square root of CD4% and square root of CD4 cell count simultaneously using chained equations and predictive mean matching, adjusting for country and year of cART start, stratifying by sex, age, income group and cohort. We generated 50 imputed data sets and combined these using Rubin's rule [30].

| Analysis of temporal trends in CD4 measurements
We used generalized additive mixed models to analyse temporal trends in CD4 measures by sex, age and country income group. We assessed three key outcomes at cART initiation: the proportion of children with severe immunodeficiency; median CD4% (aged <5 years); and median CD4 count (aged ≥5 years). Sex, age group and income group and their interactions were entered as fixed effects and country as a random intercept. Yearly trends were smoothed by sex, age group and income group. Data were aggregated by calendar year (three to sixteen years, depending on country), country (thirty-two countries), sex and age group prior to analysis: each combination of these factors corresponded to a cell in the analysis. For each cell, we calculated the number of children with severe immunodeficiency and the median CD4% and count. Each cell was entered in the model with a two-part weight. The first part incorporated the precision of the aggregated values for each cell into the model and corresponded to the number of observations contained in a cell, divided by the average number of observations in all cells in the same income group. The second part corresponded to the ratio of the number of patients that were newly enrolled in that cohort for that year and the number of patients that started cART in that country during that year [31]. The weights of the second part were also normalized by country income group. We used the data set with imputed data for the main analysis. In sensitivity analyses, we fitted the model restricted to the subset with complete data.  [1,2]. We used a generalized linear mixed effects model with income groups as fixed effects and countries as random effects. The main slope was the middle period (2006 to 2009). The variability in the country effect was expressed by a different intercept, allowing parallel deviations from the average income group and random slopes, accounting for slower or faster than average decreases. The model allowed different slopes for different periods but forced the line segments for the three periods to be continuous at the change points, and the change in slopes between periods to be the same for all countries within each income group. We The technical appendix provides further details on the multiple imputation and smoothing (see Appendix 1). Analyses were conducted in R 3.1.0 (R Core Team, Vienna, Austria).

| Descriptive analyses
Data from 67,486 children from 44 countries <16 years old at cART start were submitted to the data centre ( Figure S1). We excluded 13,383 participants who did not meet the inclusion criteria, mostly because they were not treatment-na€ ıve at cohort entry; hence, 54,103 children from 41 countries were included in descriptive analyses ( Figure 1). We excluded 1950 children from further analyses using multiple imputation (nine countries) either because their CD4 measures were before 1996 (HIC) or 2002 (LIC, LMIC and UMIC) (268 children), or from calendar years with <10 children in that country with CD4 measures (1544 children) or with no measure after May of the last calendar year for the relevant country (138 children) ( Figure S1). Data typically spanned the years 2003 to 2013 in low-and middleincome countries and 1997 to 2012 in HIC; in the USA, data were only available for 1996 to 2006, as all participants initiated cART before 2007 ( Table 1). The median year of cART initiation ranged from 1998 in the US to 2012 in Mozambique. The median (interquartile range (IQR)) age of children starting cART was 6 years (3 to 10) in LIC, 5 (2 to 10) in LMIC, 6 (2 to 10) in UMIC and 6 (2 to 10) in HIC. Median CD4 cell counts/percentages at cART initiation by country and sex are shown in Table 2. The median CD4 count at cART start for children ≥5 years old was 250 cells/lL (106 to 416) in LIC, 258 (118 to 446) (LMIC), 189 (57 to 349) (UMIC) and 348 (179 to 601) (HIC). Similar patterns were evident for CD4% in children <5 years ( Table 2). The overall percentage (95%CI) of children starting cART with severe immunodeficiency was 48% (47% to 49%) (LIC), 52% (51% to 53%) (LMIC), 57% (56% to 58%) (UMIC) and 31% (30% to 33%) (HIC).

| Analysis of temporal trends in CD4 measurements
Further analyses were based on 52,153 children, among whom CD4 measurements were imputed for 17,790 patients. (Table S1). Figure 2 shows modelled temporal trends in the prevalence of severe immunodeficiency at cART initiation. Figure 3 shows corresponding trends in median CD4 counts (children aged ≥5 years) or CD4% (children <5 years). In LIC, the estimated percentage of children starting cART with severe immunodeficiency declined from 70% in 2004 to 42% in 2013. Corresponding figures were 67% to 46% (LMIC), 61% to 43% (UMIC) and 45% to 14% (HIC, 1996 to 2012). Of note, in LIC among children aged <1 year, there was almost no decline in the proportion starting cART with severe immunodeficiency.
Temporal trends in median age at cART initiation varied among country income groups ( Figure S4). There was no clear trend in median age at cART initiation in LMIC, while there was a decrease from 2004 to 2013 in LIC (6.2 to 4.6 years) and an increase in UMIC (4.5 to 7.8 years) as well as HIC (2.1 (1996) to 7.4 years (2012)).  (Table 3B, Table S2B).

| DISCUSSION
In our study of about 52,000 children from 32 countries, we found that since the 2006 and 2010 WHO Guidelines addressing paediatric cART eligibility, there have been annual reductions in the proportion of children starting cART with severe immunodeficiency in almost all country income groups. By 2013, less than half of children had severe immunodeficiency when starting cART in all country income groups. Nevertheless, considerable global inequity in advanced HIV disease at cART initiation remains; in 2013, >40% of children in low-and middle-income countries still started cART with severe immunodeficiency compared to <20% in HIC.

| Relationship with WHO Treatment Guidelines
The goal of increasing paediatric cART coverage is one of the key reasons for the WHO 2015 recommendation of immediate cART irrespective of CD4 for all children [4]. It is therefore encouraging that previous guideline revisions that expanded paediatric cART eligibility have been temporally associated with reduction in the proportion of children with advanced HIV disease at cART initiation. Nevertheless, except in HIC, more than 40% of children continued to start cART with severe immunodeficiency in 2013. This supports expanding immediate cART to all children, and indicates that WHO guideline changes alone are insufficient to achieve optimal treatment coverage. Ongoing and proactive engagement by WHO with ministries of health should focus on implementing paediatric testing and treatment guidelines. Expansion of cART eligibility needs to be accompanied by efforts to ensure sufficient and sustainable access to HIV commodities as well as strengthened health systems, including trained health workers and infrastructure, which requires political will and adequate funding [33]. Inequities in the proportions of children starting cART with severe immunodeficiency were not only seen between different country income groups, but also among countries in the same income group. For example, in LICs, the proportion of children starting cART with severe immunodeficiency ranged across countries from 33% to 69%. This may be partly due to the particular programmes in these countries that contributed to this analysis, and the years in which those programmes initiated the majority of children on cART. It is also possible that different levels of donor and national government funding and political will contributed to these differences.
It is somewhat surprising that there were no clear decreases in median age at cART start across all country income groups since eligibility for immediate cART irrespective of CD4 values applied initially to the youngest children (<2 years) and only expanded to those <5 years in 2013 ( Figure S4) [2,3]. However, interpreting trends in age at cART initiation is complex [21]. The WHO guideline shifts towards universal treatment eligibility for progressively older groups of children have been accompanied by recommendations of more effective PMTCT, namely universal lifelong cART for all pregnant and breastfeeding women (so-called "Option B+"). By 2013, the population of two-to five-year olds eligible for immediate treatment would likely have decreased, both due to fewer new infections as well as some children already having started cART at less than two years of age following the implementation of 2010 guidelines. The median age at cART initiation thus depends on the effectiveness of PMTCT programmes in preventing new infant infections, capacity for EID and early cART, as well as the backlog of older children not yet on therapy [34,35]. There was a decrease in age at cART initiation in LIC, suggesting that the WHO 2010 guideline recommendation of universal cART for all children <2 years may have had the greatest impact in the poorest countries, where EID capacity increased substantially in this period [36][37][38][39]. Nevertheless, there was little temporal improvement in the proportion of children <1 year with severe immunosuppression,  indicating that further improvements in EID and early cART access are urgently needed. While there was an increase in overall age at cART initiation in UMIC, this may partly be due to effectiveness of PMTCT programmes preventing new infections. These countries nevertheless experienced the steepest decline in the proportion of infants with severe immunosuppression, indicating substantial progress in achieving cART initiation before the onset of advanced disease in infants. The overall increase in median age in both UMIC and HIC is likely due to reductions in new infant infections due to effective PMTCT, with a relative increase in the proportion of long-term survivors initiating cART. Indeed, in Southern Africa, we have shown an increase in the proportions of children <1 year and >10 years old initiating cART [40]. In Europe, many older patients presenting for testing and care may be immigrants, predominantly from sub-Saharan Africa [41][42][43].

| From improving paediatric cART access to improving paediatric HIV outcomes
CD4 at cART initiation is a useful indicator of responsiveness to WHO treatment initiation guidelines and indicates the extent to which we are successfully preventing advanced HIV disease through early cART initiation [44]. Early cART initiation and retention are particularly critical in infants, as disease progression without cART is rapid and associated with high mortality. However, increasing CD4 at treatment start will only result in improved paediatric outcomes if those children are retained and adherent to effective treatment. There is a much-needed emphasis on children in the UNAIDS 90:90:90 goals which focus not just on diagnosis and treatment access, but also aim for 90% retention on cART, with 90% viral suppression [44]. Indeed, the WHO 2015 treatment guidelines highlight as a research gap the effect of early cART on retention and adherence [4]. The IeDEA collaboration has previously found loss-to-follow-up rates in children by 18 months after cART initiation ranging from 4.1% in Asia to 21.8% in West Africa [45]. A recent analysis as part of an IeDEA-WHO Collaboration found that children <2 years of age and youth aged 15 to 24 years were least likely to be retained in care [46]. Expanding paediatric cART therefore requires strategies not only to improve diagnosis and cART initiation, but also to optimize retention and support adherence across the paediatric and adolescent age spectrum. As more children are initiated on treatment at younger ages, better access to viral load monitoring to assess treatment effectiveness as well as paediatric-friendly second-and third-line drugs will become more important.

| Strengths and limitations
This analysis included a large number of children from many countries and all income group settings. It is one of the first analyses with substantial numbers of children that initiated cART after the WHO 2010 guidelines, allowing enough time to examine their effects. The inclusion of European data where, at least in some countries, there were sufficient numbers of children initiating cART to include data up to 2012, allows for comparison between all income settings after the 2006 and 2010 guideline revisions. A key limitation of our study is that we looked at the impact of WHO guidelines and not individual country guidelines on changes in CD4 at cART initiation. Although in some cases country guidelines pre-date WHO guidelines, they more frequently lag behind WHO guidelines, which may explain the lack of a clearer effect of WHO guidelines on CD4 at cART start. Furthermore, in some country income groups, most data were from only one or two countries. Similarly, for some countries, data may have originated from a small number of programmes or from particular years. This may partly explain the diversity in findings within the same country income group, and our results may not be generalizable to the whole country or income group. However, to reflect reality as best as possible, we used a weighted analysis approach, with more weight assigned not only to more precise estimates obtained from countries contributing many observations to our data set but also dependent on the UNAIDS estimated number of children starting cART in a given country and year [31], so that countries with many children starting cART were adequately represented in our Figure 2. Severe immunodeficiency at the start of cART by age, sex and country income groups (colours). Severe immunodeficiency was defined according to WHO as CD4% <25% (age <12 months), <20% (12 to 35 months), <15% (36 to 59 months) and CD4 count <200 cells/lL or CD4% <15% (≥5 years). Results from generalized additive mixed effects models based on 52,153 children after imputation of missing data. 95% CIs are shown as shaded areas. Vertical lines indicate the changes in WHO guidelines on when to start cART. Age groups are shown along the right edge. Note that no data available for the analysis for 2003 in LMIC and after 2012 for HIC; hence, the periods shown on the graph differ slightly by country income group. cART, combination antiretroviral therapy; WHO, World Health Organization; LMIC, lower middle-income country; HIC, high-income country. analysis. It is also possible that the facilities included in our analysis reflect better access to paediatric cART than occurs in the whole country or income group. However, almost all sites from LMICs in our cohorts are routine care facilities rather than dedicated research cohorts, and followed the relevant national cART guidelines. The routine nature of the data is reflected in the substantial proportions of missing CD4 data at cART initiation, with almost one third of patients missing a CD4 measurement in LIC. We conducted multiple imputation to minimize bias due to missing data. Results including the imputed values were very similar to those of complete case analyses. However, it is possible that there is residual bias. For example if children with poorer health were less likely to have CD4 measurements performed, this would violate the assumption of values "missing at random" [30] and lead to an underestimation of the proportion of children starting cART with severe immunodeficiency.

| CONCLUSIONS
The results of this study represent a milestone in efforts to increase paediatric cART access worldwide. We are almost halfway there. In 2012/2013, for the first time in all country income groups, approximately half of children initiating cART did not have severe immunodeficiency. This is at least partly attributable to WHO guidelines changes that expanded paediatric cART eligibility. Nevertheless, the persistence of disparities in cART access across income group settings are also reflected in our data showing that in 2013 there were still >40% of children starting cART with severe immunosuppression in LMICs. The WHO 2016 guidelines provide an opportunity to expand paediatric cART access to infants, children and adolescents and encourage monitoring of the effect of the guidelines not only on immunological status at cART initiation, but also on retention and treatment effectiveness to comprehensively evaluate the effectiveness of paediatric cART programmes.

A C K N O W L E D G E M E N T S
We are grateful to all children, caregivers and data managers involved in the participating cohorts and treatment programmes.

F U N D I N G
The International Epidemiology Databases to Evaluate AIDS (IeDEA) collaboration is supported by the core regional IeDEA grants through the National Insti-

D I S C L A I M E R
The conclusions and opinions expressed in this article are those of the authors and do not necessarily represent the official views of any of the institutions and organizations listed above.

SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article: Appendix S1. Supplementary material. Table S1. Comparison of patients starting cART with and without CD4 cell count. Analysis of 44,480 patients included in multiple imputation and regression analyses Table S2. Decline in percentage of children starting cART with severe immunodeficiency by calendar period, reflected by (A) average change in percentage per year within each WHO guideline period and (B) rate of decrease presented as estimated slope coefficient from the segmented regression analysis. This analysis was based on complete cases (34,363 children) Figure S1. Flow chart of children included and excluded from analyses. Figure S2. Severe immunodeficiency at the start of cART by age (rows), sex (columns) and country income groups (colours). Results from generalized additive mixed effects models based on 34,363 children with complete data. 95% CIs are shown as shaded areas. Figure S3. Median CD4 cell count in children aged 5 years or older and median CD4% in children below 5 years of age at the start of cART by age (rows), sex (columns) and income group (colours). Results from generalized additive mixed effects models based on 34,363 children with complete data. 95% CIs are shown as shaded areas. Figure S4. Median age in years at start of cART by income group. Analysis based on 52,153 patients.