Mortality and losses to follow‐up among adolescents living with HIV in the IeDEA global cohort collaboration

Abstract Introduction We assessed mortality and losses to follow‐up (LTFU) during adolescence in routine care settings in the International epidemiology Databases to Evaluate AIDS (IeDEA) consortium. Methods Cohorts in the Asia‐Pacific, the Caribbean, Central, and South America, and sub‐Saharan Africa (Central, East, Southern, West) contributed data, and included adolescents living with HIV (ALHIV) enrolled from January 2003 and aged 10 to 19 years (period of adolescence) while under care up to database closure (June 2016). Follow‐up started at age 10 years or the first clinic visit, whichever was later. Entering care at <15 years was a proxy for perinatal infection, while entering care ≥15 years represented infection acquired during adolescence. Competing risk regression was used to assess associations with death and LTFU among those ever receiving triple‐drug antiretroviral therapy (triple‐ART). Results Of the 61,242 ALHIV from 270 clinics in 34 countries included in the analysis, 69% (n = 42,138) entered care <15 years of age (53% female), and 31% (n = 19,104) entered care ≥15 years (81% female). During adolescence, 3.9% died, 30% were LTFU and 8.1% were transferred. For those with infection acquired perinatally versus during adolescence, the four‐year cumulative incidences of mortality were 3.9% versus 5.4% and of LTFU were 26% versus 69% respectively (both p < 0.001). Overall, there were higher hazards of death for females (adjusted sub‐hazard ratio (asHR) 1.19, 95% confidence interval (CI) 1.07 to 1.33), and those starting treatment at ≥5 years of age (highest asHR for age ≥15: 8.72, 95% CI 5.85 to 13.02), and in care in mostly urban (asHR 1.40, 95% CI 1.13 to 1.75) and mostly rural settings (asHR 1.39, 95% CI 1.03 to 1.87) compared to urban settings. Overall, higher hazards of LTFU were observed among females (asHR 1.12, 95% CI 1.07 to 1.17), and those starting treatment at age ≥5 years (highest asHR for age ≥15: 11.11, 95% CI 9.86 to 12.53), in care at district hospitals (asHR 1.27, 95% CI 1.18 to 1.37) or in rural settings (asHR 1.21, 95% CI 1.13 to 1.29), and starting triple‐ART after 2006 (highest asHR for 2011 to 2016 1.84, 95% CI 1.71 to 1.99). Conclusions Both mortality and LTFU were worse among those entering care at ≥15 years. ALHIV should be evaluated apart from younger children and adults to identify population‐specific reasons for death and LTFU.


| Ethics review
Each region secured local regulatory approvals for participation in this analysis, including reviews by local research and ethics regulatory bodies and, where required, national-level approvals. Consent and assent requirements and procedures were regulated by the local regulatory bodies, and adherence to those standards was the responsibility of each site while being monitored and managed by regional coordinating centres. (https://www.iedea.org/regions/) [7][8][9].

| Definitions and measurements
Adolescents were defined as those 10 up to 19 years of age and the analysis focused on this period of life. The beginning of followup, referred to as the "baseline" time point, was the date of the 10th birthday for those who entered care before age 10, and the date of the first clinic visit for those who entered care at or after age 10. Follow-up time ended and data were censored at whichever of the following came first: (1) death, (2) transfer out, (3) loss to follow-up (LTFU), (4) turning 19 years of age, or (5) the closing date of the individual regional cohort database. The main outcomes of interest were death, LTFU and transfer that occurred in the year following the last clinic visit during the period of adolescence.
Specifically, adolescents without evidence of contact with the clinic for more than 12 months were classified as LTFU with their follow-up period ending 12 months after their last clinic contact. In addition, if a patient previously considered LTFU was subsequently known to have died or have been transferred (e.g. through updated reporting by their clinic), their outcomes were revised up to 24 months after their last clinic contact (and not beyond turning age 19 or database closure).
HIV disease stage was categorized as asymptomatic (CDC N or WHO 1), mild (CDC A or WHO 2), moderate (CDC B or WHO 3) and severe (CDC C or WHO 4). Weight and height measurements were converted to age-and sex-adjusted z-scores. For weight-forage z-scores, US National Center for Health Statistics and WHO International Growth Reference standards were used to allow for scoring children >10 years of age [10,11]. For height-for-age zscore we used the WHO 2006/2007 Child Growth Standards [12,13]. Severe immunodeficiency was defined according to 2006 WHO global guidelines (e.g. <15% or <200 cells/mm 3 for children ≥5 years old) [14].
For laboratory and clinical measurements, we used the closest values reported during a window of plus or minus three months from the baseline visit (i.e. at age 10 or the date of the first visit if entering care after age 10), with the pre-baseline measurement used in the case of multiple values. At antiretroviral initiation, we used a testing window of three months before and one week after start (e.g. for CD4, viral load).

| Statistical analysis
The analysis was restricted to assess outcomes during the period of adolescence. Adolescents entering care before age 15 years were compared to those with a first visit at or after age 15. Entry into care before age 15 was considered a proxy for those likely to have been infected perinatally or very early in life compared to those infected in older adolescence, predominantly assumed to be through risk behaviours and called the "late-infected" [15][16][17]. The term "late-infected" was chosen to characterize the timing of HIV infection relative to the stage of adolescence (i.e. between 15 and 19 years of age). The selection of the age threshold is consistent with UNAIDS Global AIDS Monitoring methods where those 10 to 14 years of age are considered to be in early adolescence and those 15 to 19 years in late adolescence. In addition, infections acquired through risk behaviours are not modelled in the UNAIDS Spectrum model to occur among those entering HIV care before the age of 15 years [6]. We conducted a sensitivity analysis to examine the impact of differentiating patients by age <10 versus ≥10 years at entry into care as a comparison [18].
To compare proportions, we used chi-square tests, and we compared medians with the Mann-Whitney test. We used a cumulative incidence function to estimate the probabilities of death and LTFU during adolescence. In a subset of adolescents who had received ≥3 antiretrovirals as their initial treatment regimen, we conducted separate competing risks regression analyses based on Fine and Gray's proportional sub-hazards model [19] to identify correlates of death (LTFU as a competing event) and correlates of LTFU (death as a competing event) from the start of triple-drug ART. The following variables were included in the univariate analysis: age and calendar year at first triple-drug ART, sex, facility level and facility setting (as defined by the site). CD4 count and weight-for-age z-scores were included as time-updated variables, and their values were carried forward if no subsequent measurements were recorded. In regression analyses, missing data were modelled as a separate category within each variable. We did not use multiple imputation to model missing data due to the relatively small numbers of covariates available, and the resulting lack of precision in imputation. To assess the robustness of our analyses to missing data, we undertook a sensitivity analysis based on subsets of patients with complete data. Variables were included in the multivariate model if they had a p < 0.2 in univariate analysis. We selected the final model using a backward elimination procedure and retained all variables in the model that had a p < 0.05. The adjusted subdistribution hazard ratios (asHR) were reported with their 95% confidence intervals (95% CI).
Management of the multiregional aggregated data and statistical analyses were performed at the Kirby Institute, UNSW Australia, using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and Stata (StataCorp, STATA 14.0 for Windows, College Station, TX, USA).
At baseline (age 10 years or first clinic visit if entering care later), antiretrovirals had already been started by 41% of perinatally infected adolescents and 0.9% of late-infected adolescents. The median CD4 count was 435 (IQR 196 to 745) cells/mm 3 for those perinatally infected and 329 (IQR 178 to 521) cells/mm 3 for the late-infected; higher proportions of perinatally infected were severely underweight (17% vs. 11%, p < 0.001) and severely stunted (13% vs. 4.5%, p < 0.001) with z-scores <À3 compared to lateinfected adolescents. At baseline, HIV viral load was infrequently available in both groups (19% vs. 5.4%). Of those with a viral load measurement at baseline, 38% of the perinatally infected and 22% of the late-infected were undetectable.

| Patient outcomes
During the adolescent follow-up period, 3.9% died, 30% were LTFU and 8.1% were transferred. Separated by age at cohort entry, among those entering care before age 15 years, 4.0% died, 27% were LTFU, and 9.7% transferred (Table S2). For those entering care at or after age 15 years, 3.8% died, 38% were LTFU and 4.6% were transferred. These data include 62 adolescents living with HIV (ALHIV) who were recategorized from LTFU to dead and 98 recategorized to transferred between 12 and 24 months after most recent clinic contact (and before age 19 or database closure). Among perinatally infected adolescents, the four-year cumulative incidence of death was 3.9% and of LTFU was 26%, while for late-infected adolescents it was 5.4% for death and 69% for LTFU; both outcomes were significantly higher in late-infected adolescents (p < 0.001) ( Figure 2).
In the sensitivity analysis where age at entry into care was redefined as <10 years versus ≥10 years, the median CD4 at baseline for the group entering care at <10 years of age (n = 22,168) was 673 (IQR 429 to 949) cells/mm 3 . The proportion of adolescents entering care before age 10 years who died was 1.9%, compared to 4.0% using the age 15 threshold, and the proportion who were LTFU was 18% compared to 27% (Table S3). In addition, of the 19,970 adolescents entering care between age 10 and 14 years, 1247 (6.2%) died and 7469 (37%) were LTFU by the age of 19 years. The multivariate regression model restricted to those who received triple-drug ART as their initial antiretroviral regimen showed that there was an increase in the hazard rate of death for those starting treatment at older ages compared to those <5 years of age (5 to 9 years adjusted subdistribution hazard ratios asHR 2.59, 95% CI 1.74 to 3.85; 10 to 14 years asHR 6.93, 95% CI 4.69 to 10.22; ≥15 years asHR 8.72, 95% CI 5.85 to 13.02) ( Table 2). The hazard was higher for females (asHR 1.19, 95% CI 1.07 to 1.33), and those receiving care in mostly urban (asHR 1.40, 95% CI 1.13 to 1.75) and mostly rural settings (asHR 1.39, 95% CI 1.03 to 1.87) compared to urban settings. The hazard of death was lower for those with higher CD4 count, better weight-for-age z-scores, receiving care at a district hospital and in rural settings compared to health centres and in urban settings, with a later year of starting ART, and for cohorts from the Asia-Pacific, Central Africa, East Africa, and South Africa compared to Southern Africa. Hazard rates of death were lowest overall among adolescents with a current CD4 ≥500 cells/mm 3 compared to <200 cells/ mm 3 (asHR 0.12, 95% CI 0.10 to 0.15), a weight-for-age zscore ≥À2 compared to <À3 (asHR 0. 22 Increased hazard rates of LTFU among adolescents who received triple-drug ART as their initial antiretroviral regimen were associated with female sex (asHR 1.12, 95% CI 1.07 to 1.17), older age at ART start compared to <5 years (5 to 9 years asHR 2.59, 95% CI 2.32 to 2.88; 10 to 14 years asHR 6.11, 95% CI 5.49 to 6.81; ≥15 years asHR 11.11, 95% CI 9.86 to 12.53), receiving care at a district hospital compared to a health centre (asHR 1.27, 95% CI 1.18, 1.37), receiving care in rural compared to urban settings (asHR 1.21, 95% CI 1.13, 1.29), receiving care in East Africa (asHR 1.14, 95% CI 1.01 to 1.28), South Africa (asHR 1.75, 95% CI 1.63 to 1.88), or CCA-SAnet (asHR 2.99, 95% CI 2.65 to 3.36) compared to Southern Africa, and starting triple-drug ART after 2006 (highest asHR for 2011 to 2016 1.84, 95% CI 1.71 to 1.99) ( Table 3). In contrast, lower hazard rates of LTFU were associated with CD4 count ≥350 cells/mm 3 (lowest asHR for ≥500 0.65, 95% CI 0.61 to 0.69), receiving care in regional, provincial, or university hospitals compared to health centres (asHR 0.63, 95% CI 0.58 to 0.68), receiving care in mostly urban and mostly rural compared to urban settings (lowest asHR for mostly urban 0.71, 95% CI 0.62 to 0.81), and receiving care in the Asia-Pacific compared to Southern Africa (asHR 0.19, 95% CI 0.14 to 0.25).
The sensitivity analyses excluding missing data gave qualitatively very similar results for most covariates (data not shown). The only exceptions were in mortality analyses, where survival was no longer improved in Central Africa (asHR in

| DISCUSSION
This is the first IeDEA multiregional cohort analysis to reflect the complexity of the mixed adolescent HIV epidemic including both individuals with perinatally acquired HIV and those infected later. Among the 61,242 adolescents in our analysis, 69% entered care before the age of 15 (our proxy for perinatally acquired infection), but not until a median age of 9.8 years.
While the cumulative incidence of both mortality and LTFU were higher among those entering care at ≥15 years (our proxy for infection acquired later during adolescence), the qualitative differences in mortality over this period were small. However, the sensitivity analysis demonstrated a higher burden of mortality among those perinatally infected adolescents who did not enter care until 10 to 14 years of age. Our overall four-year cumulative incidence of death of 4.2% compares to the post-ART mortality incidence rate of 0.97 per 100 person-years among children five to nine years of age in IeDEA [20], and rates among youth starting ART during the ages of 15 to 24 years from 0.8 per 100 in Nigeria up to 13.5 per 100 in Tanzania in a seven-African country analysis [2]. From our regression model, most factors found to be protective against death among those who started treatment with triple-drug ART were consistent with other studies (e.g. better immune control, higher weight-for-age z-score) [4,21,22]. Any Data are presented as n (%) unless otherwise noted. We used the 1977 WHO growth curve for weight-for-age z-score (more recent weight curves are limited to children age ≤10 years) and the 2006/2007 WHO growth curve for height-for-age z-score. 3-ART, antiretroviral therapy regimen of three or more antiretrovirals; IQR, interquartile range; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor; mono/dual, single or two drugs; SD, standard deviation. a Baseline was the date of the 10th birthday for those who entered care before age 10, and the date of the first clinic visit for those who entered care at or after age 10. b IeDEA regions utilize a common data exchange standard for harmonizing data for use in multiregional analyses that includes formats and categorizations for specific data variables that are available at iedeades.org. c This only includes those who were on antiretrovirals at baseline. Those who started before baseline and stopped before baseline and those who started antiretrovirals after baseline were not included. d 3-ART represents triple-drug regimens. The drug class following that term denotes where one of the drugs included either an NNRTI, PI, or both classes; "other" represents triple-drug regimens without an NNRTI or PI. Non-3-ART represents regimens with fewer than three individual antiretroviral drugs. e Includes other triple-drug and mono/dual antiretroviral combinations. f Duration has been calculated for adolescents who started antiretrovirals before baseline and were still on them at baseline.
CD4 category ≥200 cells/mm 3 or weight-for-age z-score better than or equal to À3 was highly protective; as was starting triple-drug ART after 2011, which may reflect scale-up and quality improvement of paediatric HIV programmes and broadening treatment access in our settings [23,24]. Among those who had not received treatment, the median duration of follow-up during adolescence was only one year and 52% were LTFU, making reliable ascertainment of mortality difficult in this sub-group. The associations with regional cohort may reflect variations in national infrastructures for HIV and availability of other supportive healthcare services in the context of background country development [2,25-27]. The overall high cumulative incidence of LTFU was concerning. Losses among those presenting to care during late adolescence rose sharply starting in the first year of follow-up, whereas losses among the perinatally infected steadily increased over time. The four-year cumulative incidence of LTFU during adolescence was 26% for the perinatally infected and almost three times that at 69% for the late-infected. This compares to data from adolescents and young adults 15 to 24 years of age at treatment initiation in seven African countries, where LTFU ranged from 7.1% in Uganda to 30% in Tanzania [2]. The rapid early losses are also consistent with levels of LTFU, approximately 30%, documented in prevention of mother-to-child HIV transmission Option B+ programmes [28,29]. While we did not have access to pregnancy data to confirm whether young women entered HIV care through antenatal care, being female was a risk factor for LTFU. While current universal treatment recommendations may reduce the early LTFU that was previously associated with delays related to CD4 testing [30][31][32], there are also studies among adolescents and adults reporting greater attrition among those started on ART at higher CD4 counts who have not experienced clinical disease progression [33][34][35].
The associations between LTFU and older age, as well as later year at cohort entry, may be related to having less time to return to care after an interruption (i.e. patient churn), survival bias among those who started treatment as younger children, or the poorer retention often seen among older adolescents and young adults [2,[36][37][38][39]. Receiving care in rural settings was associated with higher LTFU, but protective against death, which may suggest under ascertainment of mortality in rural areas where out migration in sub-Saharan Africa has been common [40,41]. The reasons for the associations with regional cohort are unclear, and could be due to varying proportions of perinatally infected youth, local patient case mix, or other socioeconomic or demographic factors [18,28,42,43]. In addition,   regional cohorts with sites within areas with a high density of ART programmes may see more "silent transfers, " where patients move between clinics without formal referrals. The median baseline CD4 count for the group entering care between 15 and 19 years was unexpectedly low at 329 cells/ mm 3 , implying that there may have been perinatally infected adolescents mixed into this group who entered care at older ages with advanced HIV disease. This may reflect the health status of older slow progressors among perinatal adolescent survivors or the potential contribution of rapid disease progressors among those infected during adolescence. Age alone may be insufficient to avoid misclassification, and more complex algorithms to assess combinations of routinely available variables (e.g. weight, height and CD4 count) would help to disaggregate data [15].
Our analysis was limited by the use of routinely collected clinical data, which were incomplete. We included children with missing data in analyses using missing value categories in covariates, which has the advantage of maximising the sample size. Sensitivity analyses which excluded missing data gave very similar results, but did result in some changes in differences in survival between regions and facility setting. These site-level covariates should be interpreted especially cautiously. Beyond variations in clinical resources and programme policies, some of the regions had more older female adolescents enrolled, which may have been associated with antenatal care programmes where early LTFU rates have been high [28]. Our focus on those 10 to 19 years of age results in a survivor bias, as individuals who were perinatally infected would have had to survive childhood in order to be eligible for inclusion. In addition, our age proxy for perinatally acquired infection may have miscategorized those presenting very late to care as being infected during adolescence. Restricting inclusion in the analysis to those with at least six months of data may have resulted in underreporting of LTFU. While we allowed for randomly collected tracing data to reclassify outcomes, we did not systematically adjust for the risk of unascertained mortality among those LTFU.

| CONCLUSIONS
In this global analysis of adolescent outcomes in IeDEA, 3.9% of ALHIV were reported to have died and 30% were LTFU, with both rates higher in those entering care after age 15. However, those entering care between 10 and 15 years were at higher risk of death than those in care before age 10, reflecting the severe immunodeficiency associated with delayed diagnoses. Greater prioritization of adolescents for clinical and social support is urgently needed to retain youth in HIV treatment programmes as they transition through adolescence into adult life.   AA, KM, DMM, and AHS were involved with the collection, preparation, and/or submission of the source data for the analysis. AK and ML conducted the analysis. AK and AHS wrote the paper. All co-authors were involved with revisions of the paper, and read and approved the final version.

A C K N O W L E D G E M E N T S
We thank the adolescents, caregivers, and staff at our participating clinics who inspire and support our work. Additional appreciation goes to IeDEA's paediatric and adolescent investigators, the IeDEA Pediatric Working Group, regional data managers, and project managers, and the IeDEA-WHO collaboration.

F U N D I N G
Funding for this work is provided through grants from the US National Institutes of Health's National Institute of Allergy and Infectious Diseases, the Eunice Kennedy Shriver National Institute of Child Health and Human Development, the National Cancer Institute, the National Institute of Mental Health, and the National Institute on Drug Abuse: U01AI069907 (Asia-Pacific); U01AI069923 (CCASAnet); U01AIQI096299 (Central Africa); U01AI069911 (East Africa); U01AI069924 (Southern Africa); U01AI069919 (West Africa). Complete investigator lists and regional acknowledgements are in the Appendix.

SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article: Caribbean, Central, and South America (CCASAnet)