INTRODUCTION

The care of children and young adolescents living with HIV (CALWH) needs to be optimized to attain equity of outcomes, ensure the realization of global HIV targets, and prevent stalling of efforts in the fight against HIV/AIDS in these key populations.1 The impact of optimizing pediatric HIV care has a ripple effect into adulthood. The presence of determinants such as a low CD4+ count in the last pediatric visit before transitioning to adult care and a history of diagnosis of an AIDS-defining event while in pediatric care have been shown to increase the likelihood of poorer outcomes and death in adulthood.2 In addition, children and young adolescents are prone to greater stigma, leaving them further vulnerable to poorer outcomes and mortality.3,4

HIV disproportionately affects children and young adolescents in sub-Saharan Africa.5 A systematic review by Abuogi et al6 reported a pooled mortality rate of 0.27/100 among 5558 children living with HIV. Mutanga et al7 in Zambia found a mortality rate of 1.6/100, with most deaths occurring in the first 3 months of treatment initiation. A 10-year study by Kebede8 in Ethiopia reported a mortality rate of 5.4/100 child years. Another study in Ethiopia showed a lower mortality rate of 0.23/100 among 426 children living with HIV, with 51% of these deaths occurring in the first 2 years of treatment.9 Studies in Kenya on CALWH have reported 5%–12.8% deaths.10,11 Robust evidence on HIV-related deaths in Africa is therefore needed to guide program implementation and evaluations and inform policy reforms.12

The necessity for an assessment of predictors of mortality in CALWH cannot be belabored. Studies have observed varying factors that increase the likelihood of death in CALWH, including younger age, shorter duration of time on antiretroviral therapy (ART), high viral load, severe immunosuppression (CD4 below 350 cells/mm3), World Health Organization (WHO) clinical stage (stages 3 and 4), nutritional status, and persistent diarrhea.13–17 However, there is a paucity of studies assessing these predictors in the Kenyan demographic, leading to a lack of evidence-based data for individualized care to CALWH in Kenya. By 2019, the implementation of the first Kenya AIDS Strategic Framework (KASF) had resulted in a 64% reduction in HIV-related deaths in the country. KASF II aims at a 75% reduction in HIV-associated mortality by 2025.18 There is thus a profound need to ascertain and monitor HIV mortality and its predictors in Kenya. This will allow the country to evaluate, learn, and adjust KASF and other national HIV programs for improvement.

Pursuant to the diverse nature of the predictors of mortality in CALWH and the need for a more comprehensive approach to care for this vulnerable group, we set out to determine the predictors of death among CALWH who die while in care at Academic Model Providing Access to Healthcare (AMPATH) HIV clinics in Western Kenya, one of the largest care programs in the country.

METHODS

This was a retrospective case–control study of children and adolescents receiving care in more than 50 HIV clinics supported by the AMPATH program in Western Kenya. The program enrolled 41,688 CALWH into care from its inception in 2001 to November 2022. Data were retrieved from the AMPATH electronic medical records system. Data were included for children and young adolescents who were HIV positive, enrolled in care during the period January 2002–November 2022, aged less than 15 completed years at enrollment in care, and ever initiated ART.

The cases (N = 1049) were those recorded dead by November 2022, while the controls (N = 5185) were alive and in care, with a clinical encounter within 3 months of the end of the study time line (November 2022). Each case was matched with 5 controls, creating optimal matching by minimizing the average distance between a case and a control.19 Matching was optimized using the date of starting ART with a 28-day window period allowed and exact matching of clinic type and gender. The clinic type was classified as either rural or urban, based on the location of the clinic.

The first viral load was the viral load done 6 months post-ART initiation. Where this was unavailable, the first viral load done closest to the 6-month mark was used in the analysis. Viral suppression was defined as a viral load of less than 1000 copies/mL, and an undetectable viral load was less than 40 copies/mL. Underweight was defined as a WHO body mass index (BMI) less than the fifth percentile for age.

Variables

The outcome of interest was time to death from the start of ART. Known exposures and confounders at enrollment in care included age, sex, orphan status, school attendance, nutritional status, clinic enrolled (urban or rural), WHO clinical stage, CD4+ count, first viral load after ART initiation, and time to starting ART after enrollment in care. These constituted the independent variables.

Statistical Analysis

Descriptive statistics (counts, percentages, means, medians, and interquartile ranges) were used to summarize the demographic and clinical characteristics of the study participants. Kruskal–Wallis and Fisher exact tests were used for comparing continuous and categorical data, respectively. Where the factor levels were more than 2, Monte Carlo simulations were used to generate the P values for the test of independence between cases and controls.

The Kaplan–Meier survival curves were used to compare and describe survival probabilities across variables of interest. The log-rank test was used to compare the survival curves. Univariable and multivariable Cox proportional hazard (PH) regression models were performed using time from initiation of ART to death as the outcome. Regression adjustment was used to control for confounding using the available data. Global (multivariable analysis) and individual (univariable analysis) tests for PH assumption were performed, where globally the assumption held (P value 0.0798) and was only violated in the gender, CD4, and time to viral load test variables in the univariable analysis. Our results relied on the adjusted (multivariable) models, and hence, there was no need for changes in the univariable models that did not satisfy the PH assumption. Multiple imputation analysis was performed to account for missing data, and a pooled Cox regression was fitted on the imputed 10 sets of data. Pooled results were also used for sensitivity analysis to gauge the univariable and multivariable results under the missing at random assumption. The missing date of death for cases was imputed with dates on the last appointment attended. The control participants were right-censored at the date of their last AMPATH HIV clinic visit. Statistical significance was assessed at P = 0.05.

Ethics

This study got approval from the Institutional Research and Ethics Committee of Moi University School of Medicine, Moi Teaching and Referral Hospital, Eldoret, Kenya, and the Indiana University Institutional Review Board.

RESULTS

The analysis cohort comprised 6234 children and young adolescents, 1049 cases and 5185 controls (see Figure 1, for participant enrollment flow chart, Supplemental Digital Content 1, https://links.lww.com/QAI/C190). Slightly more than half of CALWH in this cohort were male (51.7%), and almost half were between 0 and 5 years of age (41.5%) at the start of ART. The average age at ART initiation was 6.5 years, with cases having a slightly lower average age at 6.01 (4.37) years compared with the controls at 6.62 (4.11) years. Approximately three-quarters of the cases (73.8%) were school goers before their deaths, compared with 92.3% of the controls. In this cohort, similar proportions of cases (63.9%) and controls (61.9%) attended clinics in urban areas. A high proportion of the cases (44.7%) had both parents alive, compared with the controls (24.2%) (Table 1).

TABLE 1. - Participant Demographic Characteristics Deceased Children (N = 1049, 17%) Living Children (N = 5185, 83%) Total (N = 6234) P Gender  Female 506 (48.2%) 2502 (48.3%) 3008 (48.3%) 1.000*  Male 543 (51.8%) 2683 (51.7%) 3226 (51.7%) Age group at ART start  0–5 yrs 503 (48.0%) 2083 (40.2%) 2586 (41.5%) <0.001*  6–10 yrs 317 (30.2%) 1942 (37.5%) 2259 (36.2%)  10+ yrs 229 (21.8%) 1160 (22.4%) 1389 (22.3%) Age group at enrollment (yrs)  0–5 yrs 557 (53.1%) 2585 (49.9%) 3142 (50.4%) 0.003*  6–10 yrs 296 (28.2%) 1734 (33.4%) 2030 (32.6%)  10+ yrs 196 (18.7%) 866 (16.7%) 1062 (17.0%) Child orphan status  Orphan 266 (31.4%) 2340 (56.5%) 2606 (52.3%) < 0.001*  Single parent 202 (23.9%) 799 (19.3%) 1001 (20.1%)  Parents alive 378 (44.7%) 1002 (24.2%) 1380 (27.7%)  Missing 203 1044 1247 Schooling  No 139 (26.2%) 317 (7.7%) 456 (9.8%) < 0.001*  Yes 391 (73.8%) 3808 (92.3%) 4199 (90.2%)  Missing 519 1060 1579 School goers per age group at start of ART  0–5 yrs 34 (8.7%) 1176 (30.9%) 1210 (28.8%) < 0.001*  6–10 yrs 185 (47.3%) 1603 (42.1%) 1788 (42.6%)  10+ yrs 172 (44.0%) 1029 (27.0%) 1201 (28.6%) Clinic enrolled  Rural 379 (36.1%) 1975 (38.1%) 2354 (37.8%) 0.235*  Urban 670 (63.9%) 3210 (61.9%) 3880 (62.2%)

*The Fisher exact test for count data.

The proportion of underweight children among the cases was higher than that among the controls (44.6% vs. 30.4%, P < 0.001). The mean (SD) of CD4+ cells was significantly lower among cases than controls (371.15 [407.74] vs. 498.81 [418.53]). At ART start, almost two-thirds of the cases (62.1%) had advanced disease (WHO clinical stages 3 and 4), compared with the controls (52.8%). Viral suppression was lower among the cases (22.3%) than the controls, with more than half of the controls suppressed on the first viral load (59.8%) (Table 2).

TABLE 2. - Participants' Clinical Characteristics Deceased Children (N = 1049, 17%) Living Children (N = 5185, 83%) Total (N = 6234) P WHO stage at start of ART  Stage 1 110 (14.2%) 836 (20.1%) 946 (19.2%) <0.001*  Stage 2 183 (23.7%) 1130 (27.1%) 1313 (26.6%)  Stage 3 166 (21.5%) 923 (22.2%) 1089 (22.1%)  Stage 4 314 (40.6%) 1276 (30.6%) 1590 (32.2%)  Missing 276 1020 1296 CD4 cell count/mmJ at initiation of ART  Mean (SD) 371.15 (407.74) 498.81 (418.53) 477.31 (419.40) <0.001†  Median (Q1, Q3) 229.50 (44.75, 567.25) 382.50 (188.00, 710.00) 363.50 (161.00, 691.00)  Missing 481 2381 2862 CD4 cell count/mmJ at initiation of ART  <350 mmJ 346 (60.9%) 1282 (45.7%) 1628 (48.3%) <0.001*  >350 mmJ 222 (39.1%) 1522 (54.3%) 1744 (51.7%)  Missing 481 2381 2862 CDC/WHO BMI-for-age z score classification  Healthy weight 416 (44.5%) 2602 (55.8%) 3018 (53.9%) <0.001*  Underweight 417 (44.6%) 1416 (30.4%) 1833 (32.8%)  Overweight 52 (5.6%) 305 (6.5%) 357 (6.4%)  Obese 50 (5.3%) 338 (7.3%) 388 (6.9%)  Missing 114 524 638 Time from enrollment to initiation of ART (wk)  Mean (SD) 31.22 (55.69) 50.45 (82.48) 47.22 (78.94) <0.001†  Median (Q1, Q3) 10.00 (3.00, 32.00) 15.00 (4.00, 56.00) 14.00 (4.00, 51.00) Time from enrollment to initiation of ART (mo)  One mo 365 (34.8%) 1366 (26.3%) 1731 (27.8%) <0.001*  Two mo 130 (12.4%) 612 (11.8%) 742 (11.9%)  Three+ mo 554 (52.8%) 3207 (61.9%) 3761 (60.3%) Time from initiation of ART to death/censorship (yrs)  Mean (SD) 2.33 (2.88) 6.91 (4.52) 6.14 (4.62) <0.001†  Median (Q1, Q3) 1.00 (0.00, 4.00) 7.00 (3.00, 11.00) 6.00 (2.00, 10.00) Time to viral load after ART start in mo  Mean (SD) 33.00 (25.74) 49.59 (32.23) 48.13 (32.06) <0.001†  Median (Q1, Q3) 28.00 (13.25, 48.00) 47.00 (23.00, 71.00) 45.00 (21.00, 70.00)  Missing 703 1597 2300 Time to viral load after ART start in (>48 mo)  <4 yrs 261 (75.4%) 1858 (51.8%) 2119 (53.9%) <0.001*  4 yrs + 85 (24.6%) 1730 (48.2%) 1815 (46.1%)  Missing 703 1597 2300 First-drawn viral load  Undetectable 37 (10.7%) 1567 (43.7%) 1604 (40.8%) <0.001*  Detectable 309 (89.3%) 2021 (56.3%) 2330 (59.2%)  Missing 703 1597 2300 First-drawn viral load suppression  No 269 (77.7%) 1441 (40.2%) 1710 (43.5%) <0.001*  Yes 77 (22.3%) 2147 (59.8%) 2224 (56.5%)  Missing 703 1597 2300

*The Fisher exact test for count data.

†Kruskal–Wallis rank sum test.

Kaplan–Meier Analysis

We compared survival across different variables, age groups, orphan status, BMI for age, the WHO clinical stage at commencement of ART, CD4+ count, and viral load levels (detectable vs. undetectable; suppression vs. nonsuppressed).

At the start of ART, there was no substantial difference in mortality across the different age groups. However, over time, those who started ART when they were younger than 5 years had better survival compared with those who started ART when they were older than 5 years. Total orphans had better survival rates compared with partial orphans or nonorphans. In addition, participants with a healthy weight had better survival over time than those who were underweight. (see Figure 2, with Kaplan–Meier curves by age group at starting ART, orphan status, and BMI for age z score, Supplemental Digital Content 2, https://links.lww.com/QAI/C191.)

Participants who started ART at WHO stage 1, those with higher CD4+ counts, virally suppressed CALWH, and those with undetectable first viral loads had better survival over time (see Figure 3, demonstrating Kaplan–Meier curves by WHO disease stage, CD4+ count, and viral load, Supplemental Digital Content 3, https://links.lww.com/QAI/C192).

Multivariable Cox (PH) Regression

Table 3 summarizes the results of the Cox PH model. From the pooled multivariable model, factors that were significantly associated with time to death were age at the start of ART, orphan and nutritional status, schooling, WHO clinical stages 3 and 4, CD4+ counts above 350 cells/mmJ, and viral suppression on the first viral load (viral load <1000 copies/mL).

TABLE 3. - Univariable and Multivariable Regression Results for the Predictors of Mortality Dependent: Surv (Time. yr, Case) Hazard Ratio (HR-Univariable) adjusted HR (aHR-Multivariable) HR (Multiple Imputation, Multivariate) Gender  Female — — —  Male 0.99 (0.88–1.12, P = 0.919) 1.10 (0.71–1.71, P = 0.675) 1.07 (0.86–1.33, P = 0.520) Age group at ART start  0–5 yrs — — —  6–10 yrs 0.69 (0.60–0.79, P < 0.001) 6.88 (3.18–14.92, P < 0.001) 1.59 (1.16–2.16, P = 0.004)  10+ yrs 0.86 (0.73–1.00, P = 0.053) 8.36 (3.60–19.40, P < 0.001) 2.10 (1.45–3.05, P < 0.001) Child orphan status  Orphan — — —  Single parent 2.59 (2.15–3.11, P < 0.001) 4.31 (2.37–7.84, P < 0.001) 2.39 (1.68–3.40, P < 0.001)  Parents alive 3.96 (3.38–4.64, P < 0.001) 3.06 (1.67–5.60, P < 0.001) 3.10 (2.15–4.46, P < 0.001) Schooling  No — — —  Yes 0.18 (0.15–0.22, P < 0.001) 0.12 (0.06–0.21, P < 0.001) 0.12 (0.09–0.17, P < 0.001) Clinic enrolled  Rural — — —  Urban 1.04 (0.92–1.18, P = 0.546) 0.69 (0.44–1.10, P = 0.118) 0.86 (0.68–1.08, P = 0.185) WHO stage at start of ART  Stage 1 — — —  Stage 2 1.16 (0.91–1.46, P = 0.229) 2.26 (0.95–5.34, P = 0.064) 1.48 (0.97–2.23, P = 0.066)  Stage 3 1.25 (0.98–1.59, P = 0.067) 2.63 (1.12–6.18, P = 0.027) 1.58 (1.04–2.41, P = 0.034)  Stage 4 1.70 (1.37–2.11, P < 0.001) 2.20 (0.94–5.18, P = 0.070) 1.39 (0.92–2.09, P = 0.116) CD4 cell count/mmJ at initiation of ART  <350 mmJ — — —  >350 mmJ 0.57 (0.48–0.68, P < 0.001) 0.79 (0.48–1.29, P = 0.341) 0.64 (0.44–0.93, P = 0.020) CDC/WHO BMI-for-age z score classification  Healthy weight — — —  Underweight 1.86 (1.63–2.14, P < 0.001) 1.82 (1.14–2.92, P = 0.013) 1.38 (1.06–1.79, P = 0.017)  Overweight 1.07 (0.81–1.43, P = 0.627) 2.83 (0.95–8.37, P = 0.061) 1.30 (0.83–2.05, P = 0.250)  Obese 0.91 (0.68–1.22, P = 0.534) 1.39 (0.40–4.81, P = 0.601) 0.88 (0.52–1.50, P = 0.641) Time from enrollment to initiation of ART (mo)  1 mo — — —  2 mo 0.71 (0.58–0.87, P = 0.001) 0.37 (0.19–0.74, P = 0.005) 0.59 (0.40–0.87, P = 0.008)  3 + mo 0.57 (0.50–0.65, P < 0.001) 0.49 (0.30–0.78, P = 0.003) 0.68 (0.53–0.88, P = 0.004) Time to viral load after ART start in mo by mean  <4 yrs — — —  4 yrs+ 0.24 (0.19–0.31, P < 0.001) 0.27 (0.17–0.45, P < 0.001) 0.29 (0.23–0.38, P < 0.001) First drawn viral load suppression  No — — —  Yes 0.21 (0.17–0.28, P < 0.001) 0.24 (0.14–0.40, P < 0.001) 0.25 (0.19–0.33, P < 0.001)

Holding all other factors constant, CALWH who started ART at age more than 11 years had a higher risk of mortality at 8.36 times that of those who initiated ART at age less than 5 years (aHR: 8.36 [3.60–19.40]). CALWH with advanced disease, WHO clinical stages 3 (aHR: 2.63 [1.12–6.18]) and 4 (aHR: 2.20 [0.94–5.18]) also had a higher risk of mortality when compared with those in WHO stage 1. In addition, underweight almost doubled the mortality hazard risk by 1.82 times in comparison with normal weight (aHR: 1.82 [1.14–2.92]; Table 3).

On the contrary, protective factors for mortality were school attendance, CD4+ more than 350 cells, and viral suppression. School-going CALWH had an 88% lower risk of mortality when compared with their counterparts who had not started schooling (aHR: 0.12 [0.06–0.21]). Having a CD4+ count of more than 350 cells/mmJ at ART initiation reduced the risk of mortality by 36% (aHR: 0.79 [0.48–1.29]) while CALWH who were virally suppressed on the first viral load had a 76% reduced risk of death when compared with those who were virally nonsuppressed (aHR: 0.24 [0.14–0.40]; Table 3).

DISCUSSIONS Key Results

Both parents being alive significantly increased the risk of death in this study. We found this unusual and contrary to other studies that showed a 2- to 4-fold increase in child mortality if the mother is dead or if a parent is widowed.20–22 This variation could result from reporting discrepancies whereby deaths in children with 1 or no parents alive are less likely to be registered. Alternatively, this finding could be attributable to an upscale of services and support targeting orphans and vulnerable children. Partial and total orphans are prioritized for these services and tend to get more out-of-family support. These may contribute to an increase in their quality of life.

Children and adolescents who were school goers had lower chances of death. Although studies have reported higher rates of unintended disclosure of HIV status and stigma in schools with consequent worse outcomes in school goers, children in boarding schools are disproportionately affected than children in day schools.3,23 In Kenya, most school goers attend day schools, and therefore, CALWH staying at home while attending school may adhere more to treatment owing to their wake-up routine and a more conducive environment at home to take their medications when compared with those in boarding schools. In addition, caregiver support at home improves adherence and outcomes in pediatric HIV.24 A study in Kenya reported school attendance as a significant factor for retention in care for CALWH.25 Furthermore, there is a rise in advocacy for interventions in boarding schools such as long-interval refilling of ART medications to curb the number of missed clinic days and missed doses and improve adherence and overall quality of life of CALWH who attend boarding schools.26,27

Underweight CALWH also had higher chances of dying than those with healthy weight. Malnutrition can cause systemic inflammation and impaired cellular immune responses.28 The diminished immune response is somewhat synergistic with the action of HIV, and the dual depletion of the body's immune response could explain the higher mortality observed in underweight CALWH. When HIV and malnutrition occur concurrently, the risk of developing other diseases such as tuberculosis also rises, fostering poorer outcomes such as death. Other studies have also reported a higher number of deaths in malnourished children and recommended incorporating nutritional services into pediatric HIV care.29–31

In this cohort, children and adolescents with a CD4+ count of less than 350 cells had a higher mortality risk. This high risk could be because there is a longer lag time for CD4+ cell counts to increase after attaining critically low levels, prolonging the probability of getting new or recurrent AIDS-defining events despite adequate ART. This worsens an already vulnerable state. Therefore, we advocate early diagnosis and initiation on ART to prevent advanced disease that presents with low CD4+ counts in CALWH. In a multinational study of 17 countries, low CD4+ count was also a predictor of mortality, and the risk was highest in the first 6 months of treatment initiation.14 Siika e al32 demonstrated a direct relationship between CD4+ levels and mortality and recommended an increase in same-day ART initiation and intensification of follow-up efforts.

The risk of death was linearly associated with the WHO clinical stage of disease, and the highest risk of death was in advanced disease (WHO stages 3 and 4). This finding goes hand in glove with low CD4+ count and a greater risk of opportunistic infection seen with advanced disease. Similar results of higher mortalities in advanced disease stages have been reported by other authors.7,33,34 These point to a continued need for intensified early infant diagnosis, early initiation on ART, and increased retention in care to curb HIV-associated mortality. In a bid to increase early infant diagnosis in Kenya and to achieve the 2025 targets of the Kenya AIDS Strategic Framework II, HIV diagnostic guidelines in children below 18 months have been updated to include birth DNA PCR in addition to that performed at 6 weeks or first contact, at 12 months, and at 18 months of life.35

Children and adolescents aged 11 years and older at the start of ART were most likely to die, inferable to the fact that older perinatally infected children and adolescents have lived with HIV infection for a long time. Therefore, they may have been diagnosed late with advanced stages of disease. In addition, earlier guidelines for HIV care mandated the use of thresholds of CD4+ for ART initiation, further increasing the risk of starting ART with already advanced disease as opposed to current guidelines that advocate for commencing ART once a child is diagnosed with HIV. The Kaplan–Meier analysis complemented these findings and demonstrated better survival in children who started ART before 5 years of age. Viral load analysis revealed mirror findings where the children and adolescents who died had significantly higher viral loads. A multiregional cohort study across 31 countries reported slow progress in attaining viral suppression targets among children and adolescents.36 In Kenya, guidelines have evolved to upscale routine viral load monitoring and increase the number of virally suppressed children and adolescents. The National AIDS and STI's Control Program in Kenya now recommends monitoring of viral loads to be done 3 months after commencing treatment and subsequently every 6 months for children, adolescents, and youth between 0 and 24 years, or for those testing HIV positive in pregnancy.37 In addition, the first viral load test should be performed 3 months after enhanced adherence counseling for children and adolescents with low-level viremia and suspected treatment failure before switching regimens.37 Other studies have shown analogous findings of higher mortality with increasing age and viral nonsuppression.17,38,39 On the contrary, another study demonstrated that the highest number of deaths in children with HIV occurred in those below years.15 In a nutshell, viral suppression is regarded as the cornerstone of the success of ART