Effect of HIV infection on growth and bone density in peripubertal children in the era of antiretroviral therapy: a cross-sectional study in Zimbabwe

Background: Faltered linear growth and pubertal delay, which are both common in children with HIV in sub-Saharan Africa, might affect adolescent bone accrual and future fragility fracture risk. We investigated the association of HIV with bone density adjusted for skeletal size in peripubertal children in Zimbabwe. Methods: We did a cross-sectional study of baseline data from the IMVASK cohort, which enrolled children aged 8–16 years with HIV who had been taking antiretroviral therapy (ART) for at least 2 years, and children of the same age without HIV. Children with HIV were recruited from public sector HIV clinics at Parirenyatwa General Hospital and Harare Central Hospital (Harare, Zimbabwe), and children without HIV were recruited from six schools in the same suburbs that the hospitals serve. Sociodemographic, clinical, and anthropometric data were collected. Dual-energy X-ray absorptiometry (DXA) was used to measure the bone outcomes of total-body less-head bone mineral content for lean mass adjusted for height (TBLH-BMCLBM), and lumbar spine bone mineral apparent density (LS-BMAD), and we assessed the prevalence of low TBLH-BMCLBM and low LS-BMAD (defined by Z-scores of less than −2·0). Size adjustment techniques were used to overcome the size dependence of DXA measurement. We used linear regression models, with multiple imputation for missing data, to assess relationships between risk factors and TBLH-BMCLBM and LS-BMAD Z-scores in children with and without HIV. Findings: We recruited 303 children with HIV (mean age 12·4 years [SD 2·5]; 151 [50%] girls) and 306 children without HIV (mean age 12·5 years [SD 2·5]; 155 [51%] girls). In children with HIV, median age of HIV diagnosis was 3·0 years (IQR 1·2–5·8), and median ART duration was 8·1 years (6·2–9·5); for 102 (34%) children, ART included tenofovir disoproxil fumarate (TDF). Children with HIV had a higher prevalence of low TBLH-BMCLBM Z-score than children without HIV (29 [10%] of 279 children with available data vs 18 [6%] of 292 with available data; p=0·066) and a higher prevalence of low LS-BMAD Z-score (40 [14%] of 279 vs 17 [6%] of 293 with available data; p=0·0007). HIV and male sex were associated with earlier pubertal (Tanner) stage. The negative associations between HIV and Z-scores for TBLH-BMCLBM and LS-BMAD were more pronounced with pubertal maturation, particularly in girls. Among children with HIV, TDF exposure and orphanhood were associated with lower TBLH-BMCLBM Z-score in confounder-adjusted analysis. Current TDF use (vs non-TDF-based ART) was associated with a reduction in TBLH-BMCLBM Z-score of 0·41 (95% CI 0·08–0·74; p=0·015) and in LS-BMAD Z-score of 0·31 (0·08–0·69; p=0·12). Interpretation: Despite ART, HIV is associated with substantial skeletal deficits towards the end of puberty. The extent of bone deficits associated with TDF and its widespread use in children in sub-Saharan Africa are a concern for future adult fracture risk. Funding: Wellcome Trust. We did a cross-sectional study using baseline DXA bone measurements from the IMVASK study, which is a prospective cohort study on the impact of vertical HIV infection on child and adolescent skeletal development in Harare, Zimbabwe. The IMVASK protocol has been published elsewhere (ISRCTN registry ISRCTN12266984)11 and 12-month follow-up has been completed. Children aged 8–16 years with HIV were recruited from outpatient HIV clinics at the only two public sector general hospitals in Harare (Parirenyatwa General Hospital and Harare Central Hospital). Studies in children suggest that ART initiation is followed by an initial decrease in bone mass, which stabilises after 2 years,12 and therefore we enrolled children with HIV who had been taking ART for at least 2 years. Systematic quota-based sampling, stratified by age and sex, was used to recruit 50 male children and 50 female children in each of three age groups (8–10 years, 11–13 years, and 14–16 years). Exclusion criteria were being acutely unwell (defined as requiring immediate hospitalisation), not residing in Harare, and being unaware of one’s HIV status (to avoid inadvertent disclosure during study participation). A maximum of five children with HIV were recruited each day for logistical reasons. A comparison group of children without HIV was recruited from six government primary and secondary schools randomly selected from the 109 primary schools and 44 secondary schools within the same suburbs in Harare where the hospitals provide HIV care. A random number sequence was computer-generated and applied to a list provided by the Ministry of Primary and Secondary Education of all schools in the area. Schools were approached in sequence to seek consent. Schools which declined to participate were replaced by schools on a reserve list until the target of six schools was reached. Younger children (8–12 years) were sampled from primary schools and older children (14–16 years) from secondary schools, with children aged 13 years sampled from both schools. The number of children selected from each school was proportional to school size, thereby giving each child equal probability of being sampled. We applied a random number sequence to school registers to select participants using the same quota-based sampling approach of 50 male children and 50 female children in each of the three age strata. Letters were sent to the households of children who had been randomly selected. Children underwent HIV testing after enrolment; those testing positive and not in care were referred to HIV services. Children in schools who tested positive were considered for enrolment in the HIV cohort. Ethical and governance approvals were granted by the ethics committee of the London School of Hygiene & Tropical Medicine (London, UK; reference 15333), the institutional review board of the Biomedical Research and Training Institute in Harare (reference AP145/2018), the joint research ethics committee for the University of Zimbabwe College of Health Sciences (Harare) and the Parirenyatwa Group of Hospitals (Harare; reference 11/18), the Harare Central Hospital ethics committee (reference 170118/04), the Medical Research Council of Zimbabwe (Harare; reference MRCZ/A/2297), and the Ministry of Primary and Secondary Education of the Government of Zimbabwe (Harare; reference C/426/Harare). Parents or guardians provided written informed consent for study participation and HIV testing, and children provided written assent. All data in this study were collected at recruitment (baseline) in the IMVASK study, from May 4, 2018, to Jan 21, 2020. A questionnaire administered by research staff was used to collect sociodemographic and clinical data, including smoking status and alcohol and steroid use, from children in the company of a parent or guardian (with parents and guardians allowed to answer questions on the child’s behalf). The International Physical Activity Questionnaire, validated in multiple countries including South Africa but not Zimbabwe, was used to assess physical activity as multiples of the resting metabolic rate (MET) in MET minutes. Diet and nutrition were assessed with a tool that was based on a validated dietary diversity and food frequency tool from India and Malawi,13 and adapted to the Zimbabwean context with international guidelines applicable to sub-Saharan Africa.14 This tool quantified dietary calcium and vitamin D intake plus sunlight exposure; adaptations reflected the Zimbabwean context in which fortification of oils and margarine with vitamin D is mandated and specific vitamin D rich foods, such as Kapenta fish, are commonly eaten. Anthropometric measurements were done by trained research nurses and research assistants at the study clinics. Standing and sitting height, which was measured to the nearest 0·1 cm (with a Seca 213 stadiometer; Seca, Hamburg, Germany), and weight, which was measured to the nearest 0·1 kg (with Seca 875 weight scales), were recorded by two separate readers. If height measurements differed by more than 0·5 cm, or weight measurements by more than 0·5 kg, a third reading was taken by an additional reader, and final height and weight values were taken as means of the two or three measurements. The same staff measured the children with and without HIV. All equipment was calibrated annually. Tanner pubertal staging was done by a nurse and doctor, with an orchidometer used to assess testicular volume in males. Pubertal delay was defined as not having reached Tanner stage 2 in girls aged 13 years or older, and in boys aged 14 years or older. Details collected for participants with HIV were age at HIV diagnosis, probable mode of transmission, ART regimen and duration, and current CD4 cell count and HIV viral load. CD4 cell count was measured with an Alere PIMA CD4 Analyser (Waltham, MA, USA) and HIV viral load with the GeneXpert HIV-1 Viral Load assay (Cepheid, Sunnyvale, CA, USA), with viral suppression defined as fewer than 1000 HIV RNA copies per ml (as per WHO guidelines). DXA scans of the lumbar spine and total body were done by one of two trained radiographers according to standard procedures on a Hologic QDR Wi densitometer (Hologic, Bedford, MA, USA) with Apex software (version 4.5)15 for scan analysis. Daily calibration was done with the manufacturer-provided spine phantom. DXA scans were repeated in a subgroup (n=30) selected by convenience sampling to confirm reproducibility. The precision error was a root mean square deviation of 0·011 g/cm2 (lumbar spine) and 0·010 g/cm2 (total body) with a coefficient of variation of 1·35% (lumbar spine) and 1·22% (total body). As mentioned, an important limitation of DXA in paediatric populations with chronic disease is that the two-dimensional (areal) bone density values are highly dependent on bone size; thus DXA underestimates bone density in small children.10 We therefore used the two main size-adjustment techniques recommended by the International Society for Clinical Densitometry (ISCD) to overcome the size dependence of DXA measurement:16 we measured total-body less-head bone mineral content for lean mass adjusted for height (TBLH-BMCLBM), and lumbar spine bone mineral apparent density (LS-BMAD). LS-BMAD was calculated from DXA-measured lumbar spine data with the Carter method.17 TBLH-BMCLBM was calculated from the whole body scan with published derived equations for Hologic DXA scans, which adjust for log-transformed total body lean mass, total body fat mass, height, and age.18 Sex-matched and age-matched Z-scores were generated with Hologic UK population reference data from 1996–2012 in 4–20-year-olds as recommended by ISCD guidelines, as no local reference data were available.16 Low TBLH-BMCLBM and LS-BMAD were defined by a Z-score of less than −2·0.18 A sample size of 300 in each group was required to permit detection of a difference between children with and without HIV in DXA-measured size-adjusted bone density Z-scores of 0·23, with 80% power and a significance level of 0·05 assuming an SD of 1·3.19 The study had 80% power to detect a 4·8% difference in prevalence of low TBLH-BMCLBM between children with and without HIV, assuming a prevalence in those without HIV of 1·0%.20 Height-for-age and weight-for-age Z-scores were calculated with 1990 UK reference data,21 with Z-scores of less than −2·0 defining stunting (height-for-age score) and underweight (weight-for-age score). We derived socioeconomic status using the first component from a principal component analysis combining an asset list (detailing number in the household, head of household age, highest maternal and paternal education levels, household ownership, monthly household income, access to amenities [electricity, water, and a flush toilet or pit latrine], and household item ownership [fridge, bicycle, car, and television or radio]). Socioeconomic status was split into tertiles (low, middle, and high) for analysis. Analyses were done with Stata (version 16.1). The primary exposure was HIV, and primary outcomes were TBLH-BMCLBM and LS-BMAD Z-scores. We compared participant characteristics between those with and without HIV using independent sample t-tests for means, Wilcoxon signed-rank tests for non-parametric variables, and χ2 or Fisher’s exact tests for proportions. The same methods were used to compare children with and without missing data. To first understand the role of sex, puberty, and HIV on bone outcomes, we examined mean difference in TBLH-BMCLBM and LS-BMAD Z-scores between participants with and without HIV using linear regression with robust standard errors, overall and stratified by sex, Tanner stage (stages 1 and 2 vs stages 3–5), and age group (8–10 years, 11–13 years, and 14–16 years). Similarly, we examined mean risk difference for low TBLH-BMCLBM and low LS-BMAD Z-scores (less than −2 vs −2 or higher) using generalised linear models, with Poisson distributions and log links with robust standard errors. The models were adjusted for age, sex, and pubertal stage. Mean differences were estimated with 95% CIs. Three-way interactions between sex, pubertal stage, and HIV were assessed with Wald tests in linear regression models. In secondary analyses of absolute measures of TBLH-BMCLBM and LS-BMAD, generalised linear models with log link and gamma distribution were used, and marginal means and marginal mean differences with 95% CIs estimated. Associations between potential risk factors and TBLH-BMCLBM and LS-BMAD Z-scores were investigated with linear regression separately for participants with and without HIV. Adjustment was made for a priori confounders (age, sex, and pubertal stage22), potential risk factors (socioeconomic status,23 physical activity,24 calcium and vitamin D intake,25 and, in those with HIV, CD4 cell count, viral load, tenofovir disoproxil fumarate [TDF] exposure, and age at ART initiation26), variables associated in complete case analysis with TBLH-BMCLBM or LS-BMAD Z-score (at p<0·20), and variables associated with missing status. All enrolled children were included in the analyses. To account for missing data, including DXA-measured outcomes, we used multiple imputation by chained equations with seven imputed datasets, which allowed for imputation of categorical and continuous data jointly. It was assumed that data were missing at random. Our imputation models included all outcomes, auxiliary variables associated with missingness and with study group (with or without HIV), and variables identified in complete case analysis to be associated with TBLH-BMCLBM or LS-BMAD Z-score (p<0·20). In all analyses, the significance level was 0·05. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report, or in the decision to submit the paper for publication.