Improving Methods to Classify Perinatal Vs. Non-Perinatal HIV Acquisition in Young Adolescents 10–14 Years of Age

Abstract

Mode of HIV acquisition for adolescents with HIV is often not recorded within routine healthcare databases. Hence, age at enrolment in HIV care is often used as a proxy for perinatal vs. non-perinatal infection. Using routine cohort data from adolescents presenting for HIV care 10–14 years of age, we developed logistic regression models to predict likely mode of infection.

Introduction

The population of adolescents living with HIV (ALH) is comprised of those who acquired HIV perinatally and those who were infected non-perinatally. While similar in some ways, adolescents living with perinatally acquired HIV (ALPH) have differing socio-demographic and disease characteristics, vulnerabilities, and risk behaviors when compared to those with more recent infection. ALPH have longstanding HIV infection, and those presenting for HIV care for the first-time during adolescence would have lived with untreated HIV for prolonged periods.¹

While most children with perinatally acquired HIV are expected to have presented for the first time for HIV care as infants or children, up to a third may have slow-progressing infection and could survive into adolescence without treatment.² The Spectrum AIDS Impact Model has estimated median survival for children infected with HIV after the first year of life of up to 14 years in the absence of any antiretroviral treatment.^3,4 This means for adolescents presenting for the first time for HIV care between the ages of 10–14 years, perinatal and non-perinatal HIV acquisition are both plausible. In resource-limited settings, however, the mode of infection of children and ALH is frequently not ascertained or captured within routine healthcare databases. In an effort to distinguish these two sub-groups within data, a number of analyses have used age cut-offs at enrolment as a proxy to categorize the likely mode of HIV acquisition, with those entering HIV care at specific ages, such as <10^5,6 or <15 years⁷, assumed to have perinatally acquired HIV. The optimal age threshold to use in such situations is unclear and using age cut-offs alone could result in substantial misclassification.

Using data from the International epidemiology Databases to Evaluate AIDS (IeDEA) Southern Africa (IeDEA-SA), IeDEA Asia-Pacific (IeDEA-AP) and The Caribbean, Central and South America network for HIV epidemiology (CCASAnet) cohorts, we aimed to determine characteristics of ALH 10–14 years of age at enrolment into routine HIV care that could predict likely mode of infection and be used to better distinguish likely mode of infection.

Methods

We conducted an analysis on prospectively collected data of ALH who enrolled into HIV care 10–14 years of age within the IeDEA-SA, IeDEA-AP and CCASAnet cohorts, from 1990–2017. IeDEA is an international research consortium of seven regional collaborations of HIV observational databases (www.iedea.org). Each regional collaboration combines routine observational data from HIV care and treatment programs from several countries in their respective regions (www.iedea.org/regions). Characteristics and outcomes of ALH across the IeDEA collaboration have been described elsewhere.^7–9

Ethics

All IeDEA participating sites have local ethics approval from their local Institutional Review Boards (IRB) to contribute de-identified, anonymized, individual patient data to their IeDEA regional data centers. Each data center has its own IRB approval to receive, combine and analyze this data. Patients/caregivers either provided informed consent for inclusion of their data in the respective regional databases or regional IRBs have granted consent waivers.

Measurements and outcomes

Characteristics of adolescents presenting for HIV care 10–14 years of age included demographic, anthropometric, and recorded mode of infection. Recorded mode of infection was categorized as perinatal, non-perinatal, and unknown. Perinatally acquired HIV was as reported by the sites, and included acquisition of HIV during pregnancy, labor and delivery, or postnatally through breastfeeding. All other means of HIV acquisition were included as non-perinatally acquired HIV. Height measures were converted to age and sex-specific height-for-age z-scores (HAZ) using the “who2007” Stata macro.¹⁰

The main outcome of interest was the mode of HIV acquisition, categorized as perinatal and non-perinatal.

Statistical analysis

Characteristics of ALH enrolling into HIV care 10–14 years of age were described by mode of HIV infection (perinatal vs. non-perinatal vs. unknown) using medians with inter-quartile ranges (IQRs) and frequency distributions. To describe characteristics at entry into care, measurements from the date closest to the date of entry into care up to 6 months after were used. Using patient characteristics of adolescents with a documented mode of infection, logistic regression models to predict the likely mode of infection were developed. The predictive ability of the different models generated were compared using sensitivity, specificity and area under the receiver operating characteristic curve (AUROC). As our model was aimed at prediction, variable and model selection was based on goodness-of-fit statistics. The different predictive models were then used to classify mode of HIV infection in those with an undocumented mode of infection and the classification results from models with different combinations of patient characteristics (age alone vs. age with HAZ and sex vs. age with HAZ, sex, and CD4 count) compared.

Statistical analyses were done using Stata 15.0 (StataCorp, College Station, TX, USA).

Results

We included 10,349 adolescents (54% female) from 16 countries (9402 from IeDEA-SA, 718 from IeDEA-AP, 229 from CCASAnet). About two-thirds (65%) were <13 years of age at entry into HIV care (median [IQR] age 12.2 years [11.1;13.5]). Mode of HIV infection was documented in 20% (n=2,076); among these: 2,000 (96%) acquired HIV perinatally (53% female), and 76 (4%) acquired HIV non-perinatally (45% female). Adolescents with perinatal HIV, when compared to those with documented non-perinatal HIV, were more likely to be female (53% vs. 45%), slightly younger (median age 11.8 [10.8;13.0] vs. 12.7 [11.7;14.2] years), and had lower HAZ (median HAZ −2.27 [−3.10;−1.46] vs. −1.60 [−2.58;−0.82]) at enrolment into HIV care. There were no differences observed with regards to the median CD4 count at entry into care (186 [55;415] vs. 190 [54;431] cells/µL).

In multivariable logistic regression models using characteristics of adolescents with a documented mode of infection, perinatal HIV acquisition was associated with younger age at presentation (adjusted OR [aOR] for each increasing year 0.62, 95% CI 0.52;0.75), lower HAZ (aOR for a 1 unit increase 0.65, 95% CI 0.53;0.79), and being female (aOR 1.64, 95% CI 0.99;2.72). CD4 count at first presentation was not predictive of perinatal HIV acquisition.

When considering only one predictive variable using models of characteristics of adolescents with documented mode of infection, a model with age alone (AUROC 0.6572, 95% CI 0.5903;0.7241) had the highest predictive ability when compared to models with only HAZ (AUROC 0.6257, 95% CI 0.5568;0.6946), sex (AUROC 0.5436, 95% CI 0.4863;0.6009), or CD4 count (AUROC 0.5248, 95% CI 0.4427;0.6070) (Figure 1). Model including age and HAZ had a higher AUROC (0.7085; 95% CI 0.6420;0.7749), with a further increase in predictive ability after adding sex (0.7208; 95% CI 0.6577;0.7839) and CD4 count (0.7383; 95% CI 0.6635;0.81304) (Figure 1).

Figure 1:

Areas under the Receiver Operating Curves (ROC) for a) models with individual covariates and b) models with different covariates for predicting perinatal HIV acquisition among adolescents living with HIV presenting for HIV care 10–14 years of age

When using a model based on age alone, 75% (95% CI 74;76) of adolescents 10–14 years of age with unknown mode of infection were classified as having perinatally acquired HIV. This proportion decreased to 65% (95% CI 64;66) with the addition of HAZ and sex to the model. However, these additions increased the proportion that could not be classified due to missing HAZ data (14%, 95% CI 13;15%). Further addition of CD4 count to the model decreased the proportion classified as perinatally infected to 35% (95% CI 34;36) and increased the proportion that could not be classified to 56% (95% CI 55;57%).

Discussion

Among ALH presenting for HIV care 10–14 years of age, only <20% of adolescents had documented mode of infection, highlighting the need for predictive models to address this data gap. Among those with a recorded mode of infection, <5% were reported as having acquired HIV non-perinatally, suggesting that the majority had acquired HIV perinatally. An age threshold of <10 years at enrolment as a proxy for perinatal infection in this population would thus have misclassified 97% of these adolescents. Raising the threshold to 15 from 10 recognizes the younger adolescents presenting very late for care, with severely suppressed immune systems and stunting. These adolescents are often a neglected group which may not fit in with pediatric clinics but also might not fit with older adolescent clinics where most patients have non-perinatally acquired HIV.

Predictive models of mode of HIV acquisition based on age, sex, and HAZ had a much higher AUROC than a model using age alone and misclassified considerably fewer adolescents compared to using an arbitrary age threshold of <10 years as a proxy for likely perinatally acquired HIV. Importantly, in our routinely collected data, mode of HIV acquisition could be classified in 86% of adolescents using age, sex and HAZ. This information provides a basis on which to test additional variables and build a more reliable algorithm.

Our study is limited by a relatively small proportion of adolescents having documented mode of infection and limited additional predictor variables (e.g., no data on parental death). There is also a risk of misclassification for those incorrectly reported as not having perinatally acquired HIV due to the lack of parental HIV information. Nonetheless, predictive models based on a small number of variables that are routinely available may be useful for analysts wanting to include likely mode of HIV acquisition when the mode of HIV acquisition is not recorded. If age alone is used as a proxy for mode of HIV acquisition due to missing data on predictor variables, careful consideration should be given to the appropriate age threshold for assuming likely perinatally acquired HIV in different contexts.