An Early Infant HIV Risk Score for Targeted HIV Testing at Birth

A cross-sectional sample of newborns exposed to HIV was recruited to the study. Data abstraction and HIV testing occurred within 72 hours of birth.

Trained research staff recruited patients. Researchers identified mothers with known HIV-positive or -negative status and mothers of unknown HIV status daily from Monday through Friday during working hours in the obstetric unit pre- and postdelivery. Mothers who were HIV-negative were sent for HIV counseling and testing if their last test was done before 32 weeks’ gestation, in accordance with national protocols.

Researchers interviewed mothers and recorded data onto a case report form. Researchers reviewed medical records for maternal HIV results, infant birth weight, duration of labor, and PMTCT interventions. The following infant parameters were documented: birth history; anthropometric data: weight, length, head circumference, and mid-upper arm circumference; and the presence or absence of symptoms (see Supplemental Information for definitions) that included failing to thrive (includes low birth weight [LBW]), birth weight ≤2.5 kg, congenital pneumonia, hepatosplenomegaly, oral candidiasis, significant lymphadenopathy, and any opportunistic infections. Maternal characteristics documented included antenatal care (ANC), labor history, PMTCT interventions and the timing thereof, the last documented maternal HIV viral load (VL), in accordance with the current South African PMTCT guideline, antiretroviral drug history, and compliance (Supplemental Tables 5–8). Poor compliance was defined as <95% dose compliance by using self-reported data or visual analog scale.

Research nurses or qualified medical doctors working at KPTH drew blood from infants exposed to HIV. Blood was tested for HIV infection by using total nucleic acid PCR.

The laboratory reported all HIV PCR–positive results to study investigators on a daily basis. Newborns with positive or indeterminate HIV PCR results were referred to HIV clinicians for specialized care and managed by the immunology (HIV) clinic in accordance with national guidelines. Mothers with high VLs or other medical problems were referred to the relevant adult services.

Management at KPTH granted permission to conduct the study, and the University of Pretoria Faculty of Health Sciences Research Ethics Committee gave ethical clearance (protocol 285_2014). Each mother gave written informed consent on enrollment in the study.

Case report forms were entered directly into the Research Electronic Data Capture (REDCap) (Vanderbilt University, Nashville, TN) system hosted at the South Africa Medical Research Council.

Data were analyzed by using the statistical software SAS version 9.4 TS1M5 (SAS Institute, Inc, Cary, NC). Continuous data were expressed as means and SDs or as medians and interquartile ranges (IQRs) for skewed distributions. Discrete or categorical data were summarized by using frequencies and percentages. Normally distributed data were compared by using independent t tests, otherwise nonparametric alternatives were used.

The weight-for-age z scores were calculated by using 2000 Centers for Disease Control and Prevention growth charts adjusted for gestational age for preterm infants and World Health Organization 2006 growth charts from the Multicentre Growth Reference Study for term infants. We used univariate and multivariate logistic regression models to identify associations between HIV infection and maternal and infant parameters. The following predictor variables were considered: preterm gestational age (yes = 1, no = 0), LBW <2.5 kg (yes = 1, no = 0), maternal VL value (lower than detectable level [LDL] = 1, <1000 = 2, ≥1000 = 3), maternal VL value (LDL = 1, <1000 = 2, 1000–10 000 = 3, >10 000 = 4), maternal HIV seroconversion after 32 weeks’ gestation (yes = 1, no = 0), maternal cART duration at birth <4 weeks (yes = 1, no = 0), small for gestational age (SGA) (yes = 1, no = 0), maternal tuberculosis (yes = 1, no = 0), symptomatic (yes = 1, no = 0), and maternal CD4 cell count value (<200 cells per mm³ = 1, 200–500 cells per mm³ = 2, >500 cells per mm³ = 3). Multivariate regression was initially done on a saturated model (full model) and thereafter reduced to 2-, 3-, and 4-risk models. Model significance was measured with P values of .05 and .25.

After fitting logistic regression models, the variables with the highest predictive value were selected by using Schwarz information criterion or Bayesian information criterion (BIC). The performance of the selected models was evaluated by using the C-index or the area under the receiver operating curve. We determined the derived risk scores, sensitivity, and specificity as well as false-positive and false-negative rates for various cutoff values. We developed probability models with the end point of a positive birth HIV PCR using logistic regression of univariate and multivariate characteristics and risk factors. For modeling purposes, only positive and negative PCR test results were included.

Between August 2014 and December 2016, 15 175 live infants were born at KPTH, 3356 (22.12%) of these to mothers infected with HIV (Fig 1). Informed consent was obtained from 1759 of 1911 (92.05%) eligible patients. Patients with birth HIV PCR test results were included.

Almost half of mothers who were HIV positive were in their thirties (854 of 1759; 48.55%) and most had a secondary education (1557 of 1759; 88.52%). Although most women were not legally married (1370 of 1706; 80.30%), almost half received monetary support from a partner (809 of 1726; 46.87%) and just over a third were employed (599; 34.70%). Almost all mothers reported having access to a cell phone (1625 of 1720; 94.48%) and household assets such as a television (1415; 82.28%) and refrigerator (1252; 72.79%) (Table 1).

Most mothers knew their HIV status before delivery (98.77%) and were on cART (1626 of 1704; 95.29%), namely fixed dose combination (FDC) treatment (1473; 90.09%), with 105 (7.09%) having received FDC for <1 month before delivery. Close to a quarter of mothers who were HIV-positive (376 of 1687; 22.25%) seroconverted toward the end of pregnancy (>32 weeks’ gestation). Using laboratory data, we could trace VLs within 3 months of delivery for 935 (53.16%) mothers. Similarly, we could trace CD4 cell counts for 1196 (67.99%) mothers for the preceding 6 months. We recorded 562 (60.11%) mothers with undetectable VL (LDL), indicating virological control. Approximately 1 in 5 mothers (208; 22.24%) had a VL level of >1000 copies per mL documented within 3 months before delivery. We observed mild immunologic suppression in 588 (49.16%) and 394 (32.94%) mothers who had CD4 counts between 200 and 500 cells per mm³ and >500 cells per mm³, respectively. Maternal VL and maternal CD4 cell count values were negatively correlated −0.103 (P = .0053).

Few mothers had tuberculosis (21 of 1730; 1.21%), most of whom (13; 65%) were being treated for tuberculosis. Only 5.75% (97 of 1688) of mothers had no ANC visits, the median visits being 4 (IQR 3–5 visits). Infants with positive HIV PCR results were associated with mothers who had no ANC visits (P = .0005) (Table 1). Infants with positive HIV PCR results were associated with mothers who had mean VLs of 53 424 (SD = 116 854) copies per mL, whereas infants with negative HIV PCR results were associated with mothers who had mean VLs of 11 361 copies per mL (SD = 66 185) (P = .0511).

Of the infants enrolled in the study, 53.16% (935) were boys. More than a quarter of enrolled infants (432 of 1655; 26.10%) were born at <38 (37 completed) weeks, with a median gestation of 35 weeks (IQR 32–36 weeks) (Table 2).

Infants had a mean birth weight of 2.83 kg. LBW (<2.5 kg) was documented in 398 of 1598 (24.55%) of the infants who were HIV PCR–negative and 13 of 32 (40.63%) in the positive-result group (P = .0329 [χ²]). The median z score for weight was −0.33 (IQR −1.06 to 0.38).

Fewer than 15% (14.33%) of enrolled infants displayed clinical symptoms at birth. No newborns had generalized lymphadenopathy or extensive oral candidiasis. Evidence of growth restriction or SGA was documented in 204 of 1689 (12.08%) of the enrolled infants, of whom 6 were PCR-positive. Pneumonia and anemia were documented in 1.7% and 0.3% of infants, respectively (Table 2). Newborns that were symptomatic more frequently tested HIV-positive (P = .0042).

Of the 1691 infants with birth PCR results, 31 (1.8%) received HIV-positive results, 1646 (97.3%) received HIV-negative results, and 14 (0.83%) received indeterminate results; 3.87% samples were rejected (68 of 1759). According to the national HIV testing protocol, the indeterminate HIV PCR results were repeated in 12 of the 14 patients; 2 infants could not be traced. Half of the repeated HIV PCR tests (6 of 12) produced positive test results and half produced negative test results, increasing the positivity rate to 2.2% (37 of 1703).

The mean turnaround time for the PCR results was 68.27 hours (IQR 46.05–93.88 hours).

We used univariate regression models to identify associations between HIV PCR–positive outcome and 10 infant and maternal characteristics.

Newborns with positive PCR results were significantly associated with maternal VL levels <1000 (odds ratio [OR] = 26.53; 95% confidence interval [CI] 1.353–520.310; P = .002) and ≥1000 (OR = 123.67; 95% CI 7.385–>999.99); maternal cART duration <4 weeks (OR = 0.146, 95% CI 0.057–0.373; P < .0001); and symptomatic newborns (P = .237, 95% CI 0.100–0.561; P = .0011).

The saturated multivariate regression model only identified 2 significant risk factors (P < .05), namely maternal VL (P = .0002) and symptomatic newborns (P = .02). We retained variables with a P < .25 in additional models to increase predictive value; therefore, cART <4 weeks (P = .17) and SGA (P = .25) were added to risk models. These 4 parameters were combined in 2-, 3-, and 4-risk regression models. Table 3 illustrates risk models, starting with the individual unweighted model, followed by two 2-risk models (models 1 and 2), a 3-risk (model 3), and a 4-risk (model 4) model, and finally the full model that used 9 parameters in 1 weighted model. Maternal VL and infant symptoms were the first risks in the 2-risk model, whereas the second 2-model score modeled 2 maternal parameters: maternal VL and duration of maternal cART. The 3-risk model incorporated maternal VL, cART duration, and infant symptomatology; the 4-risk model added infant size for gestational age to the mentioned 3 risks.

Mothers with a VL ≥1000 copies per mL who deliver a symptomatic newborn have a 38% chance of a positive PCR HIV test result at birth (model 1–2 risks, Fig 2A). Another 2-risk model, containing only maternal characteristics, maternal VL and cART duration, revealed a probability of 28% for a newborn PCR-positive test result if maternal VL is ≥1000 copies per mL and the mother has not received cART treatment of >4 weeks (model 2–2 risks, Fig 2B). In model 3, newborns have a probability of 0.49 of a positive-result PCR test if the newborn is symptomatic, the mother had received cART for <4 weeks before delivery, and if the mother had a VL of ≥1000 copies per mL (Fig 2C). The 4-risk model included maternal characteristics VL and cART duration and infant characteristics symptomatic newborn and SGA; in this model, newborns have the highest probability of a positive HIV PCR result if maternal VL is ≥1000 copies per mL, cART duration is <4 weeks before delivery, and an infant is symptomatic and SGA (model 4, Fig 2D).

Each model can be used at a risk score probability cutoff to determine targeted PCR testing algorithms. When using both the 3- and 4-risk model scores and considering a probability of 0.02 and 0.04 as an indication for targeted birth testing, the sensitivity is 80% and 76%, respectively (Table 4).

In this study, we identified associations between maternal and infant characteristics and positive-result HIV PCR tests for newborns. We combined significant variables to build predictive models for early infant HIV risk scores to detect HIV infection at birth. According to our models, newborns had the highest (0.57) probability for HIV PCR positivity if newborns were symptomatic and SGA and were born to a mother who received cART for <4 weeks and had a VL ≥1000. Newborns had a 0.28 probability of a positive-result HIV PCR test if only maternal VL ≥1000 and cART <4 weeks were included in the model. Our study population had high HIV prevalence among mothers (22.12%), and overall PCR positivity in newborns was 2.2%, which is higher than the national average of 1.1%.³ 

Mothers who gave birth to newborns with in utero–acquired HIV more frequently had either no (P = .0023) or fewer than 3 ANC visits (P = .02). At the time of the study, basic ANC guidelines in South Africa recommended an early ANC visit (<12 weeks’ gestation) followed by 4 follow-up ANC visits for low-risk pregnancies. In mid 2016, 8 follow-up visits were recommended, and mothers who test HIV-positive during pregnancy should initiate cART a week later.¹¹ Mothers who are infected with HIV should receive continuous education about pregnancy-related issues. ANC should complement HIV monitoring and treatment in these mothers.

Almost all mothers knew their HIV status before delivery (99%) and were on cART (95%). Encouragingly, these values are in line with 2 of the current United Nations Programme on HIV/AIDS 90-90-90 goals; namely, 90% of all patients infected with HIV should be aware of their HIV status, and 90% of patients infected with HIV should receive cART.¹² Mothers who had been on cART for shorter than 4 weeks were significantly associated with newborns who had a positive-result PCR HIV test at birth. Despite most mothers knowing their status and receiving cART, only 60% had achieved viral suppression, far less than the target of 90% of HIV patients on cART who should be virally suppressed to reach the 90-90-90 goals.

In utero HIV infection in newborns was significantly associated with a detectable maternal VL. Both VL levels <1000 and ≥1000 copies per mL proved significant risks for in utero HIV acquisition (P = .0002 and P < .0001, respectively) in both univariate and multivariate analyses. Myer et al¹³ emphasized the importance of diligent maternal VL monitoring and management during pregnancy and breastfeeding. VL testing 4 weeks before or at delivery can provide valuable information to guide targeted interventions from birth, including birth PCR testing.^13,14 Health care facilities providing ANC services should focus on the importance of providing cART for mothers at least 4 weeks before delivery. Mothers who are HIV-positive should be enrolled in PMTCT programs, which should consider repeat HIV testing, treatment, and VL monitoring at 34 to 36 weeks’ gestation. In the future, the value of maternal VL point-of-care testing should be assessed.

In our study, newborns infected with HIV were more likely to be preterm (<38 weeks’ gestation), LBW, and SGA, although none of these parameters were significantly associated with a positive-result HIV PCR test in the univariate regression models. We included SGA values in the 4-model multivariate analysis, which added predictive value to targeted testing models. We recommend that all health care facilities with maternity services have infant gestational age tables and infant scales to determine weight-for-age measurements for newborns exposed to HIV.

The newborn HIV PCR positivity rate was 1.83% (31 of 1691) with <5% of samples rejected (68, 3.9%). After repeat testing of the indeterminate results, an additional 6 patients tested HIV-positive, increasing the positivity rate to 2.2% (37 of 1703). With a maternal HIV positivity percentage of 22% and 1.8% mother-to-child transmission of HIV (MTCT), 396 per 100 000 live births were HIV infected in this study. Goga et al¹⁵ looked at early (4–8 weeks) MTCT risk in South Africa from 2011–2012 to 2012–2013 in relation to the South African PMTCT strategy, which is in line with World Health Organization Option A. In 2011–2012, MTCT was 2.7% and 2.6% in 2012–2013, varying between provinces (1.9%–5.4%). Mothers who started cART during or before the first trimester of pregnancy had low risk of early MTCT (1.2%).² We recorded a similar rate of MTCT, indicating that most MTCT events are therefore in utero and intrapartum. The risk has not changed significantly in the past 5 to 6 years. The current South African MTCT case rate is not close to the elimination of MTCT target of ≤50 per 100 000 live births.¹⁶ 

In this study, we present models for targeted birth PCR testing, which may be useful in resource-constrained settings. Our findings indicate that maternal VL testing is vital for a targeted birth PCR approach. Our 3- and 4-risk models achieved a sensitivity of close to 80%, indicating that 1 in 5 newborns infected with HIV will not be offered targeted birth testing. Globally, EID, early treatment, and elimination of MTCT are prioritized; thus, we support universal birth testing within the South African PMTCT program.

Language editing was done by Cheryl Tosh, University of Pretoria.

ANC

antenatal care

BIC

Bayesian information criterion

cART

combined 3-drug antiretroviral therapy

CI

confidence interval

EID

early infant diagnosis

FDC

fixed dose combination

IQR

interquartile range

KPTH

Kalafong Provincial Tertiary Hospital

LBW

low birth weight

LDL

lower than detectable level

MTCT

mother-to-child transmission of HIV

OR

odds ratio

PCR

polymerase chain reaction

PMTCT

prevention of mother-to-child transmission of HIV

SGA

small for gestational age

VL

HIV viral load