In Utero Antiretroviral Exposure and Risk of Neurodevelopmental Problems in HIV-Exposed Uninfected 5-Year-Old Children

Abstract

Studies have observed neurodevelopmental (ND) challenges among young children perinatally HIV-exposed yet uninfected (CHEU) with in utero antiretroviral (ARV) exposure, without clear linkage to specific ARVs. Atazanavir (ATV) boosted with ritonavir has been a preferred protease inhibitor recommended for pregnant women, yet associations of ATV with ND problems in CHEU have been reported. Studies among early school-age children are lacking. The pediatric HIV/AIDS cohort study (PHACS) surveillance monitoring for antiretroviral therapy (ART) toxicities (SMARTT) study evaluated 5-year-old monolingual English-speaking CHEU using the behavior assessment system for children, Wechsler preschool and primary scales of intelligence, and test of language development-primary. A score ≥1.5 standard deviations worse than population norms defined a signal within each domain. Analyses of risk for signals were stratified by timing of any ARV initiation. Associations between ARV exposure and risk of ND signals were assessed using proportional odds models, adjusting for confounders. Among 230 children exposed to ARVs at conception, 15% had single and 8% had multiple ND problems; ATV exposure was not associated with higher risk of signals [adjusted cumulative odds ratio (cOR) = 0.66, confidence interval (CI): 0.28–1.56]. However, among 461 children whose mothers initiated ARVs during pregnancy, 21% had single and 12% had multiple ND problems; ATV exposure was associated with higher risk of signals (cOR = 1.70, CI: 0.82–3.54). The specific regimen tenofovir/emtricitabine/ATV was associated with higher risk (cOR = 2.31, CI: 1.08–4.97) relative to regimens using a zidovudine/lamivudine backbone combined with non-ATV ARVs. It remains important to monitor neurodevelopment of CHEU during early childhood and investigate the impact and the role of timing of in utero exposure to specific ARVs.

Introduction

Antiretroviral therapy (ART) during pregnancy has greatly reduced the risk of mother-to-child transmission of HIV. Because of this success, globally in 2020 ∼ 15.4 million children were perinatally HIV-exposed yet uninfected (CHEU) and under the age of 15 years,¹ most of them exposed to antiretroviral (ARV) drugs in utero, and that number increases annually.² ARV drugs are known to cross the placental barrier, raising the possibility of mitochondrial toxicities and other adverse effects in primates and humans.^3,4 Given the global increases in ARV uptake by pregnant women living with HIV, and mixed findings of safety, recent reports have recommended pharmacovigilance in identifying potential adverse effects of agents, particularly those prescribed to women during pregnancy.⁵

Among concerns regarding in utero ARV exposure is the potential for neurodevelopmental (ND) effects, including language delays and behavioral difficulties. Mental health problems are common in both children living with HIV and CHEU.^6–8 Several studies from different research groups and regions with notable HIV prevalence have examined the association of in utero ARV exposure with the risk of ND problems among CHEU^9–26 with mixed findings. A systemic review of 24 studies, including 19 from Africa, reported worse neurodevelopment in at least one domain among young CHEU (<5 years) compared with HIV-unexposed children.

In contrast, a review of 13 studies examining CHEU exposed to ART suggested little or no evidence of an effect of specific maternal ART regimens on neurodevelopment.⁹ In the surveillance monitoring for ART toxicities (SMARTT) study, in utero exposure to atazanavir (ATV) was observed in multiple evaluations to increase the risk for both language and social–emotional delays among 1-year-old CHEU.^13,18,23 However, in a subsequent analysis of the SMARTT cohort, longer prenatal ATV exposure duration and higher meconium ATV concentrations appeared to be associated with lower risk of late language emergence.¹⁶

In another evaluation of CHEU enrolled in SMARTT, tenofovir exposure was associated with increased risk for speech impairment at 3 years of age and decreased risk for concomitant language and cognitive impairment at 5 years of age.¹⁴ Of interest, however, among 5- to 13-year-old children in the SMARTT study, no associations were observed of any individual ARVs with poor cognitive or academic outcomes.¹⁵

Although a number of ARVs have demonstrated some association with suboptimal outcomes, the possible signals of ND effects with ATV exposure have been implicated in several large-scale studies and warrant further examination. ATV boosted with ritonavir has been recommended by the US Department of Health and Human Services (DHHS) as a preferred protease inhibitor (PI) for pregnant women living with HIV since 2012.^27–29

ATV boosted with ritonavir in combination with a dual nucleoside reverse transcriptase inhibitor (NRTI) backbone is recommended as preferred second- and third-line treatments by the World Health Organization (WHO)³⁰ for people living with HIV, including pregnant women. With substantial number of CHEU exposed to ATV worldwide, continued surveillance of their neurodevelopment as they reach early school age and beyond remains critically important.

The fetal brain is vulnerable to a confluence of biological and environmental influences, possibly related to in utero ARV exposure and other risks that may be associated with parental health, substance use, other toxins, and/or an inadequate supply of oxygen and nutrients.^31–39 Thus, potential for disruption of the normal ND trajectory exists, especially if exposures occur during early or sensitive periods of development.⁴⁰ As children age, it is important to examine whether early risks have long-term impact on individual or multiple developmental domains, given the possibility that manifestations may not be functionally obvious in the early years of dynamic brain growth and development.

The presence of delays or deficits in multiple developmental domains has significant implications for ongoing development and for the timing, type, and duration of early intervention services to ameliorate observed difficulties. Early identification of co-occurring delays and timely provision of appropriate, targeted, and integrated services can alter atypical developmental pathways and interrupt progression to more severe and persistent challenges, including mental health problems.⁴¹

Previous studies have primarily focused on outcomes within individual domains, precluding a full appreciation of the potential concomitant effects that in utero ARV exposure may have on co-occurring ND problems. Individual ND domains may have distinct or shared pathways,⁴² and the risk of problems in each may vary in severity. By evaluating the co-occurrence of problems, and how they are related to ARV exposure, we may better appreciate mechanisms by which ARVs affect ND.

Thus, in this study, we aimed to (1) describe signals of ND problems in language, behavior, and/or cognition among perinatally ART exposed CHEU at 5 years of age and (2) assess the specific association of in utero exposure to ATV versus non-ATV-based ART and different NRTI backbones with the risk of multiple concomitant signals, while accounting for interpersonal and social contextual factors as potential confounders.^6,7,33,36,43 In addition, we explored the role of timing of initiation of in utero ARV exposure on ND outcomes as a first step to understand the possible differential impact of ARV exposure by trimester.

Methods

Study population

The SMARTT study of the pediatric HIV/AIDS cohort study (PHACS) network is an ongoing prospective cohort study designed to monitor CHEU up to 17 years of age to identify adverse effects of in utero ARV exposure. Since 2007, SMARTT has enrolled >4000 CHEU, either at birth or in childhood with information on maternal ART and other exposures during pregnancy. The SMARTT study is conducted at 22 clinical sites in the United States including Puerto Rico;¹³ its protocol was approved by the institutional review board at Harvard T.H. Chan School of Public Health and all participating sites. Primary caregivers provided informed consent for their children and their own participation.

This specific analysis included monolingual English-speaking CHEU who attended SMARTT 5-year-old study visits before September 2018 and whose mothers living with HIV received ART during pregnancy. Monolingual English speakers were defined as children who lived in homes in which English was the sole spoken language and no other language was spoken to the child outside the home. We excluded four children with HIV-unrelated chromosomal anomalies known to affect neurodevelopment.

ND outcome measures

Psychologists at clinical sites administered the Wechsler Preschool and Primary Scale of Intelligence (WPPSI-III)⁴⁴ and the test of language development-primary (TOLD-P:3)⁴⁵ according to standardized procedures. Assessment results considered invalid by the site psychologist and confirmed by team study psychologists (R.S. and K.M.) and/or language specialists (M.L.R. and S.R.) were excluded. The age-adjusted normative mean (standard deviation; SD) of the WPPSI-III full scale intelligence quotient (FSIQ) and TOLD-P:3 spoken language quotient (SLQ) is each 100 (15). For cognitive and language functioning, we defined significant problems, or signals of risk, as composite scores at least 1.5 SD below the normative mean; that is, FSIQ <78 and SLQ <78, respectively.

The Behavior Assessment System for Children, Second Edition (BASC-2)⁴⁶—Parent Rating Scales, as collected during an interview with the child's parent/primary caregiver, was used to evaluate children's emotional–behavioral development. The normative mean (SD) of the behavioral symptom index (BSI) T-score of the BASC-2 is 50 (10); higher T-scores indicated more problematic behaviors. Emotional–behavioral problems, or signals of risk, were defined as BASC-2 BSI T-scores >1.5 SDs above the normative mean (i.e., >65). This range incudes the upper half of the “at-risk” range (BASC-2 T-score 60–69) and the entire clinically significant range (BASC-2 T-scores ≥70).

We examined nine clinical conditions that could be potentially associated with child ND outcomes. These conditions include three birth conditions (gestational age <37 weeks, birth weight <2500 g, and small for gestational age), two growth conditions (head circumference z-score less than −2 at age 1 year, and failure to thrive defined as weight z-score less than −2 by age 5 years), and four clinician-reported diagnoses [autism spectrum disorder by age 5 years, infections with congenital cytomegalovirus (CMV), syphilis, and perinatal herpes simplex viruses].

For each child, we counted the total number of conditions/diagnoses that were present and summarized such count by the number of signals of ND problems. In this study, the ordinal outcome of interest was defined as no signals, only one signal, or the presence of multiple (two or three) signals in ND domains.

Maternal/caregiver cognition was assessed by the Wechsler Abbreviated Scale of Intelligence (WASI),⁴⁷ where the WASI FSIQ score at least 1.5 SDs below the normative mean (<78) indicates lower cognitive functioning. Caregiver psychiatric/substance use disorders were assessed by the client diagnostic questionnaire^48,49 and by maternal self-reported functional limitations and health problems.

In utero ARV exposure

We categorized the timing of in utero ARV exposure initiation for the child as at conception, if the mother was prescribed ARV medications before her last menstrual period, or as postconception, if the mother initiated/restarted ARV use during pregnancy. We first classified the initial ARV regimens during pregnancy as either ATV containing or non-ATV containing regimens.

Based on previous SMARTT evaluations, we further compared tenofovir disoproxil fumarate/emtricitabine/atazanavir with or without ritonavir booster [TDF/FTC/ATV(r)], with or without a ritonavir booster, and regimens of TDF/FTC without ATV to regimens of zidovudine/lamivudine (ZDV/3TC) without ATV (reference group).^29,50 Because high bilirubin levels have been shown to be associated with both ND risk⁵¹ and ATV exposure,⁵² we considered available bilirubin levels of mothers receiving ATV during pregnancy.

Statistical analysis methods

The prevalence of single and multiple signals of ND problems was estimated with an exact 95% confidence interval (CI) based on a binomial distribution, by timing of first perinatal ARV exposure. Cumulative odds ratios (cORs) and 95% CIs for ATV-containing regimens compared with those for non-ATV-containing regimens were estimated using proportional odds models, with separate analyses conducted for children whose mothers initiated ART preconception or postconception.

Generalized estimating equation (GEE) models with an independent working correlation structure and robust standard errors were used to account for clustering of CHEU within clinical sites, and adjusted for gender, race (White, Black, other), ethnicity (Hispanic, non-Hispanic) of the child, and maternal age at delivery, maternal education (below high school, high school or above), household income <$20,000/year (Yes/No), maternal cognition by WASI FSIQ (<78, ≥78), and maternal-reported family history of language/reading problems (Yes/No) for all strata.

Because GEE models were utilized to fit the proportional odds model to account for clustering, the model fitting and the proportional odds assumption were tested using a score test implemented using the %GEEORD SAS macro. For the postconception stratum, the association was also adjusted for maternal self-reported substance use in the first trimester of pregnancy: tobacco, alcohol, marijuana, or other illicit drugs (Yes/No for each) and the first available HIV RNA level (<400, ≥400 copies/mL) measured during pregnancy and before ART initiation.

A category of missing was created for covariates with missing data. We considered adjustment for substance use one type at a time to avoid collinearity in fitting GEE models. These separate models had very similar results and we thus report the models that adjusted for marijuana use, in light of recent evidence of an association between prenatal cannabis exposure and neuropsychological and behavioral abnormalities in children.⁵³ Associations were not adjusted for the mentioned birth or growth clinical conditions and diagnoses, since they could lie on the causal pathway between in utero ARV exposures and ND outcomes.

In secondary analyses, the postconception period was further stratified to trimester 1 and trimesters 2 or 3 to explore the role of timing of ART initiation during pregnancy on ND outcomes. All analyses were conducted using SAS version 9.4 and SAS/STAT version 14.3 (SAS Institute, Cary, NC).

Results

Demographics and clinical characteristics

As of September 2018, 1745 CHEU enrolled in SMARTT attended 5-year-old study visits (between ages 4.5 and 5.5 years), among whom 922 were monolingual English-speaking children; 872 (95%) of these were exposed to ART in utero (Fig. 1). Approximately 80% of the children in each stratum of mothers' ART initiation period had available and valid ND assessments in all three domains; 90% of these were completed within 4 months of the fifth birthday.

FIG. 1.

Flow chart of children in the SMARTT cohort included in the analyses. ^a–cTwo, one, and two mothers, respectively, had missing information on ATV and were excluded from the regression analysis for the association of ART exposure with the risk of neurodevelopmental problems. ART, antiretroviral therapy; ATV, atazanavir; ND, neurodevelopmental; SMARTT, surveillance monitoring for ART toxicities.

Compared with those with available and valid assessments, children without valid assessments were more often Hispanic, and had mothers with less education, missing maternal cognitive evaluations, and missing information on current health conditions or family history of language problems (data not shown). The availability of valid assessments did not differ by ATV exposure.

The demographic and maternal characteristics of eligible children are summarized in Table 1 by timing of ART exposure initiation. Demographic characteristics of children were similar across all strata, with the exception of a higher proportion of Hispanics among those whose mothers initiated ART at/before conception. Compared with mothers who started ART postconception, mothers who were prescribed ART at/before conception were on average 2–3 years older at delivery, less often reported household income <$20,000 per year, and more often reported family history of language problems.

Table 1.

Child and Maternal/Caregiver Characteristics by Timing of In Utero Antiretroviral Exposure Initiation Among 868 Monolingual English-Speaking 5-Year-Old Children Enrolled in the Surveillance Monitoring for Antiretroviral Therapy Toxicities Study

	At conception (N = 293)	Timing of in utero ARV exposure initiation
		During pregnancy
		Overall (N = 575)	In trimester 1 (N = 163)	In trimesters 2/3 (N = 412)
Child
Female	134 (46%)	280 (49%)	80 (49%)	200 (49%)
Black/African American	252 (86%)	510 (89%)	141 (87%)	369 (90%)
Hispanic ethnicity	31 (11%)	32 (6%)	11 (7%)	21 (5%)
Year of birth
2001–2006	85 (29%)	176 (31%)	49 (30%)	127 (31%)
2007–2009	99 (34%)	201 (35%)	54 (33%)	147 (36%)
2010–2013	109 (37%)	198 (34%)	60 (37%)	138 (33%)
Maternal
Age at delivery (years), median (Q1, Q3)	29.0 (24.5, 33.8)	26.2 (22.7, 30.7)	26.6 (23.1, 30.9)	26.0 (22.6, 30.6)
Less than high school education	93 (32%)	193 (34%)	57 (35%)	136 (33%)
Household income at entry <$20k/year	179 (61%)	398 (69%)	118 (72%)	280 (68%)
FSIQ score <78a,b	58 (20%)	133 (23%)	34 (21%)	99 (24%)
Family history of language problemsa,c	111 (38%)	189 (33%)	57 (35%)	132 (32%)
First trimester substance use
Anya,d	84 (29%)	167 (29%)	61 (37%)	106 (26%)
Tobaccoa,e	59 (20%)	136 (24%)	51 (31%)	85 (21%)
Alcohola,d	18 (6%)	41 (7%)	17 (10%)	24 (6%)
Marijuanaa,d	28 (10%)	46 (8%)	18 (11%)	28 (7%)
Illicit druga,f	30 (10%)	54 (9%)	21 (13%)	33 (8%)
First trimester passive smoke exposurea,g	96 (33%)	244 (42%)	83 (51%)	161 (39%)
First CD4+ <200 (cells/mm3)	39 (13%)	77 (13%)	24 (15%)	53 (13%)
First RNA load ≥400 copies/mLa,h	104 (35%)	420 (73%)	93 (57%)	327 (79%)
Caregiver characteristics at 5-year visit
Less than high school education	87 (30%)	157 (27%)	50 (31%)	107 (26%)
Caregiver was biological mothera,i	260 (89%)	504 (88%)	136 (83%)	368 (89%)
Caregiver changed at least oncea,j	19 (6%)	35 (6%)	12 (7%)	23 (6%)
Psychiatric/substance use disordera,k	96 (33%)	197 (34%)	59 (36%)	138 (33%)
Difficulty caring for childa,l	37 (13%)	99 (17%)	41 (25%)	58 (14%)
Number of self-reported function limitationsa,l
1	161 (55%)	260 (45%)	73 (45%)	187 (45%)
2	35 (12%)	74 (13%)	15 (9%)	59 (14%)
>2	52 (18%)	108 (19%)	37 (23%)	71 (17%)
Number of self-reported health problemsa,l
1	136 (46%)	274 (48%)	76 (47%)	198 (48%)
2	58 (20%)	83 (14%)	25 (15%)	58 (14%)
>2	58 (20%)	119 (21%)	31 (19%)	88 (21%)

^a Overall missing rate ≥5%.

Number of missing values by stratum: at conception, in trimester 1 and trimesters 2/3—^b26, 14, 29; ^c31, 29, 56; ^d11, 8, 21; ^e12, 8, 24; ^f11, 8, 23; ^g28, 15, 52; ^h10, 56, 41; ⁱ15, 9, 23; ^j20, 9, 31; ^k18, 13, 29; ^l20, 16, 42.

ARV, antiretroviral; FSIQ, full Scale intelligence quotient; Q1, first quartile; Q3, third quartile.

Mothers who initiated ART in the first trimester had a higher proportion of self-reported substance use, passive smoke exposure during trimester 1, and difficulty caring for their child. The earliest available RNA measures during pregnancy were more frequently >400 copies/mL among those who started ART during pregnancy.

Signals of ND problems

FSIQ, SLQ, and BSI scores for children with valid assessments are summarized separately for each domain in Table 2. The pattern of concomitant signals of problems by timing of ART initiation for children with valid assessments in all three domains is displayed in Table 3. The proportion of CHEU with ND problems was lower for those whose mothers initiated ART earlier—15% versus 21% for a single signal of ND problem, and 8% versus 12% for multiple signals of ND problems, for CHEU exposed to ART at conception versus postconception.

Table 2.

Summary of Outcome Measures in Behavior, Cognition, and Language by Timing of In Utero Antiretroviral Exposure Initiation for Children with Valid Assessment Separately in Language, Cognition, and Behavior

		Timing of in utero ARV exposure initiation
		At conception (N = 293)	In trimester 1 (N = 163)	In trimesters 2/3 (N = 412)
BSI T-scorea	N	252	143	356
	Mean (SD)	51.5 (11.9)	53.64 (12.9)	54.87 (13.3)
	Minimum, maximum	31, 97	31, 100	31, 101
Behavioral problem (BSI T-score >65)a	No	223 (76%)	120 (74%)	281 (68%)
	Yes	29 (10%)	23 (14%)	75 (18%)
FSIQb	N	247	141	343
	Mean (SD)	95.9 (15.7)	95.96 (13.3)	93.27 (15.1)
	Minimum, maximum	48, 137	60, 144	45, 138
Cognitive problem (FSIQ <78)b	No	221 (75%)	132 (81%)	298 (72%)
	Yes	26 (9%)	9 (6%)	45 (11%)
SLQc	N	245	134	352
	Mean (SD)	92.8 (14.9)	92.59 (14.0)	89.66 (14.4)
	Minimum, maximum	57, 128	55, 137	50, 138
Language problem (SLQ <78)c	No	210 (72%)	115 (71%)	279 (68%)
	Yes	35 (12%)	19 (12%)	73 (18%)

Number of missing values at conception, in trimester 1 and trimesters 2/3: ^a41, 20, 56; ^b46, 22, 69; ^c48, 29, 60.

ARV, antiretroviral; BSI, behavioral symptom index; FSIQ, full scale intelligence quotient; SD, standard deviation; SLQ, spoken language quotient.

Table 3.

Pattern of Concomitant Signals by Timing of In Utero Antiretroviral Exposure Initiation for Children with Valid Assessments in Language, Cognition, and Behavior Domains

No. of signals	ND domain		Timing of in utero ARV exposure initiation
		At conception (N = 230)	Overall (N = 461)	In trimester 1 (N = 129)	In trimesters 2/3 (N = 332)
None		177 (77%)	312 (68%)	96 (74%)	216 (65%)
One		34 (15%)	95 (21%)	21 (16%)	74 (22%)
	Language	14 (6%)	35 (8%)	8 (6%)	27 (8%)
	Cognition	4 (2%)	5 (1%)	2 (2%)	3 (1%)
	Behavior	16 (7%)	55 (12%)	11 (9%)	44 (13%)
Two		12 (5%)	42 (9%)	10 (8%)	32 (10%)
	Language and cognition	9 (4%)	24 (5%)	2 (2%)	22 (7%)
	Language and behavior	0 (0%)	10 (2%)	6 (5%)	4 (1%)
	Cognition and behavior	3 (1%)	8 (2%)	2 (2%)	6 (2%)
Three		7 (3%)	12 (3%)	2 (2%)	10 (3%)
Single		34 (15%)	95 (21%)	21 (16%)	74 (22%)
95% CI		10.5–20.0%	17.0–24.6%	10.4–23.8%	17.9–27.2%
Multiple		19 (8.3%)	54 (12%)	12 (9.3%)	42 (12.7%)
95% CI		5.1–12.6%	8.9–15.0%	4.9–15.7%	9.3–16.7%

ARV, antiretroviral; CI, confidence interval; ND, neurodevelopmental.

At least one of the nine examined child clinical conditions/diagnoses, (i.e., premature birth, low-birth weight, small for gestational age, small head circumference, failure to thrive, autism spectrum disorder, perinatal herpes simplex virus, and congenital infections of CMV and syphilis) was observed in 35% of CHEU with no signals of ND problems, 35% of those with a single signal, and 52% of those with multiple signals. This observation suggested the possibility that these conditions might be on the causal pathway from in utero ARV exposures to concomitant ND problems.

Associations between exposures to ATV and risk of multiple ND signals

The results of adjusted proportional odds regression models are presented in Table 4. Among children with ART exposure at conception, there was no evidence that ATV exposure was associated with cumulative odds of more ND signals (single/multiple signals vs. no signal, or multiple signals vs. no/single signal) compared with no exposure to ATV at conception (cOR: 0.66, 95% CI: 0.28–1.56), after adjusting for confounders. For those with maternal ART initiated postconception, ATV exposure was associated with 70% higher odds of more ND signals (cOR: 1.70, 95% CI: 0.82–3.54). Further stratified analysis suggested that the adjusted odds increased more substantially among CHEU whose mothers initiated ART in trimester 1 (cOR = 5.81, 95% CI: 2.41–14.0) (Table 4).

Table 4.

Cumulative Odds Ratios of One or More Dysfunction Signals of In Utero Exposure to Atazanavir-Containing Regimens Compared with Exposure to Other Regimens

	Timing of in utero ARV exposure initiation
	At conception (N = 228)	During pregnancy (N = 458)
		Overall (N = 458)	In trimester 1 (N = 128)	In trimesters 2/3 (N = 330)
Unadjusted cumulative ORa (95% CI)	0.59 (0.29–1.20)	2.01 (0.95–4.26)	5.37 (2.04–14.13)	1.64 (0.66–4.07)
Adjusted cumulative ORb (95% CI)	0.66 (0.28–1.56)c	1.70 (0.82–3.54)d,e	5.81 (2.41–14.02)f	1.42 (0.57–3.51)d,g

^a Cumulative OR is defined as the OR between single/multiple signals versus no signal, and multiple signals versus no/single signal.

^b All GEE proportional odds models adjusted for gender, race, ethnicity, and maternal factors: age at delivery, education, family income at entry visit, FSIQ, family history of reading/language problems; additional maternal factors were adjusted for in models of postconception strata as specified hereunder.

^c Two children with unknown race were excluded due to numerical issues.

^d Adjusted additionally for first trimester marijuana use, first trimester passive cigarette smoking, and first RNA level.

^e Five children with missing birth mother's education, one child with missing family income at entry visit, and one child with unknown race were excluded due to numerical issues.

^f Adjusted additionally for first trimester passive cigarette smoking. In addition, one child with missing family income at study entry and one child with missing mother's education were excluded due to numerical issues.

^g Four children with missing birth mother's education and one child with unknown race were excluded due to numerical issues.

ARV, antiretroviral; CI, confidence interval; GEE, generalized estimating equation; OR, odds ratio.

Table 5 summarizes the specific ART regimens for CHEU included in the analyses already described. The majority of regimens included either a ZDV/3TC backbone (55%) or TDF/FTC backbone (31%). Most ATV-containing regimens, with or without a ritonavir booster, included a TDF/FTC backbone (72%), although a few (six) also included another PI, integrase strand transfer inhibitor (INSTI), or non-nucleoside reverse transcriptase inhibitor (NNRTI).

Table 5.

Most Frequent Antiretroviral Therapy Regimens Received by Mothers During Pregnancy

Timing of in utero ARV exposure initiation
			During pregnancy
Regimen	Total (N = 686)	At conception (N = 228)	Overall (N = 458)	Trimester 1 (N = 128)	Trimester 2/3 (N = 330)
ATV(r)+TDF/FTC	88	41 (18%)	47 (10%)	16 (12%)	35 (11%)
ATV(r)+TDF/FTC+other PI	3	0 (0%)	3 (1%)	1 (1%)	2 (1%)
ATV(r)+TDF/FTC+INSTI/NNRTI	3	2 (1%)	1 (0%)	1 (1%)	1 (0%)
ATV(r)+ZDV/3TC+other PI/NNRTI	13	2 (1%)	11 (2%)	1 (1%)	7 (2%)
ATV(r)+others	23	9 (4%)	14 (3%)	2 (2%)	5 (2%)
TDF/FTC+other PI	73	26 (11%)	47 (10%)	13 (10%)	34 (10%)
TDF/FTC+INSTI/NNRTI	41	27 (12%)	14 (3%)	3 (3%)	11 (3%)
TDF/FTC+others	2	2 (1%)	0 (0%)	0 (0%)	0 (0%)
ZDV/3TC+other PI	235	42 (18%)	193 (42%)	54 (42%)	147 (45%)
ZDV/3TC+INSTI/NNRTI	35	12 (5%)	23 (5%)	9 (7%)	11 (3%)
ZDV/3TC+others	95	20 (9%)	75 (16%)	21 (16%)	58 (18%)
Other non-ATV or backbone	75	45 (20%)	30 (7%)	7 (5%)	19 (6%)
ZDV/3TC backbone	378 (55%)	76 (33%)	302 (66%)	85 (66%)	223 (68%)
TDF/FTC backbone	210 (31%)	98 (43%)	112 (24%)	34 (27%)	83 (25%)
Others	98 (14%)	54 (24%)	44 (10%)	9 (7%)	24 (7%)

Other PI includes: lopinavir/ritonavir, nelfinavir, darunavir, and ritonavir.

ARV, antiretroviral; ATV(r), atazanavir with or without ritonavir booster; INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor; TDF/FTC, tenofovir disoproxil fumarate/emtricitabine; ZDV/3TC, zidovudine/lamivudine.

Relative to regimens with ZDV/3TC backbone but without ATV, TDF/FTC seemed to have a protective effect (Table 6) for children whose mothers were on ART at conception, whereas TDF/FTC was associated with more ND signals for children whose mothers initiated ART during pregnancy, particularly when TDF/FTC was combined with ATV with or without ritonavir booster ATV(r) (adjusted cOR = 2.31, 95% CI: 1.08–4.97). Similar to the analysis of ATV-containing versus non-ATV-containing regimens, further stratified analysis revealed a stronger association with TDF/FTC/ATV(r) among CHEU whose mothers initiated ART in trimester 1 (Table 6).

Table 6.

Cumulative Odds of One or More Dysfunction Signals of In Utero Exposure to Tenofovir/Emtricitabine With or Without Atazanavir Compared Wiith Exposure to Zidovudine/Lamivudine Without Atazanavir

	Unadjusted cumulative OR
	Timing of in utero ARV exposure initiation
		During pregnancy
	At conception (N = 170)a	Overall (N = 399)a	Trimester 1 (N = 116)a	Trimesters 2/3 (N = 296)a
TDF/FTC/ATV(r) vs. Referenceb (95% CI)	0.61 (0.22–1.69)	2.52 (1.22–5.23)	6.63 (2.64–16.65)	1.56 (0.72–3.41)
TDF/FTC+othersc vs. Referenceb (95% CI)	0.43 (0.15–1.25)	1.14 (0.73–1.77)	0.88 (0.22–3.61)	1.21 (0.83–1.77)

	Adjusted cumulative OR (95% CI)d
	Timing of in utero ARV exposure initiation
		During pregnancy
	At conception (N = 169)a	Overalle (N = 394)a	Trimester 1f (N = 112)a	Trimesters 2/3e (N = 293)a
TDF/FTC/ATV(r) vs. Referenceb (95% CI)	0.62 (0.17–2.20)	2.31 (1.08–4.97)	9.15 (3.37–24.88)	1.47 (0.60–3.57)
TDF/FTC+othersc vs. Referenceb (95% CI)	0.40g (0.12–1.42)	1.22h (0.74–2.03)	1.32i (0.27–6.43)	1.28j (0.74–2.24)

^a Some additional children were excluded in adjusted models due to incomplete data or numerical issues. N indicates the number of children included in each proportional odds model after exclusions.

^b ZDV/3TC+other PI or NNRTI or triple nucleoside reverse transcriptase; other PI includes lopinavir/ritonavir, nelfinavir, darunavir, or ritonavir.

^c TDF/FTC+other PI, NNRTI, or INSTI.

^d All models adjusted for gender, ethnicity, and maternal factors: age at delivery, education, family income at entry, full-scale intelligent quotient, and family history of reading/language problems. Additional maternal factors were adjusted for in models of postconception strata as specified hereunder.

^e Adjusted additionally for first trimester marijuana use, first trimester passive cigarette smoking, and first RNA level.

^f Adjusted additionally for any illicit drug use during first trimester and first RNA level.

^g One child with unknown race was excluded due to numerical issues.

^h Two, two, and one children with missing family income at study entry, missing mother's education, and unknown race, respectively, were excluded due to numerical issues.

ⁱ One and four children with missing family income at study entry and mother's marijuana use in trimester 1, respectively, were excluded due to numerical issues.

^j Two and one children with missing birth mother's education and unknown race, respectively, were excluded due to numerical issues.

ARV, antiretroviral; ATV(r), atazanavir with or without ritonavir booster; CI, confidence interval; INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase; OR, odds ratio; PI, protease inhibitor; TDF/FTC, tenofovir disoproxil fumarate/emtricitabine; ZDV/3TC, zidovudine/lamivudine.

Bilirubin levels were evaluated in 50 of 104 children whose mothers were treated with ATV during pregnancy. Among this very limited subset, we did not observe an association of grades 2–4 hyperbilirubinemia with multiple signals of ND problems (data not shown).

Discussion

In this study, we observed a higher proportion of 5-year-old children with concomitant ND problems among those whose mothers initiated/restarted ART after conception compared with those whose mothers were on ART before/at conception. Among children whose mothers initiated/restarted ART after conception, ATV, a currently preferred PI for ART-experienced pregnant women per DHHS perinatal guidelines, was associated with higher risk of ND problems if initiated during the first trimester, a sensitive period of early development.

Owing to our moderate sample size of CHEU whose mothers initiated ART in the first trimester, such findings require ongoing investigation as children in this cohort age. In addition, in this group of CHEU, compared with an NRTI backbone including ZDV/3TC, those who were exposed to TDF/FTC in utero had slightly higher developmental risk, and this risk increased when ATV was included in the regimen.

Our findings suggest that overall this sample of 5-year-old CHEU did not demonstrate significant ND impairment based on our standardized assessments (Table 2). However, we observed increased developmental risk associated with ATV exposure initiated during mother's pregnancy among 5-year-old children in the SMARTT cohort, following earlier observations of an association between ATV exposure and lower language functioning at 1 year of age.^13,18,23

Earlier studies highlight potential mechanisms of risk. For example, Sarkar et al.⁵⁴ reported delays in primitive reflexes of mouse models after in utero exposure to TDF/TFC/ATVr or ABC/3TC/ATVr, suggesting disruption in development of these neurocircuits. In Letendre's⁵⁵ central nervous system (CNS) penetration–effectiveness (CPE) rankings, ATV and TDF/FTC-based regimens had much lower penetration levels than others. Studies^56,57 have found that CPE scores correlated with improvement or decreased neuropsychological impairment in adults living with HIV.

In a review of cord-to-maternal ARV plasma concentration ratio, Cerveny et al.⁵⁸ summarized that the ratio for zidovudine, lamivudine, and emtricitabine ranged from roughly 0.8 to close to 1.8, whereas the range for ATV was 0.12–0.21. These studies suggest the possibility of inadequate or incomplete protection provided by ATV to the developing brain and central nervous system¹³ from potential exposure to HIV. Our observation that the impact among CHEU may extend to the early school-age period and across multiple domains suggests that ongoing surveillance of the safety and short- and long-term impact of ATV and other ARV medications upon developmental outcomes is warranted.

Although ATV exposure appears related to early developmental risk, the possible negative impact of adverse environmental or social exposures, if present, may also contribute in part to the observed increased risk for co-occurring ND effects among young children of mothers living with HIV. For example, mothers who initiated ART during the first trimester were more likely to report substance use, particularly tobacco and passive smoke exposure during the first trimester, and more difficulty caring for their child compared with mothers who initiated therapy before/at conception or in the second/third trimester.

Social or environmental disadvantage may alter neural development through multiple mechanisms that may ultimately contribute to differences in behavioral–emotional, language, and cognitive abilities.^33–38 In addition, unsuppressed viral load, if present early in pregnancy among mothers who initiated ART after conception, may be associated with other maternal health risks that may influence the caregiving environment during children's early years of life, possibly impacting the developmental trajectory of children.

However, we observed much higher protection with exposure to other ARVs in the first trimester; further, the increased risk associated with ATV relative to exposure to other ARVs remained significant after adjusting for these risk factors. Although ATV may be less commonly used among women initiating ART during pregnancy in the United States, a substantial number of children globally have been exposed to ATV. Further study of these children as they age is needed to determine whether similar findings are observed and extend to later stages of child development.

In addition, as the number of pregnant adolescents and young adults living with HIV in the United States increases, sometimes in the context of inadequately controlled viral load and increased substance use,⁵⁹ the choice of ART during pregnancy is critical to reduce the risk of developmental challenges in their offspring.

This investigation has inherent limitations. Given the objective of the SMARTT study to examine the safety of different ARTs within CHEU, we did not evaluate and monitor the neurodevelopment of age and sex-matched HIV-unexposed and uninfected children from similar backgrounds; such studies would potentially inform both our understanding of non-ART/HIV factors that affect early development and our ability to address modifiable risks during later childhood and adolescence. The sample sizes of the study were small, especially for those CHEU whose mother initiated ARV during the first trimester, although there is little extant literature with larger sample sizes.

Cognitive and language evaluations were available in English only, requiring us to restrict our analyses to monolingual English-speaking children. Underlying reasons regarding timing of ARV initiation may have been incompletely captured in the SMARTT database, precluding the possibility of adjusting for all possible confounders for the association between ARV exposure and ND outcomes. Stratified analyses could avoid possible biases or effects related to reasons why ARV was initiated at different periods.

In addition, we had a limited subset with maternal bilirubin levels during pregnancy, did not evaluate mothers' adherence to ARV during pregnancy, and were unable to consider the effects of environmental exposures that may impact early childhood development, such as lead, DDT, ambient air pollution, and/or maternal psychiatric medications during pregnancy. Our results may not adequately reflect ND outcomes among CHEU who do not have access to participation in longitudinal investigations or referral to early intervention services, if needed.

Our results highlight the need for comprehensive ND monitoring of CHEU during early childhood to ensure proper identification of delays, should they exist, and risk factors associated with those delays. In addition, these findings inform future decision making regarding timing and type of ARV treatment before or during pregnancy. ND monitoring also supports early initiation of educational and social support services to ameliorate specific developmental concerns. The presence of early developmental delays, even when observed in multiple domains, does not preclude the possibility of improvement and eventual normal developmental functioning, if identified early.

Children and families are often resilient when challenged, if appropriately supported.⁶⁰ Our analyses were based on relatively moderate sample sizes. Thus, it is important to continue investigation of the timing and safety of in utero exposure to all individual and classes of ARVs that are prescribed, even those prescribed less frequently. Equally important in those investigations is ongoing consideration of the role and quality of the caregiving environment upon ND outcomes among children with perinatal HIV and ART exposure, given their potential impact upon children's well-being.

Authors' Contributions

T.-J.Y. contributed to conceptualization (lead), methodology, formal analysis (lead), writing—original draft (lead), and writing—review and editing (lead). K.M. was involved in conceptualization (supporting), validation, and writing—original draft (lead). J.Z. took charge of formal analysis, software, and writing—original draft (supporting). R.S. carried out validation and writing—original draft (lead). S.R. carried out validation and writing—review and editing (supporting). M.L.R. was in charge of conceptualization (supporting), validation, and writing—review and editing (supporting).

T.F. was in charge of writing—review and editing (supporting). P.T. III carried out writing—review and editing (supporting). C.A.M. took charge of writing—review and editing (supporting). H.J.H. was in charge of writing—review and editing (supporting). P.L.W. was in charge of supervision and writing—review and editing (supporting).

Author Disclosure Statement

The authors declared no conflict of interest.

Funding Information

The study was supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development (NICHD), Office of the Director, National Institutes of Health (OD), National Institute of Dental & Craniofacial Research (NIDCR), National Institute of Allergy and Infectious Diseases (NIAID), National Institute of Neurological Disorders and Stroke (NINDS), National Institute on Deafness and Other Communication Disorders (NIDCD), National Institute of Mental Health (NIMH), National Institute on Drug Abuse (NIDA), National Cancer Institute (NCI), National Institute on Alcohol Abuse and Alcoholism (NIAAA), and National Heart, Lung, and Blood Institute (NHLBI) through cooperative agreements with the Harvard T.H. Chan School of Public Health (HD052102; Principal Investigator: George R. Seage III; Program Director: Liz Salomon) and the Tulane University School of Medicine (HD052104; Principal Investigator: Russell Van Dyke; Co-Principal Investigator: Ellen Chadwick; Project Director: Patrick Davis), and through Harvard T.H. Chan School of Public Health for the PHACS 2020 (P01HD103133; Multiple Principal Investigators: Ellen Chadwick, Sonia Hernandez-Diaz, Jennifer Jao, Paige Williams; Program Director: Liz Salomon).