Safety of In Utero Antiretroviral Exposure: Neurologic Outcomes in Children who are HIV-Exposed but Uninfected

Abstract

OBJECTIVE:

To evaluate whether there is an increased risk of neurologic diagnoses in children who are HIV-exposed but uninfected (CHEU) exposed in utero to specific antiretroviral medications.

DESIGN:

Prospective cohort study of CHEU enrolled from 2007–2017.

METHODS:

We evaluated children for neurologic case status, including microcephaly, febrile seizures, seizure disorders, ophthalmologic disorders, and other neurologic disorders. Adjusted relative risks (aRRs) were estimated for the association between in utero antiretroviral exposure and neurologic case using log-binomial regression, accounting for potential confounders. Sensitivity analyses were conducted to evaluate robustness of findings.

RESULTS:

Among 3747 eligible CHEU, 231 (6.2%) met neurologic case criteria (95% CI:5.4%, 7.0%). Most eligible children (86%) were exposed in utero to combination antiretroviral regimens. In adjusted models, children exposed to efavirenz at any time during pregnancy had higher risk of neurologic case status (aRR=1.53, 95% CI:0.94, 2.51). This association was stronger when comparing efavirenz exposure at conception to no exposure during pregnancy (aRR=1.92, 95% CI:1.09,3.36) and considering follow-up and case diagnosis only through age 2 (aRR=2.14, 95% CI:1.11,4.12). Children exposed to didanosine at conception and during the first trimester had increased risk of neurologic case status (aRR=2.28, 95% CI:1.07,4.87 and aRR=2.02, 95% CI:1.01,4.04, respectively), compared to didanosine-unexposed children. Children with dolutegravir exposure had some suggestion of increased risk of neurologic case (aRR=2.43, 95% CI:0.75,7.84), which was observed consistently across several sensitivity analyses.

CONCLUSIONS:

Efavirenz and didanosine exposure during pregnancy were associated with higher risk of neurologic abnormalities in CHEU, and dolutegravir exposure showed some suggestive associations which warrant further monitoring.

Introduction

With an estimated 1.3 million women living with HIV giving birth annually, it is critical to assess the safety of in utero antiretroviral (ARV) exposure to the fetus.(1–3) There have long been concerns regarding the potential teratogenicity of ARV medications, in particular efavirenz (EFV).(4–7) Multiple studies have now shown that in utero EFV exposure is not associated with an increased risk of central nervous system congenital anomalies (8, 9); however, data on newer ARVs are limited. Indeed the recent experience with peri-conception dolutegravir (DTG) exposure and neural tube defects has highlighted the importance of continued surveillance for potential adverse effects from in utero ARV exposure.(10–12)

Most studies of ARV safety in pregnancy have focused on outcomes at or soon after birth; however, not all adverse effects are evident at birth and data on the risk of more subtle neurologic diagnoses are limited.(13–16) Studies with follow-up beyond one month of life have not demonstrated an increased risk of neurologic diagnoses in children who are HIV-exposed but uninfected (CHEU), but have limited follow-up to the first few years of life or less.(8, 9, 17, 18) (19) Thus, concerns remain regarding adverse neurologic outcomes from possible mitochondrial dysfunction or other mechanisms in children exposed in utero to ARVs. As new ARVs are introduced in pregnancy, the safety of in utero exposure remains a critical public health issue.

The Surveillance Monitoring for ART Toxicities (SMARTT) Study is a prospective observational study established by the Pediatric HIV/AIDS Cohort Study (PHACS) network to identify potential adverse effects of in utero ARV exposure to CHEU.(20) The aim of this analysis was to determine the prevalence of neurologic diagnoses among CHEU and to evaluate whether there is an increased risk of neurologic diagnoses associated with in utero exposure to specific ARVs.

Methods

Study Population

The PHACS SMARTT study includes two cohorts of CHEU: the Static cohort (children enrolled from 2007–2009 ages 1 to 12 years who had participated in prior studies with available pregnancy and birth data) and the Dynamic cohort (enrolled during gestation or within 1 week after birth).(20) The SMARTT study is conducted at 22 clinical research sites in the U.S. including Puerto Rico and enrollment and follow-up are ongoing, with youth followed to age 18 years. The protocol was approved by Institutional Review Boards at each site and at the Harvard T.H. Chan School of Public Health. Written informed consent was obtained from the parent or legal guardian of each enrolled child.

Maternal and child clinical diagnoses and dates of prenatal ARV use were obtained from medical records and participant interview at study entry; retrospectively for the Static cohort. Birth characteristics (gestational age, birth weight, and delivery mode) were abstracted and maternal HIV disease characteristics during pregnancy were collected, including plasma HIV viral load (VL), absolute CD4 lymphocyte (CD4) cell counts, and CD4%. Information on substance use during pregnancy was obtained by self-reported questionnaire, including alcohol, tobacco, marijuana, opiates, and other substances.(21) After enrollment, children and their mothers or caregivers were followed at annual study visits. A complete physical examination was conducted on the child including anthropometric assessments (height, weight, head circumference [HC]), and any new diagnoses or illnesses that occurred since the prior study visit were abstracted from the medical chart or interview. Caregivers completed questionnaires on household composition, education and income levels, history of psychiatric diagnoses or substance use, and other information related to family environment at all study visits.

A study “trigger” for potential neurologic diagnoses was defined as a febrile or afebrile seizure, microcephaly, or other neurologic or ophthalmologic disorders, as previously described.(20) Children meeting any of these triggers either at enrollment (retrospectively for the Static cohort) or at an annual study visit were referred to an appropriate specialist (neurologist or ophthalmologist) for clinical evaluation. The results of the triggered evaluations were reviewed and children classified as cases or non-cases by the SMARTT Review Panel (SRP), a group of clinical and epidemiological experts, using a predefined “neurologic case” definition and blinded to the specific ARV regimens mothers used during pregnancy. The following diagnoses did not require a follow-up assessment by a specialist for confirmation: microcephaly, febrile seizures and diagnoses that were recorded at two or more consecutive study visits (e.g., persisted for ≥2 years). Neurologic diagnoses determined to be secondary to events occurring after birth (e.g. post-natal meningitis, trauma) were not included as cases. A small percentage of participants met a trigger but did not have follow-up assessments needed to determine case status and were considered unevaluable for case determination (Figure 1). Children were deemed eligible for inclusion in this analysis if they enrolled by April 1, 2017 and had a study visit for neurologic trigger assessment by August 1, 2017.

Figure 1.

Derivation of Study Population for Evaluating Incidence of Neurologic Diagnoses and Associations with in utero Antiretroviral Exposures

Outcome measures

As the purpose of this study was to identify any potential signal of a neurological adverse event associated with in utero ARV exposure, the primary outcome of neurologic case was broadly defined and included all of the diagnoses chosen a priori to be included in the definition of a neurologic trigger. Microcephaly was determined based on HC measurements obtained after 6 months of age and was defined as a HC Z-score <−2 for children up to age 3 years based on the Centers for Disease Control and Prevention (CDC) growth curves, and less than the 2^nd percentile according to annual thresholds based on criteria published by Nellhaus for children over 3 years.(22) We accounted for prematurity for children up to age 2 years by subtracting the gestational age in weeks from 40, and adjusting the age at HC measurement by this difference. Children were classified as a neurologic case if they met the criteria at any time during follow-up, while those never having any of the diagnoses of interest were considered to not have the outcome.

Exposure measures

The primary exposure of interest was in utero ARV exposure. Children were classified according to exposure to individual ARV agents, by trimester of maternal ARV initiation, and by pre-conception exposure versus never exposed during gestation. The reference group was children unexposed to the specific ARV agent of interest, either overall or within the time period (trimester or pre-conception) of interest. For example, for 1^st trimester exposure the exposed groups were defined as children of mothers who initiated the ARV agent of interest during the first trimester whereas the unexposed (reference) group were children who were not exposed to the ARV agent during gestation. Individual ARV agents were evaluated among a targeted at-risk subset of participants, based on the calendar year in which the ARV had been approved. Specifically, models were restricted to children born after 2007 for darunavir and raltegravir, after 2011 for rilpivirine, and after 2013 for DTG, and elvitegravir. These restrictions allowed a more appropriate comparison to contemporaneous regimens in use at the time each child was born. All other ARV drugs were evaluated in the full SMARTT cohort.

Potential confounders

We identified potential confounders using directed acyclic graphs (DAGs) based on previous literature and our own study findings. These potential confounders included maternal factors (age, race, ethnicity, chronic health conditions, obstetrical complications and substance use), birth cohort (<2011, 2011–2014, 2015–2017), and family/household factors (socioeconomic status, household income level and caregiver education level). Since maternal ARV use has been demonstrated to be associated with adverse birth outcomes and maternal health measures in pregnancy, preterm birth, low birthweight, cesarean section, maternal HIV VL and CD4 count were not considered as potential confounders (eg., not considered for inclusion in adjusted models) since they may be on the causal pathway between in utero ARV exposure and adverse neurologic outcomes. However, descriptive summaries of neurologic case status by these measures were provided.

Statistical analysis

The percent classified as neurologic cases and the corresponding exact 95% confidence intervals (CIs) were estimated under the binomial distribution. Demographic and maternal characteristics were compared between cases and non-cases using Fisher’s exact test for binary characteristics and Chi-Square tests for categorical characteristics. Log-binomial regression analysis was used to obtain relative risks (RRs) for associations with in utero ARV exposure. Due to the large number of ARV exposures evaluated, covariates included in the final adjusted models for case status were selected based on their association with neurologic case and potential role as a confounder based on the DAGs described above. Missing indicator approaches were used to account for missing information on covariates. A significance level of p=0.05 was employed, but conclusions were based on consistency across primary and multiple sensitivity analyses.

Sensitivity analyses:

A number of sensitivity analyses were conducted to assess the robustness of findings. As follow-up of the cohort is currently ongoing, children born in earlier years generally had more follow-up time during which a neurologic case could occur. We conducted analyses to account for this possible bias: 1) by restricting analysis to children enrolled in the dynamic cohort; 2) by restricting follow-up to the first two years of life (i.e. excluding events after age 2 and children enrolled after age 2); and 3) by fitting Poisson regression models for the incidence rate ratios (IRRs) accounting for person-time of follow-up by each participant. For the first two sensitivity analyses noted above, we utilized modified Poisson regression to obtain estimated RRs and their corresponding 95% CIs.(23) A sensitivity analysis adjusting for preterm birth was also conducted to confirm that associations between ARV exposure and neurologic case were not solely due to prematurity. Sensitivity analyses were also conducted to account for clustering of children within clinical research sites and to account for clustering of children within families (e.g. the same mother or caregiver). Finally, we conducted a sensitivity analysis excluding children with major congenital anomalies based on review by the SRP and outside experts.

Results

Characteristics of study population

A total of 3758 SMARTT participants had enrolled as of April 1, 2017 and had at least one study visit by August 1, 2017. Among these, 11 had an indeterminate neurologic case status and were excluded from analyses, yielding 3747 evaluable participants; their characteristics are shown in Table 1. Just under half were female (48%), 68% Black, and 31% Hispanic. Self-reported tobacco and alcohol use during pregnancy were relatively common (17% and 8%, respectively), and 8% of mothers reported marijuana use during pregnancy; however, reported cocaine/opiates use was relatively rare (3%).

Table 1.

Demographic and Maternal Characteristics of SMARTT study participants by Neurologic Case Status, 2007–2017

Characteristic	Neurologic Case Status		Total (N=3747)	P-Value
	No (N=3516)	Yes (N=231)
Child Characteristics
Duration of follow-up (years), Median (IQR)	4.2 (1.3, 6.9)	5.9 (4.1, 7.9)	4.3 (1.4, 7.0)	<0.001
Birth Cohort				<0.001
< 2002	367 (10%)	24 (10%)	391 (10%)
2002–2006	582 (17%)	48 (21%)	630 (17%)
2007–2010	1,075 (31%)	97 (42%)	1,172 (31%)
2011–2014	960 (27%)	51 (22%)	1,011 (27%)
2015–2017	532 (15%)	11 (5%)	543 (14%)
Female sex	1,681 (48%)	122 (53%)	1,803 (48%)	0.15
Race/origin				0.021
White	916 (26%)	70 (30%)	986 (26%)
Black/African American	2,397 (68%)	141 (61%)	2,538 (68%)
Other	25 (1%)	5 (2%)	30 (1%)
Not reported	178 (5%)	15 (6%)	193 (5%)
Latino/Hispanic	1,082 (31%)	86 (37%)	1,168 (31%)	0.048
Birth characteristics
Low birth weight (< 2500g)	566 (16%)	81 (35%)	647 (17%)	<0.001
Preterm birth (< 37 weeks gestation)	633 (18%)	71 (31%)	704 (19%)	<0.001
Cesarean delivery	1,899 (54%)	142 (61%)	2,041 (54%)	0.056
Caregiver/Household Characteristics
Annual household income < $20,000	2,328 (66%)	157 (68%)	2,485 (66%)	0.64
Caregiver not high school graduate	1,135 (32%)	92 (40%)	1,227 (33%)	0.030
Maternal Characteristics
Age and Marital Status
<25 years at birth of child	1,072 (30%)	67 (29%)	1,139 (30%)	0.61
Single, never married	2,284 (65%)	145 (63%)	2,429 (65%)	0.42
Maternal Immunologic and Virologic Health During Pregnancy
VL > 1000 copies/mL at delivery	463 (13%)	22 (10%)	485 (13%)	0.11
VL > 1000 copies/mL earliest available in pregnancy	1,662 (47%)	115 (50%)	1,777 (47%)	0.49
CD4 < 250 cells/mm3 at delivery	479 (14%)	37 (16%)	516 (14%)	0.28
CD4 < 250 cells/mm3 early in pregnancy	620 (18%)	41 (18%)	661 (18%)	0.93
Maternal Substance Use During Pregnancy
Cocaine/opiates	84 (2%)	11 (5%)	95 (3%)	0.047
Marijuana use	263 (7%)	23 (10%)	286 (8%)	0.20
Alcohol use	286 (8%)	24 (10%)	310 (8%)	0.27
Tobacco use	587 (17%)	65 (28%)	652 (17%)	<0.001
Maternal Sexually Transmitted Infections During Pregnancy
Gonorrhea	87 (2%)	6 (3%)	93 (2%)	0.83
Chlamydia	280 (8%)	19 (8%)	299 (8%)	0.99
Trichomonas	370 (11%)	26 (11%)	396 (11%)	0.99
Syphilis	92 (3%)	6 (3%)	98 (3%)	0.99
Any of above	668 (19%)	43 (19%)	711 (19%)	0.80

^*Combination ARV regimen defined as at least 3 ARV drugs from at least two drug classes

^**Percentages add up to more than 86% because some combination regimens included agents from 2 or more of the following ARV drug classes: PI, NNRTI, and II

IQR=interquartile range (25^th, 75^th percentiles), PI=protease inhibitor, NNRTI=non-nucleoside reverse transcriptase inhibitor, II=integrase inhibitor, VL= HIV viral load.

P-value calculated using Wilcoxon rank sum test for continuous characteristics, Fisher’s exact test for binary characteristics and a Chi-Square test for categorical characteristics. Data were unavailable for some participants for: ethnicity (n=3), birthweight (n=26), gestational age (n=45), delivery mode (n=56), household income (n=255), caregiver education (n=44), maternal ARV regimen (n=94), maternal age at delivery (n=40), marital status (n=44), maternal VL (n=170), maternal CD4 (n=178), substance use during pregnancy (n=218), and sexually transmitted infections (n=148).

Case status and neurologic diagnoses

Of the 3747 participants included, 231 (6.2%; exact 95% CI:5.4%, 7.0%) met criteria for being a neurologic case over a median follow-up of 4.3 years (interquartile range [IQR]:1.4, 7.0). The most common neurologic diagnoses were microcephaly (25.1%), febrile seizure (17.6%), eye-related abnormalities including esotropia, exotropia, strabismus, ptosis, nystagmus, ambylopia, and optic nerve abnormalities (16.5%) and non-febrile seizure (13.5%). A full list of the neurologic diagnoses is provided in Supplemental Table 1. The median age of first neurologic diagnosis was 2.1 years (IQR:1.1, 5.0).

Demographic and maternal characteristics by neurologic case status

Children were more likely to be a neurologic case if they were born in earlier years (≤ 2010), of white race or Hispanic ethnicity, with low caregiver education, and if their mother reported use of tobacco, alcohol, marijuana, or cocaine/opiates during pregnancy (Table 1). In multivariable models, tobacco use during pregnancy emerged as the substance most strongly associated with neurologic case status and, due to high rates of concurrent tobacco use, other substances (alcohol, marijuana, cocaine/opiates) were no longer related to case status after accounting for tobacco use. Low caregiver education was also no longer associated with case status in multivariable models after adjusting for the other demographic characteristics and sociodemographic measures. The final adjusted model included Hispanic ethnicity, tobacco use during pregnancy, and birth cohort (2011–2014 and 2015–2017 vs. pre-2011).

Association of in utero antiretroviral exposures with neurologic case status

In adjusted models, exposure to EFV was associated with a 53% increased risk of neurologic case (adjusted RR (aRR)=1.53, 95% CI:0.94, 2.51) (Table 2, Figure 2). There was a stronger association for EFV exposure at conception (aRR=1.92, 95% CI:1.09, 3.36) (Supplemental Table 2). In addition, didanosine (ddI) exposure at conception (aRR=2.28, 95% CI:1.07, 4.87) was associated with an increased risk of being a neurologic case. DTG exposure was associated with over two-fold increased risk of being a neurologic case (aRR=2.43, 95% CI:0.75, 7.84). The magnitude of association for DTG was stronger for exposure during the first trimester (aRR=2.95, 95% CI:0.79, 11.01) and at conception (aRR=3.47, 95% CI:0.74, 16.36).

Table 2.

Association of in utero ARV Exposures with Neurologic Case Status in SMARTT Cohort

Antiretroviral Drug or Class	Event Rates				Estimated Relative Risks
	N	% Exposed	Cases, exposed	Cases, unexposed	Unadjusted RR (95% CI)	P-value	Adjusted RR (95% CI)	P-value
By ARV Drug Class or Regimen
HAART	3623	86.0	192/3116 (6.2%)	33/507 (6.5%)	0.95 (0.65, 1.37)	0.77	1.30 (0.86, 1.97)	0.22
NRTIs	3653	97.4	218/3558 (6.1%)	8/95 (8.4%)	0.73 (0.36, 1.47)	0.38	0.88 (0.39, 1.99)	0.77
NNRTIs	3653	18.6	42/678 (6.2%)	184/2975 (6.2%)	1.00 (0.72, 1.39)	0.99	1.20 (0.86, 1.66)	0.28
PIs	3653	69.9	166/2553 (6.5%)	60/1100 (5.5%)	1.19 (0.89, 1.60)	0.24	1.09 (0.80, 1.49)	0.57
IIs	2492	15.7	13/391 (3.3%)	122/2101 (5.8%)	0.57 (0.32, 1.01)	0.056	0.88 (0.48, 1.61)	0.68
Nucleoside Reverse Transcriptase Inhibitors (NRTIs)
Lamivudine	3653	61.5	146/2245 (6.5%)	80/1408 (5.7%)	1.14 (0.88, 1.49)	0.32	0.89 (0.67, 1.18)	0.42
Abacavir	3653	17.7	38/647 (5.9%)	188/3006 (6.3%)	0.94 (0.67, 1.32)	0.72	0.92 (0.65, 1.30)	0.63
Stavudine	3653	2.8	7/104 (6.7%)	219/3549 (6.2%)	1.09 (0.51, 2.32)	0.82	0.91 (0.43, 1.95)	0.81
Didanosine	3653	2.5	9/92 (9.8%)	217/3561 (6.1%)	1.61 (0.82, 3.13)	0.16	1.36 (0.67, 2.77)	0.40
Emtricitabine	3653	39.1	70/1427 (4.9%)	156/2226 (7.0%)	0.70 (0.53, 0.92)	0.011	0.99 (0.73, 1.34)	0.93
Tenofovir DF	3653	41.6	79/1520 (5.2%)	147/2133 (6.9%)	0.75 (0.58, 0.98)	0.037	1.00 (0.75, 1.34)	0.97
Zidovudine	3653	61.5	149/2247 (6.6%)	77/1406 (5.5%)	1.21 (0.93, 1.58)	0.16	0.90 (0.67, 1.20)	0.48
Non-nucleoside Reverse Transcriptase Inhibitors (NNRTIs)
Efavirenz	3653	4.5	15/166 (9.0%)	211/3487 (6.1%)	1.49 (0.91, 2.46)	0.12	1.53 (0.94, 2.51)	0.090
Nevirapine	3653	6.8	21/250 (8.4%)	205/3403 (6.0%)	1.39 (0.91, 2.14)	0.13	1.23 (0.79, 1.92)	0.36
Rilpivirine	1261	19.2	4/242 (1.7%)	44/1019 (4.3%)	0.38 (0.14, 1.07)	0.066	0.47 (0.17, 1.31)	0.15
Protease Inhibitors (PIs)
Atazanavir	3653	18.8	38/685 (5.5%)	188/2968 (6.3%)	0.88 (0.62, 1.23)	0.44	1.02 (0.72, 1.44)	0.90
Darunavir	2492	13.0	14/325 (4.3%)	121/2167 (5.6%)	0.77 (0.44, 1.34)	0.36	0.89 (0.51, 1.57)	0.69
Indinavir	3653	1.3	1/47 (2.1%)	225/3606 (6.2%)	0.34 (0.05, 2.43)	0.28	0.30 (0.04, 2.12)	0.23
Lopinavir/r	3653	27.7	69/1012 (6.8%)	157/2641 (5.9%)	1.15 (0.87, 1.51)	0.33	1.05 (0.80, 1.39)	0.73
Nelfinavir	3653	16.3	46/594 (7.7%)	180/3059 (5.9%)	1.32 (0.96, 1.80)	0.084	1.02 (0.73, 1.42)	0.90
Saquinavir	3653	2.3	8/85 (9.4%)	218/3568 (6.1%)	1.54 (0.76, 3.12)	0.23	1.39 (0.68, 2.83)	0.37
Integrase Inhibitors (IIs)
Dolutegravir	782	12.0	4/94 (4.3%)	17/688 (2.5%)	1.72 (0.58, 5.12)	0.33	2.43 (0.75, 7.84)	0.14
Elvitegravir	782	13.8	2/108 (1.9%)	19/674 (2.8%)	0.66 (0.15, 2.82)	0.57	0.70 (0.16, 3.04)	0.63
Raltegravir	2492	8.0	7/199 (3.5%)	128/2293 (5.6%)	0.63 (0.29, 1.35)	0.23	0.71 (0.33, 1.53)	0.38

Relative risks estimated by log-binomial regression for full SMARTT cohort and by modified Poisson regression for other subgroups.

Adjusted model includes Hispanic ethnicity, tobacco use during pregnancy, and birth cohorts (2011–2014 and 2015–2017 vs <2011).

Models restricted to children born after 2007 for darunavir, raltegravir, and IIs (as drug class), after 2011 for rilpivirine, and after 2013 for dolutegravir and elvitegravir; all other ARV drugs are evaluated in the full SMARTT cohort.

Figure 2. Adjusted relative risks for associations of in utero ARV exposures with neurologic case, based on any exposure during pregnancy and first trimester exposures.

Adjusted relative risks (aRRs) and their confidence intervals (CIs) were estimated by log-binomial regression for full SMARTT cohort and by modified Poisson regression for other subgroups. Adjusted models include Hispanic ethnicity, tobacco use during pregnancy, and birth cohorts (2011–2014 and 2015–2017 vs <2011. Models restricted to children born after 2007 for darunavir and raltegravir, after 2011 for rilpivirine, and after 2013 for dolutegravir and elvitegravir; all other ARV drugs were evaluated in the full SMARTT cohort.

Sensitivity analyses

Analyses restricting to those enrolled at birth, accounting for clustering within research site and within the same family (same mother or caregiver), considering only the first two years of life, excluding children with major congenital anomalies, accounting for person years of follow-up and adjusting for prematurity for associations between EFV, ddI or DTG exposure (any exposure in pregnancy and 1^st trimester exposure only) with neurologic case status are shown in Figure 3. Results of the unadjusted analyses and sensitivity analyses for all ARVs are provided in Supplemental Tables 3–10. Of note, the magnitude of association between in utero EFV exposure and neurologic case status was similar to that seen in the primary analyses across all sensitivity analyses and was statistically significant when restricting to the first two years of life for any EFV exposure or first trimester exposure. The magnitude of association between in utero DTG exposure and neurologic case status also remained similar across all sensitivity analyses to that seen in the primary analyses and was statistically significant for first trimester DTG exposure after accounting for person years of follow-up and after accounting for clustering within site.

Figure 3. Sensitivity Analyses for Associations of Efavirenz, Dolutegravir, and Didanosine with Neurologic Case Status.

Adjusted relative risks (aRRs) are based on modified Poisson regression models and adjusted incidence rate ratios (aIRRs) are based on standard Poisson regression models accounting for person time; both types of models adjust for Hispanic ethnicity, tobacco use during pregnancy, and birth cohort (2011–2014 and 2015–2017 vs <2011). Models accounting for clustering within site or within the same mother/family are fit using generalized estimating equation models with an assumed exchangeable correlation structure. FUP=follow-up.

Discussion

We observed a high percentage of children who developed one or more neurologic diagnoses in our cohort compared to previous studies (6.2% vs. 0.3–2%).(13, 14, 16, 24) This may be in part because our study was designed to detect a broad range of adverse neurologic events, including microcephaly and eye-related abnormalities, which previous studies have not included, and because our study included longer follow-up of infants and children.

Our findings demonstrate that infants with in utero EFV exposure had over 50% increased risk of having a neurologic diagnosis. This association persisted in multiple sensitivity analyses. In utero DTG exposure was associated with an increased risk of having a neurologic diagnosis but the associations were not statistically stable except in specific sensitivity analyses due to the small number of exposed cases. The data on the association between in utero ARV exposure and neurologic diagnoses in CHEU have been conflicting; however, studies to date have been limited by their duration of follow-up, sample size and ability to adjust for confounders. The French Perinatal Cohort Study (FPCS) found that CHEU exposed to a non-nucleoside reverse transcriptase inhibitor combination, most commonly zidovudine + lamivudine, were more likely to be classified as having mitochondrial dysfunction compared to CHEU exposed to zidovudine monotherapy (RR=2.5; 95% CI:1.0, 6.5) with the majority of cases characterized by neurologic symptoms (86%); however, this analysis did not adjust for potential confounders.(14) A follow-up study of the FPCS found that CHEU exposed to EFV in utero had a higher risk of neurologic congenital anomalies but this was only statistically significant in a secondary analysis when applying a broader classification of neurologic congenital anomalies.(5) The IMPAACT 219/219C study demonstrated an increased risk of mitochondrial dysfunction associated with third trimester initiation of lamivudine alone and in combination with zidovudine.(16) In addition, ddI exposure during the peri-conception period and during the first trimester was found to be associated with an increased risk of having a neurologic diagnosis.

Subsequent studies in other cohorts have failed to confirm this potential increased risk of neurologic diagnoses associated with in utero ARV exposure; however, most of these studies have been limited by either their duration of follow-up, restricted outcome definition to only cases that resulted in death, or limited ability to compare various ARV regimens.(17, 18, 24–29) Experience with other medications has demonstrated the importance of long-term follow-up when evaluating the safety of medications administered in pregnancy.(19) Indeed, results from the FPCS have shown that symptoms of mitochondrial dysfunction are usually not present in the first few months of life. We followed children in our cohort for a median of 4 years allowing us to detect potential safety signals later in childhood. The median age of first confirmation of neurologic case status was 2.1 years, suggesting that a large proportion of cases would be missed if follow-up were restricted to the first few months of life.

Additionally, when conducting analyses of drug safety in pregnancy it is crucial to have the correct comparator groups. Comparing ARV-exposed to ARV-unexposed CHEU raises issues of confounding by indication (e.g. worse maternal HIV disease status leading to ARV initiation in the ARV-exposed group in earlier years of SMARTT). Comparing one ARV-exposed group to another can raise issues of confounding by historical bias if the years of exposure differ greatly between the two groups. Given our large sample size and the high proportion of in utero ARV-exposure in our cohort, we were able to address these issues by restricting our analyses to the specific time-period when the ARV of interest was approved for use.

Febrile seizures was one of the most commonly encountered neurologic diagnosis in our study (17.6%) with an overall prevalence of 1.3%. The prevalence in our cohort is lower than that found in the general U.S. population (2–5%); however, this does not necessarily negate an association with specific ARV exposures. Indeed, the FPCS observed a higher cumulative risk of febrile seizures by 18 months in CHEU with in utero and/or postnatal ARV exposure compared to CHEU without ARV exposure (11.0 vs. 4.1 per 1000, p=0.02).(15) Their study observed no association with specific ARV exposures, but was conducted early in the epidemic when the number of ARVs available was limited.

Our study had several limitations. First, as mentioned above, in an observational study, both known and unknown confounders could be present. We adjusted for several confounders and restricted the analyses to the time-period when the ARV of interest was approved for use. Additionally, we systematically identified and confirmed our neurological cases blinded to ARV exposure. As many participants enrolled after birth, we may have missed some major neurologic congenital anomalies that were detected on prenatal ultrasound and subsequently resulted in either spontaneous or therapeutic abortion. Missing cases due to pregnancy termination would have biased our findings towards the null and excluding children with major congenital anomalies did not change the magnitude of the associations seen in our study. Our analysis was implemented as a screening study rather than a test of any single hypothesis, with a goal of identifying potential signals of toxicity. We recognize that this approach involves multiple comparisons and may increase the risk of false positive findings. At the same time, we did not want to fail to detect true associations. Key to this screening strategy is an emphasis on consistency and robustness of findings across multiple sensitivity analyses, along with the need to confirm any potential associations in further more focused studies. We also acknowledge that sensitivity analyses conducted to examine different exposure periods do not reflect independent analyses, given the overlap in ARV-exposed cases between time periods of interest; these time-specific risk estimates must be interpreted cautiously, but can help inform whether the effect of ARV exposure differs by timing. Another limitation was that smaller numbers of children were exposed to newer ARV agents, which limited our ability to detect statistically significant signals for these agents; our data suggest a potential association between DTG and neurologic case status emphasizing the importance of continued surveillance of CHEU exposed to DTG and other new ARV agents. Finally, in an attempt to be inclusive and not miss any potential safety signals our definition of neurologic case included all potential neurologic and ophthalmologic diagnoses, which precludes us from making any conclusions about causal effects and potential long-term consequences. Despite the broad outcome measure, we feel that our results are important as they demonstrate potential safety concerns associated with specific in utero ARV exposures. Given the fact that our outcome measure represents different etiologies with potentially different causal mechanisms and severity of presentation, it is important that our findings be confirmed with larger sample sizes within specific exposure groups of interest.

In conclusion, we found that in utero EFV and ddI exposure were associated with higher risks of neurologic abnormalities in infancy and childhood. There was a suggestion of an association between in utero DTG exposure and higher risk of neurologic abnormalities in infancy and childhood but this association was not statistically stable due to the recent introduction of DTG and the small numbers of cases to date. The recommended antiretroviral therapy (ART) in pregnancy has recently been called into question by reports of increased risk of neural tube defects in CHEU exposed to DTG.(10) Statistical modelling suggests that despite this increased risk of neural tube defects DTG-based ART in pregnancy would avert more maternal deaths than EFV-based ART.(30) Our findings highlight the importance of continuing to proactively evaluate the safety of ART in pregnancy and the need for long-term follow-up of CHEU.(31) Further study is needed to confirm our findings in other cohorts, as well as to determine the clinical implications of the neurologic diagnoses and investigate the underlying mechanisms of the associations seen in our study.(32, 33)