Does Changing Antiretroviral Therapy in the First Trimester of Pregnancy for Safety Concerns Have an Impact on Viral Suppression?

Violaine Peyronnet, MD; Josiane Warszawski, MD, PhD; Jeanne Sibiude, MD, PhD; Olivia Dialla, MCS; Agnès Bourgeois-Moine, MD; Eida Bui, MD; Caroline Simon Toulza, MD; Delphine Peretti, MD; Cécile Brunet-Cartier, MD; Véronique Avettand-Fenoel, MD, PhD; Jérôme L. Chenadec, MD, PhD; Albert Faye, MD, PhD; Roland Tubiana, MD; Laurent Mandelbrot, MD; for the ANRS EPF-French Perinatal Cohort Study


J Acquir Immune Defic Syndr. 2019;80(5):574-584. 

In This Article


The French Perinatal Cohort

The French Perinatal Cohort (ANRS EPF CO1/CO11) is a prospective, observational study ongoing since 1985 in 90 perinatal centers throughout France,[7] which includes all pregnancies in women with HIV-1 and/or HIV-2 with their informed consent and ethics committee (Comité de Protection des Personnes) approval. Clinicians record maternal and obstetrical data and follow-up of the infant until 2 years of age if not infected or 18 years if HIV-infected. EPF coverage is estimated at 70% of pregnancies of HIV-positive women in France. The cohort comprised 2 components, CO1 and CO11. More detailed data are collected in the CO1 component performed in selected study sites.

Study Population

For the current analysis, we included all pregnancies in EPF CO01, for which the outcome occurred at 14 week gestation (WG) or more, from January 1, 2005, to December 31, 2015. We selected only women who were already on antiretroviral therapy at the time of conception and had a HIV-1 plasma viral load available before 14 week gestation (WG), which was <50 copies/mL. Women with HIV type 2, for whom the recommendations are different, were excluded (N = 188).


All treatment regimens were collected with the initiation and end dates for each drug. We defined a treatment switch as the change of at least one drug, that is, the addition, modification, or deletion of a drug in the combination of treatments present at conception. Changes in dose, as well as treatment interruptions, were not considered as switches. The type of treatment at conception was classified into 3 categories: recommended as first-line during the pregnancy, alternative, and not recommended (NR) in the first trimester of pregnancy, according to the French national guidelines at the time of conception for each pregnancy.[2–6] In the category of "NR" treatments, we included both contraindicated treatments and those with insufficient data in pregnancy to recommend. This classification was made considering for each treatment regimen the individual drugs and their combinations during the pregnancy (see Table A, Supplemental Digital Content,

The guideline periods were 2005–2006, 2007–2008, 2009–2010, 2011–2013, and 2014–2015.

Efavirenz was not recommended throughout the study period, as were the other NNRTIs, except for nevirapine started before conception. The integrase inhibitors (raltegravir and dolutegravir) were all in the NR group until the most recent guidelines. For NRTIs, tenofovir was in the insufficient data group until 2011, and throughout the study period, it was recommended not to use solely NRTIs, whether monotherapy, double therapy, or a triple NRTI regimen.

We also studied treatment changes beyond the first trimester, whether or not the ART regimen was switched in the first trimester.


We studied the characteristics potentially associated with treatment change, ie, sociodemographic and behavioral characteristics of the mother, clinical and obstetrical factors, HIV infection characteristics, and the antiretroviral therapy history.

We studied outcomes following switches in the first trimester of pregnancy (<14 WG). The main outcome was maternal viral load nearest to the time of delivery and also during the pregnancy to study the first occurrence of a value above a cutoff of 50 copies/mL.[30] Other pregnancy outcomes studied were the mode of delivery, preterm delivery (<37 WG), and adverse outcomes defined as neonatal deaths, terminations of pregnancy, or in uterofetal demise. Infant outcomes studied were HIV transmission and mortality before 1 year of age.

Statistical Analysis

The main analysis was restricted to first-trimester changes for safety concerns in HIV-1–infected patients on ART at conception with viral suppression documented by a viral load <50 copies/mL before 14 WG, after exclusion of changes for inefficacy (N = 23) or intolerance (N = 22) (EPF CO1, N = 1780). We examined the factors potentially associated with a change in treatment before 14 week gestation. For this, we performed χ2 or Fisher exact test and logistic regressions. Multivariate models were developed by including the noncollinear variables found associated with univariate P <20% after looking for potential interactions. A propensity score was calculated, defined by the probability of early change for safety concerns as a function of the initial characteristics, estimated by the final logistic regression model. To construct this score, we used the variables significantly associated with the probability of changing treatment after eliminating potential colinearities and interaction. Verification steps (analysis of the standardized differences and the disappearance of the association between each variable independently with the change after adjustment on this score[31]) were performed to verify the good performance of the propensity score.

Finally, we studied the association between 1st trimester changes for safety concerns and virological control, complications of pregnancy, and neonatal outcomes. For this purpose, we performed univariate and multivariate analysis, using logistic regressions for delivery outcomes or Cox models for outcomes occurring during pregnancy. In case of additional treatment changes, time for survival analysis was censored at the date of the second change. We performed additional subgroup analyses according to the treatment recommendations: recommended, alternative, or NR.

Multivariate analysis was performed with adjustment for the propensity score and factors found to be associated with outcomes in the univariate analysis at P level of 0.20. To account for the effect of enrolling woman more than once for successive pregnancies, we adjusted our analyses on the number of previous pregnancies in the cohort.

All analyses were performed using STATA 14 software.