Risk Factors for Adverse Birth Outcomes in the PROMISE 1077BF/1077FF Trial

Dorothy Sebikari, MBChB, MPH; Mona Farhad, MS; Terry Fenton, EdD; Maxensia Owor, MBChB; Jeffrey S. A. Stringer, MD; Min Qin, PhD; Nahida Chakhtoura, MD; Benjamin H. Chi, MD, MSc; Friday Saidi, MBBS; Neetal Nevrekar, MD; Avy Violari, MD; Tsungai Chipato, MBChB, MCE; James A. McIntyre, MBChB, FRCOG; Dhayendre Moodley, PhD; Taha E. Taha, MBBS, MCM, MPH, PhD; Gerhard Theron, MD; Mary Glenn Fowler, MD, MPH


J Acquir Immune Defic Syndr. 2019;81(5):521-532. 

In This Article


Study Setting and Population

PROMISE 1077BF/1077FF was a multicomponent randomized trial conducted at 14 sites in 7 countries (6 in sub-Saharan Africa and 1 in India). The present analysis focused on the antepartum component of the trial, the design and findings of which have been reported elsewhere.[11] This study enrolled HIV-infected pregnant women of at least 14-week gestation who did not meet clinical or CD4+ T-lymphocyte (CD4) count requirements for treatment initiation based on country guidelines (usually 350 cells/mm3). We excluded women with previous ART exposure, although previous receipt of 1 or 2 antiretroviral drugs to prevent perinatal HIV transmission in a previous pregnancy was permitted, as was 30 or fewer days of prerandomization exposure during the current pregnancy. We excluded those with entry hemoglobin concentration <7.5 g/dL, serious laboratory abnormalities based on DAIDS Toxicity Tables, 2004,[12] active tuberculosis or recent TB treatment, hepatitis B (HBV) treatment, and pregnancies where fetus(es) had a serious malformation.

In the antepartum component of PROMISE, women were randomly assigned to 1 of 3 regimens/study arms: (A) zidovudine (ZDV) plus intrapartum single-dose nevirapine ("ZDV alone"); (B) zidovudine/lamivudine and lopinavir–ritonavir ("ZDV-based ART"); or (C) tenofovir (TDF), emtricitabine, and lopinavir–ritonavir ("TDF-based ART"). Under protocol versions 1.0 and 2.0, hereafter called period 1, women who tested negative for hepatitis B surface antigen (HBsAg) were eligible to be randomized into study arms A and B only, while those who tested HBsAg-positive were eligible for any of the 3 arms. Under protocol version 3.0, hereafter referred to as period 2, which began in October 2012, participants were randomized with equal probability into any of the 3 study arms, irrespective of HBsAg status. This modification was made in response to evolving treatment guidelines regarding TDF safety for pregnant women.

Outcomes and Definitions

Maternal trial participants were evaluated at 2 and 4 weeks after enrollment and thereafter every 4 weeks until delivery. We used a modified Ballard newborn assessment[13,14] as the primary approach to estimate gestational age (GA) at delivery. We defined preterm delivery (PTD) and very PTD (VPTD) delivery as <37- and <34-week gestation at birth, respectively, and defined LBW and very LBW (VLBW) as <2500 and <1500 g, respectively. We also defined a composite adverse pregnancy outcome as any of the following: PTD, LBW, spontaneous abortion (<20-week gestation), or stillbirth (born dead without heart rate or respiratory effort on or after 20-week gestation). We further defined a severe composite adverse pregnancy outcome to include the following: VPTD, VLBW, spontaneous abortion, or stillbirth.

For multiple births, if any of the infants met the criteria for an adverse or severe adverse outcome, the pregnancy was classified as having the corresponding outcome, either on the single outcomes evaluating prematurity or birth weight or on the composite outcomes.

Design of the Present Analysis

This study is a secondary analysis designed to investigate demographic, baseline clinical, and postentry obstetrical risk factors associated with PTD and LBW, along with the composite outcomes defined above. Note that the analyses for PTD, VPTD, LBW, and VLBW were limited to pregnancies with at least one live birth (N = 3333), while the analyses with the composite outcomes also included 90 singleton pregnancies whose only outcomes consisted of stillbirths or spontaneous abortions, yielding a total sample of N = 3423. These analyses assessed the independent effects of known and suspected risk factors on adverse pregnancy outcomes, and evaluated the extent to which they might have mediated the ART treatment effect on adverse birth outcomes.

Predictor variables included in this secondary analysis were as follows:

  • Baseline maternal clinical and demographic factors: maternal age, maternal body mass index (BMI), GA, country, treatment group, CD4 count, HIV viral load, multiple gestation, history of previous preterm birth, history of cigarette smoking, history of alcohol use, and HBsAg status. All were assessed at entry except for maternal age, which was assessed at delivery.

  • Maternal obstetrical risk factors identified postrandomization and throughout the pregnancy: abruptio placentae, placenta previa, chronic hypertension, pregnancy-induced hypertension, polyhydramnios, oligohydramnios, intrauterine growth restriction, preterm labor, premature rupture of membranes, vaginal bleeding, lower genital tract infection, and urinary tract infection.

Statistical Analysis

The overall strategy of the data analysis was to (1) perform univariate logistic analyses to identify variables meeting a criterion of at least marginal association (P value < 0.15) with one or more of the predefined adverse pregnancy outcomes; (2) enter these variables into multivariate logistic models as predictors of pregnancy outcomes and examine their associations with these outcomes, controlling for one another (note: obstetrical risk factors with N < 5 events were excluded from multivariate analyses); (3) use a backward elimination procedure to sequentially remove the least significant variable from the model, until only those with P value ≤ 0.10 remained; (4) enter the variables retained after the backward elimination procedure into models that restrict the data to participants accrued during period 2 of the trial where there was equal randomization to each of the 3 treatment arms. This allowed us to examine whether the effects of these predictors in the period 2 analyses were consistent with those found to be at least moderately associated with 1 of these outcomes in the full sample models covering period 1 and period 2 (P value < 0.10).