RESULTS
Among 500 participants, 436 (87%) had single, non-missing, live birth
outcomes. We excluded 20 participants who experienced miscarriages, 11
who had stillbirths, seven with twin births, and 26 participants who
were lost to follow-up (Figure 1 ). The distribution of
miscarriages and stillbirths was similar across study arms. Baseline
characteristics between the analytical sample and the 26 with missing
outcome data were similar in terms of age, relationship status,
education, income, HIV status, partner HIV status, primigravida, history
of a preterm birth or low birth weight outcome, hypertension,
nationality, and alcohol use. However, the standard-of-care group was
more likely to have missing birth outcome data (8.7%) compared to the
intervention group (2.6%). Our study team was able to contact 11 of the
26 participants missing birth outcome data by phone and determined that
they were now living outside of Gaborone.
As seen in Table 1 , the median age in our analytical sample was
27 years, the median gestational age at first antenatal care visit was
17 weeks, about half of the participants were married or cohabiting,
most women were citizens of Botswana, 7% had a recorded measure of
hypertension, 17% were living with HIV, and the median number of
antenatal care visits was six. Characteristics that differed
statistically between intervention and standard-of-care arms included
education level, income, citizenship, HIV status, first pregnancy,
partner’s HIV status, clinic, and study period (p\(\leq\)0·05 for all).
The overall prevalence was 13.3% for the composite outcome of preterm
birth and/or low birthweight, 9.4% for preterm birth, and 8.5% for low
birthweight. Without controlling for potential group imbalance, the
prevalence of preterm birth or low birthweight was not significantly
different between the intervention (14%) and standard-of-care arms
(13%).
Results from the multivariable logistic regression can be found inTable 2 . In all tested models, exposure to the intervention arm
was associated with reduced odds of the composite outcome of preterm
birth or low birthweight after controlling for primigravida, antenatal
care visits, hypertension, and clinic site; however, the confidence
intervals crossed one. Further, odds ratios were similar when limiting
the sample to participants in cross-over clinics that received the
intervention and standard-of-care assignment over time.
Using post-estimation analysis from the multivariable logistic
regression model, the predicted prevalence of the composite outcome of
preterm or low birth weight was higher in the standard-of-care (15%)
group compared to the intervention group (11%) (Figure 2 );
however, the adjusted risk ratio (ARR) had confidence intervals that
crossed one (ARR: 0.67; 95% CI: 0.38 to 1.17). Results were similar for
preterm birth (ARR: 0.65; 95% CI: 0.33 to 1.28) and low birth weight
(ARR: 0.57; 95% CI: 0.28 to 1.16).
In the posthoc analysis, the stratification reduced some of the
differences between the intervention and standard of care groups. Among
nulliparous participants, the groups were now balanced on nationality,
HIV status, and time period; however, the groups continued to differ on
education level, income, and partner’s HIV status. Although the HIV
prevalence was 5% among nulliparous participants, 18% of participants
said their partners were living with HIV and 29% did not know their
partner’s status. In the multiparous group, the intervention and
standard of care arms differed on income level, nationality, and partner
HIV status. The HIV prevalence among multiparous women was 24%, and
20% reported that their partners were living with HIV and 31% did not
know. Table 2 also provides results from the multivariable
logistic regressions stratified by prior live birth. After controlling
for hypertension and clinic, the intervention reduced the odds of the
composite outcome, preterm and low birth weight, among the nulliparous
group. However, no effect was found among multiparous women and the odds
ratios switched directions.