The effect of intrapartum antibiotics on early-onset neonatal sepsis in Dhaka, Bangladesh: a propensity score matched analysis

Neonatal infections - including sepsis, pneumonia, and meningitis - account for approximately 23.4% of the world’s 3.1 million neonatal deaths each year [1]. In developing countries, where 99% of neonatal deaths occur, up to 42% of infection related deaths occur in the first week of life [2]. This narrow time period provides only a small window of opportunity for interventions.

In Bangladesh, the incidence of clinical sepsis during the first week of life defined by the World Health Organization (WHO) Young Infants criteria for very severe disease [3] was 13.4% with a case-fatality of 10.2% [4] and the incidence of community-acquired neonatal bacteremia was 1.4 per 1000 live births [5]. The most common pathogen isolated was S. aureus[5].

Maternal infections and risk factors for infection or colonization increase the possibility of early-onset neonatal infections by vertical transmission [6]. Several interventions have been proposed to decrease the transmission of bacterial pathogens from the mother to newborn, particularly in preventing Group B Streptococcus (GBS) early-onset neonatal sepsis. For example, vaccines against the nine identified GBS stereotypes have been developed [7] and are currently being tested [8]. Another strategy is vaginal washes with chlorhexidine, which have been shown to reduce GBS bacterial load but did not affect early-onset sepsis [9, 10]. The most commonly used intervention is antibiotics given during the intrapartum period.

Administration of antibiotics during the intrapartum period (most often penicillin) decreases vaginal GBS colony counts [11] and is thought to decrease the concentration of bacteria in the maternal bloodstream and amniotic fluid. Currently, in high income countries, indications for intrapartum antibiotic prophylaxis to prevent GBS early-onset neonatal sepsis are based on universal culture-based screening [12]. After implementation of the initial guidelines in 1996, a decreasing GBS incidence has been observed over time (1.7 per 1000 live births in 1993 compared to 0.34–0.37 per 1000 in recent years) [12‐14].

Although the use of antibiotics during the intrapartum period has been widely adopted in high income countries, the evidence supporting antibiotic use derives mainly from cohort studies. There are limited data from randomized controlled trials and most are on GBS. A recent Cochrane review on antibiotics during the intrapartum period for known GBS maternal colonization identified only four randomized controlled trials, most of which were from the 1980s to early 2000s. Intrapartum antibiotics appeared to reduce GBS early-onset sepsis, however these findings may have been the result of a high risk of bias in the studies. The review concluded there was insufficient evidence to recommend intrapartum antibiotics to reduce GBS early-onset neonatal sepsis [15]. Furthermore, in regions like South Asia where the incidence of GBS early-onset sepsis was 0.02 per 1000 live births [16], it is unclear whether intrapartum antibiotics would reduce sepsis from other organisms.

Because there is a lack of randomized controlled trials and a dearth of data in particular from developing countries, we estimate the causal effect of intrapartum antibiotics on early-onset neonatal sepsis using propensity score matching in a cohort study of mother-newborn pairs in Dhaka, Bangladesh. Understanding the effect of intrapartum antibiotics on early-onset neonatal sepsis may lead to strategies to prevent sepsis and its associated morbidities and mortality globally.

Several baseline characteristics were associated with receipt of intrapartum antibiotics and early-onset neonatal sepsis (Table 1). Factors associated with intrapartum antibiotic use included receipt of antenatal care from physicians (43.8% vs. 22.3%, p = 0.001), homes with roofs made of concrete, brick, or cement (33.3% vs. 21.9%, p = 0.07), drinking water sources from the tap rather than tube well (66.7% vs. 54.9%, p = 0.08), upper quintile of wealth (33.3% vs. 21.4%, p = 0.05), and rupture of membranes at presentation (60.4% vs. 39.0%, p = 0.005). Characteristics associated with early-onset sepsis were prematurity (16.9% vs. 10.1%, p = 0.06), colonized mothers (45.5% vs. 35.2%, p = 0.08), and homes with drinking water sources from a tube well rather than tap (53.3% vs. 42.6%, p = 0.08).

Across the three matching methods considered, full unconstrained matching had the best overall balance across the covariates. After matching, the absolute standardized biases ranged from −0.19 to 0.18. The variable with the maximum standardized difference (−0.19) was no hand washing before vaginal exam. See Additional file 1: Table S1 for a summary of balance for matched and unmatched data. The full unconstrained method matched 500 controls and 48 treated women (52 controls were discarded).

Using the propensity score matched dataset (n = 548), there was a reduction in sepsis rates, although not statistically significant, between newborns of mothers who received intrapartum antibiotics and newborns of mothers who did not receive intrapartum antibiotics (odds ratio [OR] 0.381, 95% confidence interval [CI] 0.115–1.258). The result was similar after adjusting for prematurity, wealth status, and maternal colonization status (OR 0.361, 95% CI 0.106–1.225) (Table 2).

We conducted a sensitivity analysis with a complete case dataset (n = 408) that excluded observations with missing data. Matching with the full unconstrained method yielded 280 controls and 38 treated women (90 controls were discarded). Again there was a reduction, not statistically significant, in sepsis rates between the antibiotic group compared to the control group (OR 0.160, 95% CI 0.021–1.197). The results were similar after adjusting for prematurity, the highest wealth quintile, and maternal colonization status (OR 0.170, 95% CI 0.022–1.295).

Analysis with traditional logistic regression models (n = 600) without propensity score matching showed similar results. There was a reduction in sepsis rates, not statistically significant, between the antibiotic and control groups (OR 0.431, 95% CI 0.130–1.421), with similar results after adjusting for prematurity, the highest wealth quintile, and maternal colonization status (OR 0.458, 95% CI 0.138–1.521).

Since the number of sepsis cases in the treated group were small, we also compared p-values from a Fisher’s exact test of treatment and sepsis (p = 0.182) with the propensity score unmatched logistic regression (p = 0.167) and found little difference.

Antibiotics during labor suggest a decreased risk, although not statistically significant, of early-onset neonatal sepsis in this population. A reduction of early-onset neonatal sepsis by 64%, if confirmed, is clinically important.

Our findings are robust across the different approaches and methods with similar point estimates and confidence intervals. The propensity score matched adjustment estimate is somewhat larger in magnitude compared to the result from traditional regression analysis. Prior to propensity score matching, the observed covariates were imbalanced between the treated and control groups, particularly rupture of membranes at presentation and antenatal care provider type. Propensity score matching reduced confounding by indication by achieving better balance of the observed covariates across the treated and control groups. We further adjusted for confounders by fitting a regression model assuming a normal logistic regression of sepsis given antibiotic use and the observed covariates.

Our sensitivity analysis, a complete case analysis rather than a single imputation of missing values, further decreased the number of sepsis cases in the treatment group (to 1) which may have contributed to a more protective odds in that sensitivity analysis suggesting that our data were missing not at random.

There are few randomized controlled trials that examined intrapartum antibiotics and early-onset neonatal sepsis. A study by Matorras et al. (1990) in Spain found that administration of intrapartum ampicillin to GBS colonized women decreased GBS positive culture cases of early-onset sepsis by 85% and cases of clinical early-onset sepsis by 78% [26]. In Finland, a study by Tuppurainen (1989) found that administering penicillin to GBS colonized women during labor reduced GBS positive culture cases of early-onset sepsis by 75% and cases of clinical early-onset sepsis by 83% [27]. However, in neither study was the reduction statistically significant (a similar scenario as what we found here). One randomized controlled trial by Gibbs (1988) found that ampicillin and gentamicin administered to women with intra-amniotic infections during labor was associated with a statistically significant reduction in early-onset neonatal sepsis (clinical or culture confirmed) (OR 0.04, 95% CI 0.00–0.75) [28]. All three of these studies targeted high risk pregnant women with intra-amniotic infections or colonization with GBS.

In our study, we included all women regardless of their risk status. In low-resource settings like Bangladesh the ability to implement a universal screen using vaginal swabs would be limited. However, a risk factor and clinical symptoms screen may be feasible and useful to test in such a setting.

Antibiotic choice depends on local knowledge of the etiology of maternal colonization and neonatal sepsis and antibiotic resistance patterns. In our facility, most women received the standard first generation cephalosporin, cephalexin, which provides excellent gram positive coverage. An earlier study at the same facility found that the most prevalent organism in the maternal vaginal tract was S. aureus, which was also one of the most common etiologies of neonatal bacteremia in Bangladesh [5]. However, approximately 25% of S. aureus isolates were resistant to cephalexin in a community-based study in Bangladesh [5].

The study has several limitations. We assume there are no unobserved differences between the treatment and control groups given the observed variables; we are only able to adjust for observed confounders. We had more than 5% missing data on gestational age, birth weight, and maternal temperature; we used sensitivity analysis to examine robustness across how we handled this missing data. We had limited neonatal blood culture data due to low compliance with referrals to tertiary care centers. Therefore, our primary outcome measure of neonatal sepsis relied on clinical signs and symptoms, which is overly sensitive and nonspecific. However, newborns of mothers with or without intrapartum antibiotics were non-differentially classified as having sepsis. This non-differential misclassification would have underestimated our effect size. As stated above, our dataset is of relatively small size. Our effective sample size following matching was 89 controls and 48 treated, implying that we had less than 80% power to detect an effect of the size that we found. But the data provides our best information at this point about these associations. Additional, and larger, studies are needed to confirm these results. The provision of antibiotics was based on clinician discretion without a set protocol, which allowed us to mimic randomization of women to antibiotics or control based on a set of similar observed characteristics.

To our knowledge, this is the first study that uses propensity scores to determine the effects of intrapartum antibiotics on early-onset neonatal sepsis in developing country settings. Given the challenges of conducting a randomized controlled trial in settings where intrapartum antibiotic prophylaxis is widely accepted for GBS colonization like the United States and the lack of preliminary data in countries like Bangladesh, this is an attractive method that represents a strong design to investigate the causal effects of antibiotic prophylaxis and early-onset neonatal sepsis [19].

Including intrapartum antibiotics as part of a comprehensive neonatal survival package has the potential to save many lives. Given the potential for tremendous benefit, a double-blind randomized controlled trial testing the effect of intrapartum antibiotics on early-onset neonatal sepsis or larger studies using propensity score matching would be ideal. Current evidence from randomized controlled trials is inconclusive and limited only to GBS [29]. In designing future studies, we need to examine the frequency and timing of antibiotic doses required, the optimal antibiotic choice depending on the geographic variability of organisms, and possible risks. Future studies would provide needed knowledge for areas where intrapartum antibiotics prophylaxis is currently given and support or nullify the use of intrapartum antibiotics in low-resource settings.

Antibiotics during labor indicated a strong protection against early-onset neonatal sepsis (OR = 0.36), but the relative size of our population did not yield a level of statistical significance. In settings where the burden of neonatal mortality is disproportionately high, additional studies with larger datasets using propensity scores or randomized controlled trials testing the effect of intrapartum antibiotics on early-onset neonatal sepsis are warranted.