This article is part of a wider study protocol entitled “Systematic estimates of the burden of GBS worldwide in pregnant women, stillbirths and infants.” It was submitted for ethical approval to the London School of Hygiene & Tropical Medicine (reference citation 11966) and approved on 30 November 2016.

There is some evidence, from this comprehensive review, that GBS is associated with preterm birth. There is a consistent increase in risk of preterm birth in women with maternal GBS colonization, which is stronger in case-control studies compared to cohort or cross-sectional studies. In addition, where there is evidence of ascending infection with maternal GBS bacteriuria, the association with preterm birth is stronger, which is biologically plausible. Our findings are potentially important, and we have made extensive attempts to consider study design and address specific biases, learning from challenges in previous reviews. However, considerable limitations remain, and these results could still be affected by bias or confounding, as discussed below.

In terms of the data included, we are limited in terms of geographical distribution, with most data from high-income contexts, and the potential sources of bias we were able to assess based on reported sampling strategies, microbiological methods, and gestational age measurement.

Our sensitivity analyses are specifically aimed to address misclassification, but the reduction in power through exclusion of studies likely limited the ability to detect true association. Interestingly, however, in terms of sensitivity of exposure, it may actually be easier to detect an association when less sensitive sampling methods are used. Rectovaginal sampling and broth enrichment increase sensitivity of detection [71] in 40% and 90%, respectively [16], but excluding less sensitive methods did not identify an association. Conversely, using nonselective medium and sampling from the cervix or upper vagina (less sensitive methods that would detect more heavily colonized women), showed a strong association with preterm delivery (RR, 1.65 [95% CI, 1.03–2.64]) for cohort and cross-sectional studies. This is consistent with studies reporting associations in women considered heavily colonized [60] and the association we identified between maternal bacteriuria (which reflects denser colonization) and ascending infection, which was more strongly associated with preterm birth.

Our sensitivity analysis suggested that antibiotic use near or at delivery may also affect findings, and could bias the results toward the null if women with complications are given antibiotics and GBS is thus not detected. In future studies it will be important to take into account the receipt and timing of antibiotics in pregnancy. In terms of the timing of the sample, earlier sampling may increase misclassification due to reacquisition of bacteria, so repeated sampling through pregnancy could be important to test the association.

In terms of outcome measurement, uncertainty in gestational age dating will increase misclassification. This may be nondifferential in cohort and cross-sectional studies, which would bias findings toward the null, but may be differential in case-control studies, with recruitment bias toward more preterm babies included in the study and thus overestimation of the effect. This may account for the differences in findings and some uncertainty in both. Another study found no differences in the effect estimates for GBS colonization and premature delivery when measured by ultrasound or by date of LMP [37], but due to the small number of studies here we could not compare results according to the method used for gestational age dating. We note that the inconsistencies in GA assessment would be even more marked in data from middle-income and especially low-income contexts, where GA measurement is challenging; usually using fundal height and sometimes LMP. Given that three-quarters of preterm births are in sub-Saharan Africa and South Asia, more data, with consistent GA, are crucial for future studies [72].

The results presented here may, however, be subject to confounding (due to factors associated with both maternal GBS colonization and preterm birth). This could change the effect in either direction. Adjustment for confounding factors has increased the effect size in several studies [27, 29, 35, 36, 62], but this may be context specific [16]. It is important that confounding factors are considered, including known risk factors for preterm birth such as low socioeconomic status, black race, low body mass index, previous preterm birth, multiple gestation, short interpregnancy interval, and the use of tobacco or illicit drugs [9], and risk factors for GBS colonization such as age at pregnancy, interpregnancy interval, previous abortions, and level of education [73, 74]. These should be incorporated into multivariable modeling strategies.

To better answer this important research question, more data are needed with optimized and standardized methodologies reported systematically, particularly from low- and middle-income contexts [75]. The optimal study design would be a large, longitudinal prospective study including multiple sites, with accurate exposure and outcome measurement, and repeat sampling at intervals in pregnancy. This should include measurement of gestational age, preferably based on first-trimester ultrasound, samples taken from rectovaginal swabs and isolated in selective enrichment broth (ideally with quantification of GBS colonization), adjustment for use of antibiotics (including timing of receipt of antibiotics during pregnancy), and measurement and adjustment of known risk factors for both preterm delivery and GBS colonization. However, this is challenging to achieve in health systems in low- and middle-income contexts where resources are limited. An alternative approach would be an intervention vaccine-probe study, which would overcome the problems of bias and confounding and could be done as part of a maternal GBS vaccine study.

We found some evidence of an association between maternal GBS colonization and preterm birth. Misclassification is likely to reduce the effect size of risk and/or odds ratios, so this may be underestimated. However, results may also be subject to confounding, which could influence the findings in either direction. Current prevention strategies (based on intrapartum antibiotic prophylaxis) are too late to prevent preterm birth associated with GBS colonization. A future maternal GBS vaccine targeted against maternal GBS colonization or the mechanism downstream of colonization leading to preterm birth, and administered during the appropriate timing of pregnancy, could be useful and should be included in any maternal GBS vaccine trial (Table 2).

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.