Confounding

Confounding is the major threat to the internal validity of observational epidemiological studies, so in a systematic review, the confounder-adjusted summary effect will provide the most useful information for questions about aetiology.7 CT and pregnancy are both sexually acquired, so many factors, including younger age, smoking, lower socioeconomic position and belonging to black or indigenous minority ethnic groups11 are associated with both and could plausibly result in systematic overestimation of exposure–outcome association. Tang et al chose to present unadjusted analyses as their primary outcome, because these were more often reported (178 effect estimates) than adjusted estimates (51 estimates). For outcomes with both univariable and multivariable analyses, the unadjusted estimate was higher than the adjusted estimate in 16 of 19 pairs, suggesting that confounding is present. For example, for preterm labour, the unadjusted and adjusted risk ratios in cohort studies were 1.54 (95% CI 1.48 to 1.60) and 1.09 (95% CI 1.03 to 1.15) and ORs in case–control studies were 1.29 (95% CI 1.11 to 1.50) and 1.15 (95% CI 0.95 to 1.39), respectively.6

The presence of other genital tract infections, which are commonly found in women with CT and are also associated with specific outcomes, could also result in confounding. Tang et al did not report on coinfections and whether these had been measured or included in multivariable analyses. They did find that unadjusted effect estimates from case–control studies were higher in low-income and middle-income countries where, plausibly, women are both more likely to have had other genital tract infections associated with adverse pregnancy outcomes and might also have been less likely to have received treatment.

Measurement bias

The investigation of associations between curable infections and pregnancy outcomes brings special challenges because treatment, at an early enough stage, could reduce or eliminate the risk of the outcome. Indeed, that is usually the intention of treatment. Recording a woman as having CT when the infection was treated will result in mismeasurement of the exposure. In this situation, a cross-sectional study of the association between CT and adverse pregnancy outcomes could, paradoxically, provide a less biased effect estimate than a cohort or case–control study in which women were treated (Figure 1). In a cross-sectional study, a positive test result for CT taken at the time of delivery means that an untreated infection was present before or during pregnancy (or there is persistence/reinfection of a treated infection). The measured association is between untreated CT and pregnancy outcome (if adjusted for confounding). In cohort and case–control studies, CT tests are either done prospectively or records of tests done during pregnancy are sought. Women with CT diagnosed during pregnancy are likely to have been treated so the measured association is between treated CT and pregnancy outcomes. Tang et al found that summary estimates in meta-analyses of preterm labour and spontaneous abortion (but not low birth weight and premature rupture of membranes) were higher for cross-sectional than either cohort or case–control studies. There were no cross-sectional study details of other pregnancy outcomes, and Tang et al reported that details of CT testing and treatment were very rarely reported in the included studies.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Observational epidemiological study designs to examine the association between CT and pregnancy outcomes, and the temporal relationship between detection of CT, with Rx if CT+, and pregnancy outcome. The thick solid line indicates the course of pregnancy, with delivery at the end. In a cohort study, CT infection status is determined during pregnancy and, if present, is usually treated. The dashed line, with arrow pointing to the end of pregnancy, shows when the outcome is ascertained. In a case-control study, the dashed line starting at the end of pregnancy shows that the outcome is ascertained at delivery. The arrow points back in time to show that CT infection status is often determined through a lookback in medical records and, if detected, is likely to have been treated. In a cross-sectional study, both outcome and CT infection status are determined at the end of pregnancy and CTI infection is likely to be untreated. Abbreviations: CT, Chlamydia trachomatis; CT+, CT-infected; CT-, CT-uninfected; Rx, treatment.

What do we want to know?

STI practitioners and researchers probably want to know whether CT infection in pregnancy is an important cause of poor reproductive health outcomes. Tang et al’s review shows associations with several adverse outcomes. Because of the wide scope of the review, there was little opportunity to discuss the questions about causality or of potential mechanisms for individual outcomes. For example, the authors suggest that there is increased evidence of associations with stillbirth and spontaneous abortion.6 However, these associations were only present in unadjusted analyses, the biological mechanism is unclear, and the certainty of the body of evidence was assessed as low or very low.

The systematic review by Tang et al does not allow a precise quantification of the causal association between CT infection and the multiple potential adverse outcomes for women’s reproductive health. Their findings will, however, be relevant to the interpretation of systematic reviews of associations between other sexually transmitted and genital tract infections and pregnancy outcomes.8 12 13 The review also highlights methodological issues that researchers can take into account for future studies. Individual studies of associations between genital tract infections and pregnancy outcomes should examine a full range of potential pathogens, including the vaginal microbiota, record the timing of testing and treatment for infection, and both measure and control for confounding factors. Improved reporting of observational studies14 will make it easier for results to be incorporated in systematic reviews.7 Doing a systematic review requires more than following a recipe. In systematic reviews of observational studies, authors should consider in advance what are the important confounding factors for each exposure–outcome pair, show results of both unadjusted and confounder-adjusted estimates, and assess the risk of bias considering that sources of bias may be specific to different research questions and study designs.

Improved reproductive health outcomes are a major international goal for women and their families, for clinical medicine and for public health. Systematic reviews can help to investigate the potential population impact of interventions. The studies identified and included will point to gaps in the evidence, and summary effect sizes can be used to estimate population-attributable fractions (bearing in mind the strong assumptions about causality). Ultimately, randomised controlled trials of different interventions, with clinical outcomes, in settings with low and high levels of CT and coinfection prevalence will be needed to determine whether screening and treatment to prevent adverse pregnancy outcomes do more good than harm at a reasonable cost.