Jump to Main Content
Observational studies of non-specific effects of Diphtheria-Tetanus-Pertussis vaccines in low-income countries: Assessing the potential impact of study characteristics, bias and confounding through meta-regression
- Bollaerts, Kaatje, Verstraeten, Thomas, Cohet, Catherine
- Vaccine 2019 v.37 no.1 pp. 34-40
- World Health Organization, developing countries, experimental design, experts, meta-analysis, mortality, observational studies, relative risk, risk estimate, systematic review, vaccination, vaccines
- It has been suggested that some vaccines have effects beyond protection against the diseases they target, called non-specific effects (NSEs). In 2016, a systematic review by Higgins et al., commissioned by the WHO Strategic Advisory Group of Experts (SAGE) on immunization, estimated the relative risk (RR) of all-cause mortality after whole-cell Diphtheria-Tetanus-Pertussis (DTwP) vaccination to be 1.38 (95% CI: 0.92–2.08), and described these potential NSEs as inconsistent. However, the selection of studies for meta-analysis, based on their proneness to bias and confounding, was debated.To identify study characteristics and postulated risks of bias and confounding that might have impacted the RR of all-cause mortality after DTwP vaccination in observational studies conducted in low-income countries.Based on methodological considerations on study design and analysis, we systematically assessed all 17 DTwP studies from the Higgins et al. review for risk of selection bias, exposure and outcome misclassification, confounding and differential co-interventions. We used meta-regression to assess the impact of study characteristics and the postulated risks of bias and confounding on the RR estimates, and looked for outlying and influential risk estimates. Permutation tests were performed to control for false-positive findings.The overall RR of all-cause mortality after DTwP vaccination including all but one outlying and influential study was 1.32 (95% CI: 0.83–2.08). Based on uni-variable meta-regression, we found that study location (p = 0.01), studies using the landmark approach (p = 0.015) and studies at high risk of exposure misclassification (p = 0.036) were significantly associated with increased RR estimates whereas studies at high risk of selection bias (p = 0.059) showed borderline significance. The results further suggest these effect modifiers are clustered in studies conducted in West-Africa.The increased RR of all-cause mortality after DTwP might be confined to West-African countries and/or certain postulated risks of bias might have inflated these RRs.