Using ‘chain of evidence’ synthesis for the use of intra-partum antibiotic prophylaxis to prevent early onset group B streptococcal disease

Session: 

Oral session: Statistical methods (2)

Date: 

Sunday 16 September 2018 - 14:00 to 14:20

Location: 

Ochil 1

All authors in correct order:

Anwer S1, Dias S1, Ades AE1
1 Population Health Sciences, University of Bristol, UK
Presenting author and contact person

Presenting author:

Sumayya Anwer

Contact person:

Sumayya Anwer
Sofia Dias
Abstract text
Background: Typically, meta-analyses assessing intra-partum antibiotic prophylaxis (IAP) for early onset group B streptococcal (EOGBS) disease in infants analyse infant colonisation conditional on maternal colonisation and blood infections conditional on either infant or maternal colonisation separately. As they are conditionally linked, only a single meta-analysis synthesising all three forms of evidence will produce a coherent set of estimates on the effect of IAP on colonisation and infections in infants.

Objectives: To conduct a Bayesian meta-analysis using all available evidence on colonisation and EOGBS in infants and use cross-validation to check the consistency of treatment effects predicted using different evidence sources.

Methods: Relative risks (RR) for the overall effects of IAP on EOGBS given maternal colonisation are modelled using a ‘chain of evidence’ structure. A random-effects model using a vague prior distribution on the treatment effect is used for infant colonisation. A fixed-effect model with a weakly informative prior distribution is used for EOGBS conditional on infant colonisation.

Results: Overall, the use of IAP reduces the risk of EOGBS in infants by 97.7% (86.6% to 99.8%) (RR 0.023, 95% credible interval (CrI) 0.002 to 0.134). Most of the treatment effect is associated with prevention of infant colonisation (RR 0.069, CrI 0.021 to 0.200). There is equivocal evidence for a treatment effect for preventing EOGBS in colonised infants (RR 0.347, CrI 0.026 to 1.617). Cross-validation confirmed consistency of the overall treatment effect predicted from two sources of evidence with the treatment effect obtained from the third (posterior predictive P value 0.942).

Conclusion: By explicitly acknowledging the relationship between outcomes, the synthesis model makes use of the three types of evidence available on the stages leading to EOGBS disease in infants. Results suggest that IAP is effective in preventing EOGBS by preventing infant colonisation. Evidence from different sources was shown to be consistent. These methods can be applied to other clinical situations where outcomes have a conditional relationship.

Patient/healthcare consumer involvement: Trials that motivated the use of chain of evidence synthesis were conducted on pregnant women. Using all available evidence can improve future patient healthcare guidelines.

Relevance to patients and consumers: 

Chain of evidence synthesis delivers a single coherent analysis of multiple evidence sources with conditional relationships and provides insights into the mechanisms of the treatment effect. Having a single set of coherent estimates improves interpretability of the results and provides better information on cost effectiveness and treatment mechanisms that can lead to improvements in patient care.