Article type
Year
Abstract
Background:
The effectiveness of any public health intervention/programme is assessed based on variety of study designs. Performing conventional meta-analysis based on study design has been usual practice in systematic reviews of these interventions. RCTs are given more importance than other study designs. If conclusions are based only on RCTs then only partial evidence summarising effectiveness is provided and this is an injustice toward the large number of high-quality observational studies. It is therefore useful to think about how best to make use of the wide range of evidence available for making sound public health recommendations and not to underestimate or undervalue the evidence from other study designs.
Objectives:
To develop and apply Bayesian methods in public health evidence synthesis.
Methods:
We adopted the Bayesian approach to meta-analysis where the prior distribution had been elicited from observational study designs and the likelihood function from RCTs. We summarised the posterior distribution in terms of posterior pooled estimates and credible intervals using the Markov Chain Monte Carlo method. We performed sensitivity analysis using different prior distributions to check the robustness of the posterior estimates. We applied the models developed in public health systematic reviews of interventions.
Results:
We used a Bayesian fixed- and random-effects model where the prior distribution was elicited from weaker study designs and the likelihood function from RCTs. We found the posterior distribution obtained to be meaningful and appropriate for evidence consolidation rather than concluding only on the basis of one study design.
Conclusions:
The use of observational studies as a prior distribution produced more precise and generalisable evidence. The whole process provides a robust method of evidence synthesis in complex public health research.
Patient or healthcare consumer involvement:
This work is around statistical methods and plays an important role in the synthesis of evidence in the field of public health. Good-quality evidence is created when evidence from all sources is combined. This caters for the needs of policy-makers and clinicians and thereby end-users or patients.
The effectiveness of any public health intervention/programme is assessed based on variety of study designs. Performing conventional meta-analysis based on study design has been usual practice in systematic reviews of these interventions. RCTs are given more importance than other study designs. If conclusions are based only on RCTs then only partial evidence summarising effectiveness is provided and this is an injustice toward the large number of high-quality observational studies. It is therefore useful to think about how best to make use of the wide range of evidence available for making sound public health recommendations and not to underestimate or undervalue the evidence from other study designs.
Objectives:
To develop and apply Bayesian methods in public health evidence synthesis.
Methods:
We adopted the Bayesian approach to meta-analysis where the prior distribution had been elicited from observational study designs and the likelihood function from RCTs. We summarised the posterior distribution in terms of posterior pooled estimates and credible intervals using the Markov Chain Monte Carlo method. We performed sensitivity analysis using different prior distributions to check the robustness of the posterior estimates. We applied the models developed in public health systematic reviews of interventions.
Results:
We used a Bayesian fixed- and random-effects model where the prior distribution was elicited from weaker study designs and the likelihood function from RCTs. We found the posterior distribution obtained to be meaningful and appropriate for evidence consolidation rather than concluding only on the basis of one study design.
Conclusions:
The use of observational studies as a prior distribution produced more precise and generalisable evidence. The whole process provides a robust method of evidence synthesis in complex public health research.
Patient or healthcare consumer involvement:
This work is around statistical methods and plays an important role in the synthesis of evidence in the field of public health. Good-quality evidence is created when evidence from all sources is combined. This caters for the needs of policy-makers and clinicians and thereby end-users or patients.