Pooling of effect estimates obtained from various study designs in systematic reviews of public health interventions

A Bayesian approach to meta-analysis

Melissa Glenda Lewis, Vasudeva Guddattu, Asha Kamath, Seena Biju, Judith Noronha, Baby Nayak, N. Sreekumaran Nair

Research output: Contribution to journalArticle

Abstract

Background Randomised controlled trials (RCTs) are gold standard in assessing the effectiveness of a clinical intervention because of their high internal validity. However, the same does not hold true for interventions conducted at the population level like public health interventions. Well-designed RCTs are not easy to conduct at population level. Similarly, well planned, high-quality non-RCTs or observational studies can complement RCTs. Because of this, several systematic reviews of public health interventions are assessed with other study designs, namely non-RCTs and observational studies. In such situations, studies of similar study design are pooled together to obtain an overall effect estimate. This is inevitable, because the principle of meta-analysis does not offer an opportunity for combining effect estimates coming from various study designs. If the meta-analysis performed for each study design provides contrasting results, then this introduces a quandary for the decision makers and public health policy makers to call for a decision. Objective The present study aims to integrate the results coming from a variety of study designs in order to obtain a single estimate of effect of intervention. Methodology Bayesian approach to meta-analysis was used by formulating prior distribution from observational studies or non-RCTs and likelihood function from RCTs. Five systematic reviews of public health intervention were used to demonstrate the methodology. Results/conclusions By formulating prior distribution from observational studies, the posterior estimates were found to be different than that from the results of RCTs or other study designs. The posterior pooled-estimate was found to be precise and the width of the credible interval narrowed. Inclusion of the relevant observational studies (or non-RCTs) in the systematic review is a potential advantage for evaluating the effectiveness of public health intervention.

Original languageEnglish
Pages (from-to)137-142
Number of pages6
JournalClinical Epidemiology and Global Health
Volume5
Issue number3
DOIs
Publication statusPublished - 01-09-2017

Fingerprint

Bayes Theorem
Observational Studies
Meta-Analysis
Randomized Controlled Trials
Public Health
Likelihood Functions
Public Policy
Health Policy
Administrative Personnel
Population
Non-Randomized Controlled Trials

All Science Journal Classification (ASJC) codes

  • Epidemiology
  • Public Health, Environmental and Occupational Health
  • Microbiology (medical)
  • Infectious Diseases

Cite this

@article{6dc310f82af1488b934378ee017bd26a,
title = "Pooling of effect estimates obtained from various study designs in systematic reviews of public health interventions: A Bayesian approach to meta-analysis",
abstract = "Background Randomised controlled trials (RCTs) are gold standard in assessing the effectiveness of a clinical intervention because of their high internal validity. However, the same does not hold true for interventions conducted at the population level like public health interventions. Well-designed RCTs are not easy to conduct at population level. Similarly, well planned, high-quality non-RCTs or observational studies can complement RCTs. Because of this, several systematic reviews of public health interventions are assessed with other study designs, namely non-RCTs and observational studies. In such situations, studies of similar study design are pooled together to obtain an overall effect estimate. This is inevitable, because the principle of meta-analysis does not offer an opportunity for combining effect estimates coming from various study designs. If the meta-analysis performed for each study design provides contrasting results, then this introduces a quandary for the decision makers and public health policy makers to call for a decision. Objective The present study aims to integrate the results coming from a variety of study designs in order to obtain a single estimate of effect of intervention. Methodology Bayesian approach to meta-analysis was used by formulating prior distribution from observational studies or non-RCTs and likelihood function from RCTs. Five systematic reviews of public health intervention were used to demonstrate the methodology. Results/conclusions By formulating prior distribution from observational studies, the posterior estimates were found to be different than that from the results of RCTs or other study designs. The posterior pooled-estimate was found to be precise and the width of the credible interval narrowed. Inclusion of the relevant observational studies (or non-RCTs) in the systematic review is a potential advantage for evaluating the effectiveness of public health intervention.",
author = "Lewis, {Melissa Glenda} and Vasudeva Guddattu and Asha Kamath and Seena Biju and Judith Noronha and Baby Nayak and Nair, {N. Sreekumaran}",
year = "2017",
month = "9",
day = "1",
doi = "10.1016/j.cegh.2016.12.001",
language = "English",
volume = "5",
pages = "137--142",
journal = "Clinical Epidemiology and Global Health",
issn = "2213-3984",
publisher = "Elsevier BV",
number = "3",

}

TY - JOUR

T1 - Pooling of effect estimates obtained from various study designs in systematic reviews of public health interventions

T2 - A Bayesian approach to meta-analysis

AU - Lewis, Melissa Glenda

AU - Guddattu, Vasudeva

AU - Kamath, Asha

AU - Biju, Seena

AU - Noronha, Judith

AU - Nayak, Baby

AU - Nair, N. Sreekumaran

PY - 2017/9/1

Y1 - 2017/9/1

N2 - Background Randomised controlled trials (RCTs) are gold standard in assessing the effectiveness of a clinical intervention because of their high internal validity. However, the same does not hold true for interventions conducted at the population level like public health interventions. Well-designed RCTs are not easy to conduct at population level. Similarly, well planned, high-quality non-RCTs or observational studies can complement RCTs. Because of this, several systematic reviews of public health interventions are assessed with other study designs, namely non-RCTs and observational studies. In such situations, studies of similar study design are pooled together to obtain an overall effect estimate. This is inevitable, because the principle of meta-analysis does not offer an opportunity for combining effect estimates coming from various study designs. If the meta-analysis performed for each study design provides contrasting results, then this introduces a quandary for the decision makers and public health policy makers to call for a decision. Objective The present study aims to integrate the results coming from a variety of study designs in order to obtain a single estimate of effect of intervention. Methodology Bayesian approach to meta-analysis was used by formulating prior distribution from observational studies or non-RCTs and likelihood function from RCTs. Five systematic reviews of public health intervention were used to demonstrate the methodology. Results/conclusions By formulating prior distribution from observational studies, the posterior estimates were found to be different than that from the results of RCTs or other study designs. The posterior pooled-estimate was found to be precise and the width of the credible interval narrowed. Inclusion of the relevant observational studies (or non-RCTs) in the systematic review is a potential advantage for evaluating the effectiveness of public health intervention.

AB - Background Randomised controlled trials (RCTs) are gold standard in assessing the effectiveness of a clinical intervention because of their high internal validity. However, the same does not hold true for interventions conducted at the population level like public health interventions. Well-designed RCTs are not easy to conduct at population level. Similarly, well planned, high-quality non-RCTs or observational studies can complement RCTs. Because of this, several systematic reviews of public health interventions are assessed with other study designs, namely non-RCTs and observational studies. In such situations, studies of similar study design are pooled together to obtain an overall effect estimate. This is inevitable, because the principle of meta-analysis does not offer an opportunity for combining effect estimates coming from various study designs. If the meta-analysis performed for each study design provides contrasting results, then this introduces a quandary for the decision makers and public health policy makers to call for a decision. Objective The present study aims to integrate the results coming from a variety of study designs in order to obtain a single estimate of effect of intervention. Methodology Bayesian approach to meta-analysis was used by formulating prior distribution from observational studies or non-RCTs and likelihood function from RCTs. Five systematic reviews of public health intervention were used to demonstrate the methodology. Results/conclusions By formulating prior distribution from observational studies, the posterior estimates were found to be different than that from the results of RCTs or other study designs. The posterior pooled-estimate was found to be precise and the width of the credible interval narrowed. Inclusion of the relevant observational studies (or non-RCTs) in the systematic review is a potential advantage for evaluating the effectiveness of public health intervention.

UR - http://www.scopus.com/inward/record.url?scp=85008462052&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85008462052&partnerID=8YFLogxK

U2 - 10.1016/j.cegh.2016.12.001

DO - 10.1016/j.cegh.2016.12.001

M3 - Article

VL - 5

SP - 137

EP - 142

JO - Clinical Epidemiology and Global Health

JF - Clinical Epidemiology and Global Health

SN - 2213-3984

IS - 3

ER -