Evidence Synthesis Methods to Evaluate the Effectiveness of Complex Public Health Interventions With Multiple Intervention Components
In the evidence synthesis of public health interventions, several challenges arise due to the heterogeneity of study design, populations and complexity of the interventions, for which standard meta-analysis methods are not always appropriate. Despite this, most public health intervention appraisals are currently based on narrative reviews or pairwise meta-analysis. This thesis explores the development and application of evidence synthesis methodology that addresses the complexity of public health evidence. Component network meta-analysis can estimate the effectiveness of intervention components and combinations of components. In this thesis, component network meta-analysis was extended to incorporate the estimation of covariate-component effects at the individual and study level.
In this thesis, a review of NICE public health guidelines explored the use of, and barriers to, evidence synthesis methodology. Application of evidence synthesis methodology, and the extension of standard methods, was demonstrated using two motivating examples in public health: a previous Cochrane review in the prevention of childhood accidents, and a systematic review conducted in this thesis investigating interventions for the prevention of type 2 diabetes mellitus. A component taxonomy for the second motivating example was developed based on delivery modality. A series of network and component network meta-analysis models were applied to the motivating examples. The extension of standard methods proposed in this thesis was applied where individual participant data was available.
Heterogeneity is a major barrier to the implementation of more advanced statistical methods in public health intervention appraisals. Evidence synthesis methodology and the proposed extended method in this thesis enables the exploration of heterogeneity in terms of intervention components, and covariate associations at the individual and study level, whilst addressing ecological bias. These methods could have implications for improving the intervention evaluation process across the wide-ranging topics in public health by providing more information to decision makers and addressing policy relevant questions.
Supervisor(s)Stephanie Hubbard; Laura Gray
Date of award2023-02-27
Author affiliationDepartment of Health Sciences
Awarding institutionUniversity of Leicester