Donegan, Sarah
(2011)
The value of individual patient data for mixed treatment comparison meta-analysis.
PhD thesis, University of Liverpool.
|
Text
570446.pdf - Unspecified Download (46MB) | Preview |
Abstract
Numerous treatments can be compared simultaneously using a single mixed treatment comparison (MTC) meta-analysis model that combines all direct and indirect evidence. Three key assumptions underlie MTC methodology: similarity, consistency, and homogeneity. Meta-analysis can be based on individual patient data (IPD) and/or aggregate data. Acquiring IPD will improve the quality of conventional pair-wise meta-analysis in various ways. The value of IPD for MTC meta-analysis is currently unknown. This thesis explores the benefits of using IPD covariate information to assess the underlying assumptions of MTCs. The methodology is illustrated using real IPD from a single multicentre trial that compared artemisinin-based combination therapies (ACTs) for treating uncomplicated malaria in African children. Existing aggregate data MTC meta-analysis models for dichotomous outcomes are extended to allow for patient-level outcomes and covariates. The potential benefits of IPD are evaluated by comparing results from IPD models including treatment by patient-level covariate interactions, with those from aggregate data models including treatment by study-level covariate interactions. The results showed that treatment effects and drug rankings based on IPD, differed from those estimated using aggregate data. The inclusion of patient-level, rather than site-level covariates, produced more precise treatment effects and regression coefficients for the interactions. Therefore, including patient-level covariates was more favourable than including site-level data. A new approach is proposed to determine whether any existing inconsistency is reduced, or explained, following the inclusion of treatment by covariate interactions in the MTC model. The same approach is followed for models involving study-level covariates and models with patient- level covariates. Using aggregate data, results showed that there were too few sites contributing direct evidence to allow consistency to be established when including treatment by covariate interactions. Based on IPD, the regression coefficients for the interactions were estimated from the within-site and across-site interactions and therefore consistency could be determined. Patient- level covariates, rather than site-level data, were clearly beneficial when judging whether inconsistency was reduced by including treatment by covariate interactions in the model. Novel MTC meta-analysis models for a dichotomous outcome are introduced that each combine IPD and aggregate data using a one-stage approach while including treatment by covariate interactions. The methodology is illustrated using the real IPD and a supplementary dataset consisting of aggregate data from a single Cochrane review that also compared ACTs. When MTC models were fitted to the aggregate dataset alone, the results were imprecise and the Markov chain Monte Carlo chains did not convergence. When MTC models were applied to the IPD and when one-stage models were fitted to all data, convergence of the chains was adequate and the credibility intervals for the treatment effects and regression coefficients were much narrower. When exploring treatment by covariate interactions, it was beneficial to obtain IPD, if only for a subset of trials, and to combine the patient-level data with the additional aggregated data in a me ta-analysis model. This thesis has shown that IPD can be extremely valuable in MTC meta-analysis.
| Item Type: | Thesis (PhD) |
|---|---|
| Depositing User: | Symplectic Admin |
| Date Deposited: | 19 Oct 2023 17:54 |
| Last Modified: | 19 Oct 2023 18:02 |
| DOI: | 10.17638/03174254 |
| Copyright Statement: | Copyright © and Moral Rights for this thesis and any accompanying data (where applicable) are retained by the author and/or other copyright owners. A copy can be downloaded for personal non-commercial research or study, without prior permission or charge |
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3174254 |
Dimensions
Dimensions
