Related Articles |
A recursive partitioning strategy for subgroup identification in particular person affected person information meta-analysis.
Stat Med. 2018 Jan 31;:
Authors: Mistry D, Stallard N, Underwood M
Summary
BACKGROUND: Motivated by the setting of scientific trials in low again ache, this work investigated statistical strategies to establish affected person subgroups for which there’s a big therapy impact (therapy by subgroup interplay). Statistical exams for interplay are sometimes underpowered. Particular person affected person information (IPD) meta-analyses present a framework with improved statistical energy to analyze subgroups. Nevertheless, standard approaches to subgroup analyses utilized in each a single trial setting and an IPD setting have quite a lot of points, one in all them being that elements used to outline subgroups are investigated one after the other. As people have a number of traits that could be associated to response to therapy, various exploratory statistical strategies are required.
METHODS: Tree-based strategies are a promising various that systematically searches the covariate area to establish subgroups outlined by a number of traits. A tree methodology specifically, SIDES, is described and prolonged for software in an IPD meta-analyses setting by incorporating fixed-effects and random-effects fashions to account for between-trial variation. The efficiency of the proposed extension was assessed utilizing simulation research. The proposed methodology was then utilized to an IPD low again ache dataset.
RESULTS: The simulation research discovered that the prolonged IPD-SIDES methodology carried out properly in detecting subgroups particularly within the presence of enormous between-trial variation. The IPD-SIDES methodology recognized subgroups with enhanced therapy impact when utilized to the low again ache information.
CONCLUSIONS: This work proposes an exploratory statistical strategy for subgroup analyses relevant in any analysis self-discipline the place subgroup analyses in an IPD meta-analysis setting are of curiosity.
PMID: 29383818 [PubMed – as supplied by publisher]