Meta-analytical approaches to ordinal outcome data in clinical interventional studies: A scoping review with...

Conducting meta-analyses on ordinal outcome data is more complex than on binary or continuous data. This study aims to summarise the current biomedical literature on meta-analytical methods for ordinal outcomes and attempts to reproduce the results of previous studies.

A systematic search was conducted in three databases, MEDLINE, EMBASE, and PsycINFO, from inception to 05/05/2024. Forward and backward citation searches were also performed. The screening was conducted in two phases using Covidence software. Studies were included if they reported or compared methods for meta-analysis of ordinal outcomes in clinical interventional studies. Relevant studies were summarised and discussed. If sufficient data for methods comparison were available, either in the retrieved reports or after contacting the authors, an attempt at reproducible research was made.

333 records were screened, yielding four methodological studies that met the inclusion criteria. These studies addressed meta-analytical methods for ordinal scales ranging from 5 to 20 ordered categories. The three primary approaches identified were (1) ordinal models (proportional odds and generalised odds), (2) binary models (dichotomisation of ordinal scales), and (3) continuous models (treating ordinal scales as continuous variables). None of the included studies provided a comprehensive comparison of all three approaches. Two studies compared different proportional odds models; one compared binary with proportional odds, but the full results were not published, and one compared continuous and generalised odds models using simulated data with one scenario. The latter study allowed for reproducibility, but our analysis produced different results, and attempts to clarify the discrepancy with the authors were unsuccessful.

A significant knowledge gap exists regarding the optimal meta-analytical method for ordinal outcomes in clinical interventional studies. Further methodological research is required to establish a robust evidence base for choosing the most appropriate approach.