Statistical approaches to identify subgroups in meta-analysis of individual participant data: a simulation study

Session:

Oral session: Statistical methods (1)

Date:

Monday 17 September 2018 - 11:00 to 11:20

Location:

Fintry Auditorium

All authors in correct order:

Belias M¹, Rovers M¹, Reitsma JB², Debray TPA², IntHout J¹

¹ Radboud UMC, Netherlands
² Utrecht UMC, Netherlands

Presenting author and contact person

Presenting author:

Michail Belias

Contact person:

Michail Belias

Thomas Debray

Abstract text

Background:
Individual participant data meta-analyses (IPD-MA) provide the best evidence regarding treatment effects and offer opportunities and benefits when investigating subgroup effects. IPD-MA may be conducted either in one or two stages. Both procedures offer a variety of methods.

Objectives:
To compare the performance of five common regression-based approaches to detect subgroups in meta-analysis (MA) of individual participant data (IPD): meta-regression; per-subgroup MA (PS-MA); MA of interaction terms (MA-IT); naive one-stage IPD-MA (ignoring potential study-level confounding); and centered one-stage IPD-MA (accounting for potential study-level confounding).

Methods:
We conducted a simulation study where we varied the magnitude of subgroup effect (0%, 25%, 50% relative reduction), between-study treatment effect heterogeneity (none, medium, large), trial size (50,100, 200), and number of trials (5,10) for binary, continuous and time-to-event outcomes. For each scenario, we assessed the power and false positive rate (FPR) of the aforementioned five approaches.

Results:
Naive and centered one-stage IPD-MA yielded the highest power, whilst preserving acceptable FPR around the nominal 5%. Similar results were obtained for MA-IT, except when analysing binary outcomes (where it yielded less power and FPR < 5%). PS-MA showed similar power as MA-IT in non-heterogeneous scenarios, but power collapsed as heterogeneity increased, and PS-MA suffered from too high FPR, in non-heterogeneous settings. Meta-regression showed poor power (<15%) in all scenarios.

Conclusions:
Our results indicate that subgroup detection in IPD-MA requires careful modelling. Naive and centered one-stage IPD-MA performed equally well, but due to the possibility of ecological bias, we recommend the latter.

Patient or healthcare consumer involvement:
Our research provides insight into commonly applied statistical methods, to make better use of existing data to the benefits of individual patients. We aim to improve patient-centered healthcare.