Identifying subgroups based on continuous measurements in individual patient data meta-analysis

Article type
Authors
Belias M1, Rovers M2, Reitsma JB3, Debray TPA3, IntHout J1
1Radboud Institute for Health Sciences, Department for Health Evidence, Radboud University Medical Center Nijmegen
2Department for Health Evidence and Department of Operating Rooms, Radboud University Medical Center Nijmegen; Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht
3Julius Center for Health Sciences and Primary Care and Dutch Cochrane Center, University Medical Center Utrecht
Abstract
Background: Individual patient data meta-analysis (IPD-MA) is increasingly used to analyse heterogeneity of treatment effects. Linearity assumptions are often made when examining subgroups based on continuous measurements. However, several more flexible methods exist.

Objectives: Our goal is to illustrate, critically review and compare state of the art methods on subgroup effects identification in IPD-MA, based on continuous measurements.

Methods: We reviewed META-STEPP, generalized additive mixed effects models, (multi-level) regression models involving fractional polynomials or splines and several tree-based approaches.

We applied the methods above to two empirical examples: prescription of antibiotics in children with otitis media and anti-platelet treatment in secondary stroke prevention.

Results: We will provide treatment effect plots to visualize subgroup effects within and across studies.

Conclusions: We will explain the advantages and limitations of the aforementioned methods.

Patient or healthcare consumer involvement: Our research will provide insight into recently developed statistical methods, to make better use of existing data for the benefit of individual patients. We aim to improve patient-centered health care.