Prespecification of subgroup analyses and examination of treatment-subgroup interactions in cancer individual participant data meta-analyses are suboptimal

Date & Time
Monday, September 4, 2023, 12:30 PM - 2:00 PM
Location Name
Pickwick
Session Type
Poster
Category
Individual patient data meta-analysis
Speakers
Ya Gao
Authors
Gao Y1, Liu M1, Zhang J2, Song F3, Tian J1
1Evidence-Based Medicine Center, School of Basic Medical Sciences, Lanzhou University, Lanzhou, China, China
2Evidence-Based Medicine Center, Tianjin University of Traditional Chinese Medicine, Tianjin, China, China
3Public Health and Health Services Research, Norwich Medical School, University of East Anglia, Norwich, UK, UK
Description

Background: The results of subgroup analyses may have a significant impact on clinical and public health decision-making. How often cancer individual participant data meta-analyses (IPDMAs) prespecify subgroup analyses, conduct planned subgroup analyses, and use daft (across-trial interaction alone), deluded (within-trial and across-trial interactions combined), or deft (within-trial interaction alone) approach to assess the treatment-subgroup interactions remain unclear.
Objectives: This study aimed to explore the pre-specification and conduct of subgroup analyses in cancer IPDMAs.
Methods: We searched PubMed, Embase.com, Cochrane Library, and Web of Science to identify IPDMAs of randomized controlled trials evaluating intervention effects for cancer. We evaluated how often cancer IPDMAs prespecify subgroup analyses and statistical approaches for examining treatment-subgroup interactions and handling continuous subgroup variables.
Results: We included 89 IPDMAs, of which 41 (46.1%) reported a statistically significant treatment-subgroup interaction (p-value < 0.05) in at least one subgroup analysis. Forty-seven (52.8%) IPDMAs prespecified methods for conducting subgroup analyses and the remaining 42 (47.2%) did not prespecify subgroup analyses. Of the 47 IPDMAs prespecified subgroup analyses, 19 performed the planned subgroup analyses, 21 added subgroup analyses, 7 reduced subgroup analyses. Eighty IPDMAs examined treatment-subgroup interactions, but 72 IPDMAs did not provide enough information to determine whether an appropriate approach that avoided aggregation bias was used. Eighty-five IPDMAs that used continuous variables in subgroup analyses categorized continuous variables and only one IPDMA examined non-linear relationships.
Conclusions: Many cancer IPDMAs did not prespecify subgroup analyses, nor did they fully perform planned subgroup analyses. Lack of details for the test of treatment-subgroup interactions and examination of non-linear interactions was suboptimal. Patient, public, and/or healthcare consumer involvement: NA.