Article type
Year
Abstract
Background:
Individual participant data meta-analysis (IPDMA) projects herald huge promise [1], but are time-consuming, often taking upwards of two years to obtain, clean, harmonise, and meta-analyse IPD. Therefore, before embarking on an IPDMA project, researchers and funders should ascertain if it is worth their investment and, if so, how best to proceed.
Objectives:
In this talk, aimed at a broad audience, I discuss the role of power calculations in IPDMA projects and how they reveal (1) whether the project is capable of identifying meaningful clinical effects and, if so, (2) which studies to prioritise for IPD collection.
Methods:
Firstly, IPDMA investigators should identify the set of studies likely to provide their IPD and extract key characteristics, such as number of participants and outcome events per group, and covariate distributions. Using this information and assumptions about true study effect sizes, Fisher’s information can be applied to derive the approximate standard error of each study’s effect estimate. Then, these are used to calculate the anticipated standard error of the IPDMA summary result, and subsequently the IPDMA power.
Results:
Real applications will be showcased (rather than statistical formula), focusing on IPDMA projects of randomised trials examining treatment-covariate interactions (treatment effect modifiers) at the individual level. Continuous, binary, and survival outcomes are covered. In one example, an IPDMA of 31 trials has 42% power for a treatment-sex interaction but 90% power for a treatment-age interaction. The calculations also reveal the contribution of each study towards power, to help inform which studies should be prioritised for IPD collection. I show that a study’s power contribution depends on covariate distributions not just total participants or events.
Conclusions:
Careful assessments of power and sample size should be an integral part of planning and commissioning IPDMA projects, in advance of IPD collection. Alongside other information (e.g., risk of bias assessments), power calculations help direct investment toward high-quality IPDMA projects most likely to provide trusted evidence that informs clinical practice to improve patient outcomes.
Patient, public, and/or healthcare consumer involvement: None.
[1] Riley RD, Tierney J, Stewart L(Eds). Individual Participant Data Meta-Analysis: A Handbook for Healthcare Research. Wiley, Chichester; 2021
Individual participant data meta-analysis (IPDMA) projects herald huge promise [1], but are time-consuming, often taking upwards of two years to obtain, clean, harmonise, and meta-analyse IPD. Therefore, before embarking on an IPDMA project, researchers and funders should ascertain if it is worth their investment and, if so, how best to proceed.
Objectives:
In this talk, aimed at a broad audience, I discuss the role of power calculations in IPDMA projects and how they reveal (1) whether the project is capable of identifying meaningful clinical effects and, if so, (2) which studies to prioritise for IPD collection.
Methods:
Firstly, IPDMA investigators should identify the set of studies likely to provide their IPD and extract key characteristics, such as number of participants and outcome events per group, and covariate distributions. Using this information and assumptions about true study effect sizes, Fisher’s information can be applied to derive the approximate standard error of each study’s effect estimate. Then, these are used to calculate the anticipated standard error of the IPDMA summary result, and subsequently the IPDMA power.
Results:
Real applications will be showcased (rather than statistical formula), focusing on IPDMA projects of randomised trials examining treatment-covariate interactions (treatment effect modifiers) at the individual level. Continuous, binary, and survival outcomes are covered. In one example, an IPDMA of 31 trials has 42% power for a treatment-sex interaction but 90% power for a treatment-age interaction. The calculations also reveal the contribution of each study towards power, to help inform which studies should be prioritised for IPD collection. I show that a study’s power contribution depends on covariate distributions not just total participants or events.
Conclusions:
Careful assessments of power and sample size should be an integral part of planning and commissioning IPDMA projects, in advance of IPD collection. Alongside other information (e.g., risk of bias assessments), power calculations help direct investment toward high-quality IPDMA projects most likely to provide trusted evidence that informs clinical practice to improve patient outcomes.
Patient, public, and/or healthcare consumer involvement: None.
[1] Riley RD, Tierney J, Stewart L(Eds). Individual Participant Data Meta-Analysis: A Handbook for Healthcare Research. Wiley, Chichester; 2021