Article type
Year
Abstract
Background: meta-analyses increase both the power and the precision of estimated treatment effects. However, meta-analyses introduce problems inherent to multiple testing as new trials are added over time. To overcome this problem, trial sequential analysis (TSA) accounts for bias and observed heterogeneity in a cumulative meta-analysis. TSA is a natural bridge between cumulative meta-analysis and sequential analysis of clinical trials, and offers methods for dealing with repeated testing bias, heterogeneity among studies, and measuring significance and futility precisely. In order to increase its uptake and usability, we implemented the TSA methods in the open-source R environment, and developed a user-friendly web-based application for binomial raw and summary effect measures.
Objectives: the goal of this study is to describe the development and testing of a user-friendly tool to conduct TSA for both raw and summary effect measures.
Methods: our TSA tool is programmed in the open-source R language building on the 'tidyverse' package (required for the R function) and 'shiny' (required to run the web-based application). This tool allows for calculation of the required information size (RIS)/sample size, monitoring boundaries for type I errors using the alpha spending function, and conditional power boundaries for futility using a beta spending function, with a predefined conditional bound gamma. To conduct TSA we need a predefined baseline risk, relative risk reduction, type I and type II error probabilities, and either a predefined - or estimated - between-study heterogeneity from the available data . A TSA plot is generated for cumulative Z-scores on the y-axis and cumulative information size from trials on the x-axis, which compares these Z-scores with the monitoring and futility boundaries (Figure 1).
Results: using data from a published systematic review, Figure 1-2 shows the results of the validation analyses. Our new tool can produce TSA analyses/plots for both raw and estimated binomial summary measures that are above and beyond what was available previously, and overcomes the limitations in the TSA-software that is currently available. We will also present how the overall power changes due to varying predefined conditional power bounds, which is not described in previous studies (Figure 1, 3). Currently, we are developing this software further to incorporate continuous, and other commonly used data types.
Conclusions: uur web-based application will aid users unfamiliar with R to conduct these analyses without prior training. This provides unhindered access to the software within a point-and-click environment. To the best of our knowledge, this is the first web-based TSA tool that overcomes the shortcomings of the software that was previously available.
Patient or healthcare consumer involvement: patients or healthcare consumers were not involved in this project.
Objectives: the goal of this study is to describe the development and testing of a user-friendly tool to conduct TSA for both raw and summary effect measures.
Methods: our TSA tool is programmed in the open-source R language building on the 'tidyverse' package (required for the R function) and 'shiny' (required to run the web-based application). This tool allows for calculation of the required information size (RIS)/sample size, monitoring boundaries for type I errors using the alpha spending function, and conditional power boundaries for futility using a beta spending function, with a predefined conditional bound gamma. To conduct TSA we need a predefined baseline risk, relative risk reduction, type I and type II error probabilities, and either a predefined - or estimated - between-study heterogeneity from the available data . A TSA plot is generated for cumulative Z-scores on the y-axis and cumulative information size from trials on the x-axis, which compares these Z-scores with the monitoring and futility boundaries (Figure 1).
Results: using data from a published systematic review, Figure 1-2 shows the results of the validation analyses. Our new tool can produce TSA analyses/plots for both raw and estimated binomial summary measures that are above and beyond what was available previously, and overcomes the limitations in the TSA-software that is currently available. We will also present how the overall power changes due to varying predefined conditional power bounds, which is not described in previous studies (Figure 1, 3). Currently, we are developing this software further to incorporate continuous, and other commonly used data types.
Conclusions: uur web-based application will aid users unfamiliar with R to conduct these analyses without prior training. This provides unhindered access to the software within a point-and-click environment. To the best of our knowledge, this is the first web-based TSA tool that overcomes the shortcomings of the software that was previously available.
Patient or healthcare consumer involvement: patients or healthcare consumers were not involved in this project.