Hello, anyone out there familiar with Trial Sequential Analysis (TSA) software developed by Copenhagen Trial Unit?
The free software is available at: http://www.ctu.dk/tsa/downloads.aspx
Manual is available here: http://www.ctu.dk/tsa/files/tsa_manual.pdf
Summary of my questions up here:
1. How do you obtain required information size (RIS) by using traditional 5% significance boundary, 'cause for what I'm aware of, RIS can only be calculated by α-spending boundary using O'Brien-Fleming method?
2. Where and how can I set "δ" value while applying α-spending boundary ?
-----------------------------------------------------------------------------------------------------------------------------------------
So I've been working on a project involving meta-analysis and trial sequential anaylsis of noninferiority randomized control trials (RCTs).
I've met some technical problems which I can't solve, so I tried to look for literature of similar study design and luckily I found this study entitled Haloperidol Versus 5-HT 3 Receptor Antagonists for Postoperative Vomiting and QTc Prolongation: A Noninferiority Meta-Analysis and Trial Sequential Analysis of Randomized Controlled Trials.
Regarding the setting of conventional 5% significance boundary and alpha-spending boundary using O'Brien Fleming method, here I quote " The first was the classical boundary, setting the α error to 5%. The required estimated information size was 799, and our cumulative value was found to be 859. Further, we used the O'Brien-Fleming α-spending function for testing for futility. For this the δ was set at 10% and β at 20%. The TSA graph clearly showed that cumulative Z scores were well past the inner wedge of futility and fell within the range of actual equivalence. The information size of 859 was well past the required 812 for this method as well"
Obviously, the equivalence of haloperidol and 5-HT 3 receptor antagonists is proved.
I tried to simulate and run the software myself using exactly the same data provided in this article, as shown in Forest plot.
After adding every single trials with the events/total, I started to set up the parameter needed for applying both conventional 5% significance boundary and α-spending boundary. The parameters needed for setting up the traditional 5% significance boundary are type I error (5%) and boundary type (two-sided) as shown in figure 1. As you can see, there's no where showing the estimated RIS, which gave rise to the first question I mentioned above. Next, I filled in parameters needed to set up α-spending boundary, including α=5%, power=80%, information axis=sample size, information size=estimate, incidence in intervention arm=85/428=19.9, incidence in control arm=91/431=21.1, heterogeneity correction=model variance based, as shown in figure 2. However, there's nowhere for me to put in "δ" value, which was mentioned and set 10% in the study mentioned above. With my setting, the estimated RIS is 35533
, far larger than 812 as mentioned in the article, and wasn't able to perform TSA due to too little information used, as shown in figure 3.
So my question is obvious, how did they obtain such results (please refer to TSA.jpg) ??????
Where did I do incorrectly ?
Thank you for your time going through this and it'd be nice to give me some help !!!
The free software is available at: http://www.ctu.dk/tsa/downloads.aspx
Manual is available here: http://www.ctu.dk/tsa/files/tsa_manual.pdf
Summary of my questions up here:
1. How do you obtain required information size (RIS) by using traditional 5% significance boundary, 'cause for what I'm aware of, RIS can only be calculated by α-spending boundary using O'Brien-Fleming method?
2. Where and how can I set "δ" value while applying α-spending boundary ?
-----------------------------------------------------------------------------------------------------------------------------------------
So I've been working on a project involving meta-analysis and trial sequential anaylsis of noninferiority randomized control trials (RCTs).
I've met some technical problems which I can't solve, so I tried to look for literature of similar study design and luckily I found this study entitled Haloperidol Versus 5-HT 3 Receptor Antagonists for Postoperative Vomiting and QTc Prolongation: A Noninferiority Meta-Analysis and Trial Sequential Analysis of Randomized Controlled Trials.
Regarding the setting of conventional 5% significance boundary and alpha-spending boundary using O'Brien Fleming method, here I quote " The first was the classical boundary, setting the α error to 5%. The required estimated information size was 799, and our cumulative value was found to be 859. Further, we used the O'Brien-Fleming α-spending function for testing for futility. For this the δ was set at 10% and β at 20%. The TSA graph clearly showed that cumulative Z scores were well past the inner wedge of futility and fell within the range of actual equivalence. The information size of 859 was well past the required 812 for this method as well"
Obviously, the equivalence of haloperidol and 5-HT 3 receptor antagonists is proved.
I tried to simulate and run the software myself using exactly the same data provided in this article, as shown in Forest plot.
After adding every single trials with the events/total, I started to set up the parameter needed for applying both conventional 5% significance boundary and α-spending boundary. The parameters needed for setting up the traditional 5% significance boundary are type I error (5%) and boundary type (two-sided) as shown in figure 1. As you can see, there's no where showing the estimated RIS, which gave rise to the first question I mentioned above. Next, I filled in parameters needed to set up α-spending boundary, including α=5%, power=80%, information axis=sample size, information size=estimate, incidence in intervention arm=85/428=19.9, incidence in control arm=91/431=21.1, heterogeneity correction=model variance based, as shown in figure 2. However, there's nowhere for me to put in "δ" value, which was mentioned and set 10% in the study mentioned above. With my setting, the estimated RIS is 35533
, far larger than 812 as mentioned in the article, and wasn't able to perform TSA due to too little information used, as shown in figure 3.So my question is obvious, how did they obtain such results (please refer to TSA.jpg) ??????
Where did I do incorrectly ?
Thank you for your time going through this and it'd be nice to give me some help !!!
