Archived Comments for:
Subgroup Analysis of Trials Is Rarely Easy (SATIRE): a study protocol for a systematic review to characterize the analysis, reporting, and claim of subgroup effects in randomized trials
Ewout Steyerberg, Dept of Public Health, Erasmus MC
24 November 2009
The SATIRE group is to be applauded for their effort to shed more light at the minefield of subgroup analysis, reporting, and interpretation. The question may however be what truly new insights will be obtained. One might predict that findings from a well-selected set of papers published in 2007 will be largely similar to what was found before in the studies shown in Table 1.
Of particular interest may be the answer to question 10 in Appendix 1: "Is the interaction consistent across closed related outcomes within the study?". In a previous individual patient data (IPD) meta-analysis of over 30,000 patients from 6 trials, we found statistically significant subgroup effects in 2 trials .. but in opposite directions [1]. Identifying such IPD meta-analyses may be difficult when the individual trial subgroup effects were reported in 2007; one might need to perform a specific search for these studies, and consider multiple years?
Another point of interest is how often interaction with prognostic risk, e.g. from a multivariable regression model, was studied, rather than an interaction with a single characteristic. Such an approach was emphasized by Pocock and Lubsen in 2008 [2], but earlier examples may well be available. For example, a similar relative effect of accelerated tissue plasminogen activator (TPA) versus streptokinase treatment was found across the full range of mortality risk from acute myocardial infarction in the GUSTO-I trial, where the authors based risk on a multivariable combination of 5 baseline characteristics in a logistic regression model [3].
I like to encourage the authors to consider these 2 specific points in their review, or perform specific studies if necessary.
References 1. Hernandez AV, Westerhout CM, Steyerberg EW, Ioannidis JP, Bueno H, White H, Theroux P, Moliterno DJ, Armstrong PW, Califf RM et al: Effects of platelet glycoprotein IIb/IIIa receptor blockers in non-ST segment elevation acute coronary syndromes: benefit and harm in different age subgroups. Heart 2007, 93(4):450-455. 2. Pocock SJ, Lubsen J: More on subgroup analyses in clinical trials. N Engl J Med 2008, 358(19):2076; author reply 2076-2077. 3. Califf RM, Woodlief LH, Harrell FE, Jr., Lee KL, White HD, Guerci A, Barbash GI, Simes RJ, Weaver WD, Simoons ML et al: Selection of thrombolytic therapy for individual patients: development of a clinical model. GUSTO-I Investigators. Am Heart J 1997, 133(6):630-639.
Trials
Searching for new insights in subgroup analyses
24 November 2009
The SATIRE group is to be applauded for their effort to shed more light at the minefield of subgroup analysis, reporting, and interpretation. The question may however be what truly new insights will be obtained. One might predict that findings from a well-selected set of papers published in 2007 will be largely similar to what was found before in the studies shown in Table 1.
Of particular interest may be the answer to question 10 in Appendix 1: "Is the interaction consistent across closed related outcomes within the study?". In a previous individual patient data (IPD) meta-analysis of over 30,000 patients from 6 trials, we found statistically significant subgroup effects in 2 trials .. but in opposite directions [1]. Identifying such IPD meta-analyses may be difficult when the individual trial subgroup effects were reported in 2007; one might need to perform a specific search for these studies, and consider multiple years?
Another point of interest is how often interaction with prognostic risk, e.g. from a multivariable regression model, was studied, rather than an interaction with a single characteristic. Such an approach was emphasized by Pocock and Lubsen in 2008 [2], but earlier examples may well be available. For example, a similar relative effect of accelerated tissue plasminogen activator (TPA) versus streptokinase treatment was found across the full range of mortality risk from acute myocardial infarction in the GUSTO-I trial, where the authors based risk on a multivariable combination of 5 baseline characteristics in a logistic regression model [3].
I like to encourage the authors to consider these 2 specific points in their review, or perform specific studies if necessary.
References
1. Hernandez AV, Westerhout CM, Steyerberg EW, Ioannidis JP, Bueno H, White H, Theroux P, Moliterno DJ, Armstrong PW, Califf RM et al: Effects of platelet glycoprotein IIb/IIIa receptor blockers in non-ST segment elevation acute coronary syndromes: benefit and harm in different age subgroups. Heart 2007, 93(4):450-455.
2. Pocock SJ, Lubsen J: More on subgroup analyses in clinical trials. N Engl J Med 2008, 358(19):2076; author reply 2076-2077.
3. Califf RM, Woodlief LH, Harrell FE, Jr., Lee KL, White HD, Guerci A, Barbash GI, Simes RJ, Weaver WD, Simoons ML et al: Selection of thrombolytic therapy for individual patients: development of a clinical model. GUSTO-I Investigators. Am Heart J 1997, 133(6):630-639.
Competing interests
No competing interests