Subgroup analysis The first-line vitality of failed clinical studies?

First, what is subgroup analysis? In randomized clinical trials, subgroup analysis refers to assessing the therapeutic effect of a particular endpoint (e.g. total lifetime) in a patient subgroup defined by baseline characteristics (e.g., age, sex, histology, and race) (e.g., risk ratio HR).
is not recommended to determine subgroups based on post-randomization measurement results, as this may affect the principle of randomization of patients assigned to subgroups.
subgroup analysis helps to explore subgroups of patients who are more likely to benefit from experimental treatments to maximize information from clinical trials.
, new assumptions and experiments can be generated based on these results.
, this can lead to changes in clinical practice.
, subgroup analysis can also be used to evaluate whether the overall therapeutic effect is consistent among the patient subgroups, which is often referred to as a "robustness checking".
for these reasons, regulators support appropriate subgroup analysis.
problems exist in sub-group analysis? Subgroup analysis has two key statistical limitations.
, they are often statistically ineffective (under-powered).
this is because sample size calculations in clinical trials typically consider sufficient statistical effectiveness in all randomized patients, not subsets of patients.
refore, the interaction effect test of whether there are significant differences in the therapeutic effects observed in one subgroup (e.g. gender) and another subset (female) is often ineffective.
, subgroup analysis tends to produce "false negative" results.
limitation of subgroup analysis is that it is particularly prone to multiplicity.
is the increased probability of obtaining "false positive" results, i.e. incorrect conclusions that there are significant differences between treatment groups.
, one or more of these comparisons are more likely to produce significant results by chance through multiple subgroup analyses of the main endpoints.
, for example, if 10 comparisons were made between the main endpoints, there was a 40 per cent chance that at least one of them would produce a false positive result.
, when multiple subgroup analyses were performed, p-values less than 0.05 in a single comparison did not provide sufficient evidence to support significant differences between treatment groups.
the harmful consequences of subgroup analysis? Subgroup analysis can sometimes be used to "save" a failed study that found that the trial group was significantly superior to the controlled drug in a particular patient subgroup, although the main purpose of the trial was not achieved.
However, it is for this reason that the bid may conduct unannued analyses of many subgroups in an attempt to find a subgroup of patients in the treatment group that is significantly superior to the control drug, which is often described as "data dredging" or "fishing trip".
misreading of subgroup analysis may lead to the initiation of clinical studies based on unproven assumptions and ultimately direct damage to the health of the subjects.
cost of understanding these harmful consequences is extremely high, but it is easy to prevent them by understanding the fundamentals of subgroup analysis.
, how to correctly implement and explain subgroup analysis? In order to properly carry out and explain subgroup analysis, it is first necessary to determine whether subgroup analysis is specified in advance.
pre-defined subgroup analysis is intended to perform hypothesis tests, in contrast, unannued (also known as exploratory, retrospective, or ex post) subgroup analysis generates new assumptions and performs "robustness checks."
need to be noted that both can provide valuable information, but there are significant differences in principle and purpose.
can only be concluded on the basis of pre-defined subgroup analysis, or lead to any subsequent changes in clinical practice.
In order to overcome the two statistical limitations of under-effectiveness (reduced degree of certainty) and multiplicity, the following five steps outline the best way to properly conduct, interpret and report pre-defined subgroup analysis: 1, pre-set subgroup analysis in scenarios and/or statistical analysis plans (SAP) Most of the time, pre-specified subgroup analysis should be documented in detail in the programme.
can also be detailed in SAP before data is revealed or before the first patient visit in an open study.
table provides an overview of the information to be recorded when pre-defined subgroup analysis.
pre-defined subgroup analysis is considered more credible because they are scheduled before any data checks.
, however, pre-defined or unplanned subgroup analysis tends to be multiple, i.e. the probability of false positive results increased as a result of testing for multiple subgroups mentioned above.
, pre-defined subgroup analysis alone does not make it automatically effective: it must still be properly interpreted, interpreted and reported in accordance with the following steps.
2, using the Interaction Test interaction test is the most appropriate statistical method for subgroup analysis, a concept that can be explained by the following hypothetical example: Figure 1: What is an interaction effect test? Suppose there are two therapeutic (Tx) groups in clinical trials: Tx Group A and Tx B.
group of patients defined by baseline characteristics also had two levels: male and female.
the regression lines of the connecting circles and squares represent the efficacy of Tx A and Tx B therapy in prolonging total survival, respectively.
, the higher the regression line, the higher the risk of death.
arrows at each level through the subgroup refer to the therapeutic effect.
if the regression lines are parallel, there is no interaction between the therapeutic effect and the sex (Figure A), i.e. the male treatment effect is the same as that of the female.
if the regression lines are not parallel or crossed (Figures B and C), there is a statistically significant interaction between the therapeutic effect and gender, i.e. there is a significant difference between the therapeutic effect of men and women.
3, estimating the therapeutic effects of each level of the subgroup Interaction effect test is usually performed as part of the regression model, the type of regression model depends on the end of the analysis.
Cox scale risk model is a standard method for analyzing clinical trials to the end of event time.
, in the case of this hypothetical example, the Cox model is used for a "therapeutic-gender" interaction effect test that provides HR (Tx A vs Tx B), 95% confidence interval, and associated p-values for each level in the subgroup.
2: Forest maps are usually used to show subgroup analysis results.
above is the result of an example of interaction effects described in Figure 1C.
diamond represents the point estimate of HR (Tx A vs Tx B) and the horizontal line represents the 95% confidence interval.
the relevant p-values for each level of HR in the subgroup should be interpreted with caution.
common mistake is to claim that there is a difference in therapeutic effects because HR-related p-values are statistically significant in men but not in women.
this is incorrect because only the p-value of the interaction test can determine whether there are significant differences in HR observed in different genders.
This is because the interaction effect test considers: (i) prognostication of patients of different levels in the subgroup, for example, the total lifetime of a woman may be better than that of a man, regardless of the treatment assigned;
4, the use of corroborative evidence to verify subgroup results In order to confirm the subgroup results of a single clinical trial, further validation is required in independent studies or meta-analyses.
important to emphasize that, until corropirative evidence is available, subgroup analysis generates only hypotheses, and the therapeutic effect observed in all randomized patients is still considered the most appropriate estimate for each level of patients in the subgroup.
5, responsible reporting of results needs to be reported resyshonshying subgroup results so that others can interpret them appropriately.
results of the primary endpoint analysis for all randomized patients should be highlighted in summaries and conclusions.
, the name of the pre-defined subgroup analysis should be specified and the number of pre-defined and unann planned subgroup analyses should be clearly stated.
the validity of subgroup analysis results should also be discussed in the context of current corroroarative evidence and scientific literature.
the concepts described above apply to any type of endpoint, such as classification (respondent or non-responder), continuity (systolic pressure), or to event time data (total lifetime).
table summarizes the key points that help clinicians correctly interpret subgroup analysis.
: Barraclough H, Govindan R. Biostatistics Primer: What a clinician ought to know: subgroup analyses. J Thorac Oncol. 2010 May; 5(5):741-6.