1 / 9

PUBH7420 – Group 7 April 23 rd , 2012

Should subgroup analyses be performed if the overall results of a clinical trial show no difference between treatment groups? . PUBH7420 – Group 7 April 23 rd , 2012. Three types of people. The Skeptic Automatically dismisses subgroup analysis as most likely due to the wiliness of chance

cybele
Download Presentation

PUBH7420 – Group 7 April 23 rd , 2012

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Should subgroup analyses be performed if the overall results of a clinical trial show no difference between treatment groups? PUBH7420 – Group 7 April 23rd, 2012

  2. Three types of people • The Skeptic • Automatically dismisses subgroup analysis as most likely due to the wiliness of chance • The Believer • Data are unlikely to play tricks; data could indicate that a subgroup truly differs from others • The Agnostic • Perhaps sides with the skeptic but has an uneasy feeling that subgroup analysis may in fact show something real Witte (2009)

  3. The Skeptic • The trial probably was not powered to find differences within subgroups, thus: • Subgroup analysis is a great way to guarantee finding a statistically significant difference. • If we divide the data into subgroups, even just 2, it is unlikely that we’ll see the same effect in the subgroups. • If the overall treatment effect seems to be approaching significance - one subgroup may appear to have a large effect despite reduced power/increased type 1 error. • These analyses are typically not pre-planned or pre-specified.

  4. The Believer • Pre-planned is not intrinsically better than post-hoc. • Is the strength of evidence for a hypothesis based upon whether someone has asked the question before or after data was collected? • Is it more likely to be right simply because somebody thought of it? • Perhaps, but then what we are doing is implicitly putting prior beliefs on hypotheses. • If estimated treatment effect differs between subgroups, the data is providing real information which we should acknowledge.

  5. The Skeptic • The overall result of an RCT is a better estimate of treatment effect in the various subgroups examined than are the observed effects in those individual subgroups. • In Statistics, we call this phenomenon Stein’s Paradox and it applies to meta-analyses as well. • A genuine difference between subgroups is not necessarily due to the classification of the subgroups. • Difference could arise due to chance.

  6. The Believer • Stein’s paradox is a heuristic. • Does this point apply to hypothesis testing? Power is greater with the larger sample size of a general sample, but if there is a statistically significant result in the subgroup analysis doesn’t that mean low power was overcome? • Randomization should diminish differences that arise by chance – confounding is controlled for and we expect confounders to be distributed similarly between treatment groups.

  7. The Believer • Challenge the authority of the principle that thou shalt adjust for multiple testing. • This principle applies uncontroversially if I have a compelling reason to group the tests/intervals as part of one decision process. • Alternatively, if you want to protect against the inflated type 1 error, multiple tests should be adjusted for.

  8. The Skeptic • Post-hoc subgroup analysis should only be considered as a hypothesis generating mechanism for further research. • Caveat: subgroup analyses can result—and have resulted—in trials that waste time/money and put patients at risk. • Course text has numerous examples of failed trials based off the findings of a subgroup analysis. • Publication bias

  9. What to do… • Distinguish between prior and data-derived hypotheses. • Do not calculate p-values for data-derived hypotheses. • Place greater emphasis on the overall result than on what may be apparent within a particular subgroup. • Use tests of “interactions,” and/or correct for multiplicity of statistical comparisons.

More Related