Are more suspicious when the budget impact is over £200 million
HTA reviewers are quite rightly trained to be suspicious of post hoc sub-group analyses of clinical data. We all know that if you do enough analyses you will eventually find something passing a test of statistical significance.
NICE published an interesting example on 17th August:
in its ACD guidance on dabigatran in atrial fibrillation.
The RE-LY trial had three arms, a dabigatran 150mg dose, a dabigatran 110mg dose and warfarin. The EMA licensed what the maker, Bohringer, call a sequence option, of 150mg dabigatran to the age of 80 and 110mg thereafter (with some scope for clinician judgement on bleeding risk in between).
NICE’s assessment obviously has to focus on the licensed sequence but the clinical data underpinning the license did not come from a pre-planned analysis, and this made the Appraisal Committee unhappy; indeed, this is the most substantive criticism in the document in my opinion, and seemingly the basis for a request for re-worked analysis.
Bohringer’s model of the sequence used data from the trial on people aged under the age of 80 for the efficacy of the 150mg dose and data from the trial on people aged over 80 for the 110mg dose. The Appraisal Committee said this was not appropriate and asked for the efficacy of the two doses to be based on data from the trial for people of all ages.
Question 1: I don’t know who first asked for this post hoc analysis split at the age of 80 – was it Bohringer or was it the EMA? If it was the EMA and they thought it was adequate as the basis for a license then does that put the post hoc sub-group analysis in a different light? Or does the potential budget impact of this medicine in this indication over-ride that?
Question 2: I think I am right in saying that bleeding risk is age-related, therefore the basis for the sequence has clinical plausibility. But if this is the case, isn’t there a trade-off to be made between what NICE sees as the inappropriateness of post hoc sub-group analysis versus its clinical plausibility and the relevance of the age-specific efficacy data? It’s not obvious from the ACD that was considered.