A paper has appeared in the J Clin Oncol comparing Bendamustine with Chlorambucil in untreated patients with CLL. It shows Bendamustine to be much better than Chlorambucil, but can we believe it?
The trial was conducted by a group of hospitals in Austria, Bulgaria, France, Germany, Italy, Spain, Sweden and the United Kingdom, though the names of the authors do not spring out as CLL experts apart from Marco Montillo from Milan. Very few if any of the patients came from the UK. The trial was funded by the manufactureres and marketers of Bendamustine. The statistics were designed by an outfit in Germany (DSH statistical services) with a record of designing trials in Homeopathy and the data were handled by Cephalon who market Bendamustine as Treanda in the US. In other words this was an industry trial. It is usual with industry trials for the drug manufacturer to pay the hospital participating a fairly large sum of money (perhaps 5000 Euros) for every patient recruited. Of course, I have no access to what financial arrangements were in place for this trial.
319 patients were randomly assigned to receive either Bendamustine or Chlorambucil (162 B, 157 C). The overall response rate was 68% for B and 31% for C. CRs were 31% for B and 2% for C. Median progression-free survival was 21.6 months for B and 8.3 months for C. As I said Bendamustine seems astonishingly better than Chlorambucil, though this does seem a very poor result with Chlorambucil, compared with say the LRF CLL4 trial.
Toxicity was worse with Bendamustine. Grade 3 or 4 adverse events occurred in 40% for B and 19% for C. This level of toxicity was regarded as acceptable.
This paper is published in JCO which probably means it was turned down by Blood. The criticism which any reviewer would have made would have been that the dose of chlorambucil was too low. They have answered this criticism by declaring that the dose of chlorambucil was equivalent to 60 mg/meter squared which compares well with comparisons with fludarabine (Rai = 40), alemtuzumab (Hillmen = 40) and FC (Catovsky = 70). That seems to be alright, but then I read the paper more carefully. The two drugs were given according to quite different formulae. Bendamustine was given in a dose of 100mg/sq meter for two days every 4 weeks, while Chlorambucil was given in a dose of 0.8 mg/kg for two days every two weeks. It seems strange that two different calculations were used and even stranger when I see that rather than weighing the patients they used something called Broca's normal weight. Now I had never heard of this so I Googled it. It turns out to be the height in centimetres minus 100. Does this give the same as weighing? By no means. I am 177cm high so my Broca weight is 77kg. Alas my scales make me 83 kg.
If I calculate the dose of chlorambucil I would have got under the LRF4 formula it would have been 143 mg, but under the Bendamustine paper formula it would have been 123 mg. I'm afraid my ideal weight is a little less than my actual weight. For my mother the discrepancy would have been even more. She is only 5ft 3, but weighs about the same as me. The Bendamustine paper would have given her 96 mg of Chlorambucil while the LRF4 calculation would have given her 139 mg.
So my original criticism stands - they seriously underdosed the Chlorambucil compared to what is optimum - just as the fludarabine and alemtuzumab trials did. Moreover they offer no description of modern prognostic markers - something I would regard as essential in a clinical trial in CLL. It may be that mutated and unmutated cases and del 11q and 17p cases were equally distributed among the two groups, but it may be that they weren't. This might be another explanation of why the Chlorambucil patients did so badly.
The introduction to this paper perpetuates the story that Bendamustine has both alkylating agent and purine analog activities. It is true that it bears superficial structural resemblances to a purine analog, but I have yet to see convincing evidence that it acts as one. I still believe that Bendamustine is just a way of getting adequate doses of an alkylating agent into a patient.
The saddest part of all of this is that most practicing physicians (I include myself) are not equipped to decipher these statistical games. How then could we expect patients to do so. This study is already being touted on various CLL patient forums.
ReplyDeleteAdd the statistical gimickery to the tendency for "ghost" writing of some medical articles (including review articles) and you have a recipe for deceit.
The final puzzle piece is the tendency for some academicians to color their advice with their own perceptions as influenced (perhaps subconsciously) by the bent of their own research.
We are dependent upon referees such as you to help us to decipher things such as this. Thank you once again.
DWCLL
Thank you for your comments on this. JCO really dropped the ball on this, and hopefully your comments will help JCO do a better job in the future.
ReplyDeleteOne simple rule for a non-statistician is to read a study's conclusion and see what qualifiers are mentioned. Usually, if a study has no qualifiers in the conclusion, then it is a misleading study.
This study concluded that, "Bendamustine offers significantly greater efficacy than chlorambucil, and a manageable toxicity profile, when used as first-line therapy in patients with advanced CLL."
The conclusion should have said something to the effect that of the different protocols tested, bendamustine was superior to chlorambucil. The absence of such a simple little qualifying statement should have been a red flag warning to JCO.
And even with a qualifier, other details of this study are questionable - notably, as you mention, the lack of stratification by prognostic markers.
So, what is the point of clinical trials?
ReplyDeleteFC was vaulted into a first line treatment based on the same kind of phoney head to head data with chlorambucil. Millions of dollars are made and lost on these phoney decisions.
This aside Bendamustine is still a viable treatment for CLL, and I would certainly consider it over FC as my option.