There is a more unfortunate scenario that often plays out. That is that they get a different statistician in who has a goal of getting positive looking results out of a negative result set though various means and then often carefully manipulating the wording of the success criteria to match their best attempt at making the data work. This is quite common, my partner (in the girlfriend sense) work within medial journal publishing and sees it quite a lot.
While not impossible, it is very tricky to do this before you actually have the data. I'm advocating that the wording of the test criteria, statistical tests to perform, etc, are all decided before the trial.
Bringing in a different statistician post-hoc is pointless, all he could do is run the exact predetermined statistical test. In fact, you can do even better: submit code to the journal before the study, and submit data after the study. The journal runs the code on the data and decides which conclusion to publish.
Accumulate this over the whole range of companies and studies and we end with a disaster.
If you were doing a private study where the results would only be for your own private usage, you would go the extra mile and pay the extra money for good, accurate science.
But the financiers here don't really care about whether the result is accurate. They usually are optimizing something else entirely and accuracy is just some minimum threshold to avoid being accused of fraud.
Also see Ionnidis's paper on why most published research is wrong: http://medicine.plosjournals.org/perlserv/?request=get-docum...
I can't say a whole lot about it (I am pretty sure I am still bound by agreements and truthfully it wouldn't be polite), but I once was employed in clinical trials and this line is a load of crap. I have seen trials stopped. Some came back, but it wasn't the same trial.
They had a belief in what the outcome should be before the trial (else you would not go through with / spend money on the trial). A deviation from the expected results really needs to be looked at and you really need to find out why all the prelim testing is not accurately predicting what is actually going on.
Yes, the deviation from expectation really needs to be looked at. The problem is that the financial reward for the companies might be greater to just 'move on' and never publish those unexpectedly negative results. But this means that the unexpectedly positive results for the next test are given undue weight.
I don't think the problem is really the science, or the publishers, or the drug companies, but the unrealistic weight placed on erroneously calculated statistics. This combined with a lot of money at stake forces the companies to "produce results" whether or not this is of long term public benefit.
This entirely destroys the assumptions underlying the supposed statistical power of a test.
Therein lies the problem. We know the solution: tweak the publishing process.
Currently: 1) Do study 2) Submit paper, including conclusions. 3) Journal accepts, based on correctness of methodology and interestingness of conclusion.
To prevent this result, just change the order of things.
1) Submit paper explaining methodology, and "alternate endings" (positive and negative). 2) Journal accepts paper based on methodology only. Methodology section of paper is published. 3) Authors do study and submit results. Journal publishes appropriate conclusion section. 4) If necessary, authors submit additional observational results.