Analysis of BLOOM-DM results and media bias

Written by Erewenguy from YMB


Analysis of BLOOM-DM results


There are a couple of problems with the media’s stand-alone
interpretation of the valvulopathy data from the bloom-dm trial. The
first problem has been pointed out repeatedly, that although the
difference between the placebo and lorcaserin mean is large, it is the
placebo mean that is out of the ordinary. The lorcaserin mean is
consistent with other trials.

Some have suggested that the 0.5% placebo value should be discarded as
an outlier, but unfortunately, the data are the data. To pick and
choose what data to analyze leads one to pre-determined conclusions.
So the data WILL be included, which is causing so much
hyperventilating in the media.

The real problem then is a problem of perception which results from
bias inherent in comparing real numbers. This is one of the reasons
the field of statistics exists.

I don’t really want to turn this into a statistics lesson, but would
like to reduce some of the “6X INCREASE IN VALVULOPATHY!” panic and to
show that the actual situation is a bit more complicated than 7
patients vs 1 patient.

For rare-event data with non-normal distributions, such as VHD, the
data should be transformed prior to analysis. A typical transformation
is log(N+1), but I don’t know what specific transformation the FDA
requires.  Transformation is performed to reduce type 1 errors, or
mistakenly rejecting a true null hypothesis. In our case, the null
hypothesis is that lorcaserin mean = placebo mean.

Actual raw data comparison is 2.9 vs 0.5 (a 5.8X difference)
Transformed comparison is 0.591  vs 0.176 (a 3.4X difference, using my
transformation value).

To further complicate the interpretation of the raw data, the FDA is
allowing a margin of error value to assess VHD, with the null
hypothesis that lorc mean is not greater than (1.5) x placebo mean.

So the actual comparison is not 2.9 vs 0.5, but 2.9 vs 0.75.
Transformed comparison now becomes 0.591 vs 0.243 (a 2.4X difference)

Even here, we cannot compare actual numbers, but we have to conduct
the statistical analysis to state with a relative degree of confidence
that these 2 values do not differ outside of what might be expected to
occur by chance.

If the FDA has an ounce of integrity, lorcaserin should be rejected if
it is proven to be unsafe. However, if a proper statistical analysis
does not reject the null hypothesis, I would expect that with that
same ounce of integrity the FDA should declare that VHD is not an
issue and we move forward.

I believe that in the cc it was stated that there was no difference
between treatments at P=0.10, and will assume arna’s statistician is
qualified and capable