I think statistical models are great - I don’t really understand how the programs are written and how it all gets calculated, but I’m sure the engineer, statistical, and math types solidly get it. A
Still, I can’t wrap my mind around how the numbers actually work out to whatever the statistical numbers were spit out. Just working with the censoring formula earlier in the year was difficult enough. So I'm left to rely on listening to the critiques of those more statistically inclined sources we have around us. And if ex thinks something actually works, given his proclivity towards the negative, I tend toward wanting to hear his take on something.
So for me, while it is not as scientific, I also gravitate towards using formulas I can understand.
Now I know that medians are what scientists look for. But I think averages are also important. I mean, if I had a cancer, I’d want to know what the average lifespan in front of me was - and notes much the median.
And averages are numbers I can much more easily calculate.
With that in mind, I’m going to start with PFS numbers. Some may actually find this interesting.
The trial protocol indicated that the investigators thought that they would achieve these medians (not averages) for PFS.
7 months for control
13.5 for treatment
Of course, bear in mind that there had been a 3 month pipeline for all of these patients so this would really be more like
10 months for control
16.5 for treatment
Still… let’s work with the protocol hypothesized numbers.
By May 2014, 165 patients were enrolled.
So by when were 248 patients enrolled?
And I want to estimate this because 248 PFS is the number of events to be counted in the trial.
From May 2014 to November 2015, 165 more patients were enrolled. That’s 19 months.
Now that was likely not so exact as others have taken the enrollment curve that Dr. Bosch presented at ASCO and had 9 in some months and 10 in others.
Still… I’m going to work with 8.68 patients being added per month.
248 - 165 = 83 more patients to be enrolled / 8.68 = 9.56 more months to enroll 248 patients
May 2014 + 9.56 months = February 2015 approximately 248 patients were enrolled.
Now lets work with the medians expected for this trial and turn them into averages.
I’ll also take the number of 248 and based on the percentage of control and percentage of treatment enrolled in the actual trial,
and it would mean that approximately:
82 control patients were in the 248 number (110/331 = .3323 - 248 x .3323 = 82)
165 treatment patients were in the 248 number (248 - 82 = 165)
Now I’ll multiply the average number of months based on the medians expected to each of these numbers.
82 x 7 PFS months = 574
165 x 13.5 PFS months = 2227.5
574 + 2227.5 = 2801.5 total months / 248 PFS events = 11.30 BLENDED AVERAGE PFS months one might expect if the hypothesized numbers work
In other words, if the averages were to work, it would mean on average, all the 248 patients evented within 11.3 months.
So if by February 2015 - 248 patients were enrolled in the trial,
11.3 months later, theoretically, if all of those patients were following the protocol hypothesized median numbers, the trial would hit its 248th event.
So February 2015 + 11.3 months = January 2017.
PFS numbers were PR’d to have been reached on February 6, 2017.
Now I know this isn’t super scientific. But I thought it was at least interesting.
To wrap up then, based on the approximate time that 248 patients were enrolled in the trial, if they, on average, evented according to what the investigators had hypothesized to reach statistical significance, those 248 patients would have evented by January 2017.