Recent

% | $
Quotes you view appear here for quick access.

ImmunoCellular Therapeutics, Ltd. Message Board

  • phishes3 phishes3 Feb 15, 2013 9:58 PM Flag

    Model Results and Questions

    So I'll start a new tread since the other is getting long...
    So this is the first "observations" out of my model and I welcome comments, critiques, and questions that may help me dial it in.
    1. I took the "screened" numbers off of the IMUC presentation X (124/278) X 0.6666 for the treatment arm and multiplied the control arm by 0.3333. I distributed them evenly throughout the month unless only one enrolled in the month - then I put them at the end of the month (worst case scenario).
    2. I used the shape of the SOC curve off of the NWBO presentation (n = 119, median OS = 17 months) to create my control survival curve and I used a median of 17.8 months – I placed the same curve in for my control and treatment arm and ran the model 100 times (worst case scenario – zero benefit for treatment arm and a 17.8 median survival – I have seen a few posts talking about 15 months OS and some 18-19 months OS and I’m currently looking at the data that poster EVSWORLD suggested to see if I can get a more accurate and up-to-date OS.)
    Result: 80% probability that the 32nd event falls between the dates of Dec 22nd, 2012 and Feb 15th, 2013 with the highest probable week being the week of Jan 26th. Not what I was expecting (would have hoped that the date would have been closer to last year) I would like for some of the other guys with models to do the same thing (use a similar median and use the “control” numbers for both arms and see what the result is).
    Info or insight that would help me - one big variable is the actual enrollment – does anyone have any notes from CC back in 2011 that gave some of the actual enrollment numbers at different times to check against my calculations? The 124 number still bugs me – the FDA website says 200 planned – why not continue until 200 is reached? Do the “under” enrollments cause any potential issues with the FDA if the trial is successful and IMUC wants to proceed with approval before a P3. Does anyone on here know how many additional months of OS would be needed to show a statistical difference given the number included in the trial? (I will try to research P values later… but if someone here already knows then please share =)
    Once the 32nd event occurs it is very useful to model different medians to see what “cumulative” OS gives the same date – then given an estimated control OS you can determine the treatment arm OS.
    I plan on studying more before I decide to make a larger investment (week before last I made a small investment based on my initial research and I’m trying to determine my next move.)

    SortNewest  |  Oldest  |  Most Replied Expand all replies
    • That's pretty similar to what I did except I stayed at the month granularity. I also randomize the assignment to the control and active vaccine arms in the Monte Carlo simulation, I also randomized the screening with the 44% chance of getting on trial (but then I re-normalized the results at the end to match the 124 on trial because the randomization to trial caused me to get different numbers).

      I like this approach because it gives you estimates of the probabilities which is more useful.

      By the way, I ran my numbers and set my curves to 17.8 months and I got an average time to 32 events of 23.5 months. Which I figure is mid January, so our models are pretty close. I used the 15 month survival curves and just stretch them in time and interpolate so I can generate approximate curves of any media OS.

      If I remember correctly the p-value is the probability of observing your hypothesis by chance given the null hypothesis. For example, given the control arm OS, what is the chance that the results are at least as good as the active arm results just by chance.

    • phishes,
      There are only 124 patients in the study not 278. Check out the criteria for acceptance in the trial program.
      1. patient must have been diagnosed within 30 days of enrollment.
      2.. Soon after surgery doctors ask patient if they want to be enrolled in a trial study.
      If yes, they are listed as enrolled.
      Not all patients want to enter a trial so this is a slow process of enrollment. Patients must go through a time consuming process to be in a trial. They must come back to the hospital constantly to get injections and follow-up and may live hours away. Then they don't even know if they are even getting the drug or the placebo.
      3. If they said OK, then the patient must fill out a complicated questionaire and accept the posible risks.
      4. the enrolled patient then gives a blood sample to be analyzed but the criteria for this study says they must be HLA-A1 or HLA-A2 positive which according to IMUC is only 45% - 50% of the GBM population.
      5. then of the accepted patients some small number of patients have side effects and drop out.

      The remainder left in the actual study are the 124 patients of 278 originally enrolled.
      Ralph

      Sentiment: Strong Buy

    • phishes,

      Could you be more explicit about your patient enrollment data. That is just put dates in one column, and right next to it put the number of patients enrolled on that date. Do we actually have explicit enrollment data, or are you and Disco making some educated guesses about how patients are enrolled. What are your assumptions.

      I am new here and one of the critical assumptions about modeling is enrollment data. How accurate is it? So far I have only seen a couple of press releases where announcements about number of patients enrolled on a certain date have been announced. Pretty lumpy data?

      • 1 Reply to frederickforrest
      • Yes - I took the number "enrolled" per month according to the IMUC slide in their presentation (estimated the number each month based on the graph) - this total is 278. I then took this monthly number and applied the following formula in excel: +ROUND((X-Y)*0.446*0.66666,0) where 'X' is the number enrolled in month N and 'Y' is the number enrolled in month N-1. The 0.446 is just the outcome of 124 (actual number in the trial) divided by 278 (number enrolled per the chart). The 0.666666 multiplication factor represents the 2/3s that enter the treatment arm and in the control arm you would use 1/3 or 0.333333.

        Bottom line - here are my monthly numbers I used in the model:

        Dates:
        Jan-11
        Feb-11
        Mar-11
        Apr-11
        May-11
        Jun-11
        Jul-11
        Aug-11
        Sep-11
        Oct-11
        Nov-11
        Dec-11
        Jan-12
        Feb-12
        Mar-12
        Apr-12
        May-12
        Jun-12
        Jul-12
        Aug-12

        Treatment arm:
        1.0
        0.0
        1.0
        0.0
        1.0
        2.0
        1.0
        4.0
        5.0
        4.0
        6.0
        5.0
        7.0
        7.0
        6.0
        7.0
        7.0
        7.0
        7.0
        5.0

        Control arm:
        0.0
        0.0
        0.0
        0.0
        0.0
        1.0
        1.0
        2.0
        3.0
        2.0
        3.0
        2.0
        4.0
        4.0
        3.0
        3.0
        4.0
        4.0
        3.0
        2.0

 
IMUC
0.115-0.0022(-1.88%)Sep 28 4:02 PMEDT