I created this simple model to try and predict how NWBO’s Phase 3 clinical trial is performing.
I consider this model at best a “rough-estimate”, from an engineer’s perspective (not a mathematicians). This model does not account for an enrollment ramp.
I used the Dose-Dense Temozolomide for newly diagnosed Glioblastoma study (D-D) as a foundation for this model. NWBO’s vaccine (DCVax-L) is getting compared to today’s SOC which is this D-D study (Arm1). The D-D study examined an increased chemotherapy dosing schedule to SOC but ultimately showed no statistical difference between the two treatments.
My model begins by assuming NWBO’s vaccine is ineffective, meaning that it has the same efficacy as today’s SOC. The model uses the D-D study’s Kaplan-Meier chart (K-M) to estimate when NWBO’s trial can be expected to reach its completion date. Note, NWBO’s trial has not reached its completion dated, it is “on-going” and “near completion” according to the company’s more recent statements.
I made a change to the D-D K-M chart so that it better represents NWBO’s trial patient population. The D-D study included a class of patients know as pseudo progressive (psPD) whereas the NWBO trial has excluded these patients. PsPD patients are believed to be the best GBM survivors and can make-up over 20% of the entire GBM population. So for my analysis I have disregarded the 20% longest survivors from the D-D study, assuming they will mainly be psPD patients.
Then I calculated the required PFS rate to reach trial completion. Using that PFS rate, I then extrapolated from the D-D K-M chart to get an expected survival time. Using that survival time, I then estimated a trial completion date. This is the point at which the trial would end had all patients only received SOC treatment.
As of September 2016, my model gives NWBO a 3 month advantage over SOC.
However NWBO’s advantage is made-up from two parts, its treatment (DCVax-L) and its control. Assuming the trial’s control has no advantage then the treatment’s advantage increases. I used NWBO’s trial randomization ratio to allocate proportions.
My model then gives DCVax-L a greater than 4 month advantage verse SOC thereby reaching a demonstrable efficacy level.
Disclosure - I am long NWBO. I do not short stocks.
foolish amateurish "analysis". You have NO idea whether the trial is ongoing. I have seen this many times before.Most recently in ONCY now ONCYF. The blind has almost certainly been broken and they arena win the process of "analyzing" --read torturing the data for subgroups or any other glimmer of hopeful data.
ReplyDeleteRandomly tossing the longest surviving 20% is also useless and unfounded on any reationt thought process. Additionally the tastea on PsD is all over the lot. Where did you get the 20% figure? Moreover, NWBO did NOT exclude PsD patients which can ONLY be done ex post facto. At the time of enrollment NWBO would have NO way of determining who would be PsD. Finally determine what is Psd is not precise science so even when exercising care many instances of PsD and true improvement would be impossible to determine. Stick to engineering.
I do not think we are going to find any common ground on this issue.
DeleteThe 20% psPD figure source is linked in the discussion of this post but here it is again http://www.ncbi.nlm.nih.gov/pubmed/18484594
Branko,
ReplyDeleteI appreciate your attempt to do the modeling here but there are several very important factors that are not mentioned which could more than make up the difference in advantage that you conclude may exist. First the trials would need to be matched for patient age, resection percentage and performance scores as a mimimum. Differences in these factors alone could easily make up the advantage you might otherwise believe exist. The pseudo percentage that you removed from the Temozolomide study also is an expected percentage in the DCVax-L trial but some late pseudos are almost a surely in the Phase 3 trial and might to probably are being diagnosed as progressors. There is plenty to debate about this point but is a factor worth giving a value to in any analysis. A range of 5%-15% reduction would be a more conservative approach. I personally believe we are currently in the "no man's land" zone right now if using only blinded data referrence points. Best wishes.
Anonymous,
DeleteThank-you for your feed-back.
When I run my model with your suggested assumptions all of the DCVax-L benefit disappears (I used a 10% psPD factor & a 4 month benefit for better patient prognostics) so the trial would be in the "no man's land" zone as you say.
I estimated a lower 2 month benefit for better patient prognostics. An accurate figure for this factor is not available.
The K-M chart for MGMT methylated tumors gives patients a 3 month PFS benefit (8.7 vs 5.7). Maybe a number like that 3 months would be better assumption for the improved patient prognostics. Regards.
There is one easy way to evaluate if DCVax's Phase 3 will be successful, and that is the fact that it is now Sept 2016, and it is the estimated date for the trial to have reached all its events, and the number of patients were held (due to enrollment ), which I believe was that there was confidence in the trial and a money saving decision. Lastly, any month after this that these events do not happen, strengthen the case that the results will be positive. You can goggle my article on NWBO statistics to evaluate my expertise in statistics. Just goggle NWBO Statistically Speaking in Seeking Alpha and read my statistical analysis of the Phase II data. Bohsie
ReplyDeleteBohsie,
ReplyDeleteThank-you for your contributions. I like your evaluation approach.
Considering how costly NWBO's DCVax-L screening suspension has been it would be quite ironic if they instigated it by themselves thinking it would save money.
And very nice work with statistical analysis back in 2013.