Northwest Biotherapeutics (OTCQB:NWBO) has been testing DCVax-L on GBM, a deadly form of brain cancer, in a Phase 3 clinical trial that has been ongoing since 2008. The trial reached its predetermined end-point of 248 Progression-Free-Survival (NYSE:PFS) events sometime in late 2015 or early 2016, but the company has elected to keep the trial blinded to capture additional clinical data from the trial. Presumably, this additional data will strengthen the statistical strength of the final data analysis, and further define the “tail” of the survival curve immunotherapies are known for.
Northwest Biotherapeutics recently published clinical data from the DCVax trial. The trial is still blinded, and therefore the data published only included blended data. In other words, the data published is the consolidated data from all patients in the trial; and the clinical benefit of the experimental treatment is still yet to be determined. The clinical data published included the following blended Survival chart which has a mOS of 23.1 months.
The previous modeling efforts suggested that the range of the mOS was perhaps between 24.8 and 28.6 months relative to surgery, thus the model results were 1.7 months too high. At the time the model was developed, the only publicly available information was the patient enrollment chart and an approximate time when 233 survival events were reached. Nothing was known about the shape of the survival curve at the time. But with the shape of the NWBO blended survival curve now in hand, the model has been revised to utilize this new information.
Before proceeding, the reader should carefully consider the following factors while evaluating the model results:
- Lost-To-Follow-Up (LTFU) provisions are not included in the mode.
- The model does not provide provisions for the Cross-Over design of the actual trial.
- The model does not provide provisions for the 165 patients receiving other treatments after disease progression. However, only 9 patients received treatments that are proven to extend survival.
- The Blended survival curve provided by NWBO is used as the basis for both the control and treatment arms in the revised model. The model uses “multipliers” to adjust the model control and treatment arms to achieve the desired goals of achieving approximately 233 events in July 2017 and simultaneously achieving a blended mOS of approximately 23.1 months.
- The trial is blinded and there is no way to precisely know the actual performance of the control or treatment arms.
- The shapes of the actual control and treatment survival curves will likely be more unique than the model curves discussed in this article which could result in more or less favorable results.
Additionally, the blended survival curve provided by NWBO inexplicably did not provide any data past 36 months. Therefore, some assumptions about the tail of the survival curve had to be drawn. NWBO previously stated that the trial should reach the trial 233 OS event endpoint in July of 2017. Using this information, a polynomial was generated using excel which traced the survival curve provided by NWBO up until 36 months, and then the polynomial tail was adjusted until the model generated close to 233 events on July 2017 while maintaining an approximate mOS of 23.1 months. The model survival curve, shown below, produced 236 events and a mOS of 23.1 months which is close enough to the desired targets for this exercise. The model survival curve is shown below overlaying the published blended survival curve with no adjustments made to the control and treatment arms (multiplier = 1.0 for both arms).
To begin the revised modeling effort, the control arm performance must be reasonably assumed to infer the possible clinical benefit of the DCVax treatment. One methodology involved utilizing an online GBM Survival Calculator to determine the expected survival for a group of patients with demographics similar to the DCVAX trial. This particular calculator uses the patient’s Age, Gender, KPS status, Extent of Resection, and MGMT status as inputs, and generates the 6-month, 12-month, and 24 month survival probability.
The NWBO published data included the demographics shown in Table 1.0 below; and Table 2.0 shows the survival probabilities generated by the Survival Calculator for the various categories. The Blended rows are weighted averages using the demographics from Table 1.0 and the Survival Calculator data from other rows. The last row is the expected survival probabilities for patients with chacteristics similar to those in the DCVAX trial.
The calculator survival probabilities are derived from a GBM trial with an assumed randomization time of 1 month relative to surgery. Therefore 1 month was added to the survival times to align the survival values with those provided by NWBO which were relative to surgery. These values were plotted, and the median survival time was determined to be approximately 18.5 months.
The model was revised to use the curve fitted polynomial previously discussed to determine the clinical benefit possible. The model was adjusted until the mOS of the control arm was 18.5 months which resulted in a treatment benefit of 7.4 months, a hazard ratio of 0.61, and p-value of less than 0.01. The Survival Chart for this scenario is shown below.
The caveat to this exercise is that the NWBO clinical trial, like most all clinical trials, have stringent inclusion/exclusion criteria which means the population of patients in the NWBO clinical trial is different from the population of patients from which the survival calculator is based. And even though, the patients prognositic charateristics were accomodated for in this exercise, it is still not an apples to apples comparison. In all likelihood, the NWBO trial enrolled healthier patients into the trial based on its enrollment criteria than the GBM trial that was used to develop the calculator. This will likely result in the control arm mOS exceeding 18.5 months predicted by the calculator.
The next scenario was to use a control arm with similar patient demographics and enrollment criteria to the NWBO trial to model the control arm. For this scenario the IMUC ICT-107 Phase 2 trial was used which had a mOS of 19.4 months relative to surgery. The patient demographics for IMUC (left side) and NWBO (right side) trials are shown below for comparison:
The reader should draw their own conclusions regarding the extent of the similarities of the two populations sets. The reader should also consider the following factors:
- IMUC screened for disease progression (an exclusion criteria) sooner than the NWBO trial which may have resulted in poorer prognosis patients in the IMUC trial.
- IMUC has speculated that their placebo was immunologically active and may have contributed to better outcomes for their control arm patients
There are other factors to consider as well, but these considerations are beyond the scope of this article.
When using 19.4 months as the mOS goal, the model resulted in a treatment benefit of 5.6 months, a hazard ratio of 0.69, and a p-value of 0.01. The Survival Chart for this scenario is shown below.
As stated previously, the revised model control and treatment arms are both based on the Blended survival curve provided by NWBO. The control arm was adjusted until the mOS matched the IMUC control arm. In other words, the IMUC control arm survival curve was not used as the basis for the model control arm. This same scenario is shown below with the DCVax Blended and IMUC survival curves for comparison.
Aside from the front-end, the model DCVax control arm closely followed the IMUC control arm. One could argue that the IMUC trial may have had better prognostic patients in its trial based on the front-end comparison shown in this chart. But the one must also consider that the IMUC control arm only had 43 control arm patients, while the NWBO control arm has 99 patients. Regardless, the front-end of all survival curves are the most accurate because more patients are represented in the front-end of the curve versus the tail-end which typically has very few patients.
The data from one of the model runs associated with Scenario #2 is provided in this link for independent validation. The data is in the form of a zip file that contains 100 spreadsheets that the model automatically generates for each unique trial.
The last scenario involved adjusting the control arm until the model generated results at the edge of statistical significance. The model was adjusted until the mOS of the control arm was 20.6 months which resulted in a treatment benefit of 4.7 months, a hazard ratio of 0.76, and p-value of 0.05. The Survival Chart for this scenario is shown below.
The statistical summary of all three scenarios is shown in the table below. The “CTRL” columns are the model control arms and the “TRMT” columns are the model DCVAX arms. The “BLEND” columns are the blended results.
The methodology of the model was discussed in the previous articles, but as a reminder, the data above is the consolidated data from 100 simulated trials for each scenario. As also explained in the previous articles, the model uses polynomial survival curves to predict each patient's survival, however, each patient is randomly selected on the y-axis (proportion) to generate the estimated survival in months. Because each patient is randomly selected, the outcome of each simulated trial is unique; which results in some of the simulated trials yielding statistically insignificant results. This is due entirely to chance because an assumed treatment effect (via the multipliers) are being applied. As the clinical benefit decreases, it becomes increasingly more difficult to detect a statistical difference between the control and treatment arms.
Survival Curve Comparisons
The following charts compare DCVax blended survival curve against other survival curves. It is left to the reader to draw their own inferences. The charts are all based on the curve fitted polynomials used by the model but are reasonably representative of the actual survival curves. Using the model generated curves allows more comparisons between survival curves because the model adjusts the curves to be relative to surgery. The reader can confirm each curve approximates the actual survival curve to gain reassurance the curves are indeed representative.
The first survival chart only shows the first 36 months and the second survival chart extends to 60 months. The reader is cautioned that the tails on some of the survival curves are extrapolated since there was insufficient data on some of the trials to plot the curves out to 60 months which is yet another advantage of using the polynomial curves. (The months of survival data available is as follows: DCVax Blended: 36m, IMUC-32m, Optune: 60m, Rintega: 48m, UCLA: 84m)
The revised modeling results presented in this article once again demonstrates the NWBO clinical trial is performing relatively well when compared to reasonable historical expectations. However, the control arm could be performing better than expected which could result in clinical failure for the NWBO clinical trial. Additionally, the final overall survival analysis will be confounded by the “cross-over” design, and the progression-free-survival analysis will be confounded by “pseudo-progression”.
The balance sheet of the company leaves a lot to be desired and additional stock dilution is imminent. This and other issues has landed the company in the cross-hairs of vulture financiers, short sellers, and social media critics. Some of the criticism of the company has certainly been warranted; however, some of the criticism has been unfair, biased, and detrimental to the company’s ability to raise capital.
There are no guarantees in life, other than that old tired cliché (D&T), and a positive outcome for the NWBO clinical trial is far from guaranteed; however, modeling suggests that the NWBO trial has a reasonable chance for success. And even if a modest improvement in survival is demonstrated; the biggest winners of this clinical trial will be cancer victims; the most important stakeholders.
The reader is cautioned that there are numerous limitations to the model. Additionally, the design of the model could be flawed; the data used to analyzed the trial could be inaccurate or uncorrelated to the actual NWBO clinical trial; and the assumptions used in the model could be incorrect. The information in this article is NOT investment advice; and is solely the opinion of the author.
- First results on survival from a large Phase 3 clinical trial of an autologous dendritic cell vaccine in newly diagnosed glioblastoma
- leprecon7777's Blog Posts
- Meeting Library | Meeting Library
Disclosure: I am/we are long NWBO.