Dear members of the CMS NCD review board,
As a concerned scientist, physician and citizen, I am writing to alert you to the multiple suppressed but significant faults of the Provenge phase 3 clinical program. In the interest of higher standards of clinical research and patient care, and in the name of many reticent colleagues, I am determined to challenge the silence on this issue and reignite the vox scientia. As a scientist, physician and citizen, it is my duty to do so. The IMPACT study explored the efficacy of Provenge (Sipuleucel-T) in the treatment of asymptomatic or minimally symptomatic metastatic castrate resistant (hormone refractory) prostate cancer. This study is presented as a randomized, double-blinded, placebo-controlled clinical study. However, the study was neither truly blinded nor truly placebo-controlled, casting doubts on the survival benefit associated with the study treatment (Provenge). Below, I will briefly describe for you some of the major flaws of the IMPACT protocol and analysis, while referencing and directing you to the appropriate public and official sources of information on this trial:
LACK OF STUDY BLINDINGA thorough examination of Dendreon’s phase 3 clinical program in its entirety and the company’s official interactions with the FDA reveal how unblinding was introduced in the old study protocols and later remained as an undisputed flaw in the IMPACT protocol. The major relevant events are summarized in the following timeline, adopted from the FDA CBER clinical review of Provenge [Reference 1, pages 11-14]:
- On 09/03/1999, the FDA agreed to the design of Studies D9901 and D9902 (including the efficacy endpoints, patient population, control arm, and study procedures) and the proposed analyses.
- On 07/20/2001, the FDA agreed that the clinical development plan (D9901 and D9902) was sufficient to support a license application for sipuleucel-T; FDA requested clarification of objective disease progression endpoint.
- In Oct 2002, the primary analysis of study D9901 demonstrated that overall study results were negative, but Provenge delayed time to objective disease progression only in the non pre-specified subgroup of patients with Gleason < 7. Study data was submitted to the FDA and discussed at the Type A Meeting on 11/22/2002. In the same meeting, the FDA agreed that Study D9902 could be split into 2 parts: D9902A would include subjects already enrolled regardless of Gleason score; D9902B (named IMPACT) would be initiated, to include subjects with Gleason scores of 7 and less. These study populations could not be combined for efficacy analysis.
- On 05/30/2003, Dendreon received a Special Protocol Assessment (SPA) agreement for Protocol D9902B (IMPACT): time to objective disease progression and time to disease related pain were co-primary endpoints.
- On 07/30/ 2003 Provenge received Fast Track designation based on the potential to prolong TTP and time to disease related pain (TDRP) in men with asymptomatic, metastatic, Gleason Sum < 7 AIPC.
- In October 2004, D9901 survival analysis demonstrated a survival increase of Provenge compared with placebo in the ITT population.
- On 11/24/2004, the D9901 and D9902A Final Statistical Analysis Plan (SAP) was submitted to the FDA; the agency agreed to the proposed D9902A SAP with primary endpoint of time to disease progression and adding overall survival as secondary endpoint.
- On 11/25/2005, Dendreon obtained Amendment 7 to the SPA of D9902B (IMPACT). Major changes included elimination of the Gleason score restriction, expansion of the eligibility criteria to include minimally symptomatic patients, and elevation of survival to the primary endpoint.
As you see, the original D9901 study was only designed to test the efficacy of Provenge in delaying disease progression. Study subjects were unblinded following disease progression, after information on this primary endpoint had been collected in the blinded mode. As such, this was a properly designed and conducted protocol, though it showed no Provenge benefit in PFS. After a retrospective analysis showed a Provenge benefit in survival, OS was chosen to replace PFS as primary endpoint of the D9901B study (IMPACT). However, the IMPACT study protocol remained unchanged, allowing the unblinding of study subjects following disease progression and before observation of the primary endpoint variable (death).
It is then important to appreciate that the final design of the IMPACT study was never approved by regulators a priori as a whole; rather it devolved via trial-and-error readjustments due to lack of regulatory vigilance. The IMPACT study schema that we are all used to viewing, as supplied by the company and as presented at scientific meetings [reference 2, slide 4], is incomplete at best and misleading at worst. The study schema must contain a clear and time-coded reference to the unblinding event occurring mid-study. This major flaw was noted by the FDA during its review, though surprisingly not followed up. However, the FDA review documents contain a better approximation of the true study schema [reference 1, page 20]. Unblinding of patients at mid-study after their disease progression is (1) unacceptable generally, for all the reasons why we have developed and maintain a double-blinded standard in clinical trial design, and (2) unacceptable specifically, as the primary root of the other overlooked flaws in IMPACT conduct and analysis described below.
INADEQUATE PLACEBO ARMA thorough examination of the study design and of study salvage treatment statistics reveals that IMPACT was inadequately controlled. The list below includes the major reasons why the group of study subjects who did not received Provenge following randomization cannot and should not be used as a placebo comparator arm.
- The post-progression unblinding is a pivotal intervention that likely impacted all placebo patients who learned that they had not received the experimental treatment prior to their cancer progression. Furthermore, on 03/29/2007, when the FDA advisory committee voted to endorse Provenge efficacy and recommend approval, IMPACT was ongoing and within 100 patients of achieving its target accrual of 512 patients [reference 3, page 317]. Therefore all patients who were alive in the study at that time as well as the 100 patients yet to be recruited were aware of these news while they were unblinded post-progression: they were aware that they had or had not received a treatment which a regulatory panel endorsed as efficacious. Additionally, after the FDA’s decision to postpone approval until IMPACT results were available, these patients as well as their physicians became aware that the approval of Provenge as a new therapy depended on the success of their individual cases and their rolling trial. These events exerted more than simple traditional “nocebo” and “placebo” effects on the unblinded patients. They constituted active albeit unintentional interference with both study arms.
- After disease progression, unblinded “placebo” subjects could choose to receive APC8015F. APC8015F was prepared from the frozen PBMC product that was collected at the time of preparation of the product used as control in the study (APC Placebo). The frozen product was later thawed and pulsed with PA2024 antigen and prepared in a similar manner as Provenge. The median survival time for subjects in the sipuleucel-T arm was 25.8 months, for subjects in the control arm who subsequently received APC8015F (63.7% of placebo group) was 23.8 months, and for subjects in the control arm who did not receive APC8015F (46.3% of placebo group) was 11.6 months. The administration of ACP8015F to more than half of the “placebo” patients post-progression and post-unblinding is a significant intervention confounding the further subgroup analysis of taxotere salvage (as described in the next section).
- The Kaplan Meier survival plots suggest flawed randomization: although there has never been evidence of anti-tumor effects and three different phase 3 trials including IMPACT have failed to observe PFS benefit [reference 2, page 14], the Provenge and placebo survival curves begin to separate surprisingly fast [reference 2, page 8]; the survival of placebo patients who did not receive APC8015F [reference 1, page 40] and the survival of placebo patients who did not received taxotere [reference 1, page 44] are surprisingly poor for asymptomatic patients with minimal metastatic disease. These placebo patients whose salvage therapy fate was undeniably influenced by unblinding at disease progression, appear to be the true drivers of the separation between the Provenge and placebo survival curves.
- Both “placebo” and Provenge patients were pheresed 3 times during the study, however the “placebo” patients were reinfused with only 1/3 of their original pheresed PBMCs load, while Provenge patients were reinfused with their entire original pheresed PBMCs load [reference 1, page 39; reference 4, page 12]. It is important to note that this “relative immunosuppression” or “precipitated immunosenescence” of the placebo patients was only partially corrected in the placebo fraction who chose to received ACP8015F (i.e. the frozen 2/3 of their original PBMC pheresed load, but only after disease progression), but it was never corrected in the placebo fraction who never received ACP8015P and who ultimately performed poorest in the survival analysis. The above interpretation obviously relies on the assumption that delayed reinfusion of the original frozen PBMC “corrects” the acquired immunosuppression in the placebo patients, though it is very possible that this is not the case. Additionally, a recent anonymous review [reference 6] suggests that the fraction of T-cells depleted in these patients are of the naive subtype which cannot be replenished from lymphocyte-storing tissues in older adults as in younger ones; this flaw would account for the drastic survival difference between patients under and above age 65. Although, unfortunately, I cannot take credit for the document referenced above, I believe that it raises valid concerns which have never been addressed. While the question whether “placebo” patients were actually harmed by being partially immunosuppressed during this study remains open to scientific and ethical debate, the lack of intervention equivalence between “placebo” and Provenge is undisputable. As such, the “placebo” arm in the IMPACT study does not provide an adequate comparator.
INVALID DATA ANALYSISThe use of on-study taxotere creates further imbalance between the study arms. 50.3% of “placebo” patients received taxotere during the study compared to 57.2% of Provenge patients [reference 2, page 11]. “Placebo” patients received taxotere a median of 2.4 months later than their Provenge counterparts following randomization [reference 1, page 43]. Meanwhile, patients who subsequently received taxotere tended to be younger with greater predicted survival and fewer bone metastases [reference 5]. Since taxotere remains the only intervention proven to extend survival in this patient population, this taxotere imbalance weighted towards the Provenge arm must face the utmost scrutiny. However, the FDA as well as many physicians were satisfied with Dendreon’s dismissal of the above imbalances: when entered as a time-dependent covariate, docetaxel use was not a significant predictor of OS. This statistical methodology relies on the assumption that taxotere is an external time-dependent covariate. However, because taxotere was only given following disease progression mid-study and after the patients had been unblinded, taxotere is nothing but a textbook example of an internal time-dependent covariate, which by definition cannot be used for prediction in the survival function [references 7, 8]. Therefore, the time-dependent covariate analysis of taxotere presented by Dendreon and the study investigators is invalid at best and misleading at worst. The taxotere imbalances remain to date unaddressed, and the Provenge arm survival benefit should be partly or fully attributed to taxotere unless proven otherwise.
It is surprising that the FDA review overlooked this basic statistical flaw. Meanwhile, the FDA statistical analysis employed another flawed argument to dismiss the significance of the taxotere imbalance. The statistical review document states: “the fact that placebo patients received taxotere much later indicates that they needed to be alive longer than the Provenge-treated patients to receive taxotere, so they are likely to be better than the Provenge-treated patients in terms of unknown characteristics” [reference 9, pages 24]. The FDA statisticians have overlooked the fact that the majority of placebo patients opted to received ACP8015F following disease progression, thus delaying taxotere treatment. The delay in taxotere treatment in the placebo patients is not an indication of their relative health, rather it is a testament to the confounding design of the IMPACT study, while simultaneous being yet another possible reason for the separation of the placebo and Provenge arm survival curves.
The poor design of the IMPACT trial, the questionable efficacy of Provenge in altering disease and extending life, the high efficacy of public/political pressure in quenching discourse, the FDA approval in the absence of robust data—all set dangerous precedents in therapeutic development and regulatory action, especially in an emerging field like cancer immunotherapy which is both high-impact and lucrative. In this case, the FDA has failed to make an independent recommendation based on clinical evidence. The FDA has failed to foster scientific excellence, and as such it has failed us and our cancer patients.
Unfortunately, as you may know, the Provenge row has extended past the clinics and the discussion tables. During the 2007 Provenge advisory committee meeting at the FDA, two of the leading urology experts in the US found that there was not substantial evidence of benefit and voted against Provenge approval. Additionally, after the panel, they send open letters to the FDA, emphatically criticizing the trial design and reliability of the outcome data (note that IMPACT inherited the same design as the D9901 study that they criticized). In response, these two physicians were bombarded with anonymous harassment including murder threats [reference 10]. Since then, no physician or scientist has ever again openly questioned Provenge efficacy. This malignant pressure has effectively smothered all scientific discourse on Provenge in the US leading to the recent FDA approval on 04/30/2010. Based on this record of harassment, I have chosen to comment anonymously out of concern for my personal safety and that of my family. Meanwhile, I must inform you that many colleagues share my concerns and objections to the IMPACT trial design, conduct and analysis but have chosen to remain reticent for fear of a backlash.
I can appreciate that while my anonymity is a necessity for me, it may undermine my trustworthiness before you. If so, then I encourage you to treat this letter only as a pretext for further diligence rather than as an independent source. I encourage you to examine the official sources referenced herein and listed at the end. They are all neglected public documents, including the FDA briefing document for the Provenge panel in 2007, the recently-released FDA clinical and statistical regulatory review documents, as well as presentations from the ASCO 2010 meeting. I have also included the official URL addresses where you will be able to immediately download these documents for free. While these sources may initially seem inconsequential as they are not officially under your review, they are essential to appreciating the major faults that render the IMPACT results invalid: the broken blinding, the inadequacy of the comparator arm, and the taxotere imbalance and flawed statistical analysis. A careful examination of these official sources shows that the IMPACT study needs to be re-designed and re-conducted, and the efficacy of of Provenge as a life-extending intervention still remains to be proven.
To emphasize the importance of the issue that lies before you, I will note that the eyes of the world are on you. After the inflated passing grade that Provenge received at the FDA, European, Canadian and Japanese regulators are closely watching for cues to the true validity of the IMPACT trial. Medicare reimbursement of Provenge will be seen as an endorsement of the study design and validation of the study’s scientific rigor. I encourage you as last gatekeepers of Provenge to boldly question its clinical program and results in a fair, independent and well-informed way, thus upholding high clinical and scientific standards. We must remember that scientific vigilance and free discourse have already been pressured into silent abstention in this case, and they need and deserve our help—your help.
 Clinical Review and Evaluation. CBER, Food and Drug Administration, PROVENGE (Sipuleucel T) BLA # 125197. (http://www.fda.gov/downloads/BiologicsBloodVaccines/CellularGeneTherapyProducts/ApprovedProducts/UCM214540.pdf)
 Updated Results of the IMPACT Trial of Sipuleucel-T for Metastatic, Castration-Resistant Prostate Cancer (CRPC). 2010 ASCO Genitourinary Cancers Symposium; Oral Presentation. March 5, 2010. (http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9Mjc5NDIxNnxDaGlsZElEPTM3MTk4OHxUeXBlPTI=&t=1)
 Meeting Transcript. Cellular, Tissue and Gene Therapies Advisory Committee Meeting. Food and Drug Administration, PROVENGE (Sipuleucel T) BLA # 125197; March 29, 2007. (http://www.fda.gov/ohrms/dockets/ac/07/transcripts/2007-4291T1.pdf)
 Clinical Briefing Document. Cellular, Tissue and Gene Therapies Advisory Committee Meeting. Food and Drug Administration, PROVENGE (Sipuleucel T) BLA # 125197; March 29, 2007. (http://www.fda.gov/ohrms/dockets/ac/07/briefing/2007-
 Persistence of immunotherapy survival effects of sipuleucel-T and relationship to postrandomization docetaxel use in phase III studies. 2010 ASCO Annual Meeting; Poster Presentation. June 7, 2010. (http://abstract.asco.org/AbstView_74_53600.html)
 "Provenge PhIII Trials – The Alternative Explanation of Survival Results." Anonymous. (http://caretolive.com/wp-content/uploads/2010/07/provenge-approval-fud.pdf)
 Fisher LD, Lin DY. Time-dependent covariates in the Cox proportional-hazards regression model. Annu Rev Public Health. 1999;20:145-57. (http://www.ncbi.nlm.nih.gov/pubmed/10352854)
 Cortese G, Andersen PK. Competing risks and time-dependent covariates. Biom J. 2010 Feb;52(1):138-58. (http://www.ncbi.nlm.nih.gov/pubmed/20029852)
 Statistical Review and Evaluation. CBER, Food and Drug Administration, PROVENGE (Sipuleucel T) BLA # 125197. (http://www.fda.gov/downloads/BiologicsBloodVaccines/CellularGeneTherapyProducts/ApprovedProducts/UCM214543.pdf)
 Doctors threatened for opposing drug. New York Times. June 4, 2007. (http://www.nytimes.com/2007/06/04/health/04drug.html)