The application for AstraZeneca PLC’s platelet inhibitor Brilinta (ticagrelor) ignited strong opinions from one FDA reviewer about the adequacy of the pivotal data and the sponsor’s conduct, drawing a forceful response from the review division director, according to agency documents that are striking for their tone and level of criticism.

In a series of memos, Division of Cardiovascular and Renal Products Cross-Discipline Team Leader Thomas Marciniak strenuously criticized AstraZeneca’s conduct of the pivotal, 18,000-patient PLATO trial underpinning the ticagrelor NDA. Among Marciniak’s criticisms were that the study had inadequate data collection and analyses, poor follow-up, faulty adverse event reporting, failure to adjudicate potential endpoints, an inappropriate primary endpoint and censoring problems.

Cross-Discipline Team Leader Marciniak concluded that problems with PLATO destroyed confidence in the validity of the pivotal trial’s results.

In a May 15, 2011 memo, Marciniak found a lack of support for AstraZeneca’s hypothesis that high doses of aspirin interfered with ticagrelor’s efficacy in PLATO’s U.S. patient subgroup (see related story, Also see "Multiple Aspirin Dose Analyses Reassured FDA On Brilinta’s U.S. Efficacy" - Pink Sheet, 1 January, 2012.). The reviewer also concluded that problems with the PLATO study design, conduct and data submitted destroyed any confidence in the validity of the trial results, meaning that they could not be relied upon to support approval for acute coronary syndromes.

The recommendation against approval absent a new clinical trial marked a change of opinion for Marciniak. In a September 2010 memo, he had supported approval except for ACS patients with ST-elevation myocardial infarction who are undergoing early percutaneous coronary intervention.

Marciniak’s strong criticisms of the PLATO data and AstraZeneca’s conduct drew a stinging rebuttal from his boss, Cardio-Renal Division Director Norman Stockbridge. In a July 8, 2011 memo that endorsed ticagrelor’s approval, Stockbridge cited several areas “in which the decisions that Dr. Marciniak makes in his review lack a persuasive rationale and often lack documentation as to their implications.” He also chastised Marciniak for using what he viewed as biased language.

Below are summaries and excerpts from Marciniak’s May 15 and June 8 reviews and Stockbridge’s July 8 memo.

Primary Endpoint Definition

Marciniak: The reviewer disagreed with AstraZeneca’s decision to include in the primary endpoint non-CV bleeding deaths, to count all unknown deaths as CV deaths, to classify multi-organ failure deaths as non-CV regardless of the initiating event, and to exclude 39 deaths reported in patients not having a primary endpoint and withdrawing consent but having data in the case report forms regarding deaths. Taking these revisions into account, Marciniak re-analyzed the data using his own endpoint, which he described as a more typical major adverse cardiovascular event endpoint than that employed by AstraZeneca.

Stockbridge: “The primary endpoint in PLATO was time to first event of cardiovascular death, myocardial infarction or stroke. Dr. Marciniak refers to this as ‘AZ’s PEP.’ Dr. Marciniak analyzes an endpoint ‘MACE,’ in which he has (a) excluded some, but not all, hemorrhagic deaths, (b) excluded some, but not all, deaths of unknown cause, (c) included some, but not all, multi-organ failure deaths, and (d) included all deaths following withdrawal of consent. All of these steps are at variance to the pre-specified analysis plan. While Dr. Marciniak crafts an alternative endpoint that appears to be reasonable, if post hoc, he describes nothing of the implications of any of these adjustments.”

Adjudication

Marciniak: “In performing my review I checked adjudication packages for 622 patients …. With rare exceptions, I judged that the … adjudications were appropriate, with my disagreements almost invariably limited to cases with equivocal data.”

The reviewer presented statistics using the adjudications for his MACE endpoint and for additional events not submitted by the sponsor for adjudication. He wrote that AstraZeneca failed to submit some potential endpoint events for adjudication and in other cases submitted incomplete adjudication packages, potentially biasing the results. He also cited examples of incorrect recording of adjudicated events that led to the exclusion of endpoint events suffered by ticagrelor patients.

Stockbridge: The division director said that Marciniak’s decision to employ his own adjudications for some cases may have introduced bias into the evaluation. “Despite generally agreeing with the adjudications by the blinded committee, Dr. Marciniak also reports analyses using unadjudicated events reported by the sites. The rationale for using unadjudicated data is unclear. He notes that missing data can be a problem, but there is no reason to believe that missingness was biased”

Data Collection, Submission And Quality

Marciniak: “I formerly recommended approval for ACS patients excluding STEMI patients undergoing early percutaneous coronary intervention. After being assigned as the primary efficacy reviewer for the complete response and completing a more intensive review of the application, I now conclude that there are sufficient problems with PLATO data quality such that, at best, the U.S. results are representative of ticagrelor’s efficacy, i.e., ticagrelor is inferior to clopidogrel in efficacy and safety.”

“PLATO follow-up was poor. While PLATO was a short CV outcomes study with fewer than 60% of the patients needing follow-up of at least 360 days, about 19% of patients alive at the end of study did not have a visit with vital signs on or after the earliest study completion date …. The excess number of ticagrelor patients without vital signs follow-up and without an AZ primary endpoint (148) exceeds the difference in AZ primary endpoint (147). The excess number of ticagrelor patients lacking final vital signs and the MACE endpoint (141) greatly exceeds the difference in MACE endpoints (87). When the rate of missing data exceeds the endpoint difference, one must be concerned about the validity of any conclusions.”

“Minimization of ticagrelor adverse events … was not rare in PLATO …. The scanty information AZ alleges to have collected [on some serious adverse events] … is appalling. Literally, this submission is the worst submission I have encountered for collecting – or at least submitting – information on SAEs. We can have little confidence that we understand the AE profile of ticagrelor and … its efficacy as well. We should also be highly concerned regarding whether AZ is executing adequately its primary responsibility of protecting the safety of participants in its clinical trials.”

“The problems I have documented with AZ’s handling of AEs have two implications: (1) AZ must improve its handling of AEs; and (2) Both safety and efficacy results from PLATO are suspect such that we need results from another trial to justify approval.”

“We, including other FDA divisions, should examine other AZ NDA and IND submissions to determine the prevalence of the AE reporting problems …. We should consider placing on hold ticagrelor clinical trials or other AZ trials in the IND phase …. Placing one or more trials on hold would force AZ to address these problems expeditiously and facilitate confirmation that AZ has addressed the problems.”

The initial NDA submission “did not include all clinical documents, including MedWatch forms and endpoint adjudication packages, with the [case report forms]. It did include an audit trail and unblinding information that, while appearing and represented as complete, later proved to be substantially incomplete.”

“This NDA submission is the worst in my experience regarding completeness of the submission and the sponsor responding completely and accurately to requests …. I believe that we still have incomplete CRFs, incomplete data sets and audit trails, and only a few hundred scanty safety reports.”

Stockbridge: “Dr. Marciniak cites some actual errors, examples of missing data, and his dissatisfaction with the sponsor’s adjudication to ‘conclude that there are sufficient problems with PLATO data quality such that, at best, the U.S. results are representative of ticagrelor’s efficacy.’ How poor data quality leads to an assertion that the true effect of treatment is at least as adverse as the U.S. results is not clear, but at least one needs to consider whether data quality issues undermine our ability to interpret anything from PLATO.

“Dr. Marciniak’s review contains anecdotes for cases he considers problematic. These amount to about one case per 1,000 subjects enrolled, but without subject IDs some of the descriptions may refer to the same subject. Importantly, Dr. Marciniak’s review gives little insight into how cases came under his scrutiny, so the possibility exists for the inadvertent introduction of bias in case selection. Thus, one has to decide whether the total number of problems identified by Dr. Marciniak is enough to call for re-evaluation or a new study, but I do not believe that one can make reliable inferences from the distribution of cases Dr. Marciniak identified by treatment groups.

“As far as I can tell, the data quality issues are not of great concern,” Cardio-Renal Division Director Stockbridge concluded.

“Although Dr. Marciniak seems suspicious, neither he nor [the Division of Scientific Investigations] found evidence of sponsor misbehavior. Considering the effort Dr. Marciniak expended on review of individual cases, he found relatively few problems of any kind. As far as I can tell, the data quality issues are not of great concern. Were we to rely upon non-inferiority results, the matter might be a closer call, but nothing that added noise here would do anything other than make the two groups more similar and reduce the apparent treatment difference. This includes frank errors, censoring times for which I or Dr. Marciniak might have made different choices, and accuracy and completeness of data provided for adjudication.”

“If the error rate were deemed too high, then no one should have used any part of these data in further analyses. Clearly, no one thought these errors to be that problematic. If the error rate was deemed acceptable, the original data should have been used in the analyses, not a dataset with spotty corrections.”

“While bias is a risk in analyses such as that that Dr. Marciniak performed, there simply is not sufficient information in the review to know whether his decisions were biased. One can, however, say that the language he employs displays bias. He refers to the sponsor’s aspirin analyses as ‘cherry-picking’ and based upon ‘post hoc, wildly post-randomization, and erratically defined ASA dosages,’ and he refers to ‘censoring abominations,’ but decisions behind his own analyses appear often post hoc and arbitrary. The sponsor ‘alleges’ and ‘claims’ various things (even when they have faithfully relayed information obtained from investigators), while Dr. Marciniak ‘asserts,’ ‘judges,’ ‘finds,’ and states opinion. Such rhetorical devices do not make his opinions any more compelling.”

Brilinta Review Sparks War Of Words Between FDA Cardio-Renal Division’s Marciniak, Stockbridge

More from Archive

More from Pink Sheet