After six years of public debate, an FDA panel put to rest many of the concerns about the cardiovascular safety of Avandia rosiglitazone using a mix of observational studies, meta-analyses and controlled trials. The panel's deliberations could shed light on how to balance and interpret such data as the agency is increasingly faced with a similar melange of real-world and clinical data to evaluate the benefit and risks of drugs in the postmarket.

At a joint meeting of the Endocrinologic and Metabolic Drugs and the Drug Safety and Risk Management advisory committees last week, panel members voted to relax the REMS on diabetes drug Avandia from GlaxoSmithKline plc.

Among the 26-member panel, 13 voted to modify the REMS, 7 to remove it, 5 to maintain it, and 1 to withdraw Avandia from the market. Twelve of the 13 who voted to modify the REMS recommended that it be eased.

The vote is an about-face from a 2010 recommendation to place restrictions on the drug after the panel determined results from the RECORD cardiovascular outcomes trial could not be trusted. While the original RECORD results showed no increased CV risk for Avandia, FDA's Thomas Marciniak, medical team leader in the Division of Cardiovascular and Renal products, picked apart the trial at that meeting and suggested impropriety by GSK in the handling of the data.

At the time, the panel struggled to balance GSK's reported results with Marciniak's critique of RECORD, along with previous meta-analyses and observational studies that suggested an increased risk.

In September 2010, FDA restricted the use of Avandia and requested that GSK have the RECORD data readjudicated by a third party (see BioCentury, July 19, 2010).

The game changer at last week's panel was the independent readjudication of RECORD by Duke Clinical Research Institute. The readjudicated results found no increased cardiovascular risk for Avandia, confirming the original analysis.

"What's different now is that the results of readjudication in RECORD, in my mind, have moved the needle," said temporary voting member Marvin Konstam.

Konstam, chief physician executive of the Cardiovascular Center at Tufts Medical Center, voted to withdraw Avandia in 2010, but last week voted to relax the drug's REMS.

While Marciniak again attempted to discredit RECORD along with the readjudication process, senior FDA officials and staff pushed back, and the committee ultimately devoted little time to Marciniak's latest critique.

Instead, the panel focused on weighing the strength of evidence from observational studies and meta-analyses alongside RECORD.

While most members acknowledged that randomized, controlled trials with robust clinical data collection processes remain the gold standard, they said observational studies and meta-analyses with large effect sizes where the hypothesis is prespecified are valuable and could be used to make clinical decisions.

The panel's deliberations could be relevant to FDA's efforts to assess benefit-risk in the postmarket setting and also could have implications for decisions that payers and clinicians will be making based on real-world data once a drug is approved.

Time warp

The debate over Avandia's CV risk has stretched across six years and three FDA advisory committee meetings.

A firestorm was sparked in 2007, when Steven Nissen and Kathy Wolski of the Cleveland Clinic published a meta-analysis of 42 trials that showed an increased risk of myocardial infarction (MI) and CV mortality for Avandia.

Nissen and Wolski identified a 43% higher risk of MI for Avandia (95% CI: 1.03, 1.98) and an increased risk of 64% for CV mortality (95% CI: 0.98, 2.74). Results were published in the June 14, 2007, New England Journal of Medicine.

In August 2007, FDA held its first panel meeting to discuss the results along with FDA's own meta-analysis and an interim look at the RECORD data. FDA's meta-analysis showed an increased risk of 50% for MI (95% CI: 0.9, 2.5) and 70% for cardiovascular death (95% CI: 0.7, 4.0). RECORD did not.

RECORD was an open-label trial that enrolled 4,447 patients and compared Avandia plus metformin or a sulfonylurea to metformin plus sulfonylurea. The primary endpoint was the composite of CV death and CV hospitalizations.

An interim analysis showed little to no effect for Avandia. The HR for CV mortality was 0.83 (95% CI: 0.51, 1.36; p=0.46), and the HR for MI was 1.16 (95% CI: 0.75, 1.81; p=0.50).

Despite the mixed signals, members of the 2007 panel concluded the benefits of Avandia outweighed the risks and voted 22-1 in favor of keeping the drug on the market (see BioCentury, Aug. 6, 2007).

In July 2010, the panel met to discuss the complete results from RECORD.

The trial met the primary endpoint with an HR of 0.99 (95% CI: 0.85, 1.16). Also, the upper bound of the 95% confidence interval was below the 1.3 threshold required under FDA's 2008 guidance for cardiovascular outcomes trials of diabetes drugs.

However, FDA's Marciniak cast considerable doubt on the results and suggested GSK mishandled the data.

Marciniak reviewed about 12% of the 4,447 case report forms and found problems with 13% of them. He reported that the problematic forms were four times more likely to favor Avandia than the comparator arm and that endpoint problems were twice as likely to favor Avandia.

He also said GSK was given "unrestricted access" to unblinded data.

Marciniak readjudicated the RECORD data with an updated definition of major adverse CV events and exclusion of cardiovascular hospitalizations, showing a MACE HR of 1.13 (95% CI: 0.85, 1.5). Marciniak's updated definition of MACE included MI, stroke and CV death.

After his analysis, the panel had little confidence in the RECORD data.

At the time, temporary voting member Michael Proschan said the imbalance in discrepancies favoring Avandia "scares the hell out of me." He concluded: "I don't feel comfortable about the results of RECORD."

Proschan is a mathematical statistician in the Biostatistics Research Branch at the National Institute of Allergy and Infectious Diseases (NIAID).

The 2010 committee split into three almost equal groups that recommended mild, moderate and extreme regulatory actions: 10 voted to keep Avandia on the market with existing or updated labeling, 10 advised adding limits on its use, and 12 wanted the drug withdrawn.

In September 2010, FDA decided to keep Avandia on the market but restricted its use to patients who could not tolerate or could not maintain adequate glucose control on any other diabetic agents. The agency also issued a REMS and asked GSK to have an independent third party readjudicate RECORD.

Simultaneously, EMA asked GSK to withdraw Avandia from the European market.

Lone wolf

Marciniak played a pivotal role in the 2010 meeting, but last week he was a sideshow.

Given a 30-minute slot at the start of the two-day meeting to provide his personal perspective on the readjudicated RECORD data, Marciniak came out swinging. Just as he had done in briefing documents ahead of the meeting, Marciniak blasted not only the integrity of GSK but also that of Duke Clinical Research Institute.

Marciniak went through a laundry list of issues with DCRI and suggested the institute had mishandled other trials where it was asked to adjudicate CV outcomes, including the Phase III PLATO trial of Brilinta ticagrelor vs. clopidogrel, and the Phase III ARISTOTLE trial of Eliquis apixaban vs. warfarin for atrial fibrillation.

He noted DCRI was paid by GSK and that the pharma gave the data directly to the researchers.

Marciniak stopped short of accusing GSK or DCRI of fraud, but said FDA should have been involved in the data transfer to ensure its completeness and integrity.

Where Marciniak's critique of RECORD received little pushback from senior agency officials in 2010, last week was a different story.

Ellis Unger, director of the Center for Drug Evaluation and Research's Office of Drug Evaluation-I, noted the data belong to GSK, not FDA.

"The sponsor generated the data. They own the data and have to protect the confidentiality of the data," he said.

Mary Parks, director in the division of Metabolism and Endocrinology Products, echoed Unger's response and added: "I think the question to ask is do we have evidence that there has been willful redaction of information or evidence provided that GSK has not abided by the expectations of FDA."

Later in the meeting, Parks and other FDA officials said there was no evidence of impropriety on GSK's part.

"We are not aware of any evidence that [GSK] did not provide DCRI with all of the data," said John Jenkins, director of the Office of New Drugs.

Also, the shoe was on the other foot for Marciniak when Karen Murry Mahoney, lead medical officer in the Division of Metabolism and Endocrinology Products, noted it was Marciniak who was unblinded to the RECORD results before he did his analysis in 2010 - an accusation Marciniak had made of GSK.

In briefing materials for last week's meeting, Mahoney called Marciniak's 2010 review "unacceptable."

"Regrettably, the case reviews of his original consult were not blinded; although he states that some type of blinding occurred, it appears that he himself did the redactions, then followed with his own review," she wrote. "This is not an acceptable review procedure if one wants to put forth one's review work as blinded."

Replaying RECORD

In last week's proceedings, panel members asked the agency and GSK a few questions about the selection of DCRI and the transfer of data, but did not dwell on this or any of Marciniak's other complaints.

Instead, the panel praised Duke's review of RECORD, and some members expressed regret about their 2010 decisions.

"I think that the readjudication by Duke really struck a chord with me in that GSK had done a really good job with the study," said Ellen Seely, permanent voting member of the Endocrinologic and Metabolic Drugs Advisory Committee.

Seely is a professor of medicine at Harvard Medical School and a practicing endocrinologist at Brigham and Women's Hospital.

Konstam agreed.

"I do think that the DCRI adjudication process went a long way to help erase events in 2010 vis-à -vis bias and potential malfeasance on ascertainment of cases. In 2010, I didn't feel like I could look to RECORD at all. I don't feel this way today," he said.

Patient representative Rebecca Killion voiced regret over the panel's 2010 decision in light of the Duke results.

"Based on DCRI's adjudication, it went a long way for me in validating the sponsor's initial data. It also tells me that we were jumping the gun in terms of concerns in 2010," she said.

Killion has been the patient representative on all three panel meetings.

Katherine Flegal, temporary voting member and senior research scientist at the National Center for Health Statistics at the Centers for Disease Control and Prevention (CDC), echoed the comment. "I wonder, if at the 2010 meeting, if there had not been so much uncertainty about RECORD that our deliberations would have been different," she said.

Nonetheless, panel members acknowledged that RECORD had weaknesses that could not be overcome by readjudication, namely the open-label design.

"Because of the lack of a blind for investigators, I have a lot more uncertainty about whether the endpoints are being captured completely," said Erica Brittain, permanent voting member of the Endocrinologic and Metabolic Drugs committee.

Brittain is a mathematical statistician in NIAID's Biostatistics Research Branch.

Konstam and others also raised concerns about the "high degree of missingness."

DCRI was able to readjudicate the mortality endpoint with data from all but 87 of the 4,447 patients. The missing data were balanced between Avandia and the control group.

However, both FDA and DCRI noted in their presentations that there were more missing data for the MACE endpoint. In the Avandia group, 346 (15.6%) patients were not followed for MI or stroke to the end of study. The rate was similar in the control group, with 398 (17.9%) patients lost to follow-up.

Observational expedition

FDA asked the panel to discuss how the different data sources should be weighed in evaluating the CV risk of Avandia. Committee members were mixed on how much weight should be placed on the meta-analyses and observational studies.

Kate Gelperin, medical officer in FDA's Division of Epidemiology, presented data from 21 observational studies of CV endpoints for Avandia, including studies that compared Avandia to other antidiabetic agents such as Actos pioglitazone from Takeda Pharmaceutical Co. Ltd. Avandia and Actos are both thiazolidinedione (TZD) peroxisome proliferation activated receptor (PPAR) gamma agonists.

The updated analysis included seven studies not in FDA's 2010 analysis.

In the 10 trials that tested Avandia vs. another diabetic agent - excluding Actos - three favored Avandia in terms of acute MI, with one that was statistically significant. Three were neutral, and the remaining four favored an alternative agent. One of the four was statistically significant.

Four out of four studies of Avandia vs. Actos in patients at least 65 years old favored Actos in terms of acute MI, but none were statistically significant. However, in another seven studies, six favored Actos in terms of all-cause mortality, including four that were statistically significant. One study favored Avandia, but it was not statistically significant.

The panel split on these observational data.

Temporary voting member Sanjay Kaul dismissed most of the observational studies because they lacked a prespecified hypothesis. "Prespecification is very important in interpretation of the data. Otherwise, you can go on a fishing expedition and the data drive the hypothesis."

Kaul is director of the Fellowship Training Program in Cardiovascular Disease at Cedars Sinai Heart Institute and professor in the Division of Cardiology at David Geffen School of Medicine at UCLA.

Only one of the studies included in FDA's observational analysis had a prespecified hypothesis.

However, Konstam and others felt the observational studies were valuable.

"Generally speaking, I would have completely dismissed all of the observational findings that I've seen, except this series of studies where pioglitazone was better than rosiglitazone in mortality," he said.

Konstam noted the consistency of the results across the different studies and the similar backgrounds of the patients helped to support the findings of the four Actos vs. Avandia studies.

Maria Suarez-Almazor, a professor at MD Anderson Cancer Center, also argued the consistency of the observational studies made them equally if not more reliable than RECORD. She also felt more comfortable with the observational studies because "they can provide a more real-world experience compared to clinical trials."

Suarez-Almazor is a permanent voting member of the Drug Safety and Risk Management committee.

META-tating

The panel also considered Nissen's 2007 meta-analysis, as well as FDA's meta-analyses of 52 randomized trials that had been presented at the 2010 panel meeting.

FDA's meta-analysis showed an odds ratio of 1.38 (95% CI: 0.72, 2.72) for all-cause death, OR of 1.46 (95% CI: 0.60, 3.77) for CV death, an OR of 1.8 (95% CI: 1.03, 3.25) for MI and an OR of 1.44 (95% CI: 0.95, 2.20) for MACE. The increase in MI was statistically significant.

A subgroup analysis of the MACE endpoint in active-controlled trials showed a lower OR of 1.05 (95% CI:0.48, 2.34).

Kaul noted the results from FDA's analysis were driven by events in placebo-controlled trials, which accounted for about 80% of the studies.

"The RECORD results for MACE are similar to what we get in FDA's meta-analyses when we only look at active-controlled trials," he said.

Kaul said he thought the consistency between the subgroup analysis and RECORD was reassuring. Brittain agreed.

"Even though the point estimates are different than in RECORD, there is some consistency, with the point estimates going in the same direction," she said.

Kaul also called FDA's analysis "high quality" because "it only included randomized controlled trials confined to diabetics, and long-term trials were excluded. It also focused on a clinically relevant endpoint - MACE - and the method used, exact odds ratio, was appropriate. Other meta-analyses don't do that."

For example, Kaul argued Nissen's analysis was flawed because, while the majority of trials included were short-term studies of about six months in duration, it also included two long-term studies, one of which didn't include actual diabetic patients. DREAM was a three-year trial in patients at risk of developing diabetes, and ADOPT was a five-year trial in diabetics.

When these trials were removed from Nissen's analysis, the odds ratio for MI increased to 1.45 from 1.43, but the lower bound of the 95% CI decreased to 0.88 from 1.03 and rendered the results non-significant with a p-value of 0.15 vs. a p-value of 0.03 from the complete analysis.

Kaul also said DREAM shouldn't have been included because it was the hypothesis-generating study.

Indeed, Nissen stated at the previous advisory meetings that the reason for doing the meta-analysis in the first place was due to the imbalance in the number of acute MIs in the Avandia arm of DREAM.

Konstam said the meta-analyses are valuable but have to be interpreted according to the context of the trials used.

In this case, "these are short-term and mostly placebo-controlled trials, so you can draw conclusions that maybe increasing glycemic control in the short term may drive short-term events," but the analyses could not be used to interpret long-term outcomes, he said.

Real-world implications

Guest speaker Hertzel Gerstein noted observational studies and meta-analyses should be interpreted with caution. Gerstein is a professor and director of the Division of Endocrinology and Metabolism in the Population Health Research Institute at McMaster University and Hamilton Health Sciences.

For example, he pointed to multiple observational studies that showed benefits of treatments like hormone replacement therapy in women that were later disproven.

"ERT therapy was touted as the therapy to give postmenopausal women until the large randomized, controlled Women's Health Initiative showed that it was harmful," Gerstein said.

FDA plans to rely more on real-world data to monitor the benefits and risks of the drugs in the postmarket setting under its new benefit-risk framework (see BioCentury, March 25).

After the meeting, Gerstein told BioCentury that as more real-world data become available, and regulators and clinicians are faced with mounds of results from observational studies, the best way to interpret the results is based on the magnitude of benefit.

"Unless it is a five- or tenfold swing in either direction, it shouldn't change the treatment decision," he said.

Gerstein also noted that regulators like FDA "understand the importance of collecting these data and are obliging companies to do that postmarket, particularly for drugs that are meant to be taken for many years."

However, just like clinicians, regulators must interpret real-world data cautiously and recognize its limitations, he said.

Easing up

At the end of the day, panel members concluded that the DCRI readjudication erased significant ambiguity about Avandia's safety and that the restrictions on the drug should be eased.

"When you have uncertainty in data, you should be allowed to exercise your own clinical judgment. These are judgment calls, and doctors should be allowed to make judgment calls in the face of uncertainty," Kaul said.

An increased risk for bladder cancer for Actos also weighed on panelists.

In September 2010, FDA said it would review the safety of Actos after interim data from a long-term observational study showed a statistically significant increased risk of bladder cancer in patients exposed to the diabetes drug for more than 24 months.

In August 2011, FDA updated the label for Actos-containing products to include dose- and duration-dependent increased risk of bladder cancer. The label also warns against using Actos in patients with active bladder cancer and to "use caution" in patients with a history of bladder cancer.

Harvard's Seely and others believed Avandia should remain on the market to provide an option to patients. However, they felt there was still some uncertainty about its safety and thus the REMS should be eased but remain in place.

The current REMS for Avandia includes patient and prescriber certification, along with a restricted distribution system.

Typical were the comments of temporary voting member Ruth Day: "I think there needs to be a lightening up of all of this intense monitoring, about the restrictive part of the program. It is a huge amount of time and energy," for doctors and patients.

Among the 26-member panel, 16 had been present at the 2010 meeting and half of them changed their vote in favor of Avandia. Two of three who voted in 2010 to withdraw the drug this time voted to loosen the restrictions; the third voted to keep the REMS as is.

Additionally, panel members like Day who voted for a REMS in 2010, now concluded it was too strict (see "RECORD Replay," A3).

Day and others said the certification programs and restricted distribution should be replaced with a communication plan and medication guide.

Too late?

If FDA follows the panel's recommendation, it could be easier for doctors and patients to get access to Avandia. But a look at prescription and sales data trends since 2007 suggest a major uptick in Avandia use is unlikely.

Prescription data presented by FDA show declining use since Nissen's 2007 meta-analysis.

According to the agency's presentation, prescriptions for Avandia from retail pharmacies peaked at 12.7 million in 2006. By the 2007 panel meeting, prescriptions had dropped below 10 million. Similarly, the number of patients who had received an Avandia-containing prescription in 2008 was 820,000; this dropped to less than 500,000 in 2010.

At last week's meeting, GSK presented data that showed a steep drop-off in prescriptions in May 2011 when the restricted distribution system became fully operational.

According to the pharma, the number of patients taking Avandia dropped 97% to 3,405 from 117,349 patients in 2010. The majority of patients taking the drug now were taking Avandia prior to the REMS and only 6% are new patients.

GSK reported Avandia sales of £6 million ($9 million) for 2012, compared to £123 million ($187 million) in 2011 and £440 million ($669 million) in 2010.

COMPANIES AND INSTITUTIONS MENTIONED

Brigham and Women's Hospital, Boston, Mass.

Cedars Sinai Heart Institute, Los Angeles, Calif.

Centers for Disease Control and Prevention (CDC), Atlanta, Ga.

Cleveland Clinic, Cleveland, Ohio

David Geffen School of Medicine at UCLA, Los Angeles, Calif.

Duke Clinical Research Institute (DCRI), Durham, N.C.

European Medicines Agency (EMA), London, U.K.

GlaxoSmithKline plc (LSE:GSK; NYSE:GSK), London, U.K.

Hamilton Health Sciences, Hamilton, Ontario

Harvard Medical School, Cambridge, Mass.

McMaster University, Hamilton, Ontario

MD Anderson Cancer Center, Houston, Texas

National Institute of Allergy and Infectious Diseases (NIAID), Bethesda, Md.

Takeda Pharmaceutical Co. Ltd. (Tokyo:4502), Osaka, Japan

Tufts Medical Center, Boston, Mass.

U.S. Food and Drug Administration (FDA), Silver Spring, Md.