Since the 1950s, the FDA has required drug companies to turn over all individual patient case reports from their studies. This permits reanalysis of how each case was coded3 and enabled Marciniak to scrutinise the RECORD trial data.3 The EMA had accepted the company’s findings that the risk of complications was the same, 14.5% for rosiglitazone and 14.4% for the comparator.3 However, when Marciniak studied 549 case reports he found many missing cases of cardiac problems that favoured rosiglitazone four to one.3,14 For one patient, there were 1438 pages, and for most of the other 4500 patients there were several hundred pages, making a review of all case reports a huge task.3 Marciniak concluded that the case report forms are essential for understanding a study and he found that rosiglitazone increased cardiovascular risk also in the RECORD trial,3 in contrast to Glaxo’s manipulated results.
Very importantly, Marciniak stated that ‘even with blinded adjudication, biased referral for adjudication of cases and data by unblinded investigators and site monitors may lead to biases in event rates’.14
The importance of this statement cannot be overestimated. The sponsor has access to the data and knows who received which drug, and biased selection of ‘unclear cases’ for review by an independent committee is an important reason why industry trials should be distrusted. (see also Chapter 5)
Grave suspicions were raised earlier. The editorial that accompanied the interim publication of the RECORD trial mentioned that the trial had found an exceptionally low event rate in a high-risk population of patients with diabetes and noted that the most likely explanation was incomplete ascertainment of events.15 The editorialists also noted that rosiglitazone increased the risk of a heart attack to the same degree as lipid-lowering statins lower the risk.
However, as always the FDA wanted it otherwise. According to the documents released by the Senate, a top official at the FDA, John Jenkins, director of the agency’s office of new drugs, preferred to continue to put patients at risk. He argued internally that rosiglitazone should remain on the market and briefed the company extensively on the agency’s internal debate. According to a sealed deposition, a top company official wrote after he spoke with Jenkins that ‘It is clear the office of new drugs is trying to find minimal language that will satisfy the office of drug safety’. In the deposition, Rosemary Johann-Liang, a former supervisor in the drug safety office who left the FDA after she was disciplined for recommending that rosiglitazone’s heart warnings be strengthened, said of Jenkins’ conversations with GlaxoSmithKline that ‘This should not happen’, and she suggested that ‘People have to make a determination about the leadership at the FDA’.
Rosiglitazone was suspended in Europe in September 2010 whereas the process at the FDA continued to be fishy. In July 2010, the FDA held a new advisory committee meeting to decide if the drug should remain on the market. This was 5 months after the damning Senate report, but that didn’t deter the higher-ups in the agency from more wrongdoing. In an unprecedented move, the FDA invited additional people to its meeting who had been involved in a similar 2007 meeting but were no longer active members of either committee.16 Most of these people had voted for keeping the drug on the market in 2007, and their addition to the 2010 meeting tipped the scale from voting for a withdrawal to voting for keeping it on the market, which was what the FDA decided.
The scandal rambled on. In 2009, Glaxo started the TIDE trial, scheduled to end in 2015.10 It is unethical, as it compares the cardiovascular safety of rosiglitazone and pioglitazone, although the company knew that rosiglitazone increases the risk of myocardial infarction compared to pioglitazone.10 Furthermore, the information given to patients being asked to volunteer for the trial was seriously misleading and therefore also unethical.17 Because US and European physicians were not willing to enrol patients, Glaxo exploited developing countries,2 but in 2010 India’s drug controller stopped the trial. Two FDA safety officers also suggested to stop the trial, as it was unethical and exploitative, and to take rosiglitazone off the market, as it causes 500 heart attacks and 300 cases of heart failure every month in the United States.11 Nothing was done initially, but later the FDA halted the trial.17
The same year, Glaxo had the nerve to say in a statement to the BMJ that the RECORD trial had shown its drug performed similarly as the comparators.3 Glaxo also said that a head-to-head trial would prove that rosiglitazone doesn’t increase the risk of myocardial infarction and that the evidence suggesting that it does was ‘not scientific’.18 Glaxo’s lies are not of this world.
In 2010, Steven Nissen published ‘The rise and fall of rosiglitazone’, an online editorial in the European Heart Journal. Glaxo’s head of research and development, Moncef Slaoui, wrote to the journal that Nissen’s editorial was ‘rife with inaccurate representations and speculation that fall well outside the realm of accepted scientific debate. We strongly disagree with several key points within the editorial, most importantly those which imply misconduct on the part of GSK.’19 Slaoui asked the journal to withdraw the editorial from its website and not to print it in the journal’s hardcopy edition ‘until the journal has investigated these inaccuracies and unsubstantiated allegations’. When the journal didn’t give in but published the editorial in print, Slaoui said that there was ‘absolutely no attempt to suppress’ the editorial. Glaxo called Nissen’s meta-analysis a hypothesis that had not been confirmed by more recent and considerably more robust evidence from prospective, long-term cardiovascular outcomes studies.20 Absolute bullshit. A meta-analysis of the randomised trials is the most reliable evidence we have and it is not a hypothesis; it provides definitive proof. Glaxo also remarked that ‘The American Heart Association and the American College of Cardiology Foundation had said that “insufficient data exist to support the choice of pioglitazone over rosiglitazone”.’ If that is true, it only shows how corrupt these organisations are. They should be the most concerned when a drug causes heart attacks.
So what did the FDA do when it didn’t want to lower the number of deaths among diabetes patients by taking the drug off the market as in Europe? It issued meaningless warnings, the standard fake fix.21 It stated that rosiglitazone should only be used in patients already being treated with the drug, and in those patients whose blood sugar cannot be controlled with other drugs and who, after consulting with their healthcare professional, do not wish to use pioglitazone.
Can you see what’s wrong with this advice? At least four things. First, why on earth should a patient continue with a harmful drug only because the patient is already on the drug? I think the patients would prefer a less harmful drug, as you never know when a myocardial infarction strikes.
Second, we don’t use drugs to control blood sugar but to lower the risk of complications to diabetes such as cardiovascular events. So, do get off the drug immediately, no matter what the FDA says!
Third, as the endocrinologists thought it was a wonder drug, it might not be a good idea for the patient to consult ‘their healthcare professional’. In fact, it has been shown that doctors who take money from manufacturers of rosiglitazone were substantially more prone to recommend the drug than other doctors, even after the FDA had warned about its cardiovascular harms.22
Fourth, what plausible reason could there be that a patient would not want to use pioglitazone when that drug seems to be safer (see below)?
The FDA’s stubbornness is a considerable threat to public health. By 2009, even the heavily industry-supported endocrinologists (see Chapter 8) had woken up and a consensus group of the US and the European diabetes associations unanimously advised against using rosiglitazone.2
These events are so bizarre that they raise uncomfortable questions. Did someone higher up in the FDA hierarchy receive a load of money from Glaxo at some secret bank account or in a suitcase that left no trails? Considering the enormous sales of rosiglitazone, even $100 million in bribes would be peanuts. I am not saying this happened, but if not, what could then be the explanation for this series of implausible events? Future rewards
?
The oddities don’t even stop there. The risk of myocardial infarction with rosiglitazone seems to be increased by about 80%, and in 2010, the FDA decided that trials of diabetes drugs should show that the risk of cardiovascular events is clearly less than 80%.23 To allow this degree of permitted risk is incredible, particularly since we use diabetes drugs to decrease the cardiovascular risk, certainly not to allow a certain increase.
The asymmetry and lack of consistency in regulatory decision making is dangerous for the patients. In 2007, there was almost unanimous agreement in the FDA advisory committee that rosiglitazone increases cardiovascular risk, but the committee nevertheless recommended the drug should stay on the market. If there had been almost unanimous agreement about the harms when the drug was first submitted for marketing approval, it would hardly have been approved.24
Assertions that a drug agency considers that a drug’s benefits outweigh its harms, which we hear all the time when troubles accumulate, also for rosiglitazone in 2007,24 are unhelpful. It’s not easy to compare benefits and harms, as they aren’t measured on the same scale, and it’s never made explicit how agencies arrive at gracious conclusions, which – more than anything else – seem to be convenience statements aimed at getting the agency off the hook and avoid disturbing their industry friends and their powerful allies among the politicians.
The FDA’s meaningless warnings about rosiglitazone are typical. If you analyse the text in package inserts, you’ll see how illogical and other-worldly it often is. For many years, I joked about the general warning that a drug should be used with caution in pregnancy. How should this be done? Either you use a drug or you don’t. I have kept a 1998 Janssen-Cilag package insert from the time when my children suffered repeatedly from pinworms and the whole family needed treatment. It says that the use of mebendazole (Vermox) during pregnancy and breast feeding should always occur in consultation with the doctor because there is no experience with the use of the drug under these conditions. Great advice. What exactly is the doctor supposed to do? In this case, the doctor was me or my wife, as we are both doctors. She wasn’t pregnant but if she had been, we would have preferred to live with anal itching rather than running an unknown risk of giving birth to a malformed baby.
Pioglitazone causes heart failure but is still on the market, as it is believed to be safer than rosiglitazone.11 However, serious questions about trial conduct have been raised also for this drug. A large trial, the PROactive study of 5238 patients comparing pioglitazone with placebo, failed to find a significant benefit (P = 0.10) for its primary outcome, which was a composite endpoint of various adverse cardiovascular events.25 This was the true result. The drug didn’t work. The trial protocol had been published and it stated that this outcome was chosen because the aim of the study was to evaluate the overall effects on macrovascular disease.26 However, when the trial was published in the Lancet, there was an additional composite outcome, which consisted of patients who died or had a non-fatal heart attack or stroke, and for which P was 0.03. This was called the main secondary endpoint, although it didn’t exist in the protocol.
Several observers commented on the discrepancy, and the authors, which included two people from the sponsors, Eli Lilly and Takeda, defended themselves by saying that the new composite outcome was introduced in the final statistical analysis plan, which was released in May 2005 and sent to the FDA.27 They also said that it’s legitimate to change outcomes during a study’s conduct provided it’s agreed ‘before any knowledge of unblinded data by the trialists’. Finally, they stated that ‘The PROactive Executive Committee was not aware of any results of the study before the official unblinding of the study on May 25, 2005.’
It is important to be the devil’s advocate here, as we know we cannot trust drug companies. The final visits for all patients were completed in January 2005, 4 months before the analysis plan was changed and a new outcome was invented. Both companies were represented at the steering committee and its executive committee. Furthermore, the statements in the authors’ defence were carefully worded, as if they had been cleared with lawyers. Could a company statistician have peeped at the data behind the academic investigators’ back before the final analysis plan was ‘suggested’ to them?
Such a scenario isn’t speculative. As noted in Chapter 5, we analysed 44 protocols for industry-sponsored trials and found it was stated explicitly in 16 cases that the sponsor had access to accumulating data while a trial was running.28 Who knows in how many other cases the sponsor had access to the data but was smart enough not to write this in the protocol? It reflects poor trial conduct and isn’t something the companies want to tell the world about, as it was only mentioned in one of the 44 publications.
If that were the case for the PROactive study, all of the statements in the Lancet letter might nonetheless have been technically correct. The trialists might not have been unblinded and the executive committee might not have known about the results. But the company statistician likely knew about the results because the trial had a Data and Safety Monitoring Board, whose job it is to warn about excessive harms that might emerge while the trial is running.
For obvious reasons, we should be deeply sceptical towards companies finalising statistical analysis plans after a lot of data have arrived. The incentive to cheat is huge and, as noted earlier, the difference between an honest data analysis and a less honest data analysis can be worth billions on the world market. It shouldn’t surprise anybody that cheating is exceedingly common, but until recently, it was difficult to prove, as trial protocols were regarded as confidential. We succeeded to get access to a cohort of protocols submitted to a research ethics committee in Copenhagen that allowed us to study the extent of cheating with the predeclared outcomes.29 We identified 102 protocols, which included both industry-funded (about three-quarters) and non-industry-funded trials that had all been published. To our great surprise, at least one protocol-defined primary outcome had been changed in 63% of the trials. And in 33% of the trials, a new primary outcome was introduced in the published report that didn’t exist in the protocol. Here comes the worst part:
Not a single publication acknowledged that primary outcomes had been changed!
The reason this is so devastating for the trustworthiness of trials is that there are often many outcomes, which may be further divided or combined, creating even more chances of hitting the bull’s eye. Imagine you fire a gun towards many targets that are partly overlapping. Even if you are a poor shot, there is a good chance you’ll hit near the centre of one of the targets. If you want to cheat, you’ll say that the target you hit was also the one you aimed at. Even better, you may wipe out some or all of the other targets before you invite the audience in to see how good a shot you are. Wiping out other targets corresponds to not mentioning outcomes stated in your protocol, another common practice in clinical trials. We found that 71% of the trials had at least one unreported outcome, and in these trials, a median of four efficacy and three harms outcomes were missing in the publications.29
We have published other revealing papers based on our cohort of trial protocols. For example, we found unacknowledged discrepancies between protocols and publications for sample size calculations (18/34 trials), methods of handling protocol deviations (19/43), missing data (39/49), primary outcome analyses (25/42), subgroup analyses (25/25) and adjusted analyses (23/28).30 Interim analyses were described in 13 protocols but mentioned in only five corresponding publications.
It is clear that trial reports cannot be trusted and that we need to have access to the full protocols and the raw data. The EMA agrees. The rosiglitazone scandal made the EMA’s new director Guido Rasi say in 2012 that the agency needs to analyse the raw data rather than accepting aggregated information submitted by drug companies seeking approval.31
Speaking of statistics, there is another issue with the PROactive trial that smells. The trial report mentions 14 cases of bladder cancer on drug and 6 on placebo. This difference wasn’t st
atistically significant (P = 0.07) and could therefore be explained away by the company’s salespeople.32 However, 4 years later it was revealed that one of the cases in the placebo group was benign, and 14 versus 5 is statistically significant (P = 0.04). The reason this smells is that such ‘errors’ always favour the company that controls data analysis and the writing of the report.
A final point that the glitazones illustrate so nicely is that we cannot rely on surrogate outcomes. Rosiglitazone and pioglitazone reduce glucose to the same degree and both increase the risk of heart failure. However, while rosiglitazone definitely increases cardiovascular events, the overall effect of pioglitazone is more uncertain.24,25,33 In 2011, four members of the EMA committee dealing with an application for generic pioglitazone gave a divergent statement: ‘It appears impossible to define a subpopulation of diabetic patients where the benefits of pioglitazone would outweigh its risks.’33
Sometimes, researchers declare they have validated a surrogate marker. Don’t believe them, as it cannot be done. All drugs have many effects, and we cannot pick just one of them and say that this effect will tell us what we need to know. For example, both rosiglitazone and pioglitazone increase body weight and fractures, and rosiglitazone has an adverse effect on LDL cholesterol, none of which are related to their effect on glucose.15,25 In the PROactive study, pioglitazone increased body weight by 4 kg compared to placebo, which isn’t a beneficial effect for patients with diabetes.25 It was also worrying that for every 62 patients treated with pioglitazone, one additional patient was admitted to hospital with heart failure, which is a serious condition. In 2011, the FDA warned that pioglitazone ‘may be associated with an increased risk of bladder cancer’.34 There it is again: may be associated with an increased risk. Three wool-in-mouth terms in just seven words. Drug agencies just won’t acknowledge the harms of drugs they have approved. Pioglitazone more than doubles the incidence of bladder cancer and was withdrawn for this reason in France in 2001.35 When I drink whiskey or have sex, I cannot say it may be associated with an increased chance of well-being. It feels good.
Deadly Medicines and Organised Crime Page 32