Delayed drug-related harms: deceptive and potentially deadly

Introduction

Drug regulation acquired high prominence in Europe as a result of the thalidomide catastrophe. In the 1950s barbiturates were commonly used as hypnotics, and because of their narrow margin of safety due to respiratory and cardiovascular depression were predictably highly lethal when taken in overdose – a type A adverse effect in the classification of Rawlins and Thomson [1] consisting of an immediate harm the lethality of which is determined by a combination of toxicology, dose and individual susceptibility [2]. Thalidomide was believed to cause less cardio-respiratory depression than barbiturates at doses that induced sleep, and was marketed as “the safe hypnotic” when introduced in 1957. Its unexpected and pharmacologically unrelated effects on foetal development led to a delayed epidemic of phocomelia in newborns. Even with such a distinctive abnormality, instantly recognisable at birth and caused in almost 100% of pregnancies exposed during the critical phase of gestation, the connection with thalidomide use during early pregnancy was appreciated only in 1961. By the time the drug was withdrawn late in that year an estimated 10,000 babies had been born with this uniquely obvious defect.

Exposure of the foetus to diethylstilbestrol during pregnancy (in well-intentioned but misguided attempts to prevent miscarriage) has been linked to a rare malignancy during early adult life (adenocarcinoma of the vagina) in the daughters of treated women – ie perhaps 20 years after exposure to the drug in utero. This begs the question of whether if such treatment had caused an increased incidence of a common malignancy (such as breast or colon cancer), rather than one where the background incidence is so rare, a causal link could have been recognised. If a drug-related harm is disguised as a commonly occurring natural event it is extremely difficult to prove the causal link. Even where evidence from randomised controlled trials (RCTs) makes such a connection convincingly, in any individual case it will be difficult to attribute an adverse event to a drug rather than to previously unrecognised disease with medico-legal implications in attributing responsibility or deciding on possible compensation.

There are several topical examples of drug harms that are both delayed and unobvious. These include evidence of cardiovascular harm caused by rosiglitazone (despite an apparently favourable effect on haemoglobin A1c) that led the European Medicines Agency (EMA) to suspend the marketing authorisation for all rosiglitazone-containing medicines in the European Union (see http://www.gsk.com/media/pressreleases/2010/2010_pressrelease_10103.htm accessed Oct 10th 2010) and the evidence that whereas sibutramine favourably influences body weight its effects on stroke and cardiovascular disease outweigh its benefits (http://www.bmj.com/content/340/bmj.c477.extract accessed on Oct 10th 2010). Here I consider further [3] the example of cyclo-oxygenase inhibitors where serious delayed adverse effects in the form of myocardial infarction are plausible yet stealthy. Widespread appreciation of serious yet delayed adverse events that take the form of increased incidence of common diseases will no doubt influence future regulatory requirements in terms of requirements for appropriately powered clinical trials to establish risk benefit (as opposed to establishing an effect on a surrogate endpoint of uncertain robustness), but in the meanwhile prescribers are left in considerable uncertainty when advising their patients. The same kind of asymmetry (multiplying a large risk by a factor x causes more events than the reduction of events caused by dividing by x) applies to the arithmetic when other drugs that may further increase cardiovascular disease (already very common) are prescribed to large populations of patients at high risk.

Cyclo-oxygenase (COX) inhibitors and myocardial infarction: first do no harm

NSAIDs are the most widely prescribed drugs worldwide. Their primary pharmacological effect is to inhibit cyclo-oxygenase (COX) [4]. The spectrum of activity of different agents depends on dose and on selectivity for COX-1 or COX-2. COX-1-selective inhibitors inhibit platelet thromboxane A2 (TXA2) synthesis and are anti-thrombotic; they can also be anti-inflammatory and analgesic, but often cause gastrointestinal toxicity. Nonselective COX-inhibitors are anti-inflammatory, analgesic and also cause gastrointestinal toxicity. COX-2 selective NSAIDs, such as rofecoxib (“Vioxx”), developed to reduce gastrointestinal harm [5], all raise cardiovascular safety concerns including increased arterial blood pressure, increased atherogenesis, or increased thrombotic tendency. The precautionary principle to first do no harm applies to European pharmaceutical regulation [6], and it has been argued that this should apply to NSAIDs as a class in terms of prescribing and labelling, unless data from appropriate clinical trials provide sufficient reassurance for specific products [3].

Vioxx was considered safe by the FDA in 2004, but was withdrawn by Merck later that year because of concern over possible cardiovascular harm. In 2005 a Texan court awarded $250 million to the widow of Robert Ernst, who died whilst taking the drug (see: http://news.bbc.co.uk/1/hi/business/4168332.stm). Estimates of the relative risk (RR) of myocardial infarction while taking Vioxx vary; RCTs were powered to investigate gastro-protection and adenoma prevention in low cardiovascular risk patients rather than cardiovascular outcomes, so the confidence intervals are wide. Meta-analyses of such trials suggest a near doubling of myocardial infarction risk with Vioxx and other COX-2 selective inhibitors [7,8]. Such RCT data have been questioned: Curfman and colleagues [9] recalculated the VIGOR data [10] with previously omitted results to give a RR (95% CI) for myocardial infarction for Vioxx compared to naproxen of 5.00 (1.68-20.13). An alternative approach using nested controls showed substantial potential toxicity with an odds ratio for serious coronary heart disease for Vioxx ≤25 mg/day of 1.47 (0.99-2.17) and for >25 mg/day 3.58 (1.27-10.11) [11].

Some 80 million people (average age approximately 68 years) took Vioxx worldwide [11,12]. US actuarial life tables predict approximately 700,000 cardiovascular deaths in one year in 80 million persons aged 65-70. If rofecoxib increased cardiovascular death only to the estimate from a meta-analysis of all RCTs of five coxibs of 1 cardiovascular death per 1000 treated in 1 year [13], it could have caused an excess of approximately 80,000 deaths. This may be an underestiamate because the RCT data were derived from a younger population, and because the larger RCTs included in the meta-analysis excluded patients with known cardiovascular disease. An excess of 2-3 million Vioxx-related myocardial events in 80 million exposed persons with a background of 700,000 expected cardiovascular deaths cannot be excluded [3]. Compare this to the modest benefit of anti-hypertensive or lipid lowering agents, where halving the risk with drugs is unattainable in large trials in subjects at moderately increased cardiovascular risk. Halving cardiovascular risk in a population of this size and age might affect approximately 350,000 lives, so avoiding drugs such as Vioxx (including other NSAIDs) where possible may be an even higher priority than prescribing drugs to prevent cardiovascular disease! This asymmetry is surely worth bearing in mind by prescribing doctors who rightly pride themselves on detecting and treating cardiovascular risk factors.