Frequency of drug-induced valvular heart disease in patients previously exposd to benfluorex: a multicentre prospective study

Professor Cooper and colleagues argue that the control group we used for computing the aetiologic fraction of valvular heart disease (VHD) associated with benfluorex use was not valid because control subject had to be unexposed to any drug known to cause VHD, and with no prior history of VHD. Cooper et al miss an essential methodological point: the two studies on ultrasonography evaluation of VHD performed by C Tribouilloy et al (2013) included four groups of patients. A first group of 376 benfluorex-treated diabetic patients referred by primary care physicians for echocardiography and a group of 376 diabetic patients never exposed to benfluorex.1 The third group included 423 non diabetic patients previously exposed to benfluorex and the fourth group included 412 diabetic patients previously exposed to benfluorex.2 In both studies and in all four groups, Tribouilloy et al (2013) took great care to avoid inclusion of patients who had been treated by other drug known to induce VHD or who had a history of VHD. Therefore, the four groups were similar in terms of VHD history and absence of use of drugs other than benfluorex susceptible to induce VHD. In this respect the group of 376 diabetic patients never exposed to benfluorex was a suitable control group for the two groups of benfluorex-exposed diabetic patients, enabling computation that the risk of VHD associated with benfluorex exposure is 22.7: thus that in patients with VHD who used benfluorex, there was a 95.6% chance that benfluorex use was the cause of the VHD. The argument on past exposure to other drugs susceptible to induce VHD is specious since past use of ergot derivatives (ergotamine and methysergide), and drugs used in the treatment of Parkinson (eg, pergolide and cabergoline) by French patients who took benfluorex is rare, in the order of 0.8%.2

The sentence "benfluorex use was uncommon in the population [4.1% of diabetics in France (3)], the proportion of DI-VHD attributable to its use at the population level will be considerably lower" suggests that Cooper et al have a limited understanding of what an attributable risk exactly means. The attributable risk represents the probability that among subjects with VHD who took benfluorex, the drug was the likely cause of these lesions. Thus the attributable risk applies to patients having at the same time a history of benfluorex use and a diagnosis of VHD, and therefore, the attributable risk does not depend neither on the prevalence of VHD nor on frequency of benfluorex use, but just on the risk of VHD associated with benfluorex use.

Cooper and colleagues raise the issue of the possible influence of confounders. An inspection of Table 1 of the 2012 article1 and of Table 5 of the 2013 article2 shows that age and sex distribution, body mass index, and smoking status were remarkably similar in the four groups. Moreover, the possibility that confounding could have an association with both benfluorex use and VHD strong enough to substantially affect the odds ratio of 22.7 we found is vanishingly small. Thus, worries of Cooper et al about confounding factors are ungrounded.

Attributable risk computations similar to ours were already published for VHD associated with exposure to fenfluramine in the USA. A meta- analysis of epidemiological data based on height controlled prevalence studies corrected for the background burden of valve regurgitation in general populations concluded that compared to subjects who never used fenfluramine, risks of mild or greater regurgitation of subjects who took the drug were increased 19.6-fold for the aortic valve and 5.9-fold for the mitral valve.3 Using these relative risks, attributable fractions were calculated: among subjects with aortic or mitral regurgitation who took fenfluramine, the drug was the likely cause of [19*6-1*0]/19*6 = 95% of aortic regurgitations and of [5*9-1*0]/5*9 = 83% of mitral regurgitations.3

Fenfluramine and benfluorex use has led to increased risks of cardiac adverse events of a magnitude rarely encountered in pharmacoepidemiology.4,5 Because of this magnitude of risk, it is entirely logical that the majority of VHD among patients who took benfluorex has been caused by the drug. Therefore, it is illusory to try to finding out the rare patients with VHD who took benfluorex whose cardiac lesions would not be due to benfluorex use.

1. Tribouilloy C, Mar?chaux S, Jobic Y, Jeu A, Ederhy S, Donal E, R?ant P, Arnalsteen E, Boulanger J, Garban T, Ennezat PV, Andr?jak M, Rusinaru D. Frequency of drug-induced valvular heart disease in patients previously exposed to benfluorex: a multicentre prospective study. Eur Heart J. 2013 Sep 6. [Epub ahead of print]

2. Christophe Tribouilloy, Sylvestre Marechaux, Yannick Jobic, Antoine Jeu, Stephane Ederhy, Erwan Donal, Patricia Reant, Elise Arnalsteen, Jacques Boulanger, Thierry Garban, Pierre-Vladimir Ennezat, Michel Andrejak, and Dan Rusinaru. Frequency of drug-induced valvular heart disease in patients previously exposd to benfluorex: a multicentre prospective study. Eur Heart J (2013) 34 (46): 3580-3587 first published online September 6, 2013.

3. Hopkins PN, Polukoff GI. Risk of valvular heart disease associated with use of fenfluramine. BMC Cardiovascular Disorders 2003; 3: 5.

4. Nissen SE, Wolski K. Rosiglitazone revisited: an updated meta- analysis of risk for myocardial infarction and cardiovascular mortality. Arch Intern Med 2010; 170: 1191-201.

5. McGettigan P, Henry D. Cardiovascular risk with non-steroidal anti -inflammatory drugs: systematic review of population-based controlled observational studies. PLoS Med 2011; 8: e1001098.

Conflict of Interest:

None declared

Autier et al. (1) attempt to use the data of Tribouilloy et al. (2) to estimate the proportion of valvular heart disease (VHD) attributable to benfluorex use amongst diabetic patients. The resulting estimated aetologic fraction of 95.6% is certainly striking, and whilst the calculations appear correct, it is clear that this very high figure is a result of a misinterpretation of the underlying data.

In their study, Tribouilloy et al. (2) assessed, using echocardiography (reviewed blind by two experienced observers), a cohort of patients (diabetic and non-diabetic) who had been exposed to benfluorex for three months or more, and documented the prevalence of drug induced- valvular heart disease (DI-VHD). In both this cohort, and the control group (diabetics with no previous exposure to benfluorex or history of VHD), prior exposure to other drugs associated with a DI-VHD was excluded. Thus the prevalence of DI-VHD in the control group, by definition, should be zero, and the single miss-classified DI-VHD case out of 376 participants is consistent with a reliable echocardiography-based identification of this diagnosis.

The control group, lacking the potential for inclusion of patients with DI-VHD, is therefore a markedly different population, in terms of risk, to that of the benfluorex exposed group and is completely unsuited to the calculation of the proportion of valvular heart disease attributable to benfluorex. Clearly, when a population exposed to benfluorex is compared with a population unexposed to any drug known to cause DI-VHD, and with no prior history of VHD, the aetiologic fraction of DI-VHD due to benfluorex has to be extremely high. As benfluorex use was uncommon in the population [4.1% of diabetics in France (3)], the proportion of DI-VHD attributable to its use at the population level will be considerably lower, but cannot be derived from these data.

However, what Autier et al. actually purport to present is the aetiologic fraction for VHD (i.e. all valvular heart disease, not just that associated with prior drug use), thus conflating these two entities, one of which (DI-VHD) is a modest subset of the other (VHD). This dataset is obviously not suited to the estimation of the aetologic fraction of VHD attributable to benfluorex, and additional considerations, including the lack of adjustment for any confounding factors, lack of matching between benfluorex and control groups, and lack of consideration of any other causes of VHD (such as degenerative disease, cardiomyopathy, ischaemic heart disease and infection) further undermine any attempts to derive such a figure in this way. Indeed, Tribouilloy et al. quite appropriately do not use the control group in any statistical sense, but merely to validate the reliability of DI-VHD diagnosis in the benfluorex-exposed group, as they indicate in their own response to these calculations. In conclusion, Autier et al.'s statement that "trying to identify the minority of patients in which VHD would not be due to benfluorex use is illusory" (1) cannot in any way be justified by their calculations, or indeed represent a meaningful interpretation of currently available published data (2).

1) Philippe Autier, Mathiu Boniol, Peter Boyle. Estimation of amounts of valvular heart diseases attributable to benfluorex. Eur Heart J published online November 8, 2013

2) Christophe Tribouilloy, Sylvestre Marechaux, Yannick Jobic, Antoine Jeu, Stephane Ederhy, Erwan Donal, Patricia Reant, Elise Arnalsteen, Jacques Boulanger, Thierry Garban, Pierre-Vladimir Ennezat, Michel Andrejak, and Dan Rusinaru. Frequency of drug-induced valvular heart disease in patients previously exposd to benfluorex: a multicentre prospective study. Eur Heart J (2013) 34 (46): 3580-3587 first published online September 6, 2013

3) Weill A, Paita M, Tuppin P, et al. Benfluorex and valvular heart disease: a cohort study of a million people with diabetes mellitus. Pharmacoepidemiol Drug Saf 2010;19:1256-62.

Conflict of Interest:

CC has received honoraria/ research funding from P and G, Alliance for Better Bone Health, Sanofi-Aventis, Schering Plough, MSD, Eli Lilly, Shire, Servier and Amgen. AJ has received honoraria/ research funding from Anthera, Servier and Roche. NH has received honoraria/ research funding from P and G, Alliance for Better Bone Health, Sanofi-Aventis, Schering Plough, MSD, Eli Lilly, Shire, Servier and Amgen.

We gratefully thank Dr Autier and colleagues for this interesting approach. Indeed, only one (0.26%) diabetic patient not exposed to benfluorex was classified as DI-VHD (+) in our study based on a blind reading of echocardiograms compared with 23 patients (5.6%) in the group of diabetic patients exposed to benfluorex; hence, these data indicate that this echocardiographic aspect is highly specific of exposure to benfluorex. Therefore one excepts the aetiologic fraction (AF) of VHD due to benfluorex in the present cohort to be very high as calculated by Dr Autier and colleagues. In addition, the agreement between readers for the diagnosis of DI-VHD was very good using this approach (Kappa value 0.83), indicating the robustness of this echocardiographic classification.

Conflict of Interest:

None declared

In their controlled study on 835 patients who took benfluorex (BF, Mediator ?), C Tribouilloy and Colleagues show the influence of long-term use of this drug on valvular heart disease (VHD) ¹ . The study has several merits like the blinding of ultrasound readers who were unaware of past BF use of patients and a standardized protocol for ultrasound examination. However the availability of a control group of diabetic patients who never took BF could have allowed more informative statistical analysis. More specifically, it is possible to estimate the probability that among patients with VHD who used BF, the VHD would be due to BF use.? This statistical quantity is called the aetiologic fraction (AF) or "attributable fraction percents" ^2,3 . Calculation of an AF can be done from knowledge of the risk of VHD in diabetic patients who used BF (n=412) as compared to diabetic patients who never used BF (n=376). To this end we re-organized data provided in Tribouilloy et al ¹ so that one can compare past use of BF among patients with and without VHD (Table 1). The risk of VHD associated with BF use can be calculated as an odds ratio (OR), i.e., [(23*366)/(1*371)] = 22.7, meaning that BF use is associated with a 22.7 increased risk of VHD. The AF can be calculated as [(22.7-1)/22.7] = 0.956, meaning that among patients with VHD who took BF there are 95.6% chance that VHD was actually due to BF use. If the lower bound of the 95% confidence interval is considered, then the AF is 70%. These computed odds ratios are not adjusted for confounders like smoking or gender. However, the magnitude of the association is such that adjustments are unlikely to materially change the fact that the vast majority of VHD are associated with BF use. Moreover, diabetes caries an increased risk of cardiovascular diseases in general, so the AF we calculated may in fact be underestimated.

The aetiologic fraction is a way to quantify causality between an exposure and an outcome, and this parameter provides an important link between causality and public health action ⁴ . When an etiologic fraction is high (>95% in this case), trying to identify the minority of patients in which VHD would not be due to exposure BF use is illusory ³ .

Fenfluramine is an appetite suppressor that has a metabolic activity and cardiac toxicity similar to that of BF ⁵ . In the United States procedure, the causal link between fenfluramine use and cardiovascular lesions was assumed for all patients presenting any degree of valve lesion who had taken the drug, since the attributable risks derived from epidemiological studies were so high that the presence of valve lesions and history of fenfluramine use had little chance to be a simple coincidence. In this respect, by 1999, procedures have granted compensation to anyone with valve lesions who could demonstrate having ever taken the drug ⁶ .

The procedure in France is more complicated. Compensation claims of BF victims are based on a complex expert judgment that attempts to establish whether for each particular patient, a cause-effect relationship would exist between BF use and VHD. This judgment is based on the clinical and ultrasonography information specific to that patient and ignores the totality of epidemiological evidence collected a large number of similar situations. However, in view of the high AF, these complex procedures are scientifically questionable and are prone to errors.

Conflict of interest: none to declare.

References

1. Tribouilloy C, Marechaux S, Jobic Y, Jeu A, Ederhy S, Donal E, Reant P, Arnalsteen E, Boulanger J, Garban T, Ennezat PV, Andrejak M, Rusinaru D. Frequency of drug-induced valvular heart disease in patients previously exposed to benfluorex: a multicentre prospective study. Eur Heart J 2013 Sep 6. [Epub ahead of print]

2. Cole P, MacMahon B. Attributable risk percent in case-control studies. Brit J Prev Soc Med 1971; 25 :242-244.

3. Rothman KJ, Greenland S. Measures of effect and measures of association. In: Rothman KJ, Greenland S, eds. Modern Epidemiology , 2d ed. New-York, NY: Lippincott Williams and Wilkins; 1998. P 53-55.

4. Northridge M. Public health methods: attributable risk as a link between causality and public health action. Am J Public Health 1995; 85 :1202-1203.

5. Roth BL. Drugs and valvular heart disease. N Engl J Med 2007; 356 :6-9.

6. SoRelle R. Diet drug maker agrees to 3.75 billion settlement. Circulation 1999; 100 :e133-134.

Conflict of Interest:

None declared