Correspondence
Appetite-Suppressant Drugs and Valvular Heart Disease
N Engl J Med 1999; 340:476-480February 11, 1999
- Article
To the Editor:
In his editorial1 accompanying the reports on appetite-suppressant drugs and valvular heart disease, by Jick, Khan, and Weissman and their colleagues (Sept. 10 issue),2-4 Dr. Devereux converted the “risks” reported by Jick et al.2 (e.g., 7.1 per 10,000 persons) to “rates” (e.g., 0.14 per 1000 patient-years), by dividing by 50 — (10,000 patients × 5 years) ÷ 1000. The use of patient-years assumes that Jick et al. knew the actual date of onset of valvulopathy, which they did not. Without knowledge of the date of onset, the use of patient-year rates assumes that all cases of valvulopathy occurred on the last day of follow-up. Why were these conversions made?
On the basis of the data reported by Jick et al., Dr. Devereux states that the prevalence at follow-up in patients who had taken dexfenfluramine or fenfluramine was only 0.12 percent higher than the expected prevalence. Dr. Devereux did not report the expected prevalence or how it was calculated. Under the null hypothesis that exposure is unrelated to valvulopathy, one can divide the 11 incident cases of valvulopathy by 19,046 (the total number of exposed and unexposed persons in the cohort) to obtain an expected risk (rate) of 5.8 per 10,000 persons (rate, 0.12 per 1000 person-years). However, we do not see how this expected risk (rate) is 0.12 percent lower than either of the risks (rates) reported by Jick et al.: risk with less than four months of exposure, 7.1 per 10,000 (rate, 0.14 per 1000 patient-years); risk with four or more months of exposure, 35.0 per 10,000 (rate, 0.70 per 1000 patient-years).
In the last paragraph of Dr. Devereux's editorial, he states that in patients meeting the criteria of the Food and Drug Administration (FDA) for valvulopathy, there is a possibility (ranging from 5 to 67 percent) that any abnormality found is a naturally occurring phenomenon, not a consequence of the use of appetite-suppressant drugs. However, the reports by Khan et al.3 and Weissman et al.4 indicate that the range of 5 to 67 percent represents the prevalence of any degree of regurgitation in unexposed patients (including trace or physiologic aortic regurgitation and trace or physiologic or mild mitral regurgitation), but that the prevalence of disease meeting the FDA criteria in unexposed patients was 1.3 percent in the study by Khan et al.3 and 4.5 percent in the study by Weissman et al.4 Therefore, our interpretation is that in exposed patients who meet the FDA criteria, there is a low probability (ranging from 1.3 to 4.5 percent) that the abnormality is a naturally occurring phenomenon and not a consequence of drug use.
4 ReferencesDavid F. Williamson, Ph.D.
Centers for Disease Control and Prevention, Atlanta, GA 30341-3724Susan Z. Yanovski, M.D.
National Institutes of Health, Bethesda, MD 208921
Devereux RB . Appetite suppressants and valvular heart disease. N Engl J Med 1998;339:765-767
Full Text | Web of Science | Medline2
Jick H ,Vasilakis C ,Weinrauch LA ,Meier CR ,Jick SS ,Derby LE . A population-based study of appetite-suppressant drugs and the risk of cardiac-valve regurgitation. N Engl J Med 1998;339:719-724
Full Text | Web of Science | Medline3
Khan MA ,Herzog CA ,St Peter JV , et al. The prevalence of cardiac valvular insufficiency assessed by transthoracic echocardiography in obese patients treated with appetite-suppressant drugs. N Engl J Med 1998;339:713-718
Full Text | Web of Science | Medline4
Weissman NJ ,Tighe JF Jr ,Gottdiener JS ,Gwynne JT . An assessment of heart-valve abnormalities in obese patients taking dexfenfluramine, sustained-release dexfenfluramine, or placebo. N Engl J Med 1998;339:725-732
Full Text | Web of Science | Medline
To the Editor:
The study by Weissman et al. of heart-valve abnormalities in obese patients receiving dexfenfluramine or placebo contains a conspicuous methodologic defect. The control group excluded only persons who had been taking dexfenfluramine during the six months preceding the study. This criterion obviously would not have excluded persons who might have been taking dexfenfluramine, fenfluramine, or combination therapy with phentermine before this six-month period.
Although there is some anecdotal evidence of the regression of valvular lesions among persons previously treated with these medications,1 it cannot be assumed that a six-month washout period would result in a normalization of valvular morphology. Thus, the methodologic defect in the study by Weissman et al. probably accounts for the significantly higher incidence of valvular heart disease in their control group (4.5 percent) than in the control group in the study by Khan et al. of the prevalence of cardiac valvular insufficiency in obese patients treated with appetite-suppressant drugs. In the latter control group (consisting of persons who had never received dexfenfluramine, fenfluramine, or combinations with phentermine), the incidence of valvular heart disease was 1.3 percent — less than one third the incidence in the study by Weissman et al.
If this lower incidence of valvular heart disease in the general population were used in the statistical analysis by Weissman et al., their results — and conclusions — would probably be quite different.
1 ReferencesMichael D. Myers, M.D.
4216 Katella Ave., Los Alamitos, CA 907201
Devereux RB . Appetite suppressants and valvular heart disease. N Engl J Med 1998;339:765-767
Full Text | Web of Science | Medline
To the Editor:
The reports by Khan, Jick, and Weissman and their colleagues and the editorial by Devereux suggest a relation between use of anorexigens and cardiac valvulopathy. Both Khan et al. and Jick et al. report strong associations between anorexigens and valvulopathy, whereas Weissman et al. report only a small and statistically nonsignificant difference. The issue is unsettled because of the enormous variation in the estimated relative risks among the three studies (range, 1.5 to 20.0). However, the three factors of dose response, study design, and statistical power may explain this apparent discrepancy and wide range.
The weaker association in the study by Weissman et al. is consistent with a dose–response relation between anorexigens and valvulopathy. The mean exposure was less than 3 months in the study by Weissman et al., as compared with 4 to 20 months in the study by Khan et al. and 0 to 10 or more months in the study by Jick et al. A similar duration-threshold relation between anorexigenic agents and pulmonary hypertension has been reported.1 Since the exposure levels in the studies by Khan et al. and Jick et al. reflect the actual patterns of use of these anorexigens, the two studies are more representative of actual experience.
The association identified by Weissman et al. was attenuated by the redirection of a clinical trial designed to answer a different question. Because the study subjects were not screened for valvular heart disease at base line, the resulting prevalence ratio included not only the incident cases occurring after exposure but also the prevalent cases at base line. Since the population prevalence is not trivial, perhaps 1 to 2 percent, the ratio reported by Weissman et al. is substantially biased in favor of the null hypothesis.
In interpreting the results of the study by Weissman et al., we must be certain that important differences in the prevalence of valvulopathy in the study population would have been detectable. The confidence interval (in Table 2 of their article) for the occurrence of either aortic or mitral valve disease in the dexfenfluramine group is wide (0.7 to 2.9). The confidence interval is consistent with almost a tripling of the risk of valvulopathy, offering little reassurance of safety. In fact, the power of the study by Weissman et al. to detect an odds ratio of 2.0 was only 25 percent. Since the study was not designed to evaluate the relation between dexfenfluramine and valvulopathy and since it was underpowered, the results are neither reassuring nor negative — merely inconclusive.
Finally, the reverse-logic presentation of attributable risk in the editorial by Devereux should be reconsidered. A more straightforward representation reveals that between 33 percent and 95 percent of cases in exposed persons are due to anorexigenic exposure if the relative risks reported by these three groups of investigators in fact reflect a causal relation.
1 ReferencesLemuel A. Moyé, M.D., Ph.D.
John F. Annegers, Ph.D.
University of Texas School of Public Health, Houston, TX 770301
Abenhaim L ,Moride Y ,Brenot F , et al. Appetite-suppressant drugs and the risk of primary pulmonary hypertension. N Engl J Med 1996;335:609-616
Full Text | Web of Science | Medline
To the Editor:
Weissman et al. actually studied the effects of two variables on the study groups: appetite suppressants and weight loss. Before one can attribute the higher incidence of even mild cardiac valvular anomalies to the medications in question, it is necessary to know the effect of weight loss in obese persons on cardiac performance in general. It is quite likely that a large decline in body mass will be accompanied by a substantial change in both cardiac mass and composition. An initial look at the data to see whether the most severe problems developed in the persons who lost the most weight would be informative. Also, if possible, it would be important to distinguish between those who lost mostly body fat and those who lost mostly body muscle. Finally, Cannistra and Cannistra's letter to the editor about the regression of multivalvular regurgitation after cessation of medication (Sept. 10 issue)1 was fascinating. Reportedly, all signs of regurgitation disappeared over a period of two years after use of the medication was stopped. There is no mention, however, of whether the patient regained the weight that had previously been lost.
1 ReferencesChristine F. Paoletti, M.D.
1245 16th St., Santa Monica, CA 904041
Cannistra LB ,Cannistra AJ . Regression of multivalvular regurgitation after the cessation of fenfluramine and phentermine treatment. N Engl J Med 1998;339:771-771
Full Text | Web of Science | Medline
To the Editor:
Bias has not been ruled out as a possible explanation of the association between fenfluramine or dexfenfluramine and cardiac valvular insufficiency reported by Khan et al. and Jick et al.
Khan et al.1 report that among 496 study subjects, the odds ratio for the association between appetite suppressants and cardiac-valve abnormalities was 21.4 (95 percent confidence interval, 6.8 to 108.2); the numerators for this estimate were not given. The study included 60 previously described patients2 who were known at the time of echocardiography to have used fenfluramine (47 patients) or dexfenfluramine (20) (total, 67 — the discrepancy1,2 is unexplained). On the basis of extrapolation from a graph,2 more than 35 percent of these patients had aortic or mitral insufficiency. Without the numerators, it is not possible to determine exactly how many persons were studied subsequently without knowledge of their exposure status. However, from the data that Khan et al. provide in Table 5 of their article,1 it can readily be estimated that the prevalence of valvular lesions in the latter group was less than 20 percent; that is, once the observers were unaware of exposure status, the number of cases of echocardiographically diagnosed valvular lesions among exposed persons declined significantly (P<0.02). This decline suggests that the data were biased. The selection of controls probably augmented the bias, since media advertisements would almost certainly have resulted in the selective recruitment of healthy persons.
Jick et al. completed their follow-up study in the United Kingdom before the 1996 publication of an association between appetite suppressants and pulmonary hypertension. The authors claim that their data were therefore not biased by awareness of any cardiovascular risk. However, starting in 1981, case reports in British journals, including the reported recurrence of pulmonary hypertension on rechallenge,3 repeatedly documented the association, most recently in 1993.4 The first of these reports3 cautioned that “any patient taking . . . [fenfluramine] should report immediately any deterioration in exercise tolerance.” The data reported by Jick et al. suggest that a bias was present: among 9281 untreated subjects, none had aortic or mitral insufficiency. After exclusions based on a review of the case histories, the absence of even a single cardiac murmur in such a large number of patients is not plausible. Thus, bias could readily have accounted for the excess of valve abnormalities observed among the 8723 subjects exposed to dexfenfluramine or fenfluramine.
4 ReferencesSamuel Shapiro, M.B.
Boston University School of Medicine, Brookline, MA 02446-43651
Khan MA ,Herzog CA ,St Peter JV , et al. The prevalence of cardiac valvular insufficiency assessed by transthoracic echocardiography in obese patients treated with appetite-suppressant drugs. N Engl J Med 1998;339:713-718
Full Text | Web of Science | Medline2
Cardiac valvulopathy associated with exposure to fenfluramine or dexfenfluramine: US. Department of Health and Human Services interim public health recommendations, November 1997. MMWR Morb Mortal Wkly Rep 1997;46:1061-1066
Medline3
Douglas JG ,Munro JF ,Kitchin AH ,Muir AL ,Proudfoot AT . Pulmonary hypertension and fenfluramine. BMJ 1981;283:881-883
CrossRef | Web of Science | Medline4
Brenot F ,Herve P ,Petitpretz P ,Parent F ,Duroux P ,Simonneau G . Primary pulmonary hypertension and fenfluramine use. Br Heart J 1993;70:537-541
CrossRef | Web of Science | Medline
Author/Editor Response
The authors reply:
To the Editor: Dr. Myers and Drs. Moyé and Annegers raise important questions about study design and conclusions. Although our comparison group was screened for any previous use of appetite suppressants, it is possible that previously exposed subjects were included in the placebo group in the study by Weissman et al. The validity of the comparison with the placebo group in their study is therefore clouded.
Dr. Shapiro suggests bias as an explanation of the prevalence of valvulopathy that we reported. Before our final analysis, at the request of the FDA, echocardiographic data for 67 patients were submitted to the agency.1 Some of these patients originally underwent echocardiography because of clinical findings. At that same time, the process of duplicate blinded readings was not complete, and valvular insufficiency was reported as a range (i.e., trace to mild, mild to moderate, and so forth) in several instances. The FDA reported the highest level of insufficiency noted. For this reason, there are cases of valvulopathy included in those data that do not appear in our report. In addition, with matching and exclusions, our final analysis included information from 60 of the 67 patients with previously reported echocardiographic data. We note in our report that only the primary reader was potentially aware of exposure status during the assessment of these 60 echocardiograms and that the secondary readers were fully blinded. The primary reader noted valve disease that met the case definition in 18 of 60 studies (30 percent), with secondary readers, who were completely unaware of exposure status, noting valve disease that met the case definition in 17 of 60 (28 percent). A randomly selected third reader independently noted the presence of valve disease in the single study that was interpreted differently in the first and second readings. We agree with Dr. Shapiro's observation of a higher prevalence of valve disease in the early reports and attribute this difference to the early assessment of symptomatic patients, not to bias from unblinding. On the basis of the data in Table 3 and Table 5 of our report, it can be seen that in spite of the early reporting of cases of valve disease and the potential awareness of the primary reader, the final interpretations were strikingly similar. In fact, Table 3 shows that as many as 12 additional cases of valve disease would have been identified had we not implemented independent, duplicate interpretation of all echocardiograms.
We believe our study, with a longer period of exposure and a reference population screened for any prior use of appetite suppressants, provides estimates of the risk of valvulopathy at one end of a spectrum. Studies such as those by Ryan and colleagues, with cardiac ultrasonography performed before drug exposure, give a clearer indication of the actual incidence.2
2 ReferencesMehmood A. Khan, M.D.
John V. St. Peter, Pharm.D.
Charles A. Herzog, M.D.
Hennepin County Medical Center, Minneapolis, MN 554151
Cardiac valvulopathy associated with exposure to fenfluramine or dexfenfluramine: US. Department of Health and Human Services interim public health recommendations, November 1997. MMWR Morb Mortal Wkly Rep 1997;46:1061-1066
Medline2
Ryan DH ,Bray GA ,Helmcke F , et al. Echocardiographic abnormalities in patients treated with fenfluramine or dexfenfluramine. Int J Obes 1998;22:Suppl 3:S77-S77 abstract.
CrossRef | Web of Science
Author/Editor Response
Some readers may conclude that early, brief references in the literature to the possibility that fenfluramines cause pulmonary hypertension may have induced general practitioners in the United Kingdom to look for and diagnose symptomatic valve disorders more carefully in users of these medicines. Many will doubt that such a bias, if present at all, had a material influence on the results of our study.
Potential cases of symptomatic valve disorders among persons unexposed to diet pills were ascertained in our study. However, after a review of the case histories, the valve disorders among the unexposed subjects were found to be due to congenital or rheumatic disorders. If the bias proposed by Shapiro existed, these potential cases would not have been present. The association was present only for persons with “idiopathic” valve disorders, particularly those who used fenfluramines for longer periods of time. The absence of diagnosed cases of symptomatic idiopathic valve disorders in nonusers of fenfluramines in our study is eminently “plausible.”
Biases are always present in observational studies, and whether such biases are relevant to the interpretation of the results of particular studies is a matter of judgment based to some degree on common sense.
I find it highly unlikely that a bias due to clinical suspicion had a material effect on our results, as Shapiro suggests.
Hershel Jick, M.D.
Boston University School of Medicine, Lexington, MA 02421-5207Author/Editor Response
We appreciate the comments of Dr. Myers and Drs. Moyé and Annegers. Dr. Myers points out that the exclusion of patients who used appetite suppressants during the six months before the study may account for the higher prevalence of regurgitation in our placebo group than in the control group in the study by Khan et al. Because our study was fully blinded and randomized, prior users were equitably distributed among the three treatment groups and would have had no influence on the magnitude of the difference in the prevalence of regurgitation between the treated and control subjects.
Furthermore, the implication by Dr. Myers and Drs. Moyé and Annegers that our placebo control group had a higher prevalence of valvular regurgitation than would be encountered in a similar population that had never been exposed to appetite suppressants is incorrect. Recent reports from the Framingham Offspring Study,1,2 in which echocardiography was performed in free-living persons, revealed prevalences of mitral and aortic regurgitation that were similar to those in our placebo group, whether the comparison was based on the FDA criteria (Figure 1Figure 1
Prevalences of Aortic and Mitral Regurgitation in the Framingham Offspring Study and in the Placebo Group in the Study by Weissman et Al., According to the FDA Criteria for Regurgitation.) or any degree of regurgitation. Therefore, we do not believe the lower prevalence of regurgitation in the study by Khan et al. is attributable to differences in previous use of appetite suppressants. Differences between two studies in the prevalence of valvular regurgitation assessed by echocardiography are more likely to arise from differences in echocardiographic-acquisition procedures, such as the ability to ensure that the sonographers are unaware of treatment status, differences in interpretation criteria, or both. Without standardization and quality control, it is not possible to use the findings in the control group in one study to interpret the findings in the active-treatment group in another study, and such comparisons should be interpreted with caution.We agree with Drs. Moyé and Annegers that our study population had a lower exposure to appetite suppressants than the populations studied by Khan et al. and Jick et al., and this difference may account for the lower prevalence of regurgitation. However, further investigation specifically designed to assess a dose-duration threshold is necessary before these conclusions can be formed.
Finally, Dr. Paoletti raises the intriguing question of the effect of weight loss on cardiac valvular function. We agree that our study was, in fact, assessing the combined effects of appetite-suppressant therapy and weight loss. Nonetheless, in an interesting study recently reported by Khan et al.,3 patients with significant weight loss due to gastric-bypass surgery (without exposure to appetite suppressants) did not have an increased prevalence of valvular regurgitation. Therefore, the findings in our study are unlikely to have been due to weight loss.
3 ReferencesNeil J. Weissman, M.D.
John S. Gottdiener, M.D.
Georgetown University Medical Center, Washington, DC 20007-2197John T. Gwynne, M.D.
Wyeth–Ayerst Research, Philadelphia, PA 191011
Singh JP ,Evans JC ,Levy D , et al. Prevalence of valvular regurgitation in a population-based cohort. Circulation 1997;96:Suppl I:I-541 abstract.2
Singh JP ,Evans JC ,Levy D , et al. Obesity and valvular regurgitation: is there an association? J Am Soc Echocardiogr 1998;11:509-509 abstract.3
Khan MA ,St Peter JV ,Herzog CA , et al. Relationship between significant weight loss and cardiac valvulopathy. Int J Obes 1998;22:Suppl 3:S76-S76 abstract.
Author/Editor Response
Drs. Williamson and Yanovski raise a question about the appropriate method of converting the “risk” of an adverse event (the number of persons in a group in whom the event will occur during a period of follow-up) into the “rate” of events for a given number of person-years. If one makes the extreme assumption that all cases of incident valvular regurgitation reported by Jick et al. developed on the first day of follow-up, there would have been 49,965 disease-free patient-years during which incident cases of disease could have developed over a period of five years rather than 50,000 per 10,000. Dividing the risk by the adjusted denominator yields a rate of 0.142 per 1000 patient-years, which is similar to that presented in my editorial. Further research is needed to determine whether the rate of development of valvular regurgitation after exposure to diet-suppressant medications varies over time or whether valvular regurgitation may resolve in some patients.
The second point raised both by Drs. Williamson and Yanovski and by Drs. Moyé and Annegers is how to interpret the difference in incidence rates of valvular regurgitation between exposed and unexposed persons in the studies by Khan et al. and Weissman et al. I stated the apparent “attributable risks” of valvular regurgitation in two ways on the basis of a comparison of data in the exposed and control groups. I made the common-sense assumption that the prevalence of valvulopathy in the unexposed group was the “expected” rate in the population, a common use of control data. Straightforward calculations of attributable risk were then made by subtracting the risk in each study's control group from that in the exposed group. As I pointed out in my editorial and as Drs. Moyé and Annegers note, this resulted in different attributable risks, a difference possibly related to the dose and duration of exposure as well as to methods of ascertaining valve disease.
The second approach involved estimating the likelihood that valvular regurgitation, defined according to the FDA criteria,1 in an exposed person was associated with the use of appetite-suppressant medication. The suggestion by Drs. Williamson and Yanovski that the background prevalence of valvular regurgitation (1.3 and 4.5 percent in the echocardiographic studies by Khan et al. and Weissman et al., respectively) be subtracted from the 100 percent certainty that regurgitation was present in each patient with documented valvulopathy leads to the logical corollary (to account for all “background” valvular regurgitation) that exposed persons without valvular abnormalities have a 1.3 to 4.5 percent chance that something they do not have is due to the prevalence of valvular disease in the general population. A more sensible approach is to consider each instance of valvular disease as having two potential causes. I presented the likelihood that valvular regurgitation after exposure to appetite-suppressant medication was unrelated to this exposure, as an antidote to previous information that only emphasized the connection with the drugs.
1 ReferencesRichard B. Devereux, M.D.
Weill College of Medicine of Cornell University, New York, NY 100211
Cardiac valvulopathy associated with exposure to fenfluramine or dexfenfluramine: US. Department of Health and Human Services interim public health recommendations, November 1997. MMWR Morb Mortal Wkly Rep 1997;46:1061-1066
Medline
Author/Editor Response
Dr. Paoletti raises the interesting question of whether the weight loss itself, as opposed to a direct effect of anorexiants, contributes to the development of valvular dysfunction in patients taking fenfluramine or dexfenfluramine. Conversely, she indirectly queries whether the reversibility of valvular disease, which we recently reported in one patient after discontinuation of the drug,1 might have been partly attributable to regained weight. In fact, our patient did regain the 87 lb she had lost while taking fenfluramine and phentermine, and this weight gain coincided with the regression of multivalvular regurgitation over a two-year period. However, this observation by itself does not allow any firm conclusions to be drawn about a cause-and-effect relation between weight change and valvular dysfunction.
Wong et al.2 surveyed 40 patients who had been taking fenfluramine or dexfenfluramine, in order to determine whether weight loss was correlated with the degree of valvular dysfunction as detected by echocardiographic screening. Although this group of investigators reported a mean weight loss of 27.7±21.1 lb, multiple regression analysis revealed no correlation between weight loss and the development or severity of valvular regurgitation. These data suggest that weight change by itself is not sufficient to explain the valvular dysfunction noted in this group of patients. Further investigation in a larger cohort of patients treated with anorexiants, as well as in patients who have lost substantial weight without diet drugs, would help address this issue.
2 ReferencesLauralyn B. Cannistra, M.D.
Anthony J. Cannistra, M.D.
Brown University School of Medicine, Providence, RI 029121
Cannistra LB ,Davis SM ,Bauman AG . Valvular heart disease associated with dexfenfluramine. N Engl J Med 1997;337:636-636
Full Text | Web of Science | Medline2
Wong J ,Lin S ,Klein A . Valvular heart disease associated with the use of anorexiant drugs. J Am Soc Echocardiogr 1998;11:559-559 abstract.
- Citing Articles (1)
Citing Articles
1
Tim Higenbottam, Liz Laude, Celia Emery, Mohamed Essener. (2004) Pulmonary hypertension as a result of drug therapy. Clinics in Chest Medicine 25:1, 123-131
CrossRef
