Possibility that certain hypnotics might cause cancer in skin

Summary

Fifteen epidemiologic studies have associated hypnotic drugs with excess mortality, especially excess cancer deaths. Until recently, insufficient controlled trials were available to demonstrate whether hypnotics actually cause any cancers. The US Food and Drug Administration (FDA) Approval History and Documents were accessed for zaleplon, eszopiclone and ramelteon. Since zolpidem was used as a comparison drug in zaleplon trials, some zolpidem data were also available. Incident cancers occurring during randomized hypnotics administration or placebo administration were tabulated. Combining controlled trials for the four drugs, there were 6190 participants given hypnotics and 2535 given placebo in parallel. There were eight mentions of incident non-melanoma skin cancers among participants receiving hypnotics but no comparable mentions of cancers among those receiving placebo (P = 0.064, one-tailed). There were also four mentions of incident tumors of uncertain malignancy among those receiving hypnotics but none among those receiving placebo, so combining uncertain and definite malignancies yielded a more significant contrast (P = 0.016). FDA files revealed that all four of the new hypnotics were associated with cancers in rodents. Three had been shown to be clastogenic. Together with the epidemiologic data and laboratory studies, the available evidence signals that new hypnotics may increase cancer risk. Due to limitations in available data, confirmatory research is needed.

Introduction

To a varying extent, 15 epidemiologic studies have indicated that increased mortality was predicted by the consumption of hypnotic drugs (Ahmad and Bath, 2005; Allgulander et al., 1987, 1990; Fukuhara et al., 2006; Hublin et al., 2007; Kojima et al., 2000; Kripke et al., 1979, 1998; Lack et al., 2006; Mallon et al., 2002; Merlo et al., 1996; Rumble and Morgan, 1992; Sundquist et al., 1996; Thorogood et al., 1992; Winkelmayer et al., 2007). Three studies have found no association between hypnotics usage and mortality (Brabbins et al., 1993; Hays et al., 1996; Phillips and Mannino, 2005), but these had insufficient power to exclude modest risk ratios. In contrast, I have located no epidemiologic evidence that hypnotic drugs prolong survival. The excess mortality associated with hypnotics could not be attributed to comorbidities of insomnia, as excess mortality was not associated with insomnia (Kripke et al., 2002; Lack et al., 2006; Phillips and Mannino, 2005).

Three epidemiologic studies observed that hypnotics consumption specifically predicted cancer deaths (Kripke et al., 1998; Mallon et al., 2002; Merlo et al., 1996). These studies controlled for numerous concomitant risk factors. Case–control studies found that benzodiazepine agonists were associated with ovarian cancer (Harlow and Cramer, 1995; Harlow et al., 1998). An uncontrolled case series recently expressed concern about the carcinogenicity of zopiclone and eszopiclone (Stebbing et al., 2005). In a study monitoring 13 177 people who had taken zopiclone, a surprising 42% of deaths were from cancer (Hajak, 1999). Despite epidemiologic association of hypnotics with cancer risk, observers have remained skeptical that hypnotic drugs actually cause any cancers. Many thought that the demonstrated association was not causal, partly because there has been virtually no published comment on potential mechanisms and animal models of hypnotics-induced cancer. Also, it is widely believed that contemporary marketed hypnotics are safer than those popular at the time when most of the epidemiologic studies were performed.

Randomized clinical trials are usually necessary to prove causality but are not always practical due to costs, duration or ethical concerns. The possibility of examining cancer causation has changed somewhat, now that three novel hypnotics have been introduced to the US market, joining zolpidem, the market leader. Several phase-3 trials of unusual size and duration have been completed to document benefits and risks of long-term hypnotics treatment. For no single drug were the trials large enough to determine whether a specific hypnotic drug caused a significant increase in cancer. However, by combining data from 22 trials of several new hypnotics, accessible from the Food and Drug Administration (FDA) public Internet site (2006), a pilot examination of hypnotics cancer causality could be compiled. Such a compilation has many serious limitations, but there are no better data available from randomizing hypnotics trials regarding cancer risk, nor are better data likely to be available in the near future. It would seem that any data that may focus concern and stimulate further studies of a serious risk should be presented, despite the limitations.

Method

Material from the FDA Approval History and Documents for Sonata (zaleplon), Lunesta (eszopiclone) and Rozerem (ramelteon) were accessed from the Drugs@FDA website (2006; http://www.accessdata.fda.gov/scripts/cder/drugsatfda/). The older Approval History and Documents for Ambien (zolpidem) were also accessed, because zolpidem remained the 2005 market leader. Despite mention of an incident neoplasm associated with zolpidem treatment, this older material had insufficient information to ascertain what malignant neoplasms occurred in those studies or to evaluate the rate. The recent Approval History for Ambien CR (zolpidem tartrate extended release) did not include a new examination of toxicology (McNeil, 2005a). However, documentation of zolpidem as a comparison drug during the zaleplon trials was available, making it possible to include small trials of zolpidem in the compendium. For both ethical and practical considerations, the investigator could not seek access to original case records or research tabulations, but rather estimated the necessary information from the web-based Approval Histories and Documents, which provided summary data.

For all randomized, placebo-controlled trials of each hypnotic compound, the number of incident cancers reported among participants randomized to that hypnotic was tabulated, along with the matching number among those receiving randomized placebo. So far as possible, results from open-label non-randomized portions of the trials were excluded from the formal comparison. Because data were selected only from studies randomizing participants to a hypnotic or to contrast placebo, it is assumed that all cancer susceptibility factors affecting the contrasts such as age, gender and duration of exposure were balanced by randomization. The FDA documents contained disagreements among observers, ambiguities and contradictions about the numbers of incident malignancies, but the tabulations were made to the author’s best estimation. For example, the ramelteon documents stated in several places that no cancers occurred in the placebo group, but a case of lung cancer occurred in a crossover study in which a participant had received several doses of ramelteon and finally received placebo, so that in one table, that cancer was assigned to placebo (McNeil, 2005c). All crossover cancer cases (because these could not be reliably allocated either to hypnotics or to placebo) were omitted from the tabulation of parallel studies. Listings of tumors of undetermined malignancy were tabulated separately, as the reports sometimes lacked conclusive pathologic documentation of the nature of incident tumors. Clearly benign neoplasms were not tabulated. Unfortunately, tabulating the total numbers of hypnotic- and placebo-exposed subjects required some estimation, even leaving out some small phase-I trials. Age ranges were not always available. Accurate exposure distributions were not generally available, but an attempt was made to estimate durations of drug (or placebo) exposure, arbitrarily assuming that dropouts received half of the planned doses.

The Fisher exact test was used to determine if the number of cancers in combined drug and placebo groups differed significantly. The significance criterion selected was P < 0.05, one-tailed, recognizing that warnings or regulatory action might be advisable before there is 95% scientific confidence that a drug is carcinogenic. One-tailed significance values were recorded, because the FDA and the sponsors considered cancers as risks, whereas no anti-cancer benefits were hypothesized, and both animal data and the prior epidemiology predicted the direction of the effect which the investigator was seeking to verify.

Experiments in rodents to explore genotoxicity and carcinogenicity, described in the Approval Documents related to each hypnotic, were briefly reviewed, along with the rates of infections in drug and placebo groups.

Results

For zaleplon, eszopiclone, ramelteon and zolpidem each taken individually, the rate of reported malignancies and the rate of reported total tumors was higher in the hypnotics groups than in matching randomized placebo groups. For trials of all four hypnotics combined, the hypnotics groups underwent an estimated 556 person-years of hypnotics exposure, combining all doses (Table 1), although the mean individual exposure was scarcely more than 1 month. The matching placebo groups underwent 230 person-years of placebo exposure, because the random allocations assigned more subjects to various drug groups than to the placebo (Table 2). Randomized drug-placebo exposures ranged from 1 day to 6 months. Related to these exposures, there were eight incident non-melanoma skin cancers (mostly basal cell cancers) reported in the combined hypnotics groups, but no cancer was reported in the combined placebo groups. In addition, four tumors of uncertain malignancy were mentioned in the hypnotics groups, but none was noted in the placebo groups. Regarding the eight definite malignancies, the incidence rate of malignancy was greater in the hypnotics groups than matching placebo groups with one-tailed P = 0.064. Considering the 12 malignancies and uncertain tumors combined, the incidence rate was greater in the hypnotics groups with P = 0.016. Because the mean exposure duration was virtually identical in drug and placebo groups, the P-values were virtually identical if person-years of exposure were substituted for persons in the Fisher exact tests.

Table 1.   Person years of exposure in randomized parallel drug versus placebo trials

Study	Drug				Placebo				Ages
	n	Retained	Duration	Net years	n	Retained	Duration	Net years
n: number of participants in the study taking study drug or placebo. Retained: a correction factor for dropouts, when available, based on the arbitrary assumption that dropouts occurred 50% through the study. Duration: the duration of exposure to the randomized drug (or placebo) in years, ranging from 0.0027 (1 day) to 0.5 (6 months). Net years: the product of n × retained × duration, yielding the person-years of exposure. In certain zaleplon trials, both zolpidem and placebo were randomized to contrast groups, so zaleplon and zolpidem shared the same placebo group.
Ramelteon
PNFP002	252	1.00	0.0055	1.38	123	1.00	0.0055	0.67	35–60
TL020	561	1.00	0.0959	53.79	287	1.00	0.0959	27.52	18–64
TL021	274	1.00	0.0959	26.27	131	1.00	0.0959	12.56	18–64
TL023	192	1.00	0.0055	1.05	97	1.00	0.0055	0.53	18–64
TL023	191	1.00	0.0027	0.52	97	1.00	0.0027	0.27	18–64
TL025	555	1.00	0.0959	53.22	274	1.00	0.0959	26.27	65+
TL031	50	1.00	0.0767	3.84	49	1.00	0.0767	3.76	18–45
TL032	57	0.72	0.5000	20.52	65	0.82	0.5000	26.49	18–45
Total ramelteon	2132			160.60	1123			98.07
Eszopiclone
46	204	1.00	0.1205	24.59	96	1.00	0.1205	11.57	21–64
49	593	0.80	0.5000	237.20	195	0.78	0.5000	76.05	21–64
47	175	0.99	0.0384	6.65	126	0.97	0.0384	4.69	65–86
48	143	0.95	0.0384	5.21	76	0.95	0.0384	2.77	64–85
26	338	1.00	0.0027	0.93	98	1.00	0.0027	0.27	25–50
Total eszopiclone	1453			274.57	591			95.35
Zaleplon and zolpidem
203 zaleplon	66	1.00	0.0384	2.53	33	1.00	0.0384	1.27	Not stated
204 zaleplon	65	1.00	0.0767	4.99	32	1.00	0.0767	2.45	Not stated
204 zolpidem	32	1.00	0.0767	2.45	Same	Same	Same	Same	Not stated
205 zaleplon	110	1.00	0.0137	1.51	27	1.00	0.0137	0.37	Not stated
210 zaleplon	178	1.00	0.0027	0.49	89	1.00	0.0027	0.24	26–60
301 zaleplon	355	0.89	0.0767	24.24	119	0.93	0.0767	8.49	18–65
301 zolpidem	116	0.88	0.0767	7.83	Same	Same	Same	Same	18–65
303 zaleplon	365	0.92	0.0767	25.76	126	0.93	0.0767	8.99	18–65
303 zolpidem	122	0.89	0.0767	8.33	Same	Same	Same	Same	18–65
306 zolpidem	331	0.93	0.0384	11.81	107	0.91	0.0384	3.73	>65
306 zolpidem	111	0.95	0.0384	4.04	Same	Same	Same	Same	>65
307 zaleplon	484	0.88	0.0384	16.34	153	0.95	0.0384	5.58	18–65
308 zaleplon	270	0.97	0.0384	10.05	135	0.97	0.0384	5.02	≥65
Total zaleplon	2224			97.70	821			36.15
Total zolpidem	381			22.66	Same			Same
Total all drugs	6190			555.53	2535			229.57

Table 2.   Exposure person-years and incident tumors

	Hypnotics	Placebo
*P = 0.064, Fisher exact test, one-tailed. **P = 0.016, Fisher exact test, one-tailed. N/A, not applicable.
Randomized parallel placebo-controlled studies
Number of participants	6190	2535
Person-years	556	230
Cancers	8*	0
Tumors of uncertain malignancy	4	0
Sum cancers plus tumors	12**	0
Open-label or crossover studies
Cancers	13	N/A
Tumors of uncertain malignancy	5	N/A

To provide perspective, there were 13 cancers reported incident during open-label hypnotics trials conducted as follow-on to the randomized studies or in crossover studies, and five further tumors of uncertain malignancy.

Review of the Approval History and Documents revealed that zaleplon, eszopiclone and ramelteon were each reported to be clastogenic. That is, high concentrations of these drugs were shown to damage chromosomes in vitro. Cancers of several kinds were observed when rodents were given very high doses of zaleplon, eszopiclone (in zopiclone) and ramelteon. Rats given high doses of zolpidem developed renal, thyroid and testicular cancers (Weissinger, 1992), but note that the recent labeling of Ambien CR stated ‘No evidence of carcinogenic potential was observed in either mice or rats ...’ (Anonymous 2005a). In the preapproval reports for some of these drugs, FDA reviewers expressed concern that the observed animal carcinogenicity might suggest risks for humans.

A trend for a higher rate of infections in drug than placebo groups was observed with some consistency.

Discussion

When compared with matching placebo, trials of the most contemporary hypnotics in combination observed that participants randomized to hypnotics had a higher rate of incident non-melanoma skin cancers and a significantly higher rate of incident skin cancers plus possibly malignant tumors. The skin cancer incidence rate for participants receiving hypnotics, 1439 per 100 000 person years, was probably higher than the general US rate, but data about the age and gender of participants and about rates for the general population were insufficient for any exact comparison. These combined data offered the first evidence from randomized trials that hypnotics cause any human cancer, most specifically, skin cancer. That these drugs apparently caused cancer in animal screening added credence to clinical trial evidence that hypnotics cause cancer. That use of hypnotics predicted cancer in epidemiologic studies likewise supported the clinical trial compilation.

The trials compiled, of durations from 1 day to 6 months, were so brief that de novo development of cancers might seem implausible to some observers, suggesting that either biases in ascertainment or errors in tabulation may have occurred. The shortest drug exposure with reported cancer observation was 14 days, and most were from 35-day or 6-month studies. On the other hand, experimental observations should not be disregarded merely because they seem inconsistent with contemporary suppositions.

The evidence suggested but did not prove that carcinogenicity might be a property shared by all four hypnotics. The trials of each hypnotic were insufficient in size to determine if each hypnotic by itself caused cancer. Similarly, the available data were not sufficient for contrasts among the four hypnotics for the rates of cancer incidence. Moreover, because of differences in the age groups and various other cancer-risk factors of trial participants, a non-randomized comparison among the four hypnotics would not be valid.

Some FDA reviewers initially had hesitated to approve certain of these drugs because of data suggesting carcinogenicity, but all four hypnotics ultimately received final approval (Andreason, 2004a,b; Atrakchi, 1999a, 2004; Katz, 2004a,b; McNeil, 2005b; Price, 2005; Weissinger, 1992). Evidently the sizes of the animal and human trials for each drug by itself were not sufficient to provide FDA convincing proof of carcinogenicity. The evidence for human carcinogenicity became adequate for statistical analysis only when the data for all four drugs were combined. Also, the incident cancers during double-blind administration were mostly or all basal cell skin cancers, which may not have seemed impressive to FDA reviewers. Parenthetically, basal cell skin cancer was reported in rats given zaleplon (Atrakchi, 1999b). However, the cancers in the open-label and crossover intervals included cancers of the brain, lung, bowel, breast and bladder, and the tumors of uncertain malignancy were likewise widely distributed.

This compilation had many gross limitations. First, the preapproval drug trials were not planned as trials of carcinogenicity, and the Approval Documents were not focused to document cancer outcomes. Even in combination, the size of the trials was hardly robust for cancer detection. This investigator did not have access to complete research protocols, the raw case reports, data bases documenting risk factors and outcomes for each trial participant, or comprehensive summary data. The intervals from first drug administration to first recognition of a tumor were not available. Because of some lack of clarity in the Approval Documents and because of differences of opinion and contradictions recorded between various observers, it is quite possible that this compilation contains imprecise tabulations or mistakes in interpretation. To recheck this compilation, it would be desirable for the FDA to release data from the case reports of all four drugs together, which the FDA has refused to do up to this writing. A more ideal statistical assessment would include data on the duration of exposure and other risk factors such as age and gender. As additional time has elapsed, it may further be now possible for FDA to resolve some of the ambiguities concerning which incident tumors were malignant. Nonetheless, it seems implausible that reinterpretation would erase the preponderance of incident malignancies among participants given hypnotics.

A limitation of this compilation was the concatenation of trials that were diverse in design and duration and that involved different hypnotics. The data for the four drugs were combined because they represented the newest hypnotics that had most recently appeared in the US market, and for which large trials were therefore available. Were adequate data available, a compendium of more similar trials would be preferred. A single trial of a single hypnotic large enough to evaluate cancer as an endpoint would be preferred.

As ramelteon does not act on benzodiazepine receptors, it might seem surprising to observe that its association with cancer was similar to that of benzodiazepine agonists, but so it appeared. Ramelteon, eszopiclone and zaleplon were all clastogenic, which may be a common mechanism of carcinogenesis. Eszopiclone, zaleplon and zolpidem appeared to promote infections, including viral infections. This may have indicated suppression of immune function. Either suppression of immunity or viral infections per se might increase cancer development.

Because the FDA requires manufacturers to report adverse as well as favorable clinical trial outcomes, the FDA NDA documents for these four hypnotics may offer a less-biased presentation than might review of the published literature, where publication bias is serious (Buscemi et al., 2007; Glass et al., 2005). I have not noted that the occurrence of any of these cancers has been previously published, although several of the trials have been published. As these NDA data were compiled, and not included in the tables derived from FDA on-line data, two publications have additionally reported three cancers occurring in clinical trials in indiplon groups, but none was reported in the parallel placebo groups (Roth et al., 2007; Scharf et al., 2007).

This report should not imply that we can have complete confidence from controlled trials that new hypnotics caused any cancer when compared with placebo. The tabulation process should not be considered that reliable until the FDA releases its review of case reports. The statistical outcome was not quite significant if we consider only confirmed malignancies reported to FDA from randomized trials. However, including the tumors of uncertain malignancy or the published indiplon parallel trials (which were not accessed from the FDA source), the statistical association becomes quite significant. Were adequate data available, a multivariate statistical analysis such as a Cox-multiple-hazards model would be preferred.

Because the compilation of cancer incidence awaits FDA confirmation, both for the number of cancers and for the durations of exposure and other risk factors, because the compilation mixes diverse studies of several drugs, and because the number of cancers observed during controlled hypnotics trials remains small, this preliminary analysis should be viewed as an investigative step, rather than sufficient proof that modern hypnotics cause cancer. The data indicate that further investigation is needed, including re-examination of these data by the FDA or European agencies. Expanded clinical trials with a specific focus on cancer incidence are needed. One might suppose that drug manufacturers would wish to prove that their products are safe from cancer risk. If they have insufficient confidence in their products to undertake adequate studies, the public interest demands that the US National Institutes of Health and similar agencies determine the risks or safety of contemporary hypnotics.

Addendum

Following peer review and acceptance of this manuscript, the author received new information from the FDA in response to his prior requests that hypnotics trial case files be reaudited for cancer cases and that the outcome be communicated. The FDA compilation did not include zolpidem, for which FDA concluded there was ‘insufficient information’. The following is quoted with permission from a May 22, 2008 letter from Russell Katz, M.D., Director, Division of Neurology Products, FDA Center for Drug Evaluation and Research. ‘We counted a total of 11 organ-specific and 2 non-specific (‘neoplasm NOS’) cancers occurring in the randomized portions of the trials. ... The 11 organ-specific cancers in the randomized portions of the trials consisted of 1 GI neoplasm, 1 uterine neoplasm, 2 ‘skin’ cancers, and 7 basal cell carcinomas.’ This new FDA compilation appears to strengthen the statistical evidence for cancers associated with hypnotics in randomized trials. Further, it augments evidence that the cancers observed in drug but not in placebo groups included cancers outside the skin. However, Dr. Katz stated, ‘... we do not believe the data discussed in our review provide sufficient evidence of a causal association to recommend a specific regulatory action at this time.’

Author contributions

Dr Kripke planned the study, researched the FDA data and wrote this manuscript.

Competing interests

Dr Kripke has no competing interests, other than a desire to confirm and extend previous work of his research group. He has been a long-time critic of hypnotic safety, e.g. in his non-profit website, http://www.DarkSideOfSleepingPills.com

Acknowledgements

Dr Kripke’s research is supported by the US National Institutes of Health (HL071123, HL07156001 and MH68545) by the Sam and Rose Stein Institute for Research on Aging, and by Scripps Clinic Academic Affairs.