|Home | About | Journals | Submit | Contact Us | Français|
We screened commonly used prescription drugs for possible carcinogenic effects.
In a large health care program we identified 105 commonly used drugs, not previously screened. Recipients were followed for up to 12½ years for incident cancer. Nested case-control analyses of 55 cancer sites and all combined included up to ten matched controls per case, with lag of at least two years between drug dispensing and cancer. Positive associations entailed a relative risk (RR) of 1.50, with p≤ 0.01 and higher risk for three or more, than for one prescription. Evaluation included further analyses, searches of the literature, and clinical judgment.
There were 101 associations of interest for 61 drugs. Sixty-six associations were judged to have involved substantial confounding. We found evidence that of the remaining 35, the following associations may not be due to chance: sulindac with gallbladder cancer and leukemia, hyoscyamine with non-Hodgkin lymphoma, nortriptyline with esophageal and hepatic cancer, oxazepam with lung cancer, both fluoxetine and paroxetine with testicular cancer, hydrochlorothiazide with renal and lip cancer, and nifedipine with lip cancer.
These preliminary findings suggest that further studies are indicated regarding sulindac, hyoscyamine, nortriptyline, oxazepam, fluoxetine, paroxetine, hydrochlorothiazide and nifedipine.
Many adults and children take medications regularly, yet relatively few drugs have undergone significant long-term post-marketing surveillance for adverse effects, including elevated cancer risk. In September 2006 the Institute of Medicine’s Committee on the Assessment of the US Drug Safety System recommended substantial increases in safety studies of marketed drugs (1).
We have been screening commonly used pharmaceuticals for possible carcinogenic effects for over 30 years, identifying drug recipients in an historical database consisting of prescriptions issued at the Kaiser Permanente San Francisco medical center to 143,574 persons during the years 1969 to 1973 (2–5). The International Agency for Research on Cancer (IARC) has issued several reports that reviewed world literature on drug carcinogenesis (6–12). The results of our previous screening studies were cited as providing data on 18 of the drugs evaluated and as the only source of data on humans for 9 of these.
With region-wide implementation achieved by August, 1994, the Kaiser Permanente Medical Care Program (KPMCP) in northern California employs a Pharmacy Information Management System (PIMS) that records all prescriptions dispensed to its subscribers, now numbering over 3 million. Drawing upon this newer resource and a region-wide cancer registry we have screened for possible carcinogenic effects 105 commonly used drugs that were not studied in the previous smaller database, mostly because they were introduced after 1973. We here present associations that are sufficiently strong and convincing, by both a few screening criteria and further evaluation, that more detailed study is suggested to help differentiate those that are causal from those that are due to unrecognized confounding or simply to chance, given the large number of possible associations generated.
The KPMCP is an integrated prepaid health care delivery system that provides comprehensive inpatient and outpatient care, including pharmacy services, to over 3 million current members, who comprise about 30 percent of the residents of the areas served surrounding San Francisco Bay and in the Central Valley of California. The membership is fairly representative of the local population except for some under-representation of both extremes of the economic spectrum (13).
Nested case-control analyses were conducted in a cohort of 6,608,681 subscribers to the KPMCP; drug coverage was identified and the subscribers were followed up starting as early as August 1994, when implementation of PIMS in all KPMCP pharmacies was completed. Entry to the study cohort began at the time of both joining the program and having drug coverage, if later than August, 1994. The ascertainment period for drug use was the same as that for incident cancer, ending on December 31, 2006. Follow-up ended earlier if a cancer of interest was diagnosed, or if the subject left KPMCP for any reason including death, whichever came first.
Ascertainment of pharmaceutical use was based on PIMS, which records all prescriptions dispensed to outpatients. Surveys of subscribers with drug coverage, i.e., at least partial payment for their prescriptions by KPMCP, indicate that they fill nearly 100% of their prescriptions at KPMCP pharmacies (14). The 105 drugs screened were drawn from 230 commonly used drugs with an arbitrarily selected cut-off number of at least 25,000 recipients in the PIMS database by the end of 2003.
Occurrence of cancer was ascertained through KPMCP’s Cancer Registry, which covers all subscribers and contributes to the local Surveillance, Epidemiology, and End Results (SEER) program (15). Persons first diagnosed with cancer before cohort entry were excluded from the analysis of that cancer site. The index date was the date of first diagnosis of the cancer.
In 2000, the middle year of the total 1994–2006 study interval, the annual age-sex-adjusted (to the US census, 2000) incidence per 100,000 of cancer for all sites and both sexes in the KPMCP was 447.2, as compared to 456.4 in California (California Cancer Registry), and 482.8 in SEER (National) data (15).
For each cancer case, ten risk-set matched controls (16) free of the cancer of interest were randomly selected from all of the program’s subscribers of the same sex, same year of birth and same year of starting drug coverage. The index date for controls was the date in the same year as their matched case’s diagnosis that provided equal follow-back time; all controls were still members on that date. Controls were not excluded if they developed the same cancer later and they could be included more than once for different cancer cases but not for the same case. Fewer than ten matched controls could be found for some very elderly cancer patients. For example, three women, age 99, 100, and 101 years at breast cancer diagnosis, could be matched with only 2, 8, and 3 controls, respectively.
Conditional logistic regression was used to calculate odds ratios as estimates of the relative risk of cancer associated with each medication studied using the SAS system (17). The three comparisons made were: any use vs. no use before index date, any use vs. no use at least two years before index date to rule out pre-diagnostic prescribing for symptoms possibly related to cancer (“2-year lag”), and one, two, and at least three prescriptions dispensed vs. no use to ascertain possibly greater risk for longer use. All subjects were considered non-users of a drug until our records showed that it was dispensed to them.
Analyses were performed for invasive cancers at each cancer site (plus non-invasive urinary bladder cancers) listed in the International Classification of Diseases for Oncology, Version 3 (ICD-O-3) (18) except for non-melanoma skin cancers. We combined each subsite in the mouth and pharynx into one site for analysis except that cancers of the lip, salivary glands, and nasopharynx were each analyzed separately.
Because the screening of 105 drugs with 55 cancer sites plus all cancers combined, is so apt to produce nominally statistically significant associations just by chance, we here report only those that met the following criteria: odds ratio at least 1.50 for three or more dispensings in the 2-year lag analysis, p < 0.01 for difference from odds ratio 1.00, and odds ratio for three or more dispensings greater than odds ratio for one dispensing as an indication of dose-response. Occasional noteworthy findings that did not meet these criteria will be mentioned. P values were rounded to three decimal places for Tables 2 and and3;3; thus p<0.001 indicates that p was less than 0.0005 because p = 0.0005 to 0.0009 were rounded up and shown as p = 0.001.
A peer reviewer questioned whether, even for chronically used drugs, three or more prescriptions represented long term use. For all drug/cancer associations not readily attributable to confounding we examined the distribution of total days’ supply dispensed before the cancer diagnosis to the exposed cases and here report the median duration, rounded to the nearest month (Table 3). For drugs with median duration less than 6 months, we repeated the analysis requiring a days’ supply of over two years, looking for associations with odds ratio of 1.50 or greater and p less than 0.01.
Drug usage statistics and subscriber surveys indicated that use of some drugs was associated with use of estrogens, progestins and other female hormone preparations for birth control, menopausal hormone therapy and other indications. Therefore, use of hormones, which may increase the risk of breast or uterine cancer and of tamoxifen or raloxifene, which may reduce the risk of breast cancer, was controlled in analytic models by adding four indicator variables: oral contraceptives, menopausal hormone therapy, raloxifene/tamoxifen, and other hormones. Use was treated as time-varying and was defined as receiving at least two prescriptions for the particular category of hormones before the index date. Estrogens, progestins and oral contraceptives were not screened because they have been studied extensively with good control for confounding variables (12)
When increased risk was noted for the HIV-associated sites, anal cancer, non-Hodgkin lymphoma, or other skin which includes Kaposi’s sarcoma, we repeated the analysis excluding subjects in the HIV registry maintained by KPMCP. Finding virtually no change in the associations with other skin cancers, we determined that they included no cases of Kaposi’s sarcoma; the predominant histological types are listed in the tables. When oropharyngeal cancer was found associated with drugs used to treat or prevent oral thrush in HIV-positive patients, these analyses were also repeated in HIV-free patients. Since anal intercourse is a major risk factor for both anal cancer and HIV infection, reduction in an association of a drug with anal cancer in HIV-free subjects suggests that a life-style that frequently includes this form of sexual behavior may be an important confounder (19).
Lacking data for all subjects on important possible confounders such as cigarette smoking, race/ethnicity and body mass index, we evaluated each association for likely confounding based on clinical judgment and, in some cases, review of computer-stored medical records. The main clinical conditions we were concerned about were: 1) hypertension, a risk factor for renal cancer (20), 2) hyperthyroidism, a risk factor for thyroid cancer (21) and possibly treated with antihypertensive drugs, 3) diabetes mellitus, a risk factor for cancer of the corpus uteri, possibly due to their mutual association with obesity (22), 4) alcoholism, a risk factor for esophageal and liver cancer (23, 24) and 5) cirrhosis of the liver, which also predisposes to liver cancer. Secondary analyses were performed for these cancers after excluding cases and controls with these risk factors or after controlling for them. Considering that cigarette smoking is so much more likely than any drug to have caused lung cancer, we chose to use increased lung cancer risk as a proxy for confounding by cigarette smoking in evaluating associations of drugs with other smoking-related sites such as esophagus, kidney and urinary bladder. In the tables, cigarette smoking will be referred to simply as smoking.
For associations that were not likely due to confounding we searched the literature for possible clinical or biological connections between the drug and risk of cancer.
This study was approved by the Kaiser Permanente Institutional Review Board.
Of the 105 drugs screened, 61 met our criteria for possible increased cancer risk and 44 did not. These two groups are listed in Table 1 subdivided by drug category.
Drug-cancer associations that met the screening criteria but where confounding seemed a likely explanation are shown in Table 2. Drug-cancer associations not readily attributed to confounding are listed Table 3. Brief comments are presented in Tables 2 and and3,3, with further evaluation of selected associations in the Discussion. For drugs with median days supply before diagnosis in the exposed cases of less than 6 months, reanalysis requiring over two years use yielded no associations with odds ratio at least 1.50 and p less than 0.01.
We have screened 105 commonly used drugs for possible carcinogenic effects with follow-up of up to 12 ½ years. Sixty-one drugs had 101 positive associations that met our criteria for consideration. Of these 66 were judged to be likely due to or substantially accentuated by confounding. The factors that we judged to be the main sources of confounding were cigarette smoking, hypertension, and life-style. Cigarette smoking led to acute and chronic respiratory conditions treated by drugs that were associated with smoking-related cancers. Hypertension contributed to the need for cardiovascular drugs which were associated with renal cancer, either due to hypertension itself or to associated diuretic treatment (20). The frequent associations with anal cancer were likely due to a life-style that involved anal intercourse. Other less frequent likely confounders are noted in Table 2.
The 35 associations that we judged less attributable substantially to confounding could all be due merely to chance, given the large number of possible associations that were examined. We attempted to reduce the number of chance associations by requiring a relative risk of at least 1.50 in persons who received three or more prescriptions and a p value of 0.01 or less. Also to rule out confounding by indication due to treatment of symptoms of cancer before diagnosis, we employed a two-year lag whereby only drugs received at least two years before the cancer was diagnosed were counted. Further, we included a dose-response criterion, admittedly not precise, requiring the relative risk with three or more prescriptions to be greater than that with just one prescription. These criteria are arbitrary, based on an attempted balance between finding too many chance associations and missing real causal connections that are weak. The relatively short duration of use of the antibiotics with three or more dispensings was expected. In only four of the other drug/cancer associations not readily attributable to confounding was the median duration of use by the cases before diagnosis less than six months, and many such median durations were well over one year. We found it reassuring that the established association between a thiazide drug and renal cancer (20) was confirmed by this screening procedure.
Strengths of this study include the large numbers of drug recipients with complete cancer follow-up while they are members, often long-term, in a comprehensive health care system. We have objective data on the filling of prescriptions not subject to failure of recall or recall bias. Limitations include our lack of readily accessible information about use of the drugs of concern before August 1994 and follow-up that may not be long enough to detect effects that are early in the carcinogenic process. We were not able to control for important confounders directly. Accordingly, we emphasize the fact that these findings are the result of screening and not of definitive epidemiologic studies that include more detailed information about cancer risk factors.
To follow, listed in alphabetical order of drug, are comments about selected drug-cancer associations that need more consideration than could be provided in Tables 2 and and3.3. Of the associations not readily attributable to confounding (Table 3) there are a few for which evidence suggesting possible causality was found, and further study is recommended. The other associations should just be considered hypotheses, most or all of which may be due to chance. If they are found in other settings, it is helpful to know that they also appeared in our database. Also to follow are a few associations where possible confounding needs additional comment. Finally, recommendations for further study are summarized.
No clinical or biological link was found between this antibiotic and the three cancers. In one study the offspring, up to age 18 months, of women treated with amoxicillin during pregnancy had a reduced risk of leukemia (25).
In our previous study of recipients of all statins, of which atorvastatin accounted for only 1.5%, we found an increased risk of renal cancer with two-year lag in men only, but in corrected sensitivity analysis accounting for possible confounding by cigarette smoking, risk was not increased (25). In one laboratory study atorvastatin exhibited anti-proliferative and pro-apoptotic effects (26), which suggest possible preventive effects. In our study of statins risk of testicular cancer was not found (25). The drug also showed no harmful effects on the testes of beagle dogs and rats (26, 27). Thus, there is little reason to pursue these associations.
The association of calcium channel blockers such as diltiazem has been subjected to considerable epidemiological study including control for confounders, mostly with negative results (30). A recent meta-analysis of randomized controlled trials found no increased risk for calcium channel blockers (31). Our screening does not contribute materially to this body of knowledge.
Antidepressants including selective serotonin reuptake inhibitors (SSRIs) such as fluoxetine have been evaluated in laboratory experiments and epidemiologic studies with mixed results, including some findings of possible prevention (32). Attention in humans has been focused primarily on breast cancer and no relationship has been established (33, 34). No human data concerning testicular cancer were found. The manufacturer (http://ehs.lilly.com/msds/msds_fluoxetine_hydrochloride_capsules_and_tablets.pdf) reported that in a juvenile toxicology study in rats, where the exposure period corresponds to human childhood and adolescence, administration of 30 mg/kg of fluoxetine hydrochloride (how many times not stated) resulted in skeletal muscle necrosis and irreversible degeneration and necrosis of the testis. The dosage to the rats was very large compared to the maximum recommended human dose of 80 mg. per day and the relevance of this experiment to testicular cancer in humans is not known. Further comment is below under paroxetine.
The association between this antihypertensive diuretic and renal cancer has been found repeatedly but there is evidence that hypertension is the risk factor rather than the drugs used to treat it (20). It is reassuring regarding our methodology and data that this known association appeared in this screening. Hydrochlorothiazide was not distinguished from other thiazides in our previous screenings in a smaller cohort but elevated risk of renal cancer was detected for thiazides as a group (5). Data concerning 12,799 thiazide users in that cohort with follow-up through 2002, up to 33 years, showed a standardized morbidity ratio of 1.36 (95% confidence interval 1.03–1.77) based on 55 cases of renal cancer observed and 40.34 expected (Friedman and Habel, unpublished).
Hydrochlorothiazide is a photosensitizing drug and has been associated with increased risk of skin cancer (35). It is very possible that photosensitization could explain the link we found with cancers of the skin and lip (36).
Five of the patients treated with this anti-spasmodic drug reported in Table 2 had ulcerative colitis and one had Crohn’s disease in their computer-stored records. We considered the possibility that these auto-immune conditions might predispose to non-Hodgkin lymphoma but the evidence for their being substantially related to lymphoma risk is weak (37), more likely for Crohn’s disease An additional seven patients had nonspecific diagnoses of non-infectious gastroenteritis or colitis that might also include the auto-immune diagnoses. We did not find any diagnoses of celiac disease in these patients, a condition more clearly related to risk of non-Hodgkin lymphoma (37). Thus, this association is not easily explained away.
Lactulose is a laxative that is broken down by colorectal bacteria into substances with high osmotic pressure, thus bringing water into the bowel. It may be taken orally undiluted and rectally when diluted with water or saline. In the latter form it has been recommended for homosexual men to cleanse the rectum in preparation for anal intercourse. (http://gayspecies.blogspot.com/2007_07_01_archive.html). We are unable to determine how lactulose had been used by the anal cancer patients, but lifestyle confounding is a strong possibility.
Exposure to ionizing radiation is known to cause both myeloid leukemia and thyroid cancer (21, 38). Persons whose thyroid gland was removed because of cancer would likely receive levothyroxine for hormone replacement therapy. We were able to identify prior thyroid cancer in only one of the 115 patients presented in Table 2 but our access to long-term prior records of cancer is limited. Exposure to ionizing radiation could be a possible confounder. Also the fact the levothyroxine is a naturally occurring hormone produced by the thyroid gland would seem to make it an unlikely carcinogen.
In randomized controlled trials and other studies, beta-adrenergic blockers such as this drug have not been found to increase cancer risk (32). One study found a reduced risk of prostate cancer (39). Laboratory data concerning metoprolol showed no evidence of chromosomal damage (40).
See comments regarding diltiazim, also a calcium channel blocker. There is evidence of photosensitivity associated with use of nifedipine, which could explain the increased risk of cancer of the lip, but not of the larynx. (41)
Tricyclic antidepressants have been suspected to be carcinogenic but, as with fluoxetine, the experimental and epidemiologic evidence has been conflicting, including some findings of possible prevention. (32). An association with ovarian cancer has been reported (42) and our early screenings found a transient association of the related drug, amitriptyline, to liver cancer (4, 43) possibly supporting our current finding. Although these two cancer sites suggest alcoholism as a confounder, this did not appear to be the case in our data. Nortriptyline and related tricyclic drugs would seem to merit further investigation.
Confounding by smoking is possible but the drug caused benign and malignant liver tumors and thyroid adenomas in mice with evidence that it was a promoter of liver cancer development in mice and rats (9).
Experimental evidence has mainly pertained to the related SSRI, fluoxetine. (32). Similarly, attention in humans has been focused primarily on breast cancer. The evidence has been mixed (30, 33, 34, 44). No human data concerning testicular cancer were found. The manufacturer reported toxicity to testicular tissue in rats receiving 4 times the maximum recommended human dose for depression for two to 52 weeks (45). The relevance of this evidence to testicular cancer in humans is not known. Given the similar findings for the related drug, fluoxetine, further study would seem to be indicated.
Most evidence concerning nonsteroidal anti-inflammatory drugs (NSAIDs) such as piroxicam suggests that they have cancer-preventive properties, particularly regarding colorectal cancer (46, 47). However, piroxicam did not show either cytostatic or cytotoxic effects on certain human skin melanoma cells, which were found for some other NSAIDs (48). Although no other studies of piroxicam and melanoma were found, large doses of this drug were found to inhibit proliferation and induce apoptosis in some canine cell lines (49), again supporting inhibition of cancer development. The positive association that we noted may well be due to chance.
We found no experimental or human evidence that would suggest carcinogenesis by this antipsychotic drug. Although some diseases are associated with increased risk of Hodgkin lymphoma, psychosis has not been reported to be one of them (50)
Rofecoxib is a NSAID and like piroxicam, believed to have cancer-preventive properties. This likely chance finding has ceased to be of concern since this drug is no longer marketed.
As noted (Table 3) the excretion of sulindac in bile (51, 52) and the presence of its metabolites in gallstones (53) indicate a likely biological connection since gallstones predispose to gallbladder cancer (54). If this association is confirmed, questions needing to be answered are whether sulindac increases risk of gallstones, whether presence of sulindac metabolites in gallstones increases their carcinogenicity, or whether sulindac in bile increases risk of both stones and cancer without stones as a necessary intermediate step to cancer.
Of the five cases of Other Leukemia, the specific diagnoses in the Cancer Registry were: three “Leukemia, not otherwise specified”, one “Acute Leukemia, not otherwise specified” and one “Aggressive NK- (natural killer) cell leukemia”. There have been a few case reports of adverse effects on the bone marrow attributed to sulindac. These included aplastic anemia, erythroblastopenia, and progression of leucopenia to aplastic anemia and acute myeloid leukemia (55–57). The relevance of these to our observation is unknown, but they may add to the desirability of further studies of sulindac and cancer risk.
See comments regarding diltiazim. The one study of elderly patients in Rotterdam showing verapamil differing from other calcium channel blockers in being associated with increased risk of cancer may stimulate more interest in this drug but our screening data add little to the extensive literature on this class of drugs (58,59).
Our preliminary findings plus the additional evidence cited above lead us to recommend further study of the following drugs for possible carcinogenic effects. Sulindac’s association with gallbladder cancer is especially provocative, given the enterohepatic circulation of this drug and its presence in gallstones. Its association with leukemia is also of interest. If there was more evidence that the auto-immune gastroenterological conditions often treated with hyoscyamine were more frequent in the patients who received it and that these conditions were associated with increased risk of non-Hodgkin lymphoma, our finding would be of less concern. Until then, we believe further study of hyoscyamine is indicated. Nortriptyline is of interest in relation to cancer of the liver and perhaps to a lesser degree to other cancers. Although oxazepam’s association with lung cancer raises suspicion of confounding by smoking, animal experiments provide further evidence of possible carcinogenicity and the advisability of further study. The fact that the two related drugs, fluoxetine and paroxetine were both associated with testicular cancer, with relatively little overlap in the patients receiving them, and testicular damage was produced by high doses in experimental animals, suggests the advisability of further study. Finally, our confirmation of increased risk of renal cancer with hydrochlorothiazide has already been amply confirmed but the question of drug vs. hypertension as the cause is still of interest. Given this drug’s photosensitizing effects and association with skin cancer, its association with lip cancer deserves further attention, as does a possible link between the photosensitizing drug, nifedipine, and lip cancer.
Supported by Grant R01 098838 from the National Cancer Institute
Work performed at: Division of Research, Kaiser Permanente Medical Care Program, Oakland, California
Dr. Friedman served on an advisory committee to Roche Laboratories in June, 2008 and in the past 3 years has consulted for law firms serving both plaintiffs and Ortho-McNeil-Janssen Pharmaceuticals regarding litigation concerning celecoxib and Ortho-Evra, respectively. During the last 5 years, Dr. Habel has had research support through contracts with Kaiser Foundation Research Institute from Eli Lilly, Inc; Genomic Health, Inc; Takeda; Merck; AviaraDx; Genentech; and Roche. None of these sponsors had any role in this manuscript; they did not sponsor the research, have any role in its study design, data collection, analysis, interpretation of results, or drafting.