- Email: [email protected]
Oral cancer and mouthwash use: Evaluation of the epidemiologic evidence
Oral cancer and mouthwash use: Evaluation of the epidemiologic evidence JOANN G. ELMORE, MO, MPH, and RALPH I. HORWITZ, MD, New Haven, Connecticut
BACKGROUND: An association between mouthwash use and cancer of the oropharynx has been the focus of many epidemiologic studies. PURPOSE: To critically review the methodology of studies of this association. METHODS: Studies were identified through a computerized MEDLINE search of English-language publications from 1976 to 1994. Eligible studies were independently reviewed, and specific criteria were applied to assess issues in the design, analysis, and results of the studies. Methodologic criteria were used that define a case-control study of acceptable epidemiologic quality: [I] histologic confirmation of cancer; (2] use of incident cases; [3) population-based cases; (4) equal exclusions of controls; (5] population-based controls; (6] exposure to mouthwash use before the cancer diagnosis; and [7] adjustments or restrictions for tobacco, alcohol, or both. RESULTS:Seven case-control studies were identified. The odds ratios for these studies ranged from a protective effect [0.82 for any use of mouthwash] to an elevation in risk (2.5 at the highest exposure]. Two studies had significant unadjusted odds ratios for the riskof oral cancer. Of the three studies with data on women adjusted for tobacco and alcohol use, one was statisticallysignificant, and none of the three similar studies in men was significant (although one was close]. The five studies with data on female nonsmokers, nondrinkers, or both had odds ratios ranging from 1.1 to 3.16 [only one study was significant at the p = 0.05 level], and the four studies with similar data on men were not significant. Two studies met only two of the methodologic criteria, three studies met three, one met four, and one met all seven. CONCLUSIONS: Few of the available studies on mouthwash use and riskof subsequent oropharynx cancer adhere to basic methodologic principles of case-control design. Neither the data for the overall association nor the analysis in patients without other clinical risk factors support a link between mouthwash use and oral cancer. [OTOLARYNGOLHEAD NECK SURG 1995;113:253-61 ,)
N u m e r o u s studies during the past several decades have established cigarette smoking and alcohol consumption as the major risk factors for oral and phar~,ngeal cancer. 1-5 Indeed, some investigators have suggested that nearly 75% of these malignancies in the United States are attributable to tobacco and alcohol exposure. 2 Oral conditions, such as poor dentition, insufficient oral hygiene, and lichen planus, have also been suspected as contributing to the risk of oral and pharyngeal cancer. 6-9 More recently, the regular use of mouthwash, From the Department of Medicine (Drs. Elmore and Horwitz) and School of Epidemiology and Public Health (Dr. Horwitz), Yale University School of Medicine. The Procter & Gamble Co. provided a consultant's honorarium to each author for reviewing the studies included in the report. Received for publication Nov. 23, 1994; accepted March 3, 1995. Reprint requests: Joann G. Elmore, MD, MPH, Department of Internal Medicine, 20 York Street, New Haven, CT 06504. Copyright © 1995 by the American Academy of OtolaryngologyHead Neck Surgery Foundation, Inc. 0194-5998/95/$3.00 + 0 23/1/64576
especially among nonusers of tobacco and alcohol, has also been proposed as a risk factor for oral cancer. This suggestion was first made by Weaver et al. in 1979 l° and has now been investigated in seven case-control studies, with the most recent report in 1991.1°-16 The odds ratios found in these studies of the association between mouthwash use and oral cancer have ranged from a protective effect (0.82 for any use of mouthwash) 13 to an elevation in risk (2.5 at the highest exposure)2 ° We provide a framework for reviewing the methods and results from these case-control studies in an attempt to understand this considerable variation in the published literature. Although we initially considered the possibility of combining the data from these studies into a single quantitative summary, as occurs in a meta-analysis, the extreme heterogeneity in patients, methods, and exposures made that scientifically unacceptable. In the remainder of this article, we describe and summarize the results of our methodologic review. 253
254
Otolaryngology Head and Neck Surgery September 1995
ELMOREand HORWITZ
METHODS Selection of Studies
A computerized MEDLINE search of Englishlanguage publications using the words m o u t h w a s h and oral cancer between 1976 and 1994 was performed. References from these articles were reviewed to identify further primary research or details of the study methods. Methodologic Standards
Before reviewing the studies, we used a consensus process that enabled the authors to develop methodologic standards for case-control studies of acceptable quality. These standards, which have been previously described, t7-21 include appropriate methods for the selection of case and control subjects, suitable specification of the exposure to mouthwash, and adjustment for the effect of confounding variables such as cigarette smoking and alcohol use. Failure to meet a methodologic standard in any individual study permits an explanation for the observed (reported) results other than that caused by the principal maneuver (e.g., mouthwash use), and hence increases the likelihood of bias being present. We selectively reduced the criteria for validity to seven main methodologic standards that define a case-control study of acceptable epidemiologic quality: (1) histologic confirmation of cancer; (2) use of incident cases; (3) population-based cases; (4) equal exclusions of controls; (5) population-based controls; (6) exposure to mouthwash use before the cancer diagnosis; and (7) adjustments or restrictions for tobacco, alcohol, or both. These criteria are further described in the following sections. Selection of Case Subjects
Case subjects in these studies are patients with oral cancer. The cancers should be histologically confirmed, and the patients should have newly diagnosed incident episodes of cancer during the study period..Patients who have a history of previous oral cancer may have distinct clinical risk characteristics; they may also have an easier time recalling past exposures (e.g., to mouthwash). In addition, cases should be identified from a population-based sample to avoid a potentially biased sampling of cases from a single selected hospital. An example of appropriate selection of case subjects would include subjects with histologically confirmed new oral cancer selected from a population-based registry.
Selection of Control Subjects
Control subjects in these studies are individuals. without oral cancer. Whenever people were excluded from a study, the exclusions should be applied equally to the candidates for both the case and control groups. If exclusions are applied to cases but not to controls, or vice versa, the exclusions can create bias (e.g., a requirement that controls, but not cases, be free of tobacco-related illnesses). 17In addition, control subjects should be identified from a population-based sample. This requirement is symmetric to the requirement that cases represent a population-based sample. Ascertainment of Mouthwash Exposure
The mouthwash use had to occur before the oral cancer developed for cases and a comparable period for control subjects. This time period was defined as 6 months before the diagnosis of oral cancer for purposes of this review. Clinical manifestations of early oral cancer could modify the subject's use of an agent such as mouthwash (e.g., mouthwash might be used by an individual to treat the initial symptoms of oral c a n c e r - in this instance, mouthwash use would have followed, not preceded, the occurrence of oral cancer). Because of this, defining exposure by mouthwash use at the time of the interview was not acceptable. Adjustment for Confounding Variables
Previous studies have suggested an association between oral cancer and cigarette smoking1-3'5 and between oral cancer and the use of alcohol?'5 In addition, individuals who use mouthwash have high prevalence rates for both cigarette smoking and alcohol use as noted in the seven case-control studies. We therefore set a methodologic standard that analyses adjust for the potential confounding effects of both cigarette smoking and alcohol use, through either stratification, multivariate statistical methods or restriction criteria in design. Review of Studies
The eligible studies were independently reviewed by both authors, and specific standards were applied to assess issues in the design and analysis. We then reviewed their findings and classified each study as to whether it met the main methodologic standards. We were not "blinded" to the study results when assessing the specific methods of each publication because details of study methods can, at times, be located in the discussion sections.
Otolaryngology Head and NeCk Surgery Volume I'13 Number 3
Table
ELMORE and HORWffZ
255
t. Seven reports of case-control studies of m o u t h w a s h use a n d oral c a n c e r Study
Weaver et al., 19791°
No. of cases
No. of controls
Time period of cancer episodes
11
50
Blot et al., 1983"
206
352
Sept. 1, 1975-Aug. 31, 1978
Wynder et al., 198312
571
571
1977-1980*
95
913
June 1981-Feb. 1983"
Young et al., 198614
317
306
Unknown
Kabat et al., 198915
125
107
1982-1986
Winn et el., 199116
866
1249
Mashberg et al., 19851~
Unknown
Jan. 1, 1984-March 31, 1985
Cancers included
Head and neck region (squamous cell cancer) Tongue, gum, floor of mouth, oropharynx, hypopharynx, other, and ill-defined sites within lip, oral cavity, and pharynx Lip, tongue, major salivary glands, gum, floor of mouth, oropharynx, nasopharynx, other, and illdefined sites within lip, oral cavity, and pharynx. Oral, pharynx, and hypopharynx (squamous cell cancer) Oral cavity, oropharyngeal, and hypopharyngeai cancer Tongue, floor of mouth/gums/gingiva, buccal mucosa, palate, tonsil, retromolar trigone, and pharynx Tongue, gum, floor of mouth, oropharynx, hypopharynx, other, and ill-defined sites within the lip, oral cavity, and pharynx
*Time period stated is the years of the interview(the time period of cancer episodes is unclear).
Statislical Analysis
C a s e Selection
The original authors' definitions for "regular use" of mouthwash were used whenever possible. If the authors presented more than one definition of "regular use" of mouthwash, the partition e v e r vs. n e v e r used mouthwash was used for this analysis. When summary odds ratios and/or confidence intervals were not provided in the primary reference but the raw data were available, the odds ratios and 95% confidence intervals were calculated with computer software. 22 Studies were given credit for fulfilling the basic methodologic criteria if the reference clearly indicated the criteria were fulfilled.
The number of cases in individual studies ranged from 11 to 866 (Table 1). The case groups for five studies had histologic confirmation of cancer. Of note, only two studies specified the type of cancer (i.e., squamous cell carcinoma), m3 Only one study specified that the cases were comprised of patients with incident cancers, 16 and the remaining studies did not make clear whether they included patients with preexisting oral cancer. All studies were hospital-based for the selection of cases except for one (Table 2). Control Selection
RESULTS General Overview
Seven case-control studies of the association between mouthwash use and oral cancer were published between 1979 and 1991 (Table 1). ~°-a6 The M E D L I N E search also identified one review article 2~and one letter to the editor. 24This article deals with tlhe seven case-control studies in detail, 1°-~6with additional references used to obtain complete information on the methodologic details of the studies when needed. 1-3,25
The number of controls in individual studies ranged from 50 to 1249 (Table 1). Six studies used hospital-based selection of controls, and one used a population-based method (Table 2). Six studies had unequal exclusions applied to control subjects but not to cases, and two studies had unequal exclusions for approximately one half of the controls. 14,16The unequal constraints included conditions thought to be related to smoking, 1215 drinking, 15 cancer of the esophagus and larynx, 11,13cancer of the lung, 13mental disorders, 1~ and other oral or pharyngeal diseaseJ ~
256
Otolaryngology Head and Neck Surgery September 1995
ELMORE and HORWITZ
Table 2. Criteriafor selection and matching of controls Source of data Study
Cases
Controls
Weaver et al., 19791° Blot et al., 198311
H H
H H
Wynder et al., 198312
H
H
Mashberg et al., 1985 ~3
H
H
Young et al., 1986 TM
H
H
Kabat et al., 198915
H
H
Winn et al., 199116
P
P
Unequal exclusionsof cases compared with controls
Mental disorder, neoplasm of esophagus or larynx, or other oral or pharyngeal diseases* History of tobaccorelated disease:l: History of cancer of the lung, larynx, or esophagus Head and neck cancer thought related to cigarette use (for one of two control groups) Conditions thought to be related to smoking or drinking Patients < 65 yr without a telephone (patients -> 65 yr were selected equally)
Matching of controls to cases Age
Gender
Race
+
+1-
+
+
+
-
-
+ t
-
-
-
-
-
+ 1-
-
+
+
+
H, Hospitalized; P, population based. *Includes the following ICD-8 codes: 290-315, 140-150, 161,500-502, 508, 523, 528, 529. l-Study restricted to one gender only. :[:Tobacco-related diseases defined as a cancer of the lung, esophagus, bladder, pancreas, liver, or kidney; myocardial infarction; stroke; peripheral vascular disease; abdominal aortic aneurysm; chronic bronchitis or chronic obstructive pulmonary disease; gastric ulcer; and cirrhosis of the liver.
In addition, one study selected controls with the method of random telephone dialing,16which means that 100% of the controls must have had a telephone. Despite this requirement for controls for this study, approximately 4% of white and 16% of black patients younger than 65 years did not have a telephone .2Thus only one study had equal exclusions for cases and controls. 1° Matching of controls to cases was performed in five of the seven studies (Table 2). Matching was performed for age in three studies, for gender in two, and by race of the patient in two studies. We also noted that an additional three studies were restricted to a single gender. M o u t h w a s h Exposure
The specific questions asked regarding mouthwash use were available for three studies, but for two of these only the initial question was available (Table 3). There was much variability regarding the amount of information asked about mouthwash use. One study did not present any specific data about mouthwash use (information on brand of mouthwash used was reportedly collected but not analyzed), whereas another study collected information on five specific questions about mouthwash use:
frequency, duration of use, brand, dilution, and retention in mouth. Overall, data were available on the frequency of mouthwash use in six studies, duration of use in five, specific brand of mouthwash in five, dilution in two, and retention in the mouth in two. Exposure prevalence ("regular use" of mouthwash) ranged from 29% to 91% in the cases and from 33% to 80% in the controls (Table 4). In most of the studies, exposure prevalence in the controls was approximately 30% to 40%. Unadjusted Risk of Oral C a n c e r
The unadjusted odds ratios for the risk of oral cancer among the mouthwash users ranged from 0.82 to 2.5 (Table 4). Four studies had 95% confidence intervals that crossed one, indicating a lack of statistical significance at the p = 0.05 level, one study did not present data for the unadjusted odds ratio or for its calculation, and two studies had odds ratios (1.43 and 1.17) that were statistically significant. Analysis by G e n d e r a n d Adjusted for T o b a c c o a n d Alcohol Use
The prevalence of mouthwash use for women ranged from 29% to 58% for cases and from 33% to
Otolaryngology Head a n d Neck Surgery
Volume 113
Number 3
ELMORE a n d HORWITZ
257
Table 3. Information obtained from cases and controls regarding past use of mouthwash
Study
Weaver et al., 1979 ~° Blot et al., t 9 8 3 "
Wynder eL al., 198312
Mashberg et al., 1 9 8 5 1 3 Young et al., 1986 T M Kabat et al., 1989 ~
Winn at at., 1991~6
Specific questions asked regarding mouthwash use
Unknown "Before 1975, did you (did she) ever use mouthwash on a regular basis?" Other questions unknown. "How frequently do you use mouthwash?" "How many years have you been using mouthwash on a regular basis?" Unknown Unknown "Have you ever used mouthwash on a regular basis?" Additional questions unknown. Unknown
Frequency
Duration of use
+ +
+ +
+/-
+
+
-
÷ +
+
+
+
+
Dilution
Retention in mouth
+
÷
÷
--
__
+
+
+
Brand
*Data on brand of mouthwash collected but not analyzed.
Table 4. Results of case-control studies: Rate of regular mouthwash use Rate of regular mouthwash use* Study
Cases
Controls
Weaver et al., 1979 ~° Blot et al., 19831~ Wynder et al., 198312 Mashberg et al., 1985 ~3 Young et al., 198614 Kabat et al., 1989 is Winn et ai., 1991165
91% (10/11) 44% (91/206) 45% (259/571) 43% (41/95) Unknown 29% (36/124) 52% (449/866)
80% (approx 40/50) 42% (149/352) 36% (205/568) 48% (440/913) Unknown 33% (35/107) 44% (551/1248)
Odds ratio [95% confidence interval)
2.5 (0.27, inexact)l 1.15 (0.8, 1.7) 1.43 (1.15, 1.88)1 0.82 (0.53, 1.25) Unknown 0.84 (0.46, 1.51) 1.17 (1.14, 1.63)1-
*The definition of "regular mouthwash use" varied between studies. lOdds ratio and/or 95% confidence intervals were not available in the primary references. Data were calculated with Epiinfo computer software. SData were abstracted from Table 2 of the reference. Regular mouthwash use defined as "ever" used vs. "never" used.
42% for controls. The proportion of mouthwash use for men ranged from 41% to 49% for cases and 37% to 48% for controls. The odds ratio for female mouthwash users ranged from 0.55 to 2.8 and for men ranged from 0.82 to 1.2. Only one of the five studies with information on women had a significant odds ratio12; none of the odds ratios were significant for the four studies with information on men. Note that one study with information for both women and men did not provide information on confidence intervals in the primary manuscript, and they did not provide sufficient raw data in the manuscript to allow calculation of the confidence intervals. 16 (It was assumed that the odds ratios found in this study were not significant.) Four studies provided data on odds ratios strati-
fled by gender and adjusted for tobacco and alcohol use (Table 5). For the three studies with data on women, one had a significant odds ratios of 1.6 (1.1, 2.3). For the three studies with data available on men, no studies were significant, although the study by Winn et al. 16had an odds ratio of 1.4, with a 95% confidence interval that nearly excluded one (1.0 to
1.8). Analysis of Nonsmokers and Nondrinkers
Three studies provided data on the subgroup of patients who neither smoked cigarettes nor drank alcoholic beverages, two studies provided data on patients who did not smoke, and one study provided data separately for the nonsmokers and the nondrinkers (Table 6). For the five studies with data on
258
OtolarYngology Head and Neck Surgery September t 995
ELMORE a n d HORWITZ
Table 5. Results of case-control studies: Odds ratios of oral cancer risk stratified by gender and adjusting
for tobacco and alcohol use Odds ratio [confidence interval] Study Weaver et Blot et al., Wynder et Mashberg
al., 19791° 198311 al., 1983 ~2 et al., 198513
Young et al., 1986 TM Kabat et al., 198915 Winn et al., 199116
Variables used in adjustment
Women
Men
Total
Tobacco
Alcohol
Unknown Unknown NS NA
Unknown NA NS No tobacco = 0.94 (0.61, 1.47); no alcohol = 0.87 (0.57, 1.35) Unknown NA 1.4 (1.0, 1.8)
Unknown Unknown NS NS
Unknown Unknown + +
Unknown Unknown + +
Unknown 0.65 (0.34, 1.22) Unknown
Unknown + +
Unknown + +
Unknown 0.65 (0.34, 1.22) 1.6 (1.1, 2.3)
NA, not applicable; hiS, not significant.
female nonsmokers and/or nondrinkers, the odds ratios ranged from 1.1 to 3.16, and one of these studies was statistically significant (p ___ 0.05 level). For the four studies with data on men, the odds ratios, ranging from 0.23 to 2.01, were not statistically significant. Methodologic Criteria
For the main methodologic criteria for case selection (histology confirmed, incident case, and population-based sampling), two studies met none of the criteria, four met one, and one study met all three of the methodologic criteria (Table 7). Only one study had equal exclusions for cases and controls, and two studies had equal exclusions for about one half of the controls. Only one study used a population-based method for the selection of controls. Six of the studies asked about mouthwash use before the diagnosis of the oral cancer. Three studies had either adjustments or restrictions for tobacco and alcohol use in both cases and controls. Overall only one study fulfilled all of the methodologic criteria. DISCUSSION Case and Control Selection
As noted in the results, the seven studies on oral cancer and mouthwash use were highly variable in nearly every feature in the design of the case-control research. The number of cases and controls ranged from as few as 11 and 50, respectively, in Weaver et al., 1° to 866 and 1249, respectively, in Winn et alJ 6 Except for the study by Winn et al., 16 cases and controls in the other six studies were selected from hospital sources. Only Winn et al. 16 sampled population-based cases and controls and used incident
cases of cancer. As can be seen from Table 1, there is also heterogeneity of the types of cancers included in the studies; for example, some studies included salivary cancer, and others did not. In addition, we assumed that all of the cancer cases were squamous cell carcinoma, although this was only specified in two studies. An important principle of case-control research is to apply criteria for inclusion and exclusion equally to the selection of cases and controls. For example, any exclusion that is applied only to the control group, without similar application to the cases, could create potential bias. Several of these studies, including those with elevated odds ratios for the relationship of mouthwash use to oral cancer, excluded controls but not cases with a history of tobacco-related disease. If mouthwash use is more common among tobacco users, this unilateral exclusion will decrease the proportion of regular mouthwash users among the remaining controls, leading to a spurious increase in the odds ratio. It is especially important to emphasize this particular scientific principle of case-control design because exclusion bias has been established as a source of bias in other studies such as the false relationship between reserpine and breast cancer. 26-28 Information on Mouthwash Use
No cause-effect relationship can be suspected unless the fundamental principle of temporal precedence has been confirmed (i.e,, that use of the suspected exposure preceded, rather than followed, the earliest manifestations of the disease). For casecontrol studies of this topic, questions on mouthwash use should clearly specify that the use of mouthwash preceded the earliest symptoms or date
Otolaryngology Head and Neck Surgery Volume 113 Number 3
ELMORE a n d HORWITZ
259
Table 6. Results of case-control studies for subgroups of patients with neither cigarette nor alcohol use No. of patients
Men
Women
Study
Subgroup studied
Case
Controls
Odds ratio [95% Cl]
Case
Controls
Odds ratio (95% Cl] i
Weaver et Blot et al., Wynder et Mashberg
al., 1979 ~° 198311 al., 1983 ~2 et al., 1985 la
Young et al., 1986 TM Kabat et al., 1 9 8 9 1 5 Winn et al., 1991 ~6
Unknown No tobacco Neither* No tobacco No alcohol No tobacco Neither* Neither*
Unknown 31 36 NA NA Unknown 29 Unknown
Unknown 138 88 NA NA Unknown 37 Unknown
Unknown 1.94 (0.8, 4.7) 3.161- (1.48, 8.92) NA NA 1.50 (0,15, 14.94) 1.38 (0.42, 4.55) 1.1 (0.5, 2.6)
Unknown NA 9 10 28 Unknown NA 11
Unknown NA 105 396 508 Unknown NA 63
Unknown NA 0.23 (0.03, 1.79) 2.01 (0.32, 7.68) 0.83 (0.39, 1.77) 0.91 (0.55, 1.49) NA 1.3 (0.3, 4.6)
CI, Confidence interval. *"Neither" indicates analysis performed for the subgroups of patients who neither smoke cigarettes nor drink alcoholic beverages. #The odds ratio listed in the article is 3.63. This appears to be a miscalculation by the authors.
of diagnosis of the oral cancer. Although most studies did collect information on frequency and duration of mouthwash use, only three studies described the actual questions asked of the subjects. Several of the studies collected data on the brand of mouthwash used. Brand information is essential because brands vary considerably in their alcohol content. A major part of the scientific hypothesis suggests that the mouthwash risk for oral cancer is linked by the high alcohol content of certain brands. Despite this intellectual basis for collecting data on brand use, little attention was paid in these studies to the reliability of these data. Information on "current" brand use may have little relevance to the risk of developing oral cancer, which may be more importantly linked to brands used 5 or even more years ago. We have no data in these studies to establish the consistency of brand use by individuals over many years, or even month-to-month variation in brands. In addition, even where current brand data were acquired, no effort was made to check brand recollection against actual bottles of the mouthwash in the subject's home. Other considerations in the quality of data on exposure relate to principles of blinded interviewers (blind to case or control status of subject) and keeping questioners unaware of the research hypothesis. Although it is often difficult to ensure blinding to case or control status, many investigators successfully disguise their research hypothesis to minimize bias by the interviewers. No attempt was made in any of these studies to achieve even this minimal level of blinding, which is the case-control equivalent of double-blinding in prospective clinical trial research.
Variation in Results
One of the characteristic features of the seven studies is the considerable heterogeneity. This is certainly evident in the wide variation in exposure rates for cases and controls. Regular use of mouthwash ranged from 29% to 91% for cases and from 33% to 80% for controls. Only the study by Weaver et al., which had unusually high exposure rates, had a substantially elevated odds ratios (2.5). However, as noted in Table 7, Weaver et al. 1° did not use standard epidemiologic procedures to minimize bias in the comparison of cases and controls. For the remaining studies, two had odds ratios slightly less than 1 (0.82 and 0.84) and three had odds ratios slightly greater than 1 (1.15, 1.17, and 1.43). Four studies had confidence intervals that encompassed one, indicating no significantly increased risk. The study by Wynder et al. in 198312 had an overall odds ratio of 1.4, which was statistically significant. The authors of this study repeated their work several years later in a different population, however, and found no evidence for a significant relationship between mouthwash use and oral cancer. 15 The 1983 article by Wynder et al. 12 also raised concern because the unadjusted odds ratio was elevated for women (2.8) but not for men (1.13). Although the odds ratio was somewhat higher for men than for women in the study by Young et al. 14 and in women than men in the study by Winn et al., 16 there was no substantial effect modification by gender in any other study. Perhaps the most intriguing argument concerning the possible association between regular mouthwash use and oral cancer focuses on the data for the
260
Table
Otolaryngology H e a d and Neck Surgery September t 995
ELMORE a n d HORWITZ
7.
Fulfillment
of basic
methodologic
criteria*
Case selection
Study W e a v e r et Blot et al., W y n d e r et Mashberg
Histology confirmed
a l . 19791° 198311 al., 198312 et al.,
. + +
1 9 8 5 ~a Y o u n g et al., 1 9 8 6 ~4 K a b a t e t al., 1 9 8 9 is W i n n et al., 1991 ~6
+ + +
Control selection
Incident case
Population based
Equal exclusions
-
+
.
. . . . +
.
Adjustments or restrictions
Population based
. .
.
+1 + +
+1+ -
+/-
-
+/-
+
+ + +
+ + +
+ + +
+ +
.
.
+
Both
+1 + + +
.
.
Alcohol
+ + +/-
. .
Tobacco
-
.
.
Mouthwash use before diagnosis
*Studies were given credit for fulfilling the basic methodologic criteria (+) if it was clearly indicated in the reference. If it was unclear in the reference, or they did not fulfill the basic criteria, they were not given credit ( - ) . 1-Restrictions for cases (nonsmoking, no alcohol use) but no restriction was carried out in the controls.
subset of patients who neither smoke nor drink. This group is important for two reasons. First, tobacco use and alcohol consumption are two major sources of confounding for the relationship between mouthwash use and oral cancer. By restricting the analysis to patients who are homogeneous for nonuse of both substances, the investigators are able to minimize this source of bias. Second, most investigators have argued that any effect of mouthwash use on the risk of oral cancer would be diluted or masked by the larger effects of tobacco and alcohol. Thus restricting the analysis to subjects without concurrent use of alcohol and tobacco removes considerable "noise" and increases the chance of detecting any "signal" that is present. For these reasons, the early reports by Blot et al. u and Wynder et al. ~2that women without tobacco or tobacco and alcohol use had an increased risk of oral cancer with mouthwash use were especially noteworthy. At the same time, in the report by Wynder et al., 12 men who neither smoked nor drank had an odds ratio of 0.23, suggesting a substantially decreased risk. The recent, large study by Winn et al., 16 is especially relevant for this analysis. The overall odds ratio for the association of mouthwash and oral cancer, adjusted for alcohol, tobacco, and other features with multiple logistic regression, was 1.6 for women (1.1 and 2.3) and 1.4 for men (1.0 and 1.8). However, in the restricted subgroup of patients who did not smoke or drink, the odds ratio for women dropped to 1.1 (0.5 and 2.6) and for men to 1.3 (0.3 and 4.6). Although the confidence limits are wide, the attenuation in the odds ratios when the data are analyzed in nonusers of alcohol and tobacco do not support the previous evidence that an increased risk
for oral cancer exists in mouthwash users who do not have other risk factors. Final Comments
When multiple studies exist for a single topic, investigators often choose to combine the data into a quantitative summary using the method known as meta-analysis. However, the widespread heterogeneity and the frequent methodologic shortcomings of the existing seven studies on mouthwash use and oral cancer preclude the use of meta-analysis. The variation in results and conflicting outcomes also make it highly unlikely that a causal relationship can be inferred from this research. Instead of a meta-analysis approach, we propose a general framework for evaluating the validity of these case-control observational studies. The framework does not address other relevant issues such as those relating to the statistical power of a study29or the appropriateness of the statistical methods used. 3°,31 The proposed methodologic criteria are stringent, and this is purposeful. We recognize that rarely will all criteria be met in an observational study. However, if ensuring that the results are true (validity) is to be the major objective in a study, the criteria used in this review serve as a minimum goal. In addition, the major purpose of the criteria is not to select a "winning" study. The analytic framework requires that the methodologic problems of a study be weighted for the potential damage they inflict on its validity. In our judgment the case-control study reported by Winn et al. 16 is methodologically superior to the other studies. Its advantages include a large sample, the use of incident cases, population-based selection
Otolaryngology Head and Neck Surgery Volume 113 Number 3
of cases and controls, avoidance of exclusion bias (for most of the controls), and an attempt to adjust for potential sources of bias. The overall odds ratio in that study of 1.17, although statistically significant at the p < 0.05 level, is not consistent with a substantial clinical risk between mouthwash use and oral cancer. Most importantly, the Winn et al. 16 report discredits the argument that the association between mouthwash and oral cancer is most worrisome because it is greatest in the patients without other risk factors for oral cancer. In fact, in the very subgroup without use of tobacco or alcohol, the odds ratios for mouthwash and oral cancer were decreased compared with those for users of tobacco and alcohol and were nonsignificant (1.1 for women and 1.3 for men). We conclude that a rigorous methodologic review and analysis of the available evidence do not support a causal association between mouthwash use and risk of oral cancer. Neither the data for the overall association nor the analysis in patients without other clinical risk factors supports a link between mouthwash use and oral cancer. W e t h a n k B a r b a r a Israel for assistance with t h e m a n u script. REFERENCES
1. Wynder EL, Stellman D. Comparative epidemiology of tobacco-related cancers. Cancer Res 1977;37:4608-22. 2. Blot WJ, McLaughlin JK, Winn DM, et al. Smoking and drinking in relation to oral and pharyngeal cancer. Cancer Res 1988;48:3282-7. 3. Winn DM, Blot W J, Shy CM, et al. Snuff dipping and oral cancer among women in the southern United States. N Engl J Med 1981;304:745-9. 4. Blot WJ. Alcohol and cancer. Cancer Res 1992;52(suppl): 2119s-23. 5. Mahboubi E, Sayed GM. Oral cavity and pharynx. In: Schottenfeld D, Fraumeni JR Jr, eds. Cancer epidemiology and prevention. Philadelphia: W.B. Saunders Co., 1982:583-95. 6. Kinnebrew M, Gettleman L, Carr RF, Beazley R. Squamous cell carcinoma of the tongue in a young woman. Oral Surg 1984;58:696-& 7. Eisenberg E, Krutchkoff DJ. Lichenoid lesions of oral mucosa. Oral Surg Oral Med Oral Pathol 1992;73:699-704. 8. Marder MZ, Deesen KC. Transformation of oral lichen planus to squamous cell carcinoma: a literature review and report of a case. J Am Dent Assoc 1982;105:55-60. 9. Pogrel MA, Weldon L L Cancer arising in erosive lichen planus in the middle of the dorsum of the tongue. Oral Surg Oral Med Oral Pathol 1983;55:62-6. 10. Weaver A, Fleming SM, Smith DB. Mouthwash and oral cancer: carcinogen or coincidence? J Oral Surg 1979;37: 250-3.
ELMORE and HORWITZ 2 6 t
11. Blot WJ, Winn DM, Fraumeni JF. Oral cancer and mouthwash. J Natl Cancer Inst 1983;70:251-3. 12. Wynder EL, Kabat G, Rosenberg S. Oral cancer and mouthwash use. J Natl Cancer Inst 1983;70:255-60. 13. Mashberg A, Barsa R Grossman ML. A study of the relationship between mouthwash use and oral and pharyngeal cancer. J Am Dent Assoc 1985;110:731-4. 14. Young TB, Ford CN, Brandenburg JH. An epidemiologic study of oral cancer in a statewide network. Am J Otolaryngol 1986;7:200-8. 15. Kabat GC, Herbert JR, Wynder EL. Risk factors for oral cancer in women. Cancer Res 1989;49:2803-6. 16. Winn DM, Blot WJ, McLaughlin JK, et al. Mouthwash use and oral conditions in the risk of oral and pharyngeal cancer. Cancer Res 1991;51:3044-7. 17. Feinstein AR. Clinical epidemiology: the architecture of clinical research. Philadelphia: W.B. Saunders Co., 1985. 18. Horwitz RI, Feinstein AR. Methodologic standards and contradictory results in case-control research. Am J Med 1979; 66:556-64. 19. Schlesselman JJ. Case-control studies: design, conduct, analysis. Cambridge: Oxford University Press, Inc., 1982. 20. Viscoli CM, Lacks MS, Horwitz RI. Bladder cancer and coffee drinking: a summary of case-control research. Lancet 1993;341:1432-7. 21. Hayden GF, Kramer MS, Horwitz RI. The case-control study: a practical review for the clinician. JAMA 1982;247:326-31. 22. Dean AG, Dean JA, Burton AH, Dicker RC. Epi Info, Version 5.01b: a word processing, database, and statistics program for epidemiology on microcomputers. Stone Mountain, Ga.: USD, Inc., 1990. 23. Zunt SL, Beiswanger BB, Niemann SS. Mouthwash and oral cancer (review). Journal of the Indiana Dental Association 1991;70:16-9. 24. Newcombe RG, Arendorf TM. Letter to the editor. J Natl Cancer Inst 1983;71:1103-5. 25. Ford CN, Rebstock KE, Brandenburg JH. Lessons from the network and the next steps: the Wisconsin Head and Neck Cancer Network. In: Engstrom PF, Anderson PN, Mortenson LE, eds. Advances in cancer control research and development. New York: Alan Liss Inc., 1983:279-90. 26. Boston Collaborative Drug Surveillance Program. Reserpine and breast cancer. Lancet 1974;2:669-71. 27. Armstrong B, Stevens N, Doll R. Retrospective study of the association between use of rauwalfia derivatives and breast cancer in English women. Lancet 1974;2:672-5. 28. Horwitz RI, Feinstein AR. Exclusion biases and the false relationship of reserpine and breast cancer. Arch of Intern Med 1985;145:1873-5. 29. Freiman JA, Chalmers TC, Smith H Jr, Kuebler RR. The importance of beta, the type II error and sample size in the design and interpretation of the randomized control trial: survey of 71 "negative" trials. N Engl J Med 1978;299:690-5. 30. Cupples LA, Heeren T, Schatzkin A, Colton T. Multiple testing of hypotheses in comparing two groups. Ann Intern Med 1984;100:122-9. 31. Felson DT, Cupples LA, Meenan RF. Misuse of statistical methods in arthritis and rheumatism: 1982 versus 1967-68. Arthritis Rheum 1984;27:1018-22.
BACKGROUND: An association between mouthwash use and cancer of the oropharynx has been the focus of many epidemiologic studies. PURPOSE: To critically review the methodology of studies of this association. METHODS: Studies were identified through a computerized MEDLINE search of English-language publications from 1976 to 1994. Eligible studies were independently reviewed, and specific criteria were applied to assess issues in the design, analysis, and results of the studies. Methodologic criteria were used that define a case-control study of acceptable epidemiologic quality: [I] histologic confirmation of cancer; (2] use of incident cases; [3) population-based cases; (4) equal exclusions of controls; (5] population-based controls; (6] exposure to mouthwash use before the cancer diagnosis; and [7] adjustments or restrictions for tobacco, alcohol, or both. RESULTS:Seven case-control studies were identified. The odds ratios for these studies ranged from a protective effect [0.82 for any use of mouthwash] to an elevation in risk (2.5 at the highest exposure]. Two studies had significant unadjusted odds ratios for the riskof oral cancer. Of the three studies with data on women adjusted for tobacco and alcohol use, one was statisticallysignificant, and none of the three similar studies in men was significant (although one was close]. The five studies with data on female nonsmokers, nondrinkers, or both had odds ratios ranging from 1.1 to 3.16 [only one study was significant at the p = 0.05 level], and the four studies with similar data on men were not significant. Two studies met only two of the methodologic criteria, three studies met three, one met four, and one met all seven. CONCLUSIONS: Few of the available studies on mouthwash use and riskof subsequent oropharynx cancer adhere to basic methodologic principles of case-control design. Neither the data for the overall association nor the analysis in patients without other clinical risk factors support a link between mouthwash use and oral cancer. [OTOLARYNGOLHEAD NECK SURG 1995;113:253-61 ,)
N u m e r o u s studies during the past several decades have established cigarette smoking and alcohol consumption as the major risk factors for oral and phar~,ngeal cancer. 1-5 Indeed, some investigators have suggested that nearly 75% of these malignancies in the United States are attributable to tobacco and alcohol exposure. 2 Oral conditions, such as poor dentition, insufficient oral hygiene, and lichen planus, have also been suspected as contributing to the risk of oral and pharyngeal cancer. 6-9 More recently, the regular use of mouthwash, From the Department of Medicine (Drs. Elmore and Horwitz) and School of Epidemiology and Public Health (Dr. Horwitz), Yale University School of Medicine. The Procter & Gamble Co. provided a consultant's honorarium to each author for reviewing the studies included in the report. Received for publication Nov. 23, 1994; accepted March 3, 1995. Reprint requests: Joann G. Elmore, MD, MPH, Department of Internal Medicine, 20 York Street, New Haven, CT 06504. Copyright © 1995 by the American Academy of OtolaryngologyHead Neck Surgery Foundation, Inc. 0194-5998/95/$3.00 + 0 23/1/64576
especially among nonusers of tobacco and alcohol, has also been proposed as a risk factor for oral cancer. This suggestion was first made by Weaver et al. in 1979 l° and has now been investigated in seven case-control studies, with the most recent report in 1991.1°-16 The odds ratios found in these studies of the association between mouthwash use and oral cancer have ranged from a protective effect (0.82 for any use of mouthwash) 13 to an elevation in risk (2.5 at the highest exposure)2 ° We provide a framework for reviewing the methods and results from these case-control studies in an attempt to understand this considerable variation in the published literature. Although we initially considered the possibility of combining the data from these studies into a single quantitative summary, as occurs in a meta-analysis, the extreme heterogeneity in patients, methods, and exposures made that scientifically unacceptable. In the remainder of this article, we describe and summarize the results of our methodologic review. 253
254
Otolaryngology Head and Neck Surgery September 1995
ELMOREand HORWITZ
METHODS Selection of Studies
A computerized MEDLINE search of Englishlanguage publications using the words m o u t h w a s h and oral cancer between 1976 and 1994 was performed. References from these articles were reviewed to identify further primary research or details of the study methods. Methodologic Standards
Before reviewing the studies, we used a consensus process that enabled the authors to develop methodologic standards for case-control studies of acceptable quality. These standards, which have been previously described, t7-21 include appropriate methods for the selection of case and control subjects, suitable specification of the exposure to mouthwash, and adjustment for the effect of confounding variables such as cigarette smoking and alcohol use. Failure to meet a methodologic standard in any individual study permits an explanation for the observed (reported) results other than that caused by the principal maneuver (e.g., mouthwash use), and hence increases the likelihood of bias being present. We selectively reduced the criteria for validity to seven main methodologic standards that define a case-control study of acceptable epidemiologic quality: (1) histologic confirmation of cancer; (2) use of incident cases; (3) population-based cases; (4) equal exclusions of controls; (5) population-based controls; (6) exposure to mouthwash use before the cancer diagnosis; and (7) adjustments or restrictions for tobacco, alcohol, or both. These criteria are further described in the following sections. Selection of Case Subjects
Case subjects in these studies are patients with oral cancer. The cancers should be histologically confirmed, and the patients should have newly diagnosed incident episodes of cancer during the study period..Patients who have a history of previous oral cancer may have distinct clinical risk characteristics; they may also have an easier time recalling past exposures (e.g., to mouthwash). In addition, cases should be identified from a population-based sample to avoid a potentially biased sampling of cases from a single selected hospital. An example of appropriate selection of case subjects would include subjects with histologically confirmed new oral cancer selected from a population-based registry.
Selection of Control Subjects
Control subjects in these studies are individuals. without oral cancer. Whenever people were excluded from a study, the exclusions should be applied equally to the candidates for both the case and control groups. If exclusions are applied to cases but not to controls, or vice versa, the exclusions can create bias (e.g., a requirement that controls, but not cases, be free of tobacco-related illnesses). 17In addition, control subjects should be identified from a population-based sample. This requirement is symmetric to the requirement that cases represent a population-based sample. Ascertainment of Mouthwash Exposure
The mouthwash use had to occur before the oral cancer developed for cases and a comparable period for control subjects. This time period was defined as 6 months before the diagnosis of oral cancer for purposes of this review. Clinical manifestations of early oral cancer could modify the subject's use of an agent such as mouthwash (e.g., mouthwash might be used by an individual to treat the initial symptoms of oral c a n c e r - in this instance, mouthwash use would have followed, not preceded, the occurrence of oral cancer). Because of this, defining exposure by mouthwash use at the time of the interview was not acceptable. Adjustment for Confounding Variables
Previous studies have suggested an association between oral cancer and cigarette smoking1-3'5 and between oral cancer and the use of alcohol?'5 In addition, individuals who use mouthwash have high prevalence rates for both cigarette smoking and alcohol use as noted in the seven case-control studies. We therefore set a methodologic standard that analyses adjust for the potential confounding effects of both cigarette smoking and alcohol use, through either stratification, multivariate statistical methods or restriction criteria in design. Review of Studies
The eligible studies were independently reviewed by both authors, and specific standards were applied to assess issues in the design and analysis. We then reviewed their findings and classified each study as to whether it met the main methodologic standards. We were not "blinded" to the study results when assessing the specific methods of each publication because details of study methods can, at times, be located in the discussion sections.
Otolaryngology Head and NeCk Surgery Volume I'13 Number 3
Table
ELMORE and HORWffZ
255
t. Seven reports of case-control studies of m o u t h w a s h use a n d oral c a n c e r Study
Weaver et al., 19791°
No. of cases
No. of controls
Time period of cancer episodes
11
50
Blot et al., 1983"
206
352
Sept. 1, 1975-Aug. 31, 1978
Wynder et al., 198312
571
571
1977-1980*
95
913
June 1981-Feb. 1983"
Young et al., 198614
317
306
Unknown
Kabat et al., 198915
125
107
1982-1986
Winn et el., 199116
866
1249
Mashberg et al., 19851~
Unknown
Jan. 1, 1984-March 31, 1985
Cancers included
Head and neck region (squamous cell cancer) Tongue, gum, floor of mouth, oropharynx, hypopharynx, other, and ill-defined sites within lip, oral cavity, and pharynx Lip, tongue, major salivary glands, gum, floor of mouth, oropharynx, nasopharynx, other, and illdefined sites within lip, oral cavity, and pharynx. Oral, pharynx, and hypopharynx (squamous cell cancer) Oral cavity, oropharyngeal, and hypopharyngeai cancer Tongue, floor of mouth/gums/gingiva, buccal mucosa, palate, tonsil, retromolar trigone, and pharynx Tongue, gum, floor of mouth, oropharynx, hypopharynx, other, and ill-defined sites within the lip, oral cavity, and pharynx
*Time period stated is the years of the interview(the time period of cancer episodes is unclear).
Statislical Analysis
C a s e Selection
The original authors' definitions for "regular use" of mouthwash were used whenever possible. If the authors presented more than one definition of "regular use" of mouthwash, the partition e v e r vs. n e v e r used mouthwash was used for this analysis. When summary odds ratios and/or confidence intervals were not provided in the primary reference but the raw data were available, the odds ratios and 95% confidence intervals were calculated with computer software. 22 Studies were given credit for fulfilling the basic methodologic criteria if the reference clearly indicated the criteria were fulfilled.
The number of cases in individual studies ranged from 11 to 866 (Table 1). The case groups for five studies had histologic confirmation of cancer. Of note, only two studies specified the type of cancer (i.e., squamous cell carcinoma), m3 Only one study specified that the cases were comprised of patients with incident cancers, 16 and the remaining studies did not make clear whether they included patients with preexisting oral cancer. All studies were hospital-based for the selection of cases except for one (Table 2). Control Selection
RESULTS General Overview
Seven case-control studies of the association between mouthwash use and oral cancer were published between 1979 and 1991 (Table 1). ~°-a6 The M E D L I N E search also identified one review article 2~and one letter to the editor. 24This article deals with tlhe seven case-control studies in detail, 1°-~6with additional references used to obtain complete information on the methodologic details of the studies when needed. 1-3,25
The number of controls in individual studies ranged from 50 to 1249 (Table 1). Six studies used hospital-based selection of controls, and one used a population-based method (Table 2). Six studies had unequal exclusions applied to control subjects but not to cases, and two studies had unequal exclusions for approximately one half of the controls. 14,16The unequal constraints included conditions thought to be related to smoking, 1215 drinking, 15 cancer of the esophagus and larynx, 11,13cancer of the lung, 13mental disorders, 1~ and other oral or pharyngeal diseaseJ ~
256
Otolaryngology Head and Neck Surgery September 1995
ELMORE and HORWITZ
Table 2. Criteriafor selection and matching of controls Source of data Study
Cases
Controls
Weaver et al., 19791° Blot et al., 198311
H H
H H
Wynder et al., 198312
H
H
Mashberg et al., 1985 ~3
H
H
Young et al., 1986 TM
H
H
Kabat et al., 198915
H
H
Winn et al., 199116
P
P
Unequal exclusionsof cases compared with controls
Mental disorder, neoplasm of esophagus or larynx, or other oral or pharyngeal diseases* History of tobaccorelated disease:l: History of cancer of the lung, larynx, or esophagus Head and neck cancer thought related to cigarette use (for one of two control groups) Conditions thought to be related to smoking or drinking Patients < 65 yr without a telephone (patients -> 65 yr were selected equally)
Matching of controls to cases Age
Gender
Race
+
+1-
+
+
+
-
-
+ t
-
-
-
-
-
+ 1-
-
+
+
+
H, Hospitalized; P, population based. *Includes the following ICD-8 codes: 290-315, 140-150, 161,500-502, 508, 523, 528, 529. l-Study restricted to one gender only. :[:Tobacco-related diseases defined as a cancer of the lung, esophagus, bladder, pancreas, liver, or kidney; myocardial infarction; stroke; peripheral vascular disease; abdominal aortic aneurysm; chronic bronchitis or chronic obstructive pulmonary disease; gastric ulcer; and cirrhosis of the liver.
In addition, one study selected controls with the method of random telephone dialing,16which means that 100% of the controls must have had a telephone. Despite this requirement for controls for this study, approximately 4% of white and 16% of black patients younger than 65 years did not have a telephone .2Thus only one study had equal exclusions for cases and controls. 1° Matching of controls to cases was performed in five of the seven studies (Table 2). Matching was performed for age in three studies, for gender in two, and by race of the patient in two studies. We also noted that an additional three studies were restricted to a single gender. M o u t h w a s h Exposure
The specific questions asked regarding mouthwash use were available for three studies, but for two of these only the initial question was available (Table 3). There was much variability regarding the amount of information asked about mouthwash use. One study did not present any specific data about mouthwash use (information on brand of mouthwash used was reportedly collected but not analyzed), whereas another study collected information on five specific questions about mouthwash use:
frequency, duration of use, brand, dilution, and retention in mouth. Overall, data were available on the frequency of mouthwash use in six studies, duration of use in five, specific brand of mouthwash in five, dilution in two, and retention in the mouth in two. Exposure prevalence ("regular use" of mouthwash) ranged from 29% to 91% in the cases and from 33% to 80% in the controls (Table 4). In most of the studies, exposure prevalence in the controls was approximately 30% to 40%. Unadjusted Risk of Oral C a n c e r
The unadjusted odds ratios for the risk of oral cancer among the mouthwash users ranged from 0.82 to 2.5 (Table 4). Four studies had 95% confidence intervals that crossed one, indicating a lack of statistical significance at the p = 0.05 level, one study did not present data for the unadjusted odds ratio or for its calculation, and two studies had odds ratios (1.43 and 1.17) that were statistically significant. Analysis by G e n d e r a n d Adjusted for T o b a c c o a n d Alcohol Use
The prevalence of mouthwash use for women ranged from 29% to 58% for cases and from 33% to
Otolaryngology Head a n d Neck Surgery
Volume 113
Number 3
ELMORE a n d HORWITZ
257
Table 3. Information obtained from cases and controls regarding past use of mouthwash
Study
Weaver et al., 1979 ~° Blot et al., t 9 8 3 "
Wynder eL al., 198312
Mashberg et al., 1 9 8 5 1 3 Young et al., 1986 T M Kabat et al., 1989 ~
Winn at at., 1991~6
Specific questions asked regarding mouthwash use
Unknown "Before 1975, did you (did she) ever use mouthwash on a regular basis?" Other questions unknown. "How frequently do you use mouthwash?" "How many years have you been using mouthwash on a regular basis?" Unknown Unknown "Have you ever used mouthwash on a regular basis?" Additional questions unknown. Unknown
Frequency
Duration of use
+ +
+ +
+/-
+
+
-
÷ +
+
+
+
+
Dilution
Retention in mouth
+
÷
÷
--
__
+
+
+
Brand
*Data on brand of mouthwash collected but not analyzed.
Table 4. Results of case-control studies: Rate of regular mouthwash use Rate of regular mouthwash use* Study
Cases
Controls
Weaver et al., 1979 ~° Blot et al., 19831~ Wynder et al., 198312 Mashberg et al., 1985 ~3 Young et al., 198614 Kabat et al., 1989 is Winn et ai., 1991165
91% (10/11) 44% (91/206) 45% (259/571) 43% (41/95) Unknown 29% (36/124) 52% (449/866)
80% (approx 40/50) 42% (149/352) 36% (205/568) 48% (440/913) Unknown 33% (35/107) 44% (551/1248)
Odds ratio [95% confidence interval)
2.5 (0.27, inexact)l 1.15 (0.8, 1.7) 1.43 (1.15, 1.88)1 0.82 (0.53, 1.25) Unknown 0.84 (0.46, 1.51) 1.17 (1.14, 1.63)1-
*The definition of "regular mouthwash use" varied between studies. lOdds ratio and/or 95% confidence intervals were not available in the primary references. Data were calculated with Epiinfo computer software. SData were abstracted from Table 2 of the reference. Regular mouthwash use defined as "ever" used vs. "never" used.
42% for controls. The proportion of mouthwash use for men ranged from 41% to 49% for cases and 37% to 48% for controls. The odds ratio for female mouthwash users ranged from 0.55 to 2.8 and for men ranged from 0.82 to 1.2. Only one of the five studies with information on women had a significant odds ratio12; none of the odds ratios were significant for the four studies with information on men. Note that one study with information for both women and men did not provide information on confidence intervals in the primary manuscript, and they did not provide sufficient raw data in the manuscript to allow calculation of the confidence intervals. 16 (It was assumed that the odds ratios found in this study were not significant.) Four studies provided data on odds ratios strati-
fled by gender and adjusted for tobacco and alcohol use (Table 5). For the three studies with data on women, one had a significant odds ratios of 1.6 (1.1, 2.3). For the three studies with data available on men, no studies were significant, although the study by Winn et al. 16had an odds ratio of 1.4, with a 95% confidence interval that nearly excluded one (1.0 to
1.8). Analysis of Nonsmokers and Nondrinkers
Three studies provided data on the subgroup of patients who neither smoked cigarettes nor drank alcoholic beverages, two studies provided data on patients who did not smoke, and one study provided data separately for the nonsmokers and the nondrinkers (Table 6). For the five studies with data on
258
OtolarYngology Head and Neck Surgery September t 995
ELMORE a n d HORWITZ
Table 5. Results of case-control studies: Odds ratios of oral cancer risk stratified by gender and adjusting
for tobacco and alcohol use Odds ratio [confidence interval] Study Weaver et Blot et al., Wynder et Mashberg
al., 19791° 198311 al., 1983 ~2 et al., 198513
Young et al., 1986 TM Kabat et al., 198915 Winn et al., 199116
Variables used in adjustment
Women
Men
Total
Tobacco
Alcohol
Unknown Unknown NS NA
Unknown NA NS No tobacco = 0.94 (0.61, 1.47); no alcohol = 0.87 (0.57, 1.35) Unknown NA 1.4 (1.0, 1.8)
Unknown Unknown NS NS
Unknown Unknown + +
Unknown Unknown + +
Unknown 0.65 (0.34, 1.22) Unknown
Unknown + +
Unknown + +
Unknown 0.65 (0.34, 1.22) 1.6 (1.1, 2.3)
NA, not applicable; hiS, not significant.
female nonsmokers and/or nondrinkers, the odds ratios ranged from 1.1 to 3.16, and one of these studies was statistically significant (p ___ 0.05 level). For the four studies with data on men, the odds ratios, ranging from 0.23 to 2.01, were not statistically significant. Methodologic Criteria
For the main methodologic criteria for case selection (histology confirmed, incident case, and population-based sampling), two studies met none of the criteria, four met one, and one study met all three of the methodologic criteria (Table 7). Only one study had equal exclusions for cases and controls, and two studies had equal exclusions for about one half of the controls. Only one study used a population-based method for the selection of controls. Six of the studies asked about mouthwash use before the diagnosis of the oral cancer. Three studies had either adjustments or restrictions for tobacco and alcohol use in both cases and controls. Overall only one study fulfilled all of the methodologic criteria. DISCUSSION Case and Control Selection
As noted in the results, the seven studies on oral cancer and mouthwash use were highly variable in nearly every feature in the design of the case-control research. The number of cases and controls ranged from as few as 11 and 50, respectively, in Weaver et al., 1° to 866 and 1249, respectively, in Winn et alJ 6 Except for the study by Winn et al., 16 cases and controls in the other six studies were selected from hospital sources. Only Winn et al. 16 sampled population-based cases and controls and used incident
cases of cancer. As can be seen from Table 1, there is also heterogeneity of the types of cancers included in the studies; for example, some studies included salivary cancer, and others did not. In addition, we assumed that all of the cancer cases were squamous cell carcinoma, although this was only specified in two studies. An important principle of case-control research is to apply criteria for inclusion and exclusion equally to the selection of cases and controls. For example, any exclusion that is applied only to the control group, without similar application to the cases, could create potential bias. Several of these studies, including those with elevated odds ratios for the relationship of mouthwash use to oral cancer, excluded controls but not cases with a history of tobacco-related disease. If mouthwash use is more common among tobacco users, this unilateral exclusion will decrease the proportion of regular mouthwash users among the remaining controls, leading to a spurious increase in the odds ratio. It is especially important to emphasize this particular scientific principle of case-control design because exclusion bias has been established as a source of bias in other studies such as the false relationship between reserpine and breast cancer. 26-28 Information on Mouthwash Use
No cause-effect relationship can be suspected unless the fundamental principle of temporal precedence has been confirmed (i.e,, that use of the suspected exposure preceded, rather than followed, the earliest manifestations of the disease). For casecontrol studies of this topic, questions on mouthwash use should clearly specify that the use of mouthwash preceded the earliest symptoms or date
Otolaryngology Head and Neck Surgery Volume 113 Number 3
ELMORE a n d HORWITZ
259
Table 6. Results of case-control studies for subgroups of patients with neither cigarette nor alcohol use No. of patients
Men
Women
Study
Subgroup studied
Case
Controls
Odds ratio [95% Cl]
Case
Controls
Odds ratio (95% Cl] i
Weaver et Blot et al., Wynder et Mashberg
al., 1979 ~° 198311 al., 1983 ~2 et al., 1985 la
Young et al., 1986 TM Kabat et al., 1 9 8 9 1 5 Winn et al., 1991 ~6
Unknown No tobacco Neither* No tobacco No alcohol No tobacco Neither* Neither*
Unknown 31 36 NA NA Unknown 29 Unknown
Unknown 138 88 NA NA Unknown 37 Unknown
Unknown 1.94 (0.8, 4.7) 3.161- (1.48, 8.92) NA NA 1.50 (0,15, 14.94) 1.38 (0.42, 4.55) 1.1 (0.5, 2.6)
Unknown NA 9 10 28 Unknown NA 11
Unknown NA 105 396 508 Unknown NA 63
Unknown NA 0.23 (0.03, 1.79) 2.01 (0.32, 7.68) 0.83 (0.39, 1.77) 0.91 (0.55, 1.49) NA 1.3 (0.3, 4.6)
CI, Confidence interval. *"Neither" indicates analysis performed for the subgroups of patients who neither smoke cigarettes nor drink alcoholic beverages. #The odds ratio listed in the article is 3.63. This appears to be a miscalculation by the authors.
of diagnosis of the oral cancer. Although most studies did collect information on frequency and duration of mouthwash use, only three studies described the actual questions asked of the subjects. Several of the studies collected data on the brand of mouthwash used. Brand information is essential because brands vary considerably in their alcohol content. A major part of the scientific hypothesis suggests that the mouthwash risk for oral cancer is linked by the high alcohol content of certain brands. Despite this intellectual basis for collecting data on brand use, little attention was paid in these studies to the reliability of these data. Information on "current" brand use may have little relevance to the risk of developing oral cancer, which may be more importantly linked to brands used 5 or even more years ago. We have no data in these studies to establish the consistency of brand use by individuals over many years, or even month-to-month variation in brands. In addition, even where current brand data were acquired, no effort was made to check brand recollection against actual bottles of the mouthwash in the subject's home. Other considerations in the quality of data on exposure relate to principles of blinded interviewers (blind to case or control status of subject) and keeping questioners unaware of the research hypothesis. Although it is often difficult to ensure blinding to case or control status, many investigators successfully disguise their research hypothesis to minimize bias by the interviewers. No attempt was made in any of these studies to achieve even this minimal level of blinding, which is the case-control equivalent of double-blinding in prospective clinical trial research.
Variation in Results
One of the characteristic features of the seven studies is the considerable heterogeneity. This is certainly evident in the wide variation in exposure rates for cases and controls. Regular use of mouthwash ranged from 29% to 91% for cases and from 33% to 80% for controls. Only the study by Weaver et al., which had unusually high exposure rates, had a substantially elevated odds ratios (2.5). However, as noted in Table 7, Weaver et al. 1° did not use standard epidemiologic procedures to minimize bias in the comparison of cases and controls. For the remaining studies, two had odds ratios slightly less than 1 (0.82 and 0.84) and three had odds ratios slightly greater than 1 (1.15, 1.17, and 1.43). Four studies had confidence intervals that encompassed one, indicating no significantly increased risk. The study by Wynder et al. in 198312 had an overall odds ratio of 1.4, which was statistically significant. The authors of this study repeated their work several years later in a different population, however, and found no evidence for a significant relationship between mouthwash use and oral cancer. 15 The 1983 article by Wynder et al. 12 also raised concern because the unadjusted odds ratio was elevated for women (2.8) but not for men (1.13). Although the odds ratio was somewhat higher for men than for women in the study by Young et al. 14 and in women than men in the study by Winn et al., 16 there was no substantial effect modification by gender in any other study. Perhaps the most intriguing argument concerning the possible association between regular mouthwash use and oral cancer focuses on the data for the
260
Table
Otolaryngology H e a d and Neck Surgery September t 995
ELMORE a n d HORWITZ
7.
Fulfillment
of basic
methodologic
criteria*
Case selection
Study W e a v e r et Blot et al., W y n d e r et Mashberg
Histology confirmed
a l . 19791° 198311 al., 198312 et al.,
. + +
1 9 8 5 ~a Y o u n g et al., 1 9 8 6 ~4 K a b a t e t al., 1 9 8 9 is W i n n et al., 1991 ~6
+ + +
Control selection
Incident case
Population based
Equal exclusions
-
+
.
. . . . +
.
Adjustments or restrictions
Population based
. .
.
+1 + +
+1+ -
+/-
-
+/-
+
+ + +
+ + +
+ + +
+ +
.
.
+
Both
+1 + + +
.
.
Alcohol
+ + +/-
. .
Tobacco
-
.
.
Mouthwash use before diagnosis
*Studies were given credit for fulfilling the basic methodologic criteria (+) if it was clearly indicated in the reference. If it was unclear in the reference, or they did not fulfill the basic criteria, they were not given credit ( - ) . 1-Restrictions for cases (nonsmoking, no alcohol use) but no restriction was carried out in the controls.
subset of patients who neither smoke nor drink. This group is important for two reasons. First, tobacco use and alcohol consumption are two major sources of confounding for the relationship between mouthwash use and oral cancer. By restricting the analysis to patients who are homogeneous for nonuse of both substances, the investigators are able to minimize this source of bias. Second, most investigators have argued that any effect of mouthwash use on the risk of oral cancer would be diluted or masked by the larger effects of tobacco and alcohol. Thus restricting the analysis to subjects without concurrent use of alcohol and tobacco removes considerable "noise" and increases the chance of detecting any "signal" that is present. For these reasons, the early reports by Blot et al. u and Wynder et al. ~2that women without tobacco or tobacco and alcohol use had an increased risk of oral cancer with mouthwash use were especially noteworthy. At the same time, in the report by Wynder et al., 12 men who neither smoked nor drank had an odds ratio of 0.23, suggesting a substantially decreased risk. The recent, large study by Winn et al., 16 is especially relevant for this analysis. The overall odds ratio for the association of mouthwash and oral cancer, adjusted for alcohol, tobacco, and other features with multiple logistic regression, was 1.6 for women (1.1 and 2.3) and 1.4 for men (1.0 and 1.8). However, in the restricted subgroup of patients who did not smoke or drink, the odds ratio for women dropped to 1.1 (0.5 and 2.6) and for men to 1.3 (0.3 and 4.6). Although the confidence limits are wide, the attenuation in the odds ratios when the data are analyzed in nonusers of alcohol and tobacco do not support the previous evidence that an increased risk
for oral cancer exists in mouthwash users who do not have other risk factors. Final Comments
When multiple studies exist for a single topic, investigators often choose to combine the data into a quantitative summary using the method known as meta-analysis. However, the widespread heterogeneity and the frequent methodologic shortcomings of the existing seven studies on mouthwash use and oral cancer preclude the use of meta-analysis. The variation in results and conflicting outcomes also make it highly unlikely that a causal relationship can be inferred from this research. Instead of a meta-analysis approach, we propose a general framework for evaluating the validity of these case-control observational studies. The framework does not address other relevant issues such as those relating to the statistical power of a study29or the appropriateness of the statistical methods used. 3°,31 The proposed methodologic criteria are stringent, and this is purposeful. We recognize that rarely will all criteria be met in an observational study. However, if ensuring that the results are true (validity) is to be the major objective in a study, the criteria used in this review serve as a minimum goal. In addition, the major purpose of the criteria is not to select a "winning" study. The analytic framework requires that the methodologic problems of a study be weighted for the potential damage they inflict on its validity. In our judgment the case-control study reported by Winn et al. 16 is methodologically superior to the other studies. Its advantages include a large sample, the use of incident cases, population-based selection
Otolaryngology Head and Neck Surgery Volume 113 Number 3
of cases and controls, avoidance of exclusion bias (for most of the controls), and an attempt to adjust for potential sources of bias. The overall odds ratio in that study of 1.17, although statistically significant at the p < 0.05 level, is not consistent with a substantial clinical risk between mouthwash use and oral cancer. Most importantly, the Winn et al. 16 report discredits the argument that the association between mouthwash and oral cancer is most worrisome because it is greatest in the patients without other risk factors for oral cancer. In fact, in the very subgroup without use of tobacco or alcohol, the odds ratios for mouthwash and oral cancer were decreased compared with those for users of tobacco and alcohol and were nonsignificant (1.1 for women and 1.3 for men). We conclude that a rigorous methodologic review and analysis of the available evidence do not support a causal association between mouthwash use and risk of oral cancer. Neither the data for the overall association nor the analysis in patients without other clinical risk factors supports a link between mouthwash use and oral cancer. W e t h a n k B a r b a r a Israel for assistance with t h e m a n u script. REFERENCES
1. Wynder EL, Stellman D. Comparative epidemiology of tobacco-related cancers. Cancer Res 1977;37:4608-22. 2. Blot WJ, McLaughlin JK, Winn DM, et al. Smoking and drinking in relation to oral and pharyngeal cancer. Cancer Res 1988;48:3282-7. 3. Winn DM, Blot W J, Shy CM, et al. Snuff dipping and oral cancer among women in the southern United States. N Engl J Med 1981;304:745-9. 4. Blot WJ. Alcohol and cancer. Cancer Res 1992;52(suppl): 2119s-23. 5. Mahboubi E, Sayed GM. Oral cavity and pharynx. In: Schottenfeld D, Fraumeni JR Jr, eds. Cancer epidemiology and prevention. Philadelphia: W.B. Saunders Co., 1982:583-95. 6. Kinnebrew M, Gettleman L, Carr RF, Beazley R. Squamous cell carcinoma of the tongue in a young woman. Oral Surg 1984;58:696-& 7. Eisenberg E, Krutchkoff DJ. Lichenoid lesions of oral mucosa. Oral Surg Oral Med Oral Pathol 1992;73:699-704. 8. Marder MZ, Deesen KC. Transformation of oral lichen planus to squamous cell carcinoma: a literature review and report of a case. J Am Dent Assoc 1982;105:55-60. 9. Pogrel MA, Weldon L L Cancer arising in erosive lichen planus in the middle of the dorsum of the tongue. Oral Surg Oral Med Oral Pathol 1983;55:62-6. 10. Weaver A, Fleming SM, Smith DB. Mouthwash and oral cancer: carcinogen or coincidence? J Oral Surg 1979;37: 250-3.
ELMORE and HORWITZ 2 6 t
11. Blot WJ, Winn DM, Fraumeni JF. Oral cancer and mouthwash. J Natl Cancer Inst 1983;70:251-3. 12. Wynder EL, Kabat G, Rosenberg S. Oral cancer and mouthwash use. J Natl Cancer Inst 1983;70:255-60. 13. Mashberg A, Barsa R Grossman ML. A study of the relationship between mouthwash use and oral and pharyngeal cancer. J Am Dent Assoc 1985;110:731-4. 14. Young TB, Ford CN, Brandenburg JH. An epidemiologic study of oral cancer in a statewide network. Am J Otolaryngol 1986;7:200-8. 15. Kabat GC, Herbert JR, Wynder EL. Risk factors for oral cancer in women. Cancer Res 1989;49:2803-6. 16. Winn DM, Blot WJ, McLaughlin JK, et al. Mouthwash use and oral conditions in the risk of oral and pharyngeal cancer. Cancer Res 1991;51:3044-7. 17. Feinstein AR. Clinical epidemiology: the architecture of clinical research. Philadelphia: W.B. Saunders Co., 1985. 18. Horwitz RI, Feinstein AR. Methodologic standards and contradictory results in case-control research. Am J Med 1979; 66:556-64. 19. Schlesselman JJ. Case-control studies: design, conduct, analysis. Cambridge: Oxford University Press, Inc., 1982. 20. Viscoli CM, Lacks MS, Horwitz RI. Bladder cancer and coffee drinking: a summary of case-control research. Lancet 1993;341:1432-7. 21. Hayden GF, Kramer MS, Horwitz RI. The case-control study: a practical review for the clinician. JAMA 1982;247:326-31. 22. Dean AG, Dean JA, Burton AH, Dicker RC. Epi Info, Version 5.01b: a word processing, database, and statistics program for epidemiology on microcomputers. Stone Mountain, Ga.: USD, Inc., 1990. 23. Zunt SL, Beiswanger BB, Niemann SS. Mouthwash and oral cancer (review). Journal of the Indiana Dental Association 1991;70:16-9. 24. Newcombe RG, Arendorf TM. Letter to the editor. J Natl Cancer Inst 1983;71:1103-5. 25. Ford CN, Rebstock KE, Brandenburg JH. Lessons from the network and the next steps: the Wisconsin Head and Neck Cancer Network. In: Engstrom PF, Anderson PN, Mortenson LE, eds. Advances in cancer control research and development. New York: Alan Liss Inc., 1983:279-90. 26. Boston Collaborative Drug Surveillance Program. Reserpine and breast cancer. Lancet 1974;2:669-71. 27. Armstrong B, Stevens N, Doll R. Retrospective study of the association between use of rauwalfia derivatives and breast cancer in English women. Lancet 1974;2:672-5. 28. Horwitz RI, Feinstein AR. Exclusion biases and the false relationship of reserpine and breast cancer. Arch of Intern Med 1985;145:1873-5. 29. Freiman JA, Chalmers TC, Smith H Jr, Kuebler RR. The importance of beta, the type II error and sample size in the design and interpretation of the randomized control trial: survey of 71 "negative" trials. N Engl J Med 1978;299:690-5. 30. Cupples LA, Heeren T, Schatzkin A, Colton T. Multiple testing of hypotheses in comparing two groups. Ann Intern Med 1984;100:122-9. 31. Felson DT, Cupples LA, Meenan RF. Misuse of statistical methods in arthritis and rheumatism: 1982 versus 1967-68. Arthritis Rheum 1984;27:1018-22.