Document yZaLZaQQgZxEY6w3BkZ65gyD

Case-control study of bladder cancer and chlorination by-products in treated water (Ontario, Canada) Will D. King and Loraine D. Marrett (Received 11 March 1996; accepted in revised form 9 May 1996) Chlorineis by far the most commonly used chemicalfor the disiiection of water suppliesin North America. However, chlorine reacts with organic material in the raw water producing a number of halogenated hydrocarbon by-products. This population-based case-control study in Ontario, Canada examined the relationship between bladder cancer and exposure to chlorination by-products in public water supplies. Residence and water source histories and data from municipal water supplies were used to estimate individual exposure according to water source, chlorination status, and by-productlevels (representedby trihalomethane concentration). Exposureswere estimated for the &year period prior to the interview, using 696 cases diagnosed with bladder cancer between 1 September 1992 and 1 May 1994 and 1 9 5 controls with at least 30 years of exposure information. Odds ratios (OR) adjusted for potential confounders were used to estimate relative risk. Those exposed to chlorinated surface water for 35 or more years had an increased risk of bladder cancer compared with those exposed for less than 10 years (OR= 1.41,95 percent contidence interval [CrJ= 1.10-1.81). Those exposed to an estimated THM level 2 50 & i t e r for 35 or more years had 1.63 times the risk of those exposed for lcss than 10 years (CI = 1.08-2.46). These results indicate that the risk of bladder cancer increases with both duration and concentration of exposure to chlorination by-products, with population attributable risks of about 14 to 16 percent. Chlorinationby-products represent a potentially important risk factor for bladder cancer. Cancer Causes and Control 1996,7,596-604 K e y words: Bladder neoplasms, Canada, chlorine, environmental exposures. Introduction Chlorine is by far the most commonly used chemical for the disinfection of water supplies in North America.' Chlorinated water sources serve over 75 percent of the population of Canada? During the chlorination process, chlorine reacts with organic material in the water producing hundreds of halogenated hydrocarbon compounds as by-products.' Organic materials include both naturally occurring substances and those resulting from industrial and municipal waste water or run-off. Of the identified by-products, trihalomethanes (THM) occur most frequently and have been the most thoroughly investigated.'~C~omprised primarily of a group of four volatile organics (chloroform, bromodichloromethane; dichlorobromomethane, and bromoform), THMs are mutagenic and are possible human carcinogens: Because the concentration of THMs in treated water is measured routinely and it is reasonable to assume that it correlates highly with the formation of other chlorination by- Dr King is with the Department of Community Health and Epidemiology, Queen's University,Kingston, Ontario, Canada. Dr Marrett is with the Ontario Cancer Treatment and Research Foundation, and the Department of Preventive Medicine and Biostatistics, University of Toronto, Canada. Address correspondence to Dr King, Department of Community Health and Epidemiology, Queen's University, Kingston, Ont., Canada, K7L 3N6. This research was funded by Health Canada and also was supported by the National Health Research and Development Program through a fellowship to R7D.K. 596 Cancer Causes and Control. Vol7. 1996 1996 Rapid Science Publishers Bladder cancer and chlorination by-prodwrs products,' it is a useful indicator of the level of chlorination by-products in treated water. Recent case-control studies consistently have reported an association between exposure to chlorination byproducts and risk of bladder cancer?9 A meta-analysis of seven ~tudies,8-w'~hich were conducted between 1978 and 1988, yielded a statisticallysignificant overall relative risk estimate of 1.21 (95% confidence interval [CI] = 1.09-1.34).j5The methods used, populations studied, and exposure indices varied among these studies. The magnitude of the association tended to increase with improved exposure assessment. In general, studies which collected more detailed information about participants' lifetime residences and sources of drinking water and those which accurately represented level of chlorination by-products produced higher risk estimate^.'^^'^ Assessment of individual exposure to chlorination by- products in water supplies is difficult because of the lack of specifichistorical exposure data. Epidemiologicstudies have employed a variety of indicators of exposure to chlorination by-products, including use of surface water, use of chlorinated surface water, and current level of THMs in the water supply. However, the composition of chlorinated water differs from location to location as a result of variations in the characteristicsof the raw water and in the treatment procedures employed. Therefore, none of these indicators necessarily represent the same exposure to chlorination by-products in different locations. Because of the misclassification inherent in these exposure assessment methods, it has been difficult to detect risks of relatively small magnitude. In addition, the methods of assessing exposure to chlorination byproducts describedabovehave not permitted examination of risk associatedwith exposureto varying concentrations of chlorination by-products, particularly cumulativelifetime exposure. The objective of this study was t o examine the relationship between bladder cancer and exposure to chlorination by-products in household water supplies. An attempt was made to build on previous methodologies by considering individual's residences and water sources over many years and estimatinghistorical concentrations of chlorination by-products in water supplies. Bladder cancer risks are examined according to duration and level of exposure to chlorination by-products in household water supplies. Materials and methods Selection of cases and controls A population-based case-control study was conducted, where cases were residents of Ontario (excluding the North), Canada, identified with a histologically confirmed diagnosis of primary cancer or carcinoma in situ of the bladder (site code 188, ICD-9);" diagnosed between 1 September 1992 and 1 May 1994, at ages between 25 and 74 years. Cases were identified from pathology reports routinely submitted to the Ontario Cancer Registry (OCR) at the Ontario Cancer Treatment and Research Foundation. Pathology reports typically are received at the OCR within six months of diagnosisand an estimated 95 percent of cancer incidence in Ontario is registered.'* Of 1,694 eligible cases identified in the OCR, an appropriate physician could not be identified for 18 patients. Thephysicianconsent process indicated that 232 patients were recently deceased or too ill to be contacted. Of the remaining 1,444 patients, consent was received to contact 1,262 (87 percent). A few physicians in the study area refused to participate in the study, thus eliminating 76 patients (five percent). The physician refused to provide consent for 51 patients (four percent) and no response was received from the physician by the end of data collection for 55 patients (four percent). Controls were an age-gender frequency-matched sample of the general population in the same area. Since the control subjects were used also to study cancers of the colon and rectum with respect to the same exposures, they were selected to have the expected age-gender distribution of the three cancer sites combined. Controls were accrued through a multistage procedure where households were selected randomly from a computerized database of residential telephone listings, a census of each household led to the identification of the appropriate study subject accordingto ageand gender,and this subject was askedwhether he/she would participate. Thedatabase of telephone numbers was a random sample of all listed residential telephone numbers in the study area. Interviewers contacted 10,219 households during control accrual, and 91 percent provided a census of residents. Because of the age-gender frequency-matching criteria employed, many households did not have a resident who was eligible and needed for the study at a particular point in time. In over 90 percent of the households with an eligible resident (n = 2,768), the person selected for the study agreed to receive a questionnaire ( n = 2,494). Exposure assessment Relevant exposure and confounding variables were collected using a mailed questionnaire in combination with a computer-assisted telephone interview. Structured interviews following computer prompts and validity checks were conducted after subjects had received the mailed questionnaire. Interviewers were blinded as to case-control status of the subject. Questions were included on demographics (e.g., gender, date of birth, and education), other potentially important confounding variables (e.g., smoking history and usual diet prior to Cancer Causes and Conuol. Vol7. 19% 597 W D.King and L.D.Marrett diagnosis) and information pertaining to the primary elrposures of interest (e.g,. residence and water source history and usual water consumption prior to diagnosis). Subjects reported their drinking-water source and other household water sources at each residence as municipal, household well, bottled water, or other. Volume of tap water consumed was calculated from the reported daily frequency of consuming beverages containingwater, two years prior to the interview, and usual source of water (tap or bottled) used to make hot and cold beverages. Individual information was supplemented by water treatment data collected through a mailed survey of historical treatment practices at plants serving the study area, The survey ascertained, for each plant currently operating in the study region, area served, water source and characteristics, and treatment practices for years of operationbetween 1950and 1990. Water treatment information was reported for an average day in August in five-year intervals and that observation was used to represent water characteristics for the years surrounding that date. Each water supply in the study area used by each participant was characterized by three parameters: source (surface cf ground), chlorination status (chlorinated cf unchlorinated), and level of chlorination by-products (estimated summer THM level). Water source and chlorination status were provided directly by treatment plant surveys. Past levels of chlorination by- products were represented by historical THM levels. These were estimated from the treatment plant survey data using a model which was developed to predict the THM level in treated water, from characteristics of the treatment process. The model was built using measurements recorded by the Ontario Drinking Water Surveillance Program (ODWSP) between 1988 and 1992 for 114 treatment plants. In the database maintained by this Program, plants have variable numbers of observations recorded at different points in time for a total of 2,494 observations. Potential predictors of THM level in treated water were identified on the basis of the literature, availability in the ODWSP database, and availability historically in treatment plant records (1950-90). The latter was determined through consultation with a number of plant operators and representatives of ODWSP. At this stage, it was indicated that two potentially important predictors of THMlevel,turbidity and pH, were not consistentlyavail- able on an historical basis. Those predictors selected for inclusion in the model-building process were: characteristics of the raw water (source-surface lake, surface river, or ground; depth of the intake pipe; water temperature); pre-treatment procedures (chlorination dose; chloramination); treatments employed (coagulation; polyelectrolytes; activated carbon); and post-treatment procedures (chlorination dose; de-chlorination).Separate 598 Cancer Causes and Control Vol7. 1996 models were created for three subgroups: surface water source with pre-treatment chlorination; surface water source without pre-treatment chlorination; and ground water source. To evaluatehow well these models predicted historical THMlevels, theywere applied to data for the independent variables available in the ODWSP database for 1986 and 1987 (n= 354). Values of total THMs predicted by the models were compared with those observed and present in the database. The correlation between observed and predicted values was 0.76. When observed and predicted THM concentrations were dichotomized at 50 pg/L, the model predicted values with a sensitivity of 84 percent and a specificity of 76 percent. These models were applied to survey data from the water treatment plants to estimate historical THM levels by time for each plant. Since the treatment plant survey requested information for August, when chlorination by- product levels are usually the highest, the THM estimate represents a summer level and generally will be close to the annual peak value for that plant. Private wells were assigned a THM concentration of zero. Study participants'water exposures were estimated by linking a subject's residence and water source history to the relevant treatment-plant data by time and geographic area. For each residence, only the subject's drinking-water source was employed, as another household source was seldom reponed. Thus, each subjectwould have, for each year of residence in the study area, indicators of the exposures of interest (source, chlorination status, and THM level). Duration of exposure estimates were calculated for each subject by summing the number of years in each exposure category. Results are presented for total years of exposure to chlorinated surface water and to water with estimated summer THM level at or above 25, 50, and 75 pg/L, in addition to quartiles of THM-years, the product of the continuous estimate of summer THM level and years at that level (analogous to pack-years of cigarette smoking). The analysis considers exposures occurring over the 40-year period preceding two years prior to the subject's interview. To reduce the level of misclassification in exposures, these analyses are restricted to subjects with 30 or more years of known water history. Analysis Odds ratios (OR) are used throughout as estimates of relative risk. For categorical variables, ORs are presented for each level of exposure in comparison with the lowest category of exposure. For the exposures of primary interest, unconditional logistic regression was used to obtain ORs and CISadjusted for potential confounder^.'^ Age and gender were included in all analyses. Other potentially important confounders were determined on Bladder cancer and chlorination by-products the basis of the bladder cancer literature and a backward stepwise logistic regression w a s used to identify a parsimonious model predicting bladder cancer risk. Factors considered to be potential confounders were: smoking; education; consumption of alcoholic beverages; coffee consumption; total fluid consumption; and dietary intake of energy (total calories), protein, fat, cholesterol, fiber, and vitamin A. Potential confounderswhich demonstrated a log linear relationship with risk in categorical form were analyzed as continuous variables. Exclusion of factors from the model was based on a score-test P-value of greater than 0.10; this conservative P-value ensured that all factors which might confound the relationship of interest were included in the model. The test for trend for categorical variables was based on the likelihood ratio test conducted by including the factor as a continuous term in a logistic regressionmodel. Results Response rates Questionnaires were mailed to 1,262 cases and 2,494 control subjects. One hundred and fifty-six cases and 62 controls were unable to complete questionnaire because of illness, death, or communication barriers. Of those remaining, questionnaires were completed,by 84 percent of cases (n= 927) and 87 percent of controls (n= 2,118). The overall response rate in cases is 73 percent, calculated as theproduct of physician consent and caseparticipation, and in controls is 72 percent, the product of the response to initial contact(ie.,agreed to givea householdcensus,and where an eligible subject was present, agreed to receive a questionnaire) and the response to the telephone interview. Of 431 water treatment-plant surveys mailed, 385 (89 percent) were returned with all or partial information. Response rates were similar across regions, and data were received from all major population centers. The Ontario Drinking Water Surveillance Program (1986-95) and the National Survey of Water Facilities (conducted in 1986) provided information on water source and treatments used for 1985 and 1990 for 44 of the 46 facilities with no returned survey. The presented analyses includes the 696 cases (75 percent) and 1,545 controls (73 percent) for whom water source characteristicswere available for at least 30 of the 40 years prior to interview, as described earlier. Potential confounders As seen in Table 1, cases are more often male (76 percent) in comparison with the control group (63 percent) and are slightly older. This occurred because controls were selected to match the total case group of bladder, colon, and rectum cancers. Potential confounders included in the parsimonious model predicting cancer risk are age, gender, pack-years of smoking (log [pack years + l]), current smoking status, highest level of education, and total energy intake (Table I), The strongest predictor of risk is pack-years of smoking and we found that the best fit with the data for this factor is obtained with the form log@ack-years + I). Using the parameter estimate from this representation, along with that for current smoking Table 1. Distribution and odds ratios (OR) and 95% confidence intervals (CI) for cancer of the bladder associated with potential confounders, Ontario Confounder Cases ( n = 696) 64.2' Controls (n = 1,545) 61 .O' OR. - - Gender Female 24% Male Smoking (log [pack-years + 11) 76% 47.8' Current smoker No 60% Yes 40% Level of education c Complete high school 51% Complete high school 18% 1 I Community college, some or complete university 3170 37% 63% 40.4' 76% 24% 39% 22% 40% -- - 1.27 (1.18-1.36) 1.o - 1.59 (1.25-2.03) 1.o - 0.76 (0.59-0.99) 0.75 (0.60-0.94) Energy (1,000 kcal/wk) 41'0 396' 1.03 (1.00-1.07) * Odds ratio (95 percent confidence interval [Cl]) adjusted for all other factors In the table. Age and gender distribution of controls was restricted by the sampling procedure, therefore odds ratios for these factors are not presented. Mean values; for smoking mean number of pack-years among ever smokers. Cancer Causes and Conrml. Vol7. 19% 599 E! D.King and L. D.Marrett status, a current smoker with a total of 40 pack-years (the mean level among control ever-smokers) has four times the risk of bladder cancer than a life-long nonsmoker. A pattern of decreasing risk with higher educational attainment is observed. Bladder cancer risk increaseswith higher levels of energy intake. Other potential confounders (consumption of alcoholic beverages, coffee, and total fluid consumption; and dietary intake of protein, fat, cholesterol, fiber and vitamin A) are not associated with bladder cancer risk after controllingfor the factors listed in Table 1. Water soIcrce Chlorinated surface water was used for 35 or more years by 35 petcent of controls, and for 20 to 34 years by 28 percent (Table 2). Among controls, exposure for 35 or more years to estimated THM levels 2 25,50 and 75 pg/L is observed for 14, five, and four percent of subjects, respectively. There is a pattern of increasing risk with years exposed to chlorinated surface-water, where exposure to chlorinated surface-watersource for 35 years or more is associated with a relative risk estimate of 1.41 (CI = 1.09-1.81) in comparison with those exposed for less than 10 years. Risk estimates are similar for each of the THM measures. While the general tendency is for relative risk to rise with increasing duration of exposure, it is only for 35 or more years of exposure that relative risks are consistentlysigdicant and of higher magnitude. Relative risks for 35 or more years of exposure to THM levels 2 25 pg/L, 2 50 pg/L, and 2 75 pg/L, are 1.58, 1.63 ,and 1.68 compared with those exposed at each level for less than 10years. A statisticallysigdicant increase in bladder cancer risk also is observed for the highest quartiles of cumulative exposure (THM-years) relative to the lowest quartiles of exposure (OR = 1.44, CI = 1.10-1.88).When THM-years is modeled as a continuous factor, bladder Table 2. Odds ratios (OR) and 95% confidence intervals (CI) for cancer of the bladder according to years of exposure to water factors, Ontario Water factor and years of exposuren No. of Cases No. of Controls Crude Adjustedb Chlorinated (surface source) 0-9 yrs 10-19 20-34 35+ THM 2 25 F ~ / L 0-9 yrs 10-19 20-34 35+ THM 2 50 pg/L 0-9 yrs 10-19 20-34 35+ THM 2 75 wg/L 0-9 yrs 10-19 20-34 35+ THM-years (quartiles) 1 (0-583pg/L-years) 2 (584-1,505) 3 (1506-1956) 4 (1957-6425) THM-years (continuous per 1,000pg/L-years) 157 55 169 315 154 121 278 143 253 226 163 54 356 233 67 40 151 160 165 220 1.63' 413 154 433 545 407 282 632 224 650 51 9 297 79 876 486 128 55 387 385 387 386 1.48' OR (CI) OR (CI) 1.o - 0.94 (0.65-1.37) 1.03 (0.79-1.34) 1.52 (1.20-1.93) 1.o - 1.04 (0.71-1.53) 1.15 (0.86-1.51) 1.41 (1.09-1.81) 1.o - 1.o - 1.13 (0.85-1.52) 1.22 (0.90-1.66) 1.16 (0.91-1.48) 1.15 (0.88-1.47) 1.69 (1.26-2.25) 1.58 (1.17-2.14) 1 .o - 1.12 (0.90-1.39) 1.41 (1.10-1.81) 1.76 (1.19-2.60) 1.o - 1.10 (0.87-1.38) 1.36 (1.05-1.76) 1.63 (1.08-2.46) 1.o - 1.1 8 (0.96-1.45) 1.29 (0.92-1.80) 1.79 (1.14-2.80) 1 .o - 1.09 (0.88-1.35) 1.26 (0.89-1.78) 1.68 (1.06-2.67) 1 .o 1.07 1 .OS 1.46 - (0.81-1.40) (0.83-1.43) (1.13-1.89) 1 .o 1.20 1.08 1.44 - (0.88-1.64) (0.82-1.42) (1.lo-1.88) 1.12 (1.04-1.21) 1.11 (1.02-1.21) a Years exposed out of the 40 years prior to study. Odds ratio adjusted for age, gender, log pack-years of smoking, current smoking, education, and calorie intake. Mean value. 600 Cancer Causes and Control. Vol 7. 1996 Bladder cancer and chlorination by-products cancer risk increases by 11percent with each ,000 pg/Lyears (OR = 1.11, CI = 1.02-1.21). Homogeneous water exposures Assignment of an individual's water exposures is complicated by the fact that people move often, and thereby accumulate exposures from various water suppliesduring the exposure period. In the previous analyses, exposure was quantified according to the number of years using water with different characteristics. Subjects within a 'years of exposure' level, particularly those shorter than 20 years,,have a heterogeneous mixture of exposures.For example, individuals with 10 to 19 years of exposure to THMs 2 25 pg/L could have had 19 years at 100 pg/L and 20 years at 24 pg/L, or IO years at 25 pg/L and 30 years at zero, etc.Table 3 presents the results of an analysis restricted to those who had relatively homogenous water exposures for a period of 30 or more years. Subjects exposed for 30 or more years to chlorinated surfacewater are compared with those exposed to ground water. Subjects with 30 or more years of exposure to estimated summer THM levels < 25 pg/L are used as the referent group in comparison with those having 30 or more years of exposure to estimated summer THM 2 25 and < 75 pg/L, and 2 75 pg/L. Exposure to chlorinated surface water for 30 or more years is associated with a significant increase in risk (OR = 1.39, CI = 1.09-1.79)in comparison with exposure to a ground water source.There is a trend towards increasing bladder cancer risk with homogeneous exposure to increasing summer THM level (P trend = 0.006). In comparison with those exposed to summer THM levels < 25 pg/L for 30 or more years, those exposed to levels between 25 and 74 pg/L have a 43 percent increase in risk (OR = 1.43, CI = 1.01-2.04),and those exposed to THM levels 2 75 pg/L have a 66 percent increase in risk (OR = 1.66, CI = 1.11-2.51). Risk by volume of water consumed and duration of exposure Evaluation of the combined effects of quantity of water ingested and duration of exposure to a water source with elevated THM levels (2 50 pg/L) is presented in Table 4. Water consumption was categorized into three levels based on thirds of the control distribution of consumption. The referentcategoryfor all OR estimatescomprised those using water with estimated summer THM level 2 50 pg/L for zero to nine years and consuming less than 1.54 liters of water per day. Overall, the pattern of risk estimates does not provide support for an interaction between volume of water consumed and years of exposure to THM level 2 SO pg/L (P-value interaction = 0.775) . However, statistically significant risk estimates representing more than a Table 3. Odds ratios (OR) and 95% confidence intervals (CI) for cancer of the bladder associated with exposure to water source characteristics for 30 or more of the 40 years prior to study, Ontario Water source No. of Cases No. of Controls Crude Adjusted' Ground Chlorinated surface THM level 0-24 p@L 25-74 pgJL 75+ pg/L P-value for trend OR (CI) OR (CI) 145 338 1.O - 1.o - 402 776 1.39 (1.10-1.75) 1.39 (1.09-1.79) 169 447 1.o - 1.o - 80 153 1.38 (0.99-1.94) 1.43 (1.OO-2.03) 63 97 1.72 (1.18-2.51) 1.66 (1.11-2.51) 0.002 0.006 e Odds ratio adjusted for age, gender, log pack-years of smoking, current smoking, education, and calorie intake. Table 4. Odds ratiosa for cancer of the bladder for tap water consumption and duration of exposure to summer THM 2 50 BgA, Ontario Tap water consumption e 1.54 pgA per day 1.54-2.08 > 2.08 0-9 1.o - 1.35 (0.91-2.01) 1.29 (0.87-1.90) Years of exposure to THM 2 50 pgA 10-19 20-34 1.28 (0.83-1.96) 1.32 (0.87-1.98) 1.37 (0.91-2.06) 1.70 (1.08-2.68) 1.54 (0.95-2.48) 1.72 (1.10-2.70) 35+ 1.26 (0.58-2.71) 2.58 (1.28-5.21) 2.28 (1.12-4.67) e Odds ratio (95 percent confidence interval [Cl]) adjusted for age, gender, log pack-years of smoking, current smoking, education, and calorie intake. Cancer Causes and Control. Vol7. 19% 601 W. D. King and L. D. Marrett Table 5. Odds ratios (OR)and 95% confidence intervals (CI) for cancer of the bladder for those with 30 or more years of exposure to Chlorinated surface water according to exposure THM factors, Ontario Water factor and years of exposure' No. of Cases No. of Controls Crude Adjustedb OR (CI) OR (CI) THM L 50 pg/L 0-19 yrs 20-34 35+ THM 2 75 pg/L 0-19 yrs 20-34 35+ THM-years (quartiies) 1 & 2 (0-1,857 pg/L-yrs) 3 (1,858-2,225) 4 (2,226-6,425) THM-years (continuous per 1,000 pg/L-years) 213 138 51 314 50 38 193 98 111 2.25' 470 231 75 634 88 54 420 168 188 2.18' 1.o - 1.32 (1.00-1.73) 1.50 (1.OO-2.26) 1.o - 1.15 (0.78-1.69) 1.42 (0.90-2.25) 1.o - 1.27 (0.93-1.74) 1.28 (0.95-1.74) 1.o - 1.32 (0.99-1.77) 1.45 (0.95-2.23) - 1.21 (0.80-1.82) 1.44 (0.89-2.31) 1.o - 1.27 (0.91-1.77) 1.32 (0.96-1.81) 1.07 (0.94-1.23) 1.08 (0.95-1.22) a Years exposed out of the 40 years prior to study. Odds ratio adjusted for age, gender, log pack-years of smoking, current smoking, education, and calorie intake. ' Mean value. doubling of risk are observed for those with 35 or more years of exposure who consume between 1.54 and 2.08 liters of tap water per day (OR = 2.58, CI = 1.28-5.21) or more than 2.08 litersper day (OR = 2.28, CI = 1.12-4.67). Independent effects of chlorination surface water and THM exposure Animportant aspect of exposuremodelingin this research was the ability to identify different levels of exposure to chlorination by-products within chlorinated surface water supplies. The risk associated with THM exposures among those subjects primarily exposed to chlorinated surface water in the past 40 years is presented in Table 5. For categoricalvariables, the lowest exposure levels have been collapsed to ensure that a substantial number of subjects are included in the referent category. Among those with 30 or more years of exposure to chlorinated surface water, results are suggestive of an increase in risk for those with many years of exposure to a THM level 2 50 pg/L. For those exposed for 35 or more years, therisk estimateis 1.45,but doesnot reach statistical significance(CI = 0.95-2.23). A similarsuggestivepattern of risk is observed for factors representing years of exposure to a THM level 2 75 pg/L, and THM-years. Discussion Theresults indicate an increasein bladder cancerrisk with exposure to chlorinated surface water and to estimated THM concentrations 2 25, 50, and 75 pg/L for many years. Duration of exposure seems to be an important 602 Cancer Causes and ConuoL Vol7. 1996 component of risk, as excess risk is found only after 20 or more years of exposure and the highest risks are observed for those exposed for 35 or more years. Higher estimates of bladder cancer risk are found with more specific indicators of exposure to chlorination byproducts (e.g., years with THM 2 50 pg/L compared with years exposed to chlorinated surface water). Within subjects primarily exposed to chlorinated surface water, results suggest an effect of exposure to higher concentra- tions of chlorination by-products. Results also show a trend towards increasing risk with increasing level of chlorination by-products in water. A primary contribution of this study is the estimation of historical levels of chlorination by-products in water supplies linked to individual residence histories. In addition, the large sample size, relatively high subjectresponse rates, and systematic data collection strategy add to the validity of study results. It is difficult to compare results of this study directly with those of other studies of chlorination by-products in water because of differences in the water-source characteristics of the particular populations under study and variationsin themethods of exposureassessment. In terms of design (population-based case-control; incident cases) and exposure classification (residence history linked to a survey of water treatment facilities), this study is most similar to the case-control studies of bladder cancer conducted by McGeehin et a17 and Cantor et aL9 The magnitude of risks found in our study are in general consistent with those reported by McGeehin, but are lower than Cantor's. Cantor et a19 observed the largest ' Bladder cancer and chlorination by-products increase in bladder cancer risk after 60 or more years of exposure, while in this study, availability of historical information on water treatment plants limited exposure estimation to the past 40 years. Each of these studies identified long-term exposure as an important determinant of bladder cancer risk. Our finding of strongest associationsafter 35 or more years of exposure to chlorination by-products is consistent with this. In addition, our results did not identify an interactionbetween volume of water consumed and level of exposure to chlorination by-products. Previous studies7v9have not been consistent onthis. , In the study by Cantor et ut9associations of bladder cancer risk with duration of exposure to chlorinated surface water were due mostly to effects among nonsmokers. In analyses not presented here, we found higher risk estimates for nonsmokersassociated with many years of exposure to chlorinated surface water; however, the difference in risk compared with smokers was not statistically significant. In addition, this pattern of higher risks in nonsmokers was not observed consistentlyfor factors representing exposure to chlorination by-products. Duringthe chlorinationprocess, organicmaterial reacts with chlorine to produce a number of halogenated hydrocarbon compounds.` THMs are the most common of these' and have known mutagenic and possibly carcinogenicproperties.6Inthis study,THM level inwater supplieswas estimated onthe basis of chlorinedose,water source, and treatment characteristicsof eachwater supply over time and has the strongest association with bladder cancer risk among the water exposure factors. However, THMs may not be the by-product responsible for the observed increase in bladder cancer risk. Other byproducts, such as haloacetonitriles, are also potentially carcinogenic" and their presence may be correlated highly with THMs in treated water. There are several methodologic issues relevant to the interpretation of these results. In any case-control study, an important aspect of the source of controls is its representativeness of the underlying population from which the cases arose. Our controls were restricted to households with listed telephones. However, 95 percent of a random sample of participatingcases were located in telephone directories,supportingthe validity of telephone listings database. Previous studies have identified different response patterns in control subjects according to rural or urban residence." Because of the relationship between urban/ rural residence and chlorination status of drinkingwater, and our differing response rates by case-control status, there is a concern about possible response bias. Control response rates were calculated according to designations based on postal codes of rural Ontario, metropolitan Toronto (the most highly urbanized area in Canada), and other; there were no differences among the three groups. Etiologic studies employing a case-control design typically must be concerned with the potential for differential recall between case and control subjects. In this study, such recall bias is not expected to affect results, since subjects were unaware of the specific research hypothesis under study, and exposure assessment was accomplished through a combination of subject information (primarily a residence history) and information obtained from water treatment facilities. Despite efforts at detailed individual exposure assessment, complete information on water source was unavailable for some subjects, particularly those residing for significanttime periods outside of the study area. The large sample size permitted analyses to be restricted to those for whom a large proportion of their exposure history was known and to those with relatively homogeneous exposures to specific water sources. Since the proportion of cases and controls included in these analyses were similar, a systematic bias is unlikely. While the analysis controlled for the effects of several factors, the possibility remains that results may be confounded by other factors. Of particular concern are bladder cancer risk factors which may be more common in urban areas associated with chlorinated surface-water supplies. The study ascertained subjects' occupation two years prior to interview and employment in a high risk occupationzat this time did not confoundthe relationship of interest. If the relationships identified in this study are causal, the public health impact can be described in terms of population attributable risk (PAR),which is the propor- tion of cases of disease in a population that can be attributed to exposureto the risk factor.The risk estimates associated with duration of exposure to chlorinated surface water, and to estimated summer THMlevels 2 25, and 2 50 pg/L, as well as THM-years, result in PAR estimates which range from 14 to 16 percent?' A lower PAR (eight percent) was obtained for exposure to a summer THM level L 75 pg/L. These results suggest that control of chlorination by-products in water supplies could lead to a meaningful reduction in bladder cancer incidence in Ontario. It is important to note, however, that the largest risk factor for bladder cancer is smoking, where a population attributable risk of 56 percent was calculated from t h i s study. The estimate of chlorination by-products used in this study represents the annual peak THM level for each plant. The relationship between peak and average THM level was examined using observed data from ODWSP between 1986 and 1992,and the average THM value was 76 percent of the peak value. Therefore, if average levels are considered, increases in bladder cancer risk are associated with lower average THM levels and ORs per Cancer Causes and Conuol. Vol7. 19% 603 W D. King and L. D. Marrett unit of average THM exposure are higher than reported in this paper. Disinfection with chlorine compounds is an effective and cost efficient means of ensuring that drinking-water is safe from bacteriologic contaminants. For this reason, water disinfection with chlorine should not be abandoned. However, the results of this study do suggest that disinfection practices which reduce the formation of chlorination by-products should be investigated. -Acknowledgements The authors would like t o thank those who gave their time and effort to participate in this study: the study staff; cancer patients; people who served as controls; physicians; water treatment facility staff; and staff of the Ontario Cancer Registry. We also thank Dr Yang Mao and Robert Semenciw of Health Canada, and Dr Andy Gilman of the Great Lakes Health Effects Program, for their support throughout this study. References 1. White GC. Handbook of Chlorination. New York, NY (USA): Van Nostrand Reinhold, 1986:256. 2. Federation of Associations on the Canadian Environment. National inventory of municipal waterworks and wastewater systems in Canada 1986. Ottawa, Ontario (Canada): Minister of Supply and Services Canada 1987. 3. Stevens AA. Moore LA. Slocum CT. Smith BL. Seeger DR. "I Ireland JC. By-products of water chlorination at ten operating utilities. In:Jolley R, Condie L, Johnson JD. et al, eds. Water Chlorination:Chemistry, EnvironmentalZmpact and Health Effects. Chelsea, Michigan (USA): Lewis Publishers, Inc, 1990 6:579-604. 4. International Agency for Research on Cancer. Chlorinated drinking-water; chlorination by-products; some other halogenated compounds; cobalt and cobalt compounds. Lyon, France: IARC, 1991; IARC Monogr Eva1 Carcinog Risk Hum, Vol. 52. 5. Cantor KP. Epidemiologic studies and risk assessment of volatile organic compounds in drinking water. In: Ram NM, ChristmanRF, Cantor KP,eds. Significanceand Treatment of Volatile Organic Compounds in Water Supplies. Chelsea, MI (USA): Lewis Publishers, 1990: 465-84. 6. Bull RJ, Robinson M, MeierJR, Stober J. Use of biological assay systems to assess the relative carcinogenichazards of disinfection by-products. Environ Health Perspect 1982; 4 6 215-27. 7. McGeehin MA, Reif JS, Becher JC, Mangione EJ. Casecontrol study of bladder cancer and water disinfection methods in Colorado. A m ] Epidemioll993; 138: 492-501. 8. Zierler S, Feingold L, Damley RA, CraunG. Bladder cancer in Massachusetts related to chlorinated and chloraminated drinking water: A case-control study. Arch Environ Health 1988; 43: 195-200. 9. Cantor KP, Hoover R, Hartge P, et al. Bladder cancer, drinking water source and tap water consumption: a casecontrol study.JNCZ 1987; 79: 1269-79. 10. Wilkins JR, Comstock GW. Source of drinking water at home and site specific cancer incidence in Washington County, Maryland. Am J Epidemiol 1981; 114: 178-90. 11. Gottlieb MS, Carr JK, Clarkson JR. Drinking water and cancer incidence in Louisiana. A m J Epidemiol 1982; 116: 652-67. 12. Young TB, Kanarek MS, Tsiatis AA. Epidemiologic study of drinking water chlorination and Wisconsinfemale cancer mortality. JNCZ 1981; 67: 1991-98. 13. Brenniman GR, Lagos J, Amsel J. Case-control study of cancer deaths in Illinois communities served by chlorinated and non-chlorinated water. In: Jolley RL, Brings W,Coamings RB, eds. Water Chlorination Environmental Impact and Health Eflects. Ann Arbor, MI (USA): Ann Arbor Science Publishers, 1980 3: 1043-57. 14. Alavanja M, Goldstein I, Susser M. A case-control study of gastrointestinal and urinary tract cancer mortality and drinking water chlorination. In: Jolley RL, Brings W, Coamings RB, eds. Water Chlorination Environmental Impact and Health Effects. Ann Arbor, MI (USA): Ann Arbor Science Publishers, 1980 3: 395-409. 15. Morris RD, Audet AM, Angelillo IF, Chalmers TC, Mosteller E Chlorination, chlorination by-products, and cancer: A meta-analysis. A m J Public Health 1992; 82: 955-63. 16. Lynch CF, Woolson RF, O'Gorman T, Cantor KP. Chlorinated drinking water and bladder cancer: effect of rnisclassification on risk estimates. Arch Environ Health 1989; 44: 252-9. 17. World Health Organization. International Cksifcation of Diseases, Ninth Reviswn. Geneva, Switzerland: WHO, 1977. 18. Robles SC, Marrett LD, Clarke EA, Risch HA. An application of capture-recapture methods to the estimation of completeness of cancer registration. J Clin Epidemwll988; 41: 495-501. 19. Breslow NE, Day NE, eds. Statistical Methods in Cancer Research, Voll. TheAnalysis nf Case-controlStudies. Lyon. France: International Agency for Research on Cancer, 1980; IARC Sci. Pub. No. 32. 20I Bull RJ, Meier JR, Robinson M, e t al. Evaluation of mutagenic and carcinogenic properties of brominated and chlorinated acetonitriles, by-products of chlorination.Fundam Appl Toxicoll985; 5: 1065-74. 21. Marrett LD, Kreiger N, Dodds L, Hilditch S.The effect on response rates of offeringa small incentive with a mailed questionnaire. Ann Epidemwll992; 2: 745-53. 22. SiemiatyckiJ, Dewar R, Nadon L, Gtrin M. Occupational risk factors for bladder cancer: Results from a case-control study in Montreal, Quebec, Canada. A m J Epidemioll994; 140: 1061-80. 23. Bruzzi P, Green SB, Byar DP, et al. Estimating the population attributable risk for multiple risk factors using case-control data. Am J Epidemwl 1985; 122: 904-14. 604 Cancer Causes and Control. Vol7. 1996