(i) Coagulation, sedimentation and filtration

Coagulation: Some impurities in natural waters cannot be removed by settling alone, either because they are dissolved or because they occur in a very finely divided (‘colloidal’) state. The addition of a chemical coagulant is needed to create large particles that can settle, called ‘floes’. The coagulants most commonly used are aluminium and ferric salts. When these chemicals are added, a precipitate of the metal hydroxide forms which removes suspended solids, algae and colour by a number of mechanisms, including adsorption and trapping. Mechanical or hydraulic mixing causes the hydroxide precipitate, together with impurities, to agglomerate into flocs a few millimetres in diameter. Other chemicals, called polyelectrolytes, can be used in addition to, or in place of, aluminium or iron coagulants to produce stronger or larger flocs. Once formed, the flocs are removed from the water by filtration, generally preceded by sedimentation.

Sedimentation and flotation: Sedimentation is used to remove the bulk of the flocs, so as to reduce the load on downstream filters. Sedimentation may take place in rectangular or circular, horizontal basins in which discrete settling of flocs occurs or, commonly in some European designs, in ‘floc blanket’ clarifiers, in which the water flows upwards through a fluidized bed of flocs and treated water is taken from the top of the clarifier. Flocs have a density only marginally greater than water, so treatment rates must be low. Typical tank loadings are < 1–5 m/h.

An alternative process to sedimentation is dissolved air flotation. In this process, water saturated with air under pressure is released into the water containing flocs, and tiny air bubbles become attached to the flocs and float them to the surface of the water. This is a faster process than sedimentation; typical loadings being 5–12 m/h. Dissolved air flotation may be particularly suitable for the treatment of coloured, low-turbidity waters and algal-laden waters.

Filtration: Deep-bed filtration through sand is employed to remove the remaining particulate matter. Water is passed through a bed of sand, typically composed of grains 0.5–1.0 mm in diameter, one-metre deep. Particles are trapped within the bed by a variety of mechanisms including straining, sedimentation, interception and electrostatic adhesion. Filtration rates are typically 4–10 m/h. As particles are trapped within the bed, the resistance to flow increases, necessitating a greater head of water (pressure) to maintain a constant rate of flow. Once a limiting head loss is reached, or solids start to be released from the filter, the filter is cleaned by backflushing with clean water.

In the treatment of turbid waters, filtration is almost always preceded by sedimentation, and filters are of the open ‘gravity filter’ type. With some low turbidity waters, including coloured moorland waters, the sedimentation stage may be omitted and direct filtration employed. With direct filtration, pressure filtration can be used to conserve a hydrostatic head.

(ii) Slow sand filtration

Slow sand filtration, which is a well-established process, is an alternative to the coagulation process for waters with little colour and a moderately low concentration of suspended solids. A slow sand filter consists of a 0.5–1.5-m-deep bed of fine (0.15–0.35 mm) sand, supported on a layer of gravel by a system of underdrains. At the low flow rates used (0.1–0.3 m/h), solids settle onto the surface of the sand. The layer formed, known as the ‘Schmutzdecke’, contains mud, organic waste, bacterial matter and algae and is biologically active. The mechanisms involved in slow sand filtration are: removal of colloidal material by straining, adsorption and bacterial action; destruction of pathogenic organisms by bacterial action; and purification of the water above the filter by bacterial action, flocculation and pathogen death.

As filtration progresses, the head loss through the bed increases to the point at which the required flow rate cannot be maintained. The filter is then taken out of action and the top layer is skimmed off manually or mechanically. The sand is washed for re-use. Eventually the depth of sand in the filter becomes insufficient for effective filtration, and more sand is added.

(iii) Other processes

A number of other processes may be employed prior to disinfection of water; these processes are applicable to groundwaters as well as surface waters.

Aeration may be employed for a variety of reasons, including removal of volatile taste- and odour-producing compounds, precipitation of iron and manganese and removal of carbon dioxide.

Oxidation may be used for purposes other than disinfection; these include precipitation of iron and manganese, taste and odour control, colour removal and oxidation of trace organic compounds. The principal oxidizing agents employed in water treatment are chlorine, chloramine, ozone and chlorine dioxide (White, 1986).

pH Adjustment, usually to more alkaline levels, is used to achieve optimal values for other processes, including coagulation and disinfection, as well as to reduce the corrosiveness of the water supply. pH can be increased by adding chemicals such as lime, caustic soda or soda ash or by placing the water in contact with a bed of sparingly soluble material, such as marble.

The pH of drinking-water is typically in the range 6.5–8.5, but levels up to 9.5 can occur.

Softening. Hardness in water results from the presence of calcium and magnesium compounds. When hardness is excessive, it can be reduced by precipitation softening or ion exchange. In precipitation softening, lime (and sometimes soda ash) is added to precipitate calcium as calcium carbonate, which is removed in a sedimentation tank. Ion-exchange softening is used only for groundwaters; the water is passed through a bed of cationic resin which exchanges sodium for calcium and magnesium. When the resin is fully loaded with calcium and magnesium, it is regenerated using a strong brine solution.

Activated carbon may be employed to remove natural and synthetic organic chemicals. It is produced by the controlled combustion of wood, coal and other material to produce a porous material with a large surface area and a high affinity for organic compounds. A slurry of powder can be added to the water and then removed by subsequent treatment processes, such as coagulation. Alternatively, granular-activated carbon can be employed in purpose-built adsorbers, or as a replacement for some of the sand in a rapid gravity filter.

(i) Trihalomethanes

Trihalomethanes were shown to be present in drinking-water as a result of chlorination (Bellar et al., 1974; Rook, 1974) using purge and trap and solvent extraction methods of concentration, followed by GC with electron capture detection (ECD; Croll et al., 1986). Subsequently, a number of analytical methods for the determination of trihalomethanes in drinking-water have been published (for review, see Croll et al., 1986); they include direct aqueous injection (Nicholson et al., 1977; Peters, 1980; Grob & Habich, 1983), liquid-liquid extraction (Dressman et al., 1979; US Environmental Protection Agency, 1979a; Standing Committee of Analysts, 1980), purge and trap (Bellar et al., 1974; Dressman et al., 1979; US Environmental Protection Agency, 1979b) and headspace analysis (Rook, 1974; Otson et al., 1979; Croll et al, 1986) with separation and detection by GC-ECD. More information is given in the respective monographs about the analysis of bromodichloromethane, chlorodibromomethane and bromoform.

(ii) Halogenated acetic acids

Halogenated acetic acids are common by-products of water chlorination. Most of the analytical methods involve extraction into a solvent at low pH (0.5–2) with addition of sodium chloride to salt out the substances, derivatization and then detection by GC-ECD (Krasner et al., 1989; Uden & Miller, 1983; Lahl et al., 1984), GC with microwave plasma detection (Miller et al., 1982) or isotope dilution MS (Norwood et al., 1986).

(iii) Halogenated acetonitriles

Dichloroacetonitrile and other halogenated analogues have been determined in drinking-water by solvent extraction with salting out using sodium chloride or sodium sulfate followed by GC-ECD (Oliver, 1983; Italia & Uden, 1988; Krasner et al., 1989). More information on analytical methods for halogenated acetonitriles is given in the monograph.

(iv) Chlorophenols

Chlorophenols are well-known chlorination by-products, since they can confer objectionable tastes and odours in the supply. A variety of techniques have been developed for their analysis, which usually involve derivatization to methyl, acetyl or pentafluorobenzoyl derivatives, followed by GC-ECD (Renberg, 1981; Abrahamsson & Xie, 1983; Standing Committee of Analysts, 1985, 1988) or, in some cases, GC-MS with specific-ion monitoring (Sithole & Williams, 1986).

(v) Chlorouracil, chlorouridine, chlororesorcinol and chlorosalicylic acid

These unchlorinated substances occur in natural waters and can become chlorinated during water treatment. They have been determined after freeze-drying or vacuum evaporation, extraction with methanol and examination by high-performance liquid chromatography with confirmation by GC-MS (Crathorne et al., 1979).

(vi) Organic chloramines

Chlorine can react extensively with organic amines, amino acids and related substances in water to produce chloramines. Specific analysis of organic chloramines in drinking-water is difficult, and, consequently, there is little detailed information on their presence and concentrations. In recent years, some specific methods have appeared which are based on derivatization followed by high-performance liquid chromatography with fluorescence detection (Scully et al., 1984) or ultraviolet/electrochemical detection (Lukasewycz et al., 1989).

(i) Toxic effects

Organic material recovered from chlorinated water by reverse osmosis (reduced in volume by 100 and 400 times) and given to 10 male and 10 female CD-1 mice in each experimental group as drinking-water was compared in a 30-day study with non-disinfected water and water treated with other disinfectants. A significant increase in liver weights was reported in female but not male mice given the high dose of chlorinated water concentrate in comparison with the concentrate from non-chlorinated control water. Male mice had reduced lung weights at both doses and decreased testicular weights at the high dose. No histological examination was performed (Miller et al., 1986).

Organic chemicals from the same waters, recovered on XAD resin, were administered as 0.3 ml of a 1000- or 4000-times concentrate by gavage three times per week for four weeks to 10 CD-1 mice of each sex per group. Treatment had no effect on organ weights in animals of either sex; however, water that had been chlorinated, filtered through granular activated carbon and then rechlorinated increased liver weight in male mice at both doses and decreased lung weight at the high dose. Both doses reduced ovary weight in female mice. No histological finding was reported (Miller et al., 1986).

A combined acid and neutral fraction of organic chemicals recovered on XAD resin, dissolved in dimethyl sulfoxide (corresponding to 100 1 of chlorinated drinking-water), was administered intraperitoneally on two consecutive days to 10-day-old Wistar rats and once to 20-day-old rats. This treatment resulted in 50% mortality at 48 h in 10-day-old rats and in 30% mortality in 20-day-old rats; it also induced various alterations in drug metabolizing enzyme activities in liver fractions obtained from surviving animals. The most consistent effect was an increase in the level of hepatic 7-ethoxyresorufin-O-deethylase compared to solvent-treated controls (Liimatainen et al., 1988). [The Working Group noted that no non-chlorinated water control was available.]

In a 90-day study in groups of 15 rats, humic acids dissolved at concentrations of 0.1, 0.5 and 1 g/l in distilled water and chlorinated with a 1:1 ratio of chlorine equivalents to organic carbon were given as drinking-water (pH 3). Renal weights were increased relative to body weight at 0.5 and 1 g/l, and there was a small increase in blood urea nitrogen. Haematuria was seen with the 1.0 g/l dose, which appeared to be related to the deposition of crystals [composition unspecified] in the renal pelvis (Condie et al., 1985).

Chlorine reacts very rapidly with purified DNA and RNA (Hayatsu et al., 1971). It chlorinates uracil to produce 5-chlorouracil at low chlorine:carbon ratios and dichlorouracil and ring cleavage at higher ratios (Gould et al., 1984). Relatively stable organic chloramines are formed with cytosine. Purines (modelled by caffeine) are chlorinated to a very small extent, with ring cleavage to a complex array of products (Gould & Hay, 1982). A peroxide of adenosine 5′-monophosphate (AMP) has been shown to form at physiological pH, and this reaction is dependent on the NaOCl:AMP ratio, reaching a plateau when this reatio is less than 1 (Bernofsky et al., 1987).

(ii) Effects on reproduction and prenatal toxicity

McKinney et al. (1976) reported an increased incidence of dead implantations and of litters with malformed fetuses among Swiss CD-1 mice given chlorinated tap water from Durham, NC, USA. The effect was reported to be seasonal [data not shown]. The control group in this study was given the same water purified by filtration (to reduce organic material and remove microparticulates), demineralization and distillation.

These observations stimulated a series of teratogenicity studies in which Durham city tap water was compared with water purified by the same method as described above. Using much larger group sizes than McKinney et al. (1976), Staples et al. (1979) found no significant overall difference in the reproductive status of pregnant mice given tap water or purified water. Month-by-month comparisons over a nine-month period (including the critical winter months suspected of being important by McKinney et al) indicated occasionally improved reproductive performance only in the tap water group. Chernoff et al. (1979) also found no significant effect on any fetal parameter in CD-1 mice, except for an increased incidence of supernumerary ribs, which the authors considered to be spurious, in the groups given Durham tap water. They considered the possibility that drinking-water quality had changed during the intervening years since the study by McKinney et al.

Kavlock et al. (1979) evaluated the effects of concentrates of organic materials from the drinking-water of five US cities representative of major sources of raw water. Because the reverse osmosis method used for concentrating these materials does not retain organohalides with a molecular weight of less than 200, an artificially constituted organohalide mixture was also prepared and evaluated. Groups of Swiss mice were given 300, 1000 or 3000 times the anticipated human dose of these materials by gavage on gestation days 7–14. No adverse effect on embryonal or fetal development was observed.

[The Working Group noted that these studies were designed to study the effects of the drinking-water of individual cities but not to investigate the developmental toxicity of chlorinated drinking-water, nor did they include a non-chlorinated water control.]

(iv) Genetic and related effects

The results obtained in a variety of short-term tests for samples of chlorinated water have been reviewed (Kraybill, 1980; Loper, 1980; Alink, 1982; Kool et al., 1982a; Nestmann, 1983; Bull, 1985; Degraeve, 1986;Fielding & Horth, 1986; Meier, 1988). Many of the studies were concerned with the mutagenicity of drinking-water and not with the influence of chlorination. As the source of mutagenicity in drinking-water may also be polluted raw water, the role of water chlorination cannot be evaluated unless a comparison is made with an unchlorinated sample. Papers lacking this aspect and those in which no reference is made to the disinfection agent used are not summarized here. In many papers, data were available to allow comparison of unchlorinated and chlorinated waters, and when the authors did not do this, the Working Group drew their own conclusions.

The Working Group also limited themselves to studies of water samples disinfected with chlorine or hypochlorite; studies on water samples disinfected with chlorine dioxide, monochloramine or ozone alone were not considered.

By far the majority of studies were with Salmonella typhimurium strains TA98 and TA100.

(1) Chlorinated water (Table 7)

Table 7.

Summary of the influence of chlorine disinfection on the genetic and related effects of unconcentrated drinking-water samples in comparison with unchlorinated water.

Chlorination did not increase the mutagenicity of drinking-water prepared from surface or spring water, as studied in fluctuation tests with S. typhimurium strains TA98 and TA100. Mutations were not induced in S. typhimurium TA100 when samples of chlorinated water were used to prepare bottom agar for the test plates. [The Working Group noted that volatile substances would be lost if the water were autoclaved.] Chlorination of a tap water sample derived from surface water did not increase the number of micronuclei in Tradescantia pollen mother cells. Chromosomal aberrations were induced in Allium cepa, however, by a river water sample chlorinated in the laboratory. Cell transformation was not induced by chlorinated tap water in cultured Syrian hamster embryo cells or by finished drinking-water from a surface water source in mouse embryo cells.

(2) Concentrates of chlorinated water (Tables 8–10)

The most widely used method for isolating organic material from water samples is adsorption to macroreticular resin (various types of Amberlite XAD) followed by elution with an organic solvent. Liquid-liquid extraction with an organic solvent is also commonly used. The different concentration methods used for mutagenicity studies have been discussed (Forster & Wilson, 1981; Harrington et al., 1983; Maruoka & Yamanaka, 1983; Monarca et al., 1985a,b; Wigilius et al, 1985; Vartiainen et al, 1987a). The methods are more or less selective and do not concentrate all organic materials, e.g., XAD adsorption and liquid-liquid extraction techniques may result in the loss of highly polar compounds. Concentration of an extract invariably means that volatile substances are removed with the solvent. The extent of loss depends upon the solvent used: use of low-boiling-point solvents, e.g., ether and acetone, leads to smaller losses (compounds with boiling-points of about 120°C should be retained unless evaporation is to dryness), while use of high-boiling-point solvents, e.g., ethyl acetate and dimethyl sulfoxide, leads to greater losses.

Chlorination of surface water usually resulted in increased mutagenicity of concentrated samples towards S. typhimurium, particularly strains TA100 and TA98 (Tables 8 and 10). In the few studies in which these strains were not used, negative responses were obtained, perhaps because the most sensitive organism was not used. Inclusion of a metabolic activation system usually resulted in a reduced response or totally abolished it. Mutagenic effects were consistently found in samples of surface water that had a high content of natural organic compounds at the time of the chlorination. Water samples in which the organic content had been reduced before chlorination by water treatment procedures tended to show reduced or no mutagenicity.

Chlorinated ground- and spring water samples (Table 9) were less frequently mutagenic than chlorinated surface water samples (Table 10).

Table 9.

Summary of the influence of chlorination upon the genetic and related effects of groundwater and spring water concentrates in comparison with concentrates of unchlorinated water.

Table 10.

Summary of the influence of chlorination in combination with either chlorine dioxide or ozone treatment upon the genetic activity of surface water concentrates in comparison with concentrates of unchlorinated water.

Much of the bacterial mutagenicity of concentrated chlorinated surface water samples is probably due to chlorination of natural constituents, such as humic and fulvic acids. Chlorination of aqueous solutions of fulvic and humic acids resulted in the formation of mutagenic compounds (Meier et al., 1983; Kowbel et al., 1984; Kopfler et al, 1985; Kronberg et al., 1985a,b; Meier et al., 1985; Kowbel et al., 1986; Maruoka, 1986; Meier et al., 1986; Van Duuren et al., 1986; Agarwal & Neton, 1989; Horth, 1989; Pommery et al., 1989). The mutagenicity of chlorinated water samples is not due to the volatile trihalomethanes known to be formed at chlorination; much of the mutagenicity is due to nonvolatile acidic and polar substances. Such compounds require acidic conditions for efficient extraction by non-polar solvents. In several studies, the greatest mutagenic activity was seen when concentration was performed at low pH (e.g., pH 2) (Kool et al., 1981; Van Der Gaag et al., 1982; Kronberg et al, 1985a,b; Vartiainen & Liimatainen, 1986; Ringhand et al., 1987; Fawell & Horth, 1990). A single organic compound, MX, has been shown to be responsible for a significant portion of the bacterial mutagenicity of some concentrated chlorinated surface water samples. This compound is unstable at high pH and in dimethyl sulfoxide (Meier et al, 1987; Kronberg & Vartiainen, 1988; Fielding & Horth, 1988).

Some concentrates of chlorinated tap water prepared from surface waters, groundwater or their mixture induced more sister chromatid exchange in Chinese hamster ovary cells than concentrates of the respective raw waters. In the only study of its kind, concentrates of chlorinated river water that had undergone extensive water treatment procedures did not increase the incidence of hprt locus mutations in Chinese hamster V79 cells. Chlorination was associated with an increase in the frequency of micronuclei in Chinese hamster ovary cells exposed to some samples of concentrated chlorinated tap water prepared from surface water and mixed ground- and surface water but not in those exposed to concentrated chlorinated groundwater. Concentrates prepared from chlorinated water from a river and a reservoir induced chromosomal aberrations in Chinese hamster ovary cells.

No studies in mammals in vivo were available.

Table 8.

Summary of the influence of chlorine disinfection on the genetic and related effects of surface water concentrates in comparison with concentrates of unchlorinated water.

(i) Toxic effects

The possible effect of drinking-water on serum lipids was examined in a cross-sectional study of about 1500 healthy persons who had resided for at least 10 years in 46 different communities in Wisconsin, USA. Alcohol consumption, smoking habits, dietary fat intake, dietary calcium intake and body mass were considered in the analyses. The public water supply varied in magnesium and calcium content (water hardness), whether it had been chlorinated or not. The prevalence of women whose serum cholesterol level exceeded 270 mg/dl was greater in communities served by chlorinated drinking-water (odds ratio, 2.0); the mean serum cholesterol concentrations in women on chlorinated and on nonchlorinated drinking-water supplies were 247.9 mg/dl and 239.8 mg/dl, respectively (a significant difference). A smaller difference in men was not significant. Water hardness did not influence the serum cholesterol levels in either women or men (Zeighami et al., 1990).

(ii) Effects on reproduction and prenatal toxicity

Rausch (1980) evaluated pregnancy outcomes during 1968–77 in several villages in New York State, USA, served by nonchlorinated groundwater, chlorinated groundwater or chlorinated surface water supplies. A significantly greater incidence of late fetal deaths and neonatal deaths was observed in villages on nonchlorinated groundwater; and a significantly higher prevalence of anencephaly was seen in villages using surface water as compared to villages using groundwater. There was, however, no difference in the prevalence of anencephaly between villages using chlorinated groundwater and those using nonchlorinated groundwater. The confounders analysed were season and year of birth, sex and cause of death of the fetus or newborn, maternal age, education, previous reproductive history and prenatal care and hospital where delivery occurred.

In a retrospective study described in detail in the monograph on sodium chlorite, Tuthill et al. (1982) compared neonatal morbidity and mortality in two similar communities in the USA, one of which used chlorination and the other of which used chlorine dioxide for disinfecting potable water. The number of infants that were judged by the attending physician to be premature or to have greater weight loss after birth was significantly greater in the community with chlorine dioxide-treated water. [The Working Group noted the difficulties associated with establishing prematurity and poor weight gain after birth, especially in a retrospective study, and that confounding factors were not controlled for.]

In a case-control study of spontaneous abortions in relation to tap-water consumption in northern California, USA, Hertz-Picciotto et al. (1989) observed a crude odds ratio of 1.7 (95% confidence interval (CI), 1.2–2.3) for drinkers of tap water as compared with drinkers of bottled water. After controlling for a large number of confounders, including demographic, reproductive and life-style variables, the results were still significant.

(iv) Genetic and related effects

No data were available to the Working Group.

(i) Surface versus groundwater

In the studies described below, the surrogate measure of exposure to chlorinated water was exposure to surface water, although the status of neither the surface nor the groundwater was known.

Page et al. (1976) studied 64 parishes in Louisiana, USA, in which 32% of the population were supplied by the Mississippi River, 56% by groundwater and 12% by other surface supplies. Age-adjusted 20-year mortality rates (Mason & McKay, 1974) for cancers of the gastrointestinal tract, urinary tract (these two groupings were necessary owing to small numbers of deaths from cancers at related individual sites in some counties), breast and prostate and for cancers at all sites were analysed by multiple regression, including as independent variables the percentage of the parish population drinking Mississippi River water, rurality, income, and occupation in the petroleum and coal, chemical and mining industries, in four subgroups: white men, white women, non-white men and non-white women. The proportion of the parish population using Mississippi River water was positively (p < 0.05) associated with cancer of the gastrointestinal tract in all four groups, for urinary tract cancer in white men and non-white women and for cancers at all sites in white men, non-white men and non-white women. [The Working Group noted that the parishes using Mississippi River water were all located in the southern part of Louisiana, and the possible effects of water supply type and social and cultural differences cannot be separated.]

A further analysis of these and other data (DeRouen & Diem, 1977) also took account of region, i.e., northern or southern Louisiana, to allow for cultural and occupational differences. In a multiple regression analysis, percentage use of Mississippi River water was significantly associated with mortality from cancers at all sites for white men, non-white men and non-white women, with gastrointestinal tract cancer mortality for non-white men and non-white women, with urinary tract cancer for non-white women, and with cancer of the lung for non-white men. Southern Louisiana parishes in which part of the population was supplied with Mississippi River water showed significantly higher mortality from all cancers for non-white women, cancer of the stomach for non-white women, cancer of the colon for both white and non-white women, cancer of the rectum for white men, cancer of the urinary bladder for white men and cancer of the lung for non-white men. In these parishes, there was significantly lower mortality from cancer of the lung in white women and from cancer of the liver in white men.

Kuzma et al. (1977) classified 88 Ohio (USA) counties by ground- or surface water source on the basis of a survey of water supplies conducted in 1960. A substantial proportion of the population was served by sources not included in the survey: in only 39 counties was more than 50% of the population on a water source that was covered. Average annual age-adjusted cancer mortality rates among whites from 1950 to 1969 (Mason & McKay, 1974) were obtained for cancers of the stomach, large intestine, rectum, biliary passages and liver, pancreas and urinary bladder; for all cancers at all sites; for lung cancer in men; and for breast cancer in women. Analysis of covariance included the water classification variable as a factor and percentage urbanization, median income, population size and percentage of the male population in manufacturing activity and agriculture-forestry-fishery activity as covariables. Adjusted mean mortality in counties classified as supplied by surface water significantly (p < 0.05) exceeded that in those supplied by groundwater for cancers at all sites combined in men, for stomach cancer in both men and women and for bladder cancer in men. When the 39 counties in which more than 50% of the population was on a water source covered by the survey were analysed, similar results were obtained. [The Working Group noted that in the absence of information on county population size and water source, it was impossible to confirm the adequacy of covariance analysis to control for population size.]

Bean et al. (1982) compared cancer incidence rates in Iowa (USA) municipalities served by water from surface sources with those from groundwater sources. They omitted municipalities with populations of fewer than 1000 and those in which less than 90% of the water used was from the classified source. Only municipalities receiving water from a single source type in 1965–79 were included. Cancer incidence data were obtained for 1969–71 and 1973–78, from the Third National Cancer Survey and from the Surveillance, Epidemiology and End Results Program, respectively. Age-standardized, sex-specific incidence rates were calculated for cancers of the urinary bladder, breast, colon, lung, rectum, prostate and stomach. Details of socioeconomic status and occupation were obtained from the 1970 census and from the Directory of Iowa Manufacturers. Using a previously conducted population-based case-control study of urinary bladder cancer which included subjects in Iowa (Hoover & Strasser, 1980), the authors derived information on several variables, including education, income, manufacturing, labour force, change in population between 1960 and 1970 and smoking habits, for residents of towns on groundwater and on surface water. The case-control study had shown that 63% of the controls over 55 years of age had been on the same water supply for at least 20 years before onset of their cancer, and 77% had been on the same supply for at least 10 years before onset. Analyses were based on log-linear models. After adjustment for population size, the incidences of lung cancer and rectal cancer were significantly greater [details not given] for men and women served by surface water than for those drinking groundwater. Trends of risk over three categories of well depth were not significant.

Kool et al. (1981) studied 19 cities in the Netherlands, representing approximately one-third of the population of that country. Directly standardized, sex-specific mortality rates for cancers of the bladder, lung, oesophagus, stomach, colon, rectum and liver were calculated for 1964–76. Organic constituents of tap water were determined in 1976. Correlation coefficients between source type (surface/groundwater) were calculated, and a transformation of the rates showed that mortality from liver and urinary bladder cancer in men and lung cancer in both men and women was significantly greater (p < 0.05) in cities supplied with surface water. [The Working Group noted that no potential confounding factor was taken into account, and the statistical methods were not adequately described.]

(ii) Chlorination and chlorination by-products

Cantor et al. (1978) calculated directly age-standardized, sex-specific, cancer mortality rates by site for whites for 1968–71 in 923 US counties with more than 50% urbanization in 1970. Chloroform and total trihalomethane (THM) levels were obtained from two drinking-water surveys carried out in 1975 by the US Environmental Protection Agency, and levels of bromine-containing trihalomethanes (BTHM) were calculated by subtraction. The proportion of each county served in 1960 by the sampled municipal water supplies was estimated. The correlation between chloroform and BTHM levels was 0.54. Weighted linear regression using all 923 counties was used to predict sex- and site-specific cancer rates in 1970 by including the following variables in the model: urbanization (%; 1970), education (1970), population size (1970), ratio of 1970:1950 population, workforce in manufacturing (%; 1970), population in each of 10 ethnic groups (%) and region. The differences between the observed and predicted values (residuals) were correlated with log-THM in the 76 counties where 50% or more of the population was served by a water supply included in either of the two surveys and in the 25 counties where 85% or more of the population was so served. All cancer sites for which the sex-specific mortality rate was more than 1.5/l0⁵ per year were studied. In the analysis of the 76 counties, the only significant correlation found was between the residual mortality rates for lung cancer in females and level of total THM (correlation coefficient, r = 0.22; p = 0.05). In the analysis of the subset of 25 counties, there were significant correlations between kidney cancer in men and chloroform level (r = 0.42, p = 0.04) and between urinary bladder cancer in women and BTHM level (r = 0.45, p = 0.02); whereas the correlations for kidney cancer in women and lung cancer in men were very low or negative, the correlation coefficient for male urinary bladder cancer and BTHM level was 0.38 (p = 0.06). Partial correlations controlling for high-risk occupation were calculated for cancers of the urinary bladder and lung. After allowance for lung cancer mortality [presumably a proxy for cigarette smoking], the partial correlations of urinary bladder cancer with log-BTHM level in counties in which 85–100% of the population was served by sampled supplies were 0.33 and 0.42 for men and women, respectively. Adjustment for occupational exposures left the correlations unchanged.

Hogan et al. (1979) used cancer mortality rates in US counties for 1950–69 (Mason & McKay, 1974) for white men and women in a multiple regression analysis. The cancers considered were of the tongue, oesophagus, stomach, large intestine, rectum, biliary passages and liver, pancreas, breast, ovary, kidney, urinary bladder and other urinary, thyroid and bone and cancers at all sites. Data on exposure to chloroform were taken from the two surveys carried out by the US Environmental Protection Agency in 1975 and referred to by Cantor et al. (1978). Weighted and unweighted analyses were carried out, which included the following independent variables: population density, percentage of urbanization, percentage of non-white, percentage of foreign born, median income, education, percentage in manufacturing industry, population size (all in 1960), region and chloroform in finished water. There were substantial differences in the results of three analyses based on different methods of weighting the units of observation. Consistent positive associations were found with chloroform exposure level in men and women in all analyses (with at least one significant result, p < 0.05) for cancers of the urinary bladder, breast, rectum and large intestine using data for the counties covered in the first survey and for cancers of the liver and tongue using data from the second survey. A significant negative association was obtained for pancreatic cancer using data from the second survey. Finally, in an analysis restricted to counties in which 50% or more of the population was on a sampled supply, significant associations were found for cancer of the large intestine in both men and women and for urinary bladder cancer in women. [The Working Group noted that the geographical areas and exposure measures were similar to those used by Cantor et al. (1978), and these results, therefore, do not provide independent evidence.]

Carlo and Mettlin (1980) obtained age-adjusted cancer incidence rates from the New York State Tumor Registry for 218 census tracts in Erie County, NY, USA, between 1973 and 1976. Nine census tracts with rates greater than three standard deviations from the mean or large institutions were excluded; cases with incomplete residence data were excluded. The cancer sites studied were oesophagus, stomach, colon, rectum, urinary bladder and pancreas; socioeconomic factors, mobility, percentage of non-white, urbanicity and occupation (only for bladder cancer) were controlled for. Total THM, derived from State records, and type of water source were included in the analysis. Use of surface water was significantly (p < 0.05) associated with the incidence of oesophageal and pancreatic cancer; total THM was not significantly related to any cancer site studied. [The Working Group noted that the quality of data on THM levels could not be assessed; it is probable that a number of neighbouring census tracts shared the same water supply; and the statistical procedure used was unclear.]

Tuthill and Moore (1980) studied communities in Massachusetts (USA) served by surface water in 1949, excluding those with a population of fewer than 10 000 persons in 1970 or with a growth rate exceeding 25% between 1950 and 1970. They calculated sex-specific standardized mortality ratios (SMRs) for 1969–76 for nine digestive and urinary tract cancers and ten other cancer sites thought unlikely to be related to water quality. Correlations were made between SMRs and three measures of water quality: average past (1949–51) chlorine dose, recent chlorine dose and recent total THM level. Data analysis included correlation and stepwise multiple regression. Potential confounding variables included were ethnic group, income, education, percentage of foreign-born, occupation in the textile, printing and chemical industries and population growth betweeen 1950 and 1970. There was no significant correlation between sex-specific SMRs and average chlorine dose in 1949–51. For recent chlorine dose and recent total THM level, there was a significant (p < 0.05) positive association with stomach cancer for women and for rectal cancer for men. For recent total THM level, there was a significant negative association with stomach cancer for men. After multiple regression analysis allowing for sociodemographic factors, the significant associations disappeared. [The Working Group noted that the number of communities studied was not explicitly stated and the methods of analysis were not fully presented.]

In the study by Kool et al. (1981) (described on p. 110), no significant relationship was found between levels of THM and cancer mortality in 19 cities in the Netherlands.

Isacson et al. (1983) extended the analysis of their earlier study (Bean et al, 1982; see p. 110) to include the chlorination status of the water supply (chlorinated prior to 1966 or never chlorinated). Directly standardized incidence rates for rectal cancer in men were significantly lower in municipalities with chlorinated water in both categories of well depth (< 150 feet and >150 feet [ < or >45.7 m]), but this difference was no longer significant after adjustment for potential confounding by other methods of water treatment (aeration, filtration, coagulation and sedimentation). When water sources were classified by chloroform content (0–96, 100–230 and 260–900 μg/l), nonsignificant increases were observed across these levels for cancers of the colon, rectum and urinary bladder in men and cancers of colon and rectum in women. [The Working Group noted that no test for trend with increasing chloroform level was presented.]

Zierler et al. (1986) compared mortality rates from cancers of the stomach, colon, rectum, urinary bladder, breast, lung, pancreas, kidney and lymphatic system in 23 Massachusetts (USA) communities provided with chlorinated water with mortality from these cancers in Massachusetts as a whole. There were higher mortality rates in communities served by a chlorinated water source for cancer of the stomach among both males and females and for cancer of the lung among males. [The Working Group noted that the exposed group provided a substantial proportion of the reference population.]

(iii) Time-trend study

Cech et al. (1987) used the introduction of a new water supply in Houston, TX, USA, as a natural experiment. Lake Houston was constructed in 1954, and much of the population of Houston, previously supplied with water from lightly chlorinated underground sources, thereafter received heavily chlorinated surface water. Fifty-six census tracts were studied: group A (138 697 residents) had used groundwater over the whole period of the study; group B (46 394) changed from ground- to surface water in 1954; and groups C (84 159) and D (163 466) changed from ground- to surface water after 1954. THM levels were measured in 1978–79, and average concentrations in source areas A, B, C and D were 4, 111, 129 and 50 μg/l, respectively. The outcome measure was age-adjusted five-year average mortality from urinary tract cancer in 1940–74; control causes of death were respiratory cancer, bronchitis-emphysema and homicide. Trends in death rates over time showed little variation that could be related to the change in water supply. The slopes of the regression lines for urinary cancer rates for 1940–59 and for 1960–74 in area B showed a significant (p < 0.05) decrease for white men and a significant increase for white women; no such difference was found in other areas. Adjustment of mortality rates for education, population density, percentage population employed in high-risk industries, percentage population foreign-born, presence of metal or petroleum industries, and presence of hospitals with oncological units had no effect on the results. A cohort analysis was also carried out: there was some evidence of a birth cohort effect for urinary cancer in white women in area B. The authors concluded that there was little evidence of an effect of chlorination.

(i) Community exposure data

Alavanja et al. (1978) carried out a death certificate case-control study in seven counties of New York State (USA), chosen because the water supplies were diverse (including chlorinated surface and chlorinated and nonchlorinated groundwater); individual supplies had been stable for at least 15 years before the date of the study, and immigration had been low during the same period. In all, 3446 deaths occurring in 1968–70 from cancers of the gastrointestinal tract (oesophagus, stomach, small intestine, large intestine, rectum, liver, intrahepatic bile ducts, gall-bladder and bile ducts, pancreas and peritoneum and retroperitoneal tissue) and urinary tract (bladder, kidney, renal pelvis, ureter and other unspecified urinary organs) and 1416 lung cancer deaths were individually matched to noncancer deaths [not further defined] by year of death, race, sex, birthplace and county of residence. The ‘usual place of residence’ on the death certificate was taken as the place of residence. Water distribution maps [from an unspecified period] were used to locate the water supply for each case and control individually. The odds ratios associated with chlorination for gastrointestinal and urinary tract cancers combined were 1.44 for women and 2.09 for men (both p < 0.005). For lung cancer in all urban and rural areas combined, the odds ratios were 1.55 (not significant) for women and 1.83 (p < 0.005) for men. For individual cancer sites among men, all odds ratios were significantly greater than 1; only the odds ratio for stomach cancer was significantly raised for women. Random samples of cases and controls were taken in order to compare possible occupational exposures; male cases were more likely to have had occupational exposure to carcinogens than male controls (odds ratio, 1.25; not statistically significant). [The Working Group noted that no allowance for potential confounders (such as occupational exposure and smoking) was made in the analysis, the statistical analysis of the data is inadequately described, and it is likely that the matching was not dealt with appropriately.]

Brenniman et al. (1980) carried out a death certifícate case-control study incorporating 3208 deaths from gastrointestinal and urinary tract cancer occurring in Illinois, USA, between 1973 and 1976 and 43 666 non-cancer deaths as controls, excluding deaths from complications of pregnancy, congenital anomalies, perinatal disorders, mental disorders, senility and infectious diseases. The study was restricted to whites and to communities served by groundwater −272 chlorinated and 270 nonchlorinated. Data on water supply were obtained from an inventory of municipal water facilities published in 1963 and verified, where possible, by a questionnaire sent to the water supply source. Allowance was made in the analysis for age, sex, urbanicity and residence in a standard metropolitan statistical area. Cancers of the oesophagus, stomach, large intestine, rectum, liver, gall-bladder and bile ducts, pancreas, bladder and other urinary organs were studied for men and women separately. No significantly elevated odds ratio was found for any individual site.

Results from a study based on southern Louisiana (USA) parishes were presented in three articles (Gottlieb et al., 1981, 1982; Gottlieb & Carr, 1982). Gottlieb et al. (1982) carried out a case-control study of 10 205 cancer deaths in 13 parishes in southern Louisiana. Deaths from the following cancers formed the case series: urinary bladder, colon, kidney, liver, non-Hodgkin’s lymphoma, rectum, stomach, breast, brain, oesophagus, pancreas, Hodgkin’s disease, leukaemia, lung, malignant melanoma, multiple myeloma and prostate. For each case, a control matched on sex, age, race and year of death was selected from among deaths from causes other than cancer, excluding causes related to each cancer. Analyses of surface versus groundwater were carried out (i) according to water source at death and (ii) restricted to subjects on the same source type at birth and at death (lifetime exposure). The former analysis revealed only three significant odds ratios: 1.79 for rectal cancer, 1.21 for breast cancer and 0.70 for multiple myeloma. The analysis of lifetime water use gave significant odds ratios of 2.50 for rectal cancer and 1.30 for breast cancer. [Confidence intervals were not given.] A dose-related response was seen for each of these cancers in the categories lifetime surface water, some surface water (only birth or death in a parish served by surface water) and lifetime groundwater use. Odds ratios were elevated for rectal cancer among men according to water use at death (2.21; 95% Cl, 1.57–3.12) and for men and women according to lifetime water use (3.18; 1.96–5.19 in men and 1.73; 0.97–3.10 in women). There were also elevated risks for lung cancer among men on surface water at death (1.30; 1.05–1.62) and for breast cancer among women both on surface water at death (1.21; 1.00–1.46) and with lifetime surface water use (1.30; 1.00–1.69). An additional analysis based on 11 349 case-control pairs from 20 parishes (Gottlieb & Carr, 1982) did not provide different results. [The Working Group noted some inconsistencies in the number of parishes studied; the analysis is inappropriate since matching was broken, and the results are not presented in an understandable format.] A further analysis of these data (Gottlieb et al., 1981) was restricted to a sample of 692 deaths from rectal cancer and 1167 from colon cancer; 1859 controls were selected from strata based on age at death (± five years), race, sex, year of death and parish. Four categories of estimated lifetime surface water use were derived as in the earlier work: mostly surface (birth and death in a surface water parish), some surface (either birth or death in a surface water parish), possible surface (death in a groundwater parish, birth outside the study area), least surface (birth and death in a groundwater parish). Of the total population, 99.2% could be classified into one of the four surface water exposure levels. For rectal cancer, relative risks of 1.61 (95% CI, 0.91–2.85) and 2.11 (1.17–3.84) were found for a residence served by surface water for 10–19 and > 30 years, respectively, as compared with residence served by groundwater. A significantly increasing trend with a relative risk of 2.07 (95% CI, 1.49–2.88) was also found for lifetime consumption of ‘mostly surface water’ relative to the ‘least surface water’ category. No elevation in odds ratios for any of the variables of interest was found for colon cancer. [The Working Group noted some internal inconsistencies in these papers in the numbers of cases and controls, and the method of selection of controls and whether they were individually matched is unclear.]

A study of cancer mortality in Wisconsin, USA, in relation to water chlorination was reported by Young et al. (1981), Kanarek and Young (1982) and Young and Kanarek (1983). Young et al. (1981) carried out a matched case-control study based on the death certificates of white females who had died of cancer in Wisconsin during 1972–77, restricted to the 28 counties in which the population was relatively stable and in which there were both chlorinated and unchlorinated water supplies. A total of 8029 deaths due to cancers of the gastrointestinal and urinary tracts, lung, breast and brain were matched to white female noncancer deaths on county of residence, year of death and birth date. The water supply serving the usual place of residence as recorded on each subject’s death certificate was obtained from a 1970 survey of the 202 water sources serving the study areas, gathered from a postal questionnaire to the water utilities. These data included type of water source (surface or ground), presence or absence of environmental factors that might influence organic content (e.g., rural run-off) and mean daily chlorination doses over the previous 20 years in four levels: none, low (< 1.00 ppm), medium (1.00–1.70 ppm) and high (1.71–7.00 ppm). In an unmatched analysis, adjusted for marital status, urban residence and high-risk occupations specific for certain cancer sites, significantly high odds ratios were found for colon cancer: 1.53 (95% CI, 1.11–2.11), 1.53 (1.08–2.00) and 1.51 (1.06–2.14) for the low, medium and high categories, respectively. No significant increase in risk was found for other cancers. In areas with a rural run-off into the water supply, the odds ratios for colon cancer were higher, and these increased slightly after adjustment for depth of groundwater source and purification. In an additional analysis using matched data, similar results were obtained (Young & Kanarek, 1983). In a further analysis of these data (Kanarek & Young, 1982), in which organic contamination, source depth and purification were taken into account, the odds ratio for colon cancer among persons using chlorinated in relation to that for people using unchlorinated water sources (1.43; p < 0.02) increased to 1.81 (p = 0.03) for chlorinated sources contaminated by organic compounds and to 2.81 (p = 0.01) for chlorinated surface water.

In a case-control study of multiple cancer sites based on death certificates, Zierler et al. (1986) (see p. 113) compared communities in Massachusetts (USA) in which surface water was disinfected by chloramine treatment (see p. 51) (20 communities) or chlorine treatment (23 communities). More than 50 000 deaths from cancers of the urinary bladder, colon, kidney, pancreas, rectum, stomach, lung and breast occurring between 1969 and 1983 in persons aged 45 years or more were identified. Over 200 000 deaths from lymphatic cancer, cardiovascular disease, cerebrovascular disease, pulmonary disease and pneumonia/influenza were used as controls. Exposure was defined as residence in a community supplied with chlorinated water at the time of death; nonexposure, in a community supplied with chloraminated water. No elevated risk for cancer at any site was observed.

(ii) Individual exposure data

Lawrence et al. (1984) carried out a case-control study based on death certificates of white women who had been members of the New York State (USA) Teachers’ Retirement System and had died from cancers of the colon and rectum. After geographical restrictions and other exclusions, 395 deaths occurring between 1962 and 1978 were included. Controls (395) were selected randomly from deaths due to any cause except malignant tumours and matched to the cases by age and year of death (within two years). Information was obtained on residence and employment 20 years prior to death, and water records were abstracted for both home and work addresses over the 20-year period. A model-based estimate of exposure to THM was derived from a study of New York State surface water systems. Potential confounding factors included in the matched and unmatched logistic regression analyses were population density, marital status, age and year of death. Only analyses for grouped colon and rectal cancers were reported. Results from the matched and unmatched analyses were identical. There was no significant finding in relation either to source type (odds ratio, 1.07; 90% CI, 0.79–1.43), to 20 years’ cumulative chloroform dose or to five other measures of exposure to chlorine or THM.

Cantor et al. (1985, 1987) conducted a population-based case-control interview study of urinary bladder cancer in 10 areas of the USA comprising 2982 cases aged 21–84 who had been newly diagnosed in 1978 (73% of the eligible pool). A total of 5782 controls were selected by random-digit dialling for those age 21–64 and by Health Care Financing Agency listings for those 65 and older. Interviews in the homes of the subjects gathered information on residential history, fluid consumption and potential confounders (smoking, occupation, lower urinary tract infection, artificial sweetener use, use of hair dyes). Data on water source and treatment since 1900 was obtained from an independent survey of water utilities. Year-by-year profiles of water source (surface and ground) and water treatment (chlorinated and not) were derived for the lifetime of each respondent by merging individual residential and water utility information; 76% of all person-years could be related to a known water source. Reported consumption of drinking-water was added to the intake of other home beverages containing tap water to estimate total daily ingestion of tap water. Since one goal of the study was to estimate the risk associated with consumption of chlorinated surface water in comparison with nonchlorinated groundwater, the primary analyses were restricted to a subset of respondents who had lived at least 50% of their lifetime prior to interview at residences served by one or both of these two types of water source [59% of all cases and controls]. Analysis was by logistic regression. In initial analyses (Cantor et al., 1985) that did not consider tap-water consumption levels, an association was found between duration of residence with a chlorinated surface source and risk of urinary bladder cancer. Only among nonsmokers was there a significant odds ratio for those exposed for more than 60 years (odds ratio, 2.3; 95% CI, 1.3–4.2); there was a nonsignificant inverse trend for current smokers. For all groups combined (controlling for smoking), odds ratios for the duration measure were close to one. In subsequent analyses (Cantor et al., 1987), current total fluid and tap-water consumption were considered in conjunction with duration of exposure to chlorinated surface water. Total fluid consumption was related to urinary bladder cancer risk, and tap water was the main risk factor (test for trend: males, χ² = 22.6, p < 0.0001; females, χ² = 3.15, p = 0.08). These findings were not modified by extent of disease. When respondents were grouped by duration of chlorinated surface water use, significant trends with tap-water intake were restricted to persons who had consumed chlorinated water for 40–59 years and ≥ 60 years. The odds ratios for the highest (≥ 1.961/day) versus the lowest (≤ 0.801/day) quintiles of intake in these two duration strata were 1.7 and 2.0, respectively, with significant trends (p = 0.006 and p = 0.014, respectively). The trends in odds ratios with tap-water intake were nonsignificant for up to 39 years’ duration. There was a significant trend with duration of residence with a chlorinated surface water supply, but only among women whose tap-water consumption was above the median (p = 0.02). The overall increase in the odds ratio with duration seen among nonsmokers in the previous analysis (Cantor et al., 1985) was more exaggerated among respondents whose tap-water consumption was above the median (p = 0.01 for trend) than in those whose consumption was below the median (p = 0.40 for trend).

Lynch et al. (1989) conducted an analysis of the Iowa respondents in the study of Cantor et al. (1987), comprising 354 cases of urinary bladder cancer and 752 controls. Chlorination was quantified in four ways, with increasing levels of specificity: (i) assuming that the respondent’s lifetime was spent consuming the type of water provided by the community of his or her most recent place of residence; (ii) assuming that the person’s most recently used water supply (whether or not his or her community’s supply) was used for life; (iii) applying the most recent water supply to the number of years of actual residence at this place; and (iv) using the entire lifetime residential/water supply history. For methods (ii), (iii) and (iv), there were significant trends with exposure to chlorination, the highest odds ratio being found for method (iv) (test for trend: χ² = 10.90, p = 0.001). The odds ratios for 1–25, 26–50 and > 50 years of exposure to chlorination relative to no exposure using method (iv) were 1.42, 1.70 (p < 0.01) and 2.14 (p < 0.01). After adjustment for age and smoking, the odds ratio for history of any exposure to chlorinated water was 1.47. The highest unadjusted odds ratio [no adjusted odds ratio reported] was found for exposure only to prechlorinated or prefiltered surface or shallow groundwater (odds ratio, 2.95; 95% CI, 1.52–5.75). In this study subset of Iowa respondents, cigarette smokers (more than 25 pack-years) who had had exposure to chlorinated drinking-water had a higher odds ratio (4.48; 95% CI, 2.47–8.13) than smokers never so exposed (2.89; 1.41–5.89), relative to nonsmokers never exposed to chlorinated drinking-water. This result contrasts with the findings from the overall study in which smokers who had used chlorinated surface water were not at excess risk (Cantor et al., 1985).

Cragle et al. (1985) identified 200 cases of colon cancer newly diagnosed between 1978 and 1980 at seven hospitals in North Carolina (USA) who had had at least 10 years’ residence in the state. At least two hospital controls were matched to each case by age, race, sex, vital status, date of diagnosis and hospital. Information on residential history and a variety of potential confounding factors was collected from the respondents by either a personal interview or by mail questionnaire. Each subject’s residence history for 1953–78 was related to data from the water company to derive estimates of the duration of residence on chlorinated and nonchlorinated supplies. A logistic regression analysis was carried out which included a chlorination variable and several potential confounders. The authors concluded that there was an association between chlorination and colon cancer in people over the age of 60. [The Working Group noted that a number of details of the study design are not adequately described: it is not stated how many deceased cases and controls were selected and what procedure was used for obtaining data on these subjects; it is not clear how the chlorination variable was treated in the analysis; and, in spite of the matched nature of the design, an unmatched analysis was apparently carried out.]

A study in Wisconsin (USA), reported by Young et al. (1987) was designed to estimate the risk for colon cancer associated with chronic ingestion of THMs occurring as by-products of water chlorination. White men and women aged 50–90 were included. Cases were identified from a state-wide hospital tumour registry over a two-year period; 347 cases (45% of those sampled) were included in the analysis. Two sets of controls were used: 639 cancer controls identified from the same source as the cases, and 611 population controls identified from driver’s license records, representing 48% of controls sampled. Self-completed questionnaires, supplemented with medical records, were used to obtain lifetime histories of residence, water use and medical and occupational histories. Water company records and contemporary measurements of THM were used to estimate the THM content of all types of water source in the past and then to construct estimates of lifetime ingestion of THM for each subject. Odds ratios for colon cancer relative to population controls, adjusted for age, sex and population size were 1.10 (95% CI, 0.68–1.78) for estimated cumulative exposure to 100–300 mg THM and 0.73 (0.44–1.21) for 300 mg or more, relative to the baseline group (less than 100 mg lifetime ingestion of THM). Analyses comparing surface with groundwater sources and chlorinated with nonchlorinated sources also showed no association with colon cancer risk. [The Working Group considered that the response rate in this study was too low to permit reliable inferences to be made.]

Zierler et al. (1988) carried out a case-control study of urinary bladder cancer based on death certificates of residents of 43 Massachusetts (USA) communities served by surface water disinfected by chlorine or chloramine. A total of 1057 deaths from urinary bladder cancer in people aged 45 or more occurring between 1978 and 1984 were identified. Controls were obtained from an age-stratified sample of deaths from the following causes: lung cancer, lymphoma, cardiovascular disease, cerebrovascular disease and chronic obstructive pulmonary disease (total, 2144). A large number of the cases and controls included in this study were also included in a previous case-control study carried out by the same authors (Zierler et al., 1986; see p. 117). Informants were found for 614 (58%) of the cases and 1074 (50%) of controls and were interviewed about the decedents’ residential and smoking history. Each subject’s residential history was linked to historical data on water source obtained from the US Environmental Protection Agency and State water authorities. Four categories of lifetime exposure to chlorinated water were defined, and each individual was placed into one of these. Information on socioeconomic status and high-risk occupations was obtained indirectly at the level of the community. Odds ratios for usual and lifetime exposure to chlorinated water with respect to lifetime exposure to chloramine were 1.4 (95% CI, 1.1–1.8) and 1.6 (1.2–2.1). When analysis was restricted to 30 communities each supplied by a single authority, the odds ratio for lifetime exposure with respect to lifetime nonexposure was 1.6 (1.1.-2.4). [The Working Group noted that the response rate was very low. It was unclear whether information on water supplies was obtained when individuals resided outside the 43 communities. The choice of controls may not have been appropriate. Confounding by city size was not addressed. Differences between the results of this study and those of Zierler et al. (1986) may be due to the fact that the exposure information in this study was more precise or to selection biases due to low response rates.]