1 2775. EPIDEMIOLOGICAL STUDIESThis chapter reviews observational and experimental epidemio-logical STUDIES that have been conducted to determine associationsbetween disinfected drinking-water and adverse health outcomes. Dis-infection practices vary throughout the world. Applied and residualconcentrations have varied over the years and from country study designs, sources of systematic and randomerror (bias), and guidelines for assessing the causality of associationsare discussed in section EPIDEMIOLOGICAL STUDIES of exposures todisinfected drinking-water and to specific DBPs are evaluated in sec-tions and , EPIDEMIOLOGICAL STUDIES have been conducted todetermine possible associations between adverse health-related out-comes and drinking-water disinfected with chlorine and drinking-water was studied most often, and studiesprimarily compared health risks associated with chlorinated drinking-water from surface water sources with those associated with unchlori-nated drinking-water from groundwater sources.
2 Also studied werespecific DBPs, including chloroform and other THMs. Only one studyconsidered DBPs other than THMs. Two STUDIES considered risks thatmay be associated with chloraminated water and chlorine mutagenic activity of drinking-water, which may representexposure to the non-volatile, acid/neutral fraction of chlorinatedorganic material in water, was also considered. Health effects studiedincluded cancer, cardiovascular disease and adverse reproductive anddevelopmental outcomes. Most of the STUDIES focused on bladdercancer risks. Also studied were risks of colon, rectal and study designs and causality ofepidemiological associationsBoth observational and experimental EPIDEMIOLOGICAL studieshave been conducted to assess the health risks associated withdrinking-water disinfection (Table 22). ehc 216 : Disinfectants and Disinfectant By-products278 Table 22. Types of EPIDEMIOLOGICAL surveillance and (follow-up) control (case comparison) Adapted from Monson (1990).
3 Experimental studiesResults of experimental EPIDEMIOLOGICAL STUDIES , which includeclinical trials, are reported in chapter 4 as appropriate under toxicityin humans. These STUDIES consider the effect of varying some charac-teristic or exposure that is under the investigator s control, much likein a toxicological study. Comparable individuals are randomlyassigned to a treatment or intervention group and observed for aspecific health-related outcome. Ethical concerns must be fullyaddressed. Several clinical trials have evaluated changes in lipid, thy-roid and haematological parameters that may be affected by consump-tion of disinfected drinking-water. studiesTwo basic kinds of observational EPIDEMIOLOGICAL STUDIES havebeen conducted to determine risks associated with disinfection ofdrinking-water: ecological and analytical. These two study approachesdiffer primarily in the supportive evidence they can provide about apossible causal association.
4 Unlike the analytical study, an ecologicalstudy does not link individual outcome events to individual exposureor confounding characteristics, and it does not link individualexposure and confounding characteristics to one another. In anecological study, information about exposure and disease is availableonly for groups of people, and critical information can be lost in theprocess of aggregating these data (Piantadosi, 1994). Results fromEpidemiological Studies279ecological STUDIES are difficult to interpret, and serious errors canresult when it is assumed that inferences from an ecological analysispertain either to the individuals within the group or to individualsacross the groups (Connor & Gillings, 1974; Piantadosi et al., 1988).Theoretical and empirical analyses have offered no consistentguidelines for the interpretation of ecological associations (Greenland& Robins, 1994a,b; Piantadosi, 1994). Investigators (Greenland &Robins, 1994a,b; Piantadosi, 1994; Poole, 1994; Susser, 1994a,b)have examined the limitations of ecological STUDIES and determinedwhen and under what assumptions this type of study may beappropriate.
5 Analytical STUDIES can provide the necessary information to helpevaluate the causality of an association and estimate the magnitude ofrisk. For each person included in the study, information is obtainedabout their disease status, their exposure to various contaminants andconfounding characteristics. Analytical STUDIES are either longitudinalor cross-sectional. In a longitudinal study, the time sequence can beinferred between exposure and disease; in other words, exposureprecedes disease. In a cross-sectional study, exposure and diseaseinformation relate to the same time period; in these STUDIES , it may notalways be correct to presume that exposure preceded disease. Thecross-sectional study design was used to investigate possible risks ofcardiovascular disease and reproductive and developmental risks. Longitudinal STUDIES are of two opposite approaches: the cohortstudy and the case control study. The cohort study begins with theidentification of individuals having an exposure of interest and a non-exposed population for comparison; disease consequences or otherhealth-related outcomes are then determined for each group.
6 In acase control study, the investigator identifies individuals having adisease or health outcome of interest and a control or comparisongroup of individuals without the disease of interest; exposures and riskfactors are evaluated in these persons. In a case control study, a vari-ety of exposures can be studied, whereas in a cohort study, a varietyof diseases can be cohort or follow-up study can be either retrospective orprospective, and sometimes a combination retrospective prospectiveapproach is used. Two or more groups of people are assembled forstudy strictly according to their exposure status. Incidence or mortalityEHC 216: Disinfectants and Disinfectant By-products280rates for the disease of interest are compared between exposed andunexposed groups. Multiple disease end-points can be evaluated, buta disadvantage is that large numbers of people must be studied,especially for environmental exposures. Because of the lengthy latentperiod for cardiovascular disease and cancer, a long follow-up periodis required for a prospective cohort, and this is usually not retrospective cohort study design was used to evaluate thepossible association of chlorinated drinking-water with cancer andcardiovascular disease a case control study, persons with the disease of interest (thecases) and persons without this disease (the controls or comparisongroup) are sampled from either the general population or a specialpopulation ( , hospitals or a select group) within a specifiedgeographic area.
7 Exposures among the cases are compared withexposures among non-diseased persons. Multiple exposures can beevaluated, and a relatively small number of study participants isneeded to obtain reasonably precise estimates of risk associated withenvironmental exposures. Retrospective exposures must beconsidered, and, because of the lengthy latency period for cancer andcardiovascular disease, exposures to water sources and contaminantsmust be assessed over the previous 20 30 years, or perhaps even aperson s lifetime. It may be difficult to assess these exposuresaccurately. Two types of case control STUDIES have been conducted toinvestigate associations between disinfected drinking-water andcancer: Decedent cases and controls without interviewing next-of-kin orsurvivors for information about residential histories, risk factorsand possible confounding characteristics. Decedent and incident cases and controls using interviews orother methods to obtain information about possible confoundingcharacteristics and document long-term mobility and changes inresidences to allow documentation of lifetime exposure to disin-fected water.
8 In several STUDIES , a person s intake of tapwater andhistorical exposures to chlorinated water, chloroform or otherTHMs were and systematic errorBiases that occur during the design and conduct of a study canlead to a false or spurious association or a measure of risk that departssystematically from the true value. All reported EPIDEMIOLOGICAL asso-ciations require evaluation of random and systematic error so thatresults can be interpreted properly. Systematic error (bias) affects thevalidity of a study s observed association; random error affects theprecision of the estimated magnitude of the risk. Random error isgoverned by chance and is influenced by the size of the study. Thelikelihood that a positive association is due to random error can beassessed by calculating the level of statistical significance ( P value)or confidence interval (CI). A small P value or a CI that does notinclude unity ( ) suggests that chance may be an unlikely explana-tion for an observed association, but the association may, nevertheless,be spurious because of systematic error.
9 Statistical significance doesnot imply causality or biological significance, nor does it mean thatrandom error or chance can ever be completely ruled out as a possibleexplanation for the observed association. Many epidemiologistsbelieve that strict reliance on statistical significance testing is notappropriate (US EPA, 1994a).Potential sources of systematic error include observation,selection, misclassification and confounding biases. Wheninformation on exposure and disease is collected by methods that arenot comparable for each participant ( , selective recall), anincorrect association will be due to observation bias. When the criteriaused to enrol individuals in the study are not comparable, the observedassociation between exposure and disease will be due to selection wrong diagnosis of disease or assessment of exposure can result inmisclassification bias. This type of bias may be randomly distributed(non-differential misclassification bias), which almost always biasesstudy results towards the direction of not observing an effect (orobserving a smaller change in risk than may actually be present), orit may be non-random (differential misclassification bias), which canresult in either higher or lower estimates of risk, depending on howthe misclassification is distributed.
10 Lynch et al. (1989) examined theeffects of misclassification of exposure using empirical data from aninterview-based case control study of bladder cancer in Iowa (USA).Bladder cancer risk estimates were found to be higher when moreEHC 216: Disinfectants and Disinfectant By-products282information was known about the study participants residentialhistory and their possible exposure to chlorinated water sources. Thissuggests that misclassification bias in EPIDEMIOLOGICAL STUDIES ofchlorinated water may be primarily non-differential, underestimatingthe risk; however, in study areas where residential mobility isdifferent from that in Iowa, the magnitude of risk may beoverestimated rather than bias may convey the appearance of an association;that is, a confounding characteristic rather than the putative cause orexposure may be responsible for all or much of the observed asso-ciation. Although negative confounding bias may occur, concern isusually with positive effects of confounding bias is confoundingbias responsible for the observed association?