Close this screen to return to the previous page.

How are the risks of water contaminants determined?

Everyone has probably heard warnings about the various alleged health risks of drinking tap water. Bottled water continues to soar in popularity due, in part, to the perceived risks and the marketing of bottled water products as a safe alternative. Just what are the risks of drinking harmful water contaminants, what health problems do they cause, how are those risks identified, how serious are the risks, and how can we, as consumers of water, know how to interpret the various claims that are made about drinking water safety.

When reading any information about health risks of drinking water contaminants, it is important to understand that for many of these contaminants, it is often difficult to assign an absolute health risk. You can't just go out and experiment on a group of people by taking a few thousand men, women, and children, dividing them into random groups, dosing them with varying amounts of potentially harmful chemicals, and recording the results on their health. Experiments may be performed on rats and mice, but then there is the argument that chemicals which cause health problems in other animals will not behave the same way in humans.

As a consequence, much of the information that is known about the health risks of various drinking water contaminants (as well as other harmful compounds, disease causing organisms, and risky behaviors) is determined from what are called epidemiological studies rather than experimental studies. Epidemiology is the branch of medical science that studies various factors which influence the incidence (rate of occurrence), distribution, and dynamics of diseases, injury or other health-related problems in populations. Epidemiology can be thought of in terms of who, where, when, what, and why. That is, who has the infection (or disease or injury), where are they located geographically and in relation to each other, when is the infection (or disease or injury) occurring, what is the cause, and why did it occur.

In epidemiological studies, health and disease data are collected from the health records and or surveys of two or more different populations of people over some period of time. The populations are chosen to be as similar as possible to each other except for the risk factor (chemical exposure, pathogen, behavior, etc.) that is being studied. The group which does not have the risk factor is the control population for the study. The goal is to try and determine what risk factors contribute to a particular health outcome, usually a disease, injury, and/or death. Examples of possible study populations and outcomes might be:

	Male smokers 40-70 vs. male non-smokers 40-70 - outcome, lung cancer.
	People in a community that drink hard water vs. people in another community that drink soft water - outcome, heart disease.
	People in a water district that has high levels of disinfection byproducts vs. those in another with low levels of disinfection byproducts - outcome, bladder cancer.
	Children in a town where fluoride is added to the water vs. children in a nearby town who drink water containing low fluoride levels - outcome, number of cavities per year.

When populations are similar, except for the risk factor being studied, any observed differences in health, death rate, or other outcome, will lead researchers to conclude that there is reason to suspect a possible association (or correlation) between the risk factor and the observed differences in outcome. The larger the difference in outcomes the greater the possible association between risk factor and outcome.

A term that is used to describe the results of an epidemiological study and the difference between the possible risk factor and the outcomes measured in the two populations is Relative Risk. Relative Risk is the number of outcomes in the population with the risk factor divided by the number of outcomes in the population that does not have the risk factor. For example, in a study quoted in one of the links below, in two populations, similar except that one group smoked and the other did not, there were 397 lung cancer deaths in the smoker group compared with 37 lung cancer deaths in non-smokers. The relative risk of smoking in this study was 397/37 = 10.73 (that is, smokers were 10.73 times more likely to die from lung cancer than non-smokers).

It is important to realize, however, that epidemiological studies can never prove causation; that is, they cannot prove that a specific risk factor actually causes the disease being studied. Epidemiological evidence can only show that this risk factor is associated (correlated) with a higher incidence of disease in the population exposed to that risk factor. The higher the correlation the more certain the association, and the more probable it might cause the disease, but it cannot prove the causation.

For instance, one of the articles listed below indicates a correlation between heavy smoking and liver cirrhosis. That correlation, however, is probably caused by the fact that many heavy smokers are also heavy drinkers. The heavy drinking is more probably the cause of cirrhosis than smoking.

In an epidemiological study which considered the likely causes of cirrhosis, smoking would be a confounding variable or, in the jargon of the mystery genre, a "red herring". It would be important when studying two populations to determine the possible effects of heavy drinking on cirrhosis, that both study groups had a similar percentage of smokers. It would also be important to select the groups so other risk variables like age, gender, economic status, diet, etc. were as similar as possible in both groups. Since the two populations that are being compared are never identical, there are various statistical methods that are used to help determine how risk factors interact with each other and contribute to the observed outcome.

In many epidemiological studies, particularly where it is difficult to isolate risk factors, difficult to find populations that are similar except for the risk factor of interest, or where there is little relative risk between the two populations, it may be challenging to demonstrate a clear correlation between a specific risk factor and the health problem it is suspected of causing. That is the case with many of the drinking water studies of the health effects of organic compounds described below.

Several issues that contribute to potential problems and limitations interpreting epidemiological drinking water studies:

	The low concentrations of the compounds suspected of causing health problems in most water supplies. If someone drinks an ounce of chloroform (one of the disinfection byproducts of adding chlorine to municipal water) the health effects will be sudden, obvious, and deadly. The health effects of drinking chloroform that is diluted in water to a few parts per billion are neither sudden or obvious. Usually nothing obvious happens even over a lifetime of exposure. For the few people who eventually develop cancer because of the increased risk of a lifetime's exposure to this compound, however, it can still be deadly.
	The difficulty in ruling out all other possible risk factors which may contribute to the observed outcomes. This is well illustrated by the ongoing debate about the reported health benefits of drinking hard water over soft water. The proposed risk factors of soft water are a lack of the obvious hardness minerals (calcium and magnesium) which appear to be correlated with an increase in heart disease in many epidemiological studies, even in areas where dietary sources of calcium and magnesium and other essential minerals appear to be adequate. There are also studies that do not find a health risk of soft water. When an apparent increased risk of heart disease is observed in populations that drink soft water instead of hard water, there are no obvious reasons to explain this observation if the body's needs for calcium, magnesium, and other essential minerals are met by the diet. Consequently, there is considerable speculation and debate about as yet unidentified or unrecognized confounding variables which are causing the reported findings.
	Related to the above issue is the difficulty of finding populations that are quite similar in all characteristics except the one risk factor of interest. The populations which are compared are typically from different towns or different locations within a town, so that, for example, if one population is drinking hard water and the other soft water, there are probably a number of other things about the two groups (the people's lifestyle, their diet, work environment, etc.) that are also different and must be accounted for in the analysis of the results.
	The relative risk that is measured is often very small between the population that drinks water containing the risk factor which is being studied (disinfection byproducts, for example) and the group that is not drinking the risk factor. Instead of having a relative risk of over 10 for smokers vs. non-smokers, the relative risk of many of the compounds in drinking water that are considered to be possible causes of health problems may be in the range of 1.3 to 2.
	Epidemiological drinking water studies are also complicated by the often long period of time before drinking water containing low levels of suspected risk factors begin to have an observable effect on health. Much of the existing research on long-term health effects of water contaminants examines historical health and water quality records to find patterns of health problems over time that correlate with recorded drinking water quality issues over the same period of time.
	The difficulty in determining the actual amount of exposure of a population to a particular risk factor. Think about how, if you were researching the health effects of disinfection byproducts, you would go about trying to determine the amount of disinfection byproducts the members of your study population had consumed over a 30 to 40 year period.
	The different ways an epidemiological study can be set up and conducted influences the results. The locations and characteristics of the populations that are selected for the study, the data the researcher decides to collect, the time period used for the study, the historical records and materials that are collected and evaluated, the methods that are used to analyze the data, and yes, even the researcher's expectations and biases, can all have an effect on the conclusions reached.

Many of the epidemiological studies read like a mystery novel; there is the killer, a list of suspects, several 'red herrings', the suspense that if the killer is not apprehended more innocents will die, and finally, the steadfast detective collecting and analyzing evidence, making deductions, and relentlessly pursuing his or her quarry. The evidence collected in an epidemiological study is nearly always circumstantial, and there is seldom a 'smoking gun' that leads conclusively to the culprit.

There are many cases where the evidence has been sufficiently convincing to overcome political and bureaucratic inertia and provide the incentive to regulate or ban certain contaminants (lead, E. coli, cryptosporidium, mercury, trihalomethanes, MTBE, and arsenic to name a few). There are also many cases where the evidence is not as conclusive, and the jury is still out on what the health risk of a chemical is, what levels of exposure are harmful, and what regulatory actions (if any) to take.

Even if there is a fairly clear indication that a specific contaminant causes a specific health problem, there is frequently debate about the levels at which the contaminant is harmful. There are, of course, also special interest groups promoting a particular conclusion about the seriousness of the health threat which can influence the decisions that are made about banning or regulating products and setting maximum contaminant levels for drinking water contaminants.

The issues discussed above are presented to help explain why, even in careful reading of a number of epidemiological studies on the health effects of a specific drinking water contaminant, you will frequently find differing results and perhaps even contradictory conclusions.

After a number of studies have been conducted, reviewed, and published on the health effects of a specific contaminant, a consensus will usually begin to emerge in the scientific community about the 'actual' health risks of the contaminant.

The strength of the consensus is based, in part, on the number of studies published that report similar results and the perceived quality of those studies.

The level of concern in the scientific community about the contaminant is based, in part, on the reported relative risks (or odds ratios - mentioned below), the severity of the health effect (death vs. skin rash), the level of exposure in typical water supplies, and the length of time it takes someone drinking the water to experience the harmful effects.

	If many studies show a clear correlation of a risk factor with a disease, and/or if the relative risk (or odds ratio) of disease from the contaminant is high, and/or if the the disease is severe, and/or if the onset of disease symptoms is rapid, there will be a fairly rapid consensus on the harmful health effects of the contaminant and a call to take regulatory or preventive action. - E. coli in drinking water causing rapid onset of disease and death is an example of this situation.
	There is also now a strong consensus among scientists for concern about the harmful effects of the disinfection byproducts, particularly their apparent contribution to the increased risk of getting several cancers after many years of drinking water containing these products. The epidemiological evidence for increased risk is strong and consistent but not nearly as conclusive as it is for E. coli.
	There are, however, many contaminants for which a consensus on their potential harm has not been reached - often because few studies have yet been conducted, study results may be inconsistent, the perceived health risk may low, the calculated relative risk (or odds ratio) may not be high enough to raise alarm, and so forth.
	Many proposed negative health effects of the disinfection byproducts, in addition to increased cancer rates, for example, fall into this category. Increased rates of miscarriage, low birth weight, chromosomal abnormalities, neural tube defects, and cardiac defects have all been attributed within the last few years to drinking water containing various levels of disinfection byproducts.

In conclusion, the jury is still out on whether many of the other proposed risks of the disinfection byproducts - or the risks of many hundreds of other compounds that can be found in drinking water - need to be taken seriously by the scientific community, the government officials in charge of regulating drinking water supplies, the drinking water professionals whose job it is to provide safe water to their customers, and the public that consumes the water.

The decision on whether to fund epidemiological studies to uncover harmful drinking water contaminants or whether the costs of regulation or cleanup of contaminants (even those identified as posing some risk) is based, in part on another branch of the sciences, Risk Assessment and Management, but that's a story for another time.

A final thought and story. Dr. John Snow (1813-1858) is often recognized as the first modern epidemiologist. During the 1854 Cholera epidemic in London, the city's water was supplied by two water companies. One drew its water out of the Thames River upstream from the main city while the other, Southwark and Vauxhall, drew its water from the lower Thames, downstream from the city, where it had become contaminated with sewage. Dr. Snow, during his investigation of the epidemic, plotted on a city map some 500 Cholera deaths that occurred within a 10 day period. He found that most of the deaths were within a 200 yard radius of a pump located at the intersection of Cambridge and Broad Streets. After he persuaded officials to remove the handle of the Broad Street Pump that supplied the water to that neighborhood, the number of cholera cases and deaths in London was dramatically reduced. At the time the actual infectious agent, the bacterium, Vibro cholerae, was unknown.

Abstracts of journal articles covering: Disinfection Byproducts and Cancer, Disinfection Byproducts and Pregnancy, and Calcium & Magnesium in Drinking Water

If you do read these abstracts, you will find that comparisons and results are frequently reported as Odds Ratios (OR) instead of relative risk. Although an epidemiologist or statistician will violently disagree, you can more-or-less interpret an odds ratio in the same way you would think of relative risk. A study result might be reported as "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 confidence interval [CI] = 1.10-1.81)." An odds ratio above 1.0 means the risk factor had an increased risk on the study outcome - the higher the OR value the greater the risk, and the confidence interval typically reported with an odds ratio (or relative risk) indicates whether the risk is statistically significant or not. More information on relative risk vs. odds ratio.

Some resources: Epidemiology - Trying to Establish Cause, Epidemiology
     A good, one-page description of Epidemiology
     A remarkable series of on-line college-level lectures on epidemiology:
       Toxicologic Epidemiology by Dr. Michael H. Dong
       Principles of Epidemiology by Dona Schneider, PhD, MPH, FACE
       A Brief Introduction to Epidemiology by Betty C. Jung RN MPH CHES
       Principles of Epidemiology by Kevin E. Kip, Ph.D

Many have seen the movie, Erin Brockovich, the true story of Ms. Brockovich, a legal assistant, who in 1993 lined up some 650 prospective plaintiffs from the tiny desert town of Hinkley, Calif., to sue Pacific Gas & Electric. PG&E's nearby plant was leaching chromium 6, a rust inhibitor, into Hinkley's water supply, and the suit blamed the chemical for dozens of symptoms, ranging from nosebleeds to breast cancer, Hodgkin's disease, miscarriages and spinal deterioration. In 1996 PG&E settled the case for $333 million. If you read this article (and the associated links, and another report), you will get a 'picture' of some of the complexity of trying to figure out the cause of specific diseases and the effects of suspected harmful chemicals. The epidemiological and scientific studies help to provide the understanding of the "actual" causes and effects in disease processes. They are just one piece of the overall puzzle of how the public perceives the problem, however.

Drinking Water Disinfection Byproducts: Review and Approach to Toxicity Evaluation - There is widespread potential for human exposure to disinfection byproducts (DBPs) in drinking water because everyone drinks, bathes, cooks, and cleans with water. The need for clean and safe water led the U.S. Congress to pass the Safe Drinking Water Act more than 20 years ago in 1974. In 1976, chloroform, a trihalomethane (THM) and a principal DBP, was shown to be carcinogenic in rodents. This prompted the U.S. Environmental Protection Agency (U.S. EPA) in 1979 to develop a drinking water rule that would provide guidance on the levels of THMs allowed in drinking water. Further concern was raised by epidemiology studies suggesting a weak association between the consumption of chlorinated drinking water and the occurrence of bladder, colon, and rectal cancer... {The article goes on to describe disinfection products of various drinking water treatment methods and the "balancing act" described above between providing safe water and creating treatment byproducts that have some health risk.}