A Framework for Applied Medical Geology: Part II. The Biological Impact Analysis

Abstract

The relationships among the fields of exposure science, toxicology, and epidemiology can be described within the framework of an Environmental Pathways/Biological Impact Analysis for risk assessment and risk management. A sequential path of events and processes lead from the introduction of a “stressor,” such as a contaminant, into the environment to the exposure–epidemiology interface, and then to the exposure–toxicology interface. The former interface occurs where a contaminant released into the environment is transferred to an exposure medium such as air, water, or soil that potentially can come into contact with a person. This interface has been discussed in Chap. 1 of this book under the topic of geoavailability, where it is described as the fraction of a contaminant released into the environment that is available for contact with a potentially exposed population. The second interface occurs where the exposure medium comes into contact with a human through an external exposure that leads to an internal exposure and dose. Biomarkers of exposure that relate external exposures to internal exposure and dose include measurement of metals and biochemical complexes in bodily fluids and tissues. The processes that link an external exposure from geogenic materials to an internal dose are covered in this chapter under the subject of bioavailability. This is defined as the fraction of an adsorbed contaminant that reaches a target site where cellular damage occurs. Biomarkers of effects are used to recognize preclinical manifestations of diseases and include a large number of molecular biomarkers and changes in cellular function and morphology. The use of these concepts in the field of environmental epidemiology is described and the role of these different tools in medical geology for exposure assessment is summarized to provide a context for the other sections of the book.

Appendices

Appendix 1: Toxicology and Risk Calculations Considering Bioavailability

• Cancer risk can be generally expressed by the following equation: Excess lifetime cancer risk = ELCR = DI × CSF, where DI is the chemical daily intake and CSF is the cancer slope factor.

• Noncancer risk can be calculated as Hazard Quotient = DI/RfD, where RfD is the reference dose.

• The effect of contaminant bioavailability from soil ingestion to human receptors can be evaluated by making adjustments to the dose using the following equation: DI_adjusted = DI × RBA.

• Alternately, RBA can be used to make site-specific risk adjustments for cancer risk by using the following equation: CSF_adjusted = CSF_IRIS × RBA, where CSF is the slope factor.

• Site-specific adjustment for noncancer risk can be calculated by the following equation: RfD_adjusted = RfD_IRIS/RBA.

Alternatively, when the exposure frequency, exposure duration, and ingestion rates are specified, the RBA value is used to calculate the excess lifetime cancer risk (ELCR) and the hazard quotient (HQ) as (see ITRC 2017, Section 9.1.3.2 for more details) (Table 2.4):

$$ \mathrm{ELCR}=\frac{\mathrm{Cs}\times \mathrm{RBA}\times \mathrm{IR}\times \mathrm{EF}\times \mathrm{ED}\times \mathrm{CSF}}{\;\mathrm{BW}\times \mathrm{AT}\times \mathrm{CF}} $$

$$ \mathrm{HQ}=\frac{\mathrm{Cs}\times \mathrm{RBA}\times \mathrm{IR}\times \mathrm{EF}\times \mathrm{ED}\;}{\;\mathrm{RfD}\times \mathrm{BW}\times \mathrm{AT}\times \mathrm{CF}} $$

Table 2.4 Terms for toxicology calculations

Appendix 2: Basic Concepts in Epidemiology for Medical Geologists

1.1 Environmental Epidemiology Basics

Environmental epidemiology is the study of relationship between exposure to environmental risk factors and the occurrence of disease. The purpose of this appendix is to provide an overview of basic concepts in environmental epidemiology to aid the reader in understanding the discussion of the subject in this book. More detailed descriptions of epidemiological study designs (Box 2.9) can be found in standard books on epidemiology (Gordis 2000; Szklo and Nieto 2000). Detailed descriptions of statistical techniques used in environmental epidemiology studies can be found in standard biostatistics texts such as Rosner (1995); an introduction to the use of Bayesian statistics in epidemiology can be found in Ashby and Hutton (1996). This technique was used in studies of exposures to uranium contamination on the Navajo Nation by Hund et al. (2015). In epidemiology, an important distinction is made between correlation (or association) and causation. The former is established using a variety of statistical depending on the type of population being examined and is subject to rigid rules of probability theory, which establish the degree of statistical significance of the association. The latter is evaluated by a set of criteria established by Hill (1965) in the mid-twentieth century as discussed here.

Box 2.9 Epidemiology Study Designs

Observational—Descriptive

• Disease surveillance and surveys

• Ecological

• Cross-sectional

Analytical—Longitudinal

• Cohort

• Case-control

Experimental

• Clinical trials

• Population (interventions)

1.1.1 Rates

Both prevalence and incidence are used to describe the patterns of disease. Prevalence is the proportion of people in a population with a specific disease at a point in time (point prevalence) or over a given time interval (period prevalence). Incidence is the rate that new cases occur in a group who don’t have disease and are at risk over time.

A variety of rates are used to describe disease occurrence, including: (1) Crude rates for the total population, (2) Mortality—or death rates (3) Morbidity—or illness rates, (4) Case Fatality—% ill who die, or (5) Adjusted or stratified rates.

Stratification may be based on age, gender-specific, or age-adjusted (compared to a population with a standardized age distribution) and be used to adjust for confounders such as smoking, race, SES (socioeconomic status), ethnicity, or residence. Adjustment for age is important because many diseases are more common in the elderly, and this trend must be factored out to see the effect of environmental exposures if the exposed and unexposed populations have different age distributions. Standardized mortality ratios (SMR) are used for age-adjusted analysis as illustrated in Box 2.10 in a later section of the Appendix.

1.1.2 Statistical Techniques

Commonly used techniques include:

• Discrete data analysis—2 × 2 tables

• Point estimation, interval estimation

• Tests based on distributions: Poisson, Normal, Chi-square, Student’s t-test

• Linear Regression

• Logistic Regression

• Techniques involving time-series analyses such as survival analysis

It is important to be aware of the correct distribution to be used in different circumstances that may important in medical geology. Thus, the Poisson distribution is used to analyze the number of occurrences of a rare disease in a population. The Normal distribution is considered the “gold standard” for comparing properties (means, variance) of two populations. This can be used to see if the health effects observed in exposed and unexposed populations are statistically different. The Chi-square and Student t-distributions are used as approximations to the Normal distribution for actual sample populations based on sample data sets, which may be relatively small for high levels of environmental exposures. Logistic regression is used to describe relations between combinations of variables (e.g., exposure, diet, income), and a dichotomous variable (e.g., disease or no-disease state) and is described in Box 2.11 in a later section of the Appendix.

1.1.3 Statistical Significance

Statistical tests are used to evaluate the probability that an observed difference in disease rates between two populations (i.e., exposed vs nonexposed) is due to chance alone. The measure of probability that the difference is due to chance is expressed as a p value. The usual standard in assessing the statistical significance of an association is p < 0.05; which means that there is 5% chance that the observed difference is due to chance. In practice, statistical tests are used to evaluate whether or not the Null Hypothesis can be rejected. The Null Hypothesis states that the mean value of the parameter of interest is the same in the two populations being compared. A p value <0.05, therefore, means that there is a 95% chance that the Null Hypothesis can be rejected. If the Null Hypothesis cannot be rejected at the p level, or there is a “failure to reject the Null Hypothesis,” then this does not mean that the disease rates in the two populations are the same. Often, interpretation of such a “failure” is that the study design was inadequate. An alternative but equivalent approach in comparing the two populations is to examine the 95% confidence interval of the distribution of the calculated ratio of the disease rates of the two populations. If the 95% confidence interval does not include the value of 1.0, then the Null Hypothesis can be rejected with 95% certainty. The traditional requirement of the 95% confidence interval has recently been challenged, however. Wasserstein et al. (2019) and Amrhein et al. (2019) provide a summary of some of the concerns around the overreliance on the concept of statistical significance as commonly used in statistical studies.

1.1.4 Epidemiology Study Designs

The purpose of this section is to compare the most commonly used study designs. The relationships between time, exposure, and health status for cross-sectional, cohort, and case–control studies are illustrated in Fig. 2.10.

Fig. 2.10

Comparison of common epidemiologic study designs

1.1.4.1 Observational Studies

In cross-sectional studies , the observations of exposure and disease status are made at one point in time. These studies are quick and inexpensive but cannot observe the temporal relationships between exposure and disease. They are often used as a preliminary analysis using comprehensive surveys of many disease states and exposure and to see if the magnitude of the health effects and the association with exposure warrant further, more expensive studies. They can be used in studies of the correlation of environmental exposures and biomarkers for internal dose or precursors of disease. For example, Kurttio et al. (2002) studied the association of uranium concentrations in drinking water with uranium concentrations in urine in over 300 subjects.

In ecological studies , the unit of analysis is group or population (e.g., county or state). The average disease status and exposure are known only at group level. Advantages of ecological studies are that they are quick, inexpensive, and can study large populations. For example, the relationship between exposure to arsenic in drinking water and bladder cancer has been examined in a number of ecological studies (National Research Council 1999, 2001). Lamm et al. (2004), Siegel (2004); and Frost (2004) used county-level populations in ecological analyses of the relationship between bladder cancer mortality and levels of arsenic in public drinking water supplies. In Lamm et al. (2004), a dose–response relationship was obtained by calculating the mortality rates as a function of the mean arsenic concentrations in 133 counties in the US. In Siegel (2004) and Frost (2004), Standardized Mortality and Incidence Ratios (SMRs and SIRs) (see Box 2.10) were calculated for populations in US counties. In these studies, the exposed population included all people living in counties in the US where public water supplies had average arsenic concentrations greater than 10μg/L. The unexposed population included different reference populations such as all people living in counties adjacent to the “exposed counties” with public drinking water sources with average arsenic concentrations below 10μg/L. More recently, an ecological analysis of a community in Chile that was exposed to high levels of arsenic in drinking is described in Smith et al. (2012) and was summarized in Sect. 2.5.1.3.

Ecological studies are useful to explore initial trends and define hypotheses for later analytical studies, but they have severe limitations. The disadvantages are that the joint distribution of disease and exposure are not known within groups. This produces the ecological fallacy whereby the relationship between exposure and disease at the group level may not be same as at the individual level. Potential problems include misclassification of exposure status, temporal ambiguity (disease in individuals in the group could precede exposure due to migration), and confounding by other exposures such as smoking, diet, other exposure to risk factors. For example, Lamm et al. (2003) and Kayajanian (2003) illustrated how the approach taken to stratify or group different populations with respect to potential confounders or other factors (such as water source or exposure level) in ecological studies can obscure the actual underlying dose–response relationships or causal factors in studies of health effects due to arsenic exposure.

Box 2.10 Calculation of Standardized Mortality Ratios

Siegel et al. (2002) and Siegel (2004) used Standardized Mortality Ratios (SMR) in an ecological analysis of the relationship between bladder cancer mortality and levels of arsenic in public drinking water supplies. The exposed population included all people living in counties in the US where public water supplies had average arsenic concentrations greater than 10μg/L. Several reference unexposed populations were used, including (1) the entire 2000 Census United States population and (2) all people living in counties adjacent to the “exposed counties” with public drinking water sources with average arsenic concentrations below 10μg/L. Bladder cancer mortality rates were obtained from the WONDER Database (2003).

The Standardized Mortality Ratio (SMR) is the ratio of the observed deaths and the expected deaths due to bladder cancer. For example, it can be calculated as:

$$ \mathrm{SMR}=\varSigma {x}_i/\varSigma {X}_i, $$

where the sums are taken over for 10-year intervals between 20 and >85 years and:

• x_i is the observed bladder cancer deaths in the exposed population for 10-year age intervals for the period 1982–1998.

• X_i is the expected number of deaths from bladder cancer for the corresponding 10-year age stratum for the unexposed (reference) population based on Year 2000 US bladder cancer age-specific mortality.

The expected number of deaths in the ith age stratum is calculated as:

• X_i = n_i × R_i,

where in the case of using the US population as the reference:

• n_i is the exposed population in the ith age stratum (in units of people × years).

• R_i is the age-specific bladder US national cancer mortality rate for the ith age stratum.

In general, if the lower limit of the 95% confidence interval for the SMR >1.00, then the risk is significantly elevated at the 0.05 significance level. In this study, however, this condition was not observed, the Null Hypothesis was not rejected, and increased risk for bladder cancer was not demonstrated for populations exposed to arsenic concentrations greater than 10μg/L.

1.1.4.2 Analytical Studies

In analytical studies , the disease and exposure status are known at the individual level. They can be used to test casual hypotheses and design intervention studies. In cohort studies, exposed and unexposed populations are identified, and the development of disease is followed over future time (prospective studies) or by reconstructing historical mortality or morbidity rates from vital statistics databases (retrospective studies). Compared to ecologic studies, these studies have less bias in exposure assessment, have known temporal relations, can measure disease incidence, and can monitor several health outcomes for a single exposure. Their disadvantages may include the required large size, long study times, high costs, and subject to loss of follow-up. In cohort studies, the relative risk is commonly calculated as:

• RR = (disease incidence in the exposed population)/ (incidence in the unexposed population).

In case–control studies, the cases of a disease in a population are identified and controls are identified who are comparable to the cases but lack the disease. The exposures of the cases and controls to the hazard of interest are then compared to estimate the odds ratio , which is calculated

• OR = (odds of exposure in the cases)/(odds of exposure in the controls).

Case–control studies are good for rare diseases and are relatively quick and inexpensive to carry out. They may suffer from biases in the selection of cases and controls and from information bias. Odds ratios can be obtained from discrete data using simple methods such as 2 × 2 tables. Logistic regression can also be used to obtain estimates of odds ratios as described in Box 2.11.

Box 2.11 Logistic Regression

Logistic regression is a mathematical modeling approach that can be used to describe the relationship of several independent variables to a dichotomous dependent variable. For example, in many epidemiological studies, the dichotomous dependent variable is the probability that has someone has a disease or not. The value of the dependent variable can vary between zero and one. The logistic function is shown in the illustration here.

The independent variables can be either categorical, ordinal, or continuous. For example, the independent variables can be gender, age, a level of exposure, or some other variable that is related to disease status.

The logistic model is derived from the logistic function, by writing z as a linear sum as shown in the figure. Here the X_i’s are the independent variables of interest and a and b are constant terms representing unknown parameters. These unknown parameters are estimated based on data obtained on the X_i’s for a group of subjects in the epidemiological study. Thus, if we knew the parameters a and b and we had determined the values of the X_i’s for a particular disease-free individual, then we could use the formula to obtain the probability that this individual would develop a disease over some defined time interval. For cohort studies, the risk can be estimated from the logistic function; however, for case–control studies, only the odds ratios, not individual risks, can be estimated from logistic model. Whereas the risk is defined as the probability that the disease will occur, the odds ratio is the ratio of the probability that the disease will occur over the probability that the disease will not occur. Detailed description of this method can be found in standard biostatistics books such as Kleinbaum (1994).

1.1.4.3 Strength and Significance of Associations

Both the strength of association (rate ratio) and its statistical significance must be considered in evaluating the results of an epidemiological study. Box 2.12 summarizes relationship between the rate (odds or risk) ratios and descriptions of strength of association as commonly accepted in epidemiology. The statistical significance of the association is evaluated by testing the Null Hypothesis (H_o): that no relationship exists between exposure and disease (e.g., the disease rate in the exposed population equals the disease rate in the unexposed population for cohort studies). Both the t-test for continuous variables and the Chi-square test for discrete observations are used in the hypothesis testing.

Box 2.12 Ratio Strengths

1.1.4.4 Confounding, Bias, and Interactions

Evaluation of the strength of association in epidemiology studies must also include consideration of likely bias, existence of confounders , and effect modification. Bias is a systematic error in the study that affects the validity of the measured association. Both selection bias in enrollment of cases and controls as well as misclassification bias , which can affect assignment of exposure and disease status, can be important. Other types of bias are described in Table 2.5. Confounding is the misleading appearance of association between the exposure variable and disease status due to the effect of another factor, which is associated with both the exposure and the disease status. A classic example of confounding is the apparent association between coffee drinking and pancreatic cancer, which is strongly influenced by the independent associations of smoking with both coffee drinking and the cancer. Confounding can be evaluated or corrected by a variety of techniques, including randomization, stratification (separate analyses on homogeneous subsets of the sample population that differ by the value of the confounding factor) or use of logistic regression, which controls for confounding factors through separate terms in the logistic equation. For example, incidence or prevalence of many diseases is related to age; therefore, subjects are often separated into distinct age groups (stratification) or disease rates are compared to those of population with standard age distribution (age-adjustment). The calculation of Standardized Mortality Rates (SMR) is an example of age adjustment and is illustrated in the Box 2.10 above.

Table 2.5 Examples of bias important in environmental epidemiology

Effect modification or interaction results when the effect of the exposure is changed by the presence of another exposure variable. The magnitude of the combined effect of the two exposures differs from the effect expected from each exposure alone (i.e., multiplicative and not additive). For example, the rates of lung cancer associated with exposure to both asbestos and smoking are greater than the sum of the individual risks.

1.1.4.5 Causal Criteria

Even if an association is strong, statistically significant, and the study is free of the factors described here, the association must be evaluated against recognized causal criteria before concluding that a cause–effect relationship has been established. The classic set of causal criteria was proposed by Hill (1965) and is described in Table 2.6. In medical geology studies, the addition of the requirement of geological plausibility is important.

Support for a proposed cause–effect relationship can also be reinforced by consideration and rejection of alternative explanations as well as by the use of different models for causality. Underlying models for causality in environmental epidemiology studies include (1) the host–agent–environment model, (2) the causal chain model (agent factors + person factors + place factors+ time factors), (3) web of causation model (disease develops as result of chains of causation composed of many links, which are the result of many antecedents), and (4) causal pies based on the multifactorial nature of causation, where some factors are identified as sufficient and others as necessary for the disease to occur .

Table 2.6 Causal criteria in epidemiologic studies in medical geology

1.2 Evaluation of Epidemiological Studies

Evaluation of an epidemiological study should include the following questions:

• What was the outcome (health effect)? How assessed?

• What was the exposure? How assessed?

• What was the study design?

• What was the study population (cases and controls)?

• What was main result? Statistically significant?

• Are results likely to be influenced by bias?

• Was confounding considered and controlled for?

• Which causal criteria are addressed?

In addition to the use of clinical manifestations of disease as an end point, the current use of -omics technologies allows the use of many other endpoints in epidemiological studies. Table 2.3 describes examples of the different endpoints used in some studies and their associated study designs related to geogenic contaminants such as arsenic and uranium.