Other articles in this chapter present general principles of medical surveillance of occupational illnesses and exposure surveillance. This article outlines some principles of epidemiological methods that may be used to fulfil surveillance needs. Application of these methods must take into account basic principles of physical measurement as well as standard epidemiological data-gathering practice.
Epidemiology can quantify the association between occupational and non-occupational exposure to chemico-physical stressors or behaviour and disease outcomes, and can thus provide information to develop interventions and prevention programmes (Coenen 1981; Coenen and Engels 1993). Availability of data and access to workplace and personnel records usually dictate the design of such studies. Under the most favourable circumstances, exposures can be determined through industrial hygiene measurements that are carried out in an operating shop or factory, and direct medical examinations of workers are used to ascertain possible health effects. Such evaluations can be done prospectively for a period of months or years to estimate risks of diseases such as cancer. However, it is more often the case that past exposures must be reconstructed historically, projecting backwards from current levels or using measurements recorded in the past, which may not completely meet informational needs. This article presents some guidelines and limitations for measurement strategies and documentation that affect epidemiological assessment of workplace health hazards.
Measurements should be quantitative wherever possible, rather than qualitative, because quantitative data are subject to more powerful statistical techniques. Observable data are commonly classified as nominal, ordinal, interval and ratio. Nominal level data are qualitative descriptors which differentiate only types, such as different departments within a factory or different industries. Ordinal variables may be arranged from “low” to “high” without conveying further quantitative relationships. An example is “exposed” vs. “unexposed”, or classifying smoking history as non-smoker (= 0), light smoker (= 1), medium smoker (= 2) and heavy smoker (= 3). The higher the numerical value, the stronger the smoking intensity. Most measurement values are expressed as ratio or interval scales, in which a concentration of 30 mg/m3 is double the concentration of 15 mg/m3. Ratio variables possess an absolute zero (like age) while interval variables (like IQ) do not.
Measurement strategy takes into account information about the measurement site, the surrounding conditions (e.g., humidity, air pressure) during the measurement, the duration of the measurement and the measurement technique (Hansen and Whitehead 1988; Ott 1993).
Legal requirements often dictate measurement of eight-hour time-weighted averages (TWAs) of levels of hazardous substances. However, not all individuals work eight-hour shifts all the time, and levels of exposures may fluctuate during the shift. A value measured for one person’s job might be considered representative of an eight-hour shift value if the exposure duration is longer than six hours during the shift. As a practical criterion, a sampling duration of at least two hours should be sought. With time intervals that are too short, the sampling in one time period can show higher or lower concentrations, thereby over- or underestimating the concentration during the shift (Rappaport 1991). Therefore, it can be useful to combine several measurements or measurements over several shifts into a single time-weighted average, or to use repeated measurements with shorter sampling durations.
Surveillance data must satisfy well-established criteria. The measurement technique should not influence the results during the measurement process (reactivity). Furthermore, the measurement should be objective, reliable and valid. The results should not be influenced either by the measurement technique used (execution objectivity) or by the reading or documentation by the measurement technician (assessment objectivity). The same measurement values should be obtained under the same conditions (reliability); the intended thing should be measured (validity) and interactions with other substances or exposures should not unduly influence the results.
Quality of Exposure Data
Data sources. A basic principle of epidemiology is that measurements made at the individual level are preferable to those made at the group level. Thus, the quality of epidemiolological surveillance data decreases in the following order:
- direct measurements taken of persons; information on exposure levels and time progression
- direct measurements taken of groups; information on current exposure levels for specific groups of workers (sometimes expressed as job-exposure matrices) and their variation over time
- measurements abstracted or reconstructed for individuals; estimation of exposure from company records, purchasing lists, descriptions of product lines, interviews with employees
- measurements abstracted or reconstructed for groups; historical estimation of group-based exposure indexes.
In principle, the most precise determination of the exposure, using documented measurement values over time, should always be sought. Unfortunately, indirectly measured or historically reconstructed exposures are often the only data available for estimating exposure-outcome relationships, even though considerable deviations exist between measured exposures and exposure values reconstructed from company records and interviews (Ahrens et al. 1994; Burdorf 1995). The quality of the data declines in the order exposure measurement, activity-related exposure index, company information, employee interviews.
Exposure scales. The need for quantitative monitoring data in surveillance and epidemiology goes considerably beyond the narrow legal requirements of threshold values. The goal of an epidemiological investigation is to ascertain dose-effect relation-ships, taking into account potentially confounding variables. The most precise information possible, which in general can be expressed only with a high scale level (e.g., ratio scale level), should be used. Separation into larger or smaller threshold values, or coding in fractions of threshold values (e.g., 1/10, 1/4, 1/2 threshold value) as is sometimes done, essentially relies on data measured on a statistically weaker ordinal scale.
Documentation requirements. In addition to information on the concentrations and the material and time of measurement, external measurement conditions should be documented. This should include a description of the equipment used, measurement technique, reason for the measurement and other relevant technical details. The purpose of such documentation is to ensure uniformity of measurements over time and from one study to another, and to permit comparisons between studies.
Exposure and health outcome data gathered for individuals are usually subject to privacy laws that vary from one country to another. Documentation of exposure and health conditions must adhere to such laws.
Epidemiological studies strive to establish a causal link between exposure and disease. Some aspects of surveillance measurements that affect this epidemiological assessment of risk are considered in this section.
Type of disease. A common starting point for epidemiological studies is the clinical observation of a surge in a particular disease in a company or area of activity. Hypotheses on potential biological, chemical or physical causal factors ensue. Depending on the availability of data, these factors (exposures) are studied using a retrospective or prospective design. The time between the beginning of the exposure and the onset of the disease (latency) also affects study design. The range of latency can be considerable. Infections from certain enteroviruses have latency/incubation times of 2 to 3 hours, whereas for cancers latencies of 20 to 30 years are typical. Therefore, exposure data for a cancer study must cover a considerably longer period of time than for an infectious disease outbreak. Exposures which began in the distant past can continue up to the onset of disease. Other diseases associated with age, such as cardiovascular disease and stroke, can appear in the exposed group after the study begins and must be treated as competing causes. It is also possible that people classified as “not sick” are merely people who have not yet manifested clinical illness. Thus, continued medical surveillance of exposed populations must be maintained.
Statistical power. As previously stated, measurements should be expressed on as high a data level (ratio scale level) as possible in order to optimize the statistical power to produce statistically significant results. Power in turn is affected by the size of the total study population, the prevalence of exposure in that population, the background rate of illness and the magnitude of risk of the disease that is caused by the exposure under study.
Mandated disease classification. Several systems are available for codifying medical diagnoses. The most common are ICD-9 (International Classification of Diseases) and SNOMED (Systematic Nomenclature of Medicine). ICD-O (oncology) is a particularization of the ICD for codifying cancers. ICD coding documentation is legally mandated in many health systems throughout the world, especially in Western countries. However, SNOMED codification can also codify possible causal factors and external conditions. Many countries have developed specialized coding systems to classify injuries and illnesses that also include the circumstances of the accident or exposure. (See the articles “Case study: Worker protection and statistics on accidents and occupational diseases—HVBG, Germany” and “Development and application of an occupational injury and illness classification system”, elsewhere in this chapter.)
Measurements that are made for scientific purposes are not bound by the legal requirements that apply to mandated surveillance activities, such as determination of whether threshold limits have been exceeded in a given workplace. It is useful to examine exposure measurements and records in such a way as to check for possible excursions. (See, for example, the article “Occupational hazard surveillance” in this chapter.)
Treatment of mixed exposures. Diseases often have several causes. Therefore it is necessary to record as completely as possible the suspected causal factors (exposures/confounding factors) in order to be able to distinguish the effects of suspected hazardous agents from one another and from the effects of other contributory or confounding factors, such as cigarette smoking. Occupational exposures are often mixed (e.g., solvent mixtures; welding fumes such as nickel and cadmium; and in mining, fine dust, quartz and radon). Additional risk factors for cancers include smoking, excess alcohol consumption, poor nutrition and age. Besides chemical exposures, exposures to physical stressors (vibration, noise, electromagnetic fields) are possible triggers for diseases and must be considered as potential causal factors in epidemiological studies.
Exposures to multiple agents or stressors may produce interaction effects, in which the effect of one exposure is magnified or reduced by another that occurs contemporaneously. A typical example is the link between asbestos and lung cancer, which is many times more pronounced among smokers. An example of the mixture of chemical and physical exposures is progressive systemic scleroderma (PSS), which is probably caused by a combined exposure to vibration, solvent mixtures and quartz dust.
Consideration of bias. Bias is a systematic error in classifying persons in the “exposed/not exposed” or “diseased/not diseased” groups. Two types of bias should be distinguished: observation (information) bias and selection bias. With observation (information) bias, different criteria may be used to classify subjects into the diseased/not diseased groups. It is sometimes created when the target of a study includes persons employed in occupations known to be hazardous, and who may already be under increased medical surveillance relative to a comparison population.
In selection bias two possibilities should be distinguished. Case-control studies begin by separating persons with the disease of interest from those without that disease, then examine differences in exposure between these two groups; cohort studies determine disease rates in groups with different exposures. In either type of study, selection bias exists when information on the exposure affects classification of subjects as sick or not sick, or when information on disease status affects classification of subjects as exposed or not exposed. A common example of selection bias in cohort studies is the “healthy worker effect”, which is encountered when disease rates in exposed workers are compared with those in the general population. This can result in underestimation of disease risk because working populations are often selected from the general population on the basis of continued good health, frequently based upon medical examination, whereas the general population contains the ill and infirm.
Confounders. Confounding is the phenomenon whereby a third variable (the confounder) alters the estimate of an association between a presumed antecedent factor and a disease. It can occur when the selection of subjects (cases and controls in a case-control study or exposed and unexposed in a cohort study) depends in some way upon the third variable, possibly in a manner unknown to the investigator. Variables associated only with exposure or disease are not confounders. To be a confounder a variable must meet three conditions:
- It must be a risk factor for the disease.
- It must be associated with the exposure in the study population.
- It must not be in the causal pathway from exposure to disease.
Before any data are collected for a study it is sometimes impossible to predict whether or not a variable is a likely confounder. A variable which has been treated as a confounder in a previous study might not be associated with exposure in a new study within a different population, and would therefore not be a confounder in the new study. For instance, if all subjects are alike with respect to a variable (e.g., sex), then that variable cannot be a confounder in that particular study. Confounding by a particular variable can be accounted for (“controlled”) only if the variable is measured along with exposure and illness outcomes. Statistical control of confounding may be done crudely using stratification by the con-founding variable, or more precisely using regression or other multivariate techniques.
The requirements of measuring strategy, measuring technology and documentation for industrial workplaces are sometimes statutorily defined in terms of threshold limit value surveillance. Data protection regulations also apply to the protection of company secrets and person-related data. These requirements call for the comparable measuring results and measurement conditions and for an objective, valid and reliable measuring technology. Additional requirements put forward by epidemiology refer to the representativeness of measurements and to the possibility of establishing links between exposures for individuals and subsequent health outcomes. Measurements may be representative for certain tasks, i.e. they may reflect typical exposure during certain activities or in specific branches or typical exposure of defined groups of persons. It would be desirable to have measurement data directly attributed to the study subjects. This would make it necessary to include with measurement documentation information about persons working at the concerned workplace during the measurement or to set up a registry allowing such direct attribution. Epidemiological data collected at the individual level are usually preferable to those obtained at the group level.