Epidemiology is an observational science. Its purpose is to examine hypotheses dealing with the distribution and causes of disease onset in a given population. An epidemiological study searches for a statistical association between a given factor and the emergence of a disease, and then determines the importance of this association. Epidemiological studies are sensitive to several types of bias.
Advantages of epidemiological studies
- Focus = humans
- Exposure to agent = real situation
- Endpoints: mortality & morbidity
- Potentially hypersensitive subjects can be investigated
- Acute and chronic exposures can be studied
Limitations of epidemiological studies
- Difficult to demonstrate causality
- Difficult to take all ‘confounders’ into account
- Difficult to obtain accurate individual measurements of real exposure
- Epidemiological studies are very expensive and time consuming (especially cohort studies)
To study the influence of the factor “electric field” and/or “magnetic field” on a disease, the epidemiologist carries out investigations on relevant populations. Two main types of epidemiological studies are usually carried out : analytical and experimental (trials) epidemiology.
1. Ecological studies
Ecological studies focus on the comparison of groups, rather than individuals: they study the association (correlation) between exposure variables and health, when researchers do not have any individual data.
Ecological studies do not reflect the exposure or the health of each individual in the group but the average level of exposure and the health of populations. Researchers can study the same population at different times (temporal variations) or more populations of different geographical areas during the same period (geographic variation).
For example, this kind of survey can be used to study the relationship between the concentrations of air pollutants (CO2, ozone,…) and mortality collected in the following days from hospital data and death certificates.
Even if the results are not precise at individual level, ecological studies are interesting because they are quickly set up and quite inexpensive since based on already existing data. It can provide the base to build other studies as case control or cohort studies. They can generate hypotheses.
2. Case control studies
One selects a group of subjects that have the studied disease (cases), and a group of subjects without this disease (controls). For each subject of the investigation, one will search for information concerning exposure to the risk factors in their relevant past. For this reason, case control studies are qualified retrospectives, since the studied disease has already occurred when one searches for the earlier exposure to the risk factor. One then compares the exposure to the risk factor in both cases and controls. The advantage of this protocol is that it is inexpensive and feasible within a short time. Its principal disadvantage comes from the difficulty of rebuilding the story of cases and controls in a comparable way and without bias. The measurement of the association is called the odds-ratio (OR). Case-control studies are a good type of study for rare diseases.
3. Cohort studies (exposed versus non exposed)
During a given period, one supervises a group of people exposed to a risk factor as well as a group similar to the first, but not exposed to the studied factor. The appearance frequencies of the disease in the two groups are compared.
- Retrospective cohort studies
In retrospective studies, researchers are looking in the past for the existence of a disease in two groups of people as alike as possible excepted for their exposure to EMF. In retrospective cohort studies, the measurement of association is called the relative risk (RR).
- Prospective cohort studies :
Because researchers are waiting for the appearance of the disease with the passage of time, this type of study is called a prospective study.
The advantage of prospective cohort studies is that it allows a better control of bias. Disadvantages are the high cost and difficulty of carrying out this type of study when the disease is rare or occurs after a long latency period. The measurement of association is called the relative risk (RR).
Trials: Experimental epidemiology
The term “Experimental” means that, contrarily to cohort studies, researchers control the exposure conditions of the subjects. Groups exposed and unexposed are monitored and compared with respect to the impact of the event studied. The assignment of a subject to a group or the other is randomized.
When properly conducted, these studies represent the ideal model to study the relationship between exposure to an agent and the occurrence of a disease, since the groups being compared differ only by one characteristic: the exposure. However, this approach is not always possible, often for ethical reasons if the exposure to which a group of subjects should be submitted is potentially harmful. Trials are mostly used to control the effectiveness of interventions (e.g. drugs).
Meta-analysis is a statistical technique that gathers the data of comparable epidemiological studies in order to analyse them and evaluate the coherence of the obtained results.
The odds ratio (case control studies) corresponds to the risk exposure of cases compared with the risk exposure of controls.
The relative risk (exposed – unexposed studies) corresponds to the risk of exposed people to the studied factor compared with the risk of non-exposed people.
If the odds ratio or the relative risk is equal to 1, this indicates a lack of increased risk in the group of cases or the exposed population. The closer the odds ratio or relative risk is to 1, the lower the risk.
The confidence interval indicates the degree of accuracy with which one measures the odds ratio or the relative risk. A 95 % confidence interval (CI 95 %) means that this interval contains the true value of the relative risk or odds ratio with a probability of 95 %. The odds ratio or the relative risk is considered as significant when the confidence interval does not contain the value 1.
Example: If a relative risk is equal to 2.7 with a confidence interval at 95 % of (2.3 – 3.1), the risk is significant, since the lower limit of the interval is higher than 1. On the other hand, a relative risk of 1.4 with a confidence interval to 95 % of (0.9 – 1.9) is not significant because the value 1 is contained in the confidence interval.
Epidemiological studies cannot determine a clear relationship of cause and effect. If one finds an association between a factor and a disease, that does not mean that this agent caused the disease. Establishing a relationship of cause and effect requires the checking of several criteria:
- strength of the association: the causal nature of an association will be all the more probable since the value of the relative risk or the odds ratio is high;
- specificity of the association: a given exposure specifically involves a given pathology;
- constancy of association and reproducibility: it is necessary to find the same results in several investigations and different populations;
- coherence with results of the studies already published in the scientific literature;
- temporal relation: exposure to the alleged causal factor must precede appearance of the disease;
- dose response relationship: the more significant the exposure, the greater the probability of an effect on health;
- plausibility of the biological mechanism highlighted.
Information bias concerns the estimation and measurement of parameters that influence the living organism. After 30 years of research, scientists have not managed to establish factor(s) of exposure to be studied in order to understand biological effects:
- What should be measured or calculated? electric field, magnetic field, electrical consumption, wiring code…
- Which parameters should be measured? peak, average, median, an accumulated dose…
- How long is it necessary to measure? specific measurement, 24h, 1 week…
- Where is it necessary to measure? inside the house, in front of the house, in the bedroom, at the workplace …
- When is it necessary to measure? during the day, during leisure time, at night
- Is the continuous or variable character of our exposure significant ?
- Which threshold should be chosen? 0.2 µT? 0.3 µT? 0.4 µT or higher?
Many studies are carried out with a threshold of 0.2 µT. Dr David Savitz first chose this threshold of 0.2 µT to establish a distinction between exposed people and non exposed people (Savitz, 1988). The goal was not to define a level of security but to establish a threshold for the study (Lynch C, 1997). The studies that followed were carried out with this threshold or other thresholds: 0.3 µT, 0.4 µT.
This relates to :
- the under-representation of subjects of under-priviledged socio-economic levels when the choice of controls is carried out by drawing names from a telephone list,
- for certain studies, there is a need for a given stability of housing for the controls: this involves lesser mobility of the controls than of the cases,
- refusal to answer a questionnaire or to authorize the measurement of fields inside the residence: non respondents can then be different from those who agree to take part in a study.
For domestic exposure, this bias primarily refers to the studies concerning the evaluation of fields radiating from surrounding power lines. High tension power lines are not laid out randomly in cities: they are often located in places where traffic congestion is considerable, air pollution is significant, and socio-economic status is low. Potential confounding factors (e.g. physical, chemical, genetic, nutritional, etc.) are numerous.
In the occupational environment, potential confounding factors frequently occur. In addition to the usual factors, such as socio-demographic characteristics, smoking, alcohol consumption, or general employment conditions, few studies consider factors such as organic solvents, poly-chlorinated biphenyls, welding fumes, or ionising radiation, which often characterize jobs exposed to electromagnetic fields (Knave B, 1988 and Gallagher RP, 1990).
Epidemiological studies suggesting an association are generally published in the scientific literature. On the other hand, epidemiological studies indicating a lack of association are not consistently published.
Documents & Links
- See information on Research methods in PDF (906 Ko)
- Principles of Epidemiology in Public Health Practice, Third Edition – An Introduction to Applied Epidemiology and Biostatistics: Centers for disease control and prevention