Navigation and service

Working methods used in radiation epidemiological research

  • Epidemiological studies are observational studies in humans under real environmental conditions.
  • There are four major types of studies of varying validity in epidemiological research: Cohort studies, case-control studies, ecological studies and cross-sectional studies.
  • There are defined criteria for assessing the validity of an epidemiological study and for the question about the cause-effect-relationship.

Epidemiology (derived from Greek epi "on, upon", demos "people", logos "study") is a scientific discipline studying the distribution of diseases in a population (descriptive epidemiology) and the factors influencing this distribution (analytical epidemiology). The designation of various subgroups of epidemiology is based on the investigated disease or influencing factor (the exposure):

  • relating to diseases such as infection epidemiology, cardiovascular or cancer epidemiology,
  • relating to influencing factors such as occupational, environmental, radiation, nutritional or genetic epidemiology.

Epidemiological studies are observational studies in humans

Epidemiological studies are observational studies in humans under real environmental conditions. They are fundamentally different from experimental studies in which test subjects are randomly exposed under controlled laboratory conditions. In the case of so-called "double blind" studies neither the scientist nor the subject knows the exposure status during the experiment.

Such experimental studies on the cause-effect relationship between an exposure as the cause and a disease as the effect can often not be conducted in humans for ethical and practical reasons. In these situations, estimates of the risk of disease in humans are based on observational studies.

In the field of radiation epidemiology, epidemiological studies in humans allow direct estimates of the risk of disease for a population group exposed to radiation. To this end, it is fundamentally necessary to determine the frequency of certain diseases or causes of death in the population groups investigated as well as to ascertain the past radiation exposure of every individual. Based on standardised study designs, relationships between the occurrence of diseases (incidence) or the causes of death (mortality) and the level of radiation exposure can be demonstrated. On this basis the radiation risk can then be estimated.

Radiation epidemiological studies

The investigations on the atomic bomb survivors of Hiroshima and Nagasaki remain the most important foundation of the knowledge on radiation risk. This knowledge is increasingly supported and enhanced by investigations on other population groups exposed to radiation. Radiation exposures occur for example during medical diagnostics, in the workplace, at home or as a result of accidents.

The Federal Office for Radiation Protection (BfS) contributes to enhancing knowledge in the field of radiation epidemiology

  • on the one hand by conducting its own studies,
  • on the other hand by commissioning research projects to third parties.

The largest study conducted by the BfS, is the Wismut uranium miners cohort study. It is one of the world's largest cohort study on uranium miners occupationally exposed to radon.

Study types

There are four major study types of varying validity in epidemiological research:

  • Cohort studies are very elaborate but also have the greatest validity. They are based on individual exposures and address specific diseases.
  • Case-control studies can also provide valid results based on smaller population groups with regard to causal relationships. Due to their study design, they are limited to the investigation of one single disease per study. They are based on the disease and retrospectively enquire past exposures.
  • Ecological studies such as geographical correlation studies are very prone to error and are difficult to interpret in the sense of causation. They can provide indications of possible causes but are basically not suitable for risk estimation.
  • In cross-sectional studies the exposure and the disease to be investigated are put into correlation simultaneously and are analysed. These studies are only useful for relatively common diseases and are generally only suitable for generating hypotheses.

All study types have to be planned, conducted and analysed carefully in order to provide meaningful and interpretable results.

In the following, the four major types are presented in detail:

Cohort study show / hide

A cohort study investigates a defined group of individuals exposed to a risk factor to varying degrees. In the process, the study participants or cohort members are observed over time. The research question in a cohort study is for example whether there is a higher incidence of certain diseases in higher-exposed individuals than in less-exposed or non-exposed individuals. At the beginning of the study which may also be in the past, only individuals who do not suffer from the disease to be investigated are admitted to the cohort.

In an observation procedure, the so-called follow-up, the disease state of the individuals is observed over time. Diseases or causes of death are recorded. The cohort members' exposures are considered as potential risk factors for the diseases or cases of death and are determined individually. This makes it possible to investigate incidence or mortality rates depending on exposure.

If the cohort members have similar levels of exposure, it is not possible to form different exposure groups within the cohort. Then it is necessary to compare the cohort with a control group outside the cohort (external control group). This control group has to be comparable in regard to other factors influencing the occurrence of the relevant diseases (e.g. age and gender) to the largest possible extent. If this is not possible, such other factors - so-called confounding variables - may result in a distortion of the observed relationship between exposure and risk of disease. The general population is often used as an external control group.

If the anticipated effect of an exposure is small, the cohorts may have to comprise several tens of thousands of individuals and the follow-up period may have to extend from several years to decades in order to achieve the statistical power required to establish a relationship.

Statistical parameters of cohort studies are the relative risk (RR), when comparisons are drawn between various groups of the cohort, or the standardised incidence or mortality ratio (SIR / SMR), when the analyses are based on a comparison with the total population. An example for a cohort study is the Wismut uranium miners cohort study on almost 60,000 former miners in Saxony and Thuringia conducted by the BfS. This study investigates mortality risks depending on past radiation exposures.

Case-control study show / hide

A case-control study is useful when it is not practicable to conduct a cohort study. A case-control study answers the question whether diseased individuals had been exposed more often or more exposed than comparable non-diseased individuals.

For a case-control study, individuals suffering from the disease to be studied (cases) are initially selected from a given population. For every case, one or several individuals are selected who do not suffer from the disease to be studied but belong to the same population group as the cases. They are referred to as controls.

The cases are then compared with the controls in regard to exposure. The main statistical parameter of case-control studies is the odds ratio (OR), which is a good approximation of the relative risk for rare diseases like tumours. Just as with cohort studies, bias in selecting the controls has to be avoided as far as possible also in case-control studies. The controls should at least be comparable to the cases relating to age and gender.

As only cases and the corresponding controls are investigated, case-control studies can be conducted with considerably smaller numbers of individuals than cohort studies. They are useful in the case of rare diseases or when the collection of data on individual exposures requires detailed and extensive fieldwork. Case-control studies are sometimes combined with cohort studies in order to collect more detailed data on specific subgroups of the cohort than on the other members of the cohort. This type of study is referred to as a case-control study nested into a cohort.

Ecological study show / hide

The term "ecological" studies refers to studies using aggregated data categorised according to place or time instead of individual data on exposure and disease. The correlation between exposure and disease is determined by means of these aggregated data. One example for this is the comparison of the lung cancer mortality rate in the administrative districts in Germany with the average radon concentration measured in housing in these administrative districts (Wichmann 1999).

Advantages of these studies are that they are cheap and easy to perform. Aggregated data on diseases or causes of death categorised according to region and time are often available from official statistical analyses, and corresponding data on average radiation exposure levels can easily be determined from existing measurement programmes.

Major disadvantages of these studies are, however, that no individual data are available, but only aggregated data at group level and that it is assumed that these groups only differ with regard to the (radiation) exposure to be investigated. Possible differences in the distribution of other risk factors (e.g. age, smoking) are often not considered. For this reason, the probability of drawing wrong conclusions from ecological studies is high.

This problem is illustrated by the example of radon in housing and lung cancer. An ecological study in Germany would wrongly suggest that the higher the average radon concentration per region, the lower the lung cancer mortality rate. The reason for this is the low proportion of smokers in areas with high average radon concentrations (rural, mountainous regions) contrasted to a high proportion of smokers in areas with low radon concentrations (in particular, large cities with a high proportion of tower blocks). In this situation there is a so-called negative correlation between the proportion of smokers in the population and the level of radon exposure.

In areas with a high proportion of smokers and low radon exposure levels, the lung cancer risk is higher than in areas with a low proportion of smokers but high levels of radon exposure. This is why the lung cancer risk due to radon exposure is not identified in a study without any individual information on smoking. Such a spurious correlation is also referred to as an "ecological fallacy".

Ecological studies are very prone to error and are not really suitable to identify causal relationships. They primarily serve to generate hypotheses.

Cross-sectional study show / hide

In a cross-sectional study, data on radiation exposure (e.g. the measured exposure due to the electromagnetic fields of mobile phone base stations) and the disease of interest (e.g. sleep disturbance or headache) are collected simultaneously and are put into correlation. These studies are only useful for relatively common diseases. The validity of this type of study is very limited in regard to assessing whether the exposure is a cause of the disease, as it is not clear whether the disease has occurred before or after the exposure. These studies therefore primarily serve to generate hypotheses.

Assessing the qualitiy of an epidemiological study

There are defined criteria for assessing the quality of an epidemiological study. The German Society for Epidemiology (DGEpi) has drawn up Guidelines and Recommendations to Assure Good Epidemiologic Practice. They contain a list of general criteria of quality for epidemiological studies. Particularly three main potential sources of error may lead to distorted risk estimates in epidemiological studies:

Selection bias show / hide

The term selection bias refers to a type of distortion resulting from a non-random selection of study participants. The willingness to participate in a cross-sectional study on health complaints due to electromagnetic fields from base stations, for example, might be higher in diseased individuals or individuals with impaired well-being than in healthy individuals. This difference in the willingness to participate may lead to overestimation of the actual risk.

Information bias show / hide

The term information bias refers to a type of distortion which may arise from a misclassification of exposure or disease. When estimating exposure, it is important to capture the exposure of interest as accurately as possible (validity) and to perform this task reliably (reliability). Moreover, it is mandatory to ascertain the disease state in an objective manner.

Confoundingshow / hide

A factor is referred to as a confounder or a confounding variable when the factor itself is a risk factor for the disease to be investigated and when it is also correlated with the exposure to be investigated, that is, with the actual risk factor of interest. The existence of such disturbance variables may result in under- or overestimation of the risk.

When assessing the quality of a study, particular attention has to be given to these potential bias mechanisms. The better such bias mechanisms are eliminated, the more validity a study has. There are methods to control for selection and information bias as well as for confounding.

From correlation to causal interpretation

That a risk factor is the cause of a disease cannot be proven by epidemiological studies in a formal sense. Rather, as much evidence as possible is gathered to prove that causation by the risk factor is the most probable explanation for the relationship. The prerequisite for a risk factor to be considered as the actual cause of a disease are the following three conditions:

  • the exposure to the risk factor precedes the disease;
  • a change in exposure is associated with a change in incidence rate (presence of a dose-response relationship, that is, the incidence rate increases with increasing exposure);
  • the relationship between the risk factor and the disease is not the result of a relationship between these factors and a third factor.

In the field of radiation epidemiology, e.g. the cause-effect-relationship between the exposure to radon and its decay products and developing lung cancer is regarded as proven.

In addition to the conditions mentioned above, the following criteria also play an important role:

  • Consistency of the findings (that is, the relationship found is also reproducible in other populations and in different study types - for these studies, coincidence, confounding or bias that could explain the observed risk should be excluded as far as possible);
  • If the findings are consistent with existing knowledge and plausible from a biological point of view, the more likely the relationship is to be causal.

Biological plausibility show / hide

Results from epidemiological research are biologically plausible when they can be explained by means of a biological mechanism of action based on results from experimental cell and animal tests. Often however, corresponding biological investigations are only initiated after epidemiological findings. The question concerning the biological plausibility of an epidemiological finding can therefore often not yet be answered upon its first availability.

Statistical significance show / hide

The criterion of statistical significance is also often regarded as necessary - sometimes it is even wrongly considered to be sufficient - for a causal interpretation of results. In other words, statistical significance is an indication of a cause-effect-relationship but certainly no proof of it, taken by itself.

When dealing with small risks, individual studies alone cannot necessarily provide significant results. The question to be asked here is rather whether the results from various investigations on the same question are inherently consistent - irrespective of whether each investigation reaches statistical significance on its own.

The insufficient size of individual studies for proving small risks is often compensated for by conducting several studies using the same study design and subsequently analysing the data in a combined analysis (so-called pooled analysis). An important example of this is the combined analysis of various case-control studies on radon in homes and the associated lung cancer risk conducted in Europe and North America (Darby et al., 2005).

As early as 1965, Sir Austin Bradford Hill formulated criteria according to which causation can be inferred from a statistical correlation.


Further information on how to obtain a causal interpretation of the results from a statistical correlation observed in an epidemiological study can be found in

State of 2017.01.10

© Bundesamt für Strahlenschutz