Search

This chapter reviews the options for population selection in undertaking investigations aimed at estimating the occurrence of a disease. In terms of measuring disease occurrence, an important issue will be determining the appropriate target population. The choice will depend on being able to access the population, how representative it is, the size of the population, and the population data accuracy. It will not always be possible to use an ‘ideal’ population, there will be different options and the advantages and disadvantages of each will need to be determined. The same principles apply, however, if the object of the study is to investigate the occurrence of a risk factor, such as cigarette smoking, or other human attributes.

There are several major ethical issues that face an epidemiological study. There is always the challenge in studying free-living individuals in the modern society, of balancing the needs for robust methods with individual freedoms. The key concerns reflect ensuring an appropriate consent process, maintaining confidentiality, and minimizing any negative consequences for a participant. The most commonly collected information for an epidemiological study come either from material already available in databases, material such as hard copies of records that can have key data items extracted, or data that is gathered directly from the subject. Occasionally a limited physical examination is undertaken. Much less often, there is a requirement to take samples of biological fluid such as blood and urine, or to undergo simple investigations such as electrocardiography or plain radiography, but even such investigations are typically associated with trivial risk to health.

Epidemiological studies can only show associations they cannot prove that a link is causative. Even in the bias free study with minimal confounding, a strong association does not mean that, for example, the presence of the risk factor has a direct biological link to the disease in question. There are several tests that can be used to increase the confidence that an association has biological meaning and needs to be considered. It is interesting always to differentiate factors that are on the ‘path’ to disease but not the underlying cause. The appropriateness of all these steps is dependent on the validity of the association.

A key aspect of epidemiology is about understanding and exploring variations in disease. In order to do so it is important that the same definitions of disease occurrence are used, no matter where a study is being conducted, and that the same approach to measurement is taken. The epidemiologist needs to be aware of which measure to use in specific circumstances. Measuring disease occurrence is the basic activity of epidemiology, and the following section provides the background to choosing the most appropriate measure(s) of disease occurrence for the aims of the study. The term ‘disease’ can also be taken in this context to describe any personal attribute. Thus, the concepts described in this chapter apply equally well to assessing the occurrence of a symptom, the development of a particular disability, or the frequency of a medical intervention.

This chapter builds on the previous one on the analysis of descriptive epidemiological studies and illustrates statistical methods appropriate for analysis of analytical epidemiological studies. It mainly focuses on data obtained from case–control and cohort studies, but also considers other study designs presented in Chapter 6. There are also several practical examples to help with the analysis and interpretation of the results of analytical epidemiological studies. In practice, relatively little mathematical calculation is done without computers. In this chapter, however, formulae are presented for the main measures of effect together with worked examples. Indeed, when data are available in tabulated form, as opposed to raw data files, it is frequently an easy task to calculate the important measures ‘by hand’. The formulae presented will permit the reader, for example, to check or further explore data published by others.

Having conducted a specific study design, the task is then to use the information collected to provide an effect measure which quantifies the magnitude of the association between the exposure(s) of interest and the disease under study. Epidemiological studies, and in particular their subsequent analysis, are therefore aimed at quantifying the level of increased risk when exposed to a particular factor, as this chapter explains. The effect measure which can be obtained to quantify the strength of the association, varies according to the type of study conducted. Just because there is a relationship between exposure and disease (even a strong one) does not mean that the relationship is causal.

Epidemiology can be considered as a suite of methodologies designed to inform on aspects of health and disease in human populations. They have been designed to cope with the variability between people in the risk for diseases and their course and response to treatments. There are many aspects of epidemiological study that are challenging and the successful study requires attention to the design, conduct of the study, and analysis of the data generated. There is another view of epidemiology, which is that it is a tool for studying diseases as they occur in human populations. The focus can also be on diagnosis, outcome, and results of interventions, as well as on causes. This broader definition allows a greater scope for the kinds of question that can be addressed both by those studying the health of populations and by those whose main focus is the study of disease in patient groups.

Although epidemiological studies are increasingly based on the analysis of existing data sets (including linked data sets), many studies still require primary data collection. Such data may come from patient questionnaires, interviews, abstraction from records, and/or the results of tests and measures such as weight or blood test results. The next stage is to analyse the data gathered from individual subjects to provide the answers required. Before commencing with the statistical analysis of any data set, the data themselves must be prepared in a format so that the detailed statistical analysis can achieve its goals. Items to be considered include the format the data are initially collected in and how they are transferred to an appropriate electronic form. This chapter explores how errors are minimized and the quality of the data set ensured. These tasks are not trivial and need to be planned as part of a detailed study methodology.

In comparing rates between populations, it is important that one is comparing ‘like with like’. One population may be considerably older than a population to which it is compared and therefore it would not be surprising that mortality rates were higher. Instead it is more useful to make comparisons taking account of differences in characteristics such as age or gender. The same considerations apply to examining disease rates over time in a given population. If the characteristics of the population change over time (e.g. the population gets older), this needs to be considered. To formulate hypotheses, the rate of a disease under study in a population may be compared with the rate in other populations, or in the same population at difierent time points. If the rates vary significantly between populations or are changing within a population, then this provides impetus for investigating the reasons underlying these differences or changes.

Although the definition of measures of disease occurrence, such as incidence and prevalence, may be relatively straightforward, there will be issues of how to operationalize these. The solutions will depend on what systems exist in the location of study such as clinical records, a prospective registration system. Alternatively, if there are no such systems, then alternative solutions will need to be investigated. This could include a proxy measure of disease or could involve undertaking one or more surveys to identify prevalence or new onset cases. Overall, the decision would need to be made on the basis of specific needs relating to an individual project where actual incidence data were not available, to ascertain whether an alternative (proxy) measure may be suitable.

Designing a study ‘on paper’ is no guarantee it will work; for example, individuals agree to be recruited or the method of data acquisition is acceptable and provides the data in the format and quality required. For this reason it is, especially when planning large/expensive studies, to test the feasibility of the different components in separate small-scale studies. Alternatively, or in addition it may be useful to do a pilot study, which reproduces the planned methods for the proposed full study, to test ‘the whole system’. A successful pilot study enhances the likelihood of grant success; indeed, grant-awarding bodies will often expect to see evidence that they have been conducted.

The preceding chapters have focused on the generation of scientific evidence using epidemiological studies. Here the emphasis is on bringing together evidence (evidence synthesis) to inform future research and health policies. The chapter considers levels and quality of scientific evidence and describes in detail how to conduct a systematic review. It reviews the ways of summarizing and evaluating evidence from epidemiological studies. Summary of evidence is needed in everyday clinical practice and for public health. We live in a time of information overload, and it is impossible to read all the available scientific journals, even on a narrow scientific topic. Simply using search terms such as ‘cancer’ will result in millions of results in Google Scholar or PubMed (a service of the US National Library of Medicine®) database. Given the ever-increasing volume of medical literature and time constraints, summary of evidence plays a big role in decision-making.

A key issue in epidemiology is to choose an appropriate study design. The design chosen will depend on the question of interest and some designs will offer clear advantages over others. However, other factors such as current state of knowledge will be important. The decision will be based not only on methodological but also on practical considerations, including the resources available, and the speed with which a result is required. For example, the most appropriate study may be too expensive or take too long to provide an answer. In such circumstances a compromise will need to be made—to undertake a study which can be conducted within the budget and time available, and which delivers information that is suitable for addressing the question of interest.

Primary data collection is challenging and with increasing electronic data capture in routine healthcare and other aspects of life, it is possible to address several epidemiological questions by robust analysis of such ‘secondary data’. There are considerable advantages in terms of scope, size, and speed of study to be balanced against the quality and depth of using primary data. Even when such direct contact is not required, there is often the need to extract necessary information from individual subject records such as medical files. There is often no alternative source of information, although the greater digitization of information is changing that scenario with the potential that the availability of such information might preclude the need for primary data.

The most important aspect that will influence the validity of any epidemiological study is the careful selection of the subjects for investigation. Separate issues relate to the sampling of subjects for disease status in case–control studies, and sampling by exposure status in cohort studies. In simplest terms, the issues are who should be the cases and, given that, who should be the controls. Thus, in each instance the needs are to identify the sampling frame and then what should be the process for selecting the specific sample or subsamples needed for study. In addition to consideration of who to study, other factors such as how to identify and verify plus the size of the planned study are all topics to be addressed.

Validity in respect of information that is collected in an epidemiological study is a measure of how accurate the data collected are, relative to an underlying truth. Classical approaches aim to ensure that the approach is both sensitive (low false negatives) and specific (low false positives) as, for example, in determining the presence or absence of a disease. A modified approach is needed where the value of an attribute is collected. Underpinning both these approaches is the need to consider what is the test of the underlying truth. The reader may already be aware of the application of validity tests in assessing the performance of diagnostic and screening tests. It is conventional to evaluate this performance in terms of the two indicators, sensitivity and specificity.

There are two major reasons to maximize participation in an epidemiological study. Firstly, low participation can lead to bias and secondly, inadequate numbers recruited can limit the power of the study. All other things being equal, lower participation will increase the influence of non-response bias in a study. However, it should be noted that even in studies with high participation there exists a possibility of selection bias if non-participants are markedly different from participants in aspects that are key to the conditions being studied. There are several ways of boosting participation which should be embedded into the study design from the start, since chasing up to increase participation is resource-expensive.

Epidemiological studies can be very expensive, especially from large populations with multicentre recruitment. The researcher will need to ensure that there are adequate resources, allowing for the fact that things will not always go to plan, but making sure that the research is value for money. What is considered a reasonable cost will also depend on how strong the rationale is for conducting the study. Although in theory the study design influences the costs, in practice the resources available will often constrain the methodological choices. Costing an epidemiological study accurately at the start is vital. There are several ways to maximize the use of resources to ensure the study is efficient.

Delahaye-Sourdeix, Manon Anantharaman, Devasena Timofeeva, Maria N Gaborieau, Valerie Chabrier, Amelie Vallee, Maxime P Lagiou, Pagona Holcatova, Ivana Richiardi, Lorenzo Kjaerheim, Kristina Agudo, Antonio Castellsague, Xavier Macfarlane, Tatiana V Barzan, Luigi Canova, Cristina Thakker, Nalin S Conway, David I Znaor, Ariana Healy, Claire M Ahrens, Wolfgang Zaridze, David Szeszenia-Dabrowska, Neonilia Lissowska, Jolanta Fabianova, Eleonora Mates, Ioan Nicolae Bencko, Vladimir Foretova, Lenka Janout, Vladimir Curado, Maria Paula Koifman, Sergio Menezes, Ana Wunsch-Filho, Victor Eluf-Neto, Jose Boffetta, Paolo Fernandez Garrote, Leticia Polesel, Jerry Lener, Marcin Jaworowska, Ewa Lubinski, Jan Boccia, Stefania Rajkumar, Thangarajan Samant, Tanuja A Mahimkar, Manoj B Matsuo, Keitaro Franceschi, Silvia Byrnes, Graham Brennan, Paul McKay, James D eng 1R03DE020116/DE/NIDCR NIH HHS/ R01CA092039 05/05S1/CA/NCI NIH HHS/ Research Support, N.I.H., Extramural 2015/04/04 06:00 J Natl Cancer Inst. 2015 Apr 2;107(5). pii: djv037. doi: 10.1093/jnci/djv037. Print 2015 May.; International audience; Deleterious BRCA2 genetic variants markedly increase risk of developing breast cancer. A rare truncating BRCA2 genetic variant, rs11571833 (K3326X), has been associated with a 2.5-fold risk of lung squamous cell carcinoma but only a modest 26% increase in breast cancer risk. We analyzed the association between BRCA2 SNP rs11571833 and upper aerodigestive tract (UADT) cancer risk with multivariable unconditional logistic regression adjusted by sex and combinations of study and country for 5942 UADT squamous cell carcinoma case patients and 8086 control patients from nine different studies. All statistical tests were two-sided. rs11571833 was associated with UADT cancers (odds ratio = 2.53, 95% confidence interval = 1.89 to 3.38, P = 3x10(-10)) and was present in European, Latin American, and Indian populations but extremely rare in Japanese populations. The association appeared more apparent in smokers (current or former) compared with never smokers (P het = .026). A robust association between a truncating BRCA2 variant and UADT cancer risk suggests that treatment strategies orientated towards BRCA2 mutations may warrant further investigation in UADT tumors.