 |
RMA Publications |
 |
|
|
 |
APPENDIX 2Background Terminology: Introduction To Some Epidemiological TermsDr Alex Bordujenko, RMA Secretariat EpidemiologyThis is the study of variations in disease frequency among population groups, and the factors that influence these variations. The principle objective of epidemiology has been to determine factors which may cause or contribute to disease processes in humans, so that preventive measures may be applied. Epidemiologic observations have a long history, with much work developed through the study of acute epidemic diseases such as cholera and typhoid. The discipline has burgeoned over the latter half of the Twentieth century, with interest in the study of the cause, treatment and prevention of cancer, cardiovascular and other chronic disease, and of course the advent of computer storage and analysis systems. Approaches for Epidemiological StudyHennekens and Buring (1987) define epidemiology as "the study of the distribution and determinants of disease frequency in human populations." "Humans" distinguishes the approach from research using animal or other systems in experiments. "Populations" contrasts the practise of individual investigation as in clinical research. "Frequency" indicates the quantification of disease occurrence and the risk attributable to various potential causes. The term "distribution and determinants" points to the two major approaches of epidemiology: 1. examination of the distribution of disease frequency in populations (this can produce hypotheses about the causes of disease) known as descriptive studies; and 2. analytical studies which test these hypotheses by reviewing personal characteristics or exposures among individuals within the populations. Descriptive studies use population based statistics on mortality, disease incidence, and survival. Other registries for example hospital based disease registries, may also be useful. Obviously the studies concern populations and not individuals and measures of any exposures are usually broad and may be subject to confounding or interfering factors. Selection of free living populations may introduce biases and confounding into the calculations. Examination of national and international trends, migrant studies and time trends has provided valuable insights into the causation of a number of chronic diseases for example breast, prostate and lung cancers. Analytical studies have provided much useful information concerning the discovery and/or confirmation of a number of lifestyle and other environmental exposures as causes of chronic disease, including cancer. Examples of these include cigarette smoking, where, for smokers of 40 or more cigarettes per day there is a risk of lung cancer of more than twenty times that of a non-smoker. Another well documented example is occupational exposure to asbestos and the development of mesothelioma, where the relative risk is well over 100 fold that of the unexposed population. Analytical studies from several international sources in the last decade have also demonstrated that both the incidence and recurrence of neural tube defects can be greatly reduced by maternal folate supplementation in early pregnancy, even in the absence of maternal folate deficiency. In chronic disease epidemiology, the types of analytical studies encountered are: A. Cohort studies identify groups of individuals with and without a particular exposure, and follow them over time to examine disease incidence and/or mortality rates. These may be current or past exposures. An association is suggested when rates of disease or death differ between the groups. These are able to directly measure incidence and mortality rates related to a particular exposure (especially with prospective design) but they require large numbers of exposed individuals particularly when considering uncommon diseases, before significant differences may be noted. B. Case-control studies or case-referent studies identify people with a particular disease (case), and a group of people without the disease (controls), and then collect information about past exposures, for example by interview or questionnaire. They provide a method of studying rare diseases but may be subject to recall and other biases, and difficulty in measuring past exposures. Data Presentation and Interpretation The odds ratio (OR) is a measure of association used in case control studies to estimate the odds of exposure in cases to the odds of exposure in controls. This approximates, but is not synonymous with, the "relative risk" (RR) the measure of association used in cohort studies. The term relative risk (RR) is used to describe the comparison of the risk of a known exposed group versus a known unexposed group developing a specific condition. Thus if the relative risk is one the risk is the same for both groups and exposure is not seen to be associated with the development of the particular condition that is, there is no increase in the risk of a studied outcome with the exposure of interest. If the RR (or OR) is 1.5 then the risk for the studied outcome in the exposed versus the unexposed group is increased by 50%. An RR (or OR) of 2 implies a doubling of risk, and an RR (or OR) of less than one implies a reduction of risk. Problems in decision making occur when the described increase in risk is weak (under a two to three fold increase) and particularly when the relative risk is close to one, for example 1.1 (10% increase) or 1.3 (30% increase) rather than the 20 fold increases for heavy cigarette consumption and the incidence of lung cancer and the much grater increases seen with occupational asbestos exposure and the incidence of mesothelioma. Many epidemiologists are reluctant to accept as real, increases in risk of less than 100% (RR<=2) as likely to be causative unless the "Bradford Hill" types of criteria are stringently applied to the body of evidence pertinent to the putative association, and overall, a considered case can then be made to support causality. Another term, the "confidence interval" (CI), is used to describe the range of relative risk (or odds ratio) rates within which the actual result lies, to within, for example, a 95% probability. Thus, if the confidence interval includes one then the result could have occurred due to chance and no true effect may exist. If the 95% confidence limits exclude one it does not exclude the possibility of a chance result, rather it indicates that chance would explain the observed (or a greater) risk estimate only one out of 20 times. Selected Measures of Disease Frequency As well as the relative risk and odds ratio a number of other measures of disease frequency need to be considered. A consideration of the basic concepts of these measures includes the formulae used to calculate such measures. In its simplest form data from a two-by-two table from a case-control or cohort study with count denominators would appear as:
a= the number of individuals who are exposed and have the disease As stated above for cohort studies the term relative risk (RR) is used to describe the comparison of the risk of a known exposed group versus a known unexposed group developing a specific condition, that is the incidence of the disease in the exposed divided by the incidence in the unexposed Ie/Io or the cumulative incidence of the disease in the exposed divided by the cumulative incidence in the unexposed CIe/CIo. The formula for calculating relative risk for cohort studies with count denominators is thus: Ie = CIe = a/(a+b) (where a,b,c,d are derived from the 2x2 table outlined above). For case control studies with count denominators the odds ratio is expressed as: a/c = ad (where a,b,c,d are derived from the 2x2 table outlined above) The odds ratio is said to provide a valid estimate of the relative risk for case-control studies where the cases are newly diagnosed, where prevalent cases are not included in the control group and where the selection of cases and controls is not based on exposure status. Attributable risk is the measure which provides information about the absolute effect of the exposure and is the excess risk of disease in those exposed compared with those who are unexposed to a specific factor. This measure is defined as the difference between the incidence rates in the exposed and unexposed groups and may be calculated in cohort studies as AR = CIe - CIo = a/(a+b) - c/(c+d) (where a,b,c,d are derived from the 2x2 table outlined above) The attributable risk percent (AR%), attributable rate percent attributable proportion or etiologic fraction is calculated as the attributable risk divided by the rate of disease among the exposed and is said to represent the proportion of disease in that group that could be prevented by absence of the exposure. AR% = AR/Ie x 100 = (Ie - Io) x 100 = (1- Io/Ie) x100 = (1-1/RR) x 100 = (RR-1/RR) x 100 Population Attributable Risk (PAR) is the measure used to estimate the excess rate of disease in the total study population of exposed and unexposed individuals that is attributable to the exposure. The PAR is calculated as the rate of disease in the population (incidence rate in total population = It) minus the rate in the unexposed group (Io): PAR = It - Io or by multiplying the AR by the proportion of exposed individuals in the population (Pe): PAR = (AR) x (Pe) Population Attributable Risk Percent (PAR%) is represented by: PAR% = 100 x (Pe) x (RR-1)/1+(Pe) x (RR-1) Epidemiologic Studies Need Careful Examination(Refer to:- Darzins PJ, Smith BJ and Heller RF (1992). How to read a journal article. The Medical Journal of Australia, Vol 157 pp 389-394.) The size of the population studied is important - the larger the sample size the greater the power (or ability) to detect a specified risk, the smaller the sample size the weaker the power. Negative results from small studies may not be conclusive as only large studies may confidently exclude or include low to moderate levels of risk. When examining any study results, consideration of the possibility of a non-causal association is necessary. The observed association between exposure and disease may result from bias, confounding, chance, or cause-and-effect. Bias is the term used for any systematic error in a study and may occur during study selection, information gathering or in reporting of the assessment of the exposure or outcome under investigation. Confounding bias is the possibility of the observed effect being due to other variables not adequately considered in study design or analysis of the results. Many types of study bias have been described including selection, information, recall, and interviewer bias. Confounding bias or confounding is due to variables which may themselves account for all or part of an apparent association between an exposure and a disease. They may also obscure an association. Chance is considered previously in the discussion of study power and Confidence Intervals. Study Types Study design has an effect on the quality of evidence which may be gained and a recognised `hierarchy' of study types exist. In developing the following the "US Preventative Services Task Force: Guidelines for Quality of Evidence" (Fisher, 1989) have been considered. Given the specific needs of the RMA some modification has been undertaken. In this instance the level of evidence available is at best observational (cohort or case control studies). The following is broadly the division of available study designs and how these may be considered in the information gathering. Analytic Studies: 1 Intervention Studies 1a Randomised Controlled Trial 2 Observational Studies 2a Cohort-Prospective 3 Case Control Studies 4 Population (Correlational) 5 Individual 5a Cross Sectional Surveys Where the numbering 1-5 refers to the grade assigned to the quality of the evidence. Quality refers here to study design rather than individual study merit that is that the evidence from cohorts is graded as higher than that from case control studies - this is the method used by the US Preventative Services Task Force. While the RMA places emphasis on primary research published in the leading peer reviewed journals of either broad or discipline specific type; published, peer reviewed, reports on the epidemiology of disease such as those produced from time to time by the International Agency for Research into Cancer, the National Academy of Science, or the Surgeon Generals' Reports concerning to smoking related disease; are considered appropriate sources for examination. Published reports from sources such as the National Health and Medical Research Council and other expert committees are also be considered where contemporary, applicable material is available. Consideration of Individual Studies(Refer to:- Darzins PJ, Smith BJ and Heller RF (1992). How to read a journal article. The Medical Journal of Australia, Vol 157 pp 389-394.) In the absence of interventional studies such as randomised controlled trials most reliance is placed on well designed and reported cohort and case control studies and Professor Heller provides a method of considering these. This forms a mental check list in consideration of materials. The following questions may be specifically addressed. 1. What is the research question? After determining these features a decision on adequacy of methods and clarity of results is made considering: bias - are the results biased in one direction. If so, what is the direction and magnitude of bias It is recognized that for many putative factors evidence may only be available in descriptive studies. This is often the case for case reports or case series of disease associations or drug reactions. Association and CausationAssociation is the term used to describe the statistical dependence between two variables. In epidemiology it is the degree to which the rate of disease in persons with an exposure of interest is either higher or lower than the rate of disease among those without that exposure. Such an association does not mean, or even imply, that the observed relationship is one of cause and effect (Hennekens and Buring, 1987). Making judgements about causality from epidemiologic data involves a logical process which addresses two major areas: 1. Whether for any individual study, the observed association between an exposure and disease is valid. An assessment of validity requires a consideration of the likelihood of alternative explanations for the results and chance (the luck of the draw), bias (any systematic error in the study for example in subject selection, information gathering or reporting), or confounding (the observed effect being due to other variables not adequately considered in study design or analysis of the results); and 2. Whether the body of the evidence considered supports a judgement of causality. In this process standard epidemiological criteria are used (Hennekens and Buring, 1987). Epidemiologic criteria used to assist in the assessment of causality The RMA considers the individual studies with respect to the above and then, in considering the available evidence uses standard epidemiological criteria to make a judgement regarding causality with regard to the reasonable hypothesis and balance of probabilities standards of proof. The Bradford Hill criteria (Bradford-Hill, 1965), and more contemporary versions, are widely accepted in the interpretation of epidemiological studies for the purpose of assessing the possibility of a causal association. Consideration of the body of evidence available for each contention against the current epidemiologic criteria will result in a judgement regarding causality. As Professor Holman (1997) notes, more than 30 different systems of causal verification have been described. In his technical appendix to the Pearce Report he outlines ten criteria for classification of evidence of causality, based on work by Mervyn Susser. The RMA has considered a number of such systems including those of Bradford Hill, Susser and those co-authored by Professor Holman in "The Quantification of Drug Caused Morbidity and Mortality in Australia, 1995" (English and Holman, 1995). The RMA recognises the underlying similarities which underpin these systems. The exact description of these epidemiologic criteria varies between authors and the RMA recognises the need to consider both internal study validity (for individual studies) and factors important in the body of evidence (the applicable evidence available from epidemiological, clinical, toxicological and other research) in these criteria. Sir Austin Bradford Hill, as well as other prominent statisticians and epidemiologists, including Mervyn Susser and Kenneth Rothman, have described how the subjective likelihood (or the correct judgement) of a causal relationship is increased when evidence relating to an association meets criteria devised to consider the available evidence. The Bradford Hill (Bradford-Hill, 1965) criteria are as follows: 1. Strength of Association The criteria used by the Expert Committee on Herbicide Exposure and Spina Bifida (1996) further refined the criteria to explicitly include consideration of bias and confounding in the criteria: 1. Statistical significance (that is the possibility of chance being responsible for an apparent association;
and study power) 1. Statistical significance and power If the criterion of statistical significance is satisfied then the evidence is supportive of an association. The failure of a test to reach statistical significance in the presence of adequate statistical power provides evidence against the association, however in the absence of adequate statistical power it may not necessarily detract from the association. 2. Strength of association The greater the strength of association the more likely it is to be causal. Confounding is less likely to explain a strong association because the strength of the association between the confounding variable and the outcome must also be strong. While a strong association is supportive of causality, a weak association may not necessarily detract from the evidence of causality however adequate consideration of potential confounding or bias is essential. 3. Consistency of replication Consistency of the evidence or the lack of evidence in the face of study diversity in time, place, circumstances and population, as well as research design, strongly supports or detracts from a causal hypothesis. 4. Possibility of bias in measurement of exposure or outcome Consideration of any systematic error in the study in information gathering or in reporting of the assessment of the exposure or outcome under investigation. Absence of bias in the studies considered to show a positive association supports the existence of a putative association. The presence of bias detracts from the conclusions which may be drawn from the information. 5. Possibility of selection or confounding bias Consideration of any systematic error in the study in subject selection; or the possibility of the observed effect being due to other variables not adequately considered in study design or analysis of the results. Absence of bias or confounding in the studies considered to show a positive association supports the existence of a putative association. The presence of bias or uncontrolled confounding detracts from the conclusions which may be drawn from the information. 6. Time sequence The exposure must precede the disease or injury. This criterion is compatible with, but does not necessarily support causality. Reversal of the order of exposure and disease or injury is the most persuasive basis available for rejection of causality. 7. Dose response A response which is in proportion to the level of exposure is strongly persuasive of a causal relation. However, its absence does not necessarily detract from the association. 8. Biological plausibility Theoretical coherence: Findings plausible in terms of pre-existing theory are supportive of the association. Conversely, findings that are implausible in terms of pre-existing theory detract from the evidence. Factual coherence: Compatibility of a new result with pre-existing facts is supportive of the association. Incompatible pre-existing facts strongly detract from evidence of causality. Biological coherence: Pre-existing knowledge which identifies a mechanism by which the chemical exposure may produce the disease or injury is supportive of case for the association being causal. Observations from species other than humans may also be used to support the potential mechanism of action. Incoherence between biological knowledge and study observations detracts from the case for a causal association. As Rothman and Greenland (1997) eloquently acknowledge inductively oriented causal criteria are not sufficient within themselves and require sound scientific judgement to traverse the path for which the criteria are "the road map through complicated territory". Bibliography1. Bradford-Hill A: The Environment and Disease: Association or Causation. Proceedings of the Royal Society of Medicine, 58:295-300, 1965. 2. Darzins PJ, Smith BJ and Heller RF (1992). How to read a journal article. The Medical Journal of Australia, Vol 157 pp 389-394. 3. English D and Holman D (1995). "The Quantification of Drug Caused Morbidity and Mortality in Australia, 1995." National Drug Strategy; Commonwealth Department of Human Services and Health. 4. Expert Committee (1996). "Report of the Expert Committee Into the Possible Connections Between Exposure to Herbicides in Vietnam and Spina Bifida in Children of Vietnam Veterans". Report prepared for the Repatriation Commission, Australia. 5. Hennekens CH and Buring JE (1987) Epidemiology in Medicine First Edition Little Brown and Co Boston 6. Holman D (1997). Technical Appendix: Criteria for assessing Causation in Pearce D, Review of the Repatriation Medical Authority and the Specialist Medical Review Council, Commonwealth of Australia, p120. 7. Rothman KJ and Greenland S Causation and Causal Inference. In: Detels R, Holland WW, McEwan J, Omenn GS eds, Oxford Textbook of Public Health, Volume 2: The Methods of Public Health, pp 616-629. New York: Oxford University Press, 1997. 8. US Preventative Services Task Force: Guidelines for Quality of Evidence Ed M Fisher. William and Wilkins. Baltimore Maryland, 1989. Methodology pp 27-37 and Appendix A pp 387-397.
This page last updated 22 March 2005. |
|||||||||||||||||||||
|
||
