Get Adobe Flash player

Main Menu

(9) Epidemiology

Epidemiology - Incidence and Prevalence



Friendly figures and hidden fallacies

Mortality rates from heart attacks in men aged 45-74 years in Europe may suggest among others the hypothesis that the disease distribution is related to some foods, a possibility which can be explored by correlating the rates of heart attacks in each European country (or even within smaller areas) with the average per capita consumption of several foods. Both the figures for heart attacks and for food consumption can be gathered easily from published statistics, making the exercise fast and friendly. If we had only two countries, one with a high consumption of, say, milk and the other with a low consumption, we could simply compute a rate ratio. A ratio different from 1 (equality of rates) would tell (if statistically significant) that there is an association or correlation between the occurrence of heart attacks and the consumption of milk. Here, however, we have not two but several countries and correspondingly several rates and differing levels of milk consumption; fortunately, the method of studying their correlation is just an extension of the rate ratio. If a correlation is indeed found, we should guard against inferring that milk consumption is a determinant of heart attacks. Not only do we have to take into account the possibility of confounding and bias encountered when interpreting associations in general, but here the association, called an ecological association, is at the level of geographical units, i.e. countries, not at the level of individuals, as in case-control and cohort studies.

In these, the consumption of milk and the health status (with or without heart attack) would have been measured and be known for each person, while in the correlation exercise all that is known is the rate for each country and the average consumption of milk (derived, for example, from sales figures). No one knows whether within each country the individuals who develop a heart attack are those who also consume more milk and the observed correlation could be an artifact. Falsely believing that it is real would result in an ecological fallacy. As is often heard, and as epidemiologists, contrary to what is also often heard, know perfectly well `correlation is not causation'.

Things, however, are even more complex as an opposite fallacy may be at work. Imagine two areas in one of which nobody smokes while in the other everybody smokes the same amount from the age of 15. The latter would have a much higher rate of lung cancer than the former, yet a study measuring smoking habits for each individual would be unable to detect any difference in lung cancer risk associated with smoking within each area. Only comparison of the two areas would reveal that in these circumstances smoking is a determinant of lung cancer but only at the population (area) level. To look only at the individual level may lead to an atomistic fallacy. Everything that has been said about geographical units applies also to time units, for example to correlations of concentrations of air pollutants during successive weeks and hospital admission rates for respiratory disease in the same weeks.

Cross-sectional surveys

Detailed information on health is gathered by special surveys of samples of a population, in which questions about health are asked or a health examination is carried out, or both. As in correctly conducted opinion polls, representative samples can be obtained by first subdividing the population by key criteria, typically gender, age, place of residence, and then extracting at random within each subdivision or `stratum' a number of subjects to be included in the survey. All data collected refer, as in population censuses, to a fixed point in time (calendar date) even if the actual duration may span several days or weeks. Some surveys may be repeated regularly to monitor trends in health. A major periodical survey is the US National Health and Nutrition Examination Survey. It was first conducted in the period 1971-75 on a nationwide random sample of more than 30,000 subjects and included an interview focused on diet and a medical examination. The survey was repeated in 1976-80 and in 1997-8. Since 1999, it has become an ongoing biennial survey including an enlarged and variable number of interview, medical examination, and laboratory test items.

While it is valuable to document in detail the health of a community and its changes in time, these surveys are usually less useful as tools to search for causes of disease. For example, blood pressure measurements and an electrocardiogram can be taken during the survey, and some electrocardiogram anomalies may be found to be more frequent among people with high blood pressure than among people with normal blood pressure. However, as both electrocardiogram and blood pressure were assessed at the same time, it is impossible to say which anomaly started first and can be a cause, direct or indirect, of the other. Indeed, they may both have begun at about the same time as the result of another common factor, such as tobacco smoking. All surveys of this type that collect data only once at a fixed point in time (cross-sectional studies) suffer from this shortcoming. Surveys of the same populations repeated at different dates but, as is often the case, on a different sample of people are not free of this limitation.

The burden of diseases

Another minefield, essential for the establishment of public health priorities, is the determination of the burden of different diseases due to different factors in a region or nation or even worldwide. In its simplest form, this may start with a frequently asked question of the type: what percentage of, for example, all cancers is due, say, to environment? Three main problems prevent a single answer. First, the definition of environment may cover all factors external to the body (sunlight, pollutants in air and water, tobacco smoke, foods, etc.) or it may be restricted to some of them, like pollutants in place of residence and occupation: the percentages will vary depending on the definition. A second reason is that these percentages obviously depend on how many people are exposed to the different components of the environment: if many smoke, the percentage of cancers due to environment, and to smoke in particular, is high; if few smoke, it is low. How many people smoke (and how much) may be relatively easy to determine, but there is usually much more uncertainty on how many people are exposed, for example, to carcinogenic air pollutants; moreover, for both exposures the numbers of exposed individuals vary from place to place, and percentages calculated for large countries or continents or the entire world hide these substantial variations as well as the uncertainty in the evaluation of the numbers of exposed people. Finally, even accurate percentages specific to a single place (e.g. a town) have the puzzling feature that they cannot be added up, as their total may exceed 100%, i.e. more than the total of cancer cases! In fact, if we knew all causes of cancer perfectly - which is far from being the case today – there would be a lot of double counting, as many cancers are due to the joint action of two or more causes, genetic and environmental. Occupational exposure to asbestos and tobacco smoking both independently increase the risk of lung cancer, but their combination further multiplies the risk for asbestos workers who smoke. It is correct to attribute the percentage of cancers due to this combined action, say 5% of all lung cancers, once to asbestos and once to smoking (because without either of the two exposures these cancers would have not occurred), but it is not correct to sum the corresponding percentages, 5% + 5%, because they refer to the same cancers.

With all these reservations, it should not be surprising if today one cannot be more precise than saying that on the grand scale of the whole world approximately one-third of cancer is attributable to environmental factors. Tobacco smoking accounts for at least 20%, alcoholic drinks for some 5%, infectious agents for at least 10% with higher percentages in developing countries, and occupational and environmental carcinogens for fractions variable from less than 1% to some 10%.

These percentages are a simplistic representation of the actual impact of a factor on the health of a population. The same percentage may reflect impacts of very different severity depending on whether the cancers affect young or old people or whether they are successfully treatable (and how and for how long) or not. These elements are taken into account in sophisticated analyses, developed in the last two decades, of the burden of disease in local, national, or world populations. Often, the results of burden of disease analyses are expressed as number of DALYs (disability-adjusted life years) lost due to a cause. One DALY corresponds to the loss of one year of life free of disability: hence the DALY unit of measurement incorporates both the loss of years of life, because of death, and the loss of quality of life, because of disability.

Working for the health of all

Epidemiology is at heart a field of applied research with the improvement of the health of all as the key aim. As such, epidemiology is an essential component of all public health activities that implement the organized efforts of society to promote, protect, and restore health. This concept of public health has no relation to how societal efforts to improve health are or should be organized; it does, however, imply that some kind of explicit organization should exist, rather than just dispersed and uncoordinated initiatives, for society to successfully tackle health problems.

Three broad activities contribute to people's health. In clinical medicine, doctors and other health personnel deal individually with each patient. They provide preventive measures such as drugs to control high cholesterol or elevated blood pressure, or deliver advice and psychological support to stop smoking. They intervene to diagnose, treat, and when possible cure, diseases with procedures ranging from the simple prescription of an antibiotic to a complex liver or heart transplant. Finally, they offer individual rehabilitation to people with disabling diseases. Prevention and early diagnosis at the population level form the second field of activity. Prevention addresses the root causes of disease, environmental or genetic. It embraces a vast array of regulations spanning control of pollutants in air, water, and the workplace, to traffic speed limits and safety requirements in home appliances. It includes compulsory and optional vaccination programmes as well as campaigns to foster healthy diet and behaviour. When it targets genetic causes of diseases, for example the screening of all newborns for genetic defects, primary prevention uses medical diagnostic tools, as do organized programmes of early diagnosis and treatment of diseases. These have proved effective and are operational in many countries for a limited number of high-impact diseases such as cancers of the uterine cervix and of the breast.

The third activity consists in the empowerment of people to exercise responsibility for their health through adoption of health promoting habits and participation in the decision processes that shape health policies. The latter in turn may reinforce or inhibit people's empowerment, the development of which depends on formal and informal education and on updated and accurate information.

Public health also coordinates these activities in relation to other societal actions, external to the health system, which strongly influence health, for example income and housing policies. In the coordination process, public health administrators and policy makers usually demand that the benefits and adverse effects of proposed policies be subject to economic analysis, in which epidemiologists play a specific role jointly with other specialists. Channelling the research results into practice, whether in clinical medicine, in population prevention, or for people's empowerment, requires as a first step the aggregation of the results of multiple studies to consolidate the total evidence available on a specific question, for example whether vitamin C protects against cancer in humans. This is done by critically reviewing the studies' reports, comparing methods and results, and drawing a general `best' answer to the question at hand. In the last two decades, the approach and methods used in a review, previously entirely left to the reviewer's discretion, have been refined and made more objective and rigorous under the heading of systematic reviews.

Systematic reviews, with and without meta-analysis

A systematic review is a review carried out using a systematic approach to minimize bias and random errors, a process which is explicitly documented in the methods section of the review itself. It usually offers a more objective appraisal of the available evidence than traditional reviews, conducted as narrative commentaries on the studies. In a systematic review, each study is scrutinized to assess its quality in respect of a number of criteria fixed in advance, e.g. how well the population is defined, whether the study responses were assessed blindly or not, and so on. This makes it possible to consider separately studies judged of higher and lower quality, rather than all of them together, and see whether the results of the lower-quality studies point in the same direction (e.g. towards a reduction or an increase in risk) as the higher-quality ones. Broadly consistent results can be combined in a statistical analysis, a meta-analysis, to provide a single summary estimate of risk. This analysis, in which each study is given a `weight' proportional to the number of disease cases it contributes, may cause a clear-cut result to emerge, while the individual studies, particularly if small in size, may each present a result statistically non-significant that is difficult to interpret.

Combining studies often permits the evaluation of rare events, too few of which occur in a single study. A typical case is that of side effects of new drugs, which occur infrequently, say once in a thousand treated patients. However, if a side effect is serious, for instance a major heart problem, it will have considerable impact when the drug is put on the market and used by hundreds of thousands or millions of people. Yet such an effect will be hard to detect in a randomized experiment of a size of, say, a few hundred subjects, which would be more than adequate to measure a much more frequent therapeutic effect. It is only by combining all available data from different randomized experiments that a sufficiently large number of patients is reached to allow the adverse event to become detectable. A telling example is Rofecoxib, a drug commercialized in 1999 as an anti-inflammatory remedy for rheumatic and muscular disorders. It was withdrawn from the market by the manufacturer in September 2004 on account of an increased risk of heart attacks, when an estimated 80 million people had already used it. However, if the manufacturer or the drug licensing authorities had conducted a timely meta-analysis, they would have detected the increased risk more than three years earlier, in 2000.

Systematic reviews complemented by meta-analyses of randomized controlled trials are most valuable for clinical medicine. They have helped to develop the continuously evolving body of evidence-based medicine which guides doctors' everyday practice. They have also helped to put the evidence from randomized preventive trials carried out in populations on a firm basis, for example the prevention of myocardial infarction with cholesterol-lowering drugs.

Meta-analyses have also been extended to observational epidemiology studies directly relevant to public health. Combining results from observational studies in which confounding factors and biases have usually been dealt with in a different way in each study in a statistical analysis is, however, problematic. As we know in randomized controlled trials bias and confounding are prevented by randomization and do not impinge on a meta-analysis, a condition that does not apply to observational studies. For these studies, systematic reviews are in any case necessary while the worth of meta-analyses has to be assessed case by case.



googleplus sm


ar bg ca zh-chs zh-cht cs da nl en et fi fr de el ht he hi hu id it ja ko lv lt no pl pt ro ru sk sl es sv th tr uk

Verse of the Day

Global Map