ILSI Epidemiology for pdfilsi.org/.../uploads/sites/29/2016/09/Nutritional-Epidemiology.pdf ·...

I L S I E U R O P E C O N C I S E M O N O G R A P H S E R I E S

NUTRITIONALEPIDEMIOLOGYPOSSIBILITIES ANDLIMITATIONS

Analytical Epidemiological Study

Observational Studies Experimental Studies

Uncontrolled assignment Controlled assignment

Sampling withregard to disease

or effect

Case-Control Studies

Sampling with regard toexposure, characteristic,

or suspected cause

Cohort Studies

Community assignment

RandomizedIndividual assignment

CommunityTrials

Randomized Clinical (Intervention) Trials

ILSI Epidemiology for pdf 13/04/99 8:44

ILSI Europe

by Lillian Langseth

NUTRITIONALEPIDEMIOLOGY:

POSSIBILITIES AND LIMITATIONS

ILSI Epidemiology for pdf 13/04/99 8:44 Page 3

ILSI EUROPE CONCISE MONOGRAPHS

This booklet is one in a series of ILSI Europe concisemonographs. These are written for those with a generalb a c k g round in the life sciences. However, the style andcontent of the concise monographs will also appeal to awider audience who seek up-to-date and authoritativereviews on a variety of important topics relating tonutrition, health and food safety.

Concise monographs present an overview of a particularscientific topic and are usually based on the pro c e e d i n g sof scientific meetings. The text of each is peer reviewed byacademic scientists of high standing. The concisemonographs make important results and conclusionswidely available.

Titles of concise monographs in preparation include:

◆ Healthy Lifestyles◆ Microwave Ovens.

C u r rent and newly published concise monographs are :

◆ Starches and Sugars: A Comparison of TheirMetabolism in Man

◆ A Simple Guide to Understanding and Applying theHazard Analysis Critical Control Point Concept

◆ Dietary Fibre◆ Food Allergy and Other Adverse Reactions to Food◆ Sweetness – The Biological, Behavioural and Social

Aspects◆ Oxidants, Antioxidants, and Disease Prevention◆ Caries Preventive Strategies◆ Nutritional and Health Aspects of Sugars –

Evaluation of New Findings◆ Food Biotechnology – An Introduction◆ Dietary Fat – Some Aspects of Nutrition and Health

and Product Development◆ Health Issues related to Alcohol Consumption.

The International Life Sciences Institute (ILSI) is aworldwide, public, non-profit scientific foundation head-q u a r t e red in Washington, DC, USA, with branches inA rgentina, Australasia, Brazil, Europe, Japan, Kore a ,Mexico, North America, Southeast Asia, and Thailand,with a focal point in China.

ILSI is affiliated with the World Health Organization as anon-governmental organization (NGO) and hasspecialised consultative status with the Food andA g r i c u l t u re Organization of the United Nations.

ILSI Europe was established in 1986 to provide a neutralf o rum through which members of industry and expertsf rom academic, medical and public institutions cana d d ress topics related to health, nutrition and food safety

throughout Europe in order to advance theunderstanding and resolution of scientific issues in thesea reas. ILSI Europe is active in the fields of nutrition,toxicology and food micro b i o l o g y. It sponsors re s e a rc h ,c o n f e rences, workshops and publications.

For more information about its programmes andactivities, please contact:

ILSI EuropeAvenue E. Mounier 83, Box 6

B-1200 BRUSSELSBelgium

Telephone (+32) 2 771 00 14Telefax (+32) 2 762 00 44

INTERNATIONAL LIFE SCIENCES INSTITUTE AND ILSI EUROPE


© 1996 International Life Sciences Institute

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in anyform or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior writtenpermission of the copyright holder. The International Life Sciences Institute (ILSI) does not claim copyright on U.S.government information.

Authorization to photocopy items for internal or personal use is granted by ILSI for libraries and other usersregistered with the Copyright Clearance Center (CCC) Transactional Reporting Services, provided that $0.50 perpage per copy is paid directly to CCC, 222 Rosewood Drive, Danvers, MA 01923. Telephone: (+1) 508 750 8400.

The use of trade names and commercial sources in this document is for purposes of identification only, and does notimply endorsement by the International Life Sciences Institute (ILSI). In addition, the views expressed herein arethose of the individual authors of the original book and/or their organizations, and do not necessarily reflect those ofILSI (or the author of the concise monograph).

ILSI Press1126 Sixteenth Street, N.W.Washington, DC 20036-4810USATelephone: (+1) 202 659 0074Telefax: (+1) 202 659 8654

ILSI EuropeAvenue E. Mounier 83, Box 6B-1200 BrusselsBelgiumTelephone: (+32) 2 771 00 14Telefax: (+32) 2 762 00 44

Printed in Belgium

ISBN 0-944398-87-1

Cover figure from Lilienfield DE, Foundations of Epidemiology, © D. Lilienfield 1994 (Used by permission of OxfordUniversity Press)


FOREWORD

Nutritional epidemiology – the study of the nutritionaldeterminants of disease in human populations – is anexciting branch of re s e a rch because it can pro v i d einsight into the potential causes and prevention ofmany health conditions. But it is also an extremelycomplex endeavour because many of the associationsbetween dietary factors and disease risks are difficult todiscern using epidemiological techniques.

With the increased interest in health-related issues, theresults of epidemiological studies easily find their wayto the public via newspapers and magazines. Thefindings, however, must be interpreted with caution.The associations identified by epidemiological studiesare important tools for generating and testing newhypotheses about the diet-disease relationship, butother lines of scientific research are needed to providethe evidence that a specific dietary factor causes a healtheffect.

If applied in appropriate ways, nutritionalepidemiology can generate information of gre a t

relevance to public health. This concise monographaims at making this complex subject understandable. Itdescribes the most common study designs used innutritional epidemiology and explains their strengthsand weaknesses. It also addresses the challenges ofaccurately measuring food intakes, the use ofbiomarkers to study nutritional status, the complexitiesof dealing with bias and confounding factors, theapplication of statistics and the interpretation of studyfindings. A series of examples illustrate the possibilitiesand limitations of nutritional epidemiology.

Public health authorities, health care pro f e s s i o n a l s ,scientists in academia and food industry who are nottrained in epidemiology, journalists and consumergroups will find this informative publication a usefuladdition to their libraries. With improved overallunderstanding of the possibilities and limitations ofnutritional epidemiology, people in all of these groupscan apply the findings of this field of research in waysthat can contribute to current efforts to improve healththrough disease prevention.

Author: Lillian LangsethScientific Editor: Frans J. Kok

Scientific Referees: Pierre Ducimetière, Lenore Kohlmeier, Barrie Margetts, Pirjo Pietinen Series Editor: Nicholas J. Jardine


CONTENTS

Introduction .................................................................................................................................................................... 1

Key features of epidemiology studies: design,exposure assessment and data analysis ................................ 6

Interpretation ................................................................................................................................................................ 24

Application of nutritional epidemiology data ........................................................................................................ 32

Summary ......................................................................................................................................................................... 35

Glossary ........................................................................................................................................................................... 38

Further reading .............................................................................................................................................................. 39


INTRODUCTION

Definition of epidemiologyEpidemiology is the study of the distribution anddeterminants of diseases and other health outcomes inhuman populations. Epidemiology also deals with thenatural history of diseases and it can provide evidencethat contributes to their prevention.

Traditionally, epidemiologists devoted most of theirattention to infectious diseases. However, even in theearly decades of the 20th century, epidemiology madeimportant contributions to the understanding ofnutrition-related diseases as well. The investigations ofpellagra are a classic example and they illustrate someof the strengths of epidemiology.

Epidemiological investigation of a deficiencydisease

In the 19th and early 20th centuries in both Europe andthe United States, pellagra occurred frequently amongpeople who relied on corn as their staple food and wholived in poverty. It was widely suspected that thedisease was caused by an infectious agent or by a toxinin spoiled corn. A study by Dr. Joseph Goldberger of theUnited States Public Health Service, however,suggested otherwise.

D r. Goldberger observed that pellagra was absentamong the staff in an institution even though it waswidespread among the residents. Because the staff hadaccess to foods, including meat and milk, that were notserved to the residents, Dr. Goldberger began to suspectthat the disease might be due to a nutritional deficiency.He obtained funds to supply the residents with meatand milk and found that pellagra rapidly disappearedafter these foods were introduced. When the funds ranout and the extra foods were discontinued, pellagrareturned.

A few years later, scientists developed an animal modelof pellagra and they eventually isolated niacin, theB vitamin that prevents this disease. Three types ofscientific investigation – animal experimentation, bio-chemical studies and epidemiology – all contributed tothe eventual full understanding of the causation ofpellagra, but it is important to note the key role ofe p i d e m i o l o g y. At a time when no other type ofinvestigation was possible (because no animal modelwas available and the biochemical basis of the diseasewas unknown), epidemiology provided the key clue tosolving the mystery of pellagra and epidemiologicalmethods identified a practical means of prevention.

Epidemiological investigation of chronic diseases

Epidemiology can play a similar role today, even thoughthe current focus is on chronic diseases rather thandeficiency syndromes. For example, epidemiologicalstudies in a wide variety of human populations haveconsistently associated low intakes of fruits andvegetables with increased risks of many types of cancer,as shown in Table 1. The mechanisms by which thesefoods may protect against cancer are incompletelyunderstood and the active agents have not beenconclusively identified. Nevertheless, the epidemi-ological evidence provides a reasonable basis for actionand a sound basis for further re s e a rch and manyauthorities now urge people to increase theirconsumption of fruits and vegetables in an effort toprevent cancer.

Specific characteristics of nutritionalepidemiologyNutritional epidemiology can be defined as the study ofthe nutritional determinants of disease in humanpopulations. It is one of the most exciting – and mostdifficult – types of epidemiological research.

Nutritional Epidemiology 1


The excitement comes from the direct relevance of thistype of research to crucial health problems of Westerncivilization, such as coronary heart disease, cancer,stroke, osteoporosis, cataracts, diabetes and congenitalmalformations. All of these diseases have been theobject of research in nutritional epidemiology and someof the findings have already been applied in ways thatmay improve public health.

For example, epidemiological studies completed in theearly 1990s established that women could substantiallyreduce their risk of bearing a child with a neural tubebirth defect (anencephaly or spina bifida) by increasingtheir intake of the B vitamin folic acid. Even though themechanism of action of folic acid is not fullyunderstood, public health authorities have begun totake action on this new knowledge. Medicalo rganizations in many nations have re c o m m e n d e di n c reased intakes of folic acid for women ofchildbearing potential and government agencies inseveral countries are planning to fortify staple foodswith folic acid.

The complex nature of diet

A major difficulty of nutritional epidemiology lies in theextremely complex nature of diet. To appreciate thiscomplexity, it is helpful to compare diet with anotherexposure that also influences the risk of many of thesame diseases – cigarette smoking.

An epidemiologist who is studying tobacco can obtain ag reat deal of useful information simply by askingpeople, "Do you smoke?" By collecting a few additionalpieces of information – the number of cigarettes smokedper day, the types or brands of cigarettes smoked, theage at which the person began (or stopped) smokingand any changes that may have occurred in the patternof cigarette use – the researcher can obtain a clear,reasonably accurate picture of an individual's smokinghistory.

2 Concise Monograph Series

Epidemiological studies showingprotection by fruits and vegetablesagainst cancer

Cancer site Fraction of studies showing protection

Epithelial

Lung 24/25Oral 9/9Larynx 4/4Esophagus 15/16Stomach 17/19Pancreas 9/11Cervix 7/8Bladder 3/5Colorectal 20/35Miscellaneous 6/8

Hormone-dependent

Breast 8/14Ovary/endometrium 3/4Prostate 1/14

Total 129/172

Source: Ames BN, Shigenaga MK, Hagen TM, Oxidants,antioxidants and the degenerative diseases of aging, Proc NatlAcad Sci USA 90:7915–7922, 1993.

TABLE 1


In contrast, one cannot learn much by asking people,"Do you eat?" Everyone eats. Everyone is exposed, tovarying degrees, to most dietary factors. Eating patternsoften evolve slowly over periods of years and peoplemay not remember when their habits changed. Thefoods that people eat consist of complex mixtures ofcompounds, with substantial differences even amongseemingly similar products. People who eat more of onetype of food must eat less of other types of foods, thusc reating a complex set of interc o r relations amongdietary components. Eating habits may be correlatedwith other factors that influence disease risk, such asethnic background, socioeconomic status and tobaccouse. Even the method of preparation of foods may beimportant. For example, boiled coffee may raise bloodc h o l e s t e rol levels; filtered coffee does not becausefiltering removes the components that may havecholesterol-raising effects.

Multifactorial causation of chronic diseases

Like diet, the diseases currently of interest toepidemiologists are also complex. They have multipledeterminants and long latent periods and they occurwith relatively low frequencies, even among peoplewith high exposure to risk factors.

For example, coronary heart disease has a wide varietyof recognized risk factors including age, gender,menopausal status in women, family history, bodyweight, blood pressure, blood cholesterol and diabetes.Other factors, such as the degree of oxidation of bloodlipoproteins and levels of the amino acid homocysteine,may also be involved. Many of these risk factors andsuspected risk factors are influenced by multipleaspects of diet. For example, intakes of several types offatty acids influence blood cholesterol levels and intakesof three different B vitamins influence homocysteinelevels. Some risk factors exert their effects over longperiods of time (usually by influencing the progressionof atherosclerosis). Others, however, may exert their

effects very quickly (by influencing the likelihood ofblood clotting). The more that scientists learn about thenature of coronary heart disease, the more complex thecauses seem to be.

In light of this complexity, it is easy to see why scientistsdo not fully understand the reasons for differences inc o ronary heart disease rates at diff e rent times andplaces. Even dramatic effects – such as the unexpectedlylow rate of coronary disease in France or the decrease ofmore than one-third in cardiovascular mortality in theUnited States in recent decades – cannot be easilyexplained.

The inherent limits of epidemiology

In recent years, it has become increasingly evident thatthe science of epidemiology has inherent limits.Although epidemiology is very effective in identifyingstrong links between an environmental factor and adisease (for example, the link between smoking andlung cancer), it is less effective in discerning weakerassociations. As Dr. Michael Thun of the AmericanCancer Society stated in a recent interview in the journalScience, "With epidemiology you can tell a little thingfrom a big thing. What's very hard to do is to tell a littlething from nothing at all." Many of the associationsbetween diet and disease are "little things": if the effectsare real, they are relatively subtle. It may be impossibleto determine, from epidemiology alone, whetherrelatively weak associations between diet and diseaseare real or whether they reflect some type of subtle biasor measurement error that the researchers were unableto eliminate.



Goals of nutritional epidemiologyNutritional epidemiology has several goals. Perhaps themost basic is monitoring the food consumption,nutrient intake and nutritional status of a population.Other key goals are to generate new hypotheses aboutdiet and disease, to produce evidence that supports orrefutes existing hypotheses and to assess the strength ofdiet-disease associations. Ultimately, the overall goal ofnutritional epidemiology is to contribute to theprevention of disease and the improvement of publichealth.

Epidemiology alone cannot reach this goal, however.Most types of epidemiological studies (with the notableexception of intervention trials, as described in a latersection) can only identify associations; they cannotprove that an exposure causes a health effect. Whenintervention trials are not possible, other scientificmethods must be combined with epidemiology toestablish a convincing causal relationship.

Advantages and disadvantages ofnutritional epidemiologyAdvantages

As several of the previously cited examples illustrate, akey advantage of nutritional epidemiology is its directrelevance to human health. Epidemiologists study reallife. They do not need to extrapolate from animalmodels or in vitro systems. The results of their work areoften used to calculate direct estimates of risk, whichcan then be translated into specific recommendationsfor changes in nutrient intakes or food consumptionpatterns.

Findings from nutritional epidemiology can even havedirect implications for food processing and technology.For example, recent epidemiological studies associatinghigh intakes of trans fatty acids (found in margarine andother processed vegetable fats) with increased risks of

coronary heart disease will probably prompt margarinemanufacturers to seek out ways to reformulate theirproducts to reduce their trans fatty acid content.

Disadvantages

Perhaps the most important disadvantage of nutritionalepidemiology is the potential for many kinds of bias.Bias is defined as systematic error, resulting in over- oru n d e restimation of the strength of an associationbetween an exposure and an outcome. As discussed inmore detail in a later section, studies in nutritionalepidemiology must be designed and executed withgreat care to minimize bias.

Another disadvantage of epidemiology is the difficultyin determining whether observed associations arecausal. If the association between a factor and a diseaseis not causal, efforts to modify exposure to that factorwill not reduce disease risk. For example, even thoughthe drinking of alcohol is associated with lung cancerrisk, efforts to discourage alcohol consumption wouldnot be likely to reduce the lung cancer death rate,because the relationship is not causal. Instead, it reflectsthe association of both alcohol intake and lung cancerwith a third factor – cigarette smoking.

The apparent simplicity and “real life” relevance ofepidemiological findings may also be a disadvantage insome circumstances because they encourage the misuseand overinterpretation of data. This is especially truewhen preliminary or unconfirmed findings come to theattention of the news media and the general public. Forexample, the reports of an association betweenmargarine intake and cardiovascular disease may haveprompted some consumers to switch back to butter,even though most experts believe that this course ofaction would not be beneficial to cardiovascular health.




Nutritional epidemiology in action: alcohol and breast cancerBreast cancer is one of the most common types of cancer among women in the industrialized countries. Effortsare being made to identify modifiable risk factors for this disease in the hope of finding ways to reduce itsincidence. One possible risk factor for breast cancer that has been singled out by epidemiological studies isalcohol intake. If any association exists between alcohol intake and breast cancer risk, it could be of hugeimportance to public health because large numbers of women drink alcoholic beverages.

More than 50 epidemiological studies have examined the relationship between alcohol intake and breast cancer.More than half of these studies have reported that drinkers have an increased risk of the disease. Typically, therelative risk in women who drink alcohol, as opposed to nondrinkers, is about 1.5.

Is this a true relationship? No one really knows. The relative consistency of the evidence tends to argue in favourof a true association. However, the increase in risk is so small that even a subtle bias might account for it.Moreover, no mechanism by which alcohol might influence breast cancer risk has been established. Somescientists argue strongly in favour of a link between alcohol and breast cancer, wheras others argue with equalfervour that no such association has been convincingly demonstrated.

It may be impossible to confirm or refute a link between alcohol and breast cancer by epidemiological means.The effect is so weak that observational studies cannot establish with certainty whether it is real. Indeed, this isan excellent example of the kind of situation that pushes observational epidemiology to its inherent limits andforces scientists to be very cautious in interpreting such studies. If an intervention trial were possible, it mightsettle the question, but such a study is unlikely to take place. It would be extremely difficult (and perhapsunethical) to randomly assign large numbers of women to drink or abstain from alcohol for years on end. In theabsence of intervention trials, a definitive answer to the alcohol–breast cancer question must await new findingsfrom other types of research – such as animal experiments or biochemical studies. Epidemiology alone isunlikely to resolve this issue.

BOX 1


KEY FEATURES OFEPIDEMIOLOGY STUDIES:DESIGN, EXPOSUREASSESSMENT AND DATAANALYSIS

Study designSeveral different study designs are used in nutritionale p i d e m i o l o g y. These include descriptive studies,ecological studies, case-control studies, cohort studiesand intervention trials. Each study design servesd i ff e rent purposes and has specific strengths andweaknesses.

It is important to distinguish between study designsthat employ aggregate data on populations and thosethat use individual data on specific members of apopulation. As a general rule, individual-based studiesare of greater value than those using aggregate data andtheir results should be given greater weight.

Descriptive epidemiology

Epidemiology studies can be classified into twocategories: descriptive and analytic. Descriptive epi-demiology is the study of the amount and distributionof diseases, exposures or other factors of interest withina population in terms of person, place and time.Analytic epidemiology is the more specific study of thedeterminants of disease in study populations.

In descriptive studies, re s e a rchers may collect data on avariety of factors including dietary intakes, biochemicalmarkers of nutritional status, risk factors for disease, theincidence of disease or rates of death from variouscauses. They may then compare the findings ford i ff e rent types of p e r s o n s: those of diff e rent ages,genders, ethnic backgrounds and socioeconomic or

occupational groups. They may compare the dataobtained from people who live in diff e rent places, that is,d i ff e rent geographic areas or diff e rent types ofcommunities (for example, urban versus rural). Ifrepeated studies are conducted, information may also begenerated on changes in nutritional variables over t i m e.

The data generated by descriptive epidemiology can beused to identify variations in the distribution of diseaseor in the distribution of nutritional factors. The findingsare also useful for comparing actual dietary intakes withestablished standards such as Recommended DietaryAllowances (RDAs). Many descriptive data arecollected for the purpose of nutritional surveillance,which is beyond the scope of this monograph.

National surveys In some countries, periodic national surveys provideextensive descriptive data that can be of great value tonutritional re s e a rchers. For example, in the UnitedKingdom, scientists can make use of the annualNational Food Survey (NFS), which re c o rds foodpurchases over a 1-week period in a representativesample of 7 500 British households. The reports from theNFS provide an excellent unbroken record of Britishfood habits since 1950.

In the United States, re s e a rchers can investigate a bro a drange of topics using the extensive data collected in thet h ree successive National Health and NutritionExamination Surveys (NHANES I, II and III). Forexample, one United States re s e a rch group recently useddata from NHANES I to investigate whether users ofvitamin supplements have lower death rates (they donot), a second group used NHANES II data to find outwhether Americans are eating the recommended numberof daily servings of fruits and vegetables (they do noteven come close) and a third group compared data fro mthe three successive NHANES surveys to see whetherAmericans are reducing their fat intake (they are, but notas rapidly as public health authorities would like).




Ecological epidemiology

Studies using aggregate data on diet and disease, alsocalled ecological studies, compare indices of nutritionalstatus with indices of health status for groups of people.This type of research is usually conducted in the earlystages of the investigation of a diet-disease relationship.Because ecological studies are relatively quick andinexpensive, often relying on data that have beenroutinely collected for other purposes, it makes sense touse them for a preliminary evaluation of a newhypothesis to determine whether more extensive andcostly investigations are warranted.

Advantages of ecological studies In some situations, ecological studies may be the onlyway to test a hypothesis. Data on a particular exposureor outcome may be available only for communities, notfor individuals. For example, it is generally not feasibleto measure individual exposure to nutrients or toxicsubstances in drinking water. Researchers can, however,easily determine the amounts of these substances in thewater supplies of different communities and comparethem with health outcomes.

By the late 1930s, it was recognized that the use of watersupplies with high fluoride concentrations led tomottling of tooth enamel. Dentists observed that peoplewith mottled teeth had low rates of dental caries andthey hypothesized that fluoride might prevent caries.The United States Public Health Service tested thishypothesis by conducting an ecological study.Researchers surveyed the dental health of children in 13cities where the fluoride concentration in the watersupply varied considerably. The results, reported in1942, indicated that dental caries decreased withincreasing fluoride content of water. A fluoride level ofone part per million appeared to be optimal; it wasassociated with a reduction in caries without anunsightly degree of mottling. These ecological findingsset the stage for later experimental studies that

established the benefit of adjusting the fluoride contentof drinking water supplies to one part per million.

Ecological studies may also be the best type ofinvestigation when the variation in exposure to adietary factor within a population is so small thatdifferences in health outcomes cannot be detected. Somescientists have argued, for example, that the relationshipbetween dietary fat intake and breast cancer risk is bestaddressed through ecological comparisons of differentpopulations, since the fat intakes of individuals within asingle population tend to be relatively homogeneous.

Limitations of ecological studies A key limitation of ecological studies is that findings forg roups of people may not be applicable to individuals.Even if aggregate data show that populations withhigher fat intakes have higher rates of coronary disease,it does not necessarily follow that the individuals whohave heart attacks are the higher consumers of fat. Nordoes it follow that diff e rences in fat intake werenecessarily responsible for the diff e rences in coro n a r ydisease rates; the populations may have diff e red in otherways that could be relevant to cardiovascular health.

Types of ecological studies One common type of ecological study, often referred toas a cross-cultural correlation study, involves thecomparison of food consumption data and disease ratesin different countries or geographic areas. Studies of thistype played an important role in the early stages of theinvestigation of the relationship between diet andcancer. More than 20 years ago, cross-cultural studiesrevealed very strong positive correlations between meatand fat consumption in various countries and the ratesof breast and colon cancer in those countries. Figure 1shows a classic set of data on fat intake and breastc a n c e r. Findings such as these prompted extensivefurther investigations of the possible link betweendietary fat and cancer.



FIGURE 1

Relation of national per capita fat intake to risk of breast cancer

Source: Willett W, Nutritional Epidemiology. (© Walter S. Willett 1989 (Used by permission of Oxford University Press)).

0 20 40 60 80 100 120 140 160 180

Total dietary fat intake (g/day)

25

20

15

10

5

Thailand

Japan Taiwan MexicoPhilippines

El Salvador

Ceylon

Colombia

PanamaVenezuelaChile Bulgaria

RomaniaYugoslavia

Puerto Rico

Hong Kong

Portugal

ItalyCzech

Hungary Finland

Poland

GreeceSpain

AustriaNorway

France

Germany

SwedenAustralia

BelgiumIreland

SwitzerlandU.S.

New ZealandDenmark

NetherlandsUK

Canada


A major problem with cross-cultural corre l a t i o n s ,however, is that many aspects of a population's lifestylemay be correlated. For example, intakes of fat and meata re highly correlated with economic development.Other measures associated with development, such asgross national product, also show strong correlationswith breast and colon cancer rates. Thus, it is unclearfrom ecological studies whether intakes of fat and meathave a true impact on cancer risk or whether theymerely serve as markers for other factors that changewith economic development, such as physical activity.Other types of epidemiological studies have notconsistently confirmed the strong associations betweendietary fat intake and breast and colon cancers seen inc ross-cultural comparisons. Despite decades ofinvestigation, it remains uncertain whether there is acausal relationship between dietary fat intake and therisk of cancers at these body sites.

Ecological studies may also be carried out within acountry if there are substantial differences in eatinghabits in diff e rent regions or subpopulations. Forexample, researchers in Belgium compared nutritionaland health indices of the Flemish-speaking and French-speaking populations within their country and foundthat those who spoke French had higher dietary fatintakes, higher serum cholesterol levels and higher ratesof coronary heart disease.

Ecological studies using individual exposure data Occasionally, ecological studies may be carried outusing data on individual members of severalpopulations. A classic example is the Seven CountriesStudy, in which individual-based studies of coronaryheart disease were carried out in 16 populations inseven countries. The data from these studies werec o m p a red in an effort to identify cro s s - c u l t u r a ldifferences.

Time trends Some ecological studies compare populations fro mdifferent times rather than different places. For example,after animal experiments suggested that saccharinmight cause bladder cancer, trends in bladder cancerrates were analyzed to see whether they could be linkedwith changes in the use of this sweetener. (No suchrelationship was found.)

Special populations Other types of ecological studies involve specialpopulations that are particularly relevant to the diseaseor dietary factor under investigation. For example, someepidemiological studies of saccharin and bladder cancerfocused on people with diabetes, because they areespecially heavy users of artificial sweeteners. (Again,no link was found between saccharin and bladdercancer.)

Other special populations of interest to epidemiologistsinclude religious groups such as Seventh-DayAdventists, who follow distinctive lifestyles. It isimportant to remember, however, that these groups maydiffer from the general population in multiple ways. Forexample, many Adventists are lacto-ovo-vegetariansand it would be tempting to conclude that this dietarypattern is responsible for their reduced risks of severalc h ronic diseases. However, Seventh-Day A d v e n t i s t salso abstain from the use of alcohol and tobacco andthese practices may be largely responsible for theirdistinctive patterns of disease.

Migrant studies Populations that have migrated from one part of theworld to another provide a unique opportunity forassessing the relative contributions of genetics andenvironmental factors to disease. Diet, of course, is oneof the environmental factors that change when peoplemove from one place to another and gradually adoptthe lifestyle of their new country or community.



Extensive studies have been conducted on severalmigrant populations including Japanese migrants to theUnited States and European migrants to A u s t r a l i a .These studies have shown that migrant gro u p seventually acquire the disease patterns of their newcountries but that the change occurs more rapidly forsome diseases than for others.

Some epidemiological terminology Incidence refers to the number of new cases of a diseasein a specific time period. It is usually expressed as anincidence rate: the number of new cases over a specifiedperiod of time divided by the number of people in thepopulation at risk of developing the disease. The use ofa rate enables populations of diff e rent sizes to becompared directly.

Incidence data are very useful because they provide adirect measure of the rate at which individuals in apopulation develop a disease and therefore provide abasis for statements about the probability or risk ofdisease. However, diff e rences in incidence can bedifficult to interpret because they may reflect patterns inthe diagnosis of a disease rather than its tru eo c c u r rence. For instance, the 46% increase in theincidence of prostate cancer in the United Statesbetween 1973 and 1987 was attributable primarily toi n c reased screening efforts and the availability ofsensitive new diagnostic tests, rather than to changes indiet or other risk factors.

Prevalence refers to the number of existing cases of adisease or other condition in a population. For example,in the initial examination of participants in theFramingham Study (a long-term study of coronaryheart disease in a United States community), 43 of the941 men 45–62 years old had coronary heart disease, fora prevalence of 46 per 1 000.

It is important to note that the prevalence of a disease

may reflect changes in the length of survival of thepatients (and therefore possibly changes in treatment) aswell as changes in incidence. The prevalence of diabetesmellitus no doubt increased dramatically after thediscovery of insulin, since patients no longer diedwithin a short time after diagnosis. There is noparticular reason to suspect, however, that the incidenceof the disease changed at that time.

Nutritional epidemiologists are interested in theprevalence of risk factors for disease as well as in theprevalence of disease itself. Data on the prevalence ofrisk factors, such as overweight and smoking, can helpauthorities set priorities for interventions. Comparisonof the data from the successive NHANES surveys hasshown that the prevalence of overweight among UnitedStates adults increased from 25% to 33% in one decade.This finding clearly indicates the need for increasedattention to this major health risk factor.

M o r t a l i t y (death) rate data are sometimes easier toobtain than incidence or prevalence data, but they areg reatly influenced by changes in the treatment ofdisease and do not simply reflect its occurrence. Forsome diseases of interest to nutritional epidemiologists,such as cancers of the lung and pancreas, mortality ratesare very similar to incidence rates because the diseasesa re rapidly and almost uniformly fatal. For otherdiseases, however, incidence and mortality patterns arevery different. Prostate cancer is a good example. The46% increase in the incidence of this disease in theUnited States was accompanied by only a 7% increase inmortality from this cause.

For some diseases, mortality data are of no valuebecause deaths are rare, even though the burden ofdisease may be very great. People do not die fromcataracts, for instance, but the treatment of thiscondition imposes a large burden on the health carebudgets of Western countries. In the United States alone,



physicians perform 1.2 million cataract extractionsannually, at a total cost of over $3.2 billion. If wayscould be found to prevent cataract, perhaps throughbetter diets, the benefits would be great. No change inmortality would be expected, however.

The need for age-standardization Most diseases do not occur with equal frequency amongpeople of all ages. Many of the diseases of greatesti n t e rest to nutritional epidemiologists, includingc o ronary heart disease, stroke and cancer, becomedramatically more frequent with increasing age. For thisreason alone, the rates of these diseases are higher inpopulations that include a large proportion of olderadults (such as the current populations of WesternEurope and North America) than in those that includefewer elderly people (such as the current populations ofdeveloping nations or the past populations ofindustrialized countries).

For a meaningful comparison of disease rates inpopulations with different age structures, it is necessaryto take account of the effects of age. This can be done bycomparing age-specific rates (for example, the incidenceof stroke among women 60–64 years old) or by using atechnique called age standardization, in which the ratesof disease in each age group are weighted by the agedistribution of a standard population. The resultingrates can then be compared in a meaningful way.

To illustrate, consider the incidence rates of stomachcancer among men in Norway and men in Cali,Colombia, as shown in Figure 2. For all age groups, theage-specific incidence of stomach cancer in Cali ishigher than that in Norway. However, because Norwayhas an older population (18% of men in Norway are 60years old or older, compared with 4.6% in Cali), thecrude incidence rate of stomach cancer in Norway isactually slightly higher than that in Cali (29.7 per100 000 versus 23.2 per 100 000). When the stomach


FIGURE 2

Comparison of crude and age-standardizedstomach cancer rates in two populations

Source: Margetts BM, Nelson, M (eds), Design Concepts inNutritional Epidemiology, (© Oxford: Oxford University Press,1991).

50

45

40

35

30

25

20

15

10

5

0Crude incidence Age-standardized

rates incidence rates

Norway Cali, Colombia


cancer incidence rates for men in Norway and Cali arestandardized by comparison to the world standardpopulation, a different and more meaningful pictureemerges. After standardization, the rate for Norway is18.1 per 100 000 and that for Cali is 48.3 per 100 000.These rates accurately reflect the age-specific rates andcorrectly suggest that men in Cali are exposed to morerisk factors for stomach cancer than are men in Norway.

Studies using individual data

For purposes of hypothesis testing, studies usingindividual data on diet and disease are generallypreferred. Figure 3 shows the principal types of studies.

Individual-based studies may be experimental( e x p o s u res are manipulated by the investigator) orobservational (they are not). The two most importantobservational study designs are the case-control studyand the cohort study.

Case-control studies For a case-control study, researchers identify peoplewho have a disease (cases) and otherwise similar peoplewho do not have it (controls) and compare theirexposures to factors that may have influenced theirdisease risk. Ideally, the cases and controls should beselected from the same population and they should berepresentative of that population. Case-control studies


FIGURE 3

Types of epidemiological studies

Source: Lilienfeld DE, Foundations of Epidemiology, © D. Lilienfield 1994 (Used by permission of Oxford University Press)

The Analytical Epidemiological Study

Observational Studies Experimental Studies

Uncontrolled assignment Controlled assignment

Sampling with regardto disease or effect

Case-Control Studies

Sampling with regard toexposure, characteristic,

or suspected cause

Cohort Studies

Communityassignment

RandomizedIndividual assignment

Community Trials Randomized Clinical (Intervention) Trials


a re re t rospective, meaning that the focus is onexposures that occurred in the past and on the ways inwhich these exposures may have affected anindividual's present health. Figure 4 illustrates the basicprinciple of a case-control study.

For example, a case-control study of chilli pepperconsumption and stomach cancer was conducted inMexico City. A total of 220 patients with stomach cancer(cases) and 752 stomach cancer–free but otherwisesimilar people selected from the general population(controls) were interviewed about their chilli pepperconsumption and other aspects of their diets. It wasfound that a higher proportion of cases than controls atechilli peppers. To state it in another way, the likelihoodof stomach cancer in chilli pepper consumers appearedto be greater than that in non-consumers.

Advantages of case-control studies. Case-control studieshave many advantages. They can investigate a widevariety of potential risk factors simultaneously, they are

relatively quick and inexpensive (in comparison tocohort studies, as described below) and they can beapplied a priori to all diseases, both common and rare.Another advantage of this type of study is that it ispossible to match subjects for important factors that arenot currently under investigation. For example, in acase-control study of lung cancer, the risk of which isprofoundly influenced by cigarette smoking, one couldselect controls with histories of smoking as similar aspossible to those of the cases, so that attention could befocused on other factors such as diet. (Even if matchingis not performed, it is important to collect informationon factors that may influence risk but are not currentlyunder investigation, so that appropriate adjustment forthese factors can be made during data analysis, asdiscussed in a later section)

Disadvantages of case-control studies. Case-control studiesalso have major disadvantages. This type ofinvestigation requires the researcher to collect inform-ation about the subjects' past exposures – a difficult task.People's memories of past dietary habits are often faultyand objective data on past exposures (for example,biological markers) may not be available.

Types of bias. Case-control studies are also subject toseveral types of bias, including selection bias andinformation bias. Selection bias occurs when studysubjects are chosen in a way that can misleadinglyincrease or decrease the magnitude of an association.This can occur when the cases and controls are selectedfrom different populations or when the subjects in eithergroup are not representative of the population fromwhich they come.

R e s e a rchers often select their control subjects fro mamong the other patients in the same hospitals wherethe cases were treated. The use of hospital controls isconvenient, but it may not be appropriate for sometypes of studies, such as those involving alcohol intake.Data indicate that hospital patients drink more alcohol


FIGURE 4

The basic principle of case-control studies

Source: Ahlbom A, Norell S, Introduction to ModernEpidemiology, (© Newton Lower Falls, MA: EpidemiologyResources, Inc., 1984).

Havecharacteristic

(exposed)

Do not havecharacteristic (unexposed)

Diseased (cases)

Not diseased (controls)


Information bias may also occur when biologicalmarkers are used as an index of nutritional status. Thelevels of some markers in the cases may be modified bythe onset of disease (for example, fatty acids after amyocardial infarction).

Another type of bias encountered in epidemiologicalstudies is re g ression dilution bias, in which thevariability in a biological measurement may lead to anunderestimation of the strength of an association. Forexample, blood cholesterol levels vary from day to daywithin a single individual. If researchers measure theirsubjects' cholesterol levels on a single occasion and thenassess the association between these levels and the riskof heart disease, the observed association is likely to beweaker than the true association because of thevariability in cholesterol measurements. This problemcan be dealt with by making repeated measurements ofparameters that tend to fluctuate or by usingmathematical methods to account for the impact ofdilution bias.

than the general public, so their hospitalization mayreflect their increased risk of alcohol-related illnessesand accidents. This in turn may lead to bias in studieswhere alcohol consumption is an important exposurevariable. According to a recent review, nine out of tencommunity-based studies of alcohol intake andcolorectal cancer risk showed a positive associationbetween these two factors, but only five out of17 hospital-based studies showed a similar association.The negative results of the hospital-based studies mayhave been due to bias.

To avoid problems of this sort, re s e a rchers may choose touse control subjects selected from the generalpopulation. Unfortunately, however, bias is also possiblein the selection of these controls. If re s e a rchers search forc o n t rols by dialing randomly chosen telephone numbers,for example, the control group will not include peoplewho do not have telephones. Because people who lacktelephones tend to be of low socioeconomic status, andbecause socioeconomic status influences the risk ofdisease, the absence of these people from the contro lg roup may bias the study's findings.

Information bias results when the method of datacollection makes the information obtained from casesand controls different in a misleading way. For example,cases may recall past events differently than healthycontrols do because they are motivated to pay moreattention to the causes of disease; this is called recallbias. If the interviewer knows whether subjects are casesor controls (for example, because cases are visibly ill),the conduct of the interview may change in subtle ways,leading to interviewer bias. Epidemiologists whoconduct case-control studies need to plan their researchso that both recall bias and interviewer bias are reducedas much as possible.


FIGURE 5

The basic principle of cohort studies

Source: Ahlbom A, Norell S, Introduction to ModernEpidemiology, (© Newton Lower Falls, MA: EpidemiologyResources, Inc., 1984).

Havecharacteristic

(exposed)

Do not havecharacteristic (unexposed)

Diseased

Not diseased


Cohort studies Unlike a case-control study, which begins with a groupof people who already have a disease, a cohort studybegins with a defined study population that is followedover a period of time. Cohort studies may be eitherprospective or retrospective. In a prospective cohortstudy, the study population is characterized at the startof the study and followed into the future. In aretrospective or historical cohort study, a cohort thatwas characterized in the past is followed to the present.The retrospective cohort study design is rarely used innutritional epidemiology and will not be discussedfurther in this monograph.

For a prospective cohort study, researchers identify apopulation of people who do not (yet) have the diseaseunder investigation and collect information on thesubjects' exposure to risk factors, including nutritionalfactors. They then follow the study subjects for a periodof time to see who develops the disease. The frequencyof disease among subjects exposed to a particular riskfactor is compared to the frequency among those whowere not exposed. Figure 5 shows the basic principle ofa prospective cohort study.

As part of the Zutphen study, a prospective cohortstudy of cardiovascular disease conducted in theNetherlands, 552 men provided information in 1970 ontheir current dietary habits, including theirconsumption of fish. During the next 15 years, 42 ofthese men suffered strokes. The researchers discoveredthat fish intake in 1970 was lower among men who laterexperienced strokes than among those who did not.

Advantages of cohort studies. A major advantage of acohort study is its prospective design, which avoidsmany of the problems common to retrospective case-control studies. Subjects are not asked to recall pastevents and bias is minimized because the data arecollected before the development of disease.


Disadvantages of cohort studies. Unlike the subjects in ac a s e - c o n t rol study, the participants in a cohort study mustbe contacted repeatedly to determine their healthoutcomes. This re q u i res long-term re c o rd keeping andsome degree of cooperation from the subjects. To facilitatethis, scientists sometimes select their cohorts from amongpeople who can be expected to cooperate well, such ashealth professionals. However, such individuals may notbe re p resentative of the general population.

Because re s e a rchers must wait for health outcomes tooccur and because many of the diseases they are studyingtake many years to develop, cohort studies are often verylong in duration. Because disease frequencies are low,even for common diseases, it may be necessary to enro l ll a rge numbers of subjects (at great expense) to obtainmeaningful results. Because scientists often do not knowhow long it might take for an exposure to modify the riskof disease, it may be necessary to determine the outcomesat several points in time. If the re s e a rchers wish to assessthe effects of changes in exposure, the exposures may alsoneed to be measured at several points in time, thusfurther increasing the cost and complexity of the study.When large numbers of subjects are used, costconsiderations usually dictate that exposures can bem e a s u red only in relatively crude ways (for example, bymailed questionnaires) rather than by the extensivepersonal interviews, physical examinations andlaboratory tests that could be used in a case-control study.

Nested case-control studies O c c a s i o n a l l y, scientists can combine some of theadvantages of a cohort study with those of a case-control study by conducting a nested case-control study,in which the cases and controls are selected from acohort about which information was collected on a prioroccasion. Studies of this type can combine the speed andlow cost of a case-control study with the re l a t i v efreedom from bias of a cohort study.



Nutritional epidemiology in action: vitamin E supplements and heart diseaseTwo cohort studies in the United States have associated the use of high-dose vitamin E supplements with substantialreductions in the risk of coronary heart disease. In one of these studies, which involved more than 87 000 women,the use of vitamin E supplements for two or more years was associated with a 41% decrease in coronary risk. In theparallel study of almost 40 000 men, vitamin E supplementation was associated with a 37% reduction in risk. In bothstudies, the types of supplements associated with risk reduction were those that provided relatively high doses ofvitamin E – 100 mg or more per day. Multivitamins and dietary vitamin E had no significant effect.

A protective effect of vitamin E against heart disease is biologically plausible. Vitamin E is an antioxidant. As such,it may protect against harmful oxidative processes that play a role in the causation of disease. One such oxidativeprocess – the oxidation of low-density lipoproteins (LDL) – is believed to contribute to the causation of athero-sclerosis. Studies in human subjects have shown that oral supplementation with large doses of vitamin E decreasesthe susceptibility of LDLto oxidation. A dose-response study has shown that this effect is observed only at doses ofat least several hundred milligrams of vitamin E per day. This dose-response relationship is consistent with the twoepidemiological studies in which significant associations were found only for relatively high dose supplements.

Although the findings of the two epidemiological studies are impressive, especially when considered in conjunctionwith the biochemical evidence, they do not constitute definitive proof that vitamin E protects against heart disease.The two studies were observational in nature; they were not randomized intervention trials. The participants in thestudies made their own decisions about whether or not to use vitamin E supplements. It is conceivable that peoplewho chose to take supplements might have differed from the others in ways that influence heart disease risk. Theseother factors, rather than vitamin use itself, could have been responsible for the observed associations. Although theresearchers made extensive efforts to measure and control for the effects of many dietary and lifestyle factors thatmay influence heart disease risk, it remains possible that some unmeasured factor may have biased the results.

In addition, the two epidemiological studies relied exclusively on questionnaires as a source of exposure data.Physical examinations of the subjects were not performed, and blood and tissue samples were not collected. For thisreason, no one knows whether the participants who reported taking vitamin E supplements actually had higherblood or tissue vitamin E levels than the others. No one knows whether the susceptibility of LDL to oxidation inthese individuals was any different from that in non-users of supplements. The full chain of events that might linkvitamin E supplementation with a reduced risk of heart disease was not demonstrated in these studies.

A definitive answer to the vitamin E–heart disease question can come only from randomized intervention trials. Twolarge-scale trials of vitamin E supplementation have already been completed, and neither showed a protective effectagainst heart disease. The doses of vitamin E used in those trials, however, were relatively low – 30 and 50 mg/day,respectively. If protective effects of vitamin E are seen only at doses of hundreds of milligrams per day, as both thecohort studies and the biochemical experiments suggest, the negative results of these trials are not really relevant.Definitive conclusions must await the completion of additional trials using higher doses of vitamin E. One such trial,involving more than 40 000 United States women, is already in progress.

BOX 2


For example, after British re s e a rch suggested thatchildren who had received an intramuscular injection ofvitamin K at birth had an increased risk of childhoodc a n c e r, a United States re s e a rch team conducted anested case-control study among the 54 000 subjects inan earlier study of infancy and childhood. Data onvitamin K exposure were obtained from the extensivemedical records collected during the original study andthe vitamin K exposure of 48 subjects who developedchildhood cancer (cases) was compared with that of 240c o n t rols randomly selected from the cohort. Noassociation was found between vitamin K injections andchildhood cancer.

Intervention trials

Intervention trials differ from all of the types of studiesdiscussed thus far in this monograph in that they areexperimental in nature; the researchers recruit subjectsand then randomly assign them to receive or not receivea treatment under investigation. Figure 6 shows the planfor an intervention trial. Ideally, the participants in atrial should not know their group assignment; this isaccomplished by giving an identical-appearing butinactive placebo to those who do not receive the activet reatment and by blinding the subjects to theirassignment (that is, not telling them until after the studyends). After a period of time, those who received thetreatment are compared with those who did not, to seeif their health outcomes were different.

Community trials A few intervention trials have been carried out at thecommunity level. The trials of water fluoridation are aclassic example. In these trials, entire communities hadfluoride added to their water supplies, while othercommunities received untreated water. After a period oftime, dental health was assessed in both groups ofcommunities and fluoride was found to be effective inreducing the risk of dental caries.


FIGURE 6

Design of a randomized intervention trial

Source: Beaglehole R, Bonita R, Kjellström T, BasicEpidemiology, (© Geneva, World Health Organization, 1993).

Study population

Selection by defined

criteria

Potential participants

Non-participants (do not meet

criteria)

Invitation toparticipate

Participants Non-participants

Randomization

Treatment Control


Individual-based trials M o re commonly, however, subjects are assignedindividually to intervention or control groups. Forexample, researchers in Hungary conducted a trial inwhich thousands of healthy women who were planningto become pregnant were randomly assigned to receivea daily multivitamin containing folic acid or anidentical-appearing placebo containing no vitamins.After more than 4 000 of the participants becamepregnant, the researchers found that six women in theplacebo group had conceived children with neural tubedefects, compared with zero women in themultivitamin group, a statistically significant difference.

Advantages of intervention trials The key advantage of intervention trials is that they canp rovide direct evidence of a cause-and-eff e c trelationship. If the subjects are randomly assigned tothe treated and control groups and if blinding issuccessful, one can assume that any difference thatdevelops between the members of the two groups isdirectly attributable to the factor under investigation.

Disadvantages of intervention trials U n f o r t u n a t e l y, however, intervention trials have a fewdisadvantages, especially when applied to certain typesof nutritional factors. One problem is that subjectscannot be blinded to some types of nutritionalinterventions. For example, in the Oslo study, men wererandomly assigned to an experimental group thatreceived individual counselling about smokingcessation and heart-healthy diets or to a control gro u pthat received no special counselling. It is impossible toadminister an intervention of this type in a blindedfashion. (Despite this and despite the fact that some ofthe controls stopped smoking or modified their diets ontheir own initiative, the incidence of heart attacks in thet reated group was 47% lower than in the control gro u p . )

Some interventions can be administered in a blindedfashion by putting the active agent into capsules, but this

a p p roach may be less than satisfactory if the real focus ofi n t e rest is foods. Supplements of ß-carotene, calcium orfish oil may be convenient to administer, but their eff e c t smay not be the same as those of the vegetables, dairyp roducts or seafoods of real interest to nutritionre s e a rc h e r s .

Intervention trials involve ethical considerations that donot apply to observational studies. Unless theexperimenters have good reason to believe that theirintervention may be helpful, it is unethical to impose iton people. At the same time, the experimenters must beless than certain that the intervention is effective or itwould not be ethical to withhold it from the controls.For example, even though scientists might like toconduct additional trials of folic acid supplementationaround the period of conception, it is extremely unlikelythat any ethics committee would approve such studies.The beneficial effect of folic acid has been established soclearly that it would be unethical to deny this treatmentto control subjects.

In many instances, dietary factors exert their effects over ap rolonged period of time. It may be necessary to continuean intervention for many years to detect an effect. Forexample, in a trial in Finland designed to evaluate thee ffect of ß-carotene supplementation on lung cancer riskin male smokers, the subjects received ß-carotene or aplacebo for an average of 6 years. (Unexpectedly, thosewho received ß-carotene showed a small but statisticallysignificant increase in lung cancer risk – a finding that hasnot yet been satisfactorily explained.)

A final limitation of intervention trials is that they canassess only one or two factors at a time. Because oflimited resources, only a few of the many intriguinghypotheses about diet and disease can be tested in thisway. In the remaining instances, scientists must makedecisions about probable causality on the basis ofobservational data, combined with evidence fro manimal and biochemical studies.



Sample size considerations For all of the study designs described above, it is crucialfor researchers to choose a sample of adequate size toallow effects to be detected. Researchers must usestatistical techniques in combination with theirknowledge of the anticipated strength of the associationunder investigation to determine the correct number. Iftoo many subjects are included, the study may becomeinordinately expensive; if too few, the study will beworthless.

Exposure assessment and data quality

Dietary intake

In all types of epidemiology, scientists need to measureboth exposures and outcomes. In nutritionalepidemiology, assessment of exposure poses particularproblems because dietary intakes are complex and mustbe measured in special ways.

Nutritional exposures may be defined in different waysfor different research purposes. Depending on the focusof the study, the exposures may be the foods that peopleeat, the nutrient or non-nutrient components of thosefoods, anthropometric measures such as height andweight, biochemical measures of nutritional status orclinical assessments. Epidemiologists need to definetheir research questions clearly to choose the mostappropriate exposures.

Because food consists of many substances, not justnutrients, it is not adequate to equate nutrient intakewith food intake. For example, even though fruits arethe most important sources of vitamin C in mostpeople's diets, it would be incorrect to equate vitamin Cintake with fruit intake. Associations between fruitintake and disease risk might be due to components offruit other than vitamin C. Similar distinctions need tobe made with regard to non-nutritive components offood. For example, coffee intake should not be equated

with caffeine intake, since coffee contains a wide varietyof chemical substances other than caffeine and sincetotal caffeine intake may come from several sources, notjust coffee.

Food consumption data

In ecological studies, researchers sometimes rely onnational food consumption or food "disappearance"statistics as a measure of dietary intake. These statistics,however, are crude. Their quality varies from country tocountry and they reflect waste and spoilage of food aswell as actual consumption.

For individual-based studies, re s e a rchers must usemore accurate methods of assessing dietary intake. Themethods in current use include dietary recalls, foodre c o rds, dietary histories and food fre q u e n c yquestionnaires.

Dietary recalls

For a dietary recall, subjects are asked to list the foodsthey consumed during a certain period of time, usuallythe preceding 24 hours. This method is relatively quickand simple and does not re q u i re the re s e a rcher to haveprior knowledge of the subjects' food habits.U n d e r reporting of food intake may be a pro b l e m ,h o w e v e r. Increasing evidence indicates that many peoplesystematically underreport their total food intakes andthat this tendency is stronger in some segments of thepopulation (especially overweight people) than in others. Dietary recalls are best suited to obtaining information onp resent diet rather than diet in the distant past, whenlapses of memory could be a problem. A single 24-hourrecall is not adequate for measurement of an individual'susual intake; multiple recalls must be used. A single 24-hour recall from each individual can be used, however, toestimate the mean nutrient intakes of groups of peoplerather than of specific individuals.



Food records

For a food record, subjects record their intake as theyeat. They may be asked to estimate portion sizes, toweigh food before they consume it or even to provide aduplicate meal for later analysis. If records are obtainedfor a sufficient number of days and if subjects cooperatewell, food records can provide a good picture of usualcurrent dietary intake. In fact, scientists often considerthe weighed food record to be the “gold standard” formeasuring food intake. However, this method imposesconsiderable burdens on the subject and can be usedonly with relatively educated, literate people. Also,subjects may not comply with the re s e a rc h e r s 'instructions to maintain their customary eating habitsduring the study period and some subjects mayunderreport their intakes.

Diet histories

For a diet history, subjects are asked open-endedquestions regarding their usual (present or past) dietaryintake. The interviewer carefully inquires about foodconsumption meal by meal, seeking information aboutvariations in intake and trying to establish the usualpattern of consumption. Interviews of this type aretime-consuming, often lasting for more than 1 hour, butthey can provide considerable detail about anindividual's eating habits, including subtle aspects suchas food preparation practices and seasonal variations infood choices. However, dietary histories, unlike recordsor recalls, require subjects to make judgments abouttheir usual food habits. Thus, the answers may reflectwhat the subjects think they eat (or what they wouldlike the interviewer to think they eat), rather than whatis actually eaten.

Food frequency questionnaires

Like diet histories, food frequency questionnaires focuson usual intake. This method is far more structured,h o w e v e r. Subjects complete an interviewer-

administered or self-administered questionnaire thatasks how frequently they consume a series of foods.Some types of food frequency questionnaires pose theirquestions in open-ended form, whereas others useclosed-ended questions with predetermined responsecategories. For example, in the open-ended type ofquestionnaire, subjects would be asked how often theyeat apples. In the closed-ended type of questionnaire,they might be asked whether they eat apples daily, fourto six times per week, one to three times per week, lessthan once per week but more than once per month, lessthan once per month or never.

Food frequency questionnaires provide a reasonablem e a s u re of usual present or past intake. They arecommonly used in epidemiological research because itis easy for the researcher (or a computer) to transformthe answers into usable data. The quality of the results,h o w e v e r, depends greatly on the quality of thequestionnaire.

To design a good questionnaire, epidemiologists needconsiderable advance knowledge of their studysubjects. If the study population is to includeimmigrants, for instance, the re s e a rchers will needinformation about their countries of origin so that theycan include appropriate ethnic dishes in theq u e s t i o n n a i re. Researchers also need an accurateknowledge of the food sources of the nutrients that theywish to study so that they can include the appropriatefoods in the questionnaire. For example, it has beenreported that some questionnaires currently used in theUnited States to assess carotenoid intake may yieldmisleading results because they omit certain mixeddishes (for example, vegetable soup) that contributesubstantially to total carotenoid intake.

Nutrient databases

If researchers wish to analyze their findings in terms ofnutrients, they need to convert the food consumption



data from their recalls or questionnaires into nutrientintake data. This requires the use of food compositiontables or nutrient databases, which can be obtainedfrom a variety of sources.

Unfortunately, the quality of food composition datavaries from nutrient to nutrient. For example, untilrecently, data on the amount of vitamin E in foods wereso inadequate that few researchers attempted to assessdietary intakes of this vitamin. Even for nutrients forwhich extensive data are available, the values indatabases may need to be refined as new scientificinformation is developed. For example, recent analysesusing high-performance liquid chro m a t o g r a p h y, thestate-of-the-art method of analysis for vitamin C, haveproduced values lower than those found in existingdatabases, indicating that the use of the databases maylead to overestimation of vitamin C intake.

Of course, if re s e a rchers express their findings in termsof foods rather than nutrients, they need not beconcerned about the potential inadequacies of nutrientdatabases. For example, the United States re s e a rc h e r swho conducted a case-control study of cataract did notneed to use a nutrient database to determine that peoplewho ate fewer than two servings of vegetables per dayw e re at increased risk of this eye disease. Expre s s i n gfindings in terms of foods also helps re s e a rchers avoidmaking unwarranted assumptions about exactly whatthe active agent might be, and it makes it easy totranslate findings into dietary recommendations. Eventhough the components of vegetables that pro t e c tagainst cataracts have not been identified, the results ofthe case-control study are of practical value because theyadd to the evidence supporting recommendations thatpeople should eat several servings of vegetables daily.

Biomarkers

With all the difficulties involved in measuring dietaryintake, it might seem simpler to assess nutritional status

using biomarkers, such as blood or urine levels of anutrient. Indeed, biomarkers are often used inepidemiological studies. In some instances, they mayeven be the preferred way to evaluate exposure to aparticular dietary factor. Sodium is a good example.Because it is difficult to determine the sodium content ofthe diet accurately, several important studies, includingthe Intersalt study (an enormous study of sodium andblood pressure involving more than 10 000 subjects in 32countries), have used urinary sodium excretion as theirmeasure of sodium intake.

U n f o r t u n a t e l y, the use of biomarkers is almost ascomplex as the assessment of diet. A wide variety offactors influence the accuracy and interpretation ofthese indicators of dietary exposure.

Most of the currently available biomarkers re f l e c tnutrient intakes over a relatively short period of time.For example, the levels of certain nutrients in urine orblood serum may change within hours or days of achange in dietary intake. Other biomarkers, such as thelevels of nutrients in red blood cells, change moreslowly, reflecting dietary intake over a period of weeksor months. Unfortunately, few of the currently availablebiomarkers reflect truly long-term intakes over a periodof years or decades. This is an important limitation,since long-term patterns of dietary intake are of greatinterest in studies of chronic diseases.

Some biomarkers are good measures of recent exposure.Urinary levels of potassium, sodium, fluoride, chlorideand iodide are considered good indicators of the intakesof these minerals. Other biomarkers, however, are oflittle value. For instance, except in deficiency situations,blood vitamin A levels are unrelated to dietary vitaminA intake. For some exposures of interest to nutritionalepidemiologists (for example, calcium, magnesium), nobiochemical indicators have been established.



Some theoretically valuable biomarkers may be of littleuse because their collection is unacceptable to studysubjects. The concentrations of nutrients in the liver areof great interest, for example, but they are rare l ymeasured in healthy individuals because few peoplewould be willing to have a sample of their liver takensolely for research purposes.

Some indicators are influenced by factors other thandietary intake. For example, smoking reduces bloodlevels of vitamin C and carotenoids. In such instances,the biomarker may still be a good indicator ofnutritional status, but it may not accurately reflectdietary intake.

Contamination of samples may occur even before thespecimens are collected and may occur after specimencollection unless appropriate precautions are taken. Forexample, specially prepared tubes must be used in thecollection of blood for zinc measurement becausestandard tubes may be contaminated with zinc.

Some biomarkers, such as vitamin C, are easilydegraded during storage. Unlike many other nutrients,vitamin C cannot be measured in blood samples thathave been subjected to prolonged frozen storage.

Finally, biochemical indicators of dietary intake are onlyas good as the laboratories that measure them. A qualityassurance programme is an essential part of anyp rogramme of laboratory analysis. Researc h e r sinvolved in multicentre or multinational studies alsoneed to pay attention to whether the assessmentsperformed in diff e rent laboratories are comparable.Comparability is so important and so difficult toachieve that some researchers choose to avoid the issuecompletely by having all analyses performed in thesame laboratory, even if this means shipping somesamples halfway around the world.

Data analysisObjectives of data analysis

The objectives of data analysis are to determine whetherassociations exist between exposures and outcomes andto assess the strength of the associations. The directionand strength of an association may be expressed as arelative risk or an odds ratio.

Relative risk

Relative risk is the ratio of the outcome rate amongpersons exposed to a certain factor divided by theoutcome rate among persons not exposed. If the re l a t i v erisk is greater than one, people exposed to the factor havean increased risk of the outcome under investigation. Ifthe relative risk is less than one, people exposed to thefactor have a decreased risk of the outcome.

For example, in a Spanish study of bladder cancer,subjects with high saturated fat intake had a relative riskof 2.25, meaning they had more than double the risk ofdeveloping bladder cancer than did those with lowsaturated fat intake. In an Italian study of colorectalcancer, subjects with high intakes of ß-carotene had arelative risk of 0.38, meaning they had about one-thirdthe risk of developing colorectal cancer compared tothose with low ß-carotene intakes.

Relative risks can be used to compare the strength ofdifferent associations. The relative risk of lung cancer inpeople with low fruit and vegetable intake compared tothose with high intake is about 2.0. The relative risk oflung cancer in smokers compared to non-smokers is atleast 10.0. Clearly, the association with smoking isstronger than the association with fruit and vegetableintake.

From a public health policy standpoint, it makes senseto focus preventive efforts on the most important riskfactors for a particular disease. For example, in the case



For example, in a study of laryngeal cancer in Italianmen, attributable risks were estimated for thre eimportant risk factors: smoking, alcohol and dietary ß-carotene intake. Smoking accounted for 77% of totalrisk, alcohol for 25%, low ß-carotene intake for 18% andthe three factors combined for 86% of all laryngealcancers. If the relationships between these risk factorsand the disease prove to be causal, efforts to modifythese three factors (especially smoking) mightsubstantially reduce the risk of laryngeal cancer in thestudy population.

Confounding

To obtain valid results from data analysis, it is necessaryto consider the possible effects of confounding factors(factors that are unequally distributed in the groupsunder study and that may give rise to spuriousassociations). Researchers may deal with this problemby analyzing data separately for subjects who fall intodifferent categories (strata) in terms of the confoundingfactor. For example, epidemiologists often analyze dataseparately for men and women, for smokers andnonsmokers or for different age groups.

In many situations, researchers use statistical techniquesto adjust for the effects of confounding factors duringanalysis. For instance, most epidemiologists who areanalyzing a study of diet and breast cancer would wantto adjust for reproductive factors known to influencebreast cancer risk, such as age at menarche, age at firstchildbirth and parity (number of births).

Energy intake is an important confounder in manynutritional epidemiology studies. Total energy intake ispositively correlated with intakes of most nutrients. Theinterpretation of epidemiological findings may changedepending on whether this fact is taken into account. Ina recent trial, researchers fed women varying numbersof eggs (or a placebo consisting of cholesterol-free eggsubstitute) and measured the effect on serum cholesterollevels. They found that serum cholesterol increased by


of lung cancer, prevention should focus on theelimination of cigarette smoking, since this factor hasthe greatest impact on the relative risk of this disease.Encouraging fruit and vegetable consumption wouldalso be helpful, but it is less important than smokingcessation.

Odds ratios

Instead of relative risk, some case-control studies usethe odds ratio, which is the ratio of the odds of exposurefor cases to the equivalent odds for controls. Forrelatively rare diseases, the odds ratio is a good estimateof the relative risk. As with relative risk, odds ratiosgreater than one indicate an increase in risk and oddsratios less than one indicate a decrease in risk.

For example, in one recent study, women with lowblood levels of ß-carotene had an odds ratio for cervicalcancer of 3.1, meaning that their likelihood of cervicalcancer was increased. In another study, subjects whohad a history of being breast-fed in infancy had an oddsratio for breast cancer of 0.74, indicating that theirlikelihood of breast cancer was decreased. Because theincidence rates of cervical and breast cancers in thestudy populations were low, these odds ratios can beused as estimates of relative risks.

Attributable risk

Another measure of the relationship between exposureand disease, with a meaning diff e rent from that of therelative risk or odds ratio, is attributable risk – the extentto which the occurrence of a disease or other outcomevariable can be attributed to a particular factor. A re l a t e dm e a s u re, the population attributable risk pro p o r t i o n ,indicates the proportion of all cases in a total definedpopulation that can be attributed to the factor. Thism e a s u re reflects both relative risk and the frequency ofthe factor in the population and it is useful fordetermining whether efforts to modify a factor would belikely to have a substantial impact on public health.


2.81 mg/dl for each 100 mg of egg cholesterol added tothe women's diets – a much greater increase than the1.47 mg/dl observed in a similar study of men. It mightappear from these values that women are moreresponsive than men to the cholesterol-raising effects ofeggs, but this interpretation is probably not correct. Thewomen had lower energy intakes than the men andt h e re f o re the added eggs re p resented a larg e rproportion of their total food consumption. To make ameaningful comparison between the results for the twosexes, it would be necessary to adjust for the differencein energy intake.

Epidemiologists must be careful not to treat a factor asa confounder when it is actually a part of a causalpathway. For instance, one of the ways in which obesitymay increase the risk of heart disease is by causing anincrease in blood pressure, so it would not be desirableto adjust for blood pressure in a study of obesity andheart disease.

Steps in data analysis

To analyze their data in a meaningful way,epidemiologists must use strategies that enable them toaccount for the effects of known confounding factorsand perhaps to discover some others, as well as toestimate the effects of the factors under investigation.The possibility of interactions among different factorsmust also be considered.

Multivariate analysis is often used in situations whereseveral variables must be accounted for simultaneously.The techniques of multivariate analysis diff e r, rangingf rom simple cross-classification and adjustment to morecomplex methods of statistical re g ression analysis.Multivariate techniques allow re s e a rchers to determinewhich of the variables has an independent associationwith the outcome, to detect interactions among variablesand to measure the relative contribution of each variableto the risk of the disease. It is important to note,h o w e v e r, that multivariate analysis, like univariate

analysis, does not distinguish causal from noncausalassociations. Also, dealing with interactions is difficult. Ifinteractions are present, it may be better to pre s e n tstratified results for diff e rent segments of the studypopulation (for example, smokers versus non-smokers).

INTERPRETATION

When interpreting the findings of epidemiologicalstudies, scientists must consider two types of validity:internal validity, or the degree to which the findings aretrue for the study subjects and external validity, or thedegree to which the findings can be extended to personsother than the study subjects.

Internal validityMany questions must be asked to assess the internalvalidity of a study. Have the exposure and outcomevariables been measured accurately? Is it likely thatselection bias or information bias may have occurred?Have relevant confounders been identified andm e a s u red accurately? Has proper adjustment beenmade for the effects of confounders? If measurementerror or bias might have occurred or if an unmeasuredconfounder might have been present, how would thisaffect the results? In what direction might the estimatesbe biased?

If bias might merely have weakened the strength of anassociation, its effect might not be crucial to theinterpretation of the study's findings. If bias might haves t rengthened an association or created a falseassociation, however, it might invalidate the study'sresults. The following two examples illustrate thesecontrasting situations.

A c a s e - c o n t rol study demonstrated a significantnegative association between dietary folate intake andthe risk of heart attack in a group of men and women



from the eastern United States (This association, whichis most likely mediated by the effect of folate nutritureon blood levels of the amino acid homocysteine, hasalso been observed in several other studies.) There isgood reason to question the accuracy of the assessmentof dietary folate intake in this study. Folate is very labile;seemingly small differences in the processing, cookingor storage of foods can lead to large differences in folatecontent. Because of the great variability in the folatecontent of foods, there might have been some non-d i ff e rential (random) misclassification of the folateintakes of the subjects. However, this misclassificationwould not invalidate the study's results, because itwould tend to weaken, rather than strengthen, anyassociation between folate and heart disease. Ifanything, the true association might be even strongerthan that reported in this study.

A large case-control study conducted by the UnitedStates National Cancer Institute showed that peoplewho regularly took vitamin E supplements had asubstantially and significantly lower risk of oral cancerthan those who did not take them (see Figure 7). Doesthis mean that vitamin E protects against oral cancer?Not necessarily. People who choose to take supplementsare generally more health conscious than those who donot; their lifestyles tend to be more healthful in manyways, only some of which can be measured andaccounted for in an epidemiological study. If otherhealth practices are responsible for some or all of thereduction in cancer risk observed in the supplementusers, the true effect of vitamin E supplementationwould be weaker than that reported in this study.Indeed, there might be no true association at all.

These two examples illustrate the general point thatnon-differential misclassification (random error that isequally likely to occur in all subgroups of studysubjects) is of less concern than d i f f e re n t i a lmisclassification (non-random error that affects somes u b g roups of study subjects more than others).

Differential misclassification can result in a wide varietyof erroneous conclusions because it can eitherstrengthen or weaken an association and it can evencreate a spurious association where none really exists. Inalmost all circumstances, however, non-diff e re n t i a lmisclassification can only weaken an association.

To appreciate the limited effect of non-diff e re n t i a lmisclassification, it may help to imagine that you areattempting to draw a picture of a horse while travellingin an automobile over a very bumpy road. The randomjostling of your hand will probably cause you to makee r rors in your drawing – errors that decrease theresemblance of your drawing to a real horse. This is


FIGURE 7

Oral cancer risk in users and non-users of vitamin E supplements

Source: Gridley G, McLaughlin JK, Block G, et al., Vitaminsupplement use and reduced risk of oral and pharyngeal cancer,American Journal of Epidemiology 135:1083–1092, 1992.

2

1.8

1.6

1.4

1.2

1

0.8

0.6

0.4

0.2

0Supplement users Non-users


similar to the weakening of an association by randomerror in an epidemiological study. The movements ofthe vehicle might even distort your picture so much thatthe animal would be unrecognizable, which is similar tothe weakening of an association to the point where it isno longer detectable. It is very unlikely, however, thatthe bouncing of the car would cause you to draw thehorse more accurately than you would have done if youwere sitting at a desk. Similarly, it's very unlikely thatnon-differential misclassification in an epidemiologicalstudy would strengthen an association. And just as therandom jostling of the automobile would not cause youto draw an elephant instead of a horse, non-differentialmisclassification almost never creates spuriousassociations.

The role of chance

Epidemiologists must also consider the possibility thatan observed association between an exposure and anoutcome might be due to chance. This possibility iscustomarily assessed using formal tests of statisticalsignificance. Most often, a finding is considere dsignificant if there is only a 5% (that is, one in 20)likelihood that a result as extreme as that observedwould have occurred by chance, in the absence of a trueassociation between the exposure and the outcome.

For example, in a Korean case-control study of stomachcancer, the use of a home refrigerator was significantlyassociated with decreased risk of the disease. Thismeans that the probability that the observed associationwas due to chance alone was less than one in 20. (Homerefrigeration might reduce risk because it helps topreserve fresh foods, which might provide protectivenutrients. The availability of a refrigerator might alsoreduce dependence on traditional preservation methodssuch as salting, smoking and pickling, which have beenassociated with increased stomach cancer risk.)

Confidence intervals

When findings are expressed as relative risks or oddsratios, the role of chance is usually evaluated using aconfidence interval. The confidence interval representsthe range within which the variable is likely to lie. Mostcommonly, researchers use a 95% confidence interval,which indicates that there is a 95% probability that thetrue value of the relative risk (or odds ratio) fallsbetween the two stated values. To put it another way, ifthe study were repeated on many different samplesfrom the same population and the confidence intervalswere calculated for each, 95% of these would include therelative risk (or odds ratio) estimated in the first study.For example, a study of United States men evaluated theeffects of multivitamin use on the later development ofcataracts. Among current smokers, the relative risk ofcataract in multivitamin users, compared withnonusers, was 0.38, with a 95% confidence interval of0.16–0.92. This means that there is a 95% probability thatthe value of the relative risk lies within this interval.

Statistical significance versus clinical significance

Even if an association is statistically significant, it mightnot be biologically or clinically significant. Some effectsare too small or too rare to be of practical importance. Inan analysis of data on more than 1 800 participants in along-term trial of vitamin A supplementation, subjectswho received supplements showed a statisticallysignificant increase in serum triglyceride levelscompared with those receiving placebo. Because thiswas an experimental study, it is almost certain that therelationship between the exposure (vitamin A) and theoutcome (increased triglycerides) was causal (althoughtriglyceride level was probably not a previously definedendpoint of the study). However, the magnitude of theeffect was very small – too small to cause a meaningfulchange in cardiovascular risk. The investigatorsconcluded that this minor effect was not clinicallysignificant even though it was statistically significant.



Conversely, it is possible for an effect to be of potentialclinical significance even if it is not statisticallysignificant. Sometimes an apparent effect may beimportant enough to justify action, even though no onecan be sure whether the effect is real. For example, inearly 1996, a United States research team discontinued arandomized trial of ß-carotene/vitamin A s u p p l e-mentation when preliminary results showed highernumbers of lung cancers and deaths in thesupplemented group than in the placebo group. Theincreases were not statistically significant and it wase n t i rely possible that they were due to chance.However, the researchers were concerned about theseincreases because they suggested the possibility of harmto the study participants and because they resembledeffects seen in a previous ß-carotene trial in Finland. Forthese reasons, a decision was made to stop the trial,even though it was uncertain whether the supplementshad truly increased lung cancer incidence or mortality.

Power

The term power refers to the relative frequency withwhich a specified effect would be detected by aproposed study if the study were to be repeated usingthe same conditions. To be reasonably sure that such anassociation will be detected, epidemiologists must plantheir studies so that they have adequate power, which islargely a matter of choosing a sufficient sample size togive a high probability that true effects will be found.For example, for a trial of vitamin A supplementation inpatients with the hereditary eye disease re t i n i t i spigmentosa, the researchers chose a sample size of 601subjects to give the study a power of greater than 95%.This means that the likelihood was more than 95% thatan effect of predefined amplitude would be detected ifit were present. Knowing the power of a study isparticularly important in the interpretation of negativefindings, that is, situations where no association isfound. (In this particular trial, however, the results werepositive. Vitamin A supplementation significantlyslowed the progression of retinitis pigmentosa.)

The power of a study also plays a role in determiningwhether the confidence interval around a particularvalue is narrow or wide. If the power of a study is low,the confidence interval is likely to be wide, making itdifficult to determine the strength of an association. Forexample, a small United States study (which includedonly 61 subjects) found that women with high dietaryintakes of animal fat had a significantly higher risk ofbone fractures than those with lower intakes. The oddsratio was 5.00, suggesting that the increase in risk wassubstantial. However, the 95% confidence limits aroundthis value were 1.33–18.82, a confidence interval so widethat it is impossible to know whether the true effect ofanimal fat was relatively small or extremely large.

Exposure range and dose-response relation

In some instances, a study may fail to detect anassociation because intakes of the dietary factor underinvestigation are too low, too uniform or not in therange where effects are likely. For example, althoughmost epidemiological studies of oral and pharyngealcancers have found significant negative associationswith fruit and vegetable intakes, it is not surprising thata case-control study in Puerto Rico failed to do so. Inthis population, intakes of fruits and vegetables wereextremely low: 75% of the subjects said that they neverate fruit and 87% ate no more than two servings ofvegetables per week. It is likely that very few people inthis population ate enough of these foods to influencetheir cancer risk substantially.

The several studies that have examined the associationbetween fish intake and heart disease risk provide adifferent type of example. In this instance, significantassociations have been demonstrated only inpopulations in which some people eat fish and othersdo not. In populations where everyone eats at leastsome fish, no significant associations between fishintake and heart disease risk have been detected. This isinterpreted as meaning that eating a small amount offish is associated with some protection against heart



disease, but that eating a lot of fish is not associatedwith any additional benefit.

External validityThe term external validity refers to the applicability ofepidemiological findings to people other than thesubjects who were studied. To assess external validity,scientists evaluate a study's findings in conjunctionwith other available evidence. They also considerpossible differences between the study subjects and thetarget population to whom they wish to generalize theresults.

When many studies in diff e rent populations yieldsimilar results, the findings are likely to be externallyvalid. The inverse association between fruit andvegetable intake and cancer risk is a good example ofthis. This association has been demonstrated in morethan 130 epidemiological studies conducted in a widevariety of populations from at least 17 nations,including both industrialized and developing societies.The evidence for this association is exceptionallyconsistent, particularly for epithelial cancers (forexample, cancers of the lung, mouth, larynx,oesophagus and stomach). Most nutrition scientists andepidemiologists would agree that this association can beextrapolated with confidence to populationsthroughout the world.

Meta-analysis

The interpretation of epidemiological findings is farmore difficult when the results of different studiesconflict with one another. In this situation, theinconsistency can sometimes be resolved through ameta-analysis – a quantitative technique in which thestatistical results of separate studies are pooled to yieldoverall conclusions. One recent meta-analysis evaluated15 epidemiological studies that had evaluated therelationship between blood lead levels and systolic

blood pressure in men. Although the original studiesdid not have consistent results, the meta-analysisshowed a significant positive association between leadand blood pressure. The researcher who conducted theanalysis concluded, therefore, that it would be prudentto continue efforts to minimize exposure to lead fromfood, water and other sources.

Meta-analysis may be more objective than traditionalcritical reviews of the literature and it can help makesense out of studies too small to provide re l i a b l eanswers when analyzed individually. However,decisions about which studies to include in a meta-analysis can be difficult. Opinions differ on whetherflawed studies and unpublished studies (often the onesthat give negative results or no effect) should beincluded in meta-analyses and on whether studies ofbetter quality should be given greater weight than thoseof lesser quality.

Studies that are to be combined in a meta-analysisshould be similar in terms of the types and amounts ofexposures and the types of outcomes assessed. If thestudies differ in these respects, the meta-analysis mayyield erroneous results. For example, a meta-analysisthat supposedly indicated that vitamin C has no effecton the common cold has been questioned because itincluded studies that employed low doses of vitamin C(200 mg/day or less) as well as those that usedmegadoses (1–5 g/day). The negative results of the low-dose studies may have diluted the more positive resultsof the high-dose studies, leading to an incorre c tconclusion. Critics have argued that it would have beenbetter to analyze the two groups of studies separately.

Extrapolation to different populations

Epidemiologists and public health authorities shoulduse caution when extrapolating findings from onegroup of people to other, very different populationg roups. For example, supplementation with a



combination of ß-carotene, vitamin E and seleniumcaused a significant reduction in cancer mortality in theparticipants in an intervention trial in Linxian, China,but it would be wrong to conclude on the basis of thistrial that people in Western Europe would necessarilyreduce their risk of cancer by taking a similar

supplement. Linxian is a poor rural area with unusualpatterns of cancer risk and a high prevalence ofmarginal nutritional deficiencies. Findings obtained inthis atypical and nutritionally deprived population maynot be generalizable to better-nourished populationswith very different patterns of disease.


Evaluating the quality of epidemiological studiesEpidemiological findings are only as good as the studies that produce them. To assess the quality ofepidemiological studies, readers should ask many critical questions about their design and execution, includingthe following:

Intervention trials

• Were the subjects assigned randomly to the treatment and control groups?• Was randomization successful (that is, were the treatment and control groups truly comparable with respect to

important variables)?• Were efforts made to determine whether subjects complied with the treatment protocol? • Was compliance good? • Were subjects kept “blind” to their treatment assignment through the use of a placebo or other means?• Was blinding successful (that is, were subjects unable to guess their group assignment correctly)?• Were the researchers kept "blind" as to the subjects' treatment assignment? • Was the assessment of outcome blinded?• Was the number of subjects large enough to yield statistically reliable results?

Cohort studies and nested case-control studies1

• Was the number of subjects large enough to allow effects to be detected?• Was the follow-up period long enough to yield meaningful results?• Did the investigators clearly define the reference population from which the cohort was drawn?• Were exposures and outcomes assessed by accurate and appropriate methods?• Were exposure assessments repeated during the study to account for possible changes over time?• Were many subjects lost to follow-up? If so, could this have affected the study outcome?• Were data collected on potential confounding variables? Were these variables taken into account in the data

analysis?

BOX 3



• Did sufficient time elapse between the initial examination and the diagnosis of disease to ensure that metaboliceffects of disease could not have influenced biological variables measured in the study?

• How were the diagnoses confirmed in the subjects who developed the disease under investigation?• If biological samples were collected at the outset of the study and stored for later analysis, were suitable storage

conditions used? Were all samples handled in the same manner?

Case-control studies1

• Was the number of subjects large enough to allow effects to be detected?• How was the diagnosis confirmed in the cases?• Were exposures and outcomes assessed by accurate and appropriate methods?• Were the rates of participation in the study (by both potential cases and potential controls) high? If not, could this

have influenced the study outcome?• Were controls selected from the same population that yielded the cases?• Were controls comparable to cases in all respects other than those under investigation?• Was information collected in the same way from cases and controls?• Did the subjects themselves provide the exposure data, or was it necessary to resort to proxy respondents?• Did the assessment cover an appropriate time frame?• Were interviewers blinded to the subjects' case or control status?• Did the researchers consider the possibility that metabolic effects of disease could have influenced measurements

such as blood pressure or blood nutrient levels?• Was the purpose of the study defined beforehand rather than after the data were collected?

Studies (of any design) that showed a statistical association between an exposure and an outcome

• Could the association have been due to chance?• Could the association have been due to bias?• Could the association have been due to confounding?• Is the association biologically plausible?• To whom does the association apply?• Is the association likely to represent a cause-and-effect relationship? (See Table 2 for a further discussion of criteria

for assessing the likelihood of causality.)

Studies (of any design) that did not show a statistical association between an exposure and anoutcome2

• Did the study have the power to detect a clinically or biologically significant effect if it were present?

BOX 3 – continued


Causality In the absence of intervention trials, scientists useseveral criteria to evaluate whether an observedassociation between an exposure and an outcome islikely to be causal. These criteria include the strengthand consistency of the association, the presence orabsence of a dose-response relationship, the temporalrelation between exposure and outcome and thebiological plausibility of a causal relationship. Table 2lists some guidelines for assessing whether anassociation is likely to be causal.

The stronger the association between an exposure andan outcome, the greater the likelihood that it is causal.An association consistently observed in diff e re n tpopulations and under diff e rent circumstances alsosuggests that it may be causal. If the intensity of aresponse increases with dose, it might be more likelythat a causal relationship exists. It is important to note,h o w e v e r, that the absence of a dose-re s p o n s erelationship does not prove that an association is not

causal; in some situations, a threshold effect exists andno dose-response relationship would be expected.

If a relationship is truly causal, an outcome should occuronly after the exposure. Temporal relationships shouldalso be considered in terms of the underlying ætiologyof the disease. For example, cancer is a chronic diseasewith a long latent period. One would not expect theincidence of cancer to increase within weeks or monthsafter the onset of exposure to a cancer-causing agent.

Finally, a causal association between an exposure and anoutcome is more plausible if there is a known orhypothesized biological mechanism by which theexposure is likely to alter the risk of a disease.

The findings of observational epidemiology studiesshould be considered in conjunction with the results ofanimal experiments, biochemical studies and othertypes of research to provide a complete picture of thecausation of a disease. Observational epidemiologyalone cannot prove that a relationship is causal.


Reviews and meta-analyses3

• Were the question(s) and methods clearly stated?• Were comprehensive search methods used to locate relevant studies? • Were explicit and appropriate methods used to determine which papers to include in the review?• Was the methodologic quality of the primary studies assessed?• Were the selection and assessment of the primary studies reproducible and free from bias?• Were differences in individual study results adequately explained?• Were the results of the primary studies combined appropriately?• Were the reviewers' conclusions supported by the data cited?

1. Recently, a group of epidemiologists developed a quantitative scoring system to judge the scientific quality of case-control and cohortstudies of nutrition and disease. In this system, case-control studies are scored in three areas (dietary assessment, recruitment of subjects andanalysis), and cohort studies are scored in four areas (dietary assessment, definition of cohort, ascertainment and analysis). For moreinformation on this scoring system and its application see Margetts BM, Thompson RL, Key T, et al, Development of a scoring system to judgethe scientific quality of information from case-control and cohort studies of nutrition and disease. Nutrition and Cancer 24:231–239, 1995.2. For more information see Friedman GD, Primer of Epidemiology, 4th ed. (© New York:McGraw-Hill, 1994), p321.3. For more information see Sackett DL, Haynes RB, Guyatt GH, Tugwell P, Clinical Epidemiology: A Basic Science for Clinical Medicine, 2nded. (© Boston: Little, Brown & Company, 1991), p380.

BOX 3 – continued


APPLICATION OFNUTRITIONALEPIDEMIOLOGY DATA

As the examples in this monograph have illustrated,nutritional epidemiology studies can generateinformation that is of great relevance to public health.The findings must, however, be interpreted withcaution. Scientists and policy makers need to appreciatethe inherent limits of epidemiology in the detection ofweak associations and the complexities involved inmeasuring dietary intake, avoiding bias, dealingappropriately with confounding factors, analyzing dataand assessing causality. If nutritional epidemiologyresearch is misinterpreted by the lay public or by non-epidemiologically trained scientists, it may do moreharm than good. However, if nutritional epidemiologyis applied in appropriate ways, it can be of great valueto public health authorities, health care professionalsand the food industry. The three situations describedbelow illustrate ways in which findings from nutritionalepidemiology might be applied by these diff e re n tgroups.

Hypertension in the United States An analysis of data from the most recent NHANESsurvey (NHANES III) has indicated that hypertension isfar more likely to go untreated or to be inadequatelyc o n t rolled in Mexican-Americans than in othersegments of the United States population (see Figure 8).This finding will probably prompt United Statesagencies to target more of their educational efforts aboutthe importance of blood pressure control toward theMexican-American community.


Guidelines for causation

Temporal relation Does the cause precedethe effect?

Plausibility Is the associationconsistent with otherknowledge?

Consistency Have similar results beenshown in other studies?

Dose-response Is increased exposure to relationship the possible cause

associated withincreased effect?

Reversibility Does the removal of apossible cause lead toreduction of disease risk?

Study design Is the evidence based ona strong study design?

Judging the evidence How many lines ofevidence lead to theconclusion?

Source: Beaglehole R, Bonita R, Kjellström T, Basic Epidemiology(© Geneva, World Health Organization, 1993).

TABLE 2


The same survey showed that substantial numbers ofAmericans with hypertension are taking advantage ofnon-pharmacological methods of treatment, includingweight control, dietary sodium restriction andmoderation of alcohol intake. Some 3.6% of Americanswith hypertension appear to have controlled their bloodpressure through these measures alone. These findingswill probably prompt health professionals to encouragem o re of their patients to try non-pharmacologicalapproaches to blood pressure control. They may alsoprompt the food industry to continue to develop andmarket reduced-calorie and reduced-sodium versionsof their products.

Folic acid and neural tube defects A study in Leeds, United Kingdom, indicated that only19% of a group of 603 women who became pregnant in1994 had increased their intake of folic acid beforeconception. A study in South Carolina, United States,showed that only six of 71 women (8.4%) whoconceived children with neural tube defects had taken afolic acid supplement during the periconceptionalperiod. Even though these studies were conductedalmost 2 years after United Kingdom and United Statesauthorities recommended increased intakes of folic acidfor women of childbearing potential, the vast majorityof women in the two study populations had notreceived or acted on this important message. Thesedisturbing findings may increase the impetus for folicacid fortification of staple foods. They may also promptindividual health professionals to make greater effortsto ensure that the women under their care receiveinformation about the need for adequate folic acidintake and guidance on how to achieve it.

ß-Carotene For some years, it was widely believed that the inverseassociation between fruit and vegetable intake and therisk of various diseases was probably attributable inlarge part to the ß-carotene content of these foods.Animal experiments and biochemical studies hadsuggested plausible mechanisms of action of ß-caroteneand dietary or blood levels of ß-carotene had beencorrelated with reduced risks of a variety of diseases inhuman poulations.

More recently, however, some epidemiological studieshave cast doubt on this idea. As previously mentioned,not only did large intervention trials in Finland and theUnited States fail to show a beneficial effect of ß-c a rotene supplementation on lung cancer risk insmokers, but they actually suggested the possibility of asmall adverse effect. A third intervention trial in a


FIGURE 8

Percentage of hypertensive United States adultswith adequately controlled hypertension

Source: Burt VL, Whelton P, Roccella EJ, et al, Prevalence ofhypertension in the U.S. adult population, Hypertension25:305–313, 1995.

25

20

15

10

5

0Non- Non- Mexican-

Hispanic Hispanic Americansblacks whites



Nutritional epidemiology in action: iron and heart diseaseDoes good iron nutriture increase the risk of coronary heart disease? A study from Finland, published in 1992,suggested that it might. In that study, blood samples were taken from 1 931 men and analyzed for serum ferritin,an indicator of iron storage. During three years of follow-up, 51 of the men suffered heart attacks. Men with highserum ferritin levels had a significantly higher risk of heart attack than those with lower serum ferritin levels(relative risk 2.2; 95% confidence interval 1.2–4.0).

This finding was understandably disturbing to nutrition scientists and policy makers. These experts have longrecommended that people should include ample amounts of iron-rich foods in their diets to avoid irondeficiency. Now it suddenly appeared that this advice might be harmful to people's health. In fact, an editorialin the journal that published the Finnish study suggested that the new findings might prompt a reexaminationof the definition of normal iron nutriture. According to the editorial writer, "Perhaps iron depletion, defined asthe absence of iron stores without anemia, should be regarded as physiologically normal iron status."

The editorial writer was not the only one to suggest that substantial changes in dietary habits and public healthpolicy might be warranted. Numerous reports in the news media raised the possibility that recommendedallowances for iron, food fortification programmes and other policies might need to be modified. Healthauthorities did not actually implement any of the suggested changes, however. It would have been premature tomake major changes in public health policy on the basis of a single epidemiological observation. Even thoughthe idea of a link between iron and heart disease is biologically plausible (in fact, the existence of such anassociation had been hypothesized more than a decade earlier), a single, unconfirmed study is not a sufficientbasis for action.

In the years since the Finnish study was published, epidemiologists have conducted additional studies of therelationship between iron status and cardiovascular disease, both in Europe and in North America. One U.S.study provided some support for the hypothesis of an association between iron and heart disease, but the patternobserved was different from that seen in Finland. (High intakes of one type of dietary iron – haem iron – wereassociated with increased risk, but total iron intake was not.) Other studies found no association between ironand cardiovascular risk, and still others suggested that high iron status might even be protective. The totalevidence currently available (as of mid-1996) does not support the hypothesis that good iron nutriture promotesheart disease.

If public health authorities had changed their dietary recommendations when the Finnish findings were firstpublished, they would have made an unwise decision. Although the Finnish report seemed alarming at first,scientists and policy makers acted prudently when they chose to wait for further data before deciding whetherto take action.

BOX 4



largely nonsmoking population showed no associationbetween ß-carotene supplementation and the risk ofc h ronic diseases. A c a s e - c o n t rol study of maculardegeneration (a degenerative disorder of the retina thatis a major cause of visual impairment in the elderly)found an inverse association with dietary intakes oflutein and zeaxanthin (two carotenoids found in leafygreen vegetables) but not with ß-carotene. Similarly, alarge cohort study of United States women indicatedthat risk of cataract was inversely associated with intakeof spinach (which is rich in lutein but not ß-carotene)but not with intake of carrots (the richest source of ß-carotene).

These findings have dampened the pre v i o u senthusiasm over the possible health benefits of ß-carotene. Many scientists now suspect that ß-carotenemay be primarily a marker for vegetable and fruitintake and that other carotenoids and non-carotenoidcomponents of these foods may prove to be moreimportant in disease prevention than ß-carotene itself.

Public health authorities and health professionals arelikely to react to these new developments by re -emphasizing the importance of vegetables and fruits ina balanced diet and by cautioning people thatsupplements of ß-carotene or other single nutrients arenot a substitute for consumption of these importantfoods. Manufacturers of foods and supplements whohave contemplated adding ß-carotene to their productsmay now prefer to take a more cautious approach andwait for additional data before making a decision onthis issue.

As these situations illustrate, the possibilities for the useof nutritional epidemiology to improve public healthare almost endless. Of course, to make the best use ofepidemiological data, the limits of this science need tobe taken into consideration and the results ofepidemiological research should be interpreted withc a re. If, however, the findings from nutritional

epidemiology are interpreted appropriately and appliedjudiciously, they can play a major role in today's effortsto improve health through disease prevention.

SUMMARY

Nutritional epidemiology is the study of the nutritionaldeterminants of disease in human populations. It is anexciting branch of epidemiological research because itcan provide insight into the causation and prevention ofmany of today's most crucial health problems, includingthe chronic diseases of aging. However, it is a veryd i fficult field of re s e a rch because the exposures ofinterest – dietary intakes – are extremely complex. Also,many of the associations between dietary factors anddisease risks are so subtle that they are difficult todiscern using epidemiological techniques.

The goals of nutritional epidemiology includemonitoring the food consumption, nutrient intake andnutritional status of a population; generating newhypotheses about diet and disease; producing evidencethat supports or refutes existing hypotheses; andassessing the strength of diet-disease associations.Ultimately, the overall goal of nutritional epidemiologyis to contribute to the prevention of disease and theimprovement of public health, but this goal cannot bereached through the sole use of epidemiology. Mosttypes of epidemiological studies (with the exception ofintervention trials) can only identify associations; theycannot prove that an exposure causes a health effect.When intervention trials are not possible, evidence fromother lines of scientific research must be combined withepidemiological findings to determine whether anassociation is causal.

Some of the most common study designs used innutritional epidemiology include:


• descriptive studies: studies that evaluate the amountand distribution of disease within a population interms of person, place and time, without the purposeof determining the causes of disease

• ecological studies: studies that evaluate the associationbetween diet and disease using aggregate data ongroups of people

• case-control studies: individual-based studies in whichresearchers identify people who have a disease andotherwise similar people who do not have it andcompare their exposures to factors that may haveinfluenced their disease risk

• cohort studies: individual-based studies in whichresearchers identify a group of people who do nothave the disease under investigation, collectinformation on their exposure to risk factors andmonitor them for a period of time to see whodevelops the disease

• intervention trials: experimental studies in whichresearchers randomly assign subjects to receive or notreceive the nutritional intervention that is underinvestigation and then observe them for a period oftime to see whether the intervention influences theoccurrence of disease

All study designs can contribute to the fullerunderstanding of diet-disease relationships. As a generalrule, however, individual-based studies are of gre a t e rvalue than those using aggregate data. Pro s p e c t i v ecohort studies are more highly re g a rded than re t ro-spective case-control studies because the pro s p e c t i v edesign minimizes the likelihood of bias. However, case-c o n t rol studies continue to play an important role innutritional epidemiology because they are less time-consuming and less expensive than cohort studies andbecause they are applicable to a wider variety of diseasesand exposures. Intervention trials are of special intere s tbecause they are the only type of epidemiological studythat can conclusively establish a causal re l a t i o n s h i p .U n f o r t u n a t e l y, many nutritional variables do not lendthemselves to investigation in intervention trials.

Techniques used to measure dietary intakes includedietary recalls, food records, diet histories and foodf requency questionnaires. Dietary recalls area p p ropriate for assessing the intakes of groups ofpeople, but a single 24-hour recall may not give anadequate picture of a specific individual's habitualintake. Food re c o rds are often considered the bestmethod of assessing dietary intake, but they imposeconsiderable burdens on the subject and the results maynot be accurate if subjects modify their food habitsduring the time of the study or if they underreport theirintakes. Diet histories can provide very detailedinformation, but they re q u i re subjects to makejudgments about their usual food habits. Foodf requency questionnaires provide less detailedinformation, but they are well suited for use with largeg roups of people. These questionnaires must bedesigned with great care to ensure that all importantfood sources of the nutrients under investigation areincluded.

Nutritional status can also be assessed usingbiomarkers, such as blood or urinary levels of nutrientsor their metabolites. However, most of the availablebiomarkers reflect short-term rather than long-termdietary intake, and this limits their value for some typesof research.

The objectives of data analysis in epidemiology are todetermine whether associations exist betweenexposures and outcomes and to assess the strength ofthe associations. Statistical methods are almost alwaysused in data analysis and care must be taken to considerthe possible effects of confounding factors. Thetechnique of multivariate analysis is often used insituations where several variables must be accountedfor simultaneously.

When evaluating the validity of epidemiologicalfindings, scientists must consider whether findings mayhave been due to chance and whether a study had



s u fficient power to detect an association if it waspresent. When the results of epidemiological studiesconflict with one another, it may be possible to resolvethe inconsistency through meta-analysis – aquantitative technique in which the statistical results ofseparate studies are pooled in an attempt to yieldoverall conclusions. Even if a study or a meta-analysisshows a statistically significant association,epidemiologists should use caution when extrapolatingfindings from one group of people to other, verydifferent population groups.

In the absence of intervention trials, scientists useseveral criteria to evaluate whether an observedassociation is likely to be causal. These criteria includethe strength and consistency of the association, thepresence or absence of a dose-response relationship, thetemporal relation between exposure and outcome andthe biological plausibility of a causal relationship.

Nutritional epidemiology studies can generateinformation of great relevance to public health.H o w e v e r, the findings must be interpreted withcaution. Scientists and policy makers need to appreciatethe inherent limits of epidemiology in the detection ofweak associations and the complexities involved inmeasuring dietary intake, avoiding bias, dealingappropriately with confounding factors, analyzing dataand assessing causality. If nutritional epidemiologyre s e a rch is misinterpreted by the lay public or byscientists not trained in epidemiology, it may do moreharm than good. However, if it is applied in appropriateways, nutritional epidemiology can be of great value topublic health authorities, health care professionals andthe food industry. All of these groups can apply thefindings of nutritional epidemiology in ways that cancontribute to current efforts to improve health throughdisease prevention.



GLOSSARY

Analytical epidemiology: Epidemiological investig-ations specifically aimed at studying the deter-minants of diseases in study populations.

B i a s : Systematic error resulting in over- orunderestimation of the strength of the associationbetween an exposure and an outcome.

Biomarkers: Measurements in the human body or itsproducts. Some biomarkers, such as the levels ofcertain vitamins in blood serum, are used as indicesof nutritional status. Others are used as indices of therisk or progression of disease.

Case-control study: A study design in which personswith a disease (cases) are compared with thosewithout the disease (controls) to see how theirexposures to causative factors may have differed.

Cohort study: A study design in which data onexposures to possible risk factors for disease arecollected from a group of people who do not have thedisease under investigation. The subjects are thenfollowed for a period of time to see whether the laterdevelopment of disease is related to the factors thatwere measured.

Confidence interval: The range of values within whicha variable is likely to lie.

Confounding factors: Factors that distort anassociation because they are associated with anexposure as well as a disease or other outcome.

Descriptive epidemiology: The study of variations inthe occurrence of disease in terms of person, placeand time, without the purpose of establishing causalinference.

Diet history: A method of dietary assessment in whichsubjects are asked open-ended questions about theirusual dietary intakes.

Dietary recall: A method of dietary assessment inwhich subjects are asked to recall their foodconsumption over a specific period of time.

Ecological study: A study that compares the rates ofexposures and diseases in different populations usinga g g regate data on exposure and disease, notindividual data.

E p i d e m i o l o g y : The study of the distribution anddeterminants of disease in human populations andthe application of this study to control healthproblems.

External validity: The generalizability of a study'sfindings to persons other than the study subjects.

Food frequency questionnaire: A method of dietaryassessment in which subjects are asked to recall howfrequently certain foods were consumed during aspecified period of time.

Food record: A method of dietary assessment in whichsubjects record the foods that they consume.

Incidence: The number of new cases of a disease duringa given period of time in a defined population.

Internal validity: The accuracy of a study's findingswith regard to the study subjects.

Intervention trial: A study in which exposure to thefactor under investigation is modified by theinvestigator; an experimental study.

Meta-analysis : A quantitative technique in which theresults of several individual studies are pooled toyield overall conclusions.

Multivariate analysis: A set of techniques for studyingthe effects of several factors simultaneously. Thesetechniques range from simple cross-classification andadjustment to more complex methods of statisticalregression analysis.

NHANES: National Health and Nutrition ExaminationSurvey, United States.



Odds ratio: The ratio of the odds of exposure for casesto the equivalent odds for controls. The interpretationof an odds ratio is often similar to that of a relativerisk.

Prevalence: The number of existing cases of a disease ina defined population at a specified time.

Relative risk: The ratio of the outcome rate amongpersons exposed to a certain factor divided by theoutcome rate among persons not exposed.

R i s k : A general term encompassing a variety ofmeasures of the probability of an outcome. The termrisk is usually used in reference to unfavourableoutcomes such as illness or death.

Univariate analysis: Techniques for studying theeffects of a single factor on an outcome variable.

FURTHER READING

A full list of references used to compile this concisemonograph is available from ILSI Europe. Moredetailed information on this subject can be found in thetexts listed below.

General referencesAhlbom A, Norell S. Introduction to Modern Epidemiology.Newton Lower Falls, MA: Epidemiology Resources,Inc., 1984

Beaglehole R, Bonita R, Kjellström T. Basic Epidemiology.Geneva: World Health Organization, 1993

F reudenheim JL. A review of study designs andmethods of dietary assessment in nutritionalepidemiology of chronic disease. Journal of Nutrition123:401–405, 1993

Friedman GD. Primer of Epidemiology, 4th ed. New York:McGraw-Hill, 1994

Lilienfeld DE, Stolley PD. Foundations of Epidemiology,3rd ed. Oxford: Oxford University Press, 1994

M a rgetts BM, Nelson M (eds). Design Concepts inNutritional Epidemiology, Oxford: Oxford UniversityPress, 1991

Mausner JS, Kramer S. Epidemiology – An IntroductoryText, 2nd ed. Philadelphia: WB Saunders Co, 1985

Willett W. Nutritional Epidemiology. Oxford: OxfordUniversity Press, 1989



Specific topics and diseasesAmes BN, Shigenaga MK, Hagen TM. Oxidants,antioxidants and the degenerative diseases of aging.P roceedings of the National Academy of Sciences USA90:7915–7922, 1993

Ascherio A, Rimm EB, Stampfer MJ, et al. Dietary intakeof marine n-3 fatty acids, fish intake and the risk ofcoronary disease among men. New England Journal ofMedicine 332:977-982, 1995

Block G, Patterson B, Subar A. Fruit, vegetables andcancer prevention: a review of the epidemiologicalevidence. Nutrition and Cancer 18:1-29, 1992

Burt VL, Whelton P, Roccella EJ, et al. Prevalence ofhypertension in the United States adult population.Hypertension 25:305-313, 1995

Carroll KK. Experimental evidence of dietary factorsand hormone-dependent cancers. Cancer Researc h35:3374–3383, 1975

Czeizel AE. Folic acid in the prevention of neural tubedefects. Journal of Pediatric Gastroenterology and Nutrition20:4–16, 1995

Eysenck HJ. Meta-analysis and its problems. BritishMedical Journal 309:789–792, 1994

Gridley G, McLaughlin JK, Block G, et al. Vitaminsupplement use and reduced risk of oral andpharyngeal cancer, American Journal of Epidemiology135:1083–1092, 1992

Hemilä H, Herman ZS. Vitamin C and the commoncold: a re t rospective analysis of Chalmers' re v i e w.Journal of the American College of Nutrition 14:116–123,1995

Henderson BE, Ross RK, Pike MC. Toward the primaryprevention of cancer, Science 254:1131–1137, 1991

Hulley SB, Newman TB. Cholesterol in the elderly: is itimportant? Journal of the American Medical Association272:1372–1374, 1994

Lichtenstein A. Trans fatty acids, blood lipids andc a rdiovascular risk: where do we stand? N u t r i t i o nReviews 51:340–343, 1993

López-Carrillo L, Hernández Avila M, Dubrow R. Chilipepper consumption and gastric cancer in Mexico: ac a s e - c o n t rol study. American Journal of Epidemiology139:263–271, 1994

Muir CS, Waterhouse JAH, Mack T, et al (eds). CancerIncidence in Five Continents, vol. 5. Lyons: InternationalAgency for Research on Cancer, 1987

Taubes G. Epidemiology faces its limits. S c i e n c e269:164–169, 1995

Taylor A. Cataract: relationships between nutrition andoxidation. Journal of the American College of Nutrition12:138–146, 1993



ILSI Epidemiology for pdfilsi.org/.../uploads/sites/29/2016/09/Nutritional-Epidemiology.pdf ·...

Documents

Transcript of ILSI Epidemiology for pdfilsi.org/.../uploads/sites/29/2016/09/Nutritional-Epidemiology.pdf ·...