ORIGINAL ARTICLES

How Much Global Ill Health Is Attributable toEnvironmental Factors?

Kirk R. Smith,1 Carlos F. Corvalán2, and Tord Kjellström3

Over the years, estimates have been made of the portions ofhuman mortality and morbidity that can be attributed toenvironmental factors. Frustratingly, however, even for asingle category of disease such as cancer, these estimateshave often varied widely. Here we attempt to explain whysuch efforts have come to such different results in the pastand to provide guidance for doing such estimates moreconsistently in the future to avoid the most important pitfalls.We do so by carefully defining what we mean by the terms"environmental," "ill health," and "attributable." Finally,based on these recommendations, we attempt our ownestimate, appropriately qualified according to the manyremaining uncertainties. Our estimate is that 25-33% of theglobal burden of disease can be

attributed to environmental risk factors. Children under 5years of age seem to bear the largest environmental burden,and the portion of disease due to environmental risks seemsto decrease with economic development. A summary ofthese estimates first appeared in the 1997 report, "Healthand Environment in Sustainable Development," which wasthe World Health Organization's contribution to the 5-yearanniversary of the Rio Earth Summit. A full explanation ofhow these estimates were made is first presented here. Weend with a call for a program of "strategic epidemiology,"which would be designed to fill important gaps in theunderstanding of major environmental health risks inimportant population groups worldwide. (Epidemiology1999;10:573-584)

Keywords: environmental exposures, morbidity, mortality, children, environmental health risks, populations, global factors,public health.

The question posed in the title is difficult to answer to alarge extent because of a lack of basic knowledge aboutdisease etiology. This is likely to be resolved only byadditional laboratory, clinical, environmental, and epi-demiological research. It is also a moving target, in thatsome environmental changes and their health impacts havelong latency periods. There is a class of difficultiesassociated with answering this question, however, that canbe usefully addressed with existing knowledge by carefulexposition of the concepts involved and the assumptions thatare needed. Here we address these conceptual issues bycarefully and coherently defining exactly what we mean bythe terms "environment," "ill health," and "attributable." Bydoing so, many of the difficulties in answering the overallquestion are reduced or eliminated. In addition, theremaining knowledge uncertainties become more carefullydelineated, thus helping focus further research.

From the 1Center for Occupational and Environmental Health, University ofCalifornia, Berkeley, CA; 2 Department of Protection of the Human Environment,World Health Organization, Geneva, Switzerland; and 3Department ofCommunity Health, University of Auckland, New Zealand.

Address correspondence to: Kirk R. Smith, Professor of Environmental HealthSciences, University of California, Warren Hall, MC-7360, Berkeley, CA 94720.

The views expressed in this article are those of the authors and do notnecessarily reflect the position of the World Health Organization.

©1999 by Epidemiology Resources Inc.

What is "Environment" in a Health Context?The strict medical definition of environmental causes ofdiseases would be all those that are not genetic. This is theclassic dichotomy between "nature" and "nurture," in whichenvironmental factors include all those that affect theorganism after conception regardless of whether they aremediated by social conditions and individual choice orthrough environmental media. It could be argued evenfurther, however, that genetic factors are actually alsoenvironmental, but merely on a different time scale. Thus,mutation, natural selection, and other mechanisms ofevolution have changed the genetic composition of humanityaccording to environmental conditions existing in the past. Inthis context, ie, in which current genes are seen as theoutcome of previous environments, all diseases are entirelyenvironmental.1

Neither of these definitions (“diseases are entirelyenvironmental" or "all nongenetic causes are environ-mental") is sufficiently useful for most purposes, and both flysomewhat in the face of common everyday understanding ofwhat constitutes an environmental factor. In terms of health,the most important difference from common perception of theterm is the inclusion of "behavioral" or "life-style" factors.2

In particular, diet, including alcohol, and smoking areextremely significant risk factors for a range of importantdiseases and thus for total health status in many parts of theworld. Yet, their inclusion as environmental risk factorswould tend to

573

574 Smith et al

TABLE 1. Traditional "Natural" Hazards in the Environment*

Constant search for sufficient food and water while avoiding natural toxinsInfections and parasites that spread from person to person or animal to person through

air, food, water, or insectsDust, damp, woodsmoke, pollen, and other airborne hazardsInjuries from falls, fires, and animal attacksHeat, cold, rain, snow, wind, natural disasters, and other adverse conditions

* Expanded from Ref. 4.

overwhelm the other, more conventionally understoodenvironmental factors.3 Furthermore, the most effectiveinterventions for mitigation tend to be somewhat differ-ent from those for more conventional environmentalfactors, such as pollution.

There at least two other major risk categories thatintroduce ambiguity. Some of the important social riskfactors for disease, including crime, stress, and war, beingexternal to the body, are sometimes classified as envi-ronmental. In addition, natural hazards, such as earth-quakes and inclement weather, although clearly "envi-ronmenta" in some contexts, are sometimes notconsidered environmental because their frequency andscale are not generally affected by human actions. Thus,there is sometimes an unstated presumption that envi-ronmental health deals only with those aspects of theenvironment that are affected measurably by humanactivities and not those due to nature in the raw. Indeed,the term "natural" has come to imply clean and safe tomany people. This view could only develop in recenthistory in rich countries, however, because most of hu-manity has spent most of history protecting itself from arange of far-from-benign natural environmental condi-tions, some of which are listed in Table 1.4

Figure 1 shows one way to represent the relationamong these major categories of risk factors. It may notbe worthwhile to attempt to be too precise, because thedemarcations between categories are not sharp. For ex,ample, even diet, alcohol, and smoking, while certainlybeing functions of individual behavior, are also functionsof the social environment.

Epidemiology September 1999, Vol. 10 No. 5

Another way to define the scope ofenvironmental health is in terms ofmeasurable stressors that penetratefrom the outer environment across thebody barriers to affect bodily health.The usual pathways mentioned are in-gestion, inhalation, and dermal/placen-tal absorption/penetration.' Clearly theeyes and ears ought to be added to thisconventional list, however. Ultraviolet

light, laser light, and noise, for example, areconventionally measured stressors that can cause illhealth without involving the lungs, gastrointestinal tract,or skin. The point of ambiguity is reached, however,when it is not toxic stressors in the usual chemical orphysical sense, but rather toxic information content thatreaches the sense organs to cause ill health, for example,in the form of verbal child abuse, racism, or televisionviolence.

Some economists take the approach that environmentalissues are those in which the effects are externalities, ie,that they occur to a group other than those making thedecisions about the activities that cause them. When thishappens, it is argued, there is justification for social(government) action to either rearrange the marketplacesuch that the externalities are internalized (for example,pollution taxes) or to regulate. Thus, urban outdoor airpollution, which is clearly an externality, would beincluded, but household indoor air pollution caused bysources within the household would not. There are severeproblems with this argument in practice, however. Forexample, it assumes perfect information, ie, thathouseholders understand the risks, whereas individualsare clearly not able to detect even rather largeenvironmental risks on their own (or discriminatebetween minuscule and significant risks). It also assumesthat the household is a rational decisionmaking unit,when it is clear in many societies that there are significantdifferences in power and risk among household members.Finally, even if making decisions rationally andunderstanding the risks, the capability of households toenact solutions individually may be sharply limited.(How, for example, can individual households changenational oil policy to make cleaner household fuelsavailable?)

Perhaps most importantly, dismissing household u, anyother risk factors because in some theoretical sense theyare not strict externalities flies in the face of public healthtradition. This tradition is to find and reduce all threats tohealth no matter where they occur. Thus, for example, allbenzene molecules, no matter from indoor or outdoorsources, are to be kept away from people's breathingzones; there is no division into acceptable andunacceptable molecules.

It is often noted that one of the strongest arguments forthe existence of significant environmental risk factors isthe large difference in rates for the same diseases indifferent parts of the world. Although these differencescan be deeply confounded by diet, smoking, and otherquasi-environmental differences, as well as geneticvariation in some cases, the difference in rates offers the

Epidemiology September 1999, Vol. 10 No. 5

best route for helping separate out the true environmen-tal component for many diseases.

As discussed below, environmental risk factors areusually addressed only as to their effect in causing dis-ease. Taking a burden-of-disease approach, however, addsto the discussion the importance of variation in diseaseprognosis. Clearly, environment plays an impor- tant rolehere as well, and probably for nearly every kind ofdisease. Even diseases the generation of which has littleto do with current environmental conditions, forexample, genetic and sexually transmitted diseases(STDs), will take different courses depending on theenvironmental conditions to which their victims aresubjected. Thus, like nutrition, the total burden, if not theincidence, of nearly every disease and injury is affected tosome degree by environmental factors.

The two most fundamental parameters affecting thedegree to which a risk factor is defined to be environ-mental are the nonindependent choices of time period andenvironmental baseline. As discussed above, if asufficiently long time horizon is taken, all disease isenvironmental, even that related to genetic factors. Ifcausation is confined to an extremely limited period,however, long-term environmental health threats such asclimate change would be excluded. Some intermediatechoice seems most appropriate, perhaps, as in the "ruleagainst perpetuities" in English Common Law, roughlythe potential length of one lifetime, ie, 100 years, withoutexcluding potential environmental risks we may becreating for future generations.6 Long-termenvironmental risks, however, are not included in theseanalyses.

Choice of the baseline is equally critical, becausedetermination of the existence and scale of environmentalrisk requires answering, directly or implicitly, thequestion "Compared with what?" Because humans havenever lived with zero environmental risk, nor is theresuch a time in prospect, the choice of a baseline is oftennot easy. For some categories of risk, for example, envi-ronmental circulation of synthetic chemicals, a baselineof 0 may be suitable. For others, however, it may not be,for example, airborne particulates, ionizing radiation,ultraviolet light, and inclement weather.

In summary, as will be explained separately for eachmajor disease category in a later section (EstimatedEnvironmental Portion of the Global Burden of Disease),we have generally taken the following approaches indealing with the question of what will be included asenvironmental risk factors in our burden of disease de-terminations:

• We do not include genetic risk factors, as we con-sider only current and future environments, not pastones.

• We do not include the major risk factors of diet andactive smoking, but do include nonnutritional ele-ments of diet, including food additives, infectiousagents, pesticides, etc, and passive smoking (envi-ronmental tobacco smoke).

ENVIRONMENTAL FACTORS IN GLOBAL ILL HEALTH 575

§

§

• We do, however, include behavioral factors related topersonal and household hygiene, just as we do forbehavioral factors leading to community and larg-er-scale pollution.We include a modest component of environmentalrisk in the direct and indirect risks of malnutrition toaccount for degraded soils, floods, and otherhuman-engendered impacts on the quantity, quality,and distribution of food.

• We assign a component of injuries to environmentalfactors on the basis of the substantial variations inrates among different parts of the world and, also,the recognition that even social nonphysical/chemical stressors are to some extent environmen-tal. The baseline is not taken as zero risk, however.We include a small component of environmentalrisk for every disease category because of the influ-ence of the environment on disease outcome.

• We include health impacts of the natural environ-ment, such as dust exposure and natural disasters,although we do not try to determine the componentcaused by human activities, such as desertificationand global warming.

What Is Meant by "Disease?"Although we endorse the ideals behind the World HealthOrganization's definition of health as not just the absenceof disease, as proposed in its constitution in 1948, herewe focus only on the disease portion, ie, theenvironmental component of the total burden of disease,not of the total burden of ill health. We also do notattempt to address the conceptual and measurementdifficulties inherent even in defining disease. Rather, weuse the Global Burden of Disease (GBD) with relativelyminor adjustments.7 The creation of this database, how-ever, required the assistance of a large number of expertsto address a range of such conceptual and measurementissues. Among the most difficult of these issues wereestablishing a unit of measurement by which death,disease, and injury can be combined; choosing appropriategroupings by age and sex; defining appropriate geographicboundaries; combining effects in different time periods;deciding whether effects at different ages should beweighted differently; reconciling cause of death anddisease when different diseases are involved [for example,AIDS victims who contract tuberculosis (TB)]; andestablishing categories and weights for the wide range ofdisabilities associated with different diseases.

To make the dataset more tractable for this article, wehave confined it to those disease groupings that eachcause at least 1% of the GBD (see Table 2). The dis-ability-adjusted life year (DALY) is used as the basic unitof ill health. The total in DALYs is determined bysumming the years of lost life due to premature deathsplus the weighted years of disability due to a particulardisease or risk factor.8 Shown also are the separate DALYsums and populations for more-developed andless-developed countries, showing the much larger bur-den of disease per capita in the latter. (The difference iseven larger if age distributions are taken into account.)

576 Smith et al Epidemiology September 1999, Vol. 10 No. 5

TABLE 2. Global Burden of Death and Diseases in 1990 Showing Those 22 Categories Causing at Least 1% of LostDisability,Adjusted Life Years (DALYs)

World LEiCs* MDCsI

Acute respiratory infections (ARI)IDiarrheaIPerinatal conditionsChild cluster (measles, pertussis, polio, tetanus, diphtheria)ICancerIDepressiontIMalnutritionlanemia (direct effects)Heart (ischemic)IStroke (cerebrovascular disease)ITuberculosisIRoad accidentsICongenital anomaliesMalariaIMaternal conditionsSexually transmitted diseases/human immunodeficiency virusChronic obstructive pulmonary disease(COPD)IFallsIWarSuicideAlcohol (direct effects)DrowningIPercentage of total in these 22 categoriesTotal global lost DALYs (million)Population (million)Total DALYs per 1000 capita (not age-adjusted)

* LDCs less developed countries. H MDCs more developed countries. I Categories used in World Health Organization, Health and Environment in Sustainable Development, 1997.18

8.57.26.75.25.14.73.73.42.82.82.52.42.32.22.22.11.91.51.41.21.172%13795260260

9.48.17.35.84.04.34.12.52.43.12.22.42.62.42.32.12.01.51.20.81.273%12184120300

1.6%0.31.9

0.00813.77.70.99.95.90.34.42.20.0030.61.32.11.50.72.34.00.564%

161 1140 140

The only choices made in the GBD that we have alteredslightly from the published GBD database relate to thegrouping of diseases. There is no unique or absolute wayof doing such grouping. Diseases in the GBD, forexample, have been grouped, inter alia, according tobiological criteria (eg, helminths), presumed commonmechanism (eg, malignant neoplasms), organ system (eg,cardiovascular), age (eg, perinatal), sex (eg, maternalconditions), type of intervention (eg, child cluster), lo-cation (eg, tropical cluster), cause (eg, road accidents),and motive (eg, intentional vs unintentional accidents).The way diseases are grouped, although often reflectinghistorical development of understanding and convenience,can have significant implications for policy.

This is illustrated by the most important grouping inTable 2, acute respiratory infections (AR1s). Up until the1970s, the many separate bacterial and viral diseases thatnow make up ARI were often listed in separategroupings, often inconsistently from one tabulation to thenext. Realization came about, however, that althoughdifferent microbiologically, the group had common riskfactors, common gross symptoms, and commonoutcomes, particularly sharing a high risk of producinglife-threatening pneumonia in children in developingcountries.9 Grouping them together gave them muchmore visibility in health policy, particularly becausetogether they exceed diarrheal diseases (another, butolder, grouping) to become the single largest category.

The child cluster grouping (measles, tetanus, pertussis,polio, and diphtheria) is another example, in which theimportance of child vaccination programs became strik-ingly emphasized by placing the five diseases together forwhich relatively inexpensive and reliable vaccines are

available. In this case, of course, the diseases do not evenhave the same risk factors, symptoms, or outcomes.

In both of these groupings, the biological or medicalclassifications would be much less useful for policy thanone focusing on common interventions.

We have made slight adjustments to the GBD in threemajor groupings, as follows.

§

§

§

Depression here includes only unipolar and bipolardepression. Other conditions listed under "neuro-psychiatric conditions" in the GBD were not in-cluded. Inclusion of all categories would bring thetotal to 10.5% of the G13D.Heart disease here only includes ischemic heartdisease and is separated from the next category,cerebrovascular disease, because these conditionshave quite different outcomes and are often inves-tigated separately in environmental epidemiologicalstudies. In the GBD, these two categories arecombined, along with the much smaller subcatego-ries of rheumatic and inflammatory heart disease,into the category of "cardiovascular" disease (theselatter two infection-related subcategories togetheradd 1.2% additional to the GBD).We have combined the separate GBD categories of"other STDs" and "human immunodeficiency virus(H1W' into one grouping called STD/HIV.

Further modifications of groupings might also be con-sidered. For example, it might be suitable to create a newgrouping, "food/water cluster," by adding diarrhea tosome fraction of intestinal nematode infections and thetropical-cluster diseases (filariasis, leishmaniasis,schisto-

Epidemiology September 1999, Vol. 10 No. 5

somiasis, trypanosomiasis, etc), which account, respec-tively, for an additional 0.4% and 0.8% of the GBD. Thiswould raise the visibility of food/water quality andhygiene as a potential intervention.

Inclusion of asthma would add another 0.8% to theGBD for chronic obstructive pulmonary disease (COPD)in a category that could be called "chronic respiratorycluster." Given some indication of common risk factors,this might be appropriate. On the other hand, the still-uncertain and mysterious apparent rise in asthma rates insome parts of the world probably argues for keeping itseparate so that increases can be more easilyhighlighted.10

What is Meant by "Attributable?"By "attributable environmental risk," we mean the per-centage of a particular disease category that would beeliminated if environmental risk factors were reduced totheir lowest feasible values. As it considers both theprevalence of the exposures and strength of the risks, thisterm corresponds to "attributable risk percent (pop-ulation)" in the Dictionary of Epidemiology.11

Most important diseases with significant environmentalcomponents actually have multiple risk factors, acharacteristic that on the one hand makes understandingmore difficult, but on the other, offers multiple routes forfinding useful interventions. Take again, for example, thelargest disease category in the world, ARI. As shown inFigure 2, in common with many diseases in thedeveloping world, malnutrition plays a criticallyimportant role. To save * lives, case management (pin-pointing and then treating ARI-afflicted children withantibiotics) can also be effective. Some percentage of ARIis thought to be preventable with vaccinations, as well.Finally, another important risk factor is other diseases,such as diarthea, AIDS, and measles. Children

ENVIRONMENTAL FACTORS IN GLOBAL ILL HEALTH 577

with these other diseases are more likely to contract ARI,which classifies these diseases as ARI risk factors.

When quantified, this multiple risk framework can be apowerful guide to designing cost-effective and timelyinterventions to reduce disease burdens. It also provides ameans to start examining possible synergisms amonginterventions.

COMPETING RISKSGiven lack of complete knowledge, it is easily seen howthe fractions due to known risk factors might add to lessthan 100%. But surely, most people might say, they couldnever add to more than 100%, as they do in Figure 2.(After all, how can we prevent more disease than thereactually is?)

In reality, however, most of the important risk factorsact not to create a certain number of disease cases, but toenhance the existing number of cases. In this way, theycan add to more than 100%. Take a relatively straight-forward hypothetical situation of 1,000 annual deathsfrom auto accidents along a stretch of dangerous high-way. Studies might have shown that these could bereduced by 20% through requiring headlights to be usedduring the day, 40% through stricter speed limits, 50% byputting in more stop lights, 90% by installing speedbumps, 98% by having a police officer accompany eachcar, and so on. Here it is easily seen that the total, 298%,is open ended and reflects the detail with which weunderstand the problem and our ingenuity in finding waysto deal with it.

The reason that risk factors often add to more than100% is that they are usually not completely independent;changes in one will affect the others. In our highwaysafety problem, for example, consider that we arechoosing whether to save 100 lives by daytime headlightsor 200 through speed limits. Can we save 300 by doingboth? No, because once one is implemented the overallsituation changes and the remaining potential benefits ofthe other risk factors will be reduced. In this case, manyof the 200 people whose lives might be saved in aspeed-limit campaign might also have been saved bydaytime headlights. Depending on the degree of non--independence, the remaining benefit of headlights mightonly be 50, for example, making the total equal 250, not300.

The same is so with ARI. Reduction of ARI deaths by25% through vaccination, for example, would reduce thenumber of lives left to be saved through interventionsaffecting the other risk factors. Better nutrition, for anillustration, might still be able to save 30%, but it wouldnow be 30% of a lower total. A somewhat coun-terintuitive corollary is that the number of deaths ac-countable to each risk factor depends on the order oflisting the risk factors. The closer to the top of the list, thehigher the savings for each factor.

The only practical way to determine the contribution ofa risk factor (attributable risk in health terminology),therefore, is in terms of the degree to which the disease.would be changed by reducing or increasing a particular

578 Smith et al

risk factor that we can conceptualize and manipulate inrelation to other existing risk factors. There may behundreds of potential interventions, but we usually donot bother describing those that have no practicalmeaning, for example, that if we could only teach peopleto breathe less, they might not be so affected by airpollution (or assign a police officer to every car toreduce traffic accidents).

It is well to remember, therefore, that the total attrib-utable risk for all the important risk factors viewedindependently often adds to more than 100%.12 Each riskfactor must be considered in light of the others. Except inrelatively rare cases, as when a specific chemical isassociated with a particular type of unusual cancer, therisk factors will interact rather than simply add. Progressin reducing one risk factor will affect the remainingpotential of the others.13 Indeed, the more potentiallyuseful interventions are known, the greater the attribut-able risk. In contrast, it is the diseases for which knownattributable risks add to less than 100% that are mosttroublesome. Breast cancer would seem to fit this de-scription at present.

Furthermore, the statement that large fractions of ARIare attributable to air pollution, poor housing, crowding,and chilling, is not incompatible with the also-incontrovertible truth that large fractions are also due tolack of breastfeeding, vitamin A deficiency, andmalnutrition in general. In addition, the more differen-tiated the major categories such as malnutrition andenvironment become, the apparently larger the totalaggregated attributable risk for the general category be-comes. When all of the subcategories are combinedtogether into each of the main categories of either nu-trition or environment, however, they cannot be simplyadded because they interact. Thus, all possible improve-ments by nutrition alone will have a limit less than ifbreastfeeding, vitamin A, protein supplements, etc, takenseparately are added together.

The most basic of risk factors, of course, is poverty.But what is poverty? just lack of money at the householdlevel is an insufficient criterion, because provision ofmoney on its own is not effective in most cases inproducing permanent and substantial improvements. Ifone assumes, however, that alleviation of poverty wouldbring along with it the advantages of better education,nutrition, environment, access to medical care, and so onthat has generally accompanied it in the currentlydeveloped countries, the attribution of disease in the thirdworld to poverty is almost a tautology.

More tellingly, much of the history of public healthcan be viewed as success in pinpointing the specificsubcategories of attributable risk in the form of betternutrition, environment, and medical care that can beeffectively modified by education, technology, and man-agement to achieve better health before people becomerich. To propose poverty alleviation as the primary meansto improve health is to ignore the huge potentialimprovements that can be achieved well before thatfar-off day when poverty is eliminated. It also fails torecognize that improved health is itself a prerequisite for

Epidemiology September 1999, Vol. 10 No. 5

achieving and maintaining viable sustainable economicdevelopment. The causality goes both ways.

WHAT IS "ATTRIBUTABLE?"The following are some basic principles related to deter-mining the fraction of a disease category that is attrib-utable to environmental risk factors.

• All known attributable risks for a disease often add tomore than 100% and, consequently, a largeattributable fraction claimed for one risk factor doesnot necessarily imply that another risk factor cannotalso have a large attributable fraction.

• Size of individual attributable risks depends on orderthat different risk factors are examined, because ifone risk factor is reduced, the remaining diseaseavailable to be reduced by other risk factors willdecrease.

• Existence and size of an attributable risk factorpresumes existence of feasible intervention, becausethere is an unlimited number of hypothetical riskfactors that cannot be manipulated with currentknowledge or feasible application of resources.

• Defining baselines can be important; ie, it is oftennecessary to specify 2 minimum possible or reason-ably attainable risk, particularly for environmentalrisk factors that cannot be feasibly reduced to zerolevels because, for example, of natural background.8

Estimated Environmental Portion of the GlobalBurden of DiseaseOn the basis of the discussions above, we lay out hereour principal assumptions made in (1) determining whatis meant by environment for each category, (2) account-ing various diseases into one category, and (3) ascribingattributable risk.

GENERAL PRINCIPLESViableInterventionEven though measles, polio, tetanus, and other importantinfectious diseases have significant environmentalcomponents (and indeed were reduced considerably in thecurrently developed countries by environment/nutritionimprovements before vaccinations were developed), wedo not account a high proportion of their risk toenvironment, because effective vaccines are nowavailable. For other such environmentally mediated in-fectious diseases that do not currently have effectivevaccines, however, we account a high proportion toenvironment (for example, ARI and malaria). TB isintermediate.Environment FirstWe assume that environmental actions are taken first; forexample, if a disease is seen to have a significantenvironmental as well as other components, we considerwhat fraction might be averted first through feasibleenvironmental interventions, before other interventionsare applied. For example, although we recognize that ARIburden can be substantially reduced through better

MinimumThe combination of disease prognosis and nutrition,above, is taken as 5%; ie, the minimum environmentalcomponent of any disease category is 5%.

HousingWe include what might be called the "housing complex,"ie, crowding/chilling/ventilation/drainage, as en-vironmental factors.WorkplaceAll occupational injuries and illnesses are counted asenvironmental, including those of military personnel.

ISSUES SPECIFIC To EACH DISEASE CATEGORYHere we examine one by one all of the disease categoriesresponsible for at least 1% of the global burden (Table2). The rough global percentage of each category weattribute to environment is shown as a range in paren-theses at the end of each disease discussion. We believethese estimates to be reasonable, but here they are notbacked by the kind of extensive review of the exposureassessment and epidemiological literature that would beneeded to be more certain.

ARIs here are taken to include upper and lower respi-ratory infections as well as otitis media. They mostly(80%) affect children under 5 years of age. ARI, partic-ularly as pneumonia in young children, is known to havebeen essentially eliminated by environmental and nutri-tional improvements in developed countries, largely be-fore the advent of relevant antibiotics and vaccines. Chiefenvironmental risk factors are ambient air pollution andhousing conditions, such as chilling, crowding, andindoor air pollution. (40-60%)

Diarrheal diseases mostly (86%) affect young childrenand are related in the long run nearly entirely to theenvironmental factors of poor sanitation, hygiene, andaccess to clean water/food. (80-90%)

Perinatal effects, which include low birth weight, neo-natal death, and still birth, have some environmentallinkage through exposures to pollution and poor housingexperienced by the mother during pregnancy and theinfant just after birth. A significant proportion of diseaseand death in this category is known to be due to diarrheaand ARIs, both of which are primarily environmental.(10-20%)

Child cluster refers to the vaccine -preventable diseases,measles, tetanus, pertussis, polio, and diphtheria, whichin principle are largely environmental. Environmentalimprovements will decrease rates and improve outcomesbut are not as effective as vaccines, and thus environmentis unlikely to be a competitive initial intervention.(5-10%)

Malignant neoplasms (cancer) encompass quite a variedset of sites and causes. Lung cancer, the largestsubcategory, accounts for about one-eighth of the total.For cancer, nutrition, active smoking, and alcohol are notcounted as environmental, but occupational causes; foodpreservatives (including salt) and chemical (includ-

ExclusionsWe have defined sexual behavior, active smoking, alco-hol and drug abuse, and most criminal violence to beprincipally nonenvironmental risk factors.

Epidemiology September 1999, Vol. 10 No. 5

nutrition, we assume the undertaking of environmentalimprovement before other interventions are undertaken.

PreventionWe are emphasizing prevention rather than curativemeasures. Thus, although diarrhea deaths are avertedeffectively through oral rehydration therapy in the im-mediate term, the long-term solution is to minimizeactual incidence through environmental sanitation, foodsafety, and related efforts.

Cost EffectivenessWe have not done cost-effectiveness calculations todetermine whether environmental interventions are moreattractive than other kinds of intervention in every case.Nevertheless, we believe the levels of risk we attribute toenvironmental factors are compatible with the extent ofinterventions that are at least within range of technicaland economic feasibility, particularly considering thenonhealth benefits that often accompany them.

Disease PrognosisAlthough STDs and other disease categories have littlecomponent of environmental risk in terms of incidence,improved environmental conditions would ameliorate tosome extent the course of these diseases; for example, thesame number of people might contract STDs, but theresulting disease would be milder and death would bedelayed longer, with better living and working conditions.

NutritionConsidering that malnutrition also serves as an indirectrisk factor for both vulnerability and recovery from es-sentially all diseases and injuries and that environmentplays a role in malnutrition, a small additional environ-mental component is included for every disease category.

MaximumEven though diarrhea and other disease categories mightbe arguably 100% environmental, we do not assign themfull weight because of the importance of medical careand nutrition in determining the course of disease. Forexample, although the incidence of diarrhea might begreatly reduced if Bangladesh were to have the water,sanitation, and hygiene situation of Sweden, without themedical care and nutrition available in Sweden, therewould still be significantly more ill health produced bythe remaining incidence than would be found in Sweden.No disease category is assigned more than 90% toenvironmental risks.

ENVIRONMENTAL FACTORS IN GLOBAL ILL HEALTH 579

AccidentsRoad accidents, falls, and drowning are considered tohave large environmental/occupational components,principally because of the large differences in rates thatexist in different parts of the world.14

ing passive smoking) and radioactive (including radon)agents are included along with half of infectious agents,which alone account for about 15% of all cancers 15,16

(20-25%)Depression does not cause much direct mortality but

significant disability and would seem to have only smalldirect environmental linkages through stress from dete-riorating living and working environments. (5-10%)

Malnutrition here only includes direct nutritional de-ficiencies: iron-deficiency anemia and protein-energy, io-dine, and vitamin A deficiencies, and not the increasedrisk of other diseases. Malnutrition has a modestenvironmental portion through natural geographicvariations, land degradation, and other environmentalproblems leading to lower food availability and quality.(8-10%)

Ischemic heart disease has a small but verifiedenvironmental linkage through air pollution, occupation,and, perhaps also water quality. (8-10%)

Cerebrovascular disease has a suspected linkagethrough air pollution and occupation. (8-10%)

TB has important household environmental risk fac-tors, including crowding, chilling, and, probably, airpollution. (20-25%)

Road accidents are considered to have a significantenvironmental/occupational component, plus a compo-nent due to environmental (land use and transport)planning. (25-30%)

Congenital effects (birth defects) may have a smallenvironmentalloccupational component through chem-.ical exposures during pregnancy. (5-10%) .

Malaria cases are nearly all attributable to environ-mental.factors, such as land management and housing.(70-90%)

Maternal causes of death would seem to have a fewdirect environmental links, through poor household andworking conditions. 0-10%)

STDs, including AIDS (HIV), would not seem to haveany direct environmental linkages, although the firstswitch of HIV from animals to humans may haveoccurred due to environmental disruption. (5%)

Chronic obstructive pulmonary disease is a knownoutcome of air pollution exposures, both environmentaland occupational, and is enhanced by early affliction withARI. Although most chronic obstructive pulmonary dis,ease in developed countries is due to smoking, thisaccounts for less than 10% of the world total. (33-50%)

pidemiology September 1999, Vol. 10 No. 5 ENVIRONMENTAL FACTORS IN GLOBAL ILL HEALTH 581

Falls are considered to have a significant occupation-l/environmental component. (25-30%)

War injuries/deaths are considered occupational formilitary personnel and to have a small environmentalomponent for civilians due to conflicts caused by peopleying to avoid degraded environments. (50-70%)Suicide has a small component due to stress induced by

egraded living and working environments. (5%)Violence has a significant occupational component plus

small environmental contribution to criminal violence in amanner similar to suicide. (15-20%)

Alcohol use here considers only direct health impact,xcept cirrhosis, and not its contribution to other diseases,otor vehicle accidents, fires, etc. The total impact is

educed somewhat, because moderate alcohol use isccounted as beneficial to health by lowering the risk ofchemic heart disease. There would seem to be only a

mall environmental component involving the synergism ofoor environmental (housing) conditions with healthutcomes. 0-10%)

Drowning is considered to have significant occupa-onal/environmental factors. (25-30%)These 22 categories account for about 72% of the global

urden. No other disease groupings come to more than 1%f the global burden, although several come close, including0.75-1.0%) ostcoarthritis, cirrhosis, fires, eye problemscataracts, glaucoma, and trachoma), diabetes, asthma,opical cluster, and inflammatory heart disease.Shown in Figure 3 is the summary of these major cate-

ories of disease and their estimated attributable environ-ental portions.

Applying the environ-mental percentages shown inFigure 3 for each majordisease category producesan overall environmentalrate of 25-33% for this 72%of the GBD. Although notaddressed individually here,the remaining 28% of theGBD consists of many dis-ease categories that alsohave significant environ-mental components (for ex-ample, asthma, cataracts,trachoma, meningitis, hep-atitis, dengue, rheumatic andinflammatory heart disease,encephalitis, digestivedisease, dental caries, poi-sonings, miscellaneous un,intentional injuries, tropicalCluster, and other parasiticdiseases), as well as somethat would seem to haverelatively few environmen-tal determinants (for exam-ple, genitourinary and mus-culoskeletal diseases).Thus, it seems reasonableand probably conservative

expect a similar overall environmental portion, ie, aboutone-quarter to one-third.

Recognizing the uncertainty due to insufficient knowledgeof etiology and lack of adequate data in many parts of theworld, as well as in the conceptual issues discussed abovein the definitions of environment, disease, and attribution,we state our final estimate of the environmental portion ofthe GBD as a range: One-fourth to one-third (25-33%). Thisrepresents a somewhat smaller fraction (23-31%) of totalworld deaths, because much of the burden falls on youngchildren. A wider definition of what constitutes"environmental" and a more flexible definition ofattributable risk could easily raise this to 40%.17

Assuming that the percentage due to environment in eachdisease is the same throughout the world, the roughdistribution of the attributable environmental portion bymajor world region is shown in Figure 4. In reality, ofcourse, given the relative age distributions and the majorimpact of environmental factors on disease affecting youngchildren, a larger fraction is likely to be due toenvironmental factors in developing countries. Thus, the truedistribution is likely even more skewed toward poor regionsthan shown in the figure.

HEALTH AND ENVIRONMENT IN SUSTAINABLE

DEVELOPMENT REPORT

The Health and Environment in Sustainable DevelopmentReport shows estimates for only 10 disease catego-

ries.18 Here, however, we address 22 categories, by addingto and separating some of the original categories [indicated(t) in Table 21, so that all categories responsible for at least1% each of the total global burden are covered. Theenvironmental fractions for the original 10 have beenretained as the high end, except that for TB, which we havehere increased from 10% to 25%.

Environmental Risks for ChildrenGlobally, about 43% of the total burden of disease due toenvironmental risks falls on children under 5 years of age,even though they make up only 12% of the population. Theabsolute burden due to environmental factors for youngchildren varies dramatically by region, however. Figure 5shows annual DALYs lost per 1,000 persons for the entirepopulation and for those under 5 years of age just for thehigh end of the 22 categories of ill health in Figure 3. Youngchildren from developed countries (Established MarketEconomies) by this measure lose only 16 DALYs per 1,000as compared with 751 per 1,000 in Sub-Saharan Africa, afactor of nearly 50.

The main reason for these large differences is theenormous impacts still imposed in poor countries by the twolargest disease categories, ARI and diarrheal diseases. Ascan be appreciated from Figure 6, the differ-

ence in absolute risks for these diseases between devel-oped countries and Sub-Saharan Africa is more than100-fold. Thus, poor air, food, and water quality at thehousehold level remain the biggest source of environmentalrisk in much of' the world today.

ConclusionsHere we list the conclusions in relation to major issues andmajor needs that can be drawn from the analyses above.

ISSUES

1. Environmental quality is an important direct andindirect determinant of human health.

2. People in the poorest countries tend to be most at riskfrom household-related environmental quality problems,which impose the largest environmental disease burden onhumanity.

3. People in middle-income countries tend to be most atrisk from community-related environmental qualityproblems.

4. Although not addressed here, it is probably activitiesin rich countries that most threaten the global environment.

5. In the world today, it is the health of children under5 years of age that is most damaged by poor environ-mental quality.

6. One-quarter to one-third of all ill health in theworld today seems to be attributable to environmentalfactors.

7. Environmental quality is a major factor both in theinfectious diseases that tend to affect the poorest popu-lation groups and, to a lesser extent, in the chronicdiseases that tend to affect richer groups.

NEEDS

1. There is an urgent need for more and better coor-dinated local and global data collection on environmentalexposures related to the major environmental healthimpacts.

2. A global, strategic epidemiological effort is neededto fill gaps in our understanding of the relation betweenmajor environmental exposures and ill health in majorpopulation groups.

3. There is a need to consider health issues acrosssectors, because activities in a number of sectors haveenvironmental impacts that lead to health impacts, andthere are number of cross,sector interactions.

4. To be most effective, there is a need to consider theentire environmental pathway from driving forces tohealth impact when designing interventions to improveenvironment and health.`

5. Needed immediately are accelerated and integratedefforts to improve household environmental conditions

in poor countries much more rapidly and more broadlyacross groups than now seems to be occurring.

6. Needed at all level of development, but particularlyin middle-income countries, are more comprehensive andtimely introduction of community environmental controlefforts.

We have confined our analysis here to current con-ventional environmental health risks. Not discussed arethose emerging risks that probably do not yet account fora large disease burden but have the disturbing potential todo so in the not-too-distant future. Among these, forexample, are changes in the "epidemiological environ-ment" of the infectious organisms themselves due tohuman activities, such as heavy use of antibiotics10 andthe potential health impacts of human-induced climatechange."

Perhaps the most striking feature of the resultspresented here is the importance of environmentallymediated infectious diseases in the global burden ofdisease. This includes not only those in which infec. tiousagents are carried via environmental pathways, but also,in the form particularly of ARI, those in whichenvironmental risk factors apparently enhance the severityand frequency of infection. This implies that many of themost critical health problems in the world today cannot besolved without major improvement in environmentalquality. Such improvements will be needed to achievetrue sustainable development on a global basis.

584 Smith et al

AcknowledgmentsWe appreciate comments by Tony McMichael, Tony Fletcher, and Chris Murrayon earlier drafts.

References

1 . Diamond JM. Guns, Germs, and Steel: The Fates of Human Societies. NewYork: WW Norton, 1997.

2. Krieger N, Zierler S. What explains the public's health? A call for epidemi-ologic theory. Epidemiology 1996;7:107-109.

3. McKeown T. The Role of Medicine. London: Nuffield Provincial HospitalsTrust, 1976.

4. Yassi A, Kjellström T, de Kok T, Guidotti T. Basic Environmental Health.WHOIEHG/98.19 Geneva: World Health Organization, 1998 (in press).

5. Moeller DW. Environmental Health. 2nd ed. Cambridge. MA: HarvardUniversity Press, 1997.

6. Nottingham H17. Arguments (on) methods of limiting the trust of a term foryears. London: George Tatarshall, 1685.

7. Murray CJ, Lopez AD. Global Burden of Disease. Cambridge, MA: HarvardSchool of Public Health, 1996.

8. Murray CJ, Lopez AD. On the Comparable Quantification of Health Risks:Lessons from the Global Burden of Disease Study. Epidemiology 1999;10:594-605.

9. Stansfield 5, Shepard D. Acute respiratory infections. In: Jameson D, MosleyWH, Measham AR, Bobadilla JL, eds. Disease Control Priorities in Devel,oping Countries. Oxford: Oxford University Press, 1993.

Epidemiology September 1999, Vol. 10 No. 5

10. Lewis S. ISAAC: a hypothesis generator for asthma? Lancet 1998:351:1220.11. Last JM, Abramson JH, Friedman G1), Porta M, Spasoff RA, Thuriaux M,eds. A Dictionary of Epidemiology. New York: Oxford University Press, 1995.12. Walter SI). Calculation of attributable risks from epidemiological data. Int j

Epidemiol 1978;7:175-182.13. Rothman KJ. Causes. Am j Epidemiol 1976;104:587-592.14. Saari J. Accident Prevention. In: Stellman JM, ed. Encyclopaedia of Occu,

pational Health and Safety. 4th ed. vol. 2. Geneva: International LabourOrganization, 1998.

15. World Health Organization. The World Health Report: Bridging the Gaps.Geneva: World Health Organization, 1995.

16. World Health Organization. The World Health Report: Conquering Suffer-ing, Enriching Humanity. Geneva: World Health Organization, 1997.

17. Pimentel D, Tort M, D'Anna L, Krawic A, Berger J, Rossman J, Mugo F,Doon N, Shriberg M, Howard E, Lee S, Talbot J. Ecology of increasingdisease. Bioscience 1998;48:817-826.

18. World Health Organization. Health and Environment in SustainableDevelopment. Geneva: World Health Organization, 1997.

19. Corvalán CF, Kjellström T, Smith KR. Health, Environment and Sustain, ableDevelopment: Identifying Links and Indicators to Promote Action.Epidemiology 1999;10:656-660.

20. Daily GC, Ehrlich PR. Impact of development and global change on theepidemiological environment. Environ Dev Econ 1996;1:311-346.

21. McMichael AJ, Haines E, Slooff R, Kovats S, eds. Climate Change andHuman Health: An Assessment Prepared by a Task Group on Behalf of theWorld Health Organization, the World Meteorological Organization, and theUnited Nations Environment Programme. Geneva: World Health Or-ganization, 1996.