Multiple-informant rarlking of the dii:sabling effectsof different health c:onditions in 14 countries

T Bedirhan Ustiin Jiirgen Rehm Somnath Chatterji Shekhar SaxenaRobert Trotter Robin Room Jerome Bickenbachand the WHO/NIH Joint Project CAR Study Group

THE LANCET 42 BEDFORD SQUARE LONDON WC1 B 3SL UK

ARTICLES

Multiple-informant ranking of the disabling effects of differenthealth conditions in 14 countries

T Bedirhan Ustiin, Jiirgen Rehm, Somnath Chatterji, Shekhar Saxena, Robert Trotter, Robin Room, Jerome Bickenbach,

and the WHO/NIH Joint Project CAR Study Group*

Summary

Background The Global Burden of Disease study provided

international statistics on the burden of diseases, combining

mortality and disability, that can be used for priority setting

and policy making. However, there are concerns about the

universality of the disability weights used. We undertook a

study to investigate the stability of such weighting in

different countries and informant g!OUps.

Methods 241 key informants (health professionals, policy

makers, people with disabilities, and their carers) from 14

countries were asked to rank 17 health conditions from

most disabling to least disabling. Kruskal-Wallis ANOVA was

used to test for differences in ranking between countries or

informant groups and Kendall t-B correlations to measure

association between different rank orders.

FIndings For 13 of 17 health conditions, there were

significant (p<0.O5) differences in ranking betweencountries; in the comparison of informant groups, there were

significant differences for five of the 17 health conditions.The overall rank order in the present study was, however,

almost identical to the ranking of the Global Burden of

Disease study, which used a different method. Most of the

rank correlations between countries were between 0.50 and

0.70 (average 0.61 [95% CI 0.59-0.64]). The average

correlation of rank orders between different informant groups

was 0.76.

Interpretation Rank order of disabling effects of health

conditions is relatively stable across countries, informant

groups, and methods. However, the differences are large

enough to ca$t doubt on the assumption of universality of

experts' judgments about disability weights. Further studiesare needed because disability weights are central to the

calculation of disability-adjusted life years.

Lancet 1999; 354: 111-15

See Commentary page 87

IntroductionThe Global Burden of Disease study' has attracted theattention of policy makers and public-health expensbecause the study provides a common unit for evaluationand prioritY setting for a wide range of health disorders.This unit, the disabilitY-adjusted life year (DALY), addsdisabilitY to mortalitY in the overall estimation of theburden of disease. The addition of disabilitY hasincreased the relative imponance of non-communicablediseases, and the result is a more realistic measure of theglobal burden of diseases than that obtained frommonalitY alone...' DALYs provide a framework fordifferent professions in determining priorities in healthand human services and in evaluating the efficacy ofinterventions since they are based on life years, auniversal measure.' Policy makers have a particularinterest in making decisions on resource alJocation andpolicy, and monitoring the impact of health-care reformsand other interventions with a common measureapplicable in cost-effectiveness studies. The DALY isuseful because it provides a common measure by makingvalid epidemiological data usable in an internallyconsistent approach.'

There have been various attempts to developcomposite health measures that combine information onmonalitY and non-fatal health outcomes to representpopulation health in a single number.' One of the mostwidely adopted methods uses the quality-adjusted lifeyears (QALY) model.".7 This model attempts to measurethe impact ora disease in terms of different combinationsof duration and qualitY of life.' DALY is the complementto QALY, the disabilitY weight being the reverse scale ofqualitY of life. Healthy life years (HLYs) and health-adjusted life years (HALYs) are other methods used tomeasure burden of disease.' The cross-culturalapplicabilitY of these methods and the equivalence ofderived preferences have not been standardised globally,regionally, or nationally.

Although DALYs improve our understanding ofinternational health statistics and can influence the policyprocess through decisions on resource allocation, greatcare must be taken in their construction, which involveschoices and value judgments.' There is great concern thatthe calculation of the disabilitY component in DALYs isbased principally on disease-specific disabilitY weightsthat indicate values for less than perfect health. That is, aperfectly healthy state has a weight of 0, and death isequivalent to a. weight of 1. Hence, disease-relateddisabilitY is placed between these endpoints: for example,angina has a weight of 0'223, major unipolar depression0'619, and quadrip)egia 0,895.' In the Global Burden ofDisease study these weights were determined withprofessional health-care providers through the persontrade-off method.!O Professional health-care providerswere chosen because they are familiar with healthconditions and their outcomes, and such familiaritY

*Members listed at end of paper

WHO, Assessment Classification and Epidemiology Group, Geneva

(T Bedirhan OstOn MD. J Rehm rhO. S Chatterji MD. S Saxena MD.

J Bickenbach rho); Addiction Research Foundation, Centre for

Addiction and Mental Health, Toronto, Canada (J Rehm.

R Room rho); Department of Anthropology, Northern Arizona

University, Flagstaff, AZ, USA (R Trotter rho); National Institute for

Alcohol and Drug Research, Oslo, Norway (R Room); and Queen's

University, Kingston, Canada (J Bickenbach)

Correspondence to: Dr T Bedirhan OstOn. Assessment Classification

and Epidemiology Group, WHO, CH 1211 Geneva 27. Switzerland

(e-mail: ustunb@who.ch)

THE LANCET. Vol 354. July 10, 1999 11

ARTICLES

Netherlands, Nigeria, Romania, Spain, Tunisia, Turkey,and the UK). Thus, all WHO regions were represented inthe study.

Key informants were defined as those who, by virtue of theirposition and knowledge, have an understanding of disabilitiesthat makes them representative spokespersons for their culture.For each country, 15 informants were selected-threeindividuals from each of five groups: medical professionals(eg, physician, psychiatrist, psychologist, nurse); allied healthprof~ssionals (eg, social worker, case worker); policy makers oropinion leaders in the area of disability services; individuals witha disabling physical health condition, and their caregivers;and individuals with a disabling health condition related toalcohol, drugs, or mental health, and their caregivers. The finalnumber of participants included in this study varied slightly bycountry, but essentially, the quota of 15 was achieved in eachcountry.

makes it easier to form the often complex comparisonsbetween the impacts of different disease states requiredby the person trade-off protocol.

Professional health-care providers were assumed to berepresentative of society's preferences with regard toresource allocations in health care. This ass~ption,however, requires empirical support, and for empiricaltesting, preference measures should also be obtainedfrom various other groups such as policy makers andindividuals with disabilities to see how these measuresconverge.

Furthermore, these disability weights are presumed tobe universal, that is, equal across countries and cultures.Yet they were in fact derived from a single person trade-off exercise carried out at the WHO in Geneva. Althoughthe results of other exercises since then suggest similarresults,' there is a clear need for more systematic testingacross different cultures and different informant groupswith other forms of measurement. This study wasmotivated by these concerns about the universality of thedisability weights used in the construction ofDALYs.

In a large study on the cross-cultural applicability of aproposed revision of the International Classification ofImpilinnents, Disabilities and Handicaps",.!2 a sub-studyexamined whether expert ratings on the disability effectsof different health conditions were universal, in the senseof being stable across methods, cultures, and informantgroups. Ranking was the chosen method rather than theperson trade-off method because ranking requires lessspecialised participants, less time, and no technicalknowledge. The original Global Burden of Disease studyprotocol also used an ordinal ranking exercise in additionto a variant of the person trade-off protocol, and requiredrespondents to reconcile the discrepancies between thetwo during a deliberative phase.' Our study attempted toreplicate independently the results of the exercise carriedout within the framework of the Global Burden ofDisease study in different cultures with differentinformants. We should point out, however, that becauseof the difference in methods, our study can only test theassumptions of the Global Burden of Disease study (eg,stability of disablement scores across cultures andinformant groups). With the ranking exercise, disabilityweights cannot be derived. The choice of method isjustified since the main ai~ of our study is to test theunderlying assumptions of the Global Burden of Diseasestudy. Ranking exercise also requires much less time(10-15 min instead of 2 days per group for the persontrade-off method).

We aimed to answer four questions. Are theresignificant differences in the ranking of the disablinghealth conditions by key informants from differentcountries? Are there significant differences in the rankingof the disabling health conditions by respondents fromdifferent informant groups (medical professionals, alliedhealth professionals, health-policy makers, consumers, orcaregivers)? Could the ranking of the disability weights ofthe Global BUfden of Disease study be replicated with adifferent method? What are the underlying patterns inrespondent ratings of disabling effects of healthconditions?

DesignInfonnants responded to a questionnaire including both open-ended items and closed-ended items. In rank-orderingdisablements, key infonnants were presented with a deck of 17cards listing different health conditions with a short description.They were instrUcted to rank conditions from the most disablingto the least disabling. The most disabling condition wasdescribed as one that would make activities such as dressing,feeding, moving around, and meeting basic day-to-dayresponsibilities very difficult. The least disabling conditon wasdescribed as one that would not interfere with the above-mentioned activities in day-to-day life. Codes were assigned tothe 17 conditions, with 01 representing the most disabling, and17 representing the least disabling. 13 of the 17 healthconditions were adapted from the 22 indicator conditions of theGlobal Burden of Disease study to measure the burden ofdisabilities and diseases within the DALY framework. Theselection of the original indicator conditions was based onresults from pretests, which showed that respondents in certaincultures found some diseases difficult to grasp. Ranking of22 conditions has also been shown to be too much of a burdenon the respondent. I' In addition to the original conditions,

we included disorders related to alcohol misuse and drugmisuse, and HIV infection, which we assumed would rank asmedium disabling health conditions and would show culruraldifferences. I'

The ranking exercise was carried out through an interview thatlasted, on average, 90 min. The interviewer showed theinfonnant a collection of demographic data, and asked questionsabout the language of disablement in the culrure, about existingcompensation systems for disablement, about the social stigmaof disabilities, and about other societal reactions to disablinghealth conditions. This process served to frame the issue of theranking exercise within the context of understanding thedisabling consequences of health conditions. The assessment ofsocial disapproval or stigma for different health and socialconditions was quantified on an II-point rating scale with theendpoints labelled as none (0) and extreme (10).

Each key infonnant was presented with a brief overview ofthe interview and its purpose. The interviewer then askedquestions about demographic variables of the infonnant andabout their experience in the area of disabilities. The intervieweralso asked open-ended questions on language and disablementand identification of and societal reactions to disablementassociated with selected conditions (eg, difficulties with walking,psychotic symptoms, low intelligence, and alcohol-related anddrug-related problems). Once the open-ended portion of theinterview was completed, the interviewer presented the keyinfonnant with the 17 different cards, and administered theinstrUctions to rank the conditions in order according to theirdisabling effects. Finally, the key infonnant was asked tocomplete a self-administered questionnaire that addressedexpected difficulties with day-to-day activities in a selection ofhealth conditions and included the assessment of stigma relatedto various disabilities.

Methods

ParticipantsInformants from 14 countries took part in this study (Canada,China, Egypt, Greece, India, Japan, Luxembourg, the

112 THE lANCET. Vo1354. July 10, 1999

ARTICLES

~

Health condition

QuadriplegiaDementia

Actwe psychosis

ParaplegiaBlindness

Major depression

Drug dependenceHIV infection

Alcohoiism

Tot~1 deafness

Mild mental retardation

Incontinence

Amputation beiow the knee

Rheumatoid arthritis

Severe migraines

InfertilityVitiligo on face

Rank

123456789

10U121314151617

241241241241241241222239241241241239241241240238238

n=number of informants.

Table 1: Rank order of disabling effect of health conditions byseverity

Statistical analysesNon-parametric statistics for ordinal variables were used toanalyse the data. Overall ranking was established on the basis ofmedian of the ranks. For conditions with the same median, thearithmetic mean of rankings was taken as a second criterion.

To test for differences between countries or betweeninformant groups, the Kruskal-Wallis rank-order ANOVA forone factor was used. Kendall "t-B coITelations were used tomeasure the association betWeen different rank orders. Thismeasure was chosen instead of Speannan's rank coITelationcoefficient because it has a clear interpretation of choice:Kendall "t is equivalent to the number of times rankings agreeabout a pair, minus the number of times rankings disagree,divided by the total number of pairs. Thus, "t can beinterpreted as the coITesponding difference betWeenprobabilities of agreement and disagreement with respect tothe ranking of each pair." Statistical analyses were calculatedwith SYSTAT 8.0 and with StatXact 3.1 (providing exactcalculations of the Kendall "t coITelation and the associatedCIs).

ResultsInterview data were collated, and the 17 healthconditions were ranked from most disabling to leastdisabling (table 1). Overall, the ranking of the conditionsat both ends of the range showed less variability betWeeninformants than that for intermediate conditions.

There were, however,' deviations from this o.rder acrosscountries (table 2). Active psychosis was ranked mostdisabling in the Netherlands and Canada but thirdoverall. HIV infection was given a low ranking by

informants from Japan, Luxembourg, Spain, Turkey, andthe UK, whereas it was ranked first in Egypt and Tunisia.HIV infection was the health condition with the mostvariation in ranking.

The differences between countries were significant(p<0'05) for 13 of the 17 health conditions on theKruskal- Wallis test. Only quadriplegia, paraplegia,amputation below the knee, and mild mental retardationdid not show differences in ranking between countries atp<0'05. Three of the four conditions that were judgeduniformly across countries were physical disabilities. Thefourth, mild mental retarda:tion, also showed littlevariation though this condition was less uniformly ratedthan the other three.

Although there were significant differences in theranking between countries, there is evidence ofconvergence in judgments. The average of the Kendall 'trank correlations between different countries was 0.61(95% CI 0'59-0'64). Within this average, there were,however, clear cultural differences for somecomparisons: for example, 't for the Netherlands andTurkey was 0.41 (0'14-0.69), whereas that for theNetherlands and Tunisia was 0.29 (0.03-0'56). ForLuxembourg and Spain 't was 0.87 (0-74-0.99), andfor Romania and India 0.82 (0.72-0-92). Most of rankcorrelations ranged between 0.50 and 0'7G (for't=0'50, typical 95% CI 0'27-0.73; for 't=0.60, typical95% CI 0.35-0.85). A table with all bivariatecorrelations between countries can be obtained fromthe corresponding author. Overall, these correlationsdid not suppon a general notion of universal disabilityweights.

Only five of 17 health conditions had significantlydifferent rank orders between different informantgroups: quadriplegia, HIV infection, total deafness,mild mental retardation, and amputation below theknee. Again, most of these were physical conditions,but in this analysis they showed most variation,whereas in the analysis by country they showed leastvariation.

The rank orders showed significant differencesbetween some informant groups that might have beenexpected to have convergent views (table 3). Forexample, the ratings between health professionalsworking in physical rehabilitation showed the largestdifferences from consumers and caregivers in the samesector with respect to rankings in the above

NA=data not collected. -.See table 1. tSignificantdifferences between countries at p<O:-O5{Kfuskal-Wallis ran~er ANOVA).

Table 2: Disability ranks associated with different health conditions by country

113THE lANCET. Vol 354. July 10,1999

ARTICLES

Allied health Healtil-pollcyprofessionals makars

(n=51) (n=35)

Consumers end carers

Physical Mental

disability disability(n=30) (n=45)

Health condition Overal rank Medical professionals treating:

Physicalillness

(n=14)

ADMdisorders(n=35)

~hysicaland ADM

disorders

(n=14)

Other

(n=ll)

All

(n=74)

123456789

10U121314151617

1253649

10131214

78

11151617

1234567

1110

81312

914151617

1243568

1197

12131015141617

1234567

1198

13121014151617

1325647

128

109

111314151617

12347596

1011

8121314151716

142658739

1110121315141716

123456879

1213111014151617

3124567

1510

98

141211131716

QuadriplegiaDementia

Active psychosis

ParaplegiaBlindness

Major depression

Drug dependenceHIV infection

Aicohoiism

Total deafness

Mild mental retardation

Incontinence

Amputation below tile knee

Rheumatoid arthritis

Severe migraines

InfertilityVitiligo on face

ADM=Aleohol, drug, and mental disorders.

Table 3: Disablement ranks of different health conditions by informant group

migraine omitted, 't was 0'97 (0-89-1-00; Z=4-39;p<O-OI)-

conditions, and physical condition.s in general.Overall, the rank orders for different informant groupshad an average correlation of 0-76.

The research design allowed us to measure theinfluence of social disapproval or stigma on disabilityrankings for a subset of the conditions. Higher degreesof stigma were associated with higher disabilityrankings; howev~r, the effect size was small «0-10 onaverage, for 't; data not shown). We conclude thatstigma, as measured on an II-point rating scale, has ameasurable but very small effect on disability rankings.

The resulting summary ranking across all judgmentswas very similar to the ranking by expens in theGlobal Burden of Disease study (table 4), derived by adifferent method_17 The rank order correlation ('t) was0-77 (0-40-1-00; Z=3-66, p<O-OI)- The only notabledifference was for severe migraine, which ranked thirdin the Global Burden of Disease study, and 11th inthis study. However, the way judgments were elicitedfor migraine in the two exercises differed. In theGlobal Burden of Disease study, expens wererepeatedly reminded that they should consider theartificial case of 1 continuous year of severe migraines,with the cons~guence of staying in bed most of thetime, when making their judgments. We did notrepeatedly remind our informants about this artificialcase; thus, the comparability of the condition sevetemigraine may have been compromised. With severe

DiscussionAlthough there was some yariation, the rankings of thedisabling effect of health conditions are stable acrosscountries and informant groups, irrespective of themethod used for the ranking, as shown by the highcorrelations betWeen the Global Burden of Disease studyand this study, and the close agreement among the 14countries and among the eight informant groups in ourstudy. Thus, the responses from the key informants showthat the relative burden of different health conditions, interms of disability, is fairly similar across the world. Theresults also, however, indicate that there are systematicand, in some cases, pronounced differences betWeencultures and informant groups. These differences 'arelarge enough to shed doubt on the assumption ofuniversality of the disability rankings, and subsequentlyon the weights, and they are large enough to beinvestigated further in a systematic way.

In theory, there is no reason why disability weightsshould be universal. First, the actual burden of disabilityor activity limitations is affected by m~y factors; forexample, the burden of disability for individuals withparaple~a is likely to depend on the assistance availablein the form of devices (wheelchairs, adapted cars, adaptedworkplaces) and social support. There are cleardifferences betWeen countries in the availability ofassistance and these differences should be reflected in thedisability weights attached to certain health conditions.Second, valuation and description in the terminology ofthe Global Burden of Disease study may not be stronglyrelated. Ranking exercises for this study were directedexclusively towards the extent of disabling conditions. Alldiseases had to be ranked solely qn the basis of theirdisabling effects. However, the person trade-off methodasks for a valuation (ie, trade-off of people with thatdisease). The basic comparisons are obtained throughquestions about the relative weights attached to the livesof 1000 healthy people for 1 year or the deaths of2000 blind people. Thus, issues other than the disablingeffects of health conditions, such as prognosis, pain,mood impact, and public opinon may drive thevaluations.

Median

disabilityweightsGBD

Health condition Rank order Rank order AverageGBD study' from this disability

study weightsGBD

123456789

10111213

0.8950.7620-7220.6710.6240.6190-3330.3610.2810-2090.7380-1910.020

0.8940.7700.7140.6720.6420.633Q.3330.4100.2720.2000.7530.1770.005

QuadriplegiaDementia

Active psychosis

ParaplegiaBlindness

Major depressionTotal deafness

Mild mental retardation

Below the knee amputation

Rheumatoid arthritis

Severe migraines

infertiiityVitiligo on face

12456798

1011

31213

'Taken from table 1.3 (page 39) of the Global Burden of Disease Study (GBD).'

Table 4: A comparison of the rank order between GlobalBurden of Disease study" and this study

THE lANCET. Vo1354. July 10, 1999114

ARllCLES

The WHO/NIH Joint Project CAR study advisers and principalinvestigatorsR Battjes (NIDA, USA), B Grant (NIAAA, USA), C Kennedy (NIMH,USA), D Regier (NIMH, USA), S Yu Cun (China), V Mavreas (Greece),R S Murthy (Bangalore, India), H Pal (Delhi, India), R Thara (Chennai,India), M Tazaki Gapan), C Pull (Luxembourg), H Hoek (Netherlands),A Odejide (Nigeria), R Vrasti (Romania), J L Vazquez-Barquero (Spain),A Chaker (Tunisia), A Gogus (Ankara, Turkey), N Dedeoglu (Antalya,Turkey), KOgel (Istanbul, Turkey), D Mumford (UK).

AcknowledgmentsThis study was made possible within the WHO-NIH Joint Project(UOIMH35883).

The condition ranked most variably by country in oursample was mv infection; the differences between expertgroups are less dramatic but still important. Awarenessand availability of information on and treatment for mvinfection in different countries vary greatly; also theinfection carries a stigma since it is generally transmirtedsexually. Physical conditions were ranked more uniformlythan were mental conditions across countries, but notacross informant groups. This finding implies that,though physical conditions such as quadriplegia,paraplegia, and amputation are, as a group, vieweddifferently from conditions such as active psychosis, thesedifferences are more likely to be a feature of theindividual respondent's characteristics than culturaldifferences. FUrthermore, the conditions with the mostvariation are mental conditions that lie in the mid-rangeof the rankings.

Any ordinal ranking can only indicate the relativeeffects of disability from different health conditions(eg, active psychosis causes more disability than majordepression), and not the absolute effects, which may varygreatly between countries because of different formal andinformal health and social-support. systems. Also, closeagreement between health professionals or other keyinformants does not necessarily indicate that the realdisability associated with the different health conditions issimilarly uniform across countries. In different instancesin the past, the consensus among experts has been atvariance with the actual behaviour ofpatients.'6,'7 Thus, thenext step should be empirical studies to examine whetherselected health conditions really do have the same disablingeffects in different countries and cultures across the world,and whether they are stable with respect to differentassessment methods and different informant groups.

Finally, an actual measurement of functioning ofpatients with different health conditions in theirrespective contexts, ought to be coupled with valuationexercises across cultures and respondent groups.'. Thismeasurement may then provide valuable insights into thedeterminants of values that people assign to disablinghealth conditions. If public policy decisions, such asresource allocation and priority setting, are to be basedon evidence, we need to make sure that they are formedby all parties and are based on real-life data obtained inaccordance with high scientific standards.ContributorsT Bedirhan Ustiin was die principal investigator and contributed to allparts of die study. J Rehtn codesigned die study, was responsible for diestatistical analyses, and contributed to die writing of die paper. S Chatterjicodesigned die study, and contributed to the interpretation of results.S Saxena codesigned die study, did part of the fieldwork, and contributedto the writing of die paper. R Trotter codesigned the study,conceptualised the qualitative parts, and analysed diem. R Room tookpan in die design and die analysis of the study and panicipated in therevision. J Bickenbach took pan in the interpretation and writing of theresults, and in die revision of the anicle.

-.References

I Murray CJL, Lopez AD. The Global Burden of Disease: acomprehensive assessment of mortality and disability from diseases,injuries, and risk factors in 1990 and projected to 2020. Cambridge,MA: HatVard University Press, 1996.

2 Murray CJL, Lopez AD. Alternative projections of mortality anddisability by cause 1990-2020. Global Burden of Disease Study.Lancer 1997; 349: 1498-504.

3 MUttay CJL, Lopez AD. Regional patterns of disability-free lifeexpectancy an<! disability-adjusted life expectancy: Global Burden ofDisease Study. Lancer 1997; 349: 1347-352.

4 Ad Hoc Committee on Health Research Relatin~ to FutureIntetVention Options: investing in health research and development.Geneva: WHO, 1996.

5 Field MJ, Gold MR. Summarising population health:directions for the development and application of population metrics.Institute of Medicine, Washington DC: National Academy Press,1998.

6 Torrence GW. Measurement of health state utilities for economicappraisal: a review.] Healrh Econ 1986; 5: 1-30.

7 Froberg DG, Kane RL. Methodology for measuring health-statepreferences-I: measurement strategies.] Clin Epidemio11989; 42:345-54.

8 Mehraz A, Gafni A. Preference-based outcome measures for economicevaluation of drug interventions: quality adjusted life years (QALYs)versus healthy years equivalents (HYEs). Pharmacoeconomics 1992; I:338-45.

9 Hyder AA, Rottlant G, Morrow RH. Measuring the burden of disease:healthy life years. Am] Public Health 1998; 88: 196--202.

10 Nord E. The person-trade off approach to valuing health careprograms. Med Dec.:. Making 1995; 15: 201-08.

II WHO. International classification ofimpairtnents, disabilities andhandicaps. Geneva: WHO, 1980.

12 WHO. International classification ofimpainnents, activities andparticipation (ICIDH-2) Beta-l draft. Geneva: WHO, 1997.

13 Trotter RT, Ustiin TB, Chattelji S, et al. Cross-cultural applicabilityresearch on disablement: models, methods and contributions torevision of the international classification. Hum Org (in press).

14 Makela K, Room R, Single E, et al. Alcohol, society and the state-I,a comparative study of alcohol control. Toronto: Addiction ResearchFoundation, 1981.

15 Hays WL. Statistics for the social sciences. New York: Holt, Rinehart& Winston, 1973.

16 Rehm J, Gadenne V. Intuitive predictions and professional forecasts:cognitive processes and social consequences. Oxford: Pergamon Press,1990.

17 Single E. The concept of hartn reduction and its application toalcohol: the sixth annual Dorothy Black lecture. Drugs: Ed, Prev Pol1997: 4: 7-22.

18 Ustiin TB, Chattelji S. Editorial: measuring functioning anddisability-a common framework. Inr] Merh Psychiatr Res 1998; 7:79-83.

15THE LANCET. Vo1354. July 10, 1999

The ~ncet isa weekly subscription journal. For further information on how to subscribe please contact our Subscription DepartmentTel: +44 (0)1714364981 Fax +44 (0)171 5808175

North America Tel: +1 212633 3800 Fax: +1 2126333850

Printed in UK

by Robquest Ltd@ 1999 The Lancet42 Bedford Square London WC1B 3SL UK