Lancet 2010; 376: 1558–65
Published Online
November 1, 2010
DOI:10.1016/S0140-
6736(10)61462-6 See Comment page 1524 Neuropsychopharmacology Unit, Imperial College, London, UK (Prof D J Nutt FMedSci); UK Expert Advir to the European Monitoring Centre for Drugs and Drug Addiction (EMCDDA), Lisbon, Portugal (L A King PhD);
and Department of Management, London School of Economics and Political
Science, London, UK
(L D Phillips PhD)
Correspondence to:
Prof David J Nutt, Neuropsychopharmacology Unit, Imperial College London,
Burlington-Danes Building, Hammersmith Hospital, Du Cane Road, London W12 0NN, UK
d.nutt@imperial.ac.uk Drug harms in the UK: a multicriteria decision analysis
David J Nutt, Leslie A King, Lawrence D Phillips, on behalf of the Independent Scientifi c Committee on Drugs
Summary
Background Proper asssment of the harms caud by the misu of drugs can inform policy makers in health, policing, and social care. We aimed to apply multicriteria decision analysis (MCDA) modelling to a range of drug harms in the UK.
Method Members of the Independent Scientifi c Committee on Drugs, including two invited specialists, met in a 1-day interactive workshop to score 20 drugs on 16 criteria: nine related to the harms that a drug produces in the individual and ven to the harms to others. Drugs were scored out of 100 points, and the criteria were weighted to indicate their relative importance.
Findings MCDA modelling showed that heroin, crack cocaine, and metamfetamine were the most harmful drugs to individuals (part scores 34, 37, and 32, respectively), whereas alcohol, heroin, and
crack cocaine were the most harmful to others (46, 21, a nd 17, respectively). Overa ll, a lcohol wa s the most ha rmful drug (overa ll ha rm score 72), with heroin (55) and crack cocaine (54) in cond and third places.
Interpretation The fi ndings lend support to previous work asssing drug harms, and show how the improved scoring and weighting approach of MCDA increas the diff erentiation between the most and least harmful drugs. However, the fi ndings correlate poorly with prent UK drug classifi cation, which is not bad simply on considerations of harm. Funding Centre for Crime and Justice Studies (UK).
Introduction
Drugs including alcohol and tobacco products are a major
cau of harms to individuals and society. For this reason,
some drugs are scheduled under the United Nations 1961
Single Convention on Narcotic Drugs and the 1971
Convention on Psychotropic Substances. The controls
are reprented in UK domestic legislation by the 1971
Misu of Drugs Act (as amended). Other drugs, notably
alcohol and tobacco, are regulated by taxation, sales, and
restrictions on the age of purcha. Newly available drugs
such as mephedrone (4-methylmethcathinone) have
recently been made illegal in the UK on the basis of
concerns about their harms, and the law on other drugs,
particularly cannabis, has been toughened becau of
similar concerns.
To provide better guidance to policy makers in health,
policing, and social care, the harms that drugs cau
need to be properly assd. This task is not easy becau
of the wide range of ways in which drugs can cau harm.
An attempt to do this asssment engaged experts to
score each drug according to nine criteria of harm,
ranging from the intrinsic harms of the drugs to social
and health-care costs.1 This analysis provoked major
interest and public debate, although it raid concerns
about the choice of the nine criteria and the abnce of
any diff erential weighting of them.2
To rectify the drawbacks we undertook a review of
drug harms with the multicriteria decision analysis
(MCDA) approach.3 This technology has been ud
successfully to lend support to decision makers facing
complex issues characterid by many, confl icting
objectives—eg, appraisal of policies for disposal of
nuclear waste.4 In June, 2010, we developed the
multicriteria model during a decision conference,5 which
is a facilitated workshop attended by key players, experts,
and specialists who work together to create the model
and provide the data and judgment inputs.
Methods
Study design
The analysis was undertaken in a two-stage process. The
choice of harm criteria was made during a special
meeting in 2009 of the UK Advisory Council on the
Misu of Drugs (ACMD), which was convened for this
purpo. At this meeting, from fi rst principles and with
the MCDA approach, members identified 16 harm
criteria (fi gure 1). Nine relate to the harms that a drug
produces in the individual and ven to the harms to
others both in the UK and overas. The harms are
clustered into five subgroups reprenting physical,
psychological, and social harms. The extent of individual
harm is shown by the criteria listed as to urs, whereas
most criteria listed as to others take account indirectly of
the numbers of urs. An ACMD report explains the
process of developing this model.6
In June, 2010, a meeting under the auspices of the
Independent Scientifi c Committee on Drugs (ISCD)—a
new organisation of drug experts independent of
government interference—was convened to develop the
MCDA model and asss scores for 20 reprentative
drugs that are relevant to the UK and which span the
range of potential harms and extent of u. The expert
group was formed from the ISCD expert committee
plus two external experts with specialist knowledge of
For more on the Independent Scientifi c Committee on Drugs e: www.drugscience.
org.uk
legal highs (webappendix). Their experience was extensive, spanning both personal and social aspects of drug harm, and many had substantial rearch experti in addiction. All provided independent advice and no conflicts of interest were declared. The meeting’s facilitator was an independent specialist in decision analysis modelling. He applied methods and techniques that enable groups to work effectively as a team, enhancing their capability to perform,7
improving the accuracy of individual judgments. The group scored each drug on each harm criterion in an open discussion and then assd the relative importance of the criteria within each cluster and across clusters. They also reviewed the criteria and the definitions developed by the ACMD. Thi
s method resulted in a common unit of harm across all the criteria, from which a new analysis of relative drugs harms was achieved. Very slight revisions of the defi nitions were adopted, and panel 1 shows the fi nal version.
Scoring of the drugs on the criteria
Drugs were scored with points out of 100, with 100 assigned to the most harmful drug on a specifi c criterion. Zero indicated no harm. Weighting sub-quently compares the drugs that scored 100 across all the criteria, thereby expressing the judgment that some drugs scoring 100 are more harmful than others.
In scaling of the drugs, care is needed to ensure that each successive point on the scale reprents equal
increments of harm. Thus, if a drug is scored at 50, then it should be half as harmful as the drug that scored 100. Becau zero reprents no harm, this scale can be regarded as a ratio scale, which helps with interpretation of weighted averages of veral scales. The group scored the drugs on all the criteria during the decision conference. Consistency checking is an esntial part of proper scoring, since it helps to minimi bias in the scores and encourages realism in scoring. Even more i
mportant is the discussion of the group, since scores are often changed from tho originally suggested as participants share their diff erent experiences and revi their views. Both during scoring and after all drugs have been scored on a criterion, it is important to look at the relativities of the scores to e whether there are any obvious discrepancies.
Weighting of the criteria
Some criteria are more important expressions of harm than are others. More precision is needed, within the context of MCDA, to enable the asssment of weights on the criteria. To ensure that assd weights are meaningful, the concept of swing weighting is applied. The purpo of weighting in MCDA is to ensure that the units of harm on the different preference scales are equivalent, thus enabling weighted scores to be compared and combined across the criteria. Weights are, esntially, scale factors.
MCDA distinguishes between facts and value judgments about the facts. On the one hand, harm express a level of damage. Value, on the other hand,
indicates how much that level of damage matters in a
particular context. Becau context can aff ect asss-
ments of value, one t of criterion weights for a
particular context might not be satisfactory for decision
making in another context. It follows then, that two
stages have to be considered. First, the added harm
going from no harm to the level of harm reprented by
a score of 100 should be considered—ie, a straight-
forward asssment of a diff erence in harm. The next
step is to think about how much that diff erence in harm
matters in a specifi c context. The question pod to the
group in comparing the swing in harm from 0 to 100 on
one scale with the swing from 0 to 100 on another scale
was: “How big is the diff erence in harm and how much
do you care about that diff erence?”
During the decision conference participants assd
weights within each cluster of criteria. The criterion
within a cluster judged to be associated with the largest
swing weight was assigned an arbitrary score of 100.
Then, each swing on the remaining criteria in the
cluster was judged by the group compared with the
100 score, in terms of a ratio. For example, in the
cluster of four criteria under the category physical
harm to urs, the swing weight for drug-related
mortality was judged to be the largest diff erence of the
four, so it was given a weight of 100. The group judged
the next largest swing in harm to be in drug-specifi c
Figure 1: Evaluation criteria organid by harms to urs and harms to others, and clustered under physical, psychological, and social eff ects
See Online for webappendix
mortality, which was 80% as great as for drug-related mortality, so it was given a weight of 80. Thus, the computer multiplied the scores for all the drugs on the drug-related mortality scale by 0·8, with the result that the weighted harm of heroin on this scale became 80 as compared with heroin’s score of 100 on drug-specifi c mortality. Next, the 100-weighted swings in each cluster were compared with each other, with the most harmful drug on the most harmful criterion to urs compared with the most harmful drug on the most harmful criterion to others. The result of asssing the weights was that the units of harm on all scales were equated. A fi nal normalisation prerved the ratios of all weights, but ensured that the weights on the criteria summed to 1·0. The weighting process enabled
harm scores to be combined within any grouping simply by adding their weighted scores.Dodgson and colleagues3 provide further guidance on swing weighting. Scores and weights were input to the Hiview computer program, which calculated the weighted scores, provided displays of the results, and enabled nsitivity analys to be done.
Role of the funding source
The sponsor of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all the data in the study, and had final responsibility for the decision to submit for publication.
Results
Figure 1 shows the 16 identifi ed harm criteria. Figure 2 shows the total harm score for all the drugs and the part-score contributions to the total from the subgroups of harms to urs and harms to others. The most harmful drugs to urs were heroin (part score 34), crack cocaine
Panel 1: Evaluation criteria and their defi nitions
Drug-specifi c mortality
Intrinsic lethality of the drug expresd as ratio of lethal do and standard do (for adults)
Drug-related mortality
The extent to which life is shortened by the u of the drug (excludes drug-specifi c mortality)—eg, road traffi c accidents, lung cancers, HIV, suicide
Drug-specifi c damage
Drug-specifi c damage to physical health—eg, cirrhosis, izures, strokes, cardiomyopathy, stomach ulcers
Drug-related damage
Drug-related damage to physical health, including conquences of, for example, xual unwanted activities and lf-harm, blood-borne virus, emphyma, and damage from cutting agents
Dependence
The extent to which a drug creates a propensity or urge to continue to u despite adver conquences (ICD 10 or DSM IV)
Drug-specifi c impairment of mental functioning
Drug-specifi c impairment of mental functioning—eg, amfetamine-induced psychosis, ketamine intoxication
Drug-related impairment of mental functioning
Drug-related impairment of mental functioning—eg, mood disorders condary to drug-ur’s lifestyle or drug u
Loss of tangibles
Extent of loss of tangible things (eg, income, housing, job, educational achievements, criminal record, imprisonment) Loss of relationships
Extent of loss of relationship with family and friends
Injury
Extent to which the u of a drug increas the chance of injuries to others both directly and indirectl
y—eg, violence (including domestic violence), traffi c accident, fetal harm, drug waste, condary transmission of blood-borne virus
(Continues in next column)(Continued from previous column)
Crime
Extent to which the u of a drug involves or leads to an increa in volume of acquisitive crime (beyond the u-of-drug act) directly or indirectly (at the population level, not the individual level)
Environmental damage
Extent to which the u and production of a drug caus environmental damage locally—eg, toxic waste from amfetamine factories, discarded needles
Family adversities
Extent to which the u of a drug caus family adversities—eg, family breakdown, economic wellbeing, emotional wellbeing, future prospects of children, child neglect International damage
Extent to which the u of a drug in the UK contributes to damage internationally—eg, deforestation, destabilisation of countries, international crime, new markets
Economic cost
Extent to which the u of a drug caus direct costs to the country (eg, health care, police, prisons, social rvices, customs, insurance, crime) and indirect costs (eg, loss of productivity, abnteeism)
Community
Extent to which the u of a drug creates decline in social cohesion and decline in the reputation of the community
ICD 10=International Classifi cation of Dias, tenth revision. DSM IV=Diagnostic and Statistical Manual of Mental Disorders, fourth revision.
For more on Hiview e
uk
(37), and metamfetamine (32), whereas the most harmful to others were alcohol (46), crack cocaine (17), and heroin (21). When the two part-scores were combined, alcohol was the most harmful drug followed by heroin and crack cocaine (fi gure 2).
Another instructive display is to look at the results parately for harm to urs and to others, but in a two-dimensional graph so that the relative contribution to the two types of harm can be en clearly (fi gure 3). The most harmful drug to others was alcohol by a wide margin, whereas the most harmful drug to urs was crack cocaine followed cloly by heroin. Metamfetamine was next most harmful to urs, but it was of little comparative harm to others. All the remaining drugs were less harmful either to urs or to others, or both, than were alcohol, heroin, and crack cocaine (fi gure 3). Becau this display shows the two axes before weighting, a score on one cannot be compared with a score on the other, without knowing their relative scale constants. Figure 4 shows the contributions that the part scores make on each criterion to the total score of each drug. Alcohol, with an overall score of 72, was judged to be most harmful, followed by heroin at 55, then crack cocaine with a score of 54. Only eight drugs scored, overall, 20 points or more. Drug-specifi c mortality was a substantial contributor to five of the drugs (alcohol, heroin, γ hydroxybutyric acid [GH B], methadone, and butane), whereas economic cost contributed heavily to alcohol, heroin, tobacco, and cannabis.Discussion
The results from this MCDA analysis show the harms of a range of drugs in the UK. Our fi ndings lend support to the conclusions of the earlier nine-criteria analysis undertaken by UK experts1 and the output of the Dutch addiction medicine expert group.8 The Pearson cor-relation coeffi cient between Nutt and colleagues’ 2007 study1 and the new analysis prented here for the 15 drugs common to both studies is 0·70. One reason for a less-than-perfect correlation is that the scores from Nutt and colleagues’ previous study were bad on four-point ratings (0=no risk, 1=some risk, 2=moderate risk, and 3=extreme risk). The ISCD scoring process was bad on 0–100 ratio scales, so they contain more information than the ratings do.
Throughout Nutt and colleagues’ 2007 paper, harm and risk are ud interchangeably, but in the ISCD work, risk was not considered becau it is susceptible to varying interpretations. For example, the British Medical Association defi nes risk as the probability that something unpleasant will happen.9 Thus, asssors from Nutt and colleagues’ 2007 work might have interpreted their rating task differently from the scoring task of the ISCD experts. Furthermore, in Nutt and co-workers’ 2007 study, ratings were simply averaged across the nine criteria (called parameters in the report), three each for physical harm, dependence, and social harms, whereas diff erential weights were applied to the criteria in this ISCD study, as is shown in the key of
Figure 2: Drugs ordered by their overall harm scores, showing the parate contributions to the overall scores of harms to urs and harm to others
The weights after normalisation (0–100) are shown in the key (cumulative in the n of the sum of all the normalid weights for all the criteria to urs, 46; and for all the criteria to others, 54). CW=c
umulative weight. GHB=γ hydroxybutyric acid. LSD=lyrgic acid diethylamide.
fi gure 4. Despite the many diff erences between the two studies, there is some degree of linear association between both ts of data.
The correlations between the Dutch addiction medicine expert group2 and ISCD results are higher: 0·80 for individual total scores and 0·84 for population total scores. As with Nutt and colleagues’ 2007 study, the Dutch experts applied four-point rating scales to 19 drugs. However, they ud fi ve criteria: acute toxicity, chronic toxicity, addictive potency, social harm at individual level, and social harm at population level. Simple averages produced two overall mean harm ratings, one each for individuals and for populations. The probable explanation for the greater correlation between the ISCD and Dutch data lies in the greater relative ranges of the overall results than in Nutt and co-workers’ 2007 study. The highest and lowest overall harm scores in the ISCD study are 72 for alcohol and 5 for mushrooms, which is a ratio of about 14:1; whereas in Nutt and colleagues’ study it was a ratio of just over 3:1, from 2·5 for heroin to 0·8 for khat. The highest and lowest scores for the Dutch individual ratings were 2·63 for crack cocaine and 0·40 for mushrooms, which is a ratio of 6·6:1; and for the population ratings 2·41 for crack cocaine and 0·31 for mushrooms, which is a ratio of 7·8:1. The ratio scaling in the ISCD study spanned a wider range, making the three most harmful
drugs—alcohol, heroin, and crack cocaine—much more harmful relative to the other drugs than can be expresd with rating scales, so that additional information stretched the scatterplot in one dimension, making it em more linear. Additionally, becau the Dutch scale attributes only a quarter of the scores to social factors, whereas in the ISCD scoring the factors compri nearly half of the scores (ven of 16 criteria), drugs such as alcohol which have a major eff ect will rank more highly in the ISCD analysis, with tobacco ranked lower becau its harms are mainly personal.
The correlations between the ISCD overall scores and the prent classifi cation of drugs bad on revisions to the UK Misu of Drugs Act (1971) is 0·04, showing that there is effectively no relation. The ISCD scores lend support to the widely accepted view10,11 that alcohol is an extremely harmful drug, both to urs and society; it scored fourth on harms to urs and top for harms to society, making it the most harmful drug overall. Even in terms of toxic eff ects alone, Gable12 has shown that, on the basis of a safety ratio, alcohol is more lethal than many
Figure 3: Drugs shown for their harm to urs and harm to others LSD=lyrgic acid diethylamide. GHB=γ hydroxybutyric acid.
