Journal of Public Health Advance Access originally published online on January 25, 2006
Journal of Public Health 2006 28(1):3-9; doi:10.1093/pubmed/fdi073
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Prevalence of problematic and injecting drug use for Drug Action Team areas in England
Martin Frisher
Martin Frisher, Senior Lecturer, Department of Medicines Management, Keele University, Keele, Staffordshire ST5 5BG, UK
Heath Heatlie
Heath Heatlie, Research Fellow, Department of Medicines Management, Keele University, Keele, Staffordshire ST5 5BG, UK
Mathew Hickman
Mathew Hickman, Senior Lecturer, Department of Social Medicine, University of Bristol, Canynge Hall, Whiteladies Road, Bristol, UK
Mathew Hickman, Senior Lecturer, Centre for Research on Drugs and Health Behaviour, Imperial Medical School, London, UK
Address correspondence to Martin Frisher. Email: m.frisher{at}mema.keele.ac.uk.
Background National and local monitoring of policies on illicit drug use requires information on the number of problematic drug users in a country. This article reports the findings from a study that estimated the number of problematic and injecting drug users for all Drug Action Teams (DATs) in England for 2001.
Methods The Multiple Indicator Method (MIM) is a statistical technique for using aggregated data to estimate numbers of drug users across a large number of areas. The MIM was used to combine eight indicators available for all DATs, with prevalence estimates available from a small number of DATs. The indicators were drug possession and supply offences, arrest referrals, people recorded in drug treatment databases, methadone prescriptions, drug-related hospital episodes, drug-related deaths and DATs’ Townsend score. The latter is a measure of material deprivation. A three-stage process involved, (i) factor analysis of the drug indicators, (ii) regression linking factor scores to known prevalence estimates and (iii) imputation of estimates to all other DATs.
Results Factor analysis yielded two statistically significant factors underlying the drug indicators in 150 DATs in England. The estimated prevalence rate of problematic drug use in the DATs varied from 0.2 to 1.5 per cent of the population. The estimated average number of problematic drug users per DAT was 1943 (standard deviation = 1300). The estimated average number of injecting drug users per DAT was 627 (standard deviation = 572). The estimates for England in 2001 were 287 670 (population rate = 0.64 per cent) problem drug users, and 93 185 (population rate = 0.23 per cent) injecting drug users.
Conclusions Although the model cannot take account of specific local factors, the results are likely to be accurate in areas that do not have these idiosyncrasies. The estimated prevalence figures provide a basis for all DATs to assess their contact rates with problematic and injecting drug users.
Keywords: prevalence, drug dependence, data analysis, statistical, regional health planning, spatial distribution
 |
Introduction |
|---|
 TOP Introduction ResultsDiscussionLimitations of this studyAcknowledgementsReferences |
|---|
Estimates of the prevalence of problem drug use in the UK are a key piece of evidence for monitoring of the national drug strategy.1 Before this study, few Drug Action Teams (DATs) had estimates for their area. Without this information, DATs are unable to assess whether they are meeting the national target for drug users in drug treatment programmes. The target is to increase participation by 55 per cent by 2004 and 100 per cent by 2008, compared to a 1998/99 baseline. Although population surveys conducted in private households have been used to estimate levels of drug use, they are likely to under-represent problematic and injecting drug use for a variety of reasons. People who are problematic drug users may be less likely to (a) live in households that are sampled by population surveys, (b) respond to population surveys or (c) self-report use of drugs that are perceived to be considered to be problematic.2
In 1998, a study commissioned by The European Monitoring Centre for Drugs and Drug Addiction (EMCDDA) concluded that the Multiple Indicator Method (MIM) was a potentially valid method for estimating the prevalence of illicit drug use.3 In contrast to surveying a population about their experience of drug use, the MIM is a form of synthetic estimation, which has been widely used in the United States since the 1970s.4 Synthetic estimation aims to determine the relationship (usually linear) between a drug indicator (e.g. drug-related deaths) and the number of known problem drug users (prevalence). The latter is usually derived from overlaps between individuals appearing on three or more lists of known drug users. When this relationship is established in a number of areas where prevalence is known (anchor points), it can be transferred to other areas where only the indicators are available to impute the number of problem drug users. In synthetic estimation, estimates have been derived using a single estimate and indicator. For example, the number of opiate users in Australia were imputed from information prevalence of opiate use in New South Wales, and the number and proportion of overdose deaths in New South Wales and the rest of Australia.5 The MIM seeks to analyse the relationship between numerous drug indicators before establishing the relationship with known prevalence estimates in several anchor point areas.
The MIM has previously been used to produce prevalence estimates for the UK, France and Italy.6 For the UK study, estimates were obtained for 1996 but only by Regional Health Authority areas and with anchor point estimates that did not have a uniform definition.7 This study is more ambitious in seeking to estimate prevalence for the 149 DATs in England based on uniformly defined anchor points.
Method
The MIM consists of the following steps.
- Obtain a range of indicators of problem drug use for all DAT areas.
- Obtain population estimates for all DAT area.
- Obtain estimates of the number of problematic and injecting drug users in selected areas (the anchor points areas). Prevalence estimates for problematic and injecting drug users in 2001 were available for 23 English DATs, through pilot studies funded by the Home Office and for sites selected on the basis that it was likely that there was a large enough population of problem drug users to estimate, and three or more large data sources (N > 100) were available for the indirect estimation.
- Convert indicators and prevalence estimates into rates per 100 000 population and then standardize (i.e. divide the difference between the value and the mean by the standard deviation) in order that each variable has equal weighting in the analysis.
- Use principal component methods to determine the number of factors underlying the indicator correlation matrix.
- Regress the factor scores on the prevalence estimates using least squares estimation for the anchor areas, and apply the regression equation to the areas without prevalence estimates.
Drug indicators
The primary data source for drug indicators was the UK Home Office that at the time of this study collated DAT data using a standard template. From the template, three indicators were obtained: (i) number of offences involving possession of controlled drugs; (ii) number of offences involving supply of controlled drugs; (iii) number of adults entering drug treatment services. An additional DAT indicator, arrest referrals, (iv) was obtained from the Home Office for the period April 2001–March 2002. Arrest referral is a scheme conducted in police custody suites where a drug worker assesses the drug problems of arrestees and refers them to appropriate drug treatment services.8
Hospital Episode Statistics (HES) data were obtained on all patients admitted to hospital (as either a day case or as an ordinary admission) with a primary diagnosis of substance use (ICD-10: F11–F19) (v). Drug-related deaths (vi) were provided by the Office of National Statistics based on their analysis of death certificates.9 Townsend scores (vii), which is an index of material deprivation, were derived for all DATs.10 The final indicator was the number of methadone prescriptions dispensed from community pharmacies (viii) based on Prescription Pricing Authority (PPA), Prescribing Analysis and Cost (PACT) data from April 2001 to March 2002.
Cross-area estimation
The primary geographical unit for this study is the DAT. However, a problem with using these areas for investigation is that some useful data sources relate to geographical areas that are not coterminous with DAT areas. The use of a Geographical Information System (GIS) can be used to overcome this problem. The technique used is referred to as ‘cross-area estimation’.11 This technique converts one set of area unit data to apply to another set. Cross-area estimation therefore allows data from areas that are non-coterminous areas to DATs to be reaggregated by DATs. In this study, cross-area estimation was only used for the volume of prescribed methadone. These data are available for primary care trust (PCT) areas that are generally smaller than DATs.
Missing data
There were no missing data for four of the eight indicators. For arrest referrals, there were no data for three DATs, for adults in treatment there were no data for 31 DATs, for supply of controlled drugs there were no data for 16 DATs and for possession of controlled drugs there were no data for 14 DATs. Missing data points were replaced by the national average of each indicator for the analysis.
Prevalence estimates in anchor point areas
Prevalence estimates for problematic and injecting drug users in 2001 were available for 23 English DATs.12,13 All estimates were derived using the capture–recapture method (CRM). The definitions of drug use in all of these CRM studies and therefore for the estimates in this study are
- Injecting drug user: Someone who has injected any drug in the previous 4 weeks.
- Problematic drug user: Someone who has injected (any drug) or used an opiate, or used a benzodiazepine or used crack-cocaine. All drugs refer to illicit use within the previous 4 weeks.
- Problematic drug user: Someone who has injected (any drug) or used an opiate, or used a benzodiazepine or used crack-cocaine. All drugs refer to illicit use within the previous 4 weeks.
Factor analysis
Factor analysis aims to identify underlying factors that explain the pattern of correlations within a set of observed variables. Factor analysis is often used in data reduction to identify a small number of factors that explain most of the variance observed in a much larger number of manifest variables. In this study, the principal components method was used to form linear combinations of the observed variables. The first principal component is the combination that accounts for the largest amount of variance in the sample. Successive components explain progressively smaller portions of the total sample variance and all are uncorrelated with each other. To decide how many factors are needed to represent the data, reference is made to the Eigenvalue. (The Eigenvalue is the variance explained by a factor.) One good rule of thumb for determining the number of factors is the ‘Eigenvalue greater than 1’ criteria. Where factors are identified, it is up to the researchers to decide what this factor actually represents. To obtain factor scores for each DAT, missing values were replaced by the variable mean. The analysis was conducted using SPSS.
Regression analysis
All 149 DATs in England were entered into the analysis (City of London and Hackney were combined into a single unit because of the small size of the City DAT). The independent variables were factor 1 and factor 2, and the dependent variables were the standardized prevalence estimates in anchor point areas for (i) problematic drug use and (ii) injecting drug use. The goodness of fit was measured by R2 which is the variance in the anchor points explained by the independent variables (factors 1 and 2).
Confidence intervals
The issue of statistical confidence intervals has hampered drug prevalence estimation studies. Further, the uncertainty expressed in the confidence interval only refers to statistical uncertainty. De Angelis and Hickman have shown that for one form of indirect estimation model, uncertainty can be far greater than statistical uncertainty.14 Based on the available data (i.e. factor scores and anchor point estimates), the regression analysis provides an estimate of how the factor scores relate to prevalence in all areas. Based on the sample estimates, a range of values were calculated, that within a designated likelihood, include the true population value. The dispersal of the anchor point estimates in relation to the MIM estimates result in fairly wide confidence intervals, particularly for injecting drug use. For the current analysis, 50 and 90 per cent confidence intervals are provided. Although these intervals are wide, it is also important to recognize that before this study the vast majority of DATs had no estimate of problematic or injecting drug use. The figures provided here therefore represent a considerable advance on prior knowledge, and at this stage, it may be prudent to use the mean estimates for each DAT.
Maps of problematic and injecting drug use
Each spatial unit is filled with a shade that represents the data for that area. The maps shown here have used a graduated shade, the lighter colours represent lower values, whereas the darker shades represent higher values. Maps have been produced to show both absolute numbers and the prevalence. The maps here have used the equal interval method, which given that drug misuse prevalence is negatively skewed, emphasize the smaller number of DATs with high prevalence data compared to the low prevalence areas.
|
Results |
|---|
|
TOPIntroductionResultsDiscussionLimitations of this studyAcknowledgementsReferences |
|---|
As anticipated, many of indicators were significantly associated (p < .05) with each other. Table 1 summarizes, for example, that drug possession offences were significantly associated with supply offences, arrest referrals, people receiving drug treatment, DAT Townsend score and drug-related deaths. From the pattern of correlations summarized in Table 1, factor analysis indicated that there were two factors associated with the drug indicators (both with Eigenvalues >1). Factor 1 (accounting for 38 per cent of the variance) was related to criminal justice indicators as well as Townsend score and drug treatment. The second factor (accounting for 14 per cent of the variance) related to hospital episodes, methadone and deaths.
|
Using factors 1 and 2 as predictor variables, the regression model explained 32 per cent of the variance for the problematic anchor points and 70 per cent for the injecting anchor points. While Fig. 1 shows that there is a reasonable association between the factors and the anchor point estimates in most areas, there were also a number of outliers. London (point 1) and Manchester (point 5) have much higher observed prevalence (derived from the capture–recapture studies) than would be anticipated from their factor scores, whereas Liverpool (point 4) and Trafford (point 11) have much lower observed prevalence than would be anticipated from their factor scores.
|
These data mean that MIM estimates may vary from the anchor points where factor scores are lower or higher than might be anticipated from the known prevalence data.
In view of the large number of DATs, MIM estimates, factor and indicator scores were divided into ten deciles. Those in decile 1 have scores in the lowest 10 per cent, those in decile 10 have estimates in the highest 10 per cent. Table 2 shows that although each DAT is assigned to a particular decile for problematic drug use, they may vary considerably in terms of individual indicators. In decile 1, for example, Bracknell Forest has one of the lowest rates of problematic drug use in England. This is reflected in the low level of hospital episodes (decile 1) and methadone prescribing (decile 2), but in this area there were a relatively high level of drug supply offences (decile 6) and drug possession offences (decile 8). Numbers of problematic and injecting drug users are displayed in the maps shown in Figs 2 and 3. Equivalent graphs for prevalence rates are available on request from the authors.
|
|
|
The average number of estimated problematic drug users per DAT is 1943 (standard deviation = 1300), therefore, 68 per cent of DATs have between 630 and 3230 problematic drug users. The average number of estimated injecting drug users per DAT is 627 (standard deviation = 572), therefore, 68 per cent of DATs have between 55 and 1199 injecting drug users.
The estimated number of problem drug users for England in 2001 was 287 670 (0.6 per cent of the total population (TP); 1.1 per cent aged 15–54). The 90 and 50 per cent confidence intervals for problem drug use are 174 117–401 224 (0.4–0.9 per cent TP; 0.7–1.6 per cent aged 15–54) and 244 140–331 201 (0.5–0.7 per cent TP; 1–1.3 per cent aged 15–54).
The estimated number of injecting drug users for England in 2001 was 93 185 (0.2 per cent of the TP; 0.4 per cent aged 15–54). The 90 and 50 per cent confidence intervals for injecting drug use are 21 372–164 333 (0.05–0.35 per cent TP; 0.08–0.64 per cent aged 15–54) and 65 253–120 452 (0.14–0.26 per cent TP; 0.25–0.47 per cent aged 15–54).
|
Discussion |
|---|
|
TOPIntroductionResultsDiscussionLimitations of this studyAcknowledgementsReferences |
|---|
Main finding of this study
Factor analyses indicated that two factors were associated with the eight drug indicators that were used in the MIM estimation procedure. The regression model explained 32 per cent of the variance for the problematic anchor points and 70 per cent for the injecting anchor points. Based on the relationships identified in the anchor point areas, the MIM indicates that the number of problematic/injecting drug users for England in 2001 was 287 670/93 185. The average number of estimated problematic/injecting drug users per DAT is 1943/627.
What is already known on this topic
Before this study, few DATs had estimates for their area. In 1996, the MIM was used in England, but estimates were only obtained for Regional Health Authority areas and with anchor point estimates that did not have a uniform definition.7
|
Limitations of this study |
|---|
|
TOPIntroductionResultsDiscussionLimitations of this studyAcknowledgementsReferences |
|---|
The anchor points are themselves estimates each with their own uncertainty. If there are errors in the anchor points, these will be reflected in the MIM estimates. The key assumption of the MIM is that the relationship between prevalence and the predictors in the calibration sample is transferable to all other areas. In this study, the association between the indicators and number of problem drug users is limited as all the prevalence estimates came from urban areas. We recognize that in some areas the relationship between drug indicators and prevalence may not fit the general pattern, and these estimates need to be interpreted cautiously and ideally corroborated with local data.
Another limitation of this study is that the types of drug use estimated were not drug specific. Future studies may be able to take advantage of developments in the recording of criminal justice and treatment contacts with drug users. As these contacts now often record the specific drug involved, it should be possible to develop drug-specific estimates using the MIM. This would, however, require some local estimates, using capture–recapture or other methods, of these types of drug use.
What this study adds
This study drew on an extensive data set to estimate the prevalence of problematic and injecting drug use for all DATs in England. The study utilized anchor point estimates from 13 areas and drug indicator data from a wide variety of sources. The regression models based on two factors, derived from factor analysis, accounted for 32 and 70 per cent of the variance of problematic and injecting drug use estimates, respectively. The fact that the injecting estimates have a higher correlation with the indicators may reflect the fact that many indicators more closely reflect more severe forms of drug use. An advantage of using the factor scores is that they eliminate some of the random variability in the indicators.
The MIM can be used to update reports using existing surveillance systems without the need for extensive new data collection to estimate prevalence on each occasion, at least over a 1- or 2-year period. To improve the precision of the MIM an expanded number of local prevalence estimation estimates for anchor points from a wider range of rural and urban geographical areas would be desirable. In relation to England, the DAT level estimates afford an opportunity for policy makers to assess the degree to which drug users are engaged with treatment services and the criminal justice system.
|
Acknowledgements |
|---|
|
TOPIntroductionResultsDiscussionLimitations of this studyAcknowledgementsReferences |
|---|
We wish to thank the following: Joanne Condon, Chris Goulden, Arun Sondhi and Nicola Singleton for their help in gaining access to data and comments on the report. Tim Miller for providing the capture–recapture estimates for Manchester. The Prescription Pricing Authority for providing the methadone prescription data. The Department of Health for the Hospital Episode Statistics (HES) data. Matthew Hickman is supported by an NHS Career Scientist Fellowship. The study was funded by Home Office Drugs Research Programme. The views expressed in this article are those of the authors not necessarily those of the funders.
|
References |
|---|
|
TOPIntroductionResultsDiscussionLimitations of this studyAcknowledgementsReferences |
|---|
- Hellawell K. First annual report and national plan. London: Cabinet Office, 2000.
- Harrison L, Hughes A, eds. The validity of self-reported drug use: improving the accuracy of survey estimates. In NIDA Research Monograph. Maryland, USA: National Institute on Drug Abuse, 1997: No. 167.
- European Monitoring Centre for Drugs and Drug Addiction. Study to obtain comparable national estimates of problem drug use prevalence for all EU member states. Lisbon: EMCDDA, 1999.
- Rhodes W. Synthetic estimation applied to the prevalence of drug use. J Drug Issues 1993; 23: 297–321.
- Hall W, Ross J, Lynskey M, Law M, Degenhardt L. How many opioid users are there in Australia? Med J Aust 2000; 173: 528–531.[ISI][Medline]
- Kraus L, Augustin R, Frisher M et al. Estimating prevalence of problem drug use at national level in countries of the European Union and Norway. Addiction 2003; 98: 471–485.[CrossRef][ISI][Medline]
- Frisher M, Hickman H, Kraus L, Mariani F, Wiessing L. A comparison of different methods for estimating the prevalence of problematic drug misuse in Great Britain. Addiction 2001; 96: 1465–1476.[CrossRef][ISI][Medline]
- Oerton J, Hunter G, Hickman M et al. Arrest referral in London police stations: characteristics of the first year. A key point of intervention for drug users? Drugs Educ Prev Policy 2003; 10: 73–85.[CrossRef]
- Baker A. ONS drug-related deaths database: first results for England and Wales, 1993–97. Health Statist Q 2000; 5: 57–60.
- Phillimore P, Beattie A, Townsend P. Widening inequality of health in northern England, 1981–91. Br Med J 1994; 308: 1125–1128.
[Abstract/Free Full Text] - Flowerdew R, Lovett S. A review of the problems of transferring data from one set of areal units to another incomparable set. University of Lancaster, Lancaster, UK: Northern Regional Research Laboratory, 1989: Research report 4.
- Hickman M, Higgins V, Hope V et al. Injecting drug use in Brighton, Liverpool, and London: best estimates of prevalence and coverage of public health indicators. JECH 2004: 58: 766–771.
- Millar T, Gemmell I, Hay G, Donmall M. (2002) The dynamics of drug misuse: assessing changes in prevalence. Available from http://www.homeoffice.gov.uk/rds/pdfs04/rdsolr3504.pdf (28 November 2005, date last accessed).
- De Angelis D, Hickman M, Yang S. Estimating long-term trends in the incidence and prevalence of opiate/injecting drug use and the number ofex-users: the use of back-calculation methods and opiate overdose deaths. Am J Epidemiol 2004; 160: 994–1004.
[Abstract/Free Full Text]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||