Journal of Public Health Advance Access originally published online on April 28, 2006
Journal of Public Health 2006 28(2):88-95; doi:10.1093/pubmed/fdl009
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The prevalence of problem drug misuse in a rural county of England
Richard Holland, Senior Lecturer in Public Health Medicine, School of Medicine, Health Policy & Practice, University of East Anglia, Norwich, NR4 7TJ, UK1
Roberto Vivancos, Honorary Lecturer, School of Medicine, Health Policy & Practice, University of East Anglia1
Vivienne Maskrey, Research Associate, School of Medicine, Health Policy & Practice, University of East Anglia1
Julie Sadler, Medical Statistician (freelance) Ipswich [previously at UEA]1
Daphne Rumball, Consultant Addictions Psychiatrist, Norfolk & Waveney Mental Health Partnership, Norwich, Norfolk NR2 2PA, UK2
Ian Harvey, Professor of Epidemiology and Public Health, School of Medicine, Health Policy & Practice, University of East Anglia1
Louise Swift, Medical Statistician, University of East Anglia, School of Medicine, Health Policy & Practice, University of East Anglia1
1 University of East Anglia—School of Medicine, Health Policy and Practice, Norwich NR4 7TJ, UK
2 Norfolk & Waveney Mental Health Partnership, Norwich, Norfolk NR2 2PA, UK
Address correspondence to Richard Holland, E-mail: r.holland{at}uea.ac.uk
Previous capture-recapture studies have estimated the prevalence of problem drug misuse in urban areas. This study estimates the prevalence in a rural county, Norfolk, using data from four sources: drug treatment agencies, probation, the arrest referral service, and police (drug-related crime with/without acquisitive crime). Careful consideration was given to methods of matching datasets and sensitivity analyses involved altering matching rules and postcode criteria. Whilst it is recognised that acquisitive crime is often related to drug use, this is the first capture-recapture study to incorporate acquisitive crime data. In further sensitivity analyses the proportion of acquisitive crime assumed to be drug-related was varied from 25–60%. The main analysis provided an estimated prevalence of problem drug use in Norfolk of 2.05% (95% confidence interval: 1.66%–2.56%) for ages 15–54 years, considerably higher than the 1.1% currently suggested for the UK. Sensitivity analyses based on varied matching and postcode criteria produced estimates ranging from 2.41%–3.37%, suggesting our estimate may be conservative. Sensitivity analyses assuming that 25–60% of acquisitive crimes were drug-related, produced estimates ranging from 2.02% to 5.73%, further supporting our main analysis. In conclusion, this study provides evidence that problem drug misuse is more prevalent in this rural population than previously thought.
Keywords: drug misuse, prevalence, capture-recapture, rural
 |
Background |
|---|
 TOP Background MethodsResultsDiscussionFundingAcknowledgementsReferences |
|---|
Estimating the prevalence of problem drug misuse is critical to determining the size of drug treatment services that should be provided. The number of known problem drug users reported by treatment agencies in the UK has steadily risen over the last decade, and the estimates of the prevalence of serious drug misuse suggest a prevalence of 1.1% of the UK population aged 15–54.1 This estimate was derived by using the so-called multiple-indicator method which combines information on prevalence that is only available in a few areas (so-called calibration samples or anchor points) with the ‘indicators’ of drug use that are available in all areas (e.g. clients in treatment).
An alternative technique to the multiple-indicator method or surveys is ‘capture–recapture’. This indirect technique, first used in animal populations, has been used widely to estimate the size of ‘hard-to-reach’ populations such as drug users or those with alcohol problems.2,3 Individual subjects are ‘captured’ or observed on two or more occasions, and estimates of the hidden or ‘unseen’ population are based on the degree of overlap between the resulting datasets. Although this technique has been frequently used to estimate the prevalence of drug use in inner-city areas, it has not been used within an English rural county since Brugha estimated the prevalence of drug use in Cheshire in 1993.4
The method of identifying matching individuals between datasets is fundamental to a capture–recapture study, but often little consideration is given to this. Many studies consider only exact matches that do not allow for data collection/transcription errors. In this study, we employed ‘fuzzy’ matching, a technique whereby some close matches are included. In a series of sensitivity analyses, we investigated how varying the matching criteria affected the estimates.
Research carried out by the UK Home Office has demonstrated that acquisitive crime is often committed to fund an underlying substance-misuse problem.5 In spite of this, previous capture–recapture studies that have used police crime data have focused on using drug-related crime within their analysis.4,6,7 Although our study’s main analysis used drug-related crime data, in a second set of sensitivity analyses we investigated whether using data from acquisitive crime provides similar estimates.
In summary, this study was novel for three key reasons: it is the first estimate using capture–recapture of rural problem drug use in England since 1993; it has tested the effect of varying matching criteria and it included data on acquisitive crime. This study, quantifying the extent of problem drug use, formed one part of a needs assessment of crack-cocaine users in Norfolk (see accompanying paper by Vivancos et al.).
|
Methods |
|---|
|
TOPBackgroundMethodsResultsDiscussionFundingAcknowledgementsReferences |
|---|
We used the following definition of problem drug misuse: ‘any person who experiences or causes social, psychological, physical or legal problems relating to their self-administration of a drug, including any form of drug use that involves injecting’.8 All individuals in contact with the agencies involved, at any time in the 6-month period from 1 April to 30 September 2002, were included in our analysis. We used data from four datasets: individuals treated by Norfolk’s drug treatment agencies (Agency set), drug users known to Norfolk Probation (Probation set), individuals charged by Norfolk Constabulary for drug-related crime offences (Police D-set) and individuals managed by the Arrest Referral Service (ARS). The latter is a service directing police arrestees with drug or alcohol problems into drug or alcohol treatment services. Individuals known to the ARS were not included if they were only using alcohol. Ethics approval was gained from Norwich Local Research Ethics Committee.
Individual identifiers and matching
Each dataset contained data on individuals identified by first initial, surname (soundex coded for anonymity),9 sex and date of birth. This information was necessary to first remove duplicate records within each dataset and then to identify matched cases between datasets. We used the same rules for duplicate searching within datasets as we used for matching individuals between datasets. Individuals were considered to be matches (or duplicates), if identifiers were identical using the four fields listed above. We also allowed for one-unit data entry errors in any one field (so-called fuzzy matching). For instance, this allowed a hypothetical William Smith to be matched to a Bill Smith if one dataset coded the first initial as a ‘W’, whereas another coded him as ‘B’. Equally, this allowed dates of birth 12/02/72 and 17/02/72 to be considered as matched. We also allowed for swaps in dates of birth (e.g. 05/03/69 and 03/05/69). It should be noted that individuals who appeared to slightly differ on two or more identifiers (e.g. first initial and date of birth) were not considered a match. Matching was carried out using a programme written specifically for this purpose within Microsoft Access©.
Address/postcode data
Data on the first half (e.g. NR13) of each individual’s postcode were also collected. These were missing for a variable proportion of each dataset (<4% for treatment agency, ARS or probation, but 50–55% of police data). As a result, it was decided not to use postcodes for matching purposes. Nevertheless, these data were still considered important, as individuals could have a Norfolk postcode, a non-Norfolk postcode or no postcode. For simplicity, all records containing no postcode were presumed to be Norfolk residents. The proportion of those with a postcode where the postcode was from outside Norfolk varied from 0.2 to 14%, with the highest proportion occurring in the Norfolk Constabulary dataset.
Estimation
Estimates of the total population in a capture–recapture study can be obtained from any pair of datasets (e.g. using the Chapman estimator).10 However, this assumes that each dataset is independent of the others, i.e. the presence of an individual in a dataset makes his/her presence in another no more or less likely which rarely holds for human populations. Frequently, individuals appearing in one dataset are more likely to appear in another dataset (e.g. those in the police dataset may be more likely to also appear in the probation dataset). Equally, in some cases, appearance in one dataset can make appearance in another dataset less likely. If such positive or negative dependency exists and is not allowed for, it will lead to an under- or over-estimate of the total population, respectively.
Log-linear modelling uses any number of datasets and allows dependencies to be modelled via interaction terms.10 How well a model fits the data is assessed using a chi-squared statistic, which reflects the squared differences between the actual frequencies and those implied by the model (the expected frequencies) and from which a P value, testing the hypothesis that the model is a good fit, can be derived. For the P value to be valid, some assumptions about the values of the expected frequencies are required, but these were satisfied by all the models reported in this article. Using SPSS, and starting with a model containing the maximum number of terms, non-significant terms (P > 0.05) were eliminated from the model one at a time until the model with the smallest number of terms, which still gave an acceptable fit (chi-squared P > 0.10), was achieved. Using the principle of hierarchy, terms could not be eliminated if higher-order terms containing the same datasets were still present in the model. Where P values were borderline (0.05 < P < 0.10), alternative eliminations were followed, with and without the borderline term.
Main analysis and sensitivity analyses
The main analysis used all four datasets including individuals with Norfolk postcodes, no postcodes and those with non-Norfolk postcodes who matched an individual with a Norfolk postcode or no postcode. Datasets were matched including ‘fuzzy’ matching, as described earlier.
Two types of sensitivity analyses were conducted. First, the type of matching was varied. The capture–recapture analysis was repeated allowing only perfect matching on the four identifiers (i.e. no ‘fuzzy’ matching). Next, it was repeated excluding all individuals with non-Norfolk postcodes. It was then repeated including all individuals irrespective of their postcode.
Second, we extended the police drug-related crime dataset to also include a proportion of those charged with acquisitive crimes. The estimated proportion of acquisitive crimes related to a substance-misuse problems varies within the UK from 36 to 66%.5 Higher estimates predominantly originate in inner-city areas. Although individuals who were charged with acquisitive crime who matched individuals in any of our other datasets would appear to have a drug problem, only a proportion of those in the acquisitive crime dataset who had no match were likely to have a drug problem. We therefore adjusted the number of unmatched acquisitive crime individuals so that the overall proportion of police acquisitive crime records that were assumed to have a drug problem was set at 30% (i.e. a conservative estimate). This analysis was then repeated assuming 25, 35, 40 or 60% of acquisitive crimes related to problem drug use. In all sensitivity analyses, except the first, ‘fuzzy’ matching was included.
|
Results |
|---|
|
TOPBackgroundMethodsResultsDiscussionFundingAcknowledgementsReferences |
|---|
Demographic data
After the removal of duplicate records from each dataset, there were 256 known drug users in contact with ARS, 452 individuals with identified drug problems in contact with probation, 1271 individuals in contact with treatment agencies and 445 individuals charged with drug-related offences by the police. There were 300 individuals who appeared in more than one dataset, of whom 68 were identified by ‘fuzzy’ matching. Across the four datasets, the mean age of individuals was 30.2 years, and 25% were female. Table 1 summarizes the mean age and sex distribution within each dataset, whereas Fig. 1 shows the age distribution of known problem drug users in Norfolk.
|
|
Drugs used
Only data from treatment agencies and the ARS reported individuals’ main drug used. For treatment agencies, almost all individuals’ main drug was reported as heroin or other opiates [1077 of 1173 (92%) of those reporting a main drug of use]. In contrast, for those from the ARS, heroin or opiate was the main drug in only 117 of 256 individuals (46%), with both alcohol (51, 20%) and cannabis (46, 18%) also being important main drugs (Fig. 2). Of those using alcohol as their main drug, 60% used cannabis secondarily, whereas 40% used other drugs. Where cannabis was reported as the main drug, 50% combined this with alcohol use and 28% used other drugs with or without alcohol. (Those ARS individuals who only used alcohol were excluded.)
|
Main analysis
The Chapman estimates obtained from pairs of datasets were between 2025 and 7043 (median 4292). Matches found between datasets are summarized in Table 2. The final log-linear model fitted to all four datasets [chi-squared 7.127, 5 degrees of freedom (df), P = 0.211] yielded an estimate of the number of problem drug users in Norfolk of 8278 (95% CI: 6729–10 338, Table 3). This implies a prevalence within the age group 15–54 years of 2.05% (95% CI: 1.66–2.56). The final model found that all pairs of datasets, except police and treatment agency were positively dependent.
|
|
Sensitivity analyses
Matches found between datasets after varying the matching rules, or postcode criteria, are summarized in Table 2. Estimates from log-linear modelling those data were all found to be higher than in the main analysis (base case estimate), with the highest estimate being 10 175 (Table 3). All the confidence intervals overlapped those of the main analysis.
The acquisitive crime dataset consisted of 2412 individuals with a mean age of 26 years (median 23 years), of whom 21% were female (Table 1). When acquisitive crime data were merged with drug-related crime data and matched against the other datasets, 515 (21%) individuals in the acquisitive crime dataset matched individuals in other datasets (Tables 2 and 4). Assuming that 30% of acquisitive crimes were drug related, our capture–recapture estimate was 11 405 (95% CI: 8039–16 714). When the proportion was reduced to 25%, the result (8171) was very close to the base case. As the proportion increased, so too did the estimates up to a maximum of 23 190 when 60% of acquisitive crimes were assumed to be drug related (Table 3). All the fitted models included positive dependency between all pairs of datasets but also required a negative three-factor dependency between Police, ARS and Probation.
|
|
Discussion |
|---|
|
TOPBackgroundMethodsResultsDiscussionFundingAcknowledgementsReferences |
|---|
Main finding of this study
Our estimate suggests a prevalence of problem drug use of 2% of the population aged 15–54 in Norfolk. All our sensitivity analyses tended to exceed the baseline estimate, suggesting that it may be a conservative estimate. This appears to be a high value when compared with the national estimate of 1.1% of the UK’s population aged 15–54 based on a multiple-indicator method. Whether this is a valid comparison is open to debate. We used a broader definition of problem drug use than that recently agreed by the European Monitoring Centre for Drugs and Drug Addiction. This group defined problem drug use as ‘injecting drug use or long duration/regular use of opiates, cocaine and/or amphetamines’.11 Applying this definition was not possible in this study, as neither the probation nor the police provided data on the drugs used by individuals in their datasets.
What is already known on this topic
Two other capture-recapture studies collected data at approximately the same time as our study. The first by Hickman concerned intravenous drug users in three areas (Liverpool, Brighton and Hove, and 12 inner-city London boroughs). This more restricted study collected data in 2000/01 and yielded estimates of between 1.2 and 2%.12 A slightly broader study population including those using opioids, crack-cocaine and benzodiazepines was included in a second study by Gemmell.6 This study collected data in 2000/01, was based in Greater Manchester and found a prevalence of 1.4%. Two older studies have included broader populations defined, as in our study, as problem drug users. Beynon found prevalences of 1.8–3.7% in the Northwest region between 1997 and 1999,2 whereas Hickman found prevalences of 3.1–3.6% in three inner London boroughs between 1992 and 1995.13 Comparing these various estimates is difficult, as drug use has changed substantially over the last decade and varies with social demographic factors. No directly comparable capture–recapture estimate of problem drug use is available for a population from a similarly rural county to Norfolk in the last 10 years. Nevertheless, a 2% prevalence of problem drug misuse for Norfolk seems reasonably consistent with the findings from the studies above. It should be noted that our finding is also consistent with findings from the British Crime Survey.14 This has shown that although drug use is most common in inner-city areas with 5.4% of 16- to 59-year-olds reporting class A drug use in the last year, 2.3% of 16- to 59-year-olds reported class A drug use in rural areas.
Recently, the multiple-indicator method has been used to provide estimates of the prevalence of problematic drug use in each English Drug Action Team area.1 This suggested a prevalence of between 2800 and 4200 in Norfolk in 2001, compared to our estimate of 8200 in 2002. It should be noted that the multiple-indicator method itself uses estimates from other capture-recapture studies to model expected prevalence in areas without such data. This modelling may not itself be accurate; indeed, the model used only accounted for 32% of the variance between problematic drug use estimates. Equally, the estimates may differ because of differing definitions of problem drug use.
What this study adds
No recent capture–recapture study has attempted to estimate the prevalence of problem drug use in a rural county of England. Furthermore, this study has been the first to use acquisitive crime data within a capture–recapture study. As we had no local data on the proportion of acquisitive crimes related to drug use, this was explored in a sensitivity analysis. Nevertheless, during the relatively brief 6-month data collection period, over one-fifth of individuals charged with an acquisitive crime appeared in one of our other datasets (e.g. treatment agency). This suggests that the minimum proportion of acquisitive crimes related to problem drug misuse in Norfolk is likely to be in excess of 20%. Estimates assuming 25% of acquisitive crimes are related to drug use yielded an estimate almost identical to our main analysis.
This study is also the first to report on the effect of varying the matching technique. Our sensitivity analysis demonstrates that had our ‘fuzzy-matching’ technique caused over-matching, it under-estimated the true prevalence by up to 25%. We suggest that future capture–recapture analyses should also report the effect of changing matching rules.
Limitations of this study
It is clear that our broad definition of problem drug use will have tended to increase our estimate of prevalence, and this should be borne in mind when interpreting this result. However, our data mainly, although not exclusively, focused on more severe drug use. Over 90% of treatment agency clients were attending for treatment of their opiate dependence although only 50% of individuals within the ARS were using opiate drugs. No data on drug use were available for those reported within the probation or police datasets. However, national data on drug-related crime suggest that the majority of offences relate to cannabis possession (70%).15 In contrast, data from the Office for National Statistics (ONS) suggest that amongst prisoners the majority of those with a drug dependence use either opiates or stimulants or both, rather than cannabis.16 Thus, although police drug-related crime data may reflect less severe drug use, probation data are likely to reflect more severe drug dependence. Nevertheless, in all cases, these individuals’ drug use was causing them physical, psychological or social problems, to the extent of the individuals being charged or arrested by the police, being recognized by probation as having a drug problem or seeking drug treatment.
In conclusion, our result suggests that the prevalence of problem drug use in rural areas is higher than that suggested by other sources. These data carry clear implications for service provision.
|
Funding |
|---|
|
TOPBackgroundMethodsResultsDiscussionFundingAcknowledgementsReferences |
|---|
This work was funded by a grant from the Norfolk Drug Action Team.
|
Acknowledgements |
|---|
|
TOPBackgroundMethodsResultsDiscussionFundingAcknowledgementsReferences |
|---|
The authors thank Clive Rennie (Norwich PCT), Xany Oliver (Norfolk Drug Action Team), managers and staff of Norfolk NHS Drug Treatment Services, Norfolk Constabulary, Norfolk Probation, Norfolk Tier 3 Your Services and Norfolk Voluntary Agencies.
|
References |
|---|
|
TOPBackgroundMethodsResultsDiscussionFundingAcknowledgementsReferences |
|---|
- Frischer M, Heatlie H, Hickman M. Prevalence of problematic and injecting drug use for Drug Action Team areas in England. J Public Health 2006;28(1):3–9.
- Beynon C, Bellis MA, Millar T et al. Hidden need for drug treatment services: measuring levels of problematic drug use in the North West of England. J Public Health Med 2001;23(4):286–91.
[Abstract/Free Full Text] - Corrao G, Bagnardi V, Vittadini G et al. Capture-recapture methods to size alcohol related problems in a population. J Epidemiol Community Health 2000;54(8):603–10.
[Abstract/Free Full Text] - Brugha RF, Swan AV, Hayhurst GK et al. A drug misuser prevalence study in a rural English district. Eur J Public Health 1998;8:34–6.
[Abstract/Free Full Text] - Matrix MHA & NACRO. Evaluation of Drug Testing in the Criminal Justice Systme in Nine Pilot Areas. London: Home Office, 2003 (Findings, vol. 180).
- Gemmell I, Millar T, Hay G. Capture-recapture estimates of problem drug use and the use of simulation based confidence intervals in a stratified analysis. J Epidemiol Community Health 2004;58(9): 758–65.
[Abstract/Free Full Text] - Wood F, Bloor M, Palmer S. indirect prevalence estimates of a national drug using population: the use of contact-recontact methods in Wales. Health Risk Soc 2000;2(1):47–58.
- Advisory council on the misuse of drugs. Treatment and Rehabilitation. London: HMSO, 1992.
- Soundex coder [Excel Visual Basic]. Howell DJ, 2000.
- Bishop MM, Fienberg SE, Holland PW. Discrete Multivariate Analysis. Cambridge, MA: MIT Press, 1975.
- European Monitoring Centre for Drugs and Drug Addiction. Problem Drug Use. Annual Report. Lisbon, Portugal: European Monitoring Centre for Drugs and Drug Addiction, 2004.
- 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. J Epidemiol Community Health 2004; 58(9):766–71.
[Abstract/Free Full Text] - Hickman M, Cvox S, Harvey J et al. Estimating the prevalence of problem drug use in inner London: a discussion of three capture-recapture studies. Addiction 1999;94(11):1653–62.[CrossRef][ISI][Medline]
- Chivite-Matthews N, Richardson A, O’Shea J et al. Drug Misuse Declared: Findings from the 2003/04 British Crime Survey. London: Home Office, 2005.
- Mwenda L, Kumari K. Drug Offenders in England and Wales 2003. London: Home Office, 2005 (Findings, vol. 256).
- Singleton N, Farrell M, Meltzer H. Substance Misuse Among Prisoners in England and Wales. London: Office for National Statistics, 1999.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||