Problem Drug Use Indicator

Estimating the Prevalence of Problem Drug Use at the Local and National Level

Problem Drug Use Indicator

• Gordon Hay - University of Glasgow, UK
• Ludwig Kraus - IFT Institute für Therapieforschung, Germany
• Lucas Wiessing - EMCDDA
• and country experts

Introduction

• EMCDDA Key Indicator
• Key issues (alcohol)
• Local prevalence methods
• National prevalence methods
• Reporting issues
• Discussion

Presentation

• Methods - Scientific experts in country / EMCDDA
• Reporting - Standard tables
• Interpretation - Workshop on Friday morning

EMCDDA 5 Key Indicators

• Drug use among the general population
• Problem drug use
• Drug-related infectious diseases
• Drug-related deaths and mortality of drug users
• Demand for drug treatment

Problem Drug Use Indicator

• Two main parts
  - National prevalence (Ludwig Kraus)
  - Local prevalence (Gordon Hay)
• Indirect methods
  - Estimates of prevalence
• Derived from research studies
• Pilot studies on incidence

National Prevalence Estimate

• Comparisons across countries
• Types of estimate
  - Problem drug use
  - Injecting drug use
• Refers to entire country
• Uses local estimates

Local Prevalence Estimates

• Types of estimates
  - Problem drug use
  - Injecting drug use
• Local area level
  - Cities
  - Regions
  - Entire nation

Documentation

• Reports from EMCDDA projects
  - Country / local estimates
  - Methodological guidelines
  - Scientific review
• EMCDDA Scientific Monograph
• UNODC - GAP Toolkits
• Scientific papers

Exercise - Alcohol use

• How many people in Lisbon drink?
  - Street survey - ask 50 people
  - 30 people say they drink (60%)
• What would happen if 500 people were asked?
• Survey carried out at night, in the Bairro Alto - does that matter?
• What does 'drink alcohol' mean?
• Sample size
  - Should not affect estimate
  - Can improve confidence intervals
• Representative (unbiased) sample
  - Area of country
  - Age, gender etc
• Case definition
  - Ever drunk alcohol
  - Drunk alcohol in last day, month, year

Extrapolation

• Drug use is largely a hidden activity
• Information can be obtained from a sample of the population
• This information can be extrapolated to provide information on the entire population

Extrapolation Population Surveys

• A sample of the population is obtained
• 40% of the sample use cannabis
• therefore
• 40% of the population use cannabis
• Main issues
  - Sample size
  - Representative sample
  - Case definition

Extrapolation

• Information can be obtained from a sample of drug users
  - Treatment services
  - HIV data, mortality
• This information can be extrapolated to provide information on all drug users
• Prevalence estimates can be obtained

Local Prevalence Estimation

• Multiplier Methods
  - Mortality Multiplier Method
  - Nomination methods
• Capture-recapture methods
  - Scotland
• Truncated Poisson model
  - Luxembourg

Multiplier Methods

• Benchmark figure
  - Number of drug related deaths
  - Number of drug users in treatment
• Multiplier
  - Derived from specific studies
  - Taken from previous studies

Example - Scottish Data

• 227 drug users died in 1997
• Studies show that approximately 2% of drug users die each year
• therefore
• For every 1 death there are 50 drug users
• 11,350 drug users in Scotland

Assumptions - benchmark

• All drug-related deaths are identified
  - Overdose (accidental / intentional)
  - Diseases (endocarditis)
• All drug-related deaths are reported
  - National register (ICD-9 codes)
• Deaths due to HIV or hepatitis

Assumptions - multiplier

• Mortality rate is known
  - drug users in treatment
  - impact of methadone
• Common rate for all drugs
  - heroin
  - stimulants
  - injecting / smoking

Other Multipliers

• Number of drug users in treatment
  - Methadone
  - Specific treatment agencies
• Number of drug-related crimes
  - possession of drugs
• HIV data
  - drug injectors

Nomination Methods

• Identify a sample of drug users
• Ask them to 'nominate' friends
• Ask how many are in treatment
  - 10 friends, two in treatment
• Derive multiplier
  - Five drug users for each one in treatment
• 'Snowball' to find more drug users
• Choice of initial sample
  - Treatment agency
  - Ethnicity / gender
• Choice of benchmark
• Time consuming and expensive
• Useful in specific populations
  - Small towns
  - Ethnic groups

Capture-recapture methods

• Simple idea:
  - Only a proportion of drug users are in contact with treatment agencies
• Examine the overlap between those in treatment and a second sample (e.g. police)
• Find the proportion in treatment
• Thus estimate the total number of drug users

Capture-recapture methods

Capture-recapture Methods

• Animal and fish populations
  - Capture a sample of fish
  - Mark them
  - Release them
  - Recapture a sample at a later date
  - Look for marks
  - Estimate population size

Example - Fish

Unknown number of fish in a lake

Unknown number of fish in a lake Catch a sample and mark them

Unknown number of fish in a lake Catch a sample and mark them Let them loose Recapture a sample and look for marks

Estimate population size

    n1 = number in first sample 15
    n2 = number in second sample 10
    n12 = number in both samples 5
    N = total population size
assume that
    n1/N = n12/n2    therefore    15/N = 5/10

     N = (10 x 15) / 5 = 30

Other Applications

• Disease registers
  - Diabetes
• Hidden populations
  - Drug users
  - Sex workers
  - Homeless people
• Under counting in census (USA)
• How many Americans lived in London, 1770-1775

Drug Use Application

• Drug users
  - Identify two samples
      - Police
      - Treatment agencies
  - Find overlap
  - Estimate population size

Drug Use Example

	N
Treatment	775
HIV Test Data	46
Police	76
Unique Individuals	855

Overlap (treatment, HIV and Police)

~ 1/3 of police sample in treatment total estimate = 3 x 775 = 2325

Main Assumption

• Samples are independent
  - Police do not stand outside agency arresting people
  - Participation in treatment does not reduce the need to commit crimes
• Samples are often not independent

Three Sample Analysis

• Statistical Analysis
  - Requires statistical packages
  - Log-linear models
  - Explain relationship between sources
• Estimate the size of the hidden population
• Estimate the total population size

Overlap (treatment, HIV and Police)

Three Sample Analysis

• Police source independent of other sources
• HIV and treatment sources related
  - Known drug users = 855
  - Hidden population = 1702
• Total population = 2557
• 95% Confidence Interval= [1974 - 3458]

Assumptions

• Closed population
  - Drug users do not migrate in or out of the area
  - People do not start to use drugs during study
• Correct identification of overlaps
• Drug users have similar behaviour
  - Drug users are not 'heterogeneous'

Truncated Poisson model

• Luxembourg treatment data
  - Single source - treatment register
  - Method uses three figures
      - number appearing once in treatment
      - number appearing twice in treatment
      - total number in treatment
• Assumes that number of times in treatment follows a Poisson distribution

Truncated Poisson model

Truncated Poisson model

Hidden population

Hidden population

Truncated Poisson model

National Prevalence

• Methods used for local estimates
  - Multiplier method
  - Capture-recapture method
• Synthetic estimation
• Multivariate Indicator Method

Synthetic Estimation

• Extension to the Multiplier Method
• Benchmark figure (indicator)
  - Drug users in treatment
  - Drug-related crime
• Multipliers
  - Derived from prevalence estimates in two (or more) cities / regions
  - Does not assume a common multiplier across areas

Example - 'City A'

• There are 200 drug users in contact with drug treatment agencies
• A research project estimated that there are 2,000 drug users in the city
• Therefore
  - For every one drug user in treatment, there are 10 living in the city

Example - 'City B'

• There are 500 drug users in contact with drug treatment agencies
• A research project estimated that there are 7,500 drug users in the city
• Therefore
  - For every one drug user in treatment, there are 15 living in the city

Synthetic Estimation

Synthetic Estimation

Synthetic Estimation

Synthetic Estimation

Example

• City C
• 400 drug users in treatment
• approximately 5,600 drug users

• Town D
• 50 drug users in treatment
• can we extrapolate to get an estimate?

Synthetic Estimation

• Need two (or more) anchor points
• Linear relationship between prevalence and indicator (linear regression)
• Can be difficult to extrapolate beyond anchor points
• National prevalence obtained by adding estimates for cities / regions

Multivariate Indicator Model

• Similar to the synthetic estimation method
• Relation between prevalence and indicators
  - More than one indicator
  - Indicators combined into one (or two) factors
  - Principle components
• Anchor points
  - Local prevalence estimates
  - At least 2
• Indicators
  - Available for entire country
  - Split by region
  - Drug-related or social

Example - UK

• Indicators
  - convictions for drug offences A
  - seizures of controlled drugs B
  - drug users in treatment C
  - drug-related cases of HIV D
  - drug-related deaths E
• All available at regional level
• Anchor points available

Regions	A	B	C	D	E	G
Northern & Yorkshire	11,356	13,285	9,722	37	344
Trent	6,451	7,010	3,580	67	207
Anglia and Oxford	3,761	4,183	3,762	79	216
North Thames	17,696	21,168	7,842	334	352	44,410
South Thames	13,987	16,530	7,774	122	346	38,140
South West	10,600	12,717	5,890	60	311
West Midlands	7,125	5,398	4,322	26	193	13,130
North West	12,557	11,804	8,958	63	402
Wales	6,110	5,870	2,282	14	139	8,357
Scotland	13,008	13,452	8,614	687	267	38,000

Reporting

• One of the 5 key indicators
• Standard tables
  - National estimates
  - Local estimates
• Used more generally to put information in context within national reports

Target Populations

• Problem drug use
  - injecting drug use or long-duration/regular use of opiates, cocaine and/or amphetamines
• Current injectors
• Other definitions
• Age range 15-64

Estimates

• Methods used
• Data sources
• Central estimates
• Range or 95% confidence intervals
• Population aged 15-64 in area / country

Reporting Issues

• Population sizes
  - Do they match with study areas?
• Definitions
  - Age range
  - Definition of drug use
• Stratified estimates
  - Age
  - Gender
  - Region

Recommendations

Convene expert group and discuss methods and data (both data and statistical experts)
• Try to come up with first estimates, even if they still are of dubious quality! (but don't publish)
• Discuss quality of estimates, necessary improvements, data needs in expert meetings
• Try to secure funds for continuous work
• Try to involve the data owners actively
• Try to establish a stable national expert group, involving all key players, be inclusive

Recommendations

• Make estimates at national and local levels
• Local level studies provide data and insights for national estimates
• Make estimates both in high-prevalence and low prevalence (rural) areas
• Use different methods
• Try to get repeated comparable estimates
• Quality of estimates depends on quality of data!

Discussion Points

• Estimates are derived from research studies
  - Links with scientific experts / researchers
• Cost of studies
  - Different research methods
• Research issues
  - Data protection / confidentiality
  - Peer review

Discussion Points

• Coverage of estimates
  - Local prevalence estimates
  - National prevalence estimates
• Cross - national Comparisons
  - Different methods
  - Different Definitions
• Workshop on Friday morning