Estimating incidence of heroin use from treatment data presentation for TDI expert meeting

1
Estimating incidence of heroin use from treatment
datapresentation for TDI expert meeting

• Lucas Wiessing (EMCDDA), Lucilla Ravà and Carla
  Rossi (Univ Rome), EMCDDA, Lisbon, 23 June 2003

2
Background

• Incidence has been developed as part of the key
  indicator prevalence and patterns of problem
  drug use - one of the five EMCDDA key indicators
• Since 1997, EMCDDA ( DG-Research-TSER / Pompidou
  Group) projects have resulted in estimated
  incidence curves for some cities and countries
  (Univ. Rome Tor Vergata - Prof. Carla Rossi)
• Guidelines have been prepared
• Recent project and expert meeting (May 2003) aims
  to obtain more data and estimates, final results
  expected by early 2004 -gt led to request to TDI
  expert group

3
What is the difference between incidence and
prevalence ?

• Prevalence is the total number of cases existing
  at a given moment in time (point prevalence),
  usually it is easier to estimate the cases that
  have existed during a one year period (one year
  period-prevalence)
• Incidence is the rate of NEW cases occurring over
  time, usually estimated for consecutive years
  (yearly incidence, either by calendar years in
  public health data, or by person-years-of-observat
  ion in a cohort study)

4
Why estimate incidence ?

• Incidence is much more sensitive to changes over
  time (trends) than prevalence
• It relates to the start of problem drug use
  careers and to prevention of initiation
• It can provide historical information regarding
  the epidemic of heroin use
• It can be used to forecast treatment needs in the
  near future
• Trends can be derived from some sentinel sites,
  it is not essential to have complete national
  coverage

5
How to estimate incidence of first heroin use
from treatment data

• The principle is simple an observed time series
  of cases entering first treatment shows a certain
  curve (incidence of first treatment).
• If we know the average duration since first
  heroin use, we can back-project this curve to
  obtain the unobserved incidence curve of first
  heroin use
• Average duration is called the latency period
  (LP) and its distribution needs to be estimated
  first

6
There are important limitations

• We cannot estimate total incidence from treatment
  data, only part of it, as a proportion of new
  heroin users will never enter treatment, we call
  this relative incidence which is a lower bound
  of total incidence the shape of the curve
  should however be unbiased
• If we include cases who started heroin use before
  the first year of observed treatment, we have
  left-truncation of LP, i.e. Some short LPs are
  left out.
• The LP is also right-truncated, i.e. we can not
  observe LPs which are longer than our time series
• Changes in incidence can affect LP estimation
  depending on the way LP is estimated

7
Two methods proposed in EMCDDA Guidelines

• 1)
• Latency period (LP) analysis
• Back-calculation method (BC)
• 2)
• Reporting delay adjustment (RDA) or
  lag-correction method (with or without separate
  LP analysis)

8
Latency period (LP) analysis (EMCDDA/Univ. Rome
pilot project)

• This needs to be done on individual data records
• Can be done in one or few sentinel sites only,
  results can then be used by BC method at national
  level and with aggregate data
• Need to understand biases resulting from
  selection of cases into the observed data
• Analysis of covariates e.g. age of first use,
  gender, route of administration (time dependent,
  see request on age at first IDU), which are
  important to take into account in incidence
  estimation
• LP results not only needed for incidence, also
  important for own sake (treatment careers)

9
(No Transcript)
10
Latency period distribution (EMCDDA /Univ. Rome
pilot project)

• Rome metropol. mean 6.5, median 5 yr
• Amsterdam mean 7.1, median 5 yr
• London mean 6.7, median 5 yr

11
Kaplan-Meyer curves Rome no difference by gender
(EMCDDA /Univ. Rome pilot project)
12
Kaplan-Meyer curves Rome strong age effect
(EMCDDA /Univ. Rome pilot project)
13
Back-calculation (BC) method (Brookmeyer and
Gail, Lancet 1986 Heisterkamp et al, Biometrical
Journal, 1999 Downs et al, AIDS 2000)

• Done with a specifically developed programme that
  applies a deconvolution function, to
  back-project observed curve of treatment
  incidence into the unobserved curve of onset
  incidence of first use
• No need for individual data records, aggregate
  time series is sufficient (but at least 8-10
  years long, ideally much longer)
• Latency period distribution is treated as a
  separate input element and can come from other/
  local data
• Very well suited to handle large regions
  /national data

14
Observed and projected treatment incidence,
Italy, BC method (EMCDDA /Univ. Rome pilot
project Ravà et al submitted)
15
Back-calculated incidence of first heroin use,
Italy, BC method (EMCDDA /Univ. Rome pilot
project Ravà et al submitted)
16
Observed and projected treatment incidence,
Amsterdam, BC method (EMCDDA /Univ. Rome pilot
project Ravà et al submitted)
17
Back-calculated incidence of first heroin use,
Amsterdam, BC method (EMCDDA /Univ. Rome pilot
project Ravà et al submitted)
EMCDDA 2001
18
Back-calculated incidence of first heroin use by
region, Italy, BC method (EMCDDA /Univ. Rome
pilot project Ravà et al submitted)
EMCDDA 2001
19
Estimated treatment incidence, Italy, BC method
with age as covariate (Ravà et al submitted)
20
Back-calculated cumulative incidence of first
heroin use by region, Italy, BC method (Ravà et
al submitted)
21
Lag-Correction / Reporting Delay Adjustment (RDA)
method (Brookmeyer and Liao, Am J Epidemiol 1990
Hickman et al, Am J Epidemiol 2001)

• Needs individual level data
• Therefore more appropriate for local level
  estimations (e.g. 1 large treatment centre) but
  complete local coverage is still important
• Easier to understand and no black-box BC
  programme needed
• LP analysis is implicit, but can also be done
  separately (analysis of covariates)

22
Relative incidence of opiate use, Belgium -
French Community, Lisbon and Budapest
(lag-correction method) EMCDDA /Univ. Rome pilot
project
Note data and analyses were carried out by the
national focal points of Portugal, Belgium and
Hungary, in collaboration with EMCDDA and Univ.
Rome
23
Request from incidence estimation expert group
can two items be added to core list ?

• Age at first injection (first priority)
• Compare with and validate age at first use
• Indicates age at first problem use or regular/
  heavy use
• Essential for infectious diseases indicator as it
  allows distinguishing prevalence of HIV/hepatitis
  in new injectors
• Age at first treatment (second priority)
• To estimate LP among those users who have already
  had their first treatment, i.e. the prevalent
  treatment cases
• To understand treatment careers of your clients
• To validate information from first treatment
  demand data
• It would be important to add instructions on how
  to ask this retrospective information always
  relate to important life stages / events (e.g.
  leaving school)

24
Conclusions

• Incidence estimation allows for potentially
  powerful use of treatment data by estimating
  heroin onset incidence and projecting future
  treatment needs
• Adding few items to the TDI protocol could much
  improve use of the data for incidence estimation
  (as is the case for infectious diseases
  surveillance)
• EMCDDA projects and guidelines have led to pilot
  estimates of incidence in the EU, but more effort
  and especially data are needed
• Join forces between TDI group and expert group on
  the key indicator prevalence of problem drug
  use at national/ EU level