Transcript Powerpoint

ASSESSING THE QUALITY OF
POPULATION SIZE ESTIMATES OF
PEOPLE WHO INJECT DRUGS
(PWID)
Waimar Tun, Population Council
20th International AIDS Conference
Melbourne, Australia
July 20 – 25, 2014
Background
• Current PWID population size estimations
(PSE) in many countries are not based on
strong data
• UNODC and World Bank requested a review
of existing PWID population size estimates in
10 countries
– Belarus, China, India, Libya, Myanmar,
Philippines, Kazakhstan, Kyrgyzstan, Tajikistan,
& Uzbekistan
Learning objectives
• Be able to critically review existing estimates
and their methodology
• Be able to identify opportunities to improve
the estimates (if required)
• Understand the strengths and weaknesses
of methods of PSE data collection
Where do I get my data?
• Published and grey literature (including HIV- and
drug- related country reports)
• Discussions with stakeholders in-country
(NAP/NAC, UNAIDS, WHO, UNODC, CDC, USAID,
PWID representatives, civil society organizations)
• Stakeholder meeting with representatives from
MoH, drug control agencies, civil society, and
implementing partners
Comment on data sources
• PSE methodologies not or poorly specified
• Reports not translated or translations of
technical terms were ambiguous
• Stakeholders not always aware of PSE
activities happening in their own country
• Time-consuming
What should I consider when I review
the quality of the estimates?
• Are the underlying assumptions of the
method met?
• What are the potential biases and how do
they impact the estimate?
• Are multiple methods used?
• What is the quality of the data used?
Common methods to estimate
PWID population size
Literature review/desk exercise
• Review published and grey literature for similar context and
geographic region
• The benchmark from literature is applied to the adult male
and female population
• Strengths:
– Low cost
– Little time/resources
required
• Weaknesses:
– Local contexts may be
very different
– Data sources may not
be based on rigorous
methods
Delphi
• Systematically solicits and reviews selected experts’
estimates
• Iterative process through a series of feedback and
revisions
• Strengths:
– Utilizes local expert
views and experience
– Does not require raw
data capture
– May be only option for
countries with limited
data sources
• Weaknesses:
– Sometimes based only
on qualitative or
anecdotal information
– Not good for identifying
trends, comparing to
other regions
Mapping/census and enumeration
• Venues where PWID congregate are identified
• Census counts all PWID at all hotspots
• Enumeration counts PWID at a sample of sites
• Strengths:
– Is a real count, not an
estimate
– Can produce a credible lower
limit
• Weaknesses:
– PWID not always accessible,
may not be exposed to
census data collectors
– Assumes you can locate all
PWID
– PWID may not wish to reveal
drug use due to stigma
and/or legal concerns
– Highly time consuming and
expensive
– Safety issues; dangerous
hotspots
Capture-recapture
• PWID are counted and ‘tagged’; a 2nd count is
conducted
• An estimate is obtained through a formula that
includes captures and overlaps between the 2
rounds
• Strengths:
– Fairly easy to
implement in short
period of time
– Usually cost-efficient to
implement
– Can be done with
multiple service sources
• Weaknesses:
– Assumptions of method
difficult to meet in
reality (independent
samples)
– May be dangerous to
implement (unsafe
hotspots)
Service multipliers
• Requires two data sources:
1.
2.
Benchmark (service data such as drug treatment or HIV testing)
Population-based survey with PWID where you obtain info on the
proportion who report that point of contact (‘benchmark)
Estimate obtained from multiplying inverse of proportion to the benchmark
• Strengths:
– Uses existing and
available data from
service providers
– Easily incorporated into
IBBSS with minimal
additional questions
• Weaknesses:
– High-quality service data
may not be available
• No duplicates
• Each target population
member must have chance
of being included in service
data
Wisdom of the crowd
• Ask PWID survey participants to estimate the
number of PWID in a given location
• Assumes that the average response
approximates the actual number
• Strengths:
– Very easy to implement.
Only one question.
– Easily incorporated into
IBBSS or other size
estimation methods
• Weaknesses:
– May be biased if large
segment of population
is not well-networked or
“hidden”
– Bias if not implemented
in a representative
survey
RDS† successive sampling size
estimator‡
•
•
In RDS-based survey, respondents indicate their network
size
Modelling is based on the assumption that those with large
networks are sampled first and that the population will be
depleted at a certain point
• Strengths:
– Easily calculated with existing
RDS survey data
†
• Weaknesses:
– Statistical validity currently
under debate
– Not recommended as a an
“only” method of estimation.
– Results may be biased
depending on number of people
surveyed and actual population
size
Respondent-driven sampling; ‡ Handcock, Gile, Mar (2012)
Network scale-up method† (NSUM)
• Uses general population survey; questions about:
•
•
Number of people they know of a known population
Number of PWID they know
• Strengths:
• Weaknesses:
– Does not ask sensitive
questions directly to
respondent
– National level estimate
†
Bernard, Killworth, Johnsen, and Robinson (1991)
– Average personal network
size difficult to estimate
– Some PWID may not interact
much with members of the
general population
– Respondent may not be
aware that someone in their
network engages in injection
drug use
Triangulation of multiple methods
• Data points from multiple methods are
desirable when possible
– Reduces bias from any single method
– May provide plausible lower and upper bounds
– Informs stakeholder debate
– Facilitates consensus on estimate ranges
PWID population size estimation
(2011 IBBSS, Nairobi)
25,000
20,833
20,000
15,000
Upper plausible 10,937 (~0.5% adults)
13,250
Median 6,107
Lower plausible 5,031
10,000
6,562
5,869
5,000
5,031
3,000
0
STD
HIV testing 1
Literature
review
Drop-in
Source: Population Council, UCSF, NASCOP/Kenya, CDC (2011)
HIV testing 2
WOTC
RDS SS-Size Added
25,000
20,833
Upper plausible 10,937 (~0.5% adults)
20,000
15,000
13,250
11,463
Median 6,107
Lower plausible 5,031
10,000
6,562
5,869
5,000
5,031
3,000
0
STD
HIV testing 1 RDS SS-Size
Literature
review
Drop-in
Source: Population Council, UCSF, NASCOP/Kenya, CDC (2011)
HIV testing 2
WOTC
Scientific rigor
Scientific rigor and costs of methods
Census
Population-based survey
Network scale up
Multiple sample recapture
Capture-recapture
Unique object multiplier
Truncated Poisson
Multipliers, multiple multipliers
RDS – Sequential Size
Unique event multiplier
Mapping, key informants, observation counting
Nomination counting
Place, RAP, ethnography
Registries, police, SHC, drug treatment, unions, workplace
Wisdom of the crowd
Delphi
Consensus
Discrepancies
Soft modeling
Borrow from thy neighbor
Conventional Wisdom
Straw man
Cost
Source: University of California, San Francisco
Scientific rigor
No resources or opportunity for data
collection
Census
Population-based survey
Network scale up
Multiple sample recapture
Capture-recapture
Unique object multiplier
Truncated Poisson
Multipliers, multiple multipliers
RDS – Sequential Size
Unique event multiplier
Mapping, key informants, observation counting
Nomination counting
Place, RAP, ethnography
Registries, police, SHC, drug treatment, unions, workplace
Wisdom of the crowds
Delphi
Consensus
Discrepancies
Soft modeling
Borrow from thy neighbor
Conventional Wisdom
Straw man
Cost
Source: University of California, San Francisco
Scientific rigor
Data collected directly from PWID for
size estimation purposes only
Census
Population-based survey
Network scale up
Multiple sample recapture
Capture-recapture
Unique object multiplier
Truncated Poisson
Multipliers, multiple multipliers
RDS – Sequential Size
Unique event multiplier
Mapping, key informants, observation counting
Nomination counting
Place, RAP, ethnography
Registries, police, SHC, drug treatment, unions, workplace
Wisdom of the crowds
Delphi
Consensus
Discrepancies
Soft modeling
Borrow from thy neighbor
Conventional Wisdom
Straw man
Cost
Source: University of California, San Francisco
Scientific rigor
Data collected from general population
(DHS, AIDS Indicator Survey)
Census
Population-based survey
Network scale up
Multiple sample recapture
Capture-recapture
Unique object multiplier
Truncated Poisson
Multipliers, multiple multipliers
RDS – Sequential Size
Unique event multiplier
Mapping, key informants, observation counting
Nomination counting
Place, RAP, ethnography
Registries, police, SHC, drug treatment, unions, workplace
Wisdom of the crowds
Delphi
Consensus
Discrepancies
Soft modeling
Borrow from thy neighbor
Conventional Wisdom
Straw man
Cost
Source: University of California, San Francisco
Scientific rigor
Data from PWID collected for other
purposes (IBBSS, registries, service data)
Census
Population-based survey
Network scale up
Multiple sample recapture
Capture-recapture
Unique object multiplier
Truncated Poisson
Multipliers, multiple multipliers
RDS – Sequential Size
Unique event multiplier
Mapping, key informants, observation counting
Nomination counting
Place, RAP, ethnography
Registries, police, SHC, drug treatment, unions, workplace
Wisdom of the crowd
Delphi
Consensus
Discrepancies
Soft modeling
Borrow from thy neighbor
Conventional Wisdom
Straw man
Cost
Source: University of California, San Francisco
IBBSS integration
• Size estimation methods increasingly being
integrated worldwide
– Leverages existing resources
– Adds value to behavioral and seroprevalence data already being
collected
• RDS increasingly used for IBBSS recruitment
– “Population-based” estimates
• Forthcoming RDS software-based estimation (SSSize)
– Provides an estimate using existing IBBSS RDS network data
– Has limitations and caveats, should not be used as a sole estimate
source
Other considerations
•
•
•
•
Ethical reviews
Administrative approvals
Safety of research assistants/study team
Involvement local drug using community
representatives
Conclusion
• Important to review how researchers arrived
at the estimate since many are not grounded
in quality data
• Multiple PSE methods should be used
• PWID size estimation should be a part of
routine surveillance
• Stakeholder consensus on estimate ranges
critical
Acknowledgement
•
•
•
•
•
Scott Geibel (Population Council)
Henry Fisher Raymond (UCSF)
Abu Abdul-Quader (CDC)
Pandu Harimurti (The World Bank)
Riku Lehtovuori (UNODC)
GROUP DISCUSSION
Country A
• Epidemic concentrated in PWID (account for
~90% of HIV transmission)
• HIV prevalence in PWID: 15-30% (up to 87% in
one city)
• Civil unrest has hindered PSE of PWID
• PSE (2,000) based on government registration of
drug users† (0.05% of population nationally)
• RDS-based IBBSS was conducted in 2013 in one
city; no PSE
• No IBBSS planned for near future
†
Registries based on treatment registers to arrest counts.
Country B
• Epidemic concentrated in PWID, FSWs and their
clients; PWID HIV prevalence: 7%
• PSE available for 25 out of the country’s 28
states
• National PSE (177,000; 0.02% of population)
obtained through:
– District-level mapping/enumeration at hotspots
– Data updated regularly (by NGOs that implement
targeted intervention at hotspots (WOTC with PWID
and gatekeepers at hotspots)
• IBBSS conducted every 2-3 years; currently
being conducted but no PSE
Country C
• HIV epidemic concentrated in PWID (account for ~60% of
HIV transmission)
• Prevalence of PWID: 100,000-200,000 (~0.9% of
population)
• Has extensive epi-behavioral data, including PSE; regulated
by government
• Latest published PSE available from 2010/11
– Methods and quality varies across region
– National PSE obtained from summing regional results
• Current/Upcoming activities:
– 2014 RDS-based IBBSS with service multiplier (6 sites); 7
multipliers being used
– Some regions will include cap-recap with independent databases
– NSUM (2012/13)
Country D
• Epidemic concentrated (PWID, MSM, FSW/clients)
– PWID HIV prevalence: 18%
• Prevalence of PWID
– Range: 60,000-195,000
– Stakeholder consensus (2002): 75,000
– Estimate based on 0.5% of the population being PWID; this
may be based on registration of drug users
• IBBSS completed in May 2014; includes PSE using
service and unique object multiplier
– Conducted in 16 sites (14 are in high opium-growing states
and two are major urban centers)
– Injection drug use is believed to be occurring outside of these
sites as well
Questions for Group Work
• What are the potential problems/biases with
the current estimate?
• What kind of opportunity can you identify for
improving the estimate? What are possible
next steps?