Page 1

Please cite this article in press as: Pinkerton, S.D. How many HIV infections are prevented by Vancouver Canada’s supervised injection facility?

International Journal of Drug Policy (2011), doi:10.1016/j.drugpo.2011.03.003

ARTICLE IN PRESS

GModel

DRUPOL-989; No.of Pages5

International Journal of Drug Policy xxx (2011) xxx–xxx

Contents lists available at ScienceDirect

International Journal of Drug Policy

journal homepage: www.elsevier.com/locate/drugpo

Commentary

How many HIV infections are prevented by Vancouver Canada’s supervised

injection facility?

Steven D. Pinkerton∗

Center for AIDS Intervention Research, Department of Psychiatry and Behavioral Medicine, Medical College of Wisconsin, 2071 North Summit Avenue,

Milwaukee, WI 53202, United States

a r t i c l ei n f o

Article history:

Received 14 May 2010

Received in revised form 14 February 2011

Accepted 1 March 2011

Available online xxx

Keywords:

Supervised injection facility

Injection drug use

HIV transmission

Mathematical modelling

Cost-effectiveness

a b s t r a c t

Mathematical modelling analyses of drug injection-related HIV risk reduction interventions can pro-

vide policy makers, researchers, and others with important information that would be difficult to obtain

through other means. The validity of the results of mathematical modelling analyses that rely on sec-

ondarydatasourcescriticallydependsonthemodel(s)employedintheanalysesandtheparametervalues

used to populate the models. A recent article in the International Journal of Drug Policy by Andresen and

Boyd (2010: 70–76) utilised four different mathematical models of injection-related HIV transmission

to estimate the number of HIV infections prevented by Vancouver Canada’s Insite supervised injection

facility (SIF). The present article reviews and critiques the mathematical models utilised in the Andresen

and Boyd article, then describes an alternative—and potentially more accurate—method for estimating

the impact of the Insite SIF. This model indicates that the SIF prevents approximately 5–6 infections per

year, with a plausible range of 4–8 prevented infections. These estimates are far smaller than suggested

by Andresen and Boyd (19–57 prevented infections).

© 2011 Elsevier B.V. All rights reserved.

Introduction

Vancouver Canada’s Insite supervised injection facility (SIF) is

the first such facility in North America (Wood, Tyndall, Montaner,

and Kerr 2006). Insite provides Vancouver’s injection drug users

(IDU) with a “safe,” supervised location to inject drugs using ster-

ile, facility-supplied syringes. Registered nurses and other health

care personnel are on hand to monitor the injection process and

intervene if necessary. Approximately 220,000 injections occur

at the Insite SIF each year (Tyndall, Kerr, Zhang, et al., 2005).

These are sterile injections, with no possibility of HIV transmis-

sion. Moreover, because Insite syringes are disposed of after use

by a single user, these syringes cannot transmit HIV to another

IDU.

The actual reduction in HIV transmission due to the Insite SIF is

difficult to quantify. Incidence surveillance data cannot separate

out the impact of the Insite SIF from the multiple other factors

that influence HIV incidence in the Vancouver area. Mathemati-

cal models of injection-associated HIV transmission, such as those

proposed by Bayoumi and Zaric (2008) and Pinkerton (2010), are

needed to estimate the isolated impact of the Insite SIF. By gener-

ating plausible estimates of the number of HIV infections averted

∗Tel.: +1 414 955 7762; fax: +1 414 287 4206.

E-mail addresses: pinkrton@mcw.edu, pinkrton44@yahoo.com

by the Insite SIF, mathematical models can provide policy mak-

ers,researchers,andotherswithimportantinformationthatwould

be difficult, if not impossible, to obtain through other means. The

validity of the results of mathematical modelling analyses that

rely on secondary data sources critically depends on the model(s)

employed in the analyses and the parameter values used to popu-

late the models.

Inparticular,publishedreports:(1)shoulddescribethemodelin

sufficient detail that the reader understands the theoretical basis

for the model and how it applies to the goal of the analysis; (2)

should specify the source and values of all model parameters; (3)

shoulddescribetheextenttowhichparametervalues“fit”withthe

particular circumstances under consideration; (4) should permit

the reader to replicate the analyses to a greater or lesser degree,

depending on the complexity of the model; and (5) should provide

coherent interpretations of any unexpected findings or deviations

from plausibility.

A recently published article by Andresen and Boyd (2010)

utilised four different mathematical models of injection-related

HIV transmission to estimate the number of HIV infections pre-

vented by Vancouver Canada’s Insite supervised injection facility

(SIF). The present article reviews and critiques the mathemat-

ical models and parameter values utilised in the Andresen

and Boyd article, then describes an alternative—and poten-

tially more accurate—method for estimating the impact of the

Insite SIF.

0955-3959/$ – see front matter © 2011 Elsevier B.V. All rights reserved.

doi:10.1016/j.drugpo.2011.03.003

Page 2

Please cite this article in press as: Pinkerton, S.D. How many HIV infections are prevented by Vancouver Canada’s supervised injection facility?

International Journal of Drug Policy (2011), doi:10.1016/j.drugpo.2011.03.003

ARTICLE IN PRESS

S.D. Pinkerton / International Journal of Drug Policy xxx (2011) xxx–xxx

GModel

DRUPOL-989; No.of Pages5

2

Review of models

The four models utilised in Andresen and Boyd’s (2010) assess-

ment of the epidemiological impact of the Insite SIF are separately

reviewed and critiqued below. This section concludes by consider-

ing how these seemingly disparate approaches can be reconciled

within a common modelling framework.

Model #1: Laufer (2001)

The number of HIV infections prevented by a particular risk

reduction intervention is A=I0−I1, where I0and I1denote the

number of incident infections without and with the intervention,

respectively. In the simplest, most intuitively appealing model

of injection-associated HIV acquisition, the expected number of

incident infections is estimated using the equation I=(1−?)Ni,

where N is the number of persons in the study population, ? is

the prevalence of existing HIV infection in this population (hence

(1−?)N is the number of persons at risk of acquiring HIV), and

i is the average risk of HIV acquisition for an at-risk person (i.e.,

the expected incidence rate). Combining the previous two equa-

tions, the number of infections averted can be expressed as:

A=I0−I1=(1−?)N(i0−i1)=(1−?)N?i,wherei0andi1denotethe

risk of HIV acquisition before and after the intervention was imple-

mented, and ?i=i0−i1. In short, this simple model posits that the

number of HIV infections prevented by an injection risk reduc-

tion intervention is a direct function of the amount by which the

intervention reduces participants’ risk of HIV acquisition. The main

challenge in applying this model to evaluate the impact of an SIF,

syringeexchangeprogramme,orotherinjectionriskreductionpro-

gramme is that it requires an estimate of the amount by which the

programme reduces participants’ average risk of acquiring HIV, or

equivalently, the amount by which it reduces the HIV incidence

rate.

In the Andresen and Boyd article this model is specified as

A=(1−?)N?i, which is correct unless ?i is interpreted as a rel-

ative rather than an absolute reduction in risk. In applying this

model, Andresen and Boyd treat ?i as a relative risk reduction—in

which case the correct equation would be A=i0(1−?)N?i, not

A=(1−?)N?i as stated in the article. It also appears that Andresen

and Boyd assumed that Insite reduced participants’ risk of acquir-

ing HIV by ?i=0.25, not ?i=0.75 as shown in Table 2. In any case,

no justification is provided for assuming that the Insite SIF would

produce a 0.25 (or 0.75) absolute or relative reduction in incidence

amongst SIF clients.

Model #2: Kaplan and O’Keefe (1993)

Kaplan and O’Keefe’s (1993) model is consistent with the model

described above, but expands upon this model by providing an

Table 1

Application of Models #1–#3 in Andresen and Boyd (2010).

Model #1 (“Laufer complex”)Model #2 (“Kaplan and O’Keefe”)Model #3 (“Laufer simple”)

Correct equationa

Equation as specified

Equation as applied

Parameter values

N (# of IDU)

? (HIV prevalence)

i0(incidence rate wo/Insite)

?i (incidence reduction)

?b (reduction in # borrows)

c (syringes contaminated w/HIV)

? (decontamination rate)

˛ (per injection transmission rate)

b (borrows per IDU per year w/Insite)

E (SIF injections per IDU per year)

A=(1−?)N?i

A=(1−?)N?i

A=i0(1−?)N?i

I=(1−?)Nbc(1−?)˛

I=(1−?)bc(1−?)˛

I=?Nbc(1−?)˛A=?N?bc(1−?)˛

A≈E/(E+b)(1−?)Ni0

E/(E+b)

A=E/(E+b)(1−?)Ni0

1700b

17%d

5.26%f

0.25g

5000c

22.54%e

1700b

17%d

5.26%f

109.6h

40.5%i

83%j

0.67%k

140l

139m

Infections prevented18.657.037.1

aIdenotesthenumberofincidentHIVinfectionsexpectedwithorwithouttheSIF;AdenotestheestimatednumberofinfectionspreventedbytheInsiteSIF.(Theremaining

parameters in these equations are described below.)

bApproximately 1700 unique IDU utilise the SIF each month (Tyndall, Kerr, Zhang, et al., 2006).

cAn estimated 5000 IDU reside in the Downtown Eastside (DTES) of Vancouver, where Insite is located (Health Canada, 2008).

dApproximately 17% of Insite clients are HIV-positive (Petrar et al., 2007; Tyndall, Wood, Zhang, et al., 2006).

eThis prevalence value was calculated as a weighted sum of the prevalence of infection amongst the estimated 1700 Insite clients (?=17%) and amongst the remaining

5000–1700=3300 DTES IDU (?=25.4%): weighted ?=(1700*17%+3300*25.4%)/5000=22.54%. The source of the 25.4% prevalence estimate for DTES IDU who do not utilise

the SIF is not specified in the target article.

fThis value for the HIV incidence in the absence of the Insite SIF is based on an early to mid-1990s estimate of the incidence of infection in New York City (Des Jarlais et

al., 1996); its applicability to the DTES is open to question.

gThe target article does not provide any justification for this value which, quite clearly, is incorrect. The ?i parameter indicates the absolute reduction in incidence, not

the proportional reduction. Because Andresen and Boyd assumed that the baseline incidence equals 0.0526, a 0.25 incidence reduction is untenable.

hFor this analysis Andresen and Boyd assumed that DTES IDU inject 913 times per year and used borrowed syringes for 30% of their injections (Andresen & Boyd, 2010).

Therefore, the baseline number of borrows is 273.9 per IDU per year. The authors assumed that the presence of the Insite SIF produced a 40% reduction in syringe borrowing

amongst DTES IDU ((1–40%)*273.9=164.3). They, therefore, concluded that the Insite SIF results in 109.6 fewer borrows per DTES IDU per year.

iThis very high syringe contamination rate is based on an early 1990s study conducted in New Haven (Kaplan & O’Keefe, 1993). Of note, the corresponding prevalence of

infection in this population was nearly 64%. One would expect the syringe contamination rate to be substantially smaller in a population in which the prevalence of infection

is no greater than 15–25%.

jThis bleaching/decontamination rate (83%) reflects the rate at initiation of the New Haven SEP, in a population in which an estimated 64% of all circulating syringes were

contaminated. Bleaching may be less common in populations with lower syringe contamination rates (Bayoumi & Zaric, 2008). Moreover, bleaching is not necessarily 100%

effective (Normand, Vlahov, & Moses, 1995).

kThis estimate is from Kaplan and O’Keefe (1993).

lAs in note h, Andresen and Boyd assumed that DTES IDU inject 913 times per year and used borrowed syringes for 30% of their injections. However, Model #3 is restricted

to the 1700 Insite SIF clients who utilise the SIF for an average of 139 injections per year (this usage estimate is based on the maximum operating capacity of the Insite SIF,

236,520 injections per year; Tyndall, Kerr, Zhang, et al., 2006). This leaves 913–139=774 “street” injections. In their base-case analysis Andresen and Boyd assumed that 30%

of these injections (232) would involve borrowed syringes. An additional 40% reduction (see note h) would reduce the effective number of borrows to 139 borrows per IDU

per year.

mSee note i.

Page 3

Please cite this article in press as: Pinkerton, S.D. How many HIV infections are prevented by Vancouver Canada’s supervised injection facility?

International Journal of Drug Policy (2011), doi:10.1016/j.drugpo.2011.03.003

ARTICLE IN PRESS

S.D. Pinkerton / International Journal of Drug Policy xxx (2011) xxx–xxx

GModel

DRUPOL-989; No.of Pages5

3

Table 2

Number of HIV infections prevented by Insite’s supervised injection facility. Per year.

Parameter (base-case value; range) HIV infections preventedParameter sources

Low value

b–c value High value

Incidence of HIV infection with Insite, i1(1.6%; 1.3–2.1%)

Prevalence of HIV infection in DTES, ? (17%; 12–22%)

Injections per year per IDU (1095; 730–1460)

4.3

5.6

8.3

5.2

5.2

5.2

6.9

4.9

3.8

Health Canada (2008); Tyndall, Kerr, Lai, et al. (2006)

McInnes et al. (2009) and Tyndall, Wood, Zhang, et al. (2006)

Health Canada (2004); McCoy, Metsch, Chitwood, Shapshak,

and Comerford (1998), Remis, Bruneau, and Hankins (1998)

and Singer, Himmelgreen, Dushay, and Weeks (1998)

Pinkerton (2010)

Health Canada (2008); Tyndall et al. (2005)

Bayoumi and Zaric (2008) and Kaplan and O’Keefe (1993)

Buxton (2007)

% of injections with borrowed syringes, ˇ (8.3%; 4.0–9.2%)

Injections per year at Insite SIF (220K; 200–236.5K)

Syringe decontamination probability, ? (42%; 28–83%)

Syringes distributed per IDU per year, s (260; 220–300)

5.4

4.7

5.3

5.2

5.2

5.2

5.2

5.2

5.2

5.6

5.0

5.3

explicit equation to estimate each participant’s risk of acquir-

ing HIV from borrowed syringes: r=bc(1−?)˛, where b is the

average number of injections with borrowed syringes during the

time period being modelled; c is the proportion of borrowed

syringes that are contaminated with HIV; ? is the probability that

a borrowed, contaminated syringe is effectively decontaminated

(e.g., through bleaching or rinsing) prior to use; and ˛ is the

probability of acquiring HIV from a single injection with a con-

taminated syringe. Thus, in this model, the expected incidence is

I=(1−?)Nr=(1−?)Nbc(1−?)˛.

In Andresen and Boyd’s application of this model to the Insite

SIF it appears that they inadvertently used ? rather than 1−? (i.e.,

they used the incorrect equation I=?Nbc(1−?)˛ to calculate the

number of prevented infections; see Table 1). Consequently, their

estimate of the number of infections prevented by the Insite SIF

must be called into question.

Model #3: Kaplan (1995)

This model was developed by Edward Kaplan (Kaplan, 1995;

Kaplan&Brandeau,1994)asanapproximationtohismorecomplex

“needles that kill” model (e.g., Kaplan & O’Keefe, 1993). According

to this model, the reduction in HIV incidence obtained by introduc-

ing a syringe exchange programme into a community in which no

SEP exists is proportional to the ratio E/(E+b), where E denotes the

number of syringes exchanged per IDU per year and b represents

the number of injections with borrowed syringes per IDU per year.

The number of HIV infections averted by the SEP, A, then can be

estimated by multiplying the “incidence reduction factor” E/(E+b)

by the expected incidence amongst uninfected IDU in the commu-

nity served by the SEP: A≈(E/(E+b))I0, where I0is the number of

incident infections that would be expected if there were no SEP.

It is important to note that this model applies only to the intro-

ductionofanSEPwherenoneexistedbefore,nottotheintroduction

ofasupervisedinjectionfacility.Kaplan’ssimplifiedmodelisbased

on “syringe circulation theory,” which posits that SEPs reduce HIV

incidence by decreasing the amount of time each syringe remains

in circulation, and consequently the likelihood that it is shared by

multiple injectors. This, in turn, reduces the proportion of circu-

lating syringes that are contaminated with HIV, hence the risk of

HIV acquisition when an IDU injects with a borrowed syringe. The

effectiveness of SEPs in reducing HIV incidence in drug-injecting

populations is due not only to the provision of sterile syringes,

but—and perhaps more importantly—to the removal of contam-

inated syringes from circulation. Supervised injection facilities

provide clients with sterile syringes but do not remove contami-

nated syringes from circulation in the greater community. This is a

critical difference.

InAndresenandBoyd’sapplicationofthismodeltotheInsiteSIF,

E is set to the number of syringes used in the SIF, per year, for each

of the estimated 1700 SIF clients (Tyndall et al., 2006). In contrast,

in Kaplan’s model, E represents the number of syringes exchanged,

on a one-to-one basis, per IDU per year. Thus, in the model, E equals

both the number of sterile syringes that enter circulation per IDU

peryearandthenumberofusedsyringesremovedfromcirculation

per IDU per year. Because the sterile syringes that are used within

the Insite SIF are not matched on a one-to-one basis by the removal

of potentially contaminated syringes from general circulation, this

model cannot be used to estimate the number of HIV infections

prevented by the SIF.

Model #4: Jacobs et al. (1999)

This is another SEP model whose applicability to the Insite SIF

could be questioned. Neither Jacobs et al. (1999) nor Andresen

and Boyd (2010) provide any description of the logic or presumed

transmission mechanisms that underlie this model. It appears

that this model can be expressed in the notation used here

as I=(1−?)Nb(1−?)[1−(1−?˛)m], where m is the number of

IDUs who use each syringe and who potentially could contam-

inate it. This model can be linearized without significant error:

I=(1−?)Nb(1−?)?˛m. This is identical to Model 2 if one equates

c (the contamination rate amongst borrowed syringes) with the

product ?m, which is the number of HIV-infected sharing partners.

However, from the Andresen and Boyd article it is not clear which

parameter they manipulated, or by how much. Consequently, the

reader cannot easily replicate their finding of 27 infections averted.

Overall critique of target article

Here we apply the 5 modelling and reporting criteria from the

Introduction to the target article by Andresen and Boyd.

Model selection and justification

The target article does not provide a rationale for the selection

of any of the 4 models used to estimate the impact of the Insite

SIF or any justification of their appropriateness vis-à-vis evaluating

the risk reduction effectiveness of a supervised injection facility. As

notedabove,atleastoneofthemodels(model#3)cannotbeusedto

estimate the number of HIV infections prevented by a supervised

injection facility without significant modifications. This model is

misapplied by Andresen and Boyd.

Parameter specification

All 4 models require an estimate of the number of syringe

borrowing episodes per IDU per year. Although the target article

provides an estimate of the proportion of injections for which bor-

rowed syringes are used, it does not specify the total number of

injections or the number of borrows.

Appropriateness of parameter values

The target articles uses contamination rate estimates from

New Haven in the early to mid 1990s; a 17% prevalence rate of

HIV infection amongst IDU in one place and a 22.54% prevalence

Page 4

Please cite this article in press as: Pinkerton, S.D. How many HIV infections are prevented by Vancouver Canada’s supervised injection facility?

International Journal of Drug Policy (2011), doi:10.1016/j.drugpo.2011.03.003

ARTICLE IN PRESS

S.D. Pinkerton / International Journal of Drug Policy xxx (2011) xxx–xxx

GModel

DRUPOL-989; No.of Pages5

4

elsewhere; and assumes without justification a 0.25 reduction in

the risk of acquiring HIV as a result of injecting, part time, at

an SIF.

Replication of results

Because models are incorrectly specified and/or the relevant

parameter values are not specified correctly, it is a challenge to

derive the results listed in Andresen and Boyd’s Table 4. The results

for model #4, for example, are a complete enigma because the arti-

cle does not specify which of the several parameter values the SIF

might affect.

Consistency and deviations from plausibility

The results obtained from fundamentally similar models should

be consistent. For example, the third model can be derived from

the first model by substituting E/(E+s) for the incidence reduction

parameter, r=?i/i0. Applying the third model, Andresen and Boyd

obtained an estimate of r=0.5 (approximately), whereas in the first

modeltheyassumedthatr=0.25.Consequently,theresultsofthese

two models differ by about 50%. Although the models themselves

are formally equivalent, the results reported by Andresen and Boyd

are not.

Model reconciliation

Models #1–3 share a common framework. Model #1 sug-

gests that: (1) HIV incidence is I=(1−?)Ni; and (2) the

numberofinfectionsprevented

A=I0−I1=(1−?)N(i0−i1).Intheseequations,(1−?)Nisthenum-

ber of at-risk persons in the population being modelled and i

denotes their average risk of HIV acquisition, either with (i1) or

without (i0) the intervention. Because the average risk of HIV

acquisition is identical to the incidence rate, these equations are

intuitively obvious. For example, the first equation simply states

that the number of incident infections equals the number of per-

sons at risk of infection times the incidence rate amongst at-risk

persons.Combined,theseequationsimplythatthenumberofinfec-

tions averted by an intervention equals the product of the number

of incident infections expected in the absence of the intervention

(I0) and the proportionate reduction in risk (i.e., incidence rate)

engendered by the intervention—that is: A=[(i0−i1)/i0]I0, which

also can be written as: [(i0−i1)/i0](1−?)Ni0.

Model #2 is more explicit than Model #1—it suggests that

i=bc(1−?)˛, but otherwise is completely consistent with the

first model. For example, holding the decontamination param-

eter (?) and the per-injection transmission probability (˛)

constant, (i0−i1)/i0=(b0c0−b1c1)/b0c0and A=[(b0c0−b1c1)/b0c0]

I0=[1−(b1/b0)(c1/c0)]I0.

InModel#3,theproportionateriskreduction((i0−i1)/i0)isesti-

mated by the ratio E/(E+b). Consequently, A≈[E/(E+b)]I0(Kaplan,

1995).

These three models are increasingly more specific. Model #1

applies to any risk reduction intervention, including sexual risk

reduction interventions. Model #2 is applicable to injection risk

reduction interventions that reduce the syringe borrowing rate,

the syringe contamination rate, and/or the bleaching (decontami-

nation) rate. The most specific model, Model #3, applies only to the

rather unique situation in which an SEP is introduced where none

existed before.

Of note, the “common model,” A=[(i0−i1)/i0]I0, also can

be written: A=[(i0−i1)/i1]I1, where I1 is the number of inci-

dent infections with the intervention in operation. Similarly,

when ? and ˛ are presumed constant, A=[(b0/b1)(c0/c1)−1]

I1=[(b0/b1)(c0/c1)−1](1−?)Ni1, where i1, as above, is the inci-

dence rate with the intervention in operation. This formulation

allows us to model the impact of Insite using current estimates

by reducing incidenceis

of the incidence of infection amongst Vancouver IDU, as described

below.

An alternative model of the impact of the Insite SIF

The modelling framework described above can be used to gen-

erateaplausibleestimateoftheimpactoftheInsiteSIF.Atthemost

basiclevel,theSIFreducesthenumberof“street”injectionsbypro-

viding IDU with a safe environment to inject with sterile syringes.

This,inturn,decreasesthenumberofborrowswithpotentiallycon-

taminated syringes, thereby reducing the risk of HIV acquisition.

Approximately5000IDUresideinVancouver’sDowntownEast-

side (DTES), where Insite is located (Health Canada, 2008). If one

assumes that each of these IDU injects 3 times per day, on aver-

age (Pinkerton, 2010), then the 5000 DTES IDU inject a total of

5,475,000timeseachyear.Anestimated220,000oftheseinjections

occur within the SIF (Health Canada, 2008; Tyndall, Kerr, Zhang,

et al., 2006). Let ˇ represents the proportion of non-SIF (“street”)

injections that involve borrowed syringes and assume that ˇ is

not affected by the Insite SIF. The average number of borrows per

IDU per year with the SIF in operation is b1=(5475–220K)ˇ/5000,

whereas without Insite it would be b0=(5475K)ˇ/5000. Conse-

quently, b0/b1=1.04.

Thereductionin“street”injectionsalsoreducestheHIVcontam-

inationrateamongstcirculatingsyringesbydecreasingtheaverage

number of times each syringe is used before being exchanged.

According to Kaplan’s “needles that kill” model (Kaplan & O’Keefe,

1993), the expected contamination rate amongst circulating

syringes can be estimated as c=b?/(b[?+?(1−?)]+s), where b is

the number of borrows per IDU per year, ? is the prevalence of HIV

infection, and ? is the probability that a borrowed syringe is effec-

tively decontaminated. Assuming that ?, ?, and s are unaffected

by injections that occur within the SIF, the HIV contamination

rates with Insite (c1) and without (c0) are primarily determined

by the corresponding number of borrows per IDU per year (b1and

b0). In particular, c0/c1=[(b0?/(b0?+s)]/[(b1?/(b1?+s)], where

?=?+?(1−?).

As described in the preceding section, the number of HIV infec-

tions prevented by the Insite SIF can be estimated using the

equation: A=[(b0/b1)(c0/c1)−1](1−?)Ni1, where i1denotes the

HIV incidence rate with the SIF in operation. We applied this

modelwiththefollowingbase-caseparameters:currentincidence,

i1=1.6% (Health Canada, 2008; Tyndall, Kerr, Lai, et al., 2006);

HIV prevalence, ?=17% (Petrar et al., 2007; Tyndall, Wood, Zhang,

et al., 2006); and probability that a borrowed syringe is effectively

decontaminated prior to injection, ? =42% (Bayoumi & Zaric, 2008).

As above, we assumed that each IDU injects 3 times per day, on

average, and that ˇ=8.3% of these injections were with borrowed

syringes (Pinkerton, 2010). These assumptions yield the following

estimates of the number of borrows per IDU, with and without

the Insite SIF, respectively: b1=(5475–220K)*8.3%/5000=87.233

andb0=(5475K)*8.3%/5000=90.885.Vancouverhasseveralactive

syringe exchange programmes. Insite’s own SEP distributes an

estimated1.3millionsyringeseachyear(Buxton,2007),orapprox-

imately s=(1.3M)/5000=260 syringes per DTES IDU.

Under the base-case conditions described above, the Insite SIF

wouldbeexpectedtopreventapproximately5.2incidentHIVinfec-

tions each year amongst the 5000 IDU living in the DTES area. To

assess the robustness of this result, we conducted univariate sen-

sitivity analyses that varied each of the main parameters across a

plausiblerangeofpotentialvalues.Theresultsoftheseanalysesare

presented in Table 1. The model was most sensitive to the number

of injections per IDU and to the presumed incidence of HIV infec-

tionwiththeInsiteSIFinoperation.Manipulatingtheseparameters

produced a greater than 10% deviation from the base-case result.

Page 5

Please cite this article in press as: Pinkerton, S.D. How many HIV infections are prevented by Vancouver Canada’s supervised injection facility?

International Journal of Drug Policy (2011), doi:10.1016/j.drugpo.2011.03.003

ARTICLE IN PRESS

S.D. Pinkerton / International Journal of Drug Policy xxx (2011) xxx–xxx

GModel

DRUPOL-989;No.of Pages5

5

We also conducted a multivariate Monte Carlo sensitivity anal-

ysis in which a triangular distribution was assumed for each of the

parameterslistedinTable1,withthebaseofthetriangleextending

from the low value considered in the univariate sensitivity analysis

to the high value, and the apex at the base-case value of the param-

eter. The Monte Carlo simulation was run 100,000 times. The mean

number of HIV infections prevented by the SIF across the 100,000

runs was 5.6, with a 90% confidence interval of 4.0–7.6.

Discussion

The target article by Andresen and Boyd (2010) suggests that

the Insite SIF prevents between 19 and 57 incident HIV infections

per year. The above analysis of the impact of the SIF indicates that

it prevents approximately 5–6 infections per year, with a plausible

rangeof4–8preventedinfections.Theseestimatessuggestthatthe

InsiteSIFreducesHIVincidenceamongstDTESIDUbyabout6–11%.

These estimates are conservative inasmuch as they take into

account only the reduction in the number of injections with bor-

rowed “street” syringes amongst uninfected IDU. They do not

account for possible behavioural changes on the part of Insite

clients that could lead them to decrease the rate at which they

inject with borrowed syringes, over and above the reduction due

to utilisation of the SIF. Insite offers risk reduction and safer injec-

tion counselling to all clients, so it is entirely possible that clients

would inject with borrowed syringes less often than non-Insite

clients. Unfortunately, no empirical data are available to quantify

themagnitudeofthisreduction(ifany).Kerr,Tyndall,Li,Montaner,

and Wood (2005) found that Vancouver IDU who reported that

“some, most, or all” of their injections took place at the Insite

SIF were 70% less likely than other IDU to have borrowed or lent

a syringe in the past 6months. However, a 70% reduction in the

prevalence of borrowing or lending does not imply an equivalent

reduction in the borrowing rate (i.e., the proportion of injections

that involve borrowed syringes). Additional research could help

clarify this issue.

Lifetime HIV-related medical care costs are approximately

$210,555 in 2008 Canadian dollars (Pinkerton, 2010). Conse-

quently, by preventing 5–6 HIV infections per year, the Insite SIF

averts more than $1,000,000 in future HIV-related medical care

costs. Andresen and Boyd (2010) estimate that the SIF generates

$660,000 in additional cost savings by preventing 1.08 overdose

deaths per year. The total savings due to averted HIV-related med-

ical care costs and prevented overdose deaths (approximately $1.7

to $1.9 million per year), in and of itself, is just slightly greater than

the estimated $1.5 million annual operating cost of the Insite SIF

(Andresen & Boyd, 2010, based on a radio interview with Thomas

Kerr, Principal Investigator of the Insite research project).

A previous analysis of the Insite SIF indicated that it prevents

approximately 2.8 new HIV infections each year (Pinkerton, 2010).

Unlike the analysis presented above, the previous analysis did

not take into account the reduction in the contamination rate

amongstcirculatingsyringesduetothereducednumberof“street”

injections. Otherwise the two models are formally equivalent. In

particular, the two models produce identical results if the con-

tamination rate ratio, c0/c1, is set to unity in the model described

above.

Other analyseshave indicated

programmes—including but not restricted to the SIF—save society

money in the long run by reducing HIV and hepatitis C-related

morbidity and mortality (Bayoumi & Zaric, 2008; Pinkerton, 2010).

The present analysis supports Andresen and Boyd’s (2010) conclu-

sion that the Insite SIF is cost saving as a stand-alone operation,

independent of Insite’s other programmes.

thatInsite’s various

Acknowledgements

This study was supported by a centre grant (P30-MH52776)

from the National Institute of Mental Health and grant number

R01-AI083035 from the National Institute of Allergy and Infectious

Diseases.

References

Andresen, M. A., & Boyd, N. (2010). A cost-benefit and cost-effectiveness analysis of

Vancouver’s Supervised injection facility. International Journal of Drug Policy, 21,

70–76.

Bayoumi, A. M., & Zaric, G. S. (2008). The cost-effectiveness of Vancouver’s

supervised injection facility. Canadian Medical Association Journal, 179, 1143–

1151.

Buxton, J., & for the Canadian Community Epidemiology Network on Drug

Use. (2007). Vancouver drug use epidemiology: Site Report for the Cana-

dian community epidemiology network on drug use. Vancouver, BC: Canadian

community Epidemiology Network on Drug Use., pp. 1–80, Available at

http://www.city.vancouver.bc.ca/fourpillars/research.htm (15.05.10)

Des Jarlais, D. C., Marmor, M., Paone, D., Titus, S., Shi, Q., Perlis, T., et al. (1996).

HIV incidence among injecting drug users in New York City syringe-exchange

programmes. Lancet, 348, 987–991.

HealthCanada.(2004).I-Track—enhancedsurveillanceofriskbehavioursamonginject-

ing drug users in Canada: Pilot survey report. Health Canada Centre for Infectious

Disease Prevention and Control.

Health Canada. (2008). Vancouver’s INSITE Service and other supervised injec-

tion sites. What has been learned from research? Final report of the Expert

Advisory Committee on Supervised Injection Site Research. Available at www.hc-

sc.gc.ca/ahc-asc/pubs/ sites-lieux/index-eng.php (23.02.10)

Jacobs, P., Calder, P., Taylor, M., Houston, S., Saunders, L. D., & Albert, T. (1999).

Cost effectiveness of streetworkers’ needle exchange program of Edmonton.

Canadian Journal of Public Health, 90, 168–171.

Kaplan, E. H. (1995). Economic analysis of needle exchange. AIDS, 9, 1113–1119.

Kaplan, E. H., & Brandeau, M. L. (1994). AIDS policy modeling by example. AIDS,

8(Suppl. 1), S333–S340.

Kaplan,E.H.,&O’Keefe,E.(1993).Lettheneedlesdothetalking:EvaluatingtheNew

Haven needle exchange. Interfaces, 23, 7–26.

Kerr, T., Tyndall, M. W., Li, K., Montaner, J. S., & Wood, E. (2005). Safer injecting

facility use and syringe sharing among injection drug users. Lancet, 366, 316–

318.

Laufer, F. N. (2001). Cost-effectiveness of syringe exchange as an HIV preven-

tion strategy. Journal of Acquired Immune Deficiency Syndromes, 28, 273–

278.

McCoy, C. B., Metsch, L. R., Chitwood, D. D., Shapshak, P., & Comerford, S. T.

(1998). Parenteral transmission of HIV among injection drug users: Assess-

ing the frequency of multiperson use of needles, syringes, cookers, cotton,

and water. Journal of Acquired Immune Deficiency Syndromes, 18(Suppl. 1), S25–

S29.

McInnes, C. W., Druyts, E., Harvard, S. S., Gilbert, M., Tyndall, M. W., Lima, V. D., et al.

(2009). HIV/AIDS in Vancouver, British Columbia: A growing epidemic. Harm

Reduction Journal, 6, 5.

Normand, J., Vlahov, D., & Moses, L. E. (Eds.). (1995). Preventing HIV transmission: The

role of sterile needles and bleach. Washington, DC: National Academy Press.

Petrar, S., Kerr, T., Tyndall, M. W., Zhang, R., Montaner, J. S. G., & Wood, E. (2007).

Injection drug users’ perceptions regarding use of a medically supervised safer

injecting facility. Addictive Behaviors, 32, 1088–1093.

Pinkerton, S. D. (2010). Is Vancouver Canada’s supervised injection facility cost-

saving? Addiction, 105, 1429–1436.

Remis, R. S., Bruneau, J., & Hankins, C. A. (1998). Enough sterile syringes to prevent

HIV transmission among injection drug users in Montreal? Journal of Acquired

Immune Deficiency Syndromes, 18(Suppl. 1), S57–S59.

Singer, M., Himmelgreen, D., Dushay, R., & Weeks, M. R. (1998). Variation in drug

injection frequency among out-of-treatment drug users in a national sample.

American Journal of Drug and Alcohol Abuse, 24, 321–341.

Tyndall, M. W., Kerr, T., Zhang, R., King, E., Montaner, J. G., & Wood, E. (2005).

Attendance, drug use patterns, and referrals made from North America’s

first supervised injection facility. Drug and Alcohol Dependence, 83, 193–

198.

Tyndall, M., Kerr, T., Lai, C., Shannon, K., Wood, E., & Montaner, J. (2006). Persistently

high rates of HIV transmission among illicit drug users in Vancouver, Canada,

1995–2004. In XVI International AIDS Conference March 13–18, 2006. Abstract:

MOPE0426.

Tyndall, M. W., Kerr, T., Zhang, R., King, E., Montaner, J. G., & Wood, E. (2006).

Attendance, drug use patterns, and referrals made from North America’s first

supervised injection facility. Drug and Alcohol Dependence, 83, 193–198.

Tyndall, M. W., Wood, E., Zhang, R., Lai, C., Montaner, J. S. G., & Kerr, T. (2006). HIV

seroprevalence among participants at a Supervised Injection Facility in Vancou-

ver, Canada: Implications for prevention, care and treatment. Harm Reduction

Journal, 3, 36.

Wood, E., Tyndall, M. W., Montaner, J. S., & Kerr, T. (2006). Summary of findings from

the evaluation of a pilot medically supervised safer injecting facility. Canadian

Medical Association Journal, 175, 1399–1404.