Wang et al.

Harm Reduction Journal 2014, 11:15

http://www.harmreductionjournal.com/content/11/1/15

RESEARCH Open Access

Do community-based strategies reduce HIV risk

among people who inject drugs in China? A
quasi-experimental study in Yunnan and Guangxi
provinces
Kai Wang1*, Hongyun Fu2, Kim Longfield3, Shilpa Modi3, Gary Mundy4 and Rebecca Firestone3

Abstract
Background: HIV transmission among people who inject drugs (PWID) is high in Yunnan and Guangxi provinces in
southwest China. To address this epidemic, Population Services International (PSI) and four cooperating agencies
implemented a comprehensive harm reduction model delivered through community-based drop-incenters (DiC)
and peer-led outreach to reduce HIV risk among PWID.
Methods: We used 2012 behavioral survey data to evaluate the effectiveness of this model for achieving changes
in HIV risk, including never sharing needles or syringes, always keeping a clean needle on hand, HIV testing and
counseling (HTC), and consistent condom use. We used respondent-driven sampling to recruit respondents. We
then used coarsened exact matching (CEM) to match respondents during analysis to improve estimation of the
effects of exposure to both DiC and outreach, only DiC, and only outreach, modeled using multivariable logistic
regression.
Results: We found a significant relationship between participating in both peer-led DiC-based activities and
outreach and having a new needle on hand (odds ratio (OR) 1.53, p < .05) and consistent condom use (OR 3.31,
p < .001). We also found a significant relationship between exposure to DiC activities and outreach and HIV testing
in Kunming (OR 2.92, p < .01) and exposure to peer-led outreach and HIV testing through referrals in Gejiu, Nanning,
and Luzhai (OR 3.63, p < .05).
Conclusions: A comprehensive harm reduction model delivered through peer-led and community-based strategies
reduced HIV risk among PWID in China. Both DiC activities and outreach were effective in providing PWID behavior
change communications (BCC) and HTC. HTC is best offered in settings like DiCs, where there is privacy for testing
and receiving results. Outreach coverage was low, especially in Guangxi province where the implementation model
required building the technical capacity of government partners and grassroot organizations. Outreach appears to
be most effective for referring PWID into HTC, especially when DiC-based HTC is not available and increasing
awareness of DiCs where PWID can receive more intensive BCC interventions.
Keywords: Social marketing, Drop-in center, Community-based outreach, Behavior change communication (BCC),
People who inject drugs (PWID), Respondent-driven sampling (RDS), Needle-syringe exchange program (NSP),
Condom use, HIV testing and counseling (HTC), Coarsened exact matching (CEM)

* Correspondence: uangkai@gmail.com
1
PSI/China, 909-9 F, M2 Building, Harmonious Society, Xiaokang Rd,
Hongyun Community, Wuhua District, Kunming 650000, China
Full list of author information is available at the end of the article

© 2014 Wang et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative
Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly credited.
Wang et al. Harm Reduction Journal 2014, 11:15 Page 2 of 9
http://www.harmreductionjournal.com/content/11/1/15

Background strategies to improve HIV risk behaviors. Evaluations of

In 1989, China reported its first cases of HIV among 146 these programs have examined process measures for ser-
heroin users in Yunnan Province, along China's southwest vice quality and have demonstrated expanded population
border [1]. Yunnan is located near the Golden Triangle, coverage [10-12], and recent evidence is available that
consisting of Thailand, Myanmar, and Laos. The Golden HIV and hepatitis B incidence may decline among PWID
Triangle is one of the three largest heroin production sites enrolled in multipronged harm reduction programs [2].
in the world. As a result, HIV spread quickly along drug More direct evidence on program effectiveness and a link
trafficking routes in Yunnan and into neighboring prov- to behavior change is still needed.
inces [2].
According to an assessment by the World Health The comprehensive prevention package in southwestern
Organization, Joint United Nations Programme on HIV/ China
AIDS, and the Chinese Ministry of Health, China's over- Since 2009, Population Services International (PSI) has
all HIV prevalence has remained relatively low between implemented a comprehensive prevention package (CPP)
0.5%–0.6% in 2011 [3,4]. However, the national preva- for PWID in Yunnan and Guangxi provinces, in collabor-
lence of HIV among people who inject drugs (PWID) ation with a consortium of partners, including FHI 360,
was 9.08% in 2010, with prevalence higher than 50% in the International HIV/AIDS Alliance, Research Triangle
parts of southwest China [3,4]. The provinces of Yunnan Institute, Pact, and local government partners with sup-
and Guangxi have two of the highest concentrations of port from the United States Agency for International
HIV cases, as well as registered drug users in China. Development (USAID) [13,14]. The CPP program is a
Yunnan and Guangxi account for 22% of new HIV cases community-based strategy, delivered through local part-
in the country, despite making up only 6.9% of the na- ners, that helps PWID access and use a core set of in-
tional population [5]. In Guangxi, there are approxi- terventions to reduce HIV infection. In Yunnan and
mately 50,000 registered PWID, with injecting drug use Guangxi, PWID access CPP interventions through drop-
accounting for over 69% of the cumulative reported in centers (DiC) and outreach implemented by PSI's peer
cases of HIV in the province [6]. educators, PSI-supported community-based organizations
Transmission of HIV through unsafe injecting prac- (CBOs), and other local partners. The following set of
tices, including sharing of needles and syringes, has CPP interventions (Figure 1) was provided through DiC
remained high since the early stages of the epidemic in and outreach: (1) peer-led behavior change communica-
China. At the national level, prevalence of ever sharing tions and education to promote condom use and prevent
injecting equipment among PWID was 25% in 2011 [7]. STIs, HIV, hepatitis B and C, and tuberculosis; (2)
Of the 780,000 people in China estimated to be living provision of condoms and lubricants; (3) referrals for anti-
with HIV (PLHIV) in 2011, 28.4% were infected through retroviral therapy (ARV) treatment, MMT, and community
injecting drug use [2]. rehabilitation; (4) needle-syringe exchange and education
In 2004, in response to the growing HIV epidemic,
China implemented methadone maintenance therapy
(MMT), needle-syringe exchange programs, and condom
promotion at the national and local levels [2]. Prior to this,
China's approach to working with PWID was largely puni-
tive. Drug users were sent to compulsory detoxification
programs and re-education through labor (RTL) centers
[8]. In 2008, China passed the Narcotic Control Law,
which supports community-based rehabilitation for PWID
rather than sending PWID to compulsory detoxification
programs and RTL centers [9]. Although recent policies
have created a more supportive environment for PWID, a
number of barriers still exist for PWID trying to access
HIV testing and counseling (HTC) and harm reduction
services, including those for preventing sexual transmis-
sion of HIV [10,11]. These barriers include community
stigma and discrimination, criminalization of illicit drug
use, and low coverage of existing programs [9].
Despite the establishment of a more supportive environ-
Figure 1 USAID comprehensive prevention package (CPP)
ment, limited evidence is available on the ability of the
model.
current network of harm reduction and HIV prevention
Wang et al. Harm Reduction Journal 2014, 11:15 Page 3 of 9
http://www.harmreductionjournal.com/content/11/1/15

regarding safe injecting practices and overdose prevention; sampling method used to recruit hard-to-reach popula-
and (5) promotion of ARV and MMT adherence. Addition- tions, including PWID [15-17]. RDS relies on the as-
ally, project partners conducted a set of essential ‘enabling sumption that, given sufficiently long referral chains (i.e.,
environment’ interventions that aimed to improve policies 3–6 waves of respondents), the sample composition be-
through advocacy, reduce stigma and discrimination, sup- comes stable or reaches ‘equilibrium’, which results in a
port community mobilization, build capacity of local orga- sample that has the characteristics of a probability sam-
nizations, support income generation for PWID, and use ple. We then used sampling weights to account for the
strategic information for decision-making. social networks used in chain referral recruiting.
In Yunnan province, PSI implemented the Huxianghao We started by recruiting 4–8 seeds in each study city.
(HXH, ‘Good for you good for me’) program in Kunming We gave each seed three coupons and asked him/her to
city, which included DiC and outreach activities to de- recruit three peers from his/her social network to partici-
liver a wide range of services for PWID. PSI also pro- pate in the study. We diversified seeds based on age, sex,
moted the replication of the established HXH program exposure to program interventions, current MMT use,
model in Kunming through technical assistance (TA) to and current injecting status (current or former PWID).
community rehabilitation programs. In Gejiu city, PSI The recruitment coupons were uniquely numbered, which
and project partners provided TA and funding support allowed us to link seeds to their referrals and track the
to PWID-led community-based organizations to conduct length of recruitment chains. Participants included in the
DiC and outreach activities. study were those who were aged 18 to 49, intravenous
In Guangxi province, PSI-implemented outreach activ- drug users in the past 12 months, residing in one of the
ities in Nanning city and provided TA to the local CDC, study cities for a minimum of 1 month, able to speak and
community rehabilitation programs, and community- comprehend Mandarin Chinese enough to respond to sur-
based organizations in Nanning and Luzhai City to con- vey questions, and not under the influence of drugs or al-
duct DiC and outreach activities. Project partners also cohol at the time of the survey.
supported community-based organizations by working We collected data through face-to-face interviews in
on MMT and ARV adherence with PWID and with the locations where respondents' privacy could be protected.
Luzhai CDC to conduct outreach. In Yunnan, the DiC The questionnaire covered several topics, including
and outreach activities included HTC, whereas in demographic characteristics, injecting behaviors, sexual
Guangxi, clients were referred to hospital or government activity, condom use, STI testing, HIV testing, and ex-
centers for testing. posure to program activities. The Yunnan Institution for
Drug Abuse Institutional Review Board provided ethical
Objective review and approval. We obtained informed consent be-
Following several years of program implementation, we fore conducting interviews. All interviewers and others
aimed to evaluate the effectiveness of this comprehen- associated with the study completed training on human
sive harm reduction model delivered through partici- subjects' protection.
pation in community-based DiCs and interpersonal
outreach for improving HIV risk behavior among PWID Measures
in southwestern China. Given the nature of the program, We assessed four outcomes of interest, all treated as binary
we considered harm reduction practices (never sharing variables: not sharing needles or syringes in the past
needles or syringes, always keeping a clean needle on 3 months, keeping a new needle on hand in the past
hand), consistent condom use with different types of 3 months, having received an HIV test in the past
sexual partners (including regular, casual, and commer- 12 months, and consistent condom use with any type of
cial partners), and use of HTC. Our evaluation is not de- sexual partner in the past 3 months. We created several
signed to compare implementation models but rather variables to assess exposure to the program model through
assess the effect of the overall package of CPP interven- its different delivery channels: participation in any pro-
tions on HIV risk among PWID. gram activity whether DiC-based or outreach in the past
12 months, participation in both DiC activities and out-
Methods reach in the past 12 months, participation in only DiC ac-
Study population and sampling tivities in the past 12 months, and participation in only
From March to April 2012, we used respondent-driven outreach in the past 12 months. Each variable was coded
sampling (RDS) to recruit 1,035 PWID from four cities as a dichotomous measure, indicating that some respon-
in Yunnan and Guangxi provinces for a behavioral sur- dents were not exposed to any program channel. Other
vey. Sample sizes were as follows: Kunming (n = 336) variables within the analysis included: age, sex, working
and Gejiu (n = 204) in Yunnan and Nanning (n = 355) hours, city of residence, ever use of MMT, ethnicity (Han
and Luzhai (n = 140) in Guangxi. RDS is a chain-referral Chinese or not), education (high school and higher vs. less
Wang et al. Harm Reduction Journal 2014, 11:15 Page 4 of 9
http://www.harmreductionjournal.com/content/11/1/15

than high school), and number of sex acts in the past in MMT centers, PWID who ever used MMT services
3 months. would have had greater opportunity to access outreach
and know about DiCs. This match yielded an L1 distance
Analysis of 2.83E−16, indicating that the matched subsample had
This analysis was intended to be representative of PWID minimal imbalance between those exposed and not ex-
in Yunnan and Guangxi. We used RDSAT to estimate posed to the program.
sampling weights within each city [18]. We then pooled We calculated descriptive statistics for both the full,
the data across all cities and calculated weights for each pooled dataset, and the matched sample. We applied
city based on population size estimates provided by pro- combined RDS and city weights to analyze pooled data.
ject partners. These weights were used in all subsequent For analysis of the matched subsample, we applied one
analyses. combined weight derived from CEM weights, RDS
Because we aimed to test the effectiveness of the CPP weights, and city weights. We then created bivariate and
model, we next used a matching technique, coarsened multivariate logistic regression models, controlling for
exact matching (CEM), to create statistically equivalent ethnicity and education. After assessing the effects of
groups of exposed and nonexposed respondents. This any program participation (not shown), we tested the
quasi-experimental approach allowed us to designate a isolated effects of participation in both DiC and out-
counterfactual (no participation in either DiC or outreach reach, participation in DiC only and participation in out-
activities) when an experimental design was not feasible, reach only. Separate logistic regression models were
given that program implementation was ongoing and was constructed for each treatment variable to isolate the ef-
not designed or implemented with intervention and con- fects of each channel. All analyses were conducted in
trol populations [19-21]. Coarsened exact matching is a Stata 11.
monotonic imbalance matching method designed to re-
duce imbalance between treatment and control groups in
observational data [22]. CEM assigns each case into one of Results
a specified set of strata in which members are exactly Population characteristics
matched on a set of coarsened, or categorized, variables. Table 1 shows the pooled and matched estimate for
Matched cases are then assigned a weight specific to that population characteristics; we only discuss matched
stratum and representative of the proportion of all cases results henceforth. The majority of respondents were
present in the stratum [19-21,23]. We chose CEM over male (67.4%), over 40 years old (51.6%), of Han ethnicity
other matching techniques, such as propensity score (83.3%), and had less than a high school education
matching, to achieve balanced groups, reduce the need for (68.9%). Only 31.1% of respondents were married: the
multiple iterations and re-matching, and maximize the majority had never been married or were widowed or
number of possible matches in our sample [22-24].
In the pooled dataset, we matched survey respondents Table 1 Population characteristics of people who inject
on age (continuous variable coarsened to <40 years, ≥40), drugs in four southwestern Chinese cities in 2012
sex, working hours (coded as end work before 5 p.m./start Population characteristics Pooled Matched
work after 5 p.m./not working vs. all other working hours), sample Sample
resident city, and ever used MMT. We selected these five (n = 1,035) (n = 975)
variables after consultation with program staff because (%) (%)
they were identified as having a substantial influence on Sex (male) 64.0 67.4
PWID's likelihood of participation in the interventions. Age (in years)
Controls were thus defined as people who did not partici- 18 to 30 4.8 7.2
pate in either DiC or outreach, but who were identified,
31 to 40 32.9 41.2
through matching, as having a similar probability of par-
41 to 49 62.3 51.6
ticipation as those who participated in any program activ-
ity. During routine data collection for the program, we Ethnicity (Han) 87.3 83.3
found that older people were more likely to use DiC ser- Education (High school education or above) 32.5 31.1
vices than youths. Peer educators reported that males were Marital status
more likely than females to use DiCs and MMT services. Never married 38.4 42.6
Respondents' working hours could affect their ability to
Currently married 29.9 31.1
visit DiCs during opening hours or participate in outreach.
Divorced or widowed 31.7 26.3
We matched on resident city account for variations in the
political environmental towards drug use and harm reduc- Ever received methadone maintenance 84.7 87.9
treatment (MMT)
tion in each city. Since peer educators conducted outreach
Wang et al. Harm Reduction Journal 2014, 11:15 Page 5 of 9
http://www.harmreductionjournal.com/content/11/1/15

divorced. The use of MMT was very high: nearly 88% of Luzhai (12.9%). CPP staff report that low levels of out-
respondents had received MMT sometime in the past. reach in Nanning are explained by the size of the city
and that there are few qualified outreach workers to
Behavioral outcomes and exposure to DiC and outreach reach the PWID community. The model for outreach is
We present behavioral outcomes and program exposure also different in Guangxi than in Kunming; CPP staff fo-
in Table 2. More than 80% of respondents had injected cused more on TA to build the capacity of community
heroin in the past 3 months. Reported rates for needle rehabilitation centers, government partners, and grass-
and syringe sharing were only 6.2%, much lower than root organizations to conduct outreach. This is a slower
rates reported in the literature. However, less than a model to roll out because on-the-job training is inten-
third (32.3%) of respondents reported keeping clean nee- sive, and it requires a great deal of supervision. The out-
dles on hand. The majority of the respondents reported reach model in Guangxi also required more consensus
having an HIV test during the past year (71.6%). More building and agreement on program goals than the
than half of the respondents reported having sex in the Kunming model, which takes time and can slow down
past 3 months (54.6%), with an average of 13 sex acts program execution. The lack of correlation between ex-
over the course of 3 months. Among those who were posure to outreach and some of the behavioral outcomes
sexually active, 49.3% reported consistent condom use may be attributable to low levels of coverage in places
with all partners during the same period. like Nanning and Luzhai.
Just over half of the respondents reported attending ei- Tables 3, 4, and 5 present results for the pooled and
ther DiC-based activities or receiving outreach during matched samples and show the correlation between pro-
the past 12 months (51.0%). Participation in DiC-based gram exposure and the behavioral outcomes. We present
activities was higher than outreach (43.3% vs. 26.1%). bivariate and multivariate analyses to demonstrate how
About one fifth of the respondents reported attending the models were constructed, but we only focus on the
both DiC-based activities and receiving outreach during matched multivariate analyses here.
the past 12 months (18.4%). Exposure to DiC activities
was the highest in Gejiu (75.5%), approximately 48% for Effectiveness of DiC and outreach on safer injecting practices
both Luzhai and Kunming, and much lower in Nanning Table 3 presents associations between program exposure
(16.9%). Exposure to outreach was highest in Kunming and safer injecting practices. There was a statistically
(36.3%), followed by Gejiu (27.9%), Nanning (18.6%), and significant relationship between participating in both

Table 2 Behavioral outcomes and exposure to drop-in center and outreach activities
Kunming Gejiu Nanning Luzhai Pooled Matched
(n = 336) (n = 204) (n = 355) (n = 140) sample Sample
(n = 1,035) (n = 975)
% % % % % %
Behavioral outcomes
Injected heroin in the past 3 months 72.6 81.4 83.9 81.4 75.8 80.2
Ever shared needles/syringes with others in the past 3 months (among 5.0 0.6 8.6 10.0 5.2 6.2
those who used heroin in the past 3 months)
(n = 244) (n = 166) (n = 298) (n = 114) (n = 822) (n = 762)
Always kept a new needle on hand during the past 3 months (among 26.2 38.8 24.8 29.4 28.0 32.3
those who used heroin in the past 3 months)
Received an HIV test in the past 12 months 67.6 75.9 52.1 66.5 66.5 71.6
Had sexual intercourse in the past 3 months 61.6 46.1 52.4 55.0 57.8 54.6
Mean number of sex acts in the past 3 months 13 10 17 19 14 13
(n = 207) (n = 94) (n = 186) (n = 77) (n = 564) (n = 500)
Used condoms consistently with all partners in the past 3 months (among 38.5 67.0 30.2 45.6 41.4 49.3
those who had sexual intercourse in the past 3 months)
(n = 207) (n = 94) (n = 186) (n = 77) (n = 564) (n = 500)
Program exposure
Participated in either DiC activities or outreach in the past 12 months 58.3 78.4 25.6 51.4 55.9 51.0
Participated in DiC activities and outreach in the past 12 months 25.9 25.0 9.9 10.0 22.8 18.4
Participated in DiC activities only in the past 12 months 22.0 50.5 7.0 38.6 24.2 24.9
Participated in outreach only in the past 12 months 10.4 2.9 8.7 2.9 8.9 7.7
Wang et al. Harm Reduction Journal 2014, 11:15 Page 6 of 9
http://www.harmreductionjournal.com/content/11/1/15

Table 3 Effectiveness of exposure to drop-in center and outreach activities on safer injecting practices
Never shared needles or syringes in the past Always kept a new needle on hand in the past
3 months 3 months
Program exposure Pooled sample (n = 823) Matched sample (n = 762) Pooled sample (n = 823) Matched sample (n = 762)
Bivariate Multivariate Bivariate Multivariate Bivariate Multivariate Bivariate Multivariate
OR Adjusted OR Adjusted OR Adjusted OR Adjusted
(95% CI) OR (95% CI) (95% CI) OR (95% CI) (95% CI) OR (95% CI) (95% CI) OR (95% CI)
Participated in DiC activities and 1.37 1.32 1.41 1.40 1.50* 1.49* 1.53* 1.53*
outreach in the past 12 months (0.61 to 3.09) (0.59 to 2.99) (0.58 to 3.42) (0.57 to 3.40) (1.05 to 2.14) (1.05 to 2.14) (1.07 to 2.19) (1.07 to 2.19)
Participated in DiC activities 0.80 0.80 0.53 0.55 1.01 1.03 1.27 1.28
only in the past 12 months (0.40 to 1.61) (0.40 to 1.62) (0.26 to 1.09) (0.27 to 1.13) (0.71 to 1.45) (0.71 to 1.47) (0.87 to 1.83) (0.88 to 1.85)
Participated in outreach only 0.53 0.56 0.48 0.53 1.07 1.08 1.03 1.06
in the past 12 months (0.22 to 1.33) (0.22 to 1.39) (0.19 to 1.23) (0.21 to 1.37) (0.62 to 1.85) (0.62 to 1.88) (0.59 to 1.79) (0.61 to 1.86)
*Significant at p < 0.05. Age, gender, respondents' working hours, resident city, and ever use of MMT were matched for the CEM model. Covariates were ethnicity
and education in the separated logistic regression models but not reported here.

DiC-based activities and outreach and having a new nee- different in the intervention sites, we segmented the ana-
dle on hand (odds ratio (OR) 1.53, p < .05). The relation- lysis by Kunming and the other three cities. In Kunming,
ship between exposure to only DiC or only outreach and the program has offered rapid HIV testing in the DiC
either of the two safer injecting practices was not statis- since 2005 and through outreach since 2009. In both set-
tically significant. There was also no relationship be- tings, test results are available in 10 min. In this analysis,
tween both DiC-based activities and outreach and we only found a statistically significant relationship be-
needle or syringe sharing, which was not surprising tween exposure to both DiC-related activities and out-
given the low levels of reported overall sharing (6.2%). reach and testing (OR 2.92, p < .01). There was no
There are several contextual factors that help explain association between exposure to only DiC in Kunming
these findings. The positive relationship between expos- and testing. There was a negative association between ex-
ure to both DiC-related activities and outreach and hav- posure to only outreach in Kunming and testing. An ex-
ing new injecting equipment on hand is likely due to planation for this may be that if referrals from outreach to
behavior change communication (BCC) in DiCs and out- DIC were specifically for testing, then those that did not
reach, especially in 2011 and 2012. There were also attend the DIC may have made a decision not to test.
needle-syringe exchange programs in the Kunming and Similarly, those that used the DIC but without referral
Geiju DiCs and the Nanning CDC, which means that in from an outreach worker may have used the facility for
some cases, BCC is supported by having new injecting services other than testing. Those that used the DIC as a
equipment to give to PWIDs. result of outreach contact, however, may have done so
specifically to be tested for HIV, having been referred to
Effectiveness of DiC and outreach on HIV testing the service by the outreach worker.
Table 4 presents associations between program exposure In Gejiu, Nanning, and Luzhai, only one DiC in Gejiu
and HIV testing. Because the model for testing was offered HIV testing. No immediate HIV testing was

Table 4 Effectiveness of exposure to drop-in center and outreach activities on HIV testing
Tested for HIV in the past 12 months Tested for HIV in the past 12 months
(Kunming) (Gejiu, Nanning, and Luzhai)
Program exposure Pooled sample (n = 335) Matched sample (n = 324) Pooled sample (n = 699) Matched sample (n = 650)
Bivariate Multivariate Bivariate Multivariate Bivariate Multivariate Bivariate Multivariate
OR Adjusted OR Adjusted OR Adjusted OR Adjusted
(95% CI) OR (95% CI) (95% CI) OR (95% CI) (95% CI) OR (95% CI) (95% CI) OR (95% CI)
Participated in DiC activities 3.51*** 3.50*** 2.79** 2.92** 2.96*** 2.74*** 1.82* 1.73
and outreach in the past (1.80 to 6.85) (1.79 to 6.86) (1.42 to 5.47) (1.47 to 5.78) (1.74 to 5.01) (1.67 to 4.86) (1.06 to 3.14) (1.00 to 2.99)
12 months
Participated in DiC activities 1.44 1.48 1.08 1.10 2.03*** 2.17*** 1.05 1.10
only in the past 12 months (0.82 to 2.54) (0.84 to 2.62) (0.61 to 1.93) (0.62 to 1.98) (1.39 to 2.95) (1.48 to 3.19) (0.71 to 1.56) (0.73 to 1.62)
Participated in outreach only 0.53 0.56 0.44* 0.47* 5.80** 5.78** 3.64* 3.63*
in the past 12 months (0.27 to 1.06) (0.28 to 1.12) (0.22 to 0.88) (0.23 to 0.95) (1.88 to 17.86) (1.87 to 17.89) (1.18 to 11.28) (1.17 to 11.28)
*Significant at p < 0.05; **significant at p < 0.01; ***significant at p < 0.001. Age, gender, respondents' working hours, resident city, and ever use of MMT were
matched for the CEM model. Covariates were ethnicity and education in the separated logistic regression models but not reported here.
Wang et al. Harm Reduction Journal 2014, 11:15 Page 7 of 9
http://www.harmreductionjournal.com/content/11/1/15

Table 5 Effectiveness of exposure to drop-in center and outreach activities on consistent condom use
Consistent condom use with all partners in the past 3 months
Program exposure Pooled sample (n = 564) Matched sample (n = 500)
Bivariate Multivariate Bivariate Multivariate
OR (95% CI) Adjusted OR (95% CI) OR (95% CI) Adjusted OR (95% CI)
Participated in DiC activities and outreach in the past 2.25*** (1.53 to 3.31) 3.32*** (1.55 to 3.48) 2.00*** (1.35 to 2.97) 3.31*** (1.52 to 3.50)
12 months
Participated in DiC activities only in the past 12 months 1.47* (1.02 to 2.14) 1.43 (0.98 to 2.10) 1.58 (1.06 to 2.33) 1.65* (1.10 to 2.47)
Participated in outreach only in the past 12 months 0.76 (0.42 to 1.39) 0.86 (0.46 to 1.60) 0.83 (0.43 to 1.58) 0.93 (0.48 to 1.80)
*Significant at p < 0.05; ***significant at p < 0.001. Age, gender, respondents' working hours, resident city, and ever use of MMT were matched for the CEM model.
Covariates were ethnicity and education in the separated logistic regression models but not reported here.

offered during outreach activities; referrals were given influence just not injecting practices but the use of HTC
for testing in hospitals and government testing sites. We along with condoms to prevent sexual transmission.
found a statistically significant relationship between ex-
posure to only outreach and testing in these three cities Limitations
(OR 3.63, p < .05). This model of only outreach with re- Our study has several limitations. Study respondents self-
ferrals for testing appears to be working and an appro- reported HIV risk behaviors, and results are subject to so-
priate strategy when testing is not available in DiCs or in cial desirability bias. A limitation of RDS was that it was
areas where there are no DiCs. However, participation in not possible to identify the nonresponse rate in the sample
DiC-based activities in Gejiu, Nanning, and Luzhai was [17,27,28]. For example, we do not know how many at-
not associated with referrals for testing, which suggests tempts were made before three coupons were distributed
that DiC staff did not sufficiently refer PWID to public successfully within PWID social networks. If particular
hospitals or government testing sites. profiles of PWID were more likely to refuse study partici-
pation, there may have been sampling bias. The other limi-
Effectiveness of DiC and outreach on consistent condom use tation of RDS is that although it has features of probability
Table 5 shows associations between program exposure sampling, we could not estimate the extent to which it
and consistent condom use. There was a statistically sig- was truly representative of PWID across the four study cit-
nificant relationship between participating both DiC ac- ies. However, we calculated and applied city weights to im-
tivities and outreach and consistent condom use (OR prove representativeness of the sample. All successful
3.31, p < 0.001), and a statistically significant relationship recruitment waves were between 5 and 12, which suggests
between participating in only DiC-based activities and that the study achieved equilibrium [15,29].
consistent condom use (OR 1.65, p < .05), likely ex- A further limitation is that we may not have fully
plained by the factors noted previously: intensive BCC accounted for factors that influenced selection into the
and the option of speaking with CPP staff in a private CPP program. We attempted to address risks of selection
setting. There was no association between exposure to bias through matching on age, sex, working hours, city resi-
only outreach and consistent condom use. These find- dence, and use of MMT. However, these findings still face
ings indicate that there is a subset of PWID at increased the possibility of omitted variable bias. We were not able to
risk for HIV infection through unprotected sex. account for other factors such as distance to DiCs or indi-
These findings are in line with evidence from else- vidual propensities to participate in health-promoting activ-
where in China that comprehensive, community-based ities. Finally, with the matching technique we used, we may
harm reduction programs can contribute to HIV preven- have increased the number of pairs to be matched at the
tion when designed with the needs of PWID in mind. expense of less exact matching. This is less of a concern
Social marketing of needles in Guangdong and Guangxi with categorical variables than continuous variables. We
has been demonstrated to increase access to clean nee- coarsened continuous variables based on programmatically
dles and safe injection practices [25]. HIV prevalence meaningful categories, and diagnostics of the matching
along the China-Vietnam border declined in the context procedure indicated a high quality match.
of peer outreach and distribution of safe injection equip-
ment [26]. Similarly, HIV and hepatitis B incidences were Conclusion
found to be on the decline among a cohort of PWID in Our analysis demonstrates that comprehensive peer-led
Sichuan providence who received community-based harm programs reduced HIV risk among PWID. Overall, both
reduction programs [2]. Findings from this study suggest DiC-based activities and outreach were successful at in-
how delivery channels for a comprehensive program can creasing safe injecting practices, HTC, and consistent
Wang et al. Harm Reduction Journal 2014, 11:15 Page 8 of 9
http://www.harmreductionjournal.com/content/11/1/15

condom use. Outreach coverage was low, especially in Received: 27 May 2013 Accepted: 10 April 2014
Guangxi province where the implementation model re- Published: 6 May 2014

quired building the technical capacity of government part-

ners and grassroots organizations. References
1. Jia M, Luo H, Ma Y, Wang N, Smith K, Mei J, Lu R, Lu J, Fu L, Zhang Q: The
The findings raise questions about the independent HIV epidemic in Yunnan province, China, 1989–2007. J Acquir Immune
and combined effects of different intervention activities Defic Syndr 2010, 53(1):S34–S40.
for addressing HIV risk behaviors among PWID. While 2. Ruan Y, Liang S, Zhu J, Li X, Pan SW, Liu Q, Song B, Wang Q, Xing H, Shao Y:
Evaluation of harm reduction programs on seroincidence of HIV,
a combination of outreach and DIC was more com- hepatitis B and C, and syphilis among intravenous drug users in
monly found to have an association with reduced risk southwest China. Sex Transm Dis 2013, 40(4):323–328.
behaviors, it was notable that, in areas where DICs did 3. Ministry of Health of the People’s Republic of China: China AIDS response
progress report. http://www.unaids.org/en/dataanalysis/knowyourresponse/
not offer testing services, outreach alone was strongly countryprogressreports/2012countries/ce_CN_Narrative_Report[1].pdf.
associated with an increased likelihood of testing. This 4. Zhang L, Chow EP, Jing J, Zhuang X, Li X, He M, Sun H, Li X, Gorgens M,
was, presumably, a result of referral to local testing ser- Wilson D, Wang L, Guo W, Li D, Cui Y, Wang L, Wang N, Wu Z, Wilson D:
HIV prevalence in China: integration of surveillance data and a
vices that were not available at DICs. This suggests that systematic review. Lancet Infect Dis 2013, 13(11):955–963.
outreach in itself may have an important role to play in 5. Xu J, Smith MK, Ding G, Chu J, Wang H, Li Q, Chang D, Wang G, Shang H,
increasing testing behavior when accessible testing ser- Jiang Y, Wang N: Drug use and sex work: competing risk factors for
newly acquired HIV in Yunnan, China. PloS one 2013, 8(3):e59050.
vices are available, regardless of whether those services 6. Wei L, Chen J, Rodolph M, Beauchamp G, Masse B, Li R, Wang S, Ruan Y, Liu
are themselves ‘community based’, i.e., run from drop-in S, Zhang L, Zhou F, Rose SM, Perdue T, Lai S, Shao Y, Jackson JB: HIV
centers or other locations providing testing targeted at incidence, retention, and changes of high-risk behaviors among rural
injection drug users in Guangxi China. Subst Abus 2007, 27(4):53–61.
PWID. These strategies require further expansion to 7. Ministry of Health of People's Republic of China, UNAIDS, WHO: Estimates
fully address the needs of PWID in southwestern China. for the HIV/AIDS epidemic in China. 2011. http://www.unaids.org.cn/pics/
20130521161757.pdf.
Competing interests 8. Qian HZ, Schumacher JE, Chen HT, Ruan YH: Injection drug use and HIV/
The authors declare that they have no competing interests. AIDS in China: review of current situation, prevention and policy
implications. Harm Reduct J 2006, 3:4.
Authors’ contributions 9. Li J, Ha T, Zhang C, Liu H: The Chinese government's response to drug
KW designed the study, managed fieldwork, analyzed data and created use and HIV/AIDS: a review of policies and programs. Harm Reduct J
tables, and wrote portions of the manuscript. HY designed the study, 2010, 7(1):1–6.
analyzed data, and helped outline the manuscript. KL and SM wrote and 10. Rou K, Sullivan SG, Liu P, Wu Z: Scaling up prevention programmes to
edited the manuscript. GM helped with study design, data analysis, and reduce the sexual transmission of HIV in China. Int J Epidemiol 2010,
reviewed the manuscript. RF helped with data analysis and wrote sections of 39(2):ii38–ii46.
the manuscript. All authors read and approved the final manuscript. 11. Yin W, Hao Y, Sun X, Gong X, Li F, Li J, Rou K, Sullivan SG, Wang C, Cao X:
Scaling up the national methadone maintenance treatment program in
Acknowledgements China: achievements and challenges. Int J Epidemiol 2010, 39(2):ii29–ii37.
USAID funded this study under the CAP-3D project. Several colleagues at PSI 12. Liu B, Sullivan SG, Wu Z: An evaluation of needle exchange programmes
provided support for this study, including Andrew Miller (study conception, in China. AIDS 2007, 21(8):S123–S128.
manuscript review), Shaojie Shi, Jing Wang, and Bing Zhao (study design, 13. USAID Health Policy Initiatives: China: leading the way - a comprehensive
interviewer recruitment and training, and fieldwork). Shaojie Shi, Amy approach to HIV prevention, care, and treatment. http://www.healthpolicyinitiative.
Gregowsky, and Lung Vu helped with data analysis. Lung Vu also provided com/Publications/Documents/1470_1_ComprehensiveApproach_07_14_11_
comments on early drafts of the manuscript. We would like to acknowledge r12.pdf.
the other cooperating agencies that conducted programming activities 14. PEPFAR: China operational plan report FY 2010. In Office of U.S. Global
under the CPP model: Pact, Research Triangle Institute (RTI), Family Health AIDS Coordinator. 2010.
International (FHI 360), and the International HIV/AIDS Alliance. In addition, 15. Johnston LG, Chen YH, Silva-Santisteban A, Raymond HF: An empirical
Shanthi Noriega (FHI360) assisted with the study design, and Matt Avery examination of respondent driven sampling design effects among HIV
(FHI360) helped with field training. Jianhua Li at the Yunnan Institute of Drug risk groups from studies conducted around the world. AIDS Behav 2013,
Abuse facilitated the IRB process. The following individuals are thanked for 17(6):2202–2210.
their role in fieldwork logistics and management: Rong Zhao, DajiMin, 16. Heckathorn DD: Respondent-driven sampling II: deriving valid population
Tingting Huang, Ling Li, Rong Liu, Qiu Li, Shuzhong Qiu, Hong Zhang, estimates from chain-referral samples of hidden populations. Soc Probl
Ruiwei Han, and Jiafeng Yan. Finally, graduate students from Yunnan 2002, 49(1):11–34.
Nationality University conducted interviews and data entry: Yicheng Wang, 17. Salganik MJ, Heckathorn DD: Sampling and estimation in hidden
Jinlong Tian, Jian Sun, Pengjie Yan, Hongjun Wu, Jianfeng Tian, Lifa Tan, Yu populations using respondent‐driven sampling. Sociol Methodo 2004,
Zeng, Chunchun Zhangqu, and Haiqing Zhong. Graduate students from 34(1):193–240.
Guangxi Medical University did the same: Longxiang Wei, Zijuan Meng, 18. Michael W, Spiller CC, Heckathorn D: RDSTAT: RDS analysis tool 7.1 user
Yanfei Wang, Lilin Qin, Meihong Yao, Rongjun Zhou, Yiping Yang, Yun manual. Ithaca, NY: Cornell University; 2012.
Ouyang, and Jian Zhang. 19. Stuart EA: Matching methods for causal inference: a review and a look
forward. Stat Sci 2010, 25(1):1–21.
Funding 20. Wells AR, Hamar B, Bradley C, Gandy WM, Harrison PL, Sidney JA, Coberley
USAID funded this study under the CAP-3D project. CR, Rula EY, Pope JE: Exploring robust methods for evaluating treatment
and comparison groups in chronic care management programs.
Author details Popul Health Manag 2013, 16(1):35–45.
1
PSI/China, 909-9 F, M2 Building, Harmonious Society, Xiaokang Rd, Hongyun 21. Shadish W, Cook T, Campbel D: Experimental and quasi-experimental designs
Community, Wuhua District, Kunming 650000, China. 2PSI/China, 12711 Misty for generalized causal inference. Boston: Houghton Mifflin; 2002.
Creek Drive, Little Rock, AR 72211, USA. 3PSI, 1120 19th Street NW, Suite 600, 22. King G, Nielsen R, Coberley C, Pope JE, Wells A: Comparative effectiveness of
Washington, DC 20036, USA. 4PSI Asia Region, Unit 12A, 12th Floor, Q House matching methods for causal inference. 2011. http://gking.harvard.edu/files/
Convent Building, 38 Convent Road, Silom, Bangrak, Bangkok 10500, Thailand. gking/files/psparadox.pdf.
Wang et al. Harm Reduction Journal 2014, 11:15 Page 9 of 9
http://www.harmreductionjournal.com/content/11/1/15

23. Blackwell M, Iacus S, King G, Porro G: CEM: coarsened exact matching in

stata. Stata J 2009, 9(4):524–546.
24. Haider AH, David JS, Zafar SN, Gueugniaud PY, Efron DT, Floccard B,
Mackenzie EJ, Voiglio E: Comparative effectiveness of inhospital trauma
resuscitation at a French trauma center and matched patients treated in
the United States. Ann Surg 2013, 258(1):178–183.
25. Wu Z, Luo W, Sullivan SG, Rou K, Lin P, Liu W, Ming Z: Evaluation of a
needle social marketing strategy to control HIV among injecting drug
users in China. AIDS 2007, 21(8):S115–S122.
26. Des Jarlais DC, Kling R, Hammett TM, Ngu D, Liu W, Chen Y, Binh KT,
Friedmann P: Reducing HIV infection among new injecting drug users in
the China-Vietnam cross border project. AIDS 2007, (Suppl 8):S109–S114.
27. Johnston LG, Malekinejad M, Kendall C, Iuppa IM, Rutherford GW:
Implementation challenges to using respondent-driven sampling meth-
odology for HIV biological and behavioral surveillance: field experiences
in international settings. AIDS Behav 2008, 12(4 Suppl):S131–S141.
28. McCreesh N, Frost SD, Seeley J, Katongole J, Tarsh MN, Ndunguse R, Jichi F,
Lunel NL, Maher D, Johnston LG: Evaluation of respondent-driven
sampling. Epidemiology (Cambridge, Mass) 2012, 23(1):138–147.
29. Wejnert C: An empirical test of respondent-driven sampling: point
estimates, variance, degree measures, and out-of-equilibrium data.
Sociol Methodol 2009, 39(1):73–116.

doi:10.1186/1477-7517-11-15
Cite this article as: Wang et al.: Do community-based strategies reduce
HIV risk among people who inject drugs in China? A quasi-experimental
study in Yunnan and Guangxi provinces. Harm Reduction Journal
2014 11:15.

Submit your next manuscript to BioMed Central

and take full advantage of:

• Convenient online submission

• Thorough peer review
• No space constraints or color figure charges
• Immediate publication on acceptance
• Inclusion in PubMed, CAS, Scopus and Google Scholar
• Research which is freely available for redistribution

Submit your manuscript at

www.biomedcentral.com/submit