- Email: [email protected]
Cost-effectiveness of center-based compulsory rehabilitation compared to community-based voluntary methadone maintenance treatment in Hai Phong City, Vietnam
Drug and Alcohol Dependence 168 (2016) 147–155
Contents lists available at ScienceDirect
Drug and Alcohol Dependence journal homepage: www.elsevier.com/locate/drugalcdep
Full length article
Cost-effectiveness of center-based compulsory rehabilitation compared to community-based voluntary methadone maintenance treatment in Hai Phong City, Vietnam Thu Vuong a,∗ , Marian Shanahan a , Nhu Nguyen b , Giang Le c , Robert Ali a,d , Khue Pham e , Thu T.A. Vuong b , Thuy Dinh c , Alison Ritter a a
National Drug and Alcohol Research Centre/UNSW, Australia FHI360, Viet Nam c Hanoi Medical University, Viet Nam d University of Adelaide, Australia e Haiphong University of Medicine and Pharmacy, Viet Nam b
a r t i c l e
i n f o
Article history: Received 21 June 2016 Received in revised form 21 August 2016 Accepted 6 September 2016 Available online 13 September 2016 Keywords: Cost-effectiveness Economic evaluation Compulsory rehabilitation Methadone treatment Heroin dependence East and Southeast Asia
a b s t r a c t Introduction: In Vietnam, two dominant approaches for heroin treatment are center-based compulsory rehabilitation (CCT), funded by the Vietnamese government and community-based voluntary methadone maintenance treatment (MMT), funded primarily by international donors. Recent reduction in international funding requires more efficient allocation of government funding for public health programs. A cost-effectiveness analysis comparing two approaches provides a useful source of evidence to inform the government about funding reallocation. Methods: The study was a combined retrospective and prospective, non-randomized cohort comparison over three years of CCT and MMT in Vietnam, conducted between 2012 and 2014, involving 208 CCT participants and 384 MMT participants with heroin dependence. The primary end-point was drug-free days over three years. Total costs, including both program and participant personal costs were measured and cost-effectiveness compared. Mixed effects regression analyses were used to analyze effectiveness data and non-parametric bootstrapping method was used to compare cost-effectiveness. Results: Over three years, MMT costed on average VND85.73 million (US$4108) less than CCT (95% CI: -VND76.88 million, -VND94.59 million). On average, a MMT participant had 344.20 more drug-free days compared to a CCT participant (p < 0.001). The incremental cost-effectiveness ratio for MMT was -VND0.25 million (US$11.99) (95% CI: -VND0.34 million, -VND0.19 million) per drug-free day suggesting MMT is the more cost effective alternative. Conclusions: Compared to CCT, MMT is both less expensive and more effective in achieving drug-free days. If the government of Vietnam invests in MMT instead of CCT, it is potentially a cost-saving strategy for reducing illicit drug use among heroin dependent individuals. © 2016 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Over the past decade, Vietnam has made significant advances toward reducing the harms associated with illicit drug use. The HIV Law passed in 2006 paved the way for methadone maintenance treatment (MMT), with subsequent rapid scale-up. By March,
∗ Corresponding author at: The Drug Policy Modelling Program, National Drug and Alcohol Research Centre, The University of New South Wales, 22-32 King Street, Randwick, Sydney, New South Wales 2052, Australia. E-mail address: [email protected] (T. Vuong). http://dx.doi.org/10.1016/j.drugalcdep.2016.09.008 0376-8716/© 2016 Elsevier Ireland Ltd. All rights reserved.
2016, a total of 44,479 people had received MMT (27.3% of 163,000 dependent heroin users) (Vietnam Ministry of Health, 2015). MMT has been internationally recognized as an efficacious treatment for heroin dependence (Connock et al., 2007). The medication, methadone, is a synthetic opioid which is provided in maintenance doses to the patients and prevents cravings for heroin whilst allowing the resumption of participation in the community, through employment, restoration of family relationships and everyday life activities. However, the primary approach to dealing with illicit drug use continues to be center-based compulsory rehabilitation (CCT). CCT centers are institutions in which people who use drugs are confined for up to two years (Hall et al., 2012). Although CCTs
148
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
resemble low security prisons where the main activities are drug education, moral teaching, basic health care and non-voluntary labor work (Clark et al., 2013) these centers are not part of the criminal justice system or subject to judicial oversight and their detainees have not necessarily been convicted of any crime (Clark et al., 2013). Maintenance program such as MMT are not provided in these CCT centers. The number of people who have ever been in CCT centers in Vietnam is not known, however, at any point in time, there is an average of 45,000 people in a total of 121 CCT centers (Vietnam Ministry of Labor Invalid and Social Affairs, 2013). The CCT approach is also common in many other East and Southeast Asian countries (Amon et al., 2014). While MMT has been mainly funded by international donors, CCT has been funded by the Vietnam government. While there have been serious concerns about human rights violations with the CCT approach, cultural differences in the interpretation of human rights standards add a layer of complexity to international debates around the ethics of CCT (Birgin, 2013; Clark et al., 2013; Kausikan, 1993; Wu, 2013). A different and yet mutually beneficial approach is to assess the economic value of CCT. Given that Vietnam has recently become a middle-income country (Inter-Parliamentary Union, 2010), with the associated reductions in international donor funding to public health programs, providing economic evaluations of drug dependence treatment to inform the investment mix is timely. This study is the first economic evaluation that compared the cost-effectiveness of CCT to MMT in Vietnam, with generalizability to similarly situated countries in East and Southeast Asia. Systematic review of economic evaluations conducted in developed countries (Connock et al., 2007) and subsequent studies in Asian countries (Naning et al., 2014; Ni et al., 2012; Tran et al., 2012a,b; Wammes et al., 2012; Xing et al., 2012) have shown that MMT was cost-effective compared to no drug therapy. Specifically, the studies in Malaysia (Naning et al., 2014) and Vietnam (Tran et al., 2012a,b) concluded that MMT was a cost-effective strategy compared to no intervention. The studies in China (Ni et al., 2012; Xing et al., 2012) and Indonesia (Wammes et al., 2012) concluded that the cost associated with one averted HIV-infected case ranged from US$745 to US$4609. Evidence of the economic value of CCT is unknown. To date, only one study has been conducted in Thailand (Johnson et al., 2012), evaluating the effectiveness of Therapeutic Communities program delivered in closed settings. This study found a large reduction in illicit drug use over 6-months. However, this was a non-comparative study and highly significant contextual factors may account for the results (Vuong and Ritter, 2013). Two hypotheses were tested in the current study: 1) that MMT increases the number of drug-free days to a greater extent than CCT; and 2) that MMT is more cost-effective in achieving incremental drug-free days compared to CCT.
2. Methods 2.1. Interventions Two interventions were examined in this study: center-based compulsory drug rehabilitation (CCT) and community-based voluntary methadone maintenance treatment (MMT). The study was conducted in Hai Phong City, located in the northeast of Vietnam. There are three CCT centers in Hai Phong City and the current study involved all three centers. Participants who were admitted to these three CCT centers were from all 14 districts throughout Hai Phong City. MMT in Vietnam follows the clinical guidelines of developed countries including the USA and Australia (Vietnam Ministry of
Health, 2007) with MMT patients required to attend the MMT clinic daily for administration of methadone, supplemented by drug dependence counselling. In Hai Phong City, the first three MMT clinics started operation in 2008 and the current study involved all three clinics. In order to be eligible for MMT treatment, one had to be a permanent resident of the districts where the three clinics are located. 2.2. Outcome measures The current study had two primary outcomes: self-reported heroin abstinence and self-reported drug-free days (DFDs). In this paper, DFDs was reported in two different time-frames. DFDs in the preceding 30 days was used for effectiveness comparison and DFDs over three years was used for the cost-effectiveness comparison. Urine drug screening was undertaken to verify the validity of self-reported heroin use. 2.3. Study design and procedures For this study, it was neither ethical (CCT has been internationally criticized for violation of human rights; Amon et al., 2014; World Health Organization, 2009) nor feasible (those assigned to CCT would likely cease to participate) to conduct a randomized clinical trial. Therefore observational data and subsequent complex statistical methods ((a) stepwise procedures for selection of appropriate baseline predictors; and (b) mixed effects regression analysis with three steps in modeling building for random effects, variancecovariance structure, and fixed effects) to account for observable differences and time variances were used. Fig. 1 depicts the study design, sample sizes and follow-up rates. The study design was a combined retrospective and prospective three-year cohort study with data collected at 5 time-points (baseline, two years after treatment commencement, and then 3 months, 6 months and 12 months after the initial two years). Between 2008 and 2011, the Vietnam Ministry of Health conducted an MMT effectiveness study. A total of 462 MMT patients (97% of eligible participants) were recruited at treatment entry and reassessed at 3, 6, 9, 12, 18 and 24 months. At 24-months, 384 patients (83% of 462) were interviewed: this represents the data for Baseline and T1 in Fig. 1. The current study invited all 384 participants, of whom 314 (81.7%) agreed to participate, and were interviewed at T2 (January–March, 2013), T3 and T4. At the end of the study, 298 MMT participants were interviewed, representing a follow-up rate of 78%. All CCT participants who were newly released (after two years of rehabilitation) from three CCT centers during January–November, 2013 were invited to take part in the study (385 eligible participants). Interviews with 208 CCT participants (54% of 385) were conducted during July–November, 2013: this represents the data for Baseline and T1 in Fig. 1. At the first interview (T1), data on behaviors three months prior to treatment were collected retrospectively and current (two years after treatment) data were also collected. At the end of the study, 166 CCT participants were interviewed, representing a follow-up rate of 80%. Due to substantial delays in obtaining the required political endorsement from the Hai Phong City Government, there were variations between the design and the actual recruitment and interview timelines. Specifically, the time between T1 and T2 for the MMT group was 19 months while it was three months for the CCT group. The time gap of 19 months for MMT was the length of the time between the end of the previous MMT study and the beginning of the current study. In addition, for the CCT group, there was an average 4-month gap between the time of release from the CCT centers to the T1 interview. This limitation was managed by including ‘Time’ as a covariate in the data analysis, allowing the mixed effects
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
2 years in CCT (retrospecve primary data)
CCT
MMT
1 year follow-up (prospecve primary data) T1 n=208 (100%)
Baseline (B) n=208 (100%)
2 years in MMT (secondary data - previous cohort study)
Baseline (B) n=384 (100%)
149
T2 n=182 (88%)
T3 n=173 (83%)
T4 n=166 (80%)
1 year follow-up (prospecve primary data) T1 n=384 (100%)
T2 n=314 (82%)
T3 n=304 (80%)
T4 n=298 (78%)
Fig. 1. Three-year-horizon comparison of cost-effectiveness.
regression model to both account for ‘Time’ as a covariate, and to assess the statistical significance of the difference between the two groups in relation to time trends. Inclusion criteria for CCT participants were: (1) 18 years or older; (2) heroin dependence (daily heroin use during 3 months prior to treatment as a proxy); (3) in CCT under the compulsory track; (4) official confirmation of recent release from CCT centers; and (5) consented voluntarily to participate in the study. Inclusion criteria for MMT participants were: (1) 18 years or older; (2) participated in the previous MMT cohort study; and (3) consented voluntarily to participate in this study. The study received ethical approval from the UNSW Australia Human Research Ethics Committee, US-based FHI360 Internal Review Board, Hanoi Medical University Ethics Review Committee and the Hai Phong Government Study Review Board. 2.4. Sample size Determining the appropriate sample size for this study needed to address three aspects: 1) the lack of previous research on CCT; 2) potential loss to follow-up; and 3) a non-randomized design. The primary outcome measure used in the power analysis to determine the appropriate sample size was the proportion of people who were heroin abstinent at the final time-point (T4). The power analysis used an alpha of 0.05, beta of 0.20 (power 80%) and an estimated proportion of successful treatment rate of 20% (heroin-free) 12 months after being released from CCT centers (T4), as reported by governments of many provinces/cities in Vietnam, using administrative data. We assumed that the CCT in Hai Phong City was more effective compared to the rest of Vietnam with a success rate of 42% because of a higher level of investment in facilities and human resources. For MMT treatment, the 2009 MMT cohort study showed that 85.4% of the MMT participants were heroin abstinent at 12 months in MMT treatment. The power analysis assumed that the proportion of MMT participants who were free of illicit drug use could be 62.4% if the data analysis included both missing and censored data. The resultant power analysis revealed a required minimum sample size of 136 per group. Given an estimated loss to follow-up of 30%, this was increased to 186. Because this was a nonrandomized study, baseline differences between the two groups were expected. Therefore, the final minimum target sample of 208 (for each group) was determined to allow analysis that managed baseline differences and potential sub-group analysis. Additional consideration was given to the selection of the appropriate starting sample for the MMT group: a choice between the 384 MMT participants at T1 or only the 314 who then consented to participate in the further follow-up time periods. A decision was made to select the 384 people who were interviewed at two-year posttreatment as the starting sample. This strategy aimed to equalize the starting samples of both the CCT and MMT groups to the number
at the end of two-year post treatment. The consequence is that this potentially biases against MMT, inasmuch as there are no follow-up data for the 70 participants who were not able to be contacted to participate (whose outcome data were then imputed using worst case scenario assumption). 2.5. Sources and quantification of cost and effectiveness data The perspective taken for this CEA was that of the treatment sector, with average program costs (total program costs for each MMT clinic and each CCT center divided by the number of participants) plus individual-specific participant personal costs being estimated to derive individual-specific total costs. Although both participant and program costs were included, it was not a full social perspective as it did not include other health care or criminal justice costs. Program costs included all costs required for the functioning of a CCT center and a MMT clinic, excluding land costs. Participant personal costs included costs paid by the participants and the opportunity costs (travel costs and loss of productivity) of attending treatment. Loss of productivity was calculated using actual time reported employed for both groups and individual-specific monthly wage. MMT program cost data came from three sources: 1) a previous MMT cost study that provided 2009 program costs (Tran et al., 2012a,b); 2) a MMT program 2013 expenditure report provided by the Hai Phong City Department of Health; and 3) individual-specific methadone dose from which the individual-specific methadone syrup cost was calculated. CCT program costs were 2010, 2011, and 2012 financial data, provided by three CCT centers through a modified DATCAP questionnaire (Salomé et al., 2003). A detailed description of the cost components and the sources of data are presented in Supplementary material 1. The annual average MMT and CCT program costs were averaged across the different years, after adjustment for inflation using the general Consumer Price Index to 2013 Vietnamese Dongs (VND; General Statistics Office of Vietnam, 2013) and adjustment for participant drop-outs of treatment at each time-point. Where US dollars are reported, the annual average exchange rate of US$1 = VND20,780 in 2013 was used. Except for CCT program cost, all cost elements were participant-level cost. As MMT treatment is an on-going intervention, the average three-year program costs were calculated by multiplying the average annual program costs by three. CCT is time-limited (two years) with no costs incurred in the third year. Therefore, the average three-year program costs (the comparative time horizon) were calculated by multiplying the average annual program costs by two. Outcome data used for this analysis was DFDs. For politically sensitive reasons, it was not possible to ask the CCT participants about their drug-using behaviors while they were in CCT centers. Yet it is common knowledge that drug use occurs in the Vietnam CCTs (Open Society Institute, 2009; The Laborer Online, 2015). Therefore, for the main analysis, proxy data were derived from
150
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
reported prevalence and frequency of illicit drug use of incarcerated populations in other countries (Bird et al., 1997; Bullock, 2003; Calzavara et al., 2003; Darke et al., 1998; Gaughwin et al., 1991; Stark et al., 2006). Specifically, the above six referenced studies provided evidence that the illicit drug use prevalence among drugusing prison inmates ranged from 31% to 74% and the frequency of use ranged from 1 to 8 days (in the 30-day time window). Derivation of a number that could be perfectly representative of these studies was impossible. Therefore, a decision was made to use a generalized average number of 26 DFDs out of 30 for imputation for 67% of the CCT participants (those who reported drug use at T1) for their two-year stay in the CCT centers. The remaining 33% of the CCT participants did not report illicit drug use at T1, hence were assumed to be less likely to have used illicit drugs while in the CCT and no imputation was therefore made. This assumption was tested in the sensitivity analyses (which confirmed that the main results were robust). All other DFDs data were derived from self-report questionnaire. A three-year time horizon was used for the CEA (Fig. 1) therefore an aggregate three-year DFDs measure was calculated for each participant (following Harris et al., 2005). This assumes that the levels in the intervening months were equal to the average of those reported at the adjacent interviews. In addition, we followed Ascher-Svanum et al. (2010) in using the predicted values (e.g., LSMEANS) of DFDs derived from mixed effects regression analysis when calculating the total change in DFDs over the three years. Predicted values were obtained from the mixed effects regression analysis of the raw data. The use of predicted values is superior to raw data because it accounts for: 1) the difference in baseline demographics characteristics; 2) the difference in baseline DFDs between participants; and 3) within-participant correlation from the repeated measurements. The formula and specific calculation of the mean ‘DFDs (over 36 months)’ for each group are presented in Supplementary material 2. 2.6. Statistical analysis Statistical analyses were performed using SAS software, version 9.4. Mixed effects regression analyses, assuming autoregressive covariance structure was used to assess between-group effects on rate of change over time in DFDs, adjusting for possible confounding factors including baseline predictors and differences at baseline on DFDs. Step-wise procedures, involving diagnosis of collinearity and univariate analysis (using Ordinary Least Squares) were used to select the most appropriate baseline predictors for the data. Nine predictors were selected and are listed in Table 2. Mixed effects regression analysis can also account for potential correlation within participant from the repeated measurements, allow for irregularlyspaced measurement across time, which is the case for the data in this study. The interaction terms of “Group” and “Time” were included in the models to check if trend over time differed by group. In mixed effects regression analysis, goodness of fit is determined by the -2 restricted log likelihood (-2RELL) for random effects and variance-covariance structure model building and -2 log likelihood (-2LL) for fixed effects modeling building. These criteria were used in determining the best fitting model as each progressive model improved upon the fit of the previous model. The goodness of fit of the final models are reported in Table 2. All statistical tests used two-sided p-values, with significance set at p < 0.05. All analyses were performed on an intention-to-treat basis. Missing data were due entirely to drop-out from the study and only pertained to outcome variables (the rates of drop-out for each time-point of both groups are presented in Fig. 1). Little’s MCAR test (Little, 1988) showed that the missing data was not missing completely at random (MCAR) (X2 = 24.598, DF = 9, p = 0.003). It is impossible to test whether the data are missing at random (MAR)
or missing not at random (MNAR). Therefore, two methods of handling missing data were used: maximum likelihood (ML) and worst case scenario (WCS). ML assumes that data are MAR and handles missing data by “borrowing” information from the values of the covariates, predictor variables and values of the outcome variable of the earlier time-points to project values for later time-points (Allison, 2012). ML is the estimation method used in mixed effects regression analysis therefore no actual imputation work is required. The WCS method assumes that the lost-to-follow-up participants had outcomes (DFDs) identical to their baseline values and is a conservative test, valid for data that are MNAR (Myers, 2000). Given the high loss-to-follow-up rate of MMT participants, WCS was used for the main analysis to ensure attrition bias was addressed. 2.7. Cost-effectiveness analysis (CEA) The central measure used in CEA is the incremental costeffectiveness ratio (ICER), which is: ICER =
Mean cost of MMT − Mean cost of CCT Mean effectiveness of MMT − Mean effectiveness of CCT
Bootstrapping was used for comparing the costs and cost-effectiveness. Non-parametric bootstrapping has gained widespread use in economic evaluation alongside clinical trials because it makes no assumptions about the underlying distribution of the data (Gray et al., 2010). The bootstrapping procedure with 5000 replications was conducted on the individual-specific total cost, the individual-specific outcome (DFDs over three years) and the ICER to generate confidence intervals. 2.8. Sensitivity analysis The robustness of the CEA results was tested in four different sensitivity analyses with more conservative assumptions on both cost and effectiveness. The first sensitivity analysis (Scenario 1A) increased the estimated MMT program costs by using the program costs of 2009 only (adjusted to 2013 by CPI), instead of using the average program cost of 2009 and 2013 as in the main analysis. The second sensitivity analysis (Scenario 2A and 2B) focused on the uncertainty around CCT program costs as some CCT centers had higher program costs than others due to different levels of investment from the government and the number of participants in these centers. The third sensitivity analysis (Scenario 3A) involved exploring the uncertainty around data on drug-using behaviors of CCT participants during their two years in CCT centers. While the main analysis assumed 26 (out of every 30 days) drug free, the sensitivity analyses assumed zero drug use for the two years (30 out of 30 days). The final sensitivity analysis (Scenario 4A) managed the uncertainty pertaining to the specification of the mixed effects regression analysis. Details on the systematic framework and the rationale for the sensitivity analyses are provided in Supplementary material 3. 3. Results The characteristics of the participants are found in Table 1. Most participants had long histories of heroin dependence and of previous treatment episodes. At baseline the CCT participants appeared to have lower heroin problems (years of use, daily heroin use, ever injected and spending on heroin), even though almost every participant in both groups was using heroin on a daily basis at baseline. A smaller proportion of CCT participants had ever sought treatment and for those who had, they experienced fewer number of treatment episodes. These differences were managed in the mixed effects regressions analyses.
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
151
Table 1 Participants’ demographics, drug use history and behavioral characteristics . CCT (n = 208)
MMT (n = 384)
P-value
Male (%) Mean Age (SD)
100 33.26 (7.60)
100 37.32 (8.23)
n/a p < 0.001
Marital status (%) Single Married, divorced, separate or widow
51.69 48.31
43.49 56.51
P = 0.04
Employment (%) Unemployed Full-time Part-time (casual work) Average legal monthly income (median & range)
23.56 17.31 59.14
33.33 21.35 45.32
P = 0.003
3.00 million (0.80–15.00 million)
1.50 million (0–35.00 million)
p < 0.001
39.90 60.10
46.09 53.91
p = 0.16
22.35 (5.79)
23.22 (6.42)
p = 0.09
85.58 10.58 4.81
80.73 16.41 2.34
p = 0.03
11.01 (6.22) 97.54 3 (0–20) VND4.55 million (0–84 million) 66.43 50.48
13.17 (4.63) 99.96 4 (0–18) VND6.20 million (0–225 million) 82.55 28.91
p < 0.001 p = 0.06 p < 0.001 p < 0.001
3 (1.25) 80.77 2 (1–50)
2 (0.82) 96.61 5 (1–40)
p < 0.001 p < 0.001 p < 0.001
37.98 66.70 35.20 16.35 18.75
49.74 84.40 30.30 26.30 12.10
p = 0.004 p < 0.001 p = 0.13 p = 0.006 p = 0.06
Education (%) ≥ High school < High school (Year 12) Mean age of 1st intoxication (SD) % drugs of first intoxication Heroin Opium Other drugs Mean number of years used (SD) Ever reported daily heroin use (%) Median heroin use frequency of those who used every day (range) Median monthly drug spending (range) % heroin injection ever % poly drug use ever *Methamphetamine second drug of choice 28.30% for CCT and 11.60% for MMT Mean number of drug classes used (SD) of those who were poly users % ever sought drug dependence treatment Median number of treatment episodes (range) *for those who ever sought treatment % ever been to CCT centers % ever home detoxed % ever committed illegal behaviors % ever been to prison % overdose incident ever
p < 0.001 p < 0.001
Note: Chi-squared test was performed for categorical and t-test was performed for continuous variables. Table 2 Mixed effects regression analysis (with ‘drug-free days (in the preceding 30 days)’ as outcome variable (SE)), ML method versus WCS method for handling missing data. Fixed Effects
ML Method
WCS Method
Model 9
Model 10 (Final Model)
Model 8
Model 9 (Final Model)
Intercept Group = CCT Time Time*Group = CCT Time*Time Time*Time*Group = CCT Age Prison history Education level Marital status Poly drug use history Ever injected heroin Ever been to CCT center Number of treatment episodes Employed Model Fit (-2LL)
5.49*** (0.52) 3.78*** (0.56) 21.87*** (0.61) −11.27***(1.03) −4.17*** (0.14) 1.67*** (0.23) 0.02 (0.01) −0.44* (0.22) 0.07 (0.10) −0.23 (0.19) −0.02 (0.19) −0.50* (0.23) −0.23 (0.19) −0.003 (0.02) −0.17 (0.20) 18082
5.97*** 3.71*** 21.88*** −11.29*** −4.16*** 1.67*** – −0.41* – – – −0.50* – – – 18082
7.27*** (0.61) 2.92*** (0.61) 17.08***(0.77) −8.85***(1.29) −3.54***(0.15) 1.34*** (0.25) 0.02 (0.01) −0.43* (0.21) 0.04 (0.18) −0.30 (0.18) −0.20 (0.19) −0.35 (0.23) −0.04 (0.18) −0.0004(0.02) −0.30 (0.20) 21743
7.55*** (0.36) 2.93*** (0.60) 17.07*** (0.77) −8.85*** (1.29) −3.54*** (0.15) 1.34*** (0.25) – 0.45* (0.21) – – –
Random Effects Residual (within-subject) variation) Between-subject intercept variation Between-subject slope variation 2.86
66.32 0
66.32 0 2.86
88.29 0 3.60
88.38 0 3.62
(0.38) (0.55) (0.61) (1.03) (0.14) (0.23) (0.22)
(0.23)
– – – 21726
Note: Mixed effects coefficients are listed under fixed effects, and variance components are listed under random effects; p< 0.05*, p< 0.01**, p< 0.001***; Standard errors in brackets.
3.1. Intervention effectiveness Using the WCS missing data imputation method for the main analysis, mixed effects regression analyses supported the
hypothesis that MMT increased the number of DFDs (in the preceding 30 days) to a greater extent than CCT ( = 8.85, SE = 1.29, p < 0.0001) (Table 2, Final Model). On average, the rate of change in the number of DFDs for one MMT participant was 8.85 days (out of
152
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
Fig. 2. Fitted trajectories of change* in “drug-free days (in the preceding 30 days)” by group: ML versus WCS imputation methods.
30 preceding days) higher compared to the rate of change for one CCT participant. Nevertheless, both CCT and MMT modalities were associated with strong effects in increasing the number of DFDs. After two years of treatment, MMT participants were able to maintain an average number of DFDs ranging from 22.97 days to 29.58 days whereas CCT participants fluctuated between 8.30 and 27.38 days (Fig. 2). The other missing data technique (ML) confirmed that MMT treatment was consistently more effective in increasing the number of DFDs across all follow-up time-points. As seen in Table 2, the ‘Time’ variables and interaction terms are statistically significant: these show, respectively that DFDs changed over time (Time); that the two groups differ in relation to change in DFDs over time in favor of MMT (Time*Group = CCT); that this change is quadratic over time (Time*Time), and finally that the curvature change differs between the two groups in favor of MMT (Time*Time*Group). Summaries of mixed effects regression model outputs (for both WCS and ML) are presented in Supplementary material 4.
3.2. Intervention cost The average annual program cost for providing treatment to one CCT participant was estimated to be VND19.67 million (US$943) which was more than double the average annual program cost for providing treatment to one MMT participant of VND7.88 million (US$378). The average participant personal costs were similarly significantly greater for CCT than MMT (see Table 3). The opportunity cost of work time accounted for the largest share of CCT participant personal costs, implying that the largest cost associated with being placed in the CCT centers was the opportunity cost of labor time forgone. The average annual total cost amounted to VND61.52 million (US$2948) for CCT and VND13.35 million (US$640) for MMT. The average thee-year total cost for CCT was VND123.04 million (US$5896), which was more than three times higher than that for MMT (VND37.31 million (US$1788)). The last row of Table 3 indicates the main bootstrap results for comparing the average three-year total costs. Specifically, in
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
153
Table 3 Program costs and participant personal costs: MMT and CCT. Unit: VND million Mean (SD)
Average annual program costs Average annual participant personal costs Average annual total costs Three-year total costs a
Bootstrap results (mean difference & CI)
MMT
CCT
7.88 (1.84) 5.47 (4.11) 13.35 (4.31) 37.31a (13.94)
19.67 (5.78) 41.85 (31.49) 61.52 (32.18) 123.04 (64.36)
Mean difference = −85.73 (CI: −76.88, −94.59)
Costs applied to months in treatment by adjustment for participant drop-outs at each time-point.
Table 4 CEA sensitivity analyses results: ‘drug-free days (over 36 months)’, using average total costs. Unit: VND Comparison
Cost MMT
Cost CCT
Cost difference
Effect MMT
Effect CCT
Effect difference
ICER (95% CI)
Main results:
37.31 mil
123.04 mil
−85.73 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
Scenario 1A: MMT most expensive: Using 2009 MMT program cost Scenario 2A: CCT most expensive: Using cost data of CCT center with highest program cost Scenario 2B: CCT least expensive: Using cost data of CCT center with smallest program cost Scenario 3A: CCT more effective at T1: No drug use in CCT centers Scenario 4A: Similar scenario as main analysis but with raw data (instead of predicted data)
45.85 mil
123.04 mil
−77.19 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
-VND0.25 mil (-VND0.34 mil to -VND0.19 mil) -VND0.22 mil (-VND0.31 mil to -VND0.17 mil)
37.31 mil
141.53 mil
−104.22 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
-VND0.31 mil (-VND0.41 mil to -VND0.24 mil)
37.31 mil
109.82 mil
−72.51 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
-VND0.21 mil (-VND0.29 mil to -VND0.16 mil)
37.31 mil
123.04 mil
−85.73 mil
507.36 DFDs
144.26 DFDs
363.10 DFDs
37.31 mil
123.04 mil
−85.73 mil
976.32 DFDs
742.59 DFDs
233.73 DFDs
-VND0.24 mil (-VND0.32 mil to -VND0.19 mil) -VND0.37 mil (-VND0.42 mil to -VND0.12 mil)
ICER: Incremental cost-effectiveness ratio. dif. = difference = MMT minus CCT. mil = million.
the main result, the average three-year total cost of MMT was VND85.73 million (US$4108) lower than CCT (95% CI: -VND76.88 million, -VND94.59 million). 3.3. Incremental cost-effectiveness ratio and cost-effectiveness plane Table 4 displays the CEA main results, which were the three-year incremental costs and the corresponding incremental effectiveness (DFDs), produced by MMT compared to CCT, using average total costs. The ICER of -VND0.25 million (US$11.97) was the product of the mean difference in average total costs (-VND85.73 million (US$4108)) divided by the mean difference in DFDs (344.20). The main results showed that the ICER for one incremental DFD was -VND0.25 million (95% CI: -VND0.34 million, -VND0.19 million). This result showed that MMT was significantly more cost-effective than CCT. The average total costs for MMT were lower than CCT (the difference was −VND85.73 million (US$4108) per participant) while achieving 344.20 more DFDs, which indicated that MMT was dominant over CCT. These results suggest that if the government of Vietnam invests in MMT treatment (instead of CCT), it will result in an additional 344.20 DFDs per average dependent heroin user plus it will cost VND85.73 million (US$4108) less for the government (for treatment) and entail lower personal costs per average dependent heroin user. The CEA results supported the hypothesis
that MMT is more cost-effective in achieving an incremental DFD compared to CCT. Table 4 presents both the main result and sensitivity analyses of the CEA using average total costs. The results of the sensitivity analyses confirmed the robustness of the main results. The cost-effectiveness planes (corresponding to the main CEA result and the sensitivity analyses results) are presented in Fig. 3, confirming that 100% of the ICERs are located in the southeast quadrant with absolute certainty (no cross-over of the ICERs with other quadrants). 4. Discussion To the best of our knowledge, this is the first economic evaluation comparing the cost effectiveness of MMT and CCT in East and Southeast Asia. Confidence in our findings is strengthened because of the use of the most conservative assumptions about both costs and effectiveness. For example, worst case scenario imputation for all those lost-to-follow-up were used in the analyses and tested in the sensitivity analyses. Thus, it is compelling that MMT treatment was dominant over CCT in the main analysis for both total costs and program costs, and that this result did not change under a variety of more conservative assumptions about both costs and effectiveness. This study provides evidence that MMT was not only less costly but also more effective compared to CCT in achieving the number of DFDs for heroin dependent users over three years in Vietnam. By investing in MMT instead of CCT, not only would one dependent
154
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
Fig. 3. Cost-effectiveness planes: CEA main result and sensitivity analyses results: ‘drug-free days (over 36 months)’, using total costs.
heroin user in Hai Phong City achieve on average 344.20 more DFDs over three years, but the government of Vietnam would also save VND17.35 million (US$831) and an average dependent heroin user would save VND68.39 million (US$3277) (VND85.73 million minus VND17.35 million) to achieve these additional 344.20 DFDs. The strength of this study is the use of empirical data (rather than modelled data) to estimate the changes in drug-using behaviors among dependent heroin users. In addition, a longitudinal study design was employed with repeated assessments of a sufficient sample that increased the level of confidence when comparing rate of change over time between groups. However, the results of these analyses should be interpreted in light of several limitations. First, the study is politically sensitive and as a result some relevant data could not be collected. This included data regarding drug-using behaviors of CCT participants while they were in CCT centers; and income generated by the CCT center management by using labor of the CCT participants. While it is unclear if those in CCT are paid for their labor and how much they are paid, the income generated by this activity for the centers could be significant. However, it is not possible to know whether this income is used to reinvest in the center activities. Second, three cost components were not included in the analysis: a) costs incurred by the CCT trainees’ families for purchasing and sending food and prescribed medicines and time to visit the trainees; b) cost of land for both treatment modalities; and b) costs of loss of freedom by CCT participants. In all three cases, these design limitations potentially bias in favour of CCT. Third, potential recall bias could be realised due to the 24-month recall window of baseline drug use behaviors among CCT participants. This limitation was identified during the design phase of this study. All interviewers were provided comprehensive training with a focus on managing recall difficulty. The interviewers gave participants adequate time to recall and/or rephrased the questions to facilitate recall. In addition, the literature suggests that although the absolute levels obtained through retrospective self-report may not be as accurate as one would desire, the relative levels are quite valid (Buavirat et al., 2003; Hser and Wickens, 1989; Napper et al., 2010). For example, one study evaluating the agreement between two recall points of 10-year interval found that the agreement was exceptional with the correlation coefficients ranging as high as 0.86 and 0.90 (Anglin et al., 1993). For data in this study, both Table 1 and
Fig. 2 suggest that under-reporting of drug use behaviors at baseline by CCT participants was unlikely given the substantial and similar level of drug use compared to MMT participants (mean number of years using drugs, daily heroin use, monthly drug spending, number of DFDs). Finally, while the findings of this study confirmed that MMT was more cost-effective than CCT, this was a single study conducted in one city in Vietnam. The findings of this study should be confirmed through further economic evaluation studies in other cities and countries in East and Southeast Asia. Conflict of interest FHI360 Vietnam and Hanoi Medical University receive funding from the United States government agencies to provide technical assistance to the Vietnamese government for the operational and clinical aspects of the national methadone treatment program in Vietnam. Robert Ali has received untied educational grants from Reckitt Benckiser and Mundipharm for the post-marketing surveillance of opiate substitution treatment (OST) medications in Australia and untied educational grants from Reckitt Benckiser for conducting a study into the pharmacogenetics predictors of treatment success. Contributors All authors participated in different important roles the study, which provided the content for this manuscript. Author Thu Vuong drafted the manuscript. All authors contributed to and have approved of the final manuscript. Author disclosures The study received funding from Atlantic Philanthropies through FHI360 Vietnam and Hanoi Medical University. Other sources of funding were the Australian Government’s Endeavour PhD Scholarship and UNSW Australia/NDARC PhD Scholarship for Thu Vuong. Alison Ritter is funded through an NHMRC Senior Research Fellowship (APP1021988). Marian Shanahan is supported by funding from the Australian Government under the Substance Misuse Prevention and Service Improvements Grants Fund.
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
Acknowledgements The study team acknowledges the invaluable contribution of the researchers at the Hai Phong Medical University for their dayto-day field work. However the views expressed are those of the authors and should not be attributed to any other agency or individual. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.drugalcdep.2016. 09.008. References Allison, P.D., 2012. Handling missing data by maximum likelihood. Proceedings of the SAS Global Forum Vol. 312. Amon, J., Pearshouse, R., Cohen, J.E., Schleifer, R., 2014. Compulsory drug detention in East and Southeast Asia: evolving government, UN and donor responses. Int. J. Drug Policy 25, 13–20. Anglin, M.D., Hser, Y.-I., Chou, C.-P., 1993. Reliability and validity of retrospective behavioral self-report by narcotics addicts. Eval. Rev. 17, 91–108. Ascher-Svanum, H., Stensland, M.D., Peng, X., Faries, D.E., Stauffer, V.L., Osuntokun, O.O., Kane, J.M., 2010. Cost-effectiveness of olanzapine vs. aripiprazole in the treatment of schizophrenia. Curr. Med. Res. Opin. 27, 115–122. Bird, A.G., Gore, S.M., Hutchinson, S.J., Lewis, S.C., Cameron, S., Burns, S., 1997. Harm reduction measures and injecting inside prison versus mandatory drugs testing: results of a cross sectional anonymous questionnaire survey. BMJ 315, 21–24. Birgin, R., 2013. Arguments against the compulsory treatment of opioid dependence. Bull. World Health Organ. 91, 239–239a. Buavirat, A., Page-Shafer, K., Van Griensven, G., Mandel, J., Evans, J., Chuaratanaphong, J., Chiamwongpat, S., Sacks, R., Moss, A., 2003. Risk of prevalent HIV infection associated with incarceration among injecting drug users in Bangkok, Thailand: case-control study. BMJ 326, 308. Bullock, T., 2003. Changing Levels Of Drug Use Before, During And After Imprisonment. Home Office Research Study, London, pp. 23–48. Calzavara, L.M., Burchell, A.N., Schlossberg, J., Myers, T., Escobar, M., Wallace, E., Major, C., Strike, C., Millson, M., 2003. Prior opiate injection and incarceration history predict injection drug use among inmates. Addiction 98, 1257–1265. Clark, N., Busse, A., Gerra, G., 2013. Voluntary treatment, not detention, in the management of opioid dependence. Bull. World Health Organ. 91, 146–147. Connock, M., Juarez-Garcia, A., Jowett, S., Frew, E., Liu, Z., Taylor, R.J., Fry-Smith, A., Day, E., Lintzeris, N., Roberts, T., Burls, A., Taylor, R.S., 2007. Methadone and buprenorphine for the management of opioid dependence: a systematic review and economic evaluation. Health Technol. Assess. 11, 1–171. Darke, S., Kaye, S., Finlay-Jones, R., 1998. Drug use and injection risk-taking among prison methadone maintenance patients. Addiction 93, 1169–1175. Gaughwin, M.D., Douglas, R.M., Liew, C., Davies, L., Mylvaganam, A., Treffke, H., Edwards, J., Ali, R., 1991. HIV prevalence and risk behaviours for HIV transmission in South Australian prisons. AIDS 5, 845–852. General Statistics Office of Vietnam, 2013. Consumer Price Index. General Statistics Office of Vietnam, Hanoi. Gray, A.M., Clarke, P.M., Wolstenholme, J.L., Wordsworth, S., 2010. Applied Methods Of Cost-Effectiveness Analysis In Healthcare. Oxford University Press, Oxford. Hall, W., Babor, T., Edwards, G., Laranjeira, R., Marsden, J., Miller, P., Obot, I., Petry, N., Thamarangsi, T., West, R., 2012. Compulsory detention, forced detoxification and enforced labor are not ethically acceptable or effective ways to treat addiction. Addiction 107, 1891–1893. Harris, A.H., Gospodarevskaya, E., Ritter, A., 2005. A randomised trial of the cost effectiveness of buprenorphine as an alternative to methadone maintenance treatment for heroin dependence in a primary care setting. Pharmacoeconomics 23, 77–91. Hser, Y.I., Wickens, T.D., 1989. The effects of the spacing of test trials and study trials in paired-association learning. J. Educ. Psychol. 9, 99–120. Inter-Parliamentary Union, 2010. Representation From The Top: Ethnic Minorities In The National Assembly Of Viet Nam. Hanoi. Johnson, K.W., Young, L., Shamblen, S., Suresh, G., Browne, T., Chookhare, K.W., 2012. Evaluation of the therapeutic community treatment model in Thailand: policy implications for compulsory and prison-based treatment. Subst. Use Misuse 47, 889–909.
155
Kausikan, B., 1993. Asia’s different standard. Foreign Policy Washingtonpost 92, 24–41. Little, R.J., 1988. A test of missing completely at random for multivariate data with missing values. J. Am. Stat. Assoc. 83, 1198–1202. Myers, W.R., 2000. Handling missing data in clinical trials: an overview. Drug Inf. J. 34, 525–533. Naning, H., Kerr, C., Kamarulzaman, A., Osornprasop, S., Dahlui, M., Ng, C.-W., Wilson, D., 2014. Return On Investment And Cost-Effectiveness Of Harm Reduction Program In Malaysia, http://documents.worldbank.org/curated/en/ 2014/07/19764976/return-investment-cost-effectiveness-harm-reductionprogram-malaysia (accessed 18.10.14.). Napper, L.E., Fisher, D.G., Reynolds, G.L., Johnson, M.E., 2010. HIV risk behavior self-report reliability at different recall periods. AIDS Behav. 14, 152–161. Ni, M.J., Fu, L.P., Chen, X.L., Hu, X.Y., Wheeler, K., 2012. Net financial benefits of averting HIV infections among people who inject drugs in Urumqi, Xinjiang, Peoples Republic of China (2005–2010). BMC Public Health 12, 572–579. Open Society Institute, 2009. At What Cost? HIV And Human Rights Consequences Of The Global ‘War On Drugs’. International Harm Reduction Development Program, Public Health Program. Open Society Institute, New York. Salomé, H.J., French, M.T., Miller, M., McLellan, A.T., 2003. Estimating the client costs of addiction treatment: first findings from the client drug abuse treatment cost analysis program (Client DATCAP). Drug Alcohol Depend. 71, 195–206. Stark, K., Herrmann, U., Ehrhardt, S., Bienzle, U., 2006. A syringe exchange programme in prison as prevention strategy against HIV infection and hepatitis B and C in Berlin, Germany. Epidemiol. Infect. 134, 814–819. The Laborer Online, 2015. Tham lau ma tuy vao trai giam khong con la chuyen la (smuggling illicit drugs into closed settings is not new). http://nld.com.vn/ thoi-su-trong-nuoc/tham-lau-ma-tuy-vao-trai-giam-khong-con-la-chuyen-la20150414122935841.htm (accessed 12.04.15.). Tran, B.X., Ohinmaa, A., Duong, A.T., Do, N.T., Nguyen, L.T., Mills, S., Houston, S., Jacobs, P., 2012a. Cost-effectiveness of methadone maintenance treatment for HIV-positive drug users in Vietnam. AIDS Care 24, 283–290. Tran, B.X., Ohinmaa, A., Duong, A.T., Nguyen, L.T., Vu, P.X., Mills, S., Jacobs, P., 2012b. The cost-effectiveness and budget impact of Vietnam’s methadone maintenance treatment programme in HIV prevention and treatment among injection drug users. Global Public Health 7, 1080–1094. – nh soˆ´ 5076/QÐ-BYT, 2007, ban hành Vietnam Ministry of Health, 2007. Quyeˆ´ t di . – eˆ` u tri thay theˆ´ nghiˆen các chaˆ´ t dang thuoˆ´ c phiˆen ba˘` ng thuoˆ´ c ´ hu ong da˜ˆ n di . . . .
ˆ chuc ´ ´ thu. c hiˆe.n (English translation: ’Ministry of da˜ˆ n to methadone và hu ong Health Decision no. 5076/QD-BYT/2007, providing guideline on implementation of methadone maintenance treatment’). http:// thuvienphapluat.vn/van-ban/The-thao-Y-te/Quyet-dinh-5076-QD-BYT-2007huong-dan-dieu-tri-thay-the-nghien-do-Bo-Y-te-ban-hanh-205168.aspx (accessed 21.09.15.). – eˆ` u tri Vietnam Ministry of Health, 2015. Cˆa.p nhˆa.t tình hình thu. c hiˆe.n chı tiêu di . nghiˆe.n các chaˆ´ t da.ng thuoˆ´ c phiˆe.n ba˘` ng Methadone (MMT) d–eˆ´ n tháng 10 n˘am 2015 (English translation: ’Update of progress in expansion of MMT program until October 2015’) by the Minitry of Health (MOH) of Vietnam. http://vaac. gov.vn/solieu/Detail/?userkey=Cap-nhat-tinh-hinh-thuc-hien-chi-tieu-dieutri-nghien-cac-chat-dang-thuoc-phien-bang-Methadone-MMT-den-thang-9nam-2015 (accessed 15.11.15.). Vietnam Ministry of Labor Invalid and Social Affairs, 2013. Renovation Plan for Drug Addiction Treatment for 2013–2020 in Vietnam. Ministry of Labor, Invalid and Social Affairs (MOLISA), Hanoi. Vuong, T., Ritter, A., 2013. Not to lose sight of the bigger picture: the case of therapeutic communities (TC) research in closed settings in Thailand. Subst. Use Misuse 48, 793–794. Wammes, J.J., Siregar, A.Y., Hidayat, T., Raya, R.P., van Crevel, R., van der Ven, A.J., Baltussen, R., 2012. Cost-effectiveness of methadone maintenance therapy as HIV prevention in an Indonesian high-prevalence setting: a mathematical modeling study. Int. J. Drug Policy 23, 358–364. World Health Organization, 2009. Assessment Of Compulsory Treatment Of People Who Use Drugs In Cambodia, China, Malaysia And Viet Nam: An Application Of Selected Human Rights Principles. A Report By WHO Western Pacific Regional Office In Manila , WHO, Geneva. Wu, Z., 2013. Arguments in favor of compulsory treatment of opioid dependence. Bull. World Health Organ. 91, 142–145. Xing, Y., Sun, J., Cao, W., Lee, L., Guo, H., Li, H., Duan, S., 2012. Economic evaluation of methadone maintenance treatment in HIV/AIDS control among injecting drug users in Dehong, China. AIDS Care 24, 756–762.
Contents lists available at ScienceDirect
Drug and Alcohol Dependence journal homepage: www.elsevier.com/locate/drugalcdep
Full length article
Cost-effectiveness of center-based compulsory rehabilitation compared to community-based voluntary methadone maintenance treatment in Hai Phong City, Vietnam Thu Vuong a,∗ , Marian Shanahan a , Nhu Nguyen b , Giang Le c , Robert Ali a,d , Khue Pham e , Thu T.A. Vuong b , Thuy Dinh c , Alison Ritter a a
National Drug and Alcohol Research Centre/UNSW, Australia FHI360, Viet Nam c Hanoi Medical University, Viet Nam d University of Adelaide, Australia e Haiphong University of Medicine and Pharmacy, Viet Nam b
a r t i c l e
i n f o
Article history: Received 21 June 2016 Received in revised form 21 August 2016 Accepted 6 September 2016 Available online 13 September 2016 Keywords: Cost-effectiveness Economic evaluation Compulsory rehabilitation Methadone treatment Heroin dependence East and Southeast Asia
a b s t r a c t Introduction: In Vietnam, two dominant approaches for heroin treatment are center-based compulsory rehabilitation (CCT), funded by the Vietnamese government and community-based voluntary methadone maintenance treatment (MMT), funded primarily by international donors. Recent reduction in international funding requires more efficient allocation of government funding for public health programs. A cost-effectiveness analysis comparing two approaches provides a useful source of evidence to inform the government about funding reallocation. Methods: The study was a combined retrospective and prospective, non-randomized cohort comparison over three years of CCT and MMT in Vietnam, conducted between 2012 and 2014, involving 208 CCT participants and 384 MMT participants with heroin dependence. The primary end-point was drug-free days over three years. Total costs, including both program and participant personal costs were measured and cost-effectiveness compared. Mixed effects regression analyses were used to analyze effectiveness data and non-parametric bootstrapping method was used to compare cost-effectiveness. Results: Over three years, MMT costed on average VND85.73 million (US$4108) less than CCT (95% CI: -VND76.88 million, -VND94.59 million). On average, a MMT participant had 344.20 more drug-free days compared to a CCT participant (p < 0.001). The incremental cost-effectiveness ratio for MMT was -VND0.25 million (US$11.99) (95% CI: -VND0.34 million, -VND0.19 million) per drug-free day suggesting MMT is the more cost effective alternative. Conclusions: Compared to CCT, MMT is both less expensive and more effective in achieving drug-free days. If the government of Vietnam invests in MMT instead of CCT, it is potentially a cost-saving strategy for reducing illicit drug use among heroin dependent individuals. © 2016 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Over the past decade, Vietnam has made significant advances toward reducing the harms associated with illicit drug use. The HIV Law passed in 2006 paved the way for methadone maintenance treatment (MMT), with subsequent rapid scale-up. By March,
∗ Corresponding author at: The Drug Policy Modelling Program, National Drug and Alcohol Research Centre, The University of New South Wales, 22-32 King Street, Randwick, Sydney, New South Wales 2052, Australia. E-mail address: [email protected] (T. Vuong). http://dx.doi.org/10.1016/j.drugalcdep.2016.09.008 0376-8716/© 2016 Elsevier Ireland Ltd. All rights reserved.
2016, a total of 44,479 people had received MMT (27.3% of 163,000 dependent heroin users) (Vietnam Ministry of Health, 2015). MMT has been internationally recognized as an efficacious treatment for heroin dependence (Connock et al., 2007). The medication, methadone, is a synthetic opioid which is provided in maintenance doses to the patients and prevents cravings for heroin whilst allowing the resumption of participation in the community, through employment, restoration of family relationships and everyday life activities. However, the primary approach to dealing with illicit drug use continues to be center-based compulsory rehabilitation (CCT). CCT centers are institutions in which people who use drugs are confined for up to two years (Hall et al., 2012). Although CCTs
148
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
resemble low security prisons where the main activities are drug education, moral teaching, basic health care and non-voluntary labor work (Clark et al., 2013) these centers are not part of the criminal justice system or subject to judicial oversight and their detainees have not necessarily been convicted of any crime (Clark et al., 2013). Maintenance program such as MMT are not provided in these CCT centers. The number of people who have ever been in CCT centers in Vietnam is not known, however, at any point in time, there is an average of 45,000 people in a total of 121 CCT centers (Vietnam Ministry of Labor Invalid and Social Affairs, 2013). The CCT approach is also common in many other East and Southeast Asian countries (Amon et al., 2014). While MMT has been mainly funded by international donors, CCT has been funded by the Vietnam government. While there have been serious concerns about human rights violations with the CCT approach, cultural differences in the interpretation of human rights standards add a layer of complexity to international debates around the ethics of CCT (Birgin, 2013; Clark et al., 2013; Kausikan, 1993; Wu, 2013). A different and yet mutually beneficial approach is to assess the economic value of CCT. Given that Vietnam has recently become a middle-income country (Inter-Parliamentary Union, 2010), with the associated reductions in international donor funding to public health programs, providing economic evaluations of drug dependence treatment to inform the investment mix is timely. This study is the first economic evaluation that compared the cost-effectiveness of CCT to MMT in Vietnam, with generalizability to similarly situated countries in East and Southeast Asia. Systematic review of economic evaluations conducted in developed countries (Connock et al., 2007) and subsequent studies in Asian countries (Naning et al., 2014; Ni et al., 2012; Tran et al., 2012a,b; Wammes et al., 2012; Xing et al., 2012) have shown that MMT was cost-effective compared to no drug therapy. Specifically, the studies in Malaysia (Naning et al., 2014) and Vietnam (Tran et al., 2012a,b) concluded that MMT was a cost-effective strategy compared to no intervention. The studies in China (Ni et al., 2012; Xing et al., 2012) and Indonesia (Wammes et al., 2012) concluded that the cost associated with one averted HIV-infected case ranged from US$745 to US$4609. Evidence of the economic value of CCT is unknown. To date, only one study has been conducted in Thailand (Johnson et al., 2012), evaluating the effectiveness of Therapeutic Communities program delivered in closed settings. This study found a large reduction in illicit drug use over 6-months. However, this was a non-comparative study and highly significant contextual factors may account for the results (Vuong and Ritter, 2013). Two hypotheses were tested in the current study: 1) that MMT increases the number of drug-free days to a greater extent than CCT; and 2) that MMT is more cost-effective in achieving incremental drug-free days compared to CCT.
2. Methods 2.1. Interventions Two interventions were examined in this study: center-based compulsory drug rehabilitation (CCT) and community-based voluntary methadone maintenance treatment (MMT). The study was conducted in Hai Phong City, located in the northeast of Vietnam. There are three CCT centers in Hai Phong City and the current study involved all three centers. Participants who were admitted to these three CCT centers were from all 14 districts throughout Hai Phong City. MMT in Vietnam follows the clinical guidelines of developed countries including the USA and Australia (Vietnam Ministry of
Health, 2007) with MMT patients required to attend the MMT clinic daily for administration of methadone, supplemented by drug dependence counselling. In Hai Phong City, the first three MMT clinics started operation in 2008 and the current study involved all three clinics. In order to be eligible for MMT treatment, one had to be a permanent resident of the districts where the three clinics are located. 2.2. Outcome measures The current study had two primary outcomes: self-reported heroin abstinence and self-reported drug-free days (DFDs). In this paper, DFDs was reported in two different time-frames. DFDs in the preceding 30 days was used for effectiveness comparison and DFDs over three years was used for the cost-effectiveness comparison. Urine drug screening was undertaken to verify the validity of self-reported heroin use. 2.3. Study design and procedures For this study, it was neither ethical (CCT has been internationally criticized for violation of human rights; Amon et al., 2014; World Health Organization, 2009) nor feasible (those assigned to CCT would likely cease to participate) to conduct a randomized clinical trial. Therefore observational data and subsequent complex statistical methods ((a) stepwise procedures for selection of appropriate baseline predictors; and (b) mixed effects regression analysis with three steps in modeling building for random effects, variancecovariance structure, and fixed effects) to account for observable differences and time variances were used. Fig. 1 depicts the study design, sample sizes and follow-up rates. The study design was a combined retrospective and prospective three-year cohort study with data collected at 5 time-points (baseline, two years after treatment commencement, and then 3 months, 6 months and 12 months after the initial two years). Between 2008 and 2011, the Vietnam Ministry of Health conducted an MMT effectiveness study. A total of 462 MMT patients (97% of eligible participants) were recruited at treatment entry and reassessed at 3, 6, 9, 12, 18 and 24 months. At 24-months, 384 patients (83% of 462) were interviewed: this represents the data for Baseline and T1 in Fig. 1. The current study invited all 384 participants, of whom 314 (81.7%) agreed to participate, and were interviewed at T2 (January–March, 2013), T3 and T4. At the end of the study, 298 MMT participants were interviewed, representing a follow-up rate of 78%. All CCT participants who were newly released (after two years of rehabilitation) from three CCT centers during January–November, 2013 were invited to take part in the study (385 eligible participants). Interviews with 208 CCT participants (54% of 385) were conducted during July–November, 2013: this represents the data for Baseline and T1 in Fig. 1. At the first interview (T1), data on behaviors three months prior to treatment were collected retrospectively and current (two years after treatment) data were also collected. At the end of the study, 166 CCT participants were interviewed, representing a follow-up rate of 80%. Due to substantial delays in obtaining the required political endorsement from the Hai Phong City Government, there were variations between the design and the actual recruitment and interview timelines. Specifically, the time between T1 and T2 for the MMT group was 19 months while it was three months for the CCT group. The time gap of 19 months for MMT was the length of the time between the end of the previous MMT study and the beginning of the current study. In addition, for the CCT group, there was an average 4-month gap between the time of release from the CCT centers to the T1 interview. This limitation was managed by including ‘Time’ as a covariate in the data analysis, allowing the mixed effects
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
2 years in CCT (retrospecve primary data)
CCT
MMT
1 year follow-up (prospecve primary data) T1 n=208 (100%)
Baseline (B) n=208 (100%)
2 years in MMT (secondary data - previous cohort study)
Baseline (B) n=384 (100%)
149
T2 n=182 (88%)
T3 n=173 (83%)
T4 n=166 (80%)
1 year follow-up (prospecve primary data) T1 n=384 (100%)
T2 n=314 (82%)
T3 n=304 (80%)
T4 n=298 (78%)
Fig. 1. Three-year-horizon comparison of cost-effectiveness.
regression model to both account for ‘Time’ as a covariate, and to assess the statistical significance of the difference between the two groups in relation to time trends. Inclusion criteria for CCT participants were: (1) 18 years or older; (2) heroin dependence (daily heroin use during 3 months prior to treatment as a proxy); (3) in CCT under the compulsory track; (4) official confirmation of recent release from CCT centers; and (5) consented voluntarily to participate in the study. Inclusion criteria for MMT participants were: (1) 18 years or older; (2) participated in the previous MMT cohort study; and (3) consented voluntarily to participate in this study. The study received ethical approval from the UNSW Australia Human Research Ethics Committee, US-based FHI360 Internal Review Board, Hanoi Medical University Ethics Review Committee and the Hai Phong Government Study Review Board. 2.4. Sample size Determining the appropriate sample size for this study needed to address three aspects: 1) the lack of previous research on CCT; 2) potential loss to follow-up; and 3) a non-randomized design. The primary outcome measure used in the power analysis to determine the appropriate sample size was the proportion of people who were heroin abstinent at the final time-point (T4). The power analysis used an alpha of 0.05, beta of 0.20 (power 80%) and an estimated proportion of successful treatment rate of 20% (heroin-free) 12 months after being released from CCT centers (T4), as reported by governments of many provinces/cities in Vietnam, using administrative data. We assumed that the CCT in Hai Phong City was more effective compared to the rest of Vietnam with a success rate of 42% because of a higher level of investment in facilities and human resources. For MMT treatment, the 2009 MMT cohort study showed that 85.4% of the MMT participants were heroin abstinent at 12 months in MMT treatment. The power analysis assumed that the proportion of MMT participants who were free of illicit drug use could be 62.4% if the data analysis included both missing and censored data. The resultant power analysis revealed a required minimum sample size of 136 per group. Given an estimated loss to follow-up of 30%, this was increased to 186. Because this was a nonrandomized study, baseline differences between the two groups were expected. Therefore, the final minimum target sample of 208 (for each group) was determined to allow analysis that managed baseline differences and potential sub-group analysis. Additional consideration was given to the selection of the appropriate starting sample for the MMT group: a choice between the 384 MMT participants at T1 or only the 314 who then consented to participate in the further follow-up time periods. A decision was made to select the 384 people who were interviewed at two-year posttreatment as the starting sample. This strategy aimed to equalize the starting samples of both the CCT and MMT groups to the number
at the end of two-year post treatment. The consequence is that this potentially biases against MMT, inasmuch as there are no follow-up data for the 70 participants who were not able to be contacted to participate (whose outcome data were then imputed using worst case scenario assumption). 2.5. Sources and quantification of cost and effectiveness data The perspective taken for this CEA was that of the treatment sector, with average program costs (total program costs for each MMT clinic and each CCT center divided by the number of participants) plus individual-specific participant personal costs being estimated to derive individual-specific total costs. Although both participant and program costs were included, it was not a full social perspective as it did not include other health care or criminal justice costs. Program costs included all costs required for the functioning of a CCT center and a MMT clinic, excluding land costs. Participant personal costs included costs paid by the participants and the opportunity costs (travel costs and loss of productivity) of attending treatment. Loss of productivity was calculated using actual time reported employed for both groups and individual-specific monthly wage. MMT program cost data came from three sources: 1) a previous MMT cost study that provided 2009 program costs (Tran et al., 2012a,b); 2) a MMT program 2013 expenditure report provided by the Hai Phong City Department of Health; and 3) individual-specific methadone dose from which the individual-specific methadone syrup cost was calculated. CCT program costs were 2010, 2011, and 2012 financial data, provided by three CCT centers through a modified DATCAP questionnaire (Salomé et al., 2003). A detailed description of the cost components and the sources of data are presented in Supplementary material 1. The annual average MMT and CCT program costs were averaged across the different years, after adjustment for inflation using the general Consumer Price Index to 2013 Vietnamese Dongs (VND; General Statistics Office of Vietnam, 2013) and adjustment for participant drop-outs of treatment at each time-point. Where US dollars are reported, the annual average exchange rate of US$1 = VND20,780 in 2013 was used. Except for CCT program cost, all cost elements were participant-level cost. As MMT treatment is an on-going intervention, the average three-year program costs were calculated by multiplying the average annual program costs by three. CCT is time-limited (two years) with no costs incurred in the third year. Therefore, the average three-year program costs (the comparative time horizon) were calculated by multiplying the average annual program costs by two. Outcome data used for this analysis was DFDs. For politically sensitive reasons, it was not possible to ask the CCT participants about their drug-using behaviors while they were in CCT centers. Yet it is common knowledge that drug use occurs in the Vietnam CCTs (Open Society Institute, 2009; The Laborer Online, 2015). Therefore, for the main analysis, proxy data were derived from
150
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
reported prevalence and frequency of illicit drug use of incarcerated populations in other countries (Bird et al., 1997; Bullock, 2003; Calzavara et al., 2003; Darke et al., 1998; Gaughwin et al., 1991; Stark et al., 2006). Specifically, the above six referenced studies provided evidence that the illicit drug use prevalence among drugusing prison inmates ranged from 31% to 74% and the frequency of use ranged from 1 to 8 days (in the 30-day time window). Derivation of a number that could be perfectly representative of these studies was impossible. Therefore, a decision was made to use a generalized average number of 26 DFDs out of 30 for imputation for 67% of the CCT participants (those who reported drug use at T1) for their two-year stay in the CCT centers. The remaining 33% of the CCT participants did not report illicit drug use at T1, hence were assumed to be less likely to have used illicit drugs while in the CCT and no imputation was therefore made. This assumption was tested in the sensitivity analyses (which confirmed that the main results were robust). All other DFDs data were derived from self-report questionnaire. A three-year time horizon was used for the CEA (Fig. 1) therefore an aggregate three-year DFDs measure was calculated for each participant (following Harris et al., 2005). This assumes that the levels in the intervening months were equal to the average of those reported at the adjacent interviews. In addition, we followed Ascher-Svanum et al. (2010) in using the predicted values (e.g., LSMEANS) of DFDs derived from mixed effects regression analysis when calculating the total change in DFDs over the three years. Predicted values were obtained from the mixed effects regression analysis of the raw data. The use of predicted values is superior to raw data because it accounts for: 1) the difference in baseline demographics characteristics; 2) the difference in baseline DFDs between participants; and 3) within-participant correlation from the repeated measurements. The formula and specific calculation of the mean ‘DFDs (over 36 months)’ for each group are presented in Supplementary material 2. 2.6. Statistical analysis Statistical analyses were performed using SAS software, version 9.4. Mixed effects regression analyses, assuming autoregressive covariance structure was used to assess between-group effects on rate of change over time in DFDs, adjusting for possible confounding factors including baseline predictors and differences at baseline on DFDs. Step-wise procedures, involving diagnosis of collinearity and univariate analysis (using Ordinary Least Squares) were used to select the most appropriate baseline predictors for the data. Nine predictors were selected and are listed in Table 2. Mixed effects regression analysis can also account for potential correlation within participant from the repeated measurements, allow for irregularlyspaced measurement across time, which is the case for the data in this study. The interaction terms of “Group” and “Time” were included in the models to check if trend over time differed by group. In mixed effects regression analysis, goodness of fit is determined by the -2 restricted log likelihood (-2RELL) for random effects and variance-covariance structure model building and -2 log likelihood (-2LL) for fixed effects modeling building. These criteria were used in determining the best fitting model as each progressive model improved upon the fit of the previous model. The goodness of fit of the final models are reported in Table 2. All statistical tests used two-sided p-values, with significance set at p < 0.05. All analyses were performed on an intention-to-treat basis. Missing data were due entirely to drop-out from the study and only pertained to outcome variables (the rates of drop-out for each time-point of both groups are presented in Fig. 1). Little’s MCAR test (Little, 1988) showed that the missing data was not missing completely at random (MCAR) (X2 = 24.598, DF = 9, p = 0.003). It is impossible to test whether the data are missing at random (MAR)
or missing not at random (MNAR). Therefore, two methods of handling missing data were used: maximum likelihood (ML) and worst case scenario (WCS). ML assumes that data are MAR and handles missing data by “borrowing” information from the values of the covariates, predictor variables and values of the outcome variable of the earlier time-points to project values for later time-points (Allison, 2012). ML is the estimation method used in mixed effects regression analysis therefore no actual imputation work is required. The WCS method assumes that the lost-to-follow-up participants had outcomes (DFDs) identical to their baseline values and is a conservative test, valid for data that are MNAR (Myers, 2000). Given the high loss-to-follow-up rate of MMT participants, WCS was used for the main analysis to ensure attrition bias was addressed. 2.7. Cost-effectiveness analysis (CEA) The central measure used in CEA is the incremental costeffectiveness ratio (ICER), which is: ICER =
Mean cost of MMT − Mean cost of CCT Mean effectiveness of MMT − Mean effectiveness of CCT
Bootstrapping was used for comparing the costs and cost-effectiveness. Non-parametric bootstrapping has gained widespread use in economic evaluation alongside clinical trials because it makes no assumptions about the underlying distribution of the data (Gray et al., 2010). The bootstrapping procedure with 5000 replications was conducted on the individual-specific total cost, the individual-specific outcome (DFDs over three years) and the ICER to generate confidence intervals. 2.8. Sensitivity analysis The robustness of the CEA results was tested in four different sensitivity analyses with more conservative assumptions on both cost and effectiveness. The first sensitivity analysis (Scenario 1A) increased the estimated MMT program costs by using the program costs of 2009 only (adjusted to 2013 by CPI), instead of using the average program cost of 2009 and 2013 as in the main analysis. The second sensitivity analysis (Scenario 2A and 2B) focused on the uncertainty around CCT program costs as some CCT centers had higher program costs than others due to different levels of investment from the government and the number of participants in these centers. The third sensitivity analysis (Scenario 3A) involved exploring the uncertainty around data on drug-using behaviors of CCT participants during their two years in CCT centers. While the main analysis assumed 26 (out of every 30 days) drug free, the sensitivity analyses assumed zero drug use for the two years (30 out of 30 days). The final sensitivity analysis (Scenario 4A) managed the uncertainty pertaining to the specification of the mixed effects regression analysis. Details on the systematic framework and the rationale for the sensitivity analyses are provided in Supplementary material 3. 3. Results The characteristics of the participants are found in Table 1. Most participants had long histories of heroin dependence and of previous treatment episodes. At baseline the CCT participants appeared to have lower heroin problems (years of use, daily heroin use, ever injected and spending on heroin), even though almost every participant in both groups was using heroin on a daily basis at baseline. A smaller proportion of CCT participants had ever sought treatment and for those who had, they experienced fewer number of treatment episodes. These differences were managed in the mixed effects regressions analyses.
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
151
Table 1 Participants’ demographics, drug use history and behavioral characteristics . CCT (n = 208)
MMT (n = 384)
P-value
Male (%) Mean Age (SD)
100 33.26 (7.60)
100 37.32 (8.23)
n/a p < 0.001
Marital status (%) Single Married, divorced, separate or widow
51.69 48.31
43.49 56.51
P = 0.04
Employment (%) Unemployed Full-time Part-time (casual work) Average legal monthly income (median & range)
23.56 17.31 59.14
33.33 21.35 45.32
P = 0.003
3.00 million (0.80–15.00 million)
1.50 million (0–35.00 million)
p < 0.001
39.90 60.10
46.09 53.91
p = 0.16
22.35 (5.79)
23.22 (6.42)
p = 0.09
85.58 10.58 4.81
80.73 16.41 2.34
p = 0.03
11.01 (6.22) 97.54 3 (0–20) VND4.55 million (0–84 million) 66.43 50.48
13.17 (4.63) 99.96 4 (0–18) VND6.20 million (0–225 million) 82.55 28.91
p < 0.001 p = 0.06 p < 0.001 p < 0.001
3 (1.25) 80.77 2 (1–50)
2 (0.82) 96.61 5 (1–40)
p < 0.001 p < 0.001 p < 0.001
37.98 66.70 35.20 16.35 18.75
49.74 84.40 30.30 26.30 12.10
p = 0.004 p < 0.001 p = 0.13 p = 0.006 p = 0.06
Education (%) ≥ High school < High school (Year 12) Mean age of 1st intoxication (SD) % drugs of first intoxication Heroin Opium Other drugs Mean number of years used (SD) Ever reported daily heroin use (%) Median heroin use frequency of those who used every day (range) Median monthly drug spending (range) % heroin injection ever % poly drug use ever *Methamphetamine second drug of choice 28.30% for CCT and 11.60% for MMT Mean number of drug classes used (SD) of those who were poly users % ever sought drug dependence treatment Median number of treatment episodes (range) *for those who ever sought treatment % ever been to CCT centers % ever home detoxed % ever committed illegal behaviors % ever been to prison % overdose incident ever
p < 0.001 p < 0.001
Note: Chi-squared test was performed for categorical and t-test was performed for continuous variables. Table 2 Mixed effects regression analysis (with ‘drug-free days (in the preceding 30 days)’ as outcome variable (SE)), ML method versus WCS method for handling missing data. Fixed Effects
ML Method
WCS Method
Model 9
Model 10 (Final Model)
Model 8
Model 9 (Final Model)
Intercept Group = CCT Time Time*Group = CCT Time*Time Time*Time*Group = CCT Age Prison history Education level Marital status Poly drug use history Ever injected heroin Ever been to CCT center Number of treatment episodes Employed Model Fit (-2LL)
5.49*** (0.52) 3.78*** (0.56) 21.87*** (0.61) −11.27***(1.03) −4.17*** (0.14) 1.67*** (0.23) 0.02 (0.01) −0.44* (0.22) 0.07 (0.10) −0.23 (0.19) −0.02 (0.19) −0.50* (0.23) −0.23 (0.19) −0.003 (0.02) −0.17 (0.20) 18082
5.97*** 3.71*** 21.88*** −11.29*** −4.16*** 1.67*** – −0.41* – – – −0.50* – – – 18082
7.27*** (0.61) 2.92*** (0.61) 17.08***(0.77) −8.85***(1.29) −3.54***(0.15) 1.34*** (0.25) 0.02 (0.01) −0.43* (0.21) 0.04 (0.18) −0.30 (0.18) −0.20 (0.19) −0.35 (0.23) −0.04 (0.18) −0.0004(0.02) −0.30 (0.20) 21743
7.55*** (0.36) 2.93*** (0.60) 17.07*** (0.77) −8.85*** (1.29) −3.54*** (0.15) 1.34*** (0.25) – 0.45* (0.21) – – –
Random Effects Residual (within-subject) variation) Between-subject intercept variation Between-subject slope variation 2.86
66.32 0
66.32 0 2.86
88.29 0 3.60
88.38 0 3.62
(0.38) (0.55) (0.61) (1.03) (0.14) (0.23) (0.22)
(0.23)
– – – 21726
Note: Mixed effects coefficients are listed under fixed effects, and variance components are listed under random effects; p< 0.05*, p< 0.01**, p< 0.001***; Standard errors in brackets.
3.1. Intervention effectiveness Using the WCS missing data imputation method for the main analysis, mixed effects regression analyses supported the
hypothesis that MMT increased the number of DFDs (in the preceding 30 days) to a greater extent than CCT ( = 8.85, SE = 1.29, p < 0.0001) (Table 2, Final Model). On average, the rate of change in the number of DFDs for one MMT participant was 8.85 days (out of
152
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
Fig. 2. Fitted trajectories of change* in “drug-free days (in the preceding 30 days)” by group: ML versus WCS imputation methods.
30 preceding days) higher compared to the rate of change for one CCT participant. Nevertheless, both CCT and MMT modalities were associated with strong effects in increasing the number of DFDs. After two years of treatment, MMT participants were able to maintain an average number of DFDs ranging from 22.97 days to 29.58 days whereas CCT participants fluctuated between 8.30 and 27.38 days (Fig. 2). The other missing data technique (ML) confirmed that MMT treatment was consistently more effective in increasing the number of DFDs across all follow-up time-points. As seen in Table 2, the ‘Time’ variables and interaction terms are statistically significant: these show, respectively that DFDs changed over time (Time); that the two groups differ in relation to change in DFDs over time in favor of MMT (Time*Group = CCT); that this change is quadratic over time (Time*Time), and finally that the curvature change differs between the two groups in favor of MMT (Time*Time*Group). Summaries of mixed effects regression model outputs (for both WCS and ML) are presented in Supplementary material 4.
3.2. Intervention cost The average annual program cost for providing treatment to one CCT participant was estimated to be VND19.67 million (US$943) which was more than double the average annual program cost for providing treatment to one MMT participant of VND7.88 million (US$378). The average participant personal costs were similarly significantly greater for CCT than MMT (see Table 3). The opportunity cost of work time accounted for the largest share of CCT participant personal costs, implying that the largest cost associated with being placed in the CCT centers was the opportunity cost of labor time forgone. The average annual total cost amounted to VND61.52 million (US$2948) for CCT and VND13.35 million (US$640) for MMT. The average thee-year total cost for CCT was VND123.04 million (US$5896), which was more than three times higher than that for MMT (VND37.31 million (US$1788)). The last row of Table 3 indicates the main bootstrap results for comparing the average three-year total costs. Specifically, in
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
153
Table 3 Program costs and participant personal costs: MMT and CCT. Unit: VND million Mean (SD)
Average annual program costs Average annual participant personal costs Average annual total costs Three-year total costs a
Bootstrap results (mean difference & CI)
MMT
CCT
7.88 (1.84) 5.47 (4.11) 13.35 (4.31) 37.31a (13.94)
19.67 (5.78) 41.85 (31.49) 61.52 (32.18) 123.04 (64.36)
Mean difference = −85.73 (CI: −76.88, −94.59)
Costs applied to months in treatment by adjustment for participant drop-outs at each time-point.
Table 4 CEA sensitivity analyses results: ‘drug-free days (over 36 months)’, using average total costs. Unit: VND Comparison
Cost MMT
Cost CCT
Cost difference
Effect MMT
Effect CCT
Effect difference
ICER (95% CI)
Main results:
37.31 mil
123.04 mil
−85.73 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
Scenario 1A: MMT most expensive: Using 2009 MMT program cost Scenario 2A: CCT most expensive: Using cost data of CCT center with highest program cost Scenario 2B: CCT least expensive: Using cost data of CCT center with smallest program cost Scenario 3A: CCT more effective at T1: No drug use in CCT centers Scenario 4A: Similar scenario as main analysis but with raw data (instead of predicted data)
45.85 mil
123.04 mil
−77.19 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
-VND0.25 mil (-VND0.34 mil to -VND0.19 mil) -VND0.22 mil (-VND0.31 mil to -VND0.17 mil)
37.31 mil
141.53 mil
−104.22 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
-VND0.31 mil (-VND0.41 mil to -VND0.24 mil)
37.31 mil
109.82 mil
−72.51 mil
512.96 DFDs
168.76 DFDs
344.20 DFDs
-VND0.21 mil (-VND0.29 mil to -VND0.16 mil)
37.31 mil
123.04 mil
−85.73 mil
507.36 DFDs
144.26 DFDs
363.10 DFDs
37.31 mil
123.04 mil
−85.73 mil
976.32 DFDs
742.59 DFDs
233.73 DFDs
-VND0.24 mil (-VND0.32 mil to -VND0.19 mil) -VND0.37 mil (-VND0.42 mil to -VND0.12 mil)
ICER: Incremental cost-effectiveness ratio. dif. = difference = MMT minus CCT. mil = million.
the main result, the average three-year total cost of MMT was VND85.73 million (US$4108) lower than CCT (95% CI: -VND76.88 million, -VND94.59 million). 3.3. Incremental cost-effectiveness ratio and cost-effectiveness plane Table 4 displays the CEA main results, which were the three-year incremental costs and the corresponding incremental effectiveness (DFDs), produced by MMT compared to CCT, using average total costs. The ICER of -VND0.25 million (US$11.97) was the product of the mean difference in average total costs (-VND85.73 million (US$4108)) divided by the mean difference in DFDs (344.20). The main results showed that the ICER for one incremental DFD was -VND0.25 million (95% CI: -VND0.34 million, -VND0.19 million). This result showed that MMT was significantly more cost-effective than CCT. The average total costs for MMT were lower than CCT (the difference was −VND85.73 million (US$4108) per participant) while achieving 344.20 more DFDs, which indicated that MMT was dominant over CCT. These results suggest that if the government of Vietnam invests in MMT treatment (instead of CCT), it will result in an additional 344.20 DFDs per average dependent heroin user plus it will cost VND85.73 million (US$4108) less for the government (for treatment) and entail lower personal costs per average dependent heroin user. The CEA results supported the hypothesis
that MMT is more cost-effective in achieving an incremental DFD compared to CCT. Table 4 presents both the main result and sensitivity analyses of the CEA using average total costs. The results of the sensitivity analyses confirmed the robustness of the main results. The cost-effectiveness planes (corresponding to the main CEA result and the sensitivity analyses results) are presented in Fig. 3, confirming that 100% of the ICERs are located in the southeast quadrant with absolute certainty (no cross-over of the ICERs with other quadrants). 4. Discussion To the best of our knowledge, this is the first economic evaluation comparing the cost effectiveness of MMT and CCT in East and Southeast Asia. Confidence in our findings is strengthened because of the use of the most conservative assumptions about both costs and effectiveness. For example, worst case scenario imputation for all those lost-to-follow-up were used in the analyses and tested in the sensitivity analyses. Thus, it is compelling that MMT treatment was dominant over CCT in the main analysis for both total costs and program costs, and that this result did not change under a variety of more conservative assumptions about both costs and effectiveness. This study provides evidence that MMT was not only less costly but also more effective compared to CCT in achieving the number of DFDs for heroin dependent users over three years in Vietnam. By investing in MMT instead of CCT, not only would one dependent
154
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
Fig. 3. Cost-effectiveness planes: CEA main result and sensitivity analyses results: ‘drug-free days (over 36 months)’, using total costs.
heroin user in Hai Phong City achieve on average 344.20 more DFDs over three years, but the government of Vietnam would also save VND17.35 million (US$831) and an average dependent heroin user would save VND68.39 million (US$3277) (VND85.73 million minus VND17.35 million) to achieve these additional 344.20 DFDs. The strength of this study is the use of empirical data (rather than modelled data) to estimate the changes in drug-using behaviors among dependent heroin users. In addition, a longitudinal study design was employed with repeated assessments of a sufficient sample that increased the level of confidence when comparing rate of change over time between groups. However, the results of these analyses should be interpreted in light of several limitations. First, the study is politically sensitive and as a result some relevant data could not be collected. This included data regarding drug-using behaviors of CCT participants while they were in CCT centers; and income generated by the CCT center management by using labor of the CCT participants. While it is unclear if those in CCT are paid for their labor and how much they are paid, the income generated by this activity for the centers could be significant. However, it is not possible to know whether this income is used to reinvest in the center activities. Second, three cost components were not included in the analysis: a) costs incurred by the CCT trainees’ families for purchasing and sending food and prescribed medicines and time to visit the trainees; b) cost of land for both treatment modalities; and b) costs of loss of freedom by CCT participants. In all three cases, these design limitations potentially bias in favour of CCT. Third, potential recall bias could be realised due to the 24-month recall window of baseline drug use behaviors among CCT participants. This limitation was identified during the design phase of this study. All interviewers were provided comprehensive training with a focus on managing recall difficulty. The interviewers gave participants adequate time to recall and/or rephrased the questions to facilitate recall. In addition, the literature suggests that although the absolute levels obtained through retrospective self-report may not be as accurate as one would desire, the relative levels are quite valid (Buavirat et al., 2003; Hser and Wickens, 1989; Napper et al., 2010). For example, one study evaluating the agreement between two recall points of 10-year interval found that the agreement was exceptional with the correlation coefficients ranging as high as 0.86 and 0.90 (Anglin et al., 1993). For data in this study, both Table 1 and
Fig. 2 suggest that under-reporting of drug use behaviors at baseline by CCT participants was unlikely given the substantial and similar level of drug use compared to MMT participants (mean number of years using drugs, daily heroin use, monthly drug spending, number of DFDs). Finally, while the findings of this study confirmed that MMT was more cost-effective than CCT, this was a single study conducted in one city in Vietnam. The findings of this study should be confirmed through further economic evaluation studies in other cities and countries in East and Southeast Asia. Conflict of interest FHI360 Vietnam and Hanoi Medical University receive funding from the United States government agencies to provide technical assistance to the Vietnamese government for the operational and clinical aspects of the national methadone treatment program in Vietnam. Robert Ali has received untied educational grants from Reckitt Benckiser and Mundipharm for the post-marketing surveillance of opiate substitution treatment (OST) medications in Australia and untied educational grants from Reckitt Benckiser for conducting a study into the pharmacogenetics predictors of treatment success. Contributors All authors participated in different important roles the study, which provided the content for this manuscript. Author Thu Vuong drafted the manuscript. All authors contributed to and have approved of the final manuscript. Author disclosures The study received funding from Atlantic Philanthropies through FHI360 Vietnam and Hanoi Medical University. Other sources of funding were the Australian Government’s Endeavour PhD Scholarship and UNSW Australia/NDARC PhD Scholarship for Thu Vuong. Alison Ritter is funded through an NHMRC Senior Research Fellowship (APP1021988). Marian Shanahan is supported by funding from the Australian Government under the Substance Misuse Prevention and Service Improvements Grants Fund.
T. Vuong et al. / Drug and Alcohol Dependence 168 (2016) 147–155
Acknowledgements The study team acknowledges the invaluable contribution of the researchers at the Hai Phong Medical University for their dayto-day field work. However the views expressed are those of the authors and should not be attributed to any other agency or individual. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.drugalcdep.2016. 09.008. References Allison, P.D., 2012. Handling missing data by maximum likelihood. Proceedings of the SAS Global Forum Vol. 312. Amon, J., Pearshouse, R., Cohen, J.E., Schleifer, R., 2014. Compulsory drug detention in East and Southeast Asia: evolving government, UN and donor responses. Int. J. Drug Policy 25, 13–20. Anglin, M.D., Hser, Y.-I., Chou, C.-P., 1993. Reliability and validity of retrospective behavioral self-report by narcotics addicts. Eval. Rev. 17, 91–108. Ascher-Svanum, H., Stensland, M.D., Peng, X., Faries, D.E., Stauffer, V.L., Osuntokun, O.O., Kane, J.M., 2010. Cost-effectiveness of olanzapine vs. aripiprazole in the treatment of schizophrenia. Curr. Med. Res. Opin. 27, 115–122. Bird, A.G., Gore, S.M., Hutchinson, S.J., Lewis, S.C., Cameron, S., Burns, S., 1997. Harm reduction measures and injecting inside prison versus mandatory drugs testing: results of a cross sectional anonymous questionnaire survey. BMJ 315, 21–24. Birgin, R., 2013. Arguments against the compulsory treatment of opioid dependence. Bull. World Health Organ. 91, 239–239a. Buavirat, A., Page-Shafer, K., Van Griensven, G., Mandel, J., Evans, J., Chuaratanaphong, J., Chiamwongpat, S., Sacks, R., Moss, A., 2003. Risk of prevalent HIV infection associated with incarceration among injecting drug users in Bangkok, Thailand: case-control study. BMJ 326, 308. Bullock, T., 2003. Changing Levels Of Drug Use Before, During And After Imprisonment. Home Office Research Study, London, pp. 23–48. Calzavara, L.M., Burchell, A.N., Schlossberg, J., Myers, T., Escobar, M., Wallace, E., Major, C., Strike, C., Millson, M., 2003. Prior opiate injection and incarceration history predict injection drug use among inmates. Addiction 98, 1257–1265. Clark, N., Busse, A., Gerra, G., 2013. Voluntary treatment, not detention, in the management of opioid dependence. Bull. World Health Organ. 91, 146–147. Connock, M., Juarez-Garcia, A., Jowett, S., Frew, E., Liu, Z., Taylor, R.J., Fry-Smith, A., Day, E., Lintzeris, N., Roberts, T., Burls, A., Taylor, R.S., 2007. Methadone and buprenorphine for the management of opioid dependence: a systematic review and economic evaluation. Health Technol. Assess. 11, 1–171. Darke, S., Kaye, S., Finlay-Jones, R., 1998. Drug use and injection risk-taking among prison methadone maintenance patients. Addiction 93, 1169–1175. Gaughwin, M.D., Douglas, R.M., Liew, C., Davies, L., Mylvaganam, A., Treffke, H., Edwards, J., Ali, R., 1991. HIV prevalence and risk behaviours for HIV transmission in South Australian prisons. AIDS 5, 845–852. General Statistics Office of Vietnam, 2013. Consumer Price Index. General Statistics Office of Vietnam, Hanoi. Gray, A.M., Clarke, P.M., Wolstenholme, J.L., Wordsworth, S., 2010. Applied Methods Of Cost-Effectiveness Analysis In Healthcare. Oxford University Press, Oxford. Hall, W., Babor, T., Edwards, G., Laranjeira, R., Marsden, J., Miller, P., Obot, I., Petry, N., Thamarangsi, T., West, R., 2012. Compulsory detention, forced detoxification and enforced labor are not ethically acceptable or effective ways to treat addiction. Addiction 107, 1891–1893. Harris, A.H., Gospodarevskaya, E., Ritter, A., 2005. A randomised trial of the cost effectiveness of buprenorphine as an alternative to methadone maintenance treatment for heroin dependence in a primary care setting. Pharmacoeconomics 23, 77–91. Hser, Y.I., Wickens, T.D., 1989. The effects of the spacing of test trials and study trials in paired-association learning. J. Educ. Psychol. 9, 99–120. Inter-Parliamentary Union, 2010. Representation From The Top: Ethnic Minorities In The National Assembly Of Viet Nam. Hanoi. Johnson, K.W., Young, L., Shamblen, S., Suresh, G., Browne, T., Chookhare, K.W., 2012. Evaluation of the therapeutic community treatment model in Thailand: policy implications for compulsory and prison-based treatment. Subst. Use Misuse 47, 889–909.
155
Kausikan, B., 1993. Asia’s different standard. Foreign Policy Washingtonpost 92, 24–41. Little, R.J., 1988. A test of missing completely at random for multivariate data with missing values. J. Am. Stat. Assoc. 83, 1198–1202. Myers, W.R., 2000. Handling missing data in clinical trials: an overview. Drug Inf. J. 34, 525–533. Naning, H., Kerr, C., Kamarulzaman, A., Osornprasop, S., Dahlui, M., Ng, C.-W., Wilson, D., 2014. Return On Investment And Cost-Effectiveness Of Harm Reduction Program In Malaysia, http://documents.worldbank.org/curated/en/ 2014/07/19764976/return-investment-cost-effectiveness-harm-reductionprogram-malaysia (accessed 18.10.14.). Napper, L.E., Fisher, D.G., Reynolds, G.L., Johnson, M.E., 2010. HIV risk behavior self-report reliability at different recall periods. AIDS Behav. 14, 152–161. Ni, M.J., Fu, L.P., Chen, X.L., Hu, X.Y., Wheeler, K., 2012. Net financial benefits of averting HIV infections among people who inject drugs in Urumqi, Xinjiang, Peoples Republic of China (2005–2010). BMC Public Health 12, 572–579. Open Society Institute, 2009. At What Cost? HIV And Human Rights Consequences Of The Global ‘War On Drugs’. International Harm Reduction Development Program, Public Health Program. Open Society Institute, New York. Salomé, H.J., French, M.T., Miller, M., McLellan, A.T., 2003. Estimating the client costs of addiction treatment: first findings from the client drug abuse treatment cost analysis program (Client DATCAP). Drug Alcohol Depend. 71, 195–206. Stark, K., Herrmann, U., Ehrhardt, S., Bienzle, U., 2006. A syringe exchange programme in prison as prevention strategy against HIV infection and hepatitis B and C in Berlin, Germany. Epidemiol. Infect. 134, 814–819. The Laborer Online, 2015. Tham lau ma tuy vao trai giam khong con la chuyen la (smuggling illicit drugs into closed settings is not new). http://nld.com.vn/ thoi-su-trong-nuoc/tham-lau-ma-tuy-vao-trai-giam-khong-con-la-chuyen-la20150414122935841.htm (accessed 12.04.15.). Tran, B.X., Ohinmaa, A., Duong, A.T., Do, N.T., Nguyen, L.T., Mills, S., Houston, S., Jacobs, P., 2012a. Cost-effectiveness of methadone maintenance treatment for HIV-positive drug users in Vietnam. AIDS Care 24, 283–290. Tran, B.X., Ohinmaa, A., Duong, A.T., Nguyen, L.T., Vu, P.X., Mills, S., Jacobs, P., 2012b. The cost-effectiveness and budget impact of Vietnam’s methadone maintenance treatment programme in HIV prevention and treatment among injection drug users. Global Public Health 7, 1080–1094. – nh soˆ´ 5076/QÐ-BYT, 2007, ban hành Vietnam Ministry of Health, 2007. Quyeˆ´ t di . – eˆ` u tri thay theˆ´ nghiˆen các chaˆ´ t dang thuoˆ´ c phiˆen ba˘` ng thuoˆ´ c ´ hu ong da˜ˆ n di . . . .
ˆ chuc ´ ´ thu. c hiˆe.n (English translation: ’Ministry of da˜ˆ n to methadone và hu ong Health Decision no. 5076/QD-BYT/2007, providing guideline on implementation of methadone maintenance treatment’). http:// thuvienphapluat.vn/van-ban/The-thao-Y-te/Quyet-dinh-5076-QD-BYT-2007huong-dan-dieu-tri-thay-the-nghien-do-Bo-Y-te-ban-hanh-205168.aspx (accessed 21.09.15.). – eˆ` u tri Vietnam Ministry of Health, 2015. Cˆa.p nhˆa.t tình hình thu. c hiˆe.n chı tiêu di . nghiˆe.n các chaˆ´ t da.ng thuoˆ´ c phiˆe.n ba˘` ng Methadone (MMT) d–eˆ´ n tháng 10 n˘am 2015 (English translation: ’Update of progress in expansion of MMT program until October 2015’) by the Minitry of Health (MOH) of Vietnam. http://vaac. gov.vn/solieu/Detail/?userkey=Cap-nhat-tinh-hinh-thuc-hien-chi-tieu-dieutri-nghien-cac-chat-dang-thuoc-phien-bang-Methadone-MMT-den-thang-9nam-2015 (accessed 15.11.15.). Vietnam Ministry of Labor Invalid and Social Affairs, 2013. Renovation Plan for Drug Addiction Treatment for 2013–2020 in Vietnam. Ministry of Labor, Invalid and Social Affairs (MOLISA), Hanoi. Vuong, T., Ritter, A., 2013. Not to lose sight of the bigger picture: the case of therapeutic communities (TC) research in closed settings in Thailand. Subst. Use Misuse 48, 793–794. Wammes, J.J., Siregar, A.Y., Hidayat, T., Raya, R.P., van Crevel, R., van der Ven, A.J., Baltussen, R., 2012. Cost-effectiveness of methadone maintenance therapy as HIV prevention in an Indonesian high-prevalence setting: a mathematical modeling study. Int. J. Drug Policy 23, 358–364. World Health Organization, 2009. Assessment Of Compulsory Treatment Of People Who Use Drugs In Cambodia, China, Malaysia And Viet Nam: An Application Of Selected Human Rights Principles. A Report By WHO Western Pacific Regional Office In Manila , WHO, Geneva. Wu, Z., 2013. Arguments in favor of compulsory treatment of opioid dependence. Bull. World Health Organ. 91, 142–145. Xing, Y., Sun, J., Cao, W., Lee, L., Guo, H., Li, H., Duan, S., 2012. Economic evaluation of methadone maintenance treatment in HIV/AIDS control among injecting drug users in Dehong, China. AIDS Care 24, 756–762.