The cost-effectiveness of adapting and implementing a brief intervention to target frequent alcohol use among persons with HIV in Vietnam

Abstract

Brief interventions to reduce frequent alcohol use among persons with HIV (PWH) are evidence-based, but resource-constrained settings must contend with competition for health resources. We evaluated the cost-effectiveness of two intervention arms compared to the standard of care (SOC) in a three-arm randomized control trial targeting frequent alcohol use in PWH through increasing the percent days abstinent from alcohol and viral suppression. We estimated incremental cost per quality-adjusted life year [QALY] gained from a modified societal perspective and a one-year time horizon using a Markov model of health outcomes. The two-session brief intervention (BI), relative to the six-session combined intervention (CoI), was more effective and less costly; the estimated incremental cost-effectiveness of the BI relative to the SOC, was $525 per QALY gained. The BI may be cost-effective for the HIV treatment setting; the health utility gained from viral suppression requires further exploration.

Introduction

In 2013 the Joint United Nations Programme on HIV/AIDS (UNAIDS) called for a 90-90-90 goal by 2020: 90% of PWH know their HIV status through testing, 90% of those persons initiating care; and 90% of those initiating care being on antiretroviral therapy(1). Behavioral interventions incorporating cognitive-behavioral therapy (CBT) and motivational enhancement therapy (MET) have been demonstrated to be effective at reducing hazardous alcohol use among PWH, thus improving adherence and maintenance in care (2). Among PWH, alcohol use is associated with lower adherence to HIV medications (3-5). Ultimately, poor medication adherence and alcohol use both have impacts not only on longevity, but also on quality of life and productivity (6).

Behavioral interventions are effective at reducing hazardous alcohol use among PWH(2), yet they can be cost intensive, partly due to extensive training of staff and start-up costs. The World Health Organization (WHO) has called for the implementation of behavioral interventions aimed at all persons with hazardous alcohol use (7) and deems economic analyses as essential for policymakers to determine efficient allocation of health monies. Coupled with this, shifting funding priorities among global donors and programs such as the US-based President Emergency Plan for AIDS Relief (PEPFAR) have impacted in-country budgets and allocation of healthcare resources (8). In Vietnam, for example, external funding fueled dramatic growth in the number of ART clinics since 2005 (9), whereas evidence-based alcohol prevention programs have not been integrated (10).

As such, brief interventions, defined in a Cochrane database systematic review as one to five educational sessions (by phone or in-person) (11), have gained popularity in primary care and emergency rooms in high-resource settings. Interest in expanding these models to resource-constrained settings have increased in recent years. Such is the case in Vietnam where from 2016-2018, the Reducing Hazardous Alcohol Use & HIV Viral Load (REDART) randomized control trial was implemented to compare a brief intervention format and longer combined CBT/MET-based intervention to reduce alcohol use among PWH receiving antiretroviral therapy in Thai Nguyen, Vietnam. Recent results from Go and colleagues of the REDART randomized control trial (RCT) identified both behavioral interventions as effective in reducing frequent and heavy alcohol use. In this trial Go and colleagues found that 65% of participants in either intervention reported alcohol abstinence at the 12-month follow up visit compared to 50% among those who received no intervention (12). Furthermore, the authors found that those receiving the brief intervention had significantly higher viral suppression at 12-month follow-up than those receiving the standard of care.

Compared to traditional behavioral interventions applying CBT and MET methods, brief interventions require less direct time with a patient from the health care staff delivering the intervention, and therefore may be less costly (13). Furthermore, the brief format of the behavioral intervention, while still requiring training, likely require fewer resources to implement. In resource-constrained HIV clinic settings, limited evidence exists on the cost-effectiveness of this brief intervention compared to more traditional interventions; only one study has been done to assess the cost-effectiveness of brief interventions for individuals who have co-occurring alcohol use disorders and HIV (14). Kessler and colleagues estimated an incremental cost-effectiveness ratio (ICER) of $400 for a CBT-informed intervention integrated into HIV care in East Africa.

In order to better understand the overall potential budgetary impacts on the healthcare system and the intervention-related activities requiring the greatest costs this study extends the trial results of Go et al., to calculate and compare the cost-effectiveness of the REDART brief intervention and a combined intervention to the standard of care. Reflecting previous research on cost-effectiveness in alcohol intervention programs, a Markov model allowed for us to reflect the cyclical nature of alcohol use (15, 16). Thus, we constructed a Markov model with monthly cycles and a one-year time horizon, simulating 1000 persons per trial arm examining changes in cost and in quality of life among persons with frequent drinking and receiving antiretroviral therapy (ART) for HIV in Vietnam.

Methods

Overview

We estimated the cost per health-related gains of the REDART RCT. We compared costs and cost-effectiveness of three alcohol use reduction strategies used in the three arms of the trial: the brief intervention (BI), the combined CBT/MET intervention (CoI), and the standard of care (SOC). Using the effectiveness data from the primary trial and the incremental costs of implementing each of the two behavioral interventions, we built a Markov Model of the effect of these two alcohol interventions on a hypothetical cohort of persons with HIV receiving anti-retroviral therapy (ART) and engaging in frequent drinking, and estimated the costs associated with such an intervention over a one-year time horizon. For this simulated cohort, we applied a modified societal perspective that included not only costs to the health system but opportunity costs (e.g. lost wages due to attending appointments) of patients to determine the cost-effectiveness of the two intervention arms as compared to the standard of care.

Primary Trial Population and Study Design

The REDART trial outcomes and methods have been described elsewhere (12, 17). Briefly, the primary trial population were persons living with HIV (PWH) on ART and receiving care from one of seven government-run ART clinics in Thai Nguyen, a semi-mountainous province in Northern Vietnam. From March 2016 to May 2017, all persons who accessed care in the study clinics were informed of the study by their clinician and referred to study staff if they expressed interest. Study staff assessed eligibility using the Alcohol Use Disorders Identification Test-Consumption (AUDIT-C); eligible participants were those with at-risk alcohol use, defined as a score greater than 4 for men or greater than 3 for women (18). A total of 1559 persons were screened and 440 persons enrolled; participants were then administered a timeline follow-back (TLFB) interview by a trained staff member who queried number of alcoholic drinks per day in the past 30 days and converted these to standard drinks. The TLFB is considered a robust measure of alcohol use, though is time-intensive (i.e. requires 30-60 minutes of participant’s time) (19); thus the AUDIT-C was used for screening. Participants were randomized in a 1:1:1 ratio to the SOC, the BI, or the CoI. The current SOC in Vietnam for alcohol use treatment in ART clinics is a verbal referral to a substance use provider; for simplicity in this analysis, we assumed that the costs of such referrals (and subsequent treatment) would be similar in all three arms. The interventions, culturally adapted for this setting through a process of core content identification and tailoring for relevance to the clinical setting, included an intensive six-session CBT and MET combined intervention, a shorter four-session (2 in-person + 2 phone call) brief intervention, or the provider-referral standard of care (verbal referral to external alcohol treatment resources)(17). Additional detail of each intervention is provided in Appendix I. Of the persons enrolled (n=440) in the main trial, 97% were men, the median age was 40 years (IQR 36-44), 54% reported working full-time, and 84% were virally suppressed (defined as less than 20 copies of HIV-1 RNA per millimeter) at baseline. At 12 months, 89% of participants completed their final visit.

Cost Measures

Methods of collecting and analyzing the costs of the REDART trial have been described elsewhere (20). The incremental costs of the two interventions were estimated from a modified societal perspective, as described above, and using a micro-costing (i.e. ingredients-based or “bottom-up”) approach. Micro-costing involves “direct enumeration and costing of every input consumed in the treatment of a particular patient” (21). Such an approach can be compared to gross (“top-down”) costing in which high-level costs are estimated but not decomposed into their component elements (22). Patient opportunity costs were estimated from questionnaire data completed by the primary trial participants Time-reported measures included the amount of time spent traveling to and from appointments as well as time spent in the clinic. The mean reported salary was $38 per week and mean reported lost wages per clinic visit (from missed work) were $0.75 (13). Human resource costs from the health system perspective were estimated by direct observation using time-and-motion studies (23) performed on three different work days per counselor, including minutes spent on individual activities such as preparing for counseling sessions, completing paperwork for follow-up, and traveling to appointments. Detailed budget reports were collected with itemized building and supply costs including room rentals, monthly utilities, and computers. To demonstrate the additional impact on HIV-treatment services due to the ART clinic setting, we incorporated HIV treatment costs. Thus we estimated the annual cost of ART in Vietnam from the published literature (24) but did not include costs for HIV testing. Furthermore, since existing literature is mixed on whether adherence increases or decreases costs from the patient or provider perspectives (25), we assumed in the base case that costs were not affected by adherence. We did not include costs associated with the standard of care (referral by a provider) because it is existing in the Vietnam healthcare infrastructure. All cost data were collected in the local currency, Vietnamese Dong (VND), and then converted to United States Dollars (USD) based on 2015 currency conversion rates, as this was the year in which training and manual development began.

Model Outcomes

In line with the effectiveness outcomes of the primary trial (12), we modeled a one-year time horizon of intervention effectiveness on change in number of persons with alcohol abstinence and change in person-years of viral suppression. Alcohol use as measured via self-report using the TLFB in the primary trial, and then categorized into three drinking categories (abstinent, semi-frequent, and frequent). Additional details about these categories are provided below. Viral suppression was measured through blood sample collection at three-month intervals, with a viral load <20 copies/mL defined as virally suppressed(26).

Quality of Life and Quality-Adjusted Life Years

Quality-adjusted life years (QALYs) can be used as outcome to aggregate the impact of a health intervention and allow for standard comparison of health outcomes across different studies(27). In this study, we incorporated quality of life gains from both change in drinking status and change in viral suppression status. Participants’ quality of life based on their drinking status was measured by self-reported health utility, using the five-level EuroQol five-dimensional (EQ-5D-5L) (28, 29) at the 6-month follow-up visit. To convert EQ-5D-5L responses to an individual health utility score, we used the Crosswalk Index Value Calculator, as developed by the EuroQol group (30, 31). A value is derived from a nationally representative sample such that individuals put a value on their health, and then those values are standardized to a particular population to reflect the cultural differences of different regions around the world. Currently there is no standardized value set for Vietnam, but using the guidance provided by the EuroQol Group and a previously published Vietnam-based study (32), we used the standardized value set for Thailand, given its economic, geographic and cultural proximity to Vietnam.

We also sought to include a measure of quality of life gained from becoming virally suppressed. Given our limited sample size and high percentage of individuals virally suppressed at baseline (which we took to be 84% equally in all three groups), we sought estimates of quality of life years gained from viral suppression using secondary data so as to better generalize to a non-virally suppressed population. We converted months of viral suppression into estimated QALYs using data from a published cost-effectiveness analysis of US-based PLWH engaged in care (33). This study estimated that a two-year increase in viral suppression of 3.8% equated to a gain of 0.04 QALYs over a lifetime time horizon. We therefore assumed a 0.5 QALY gain per year of viral suppression.

Markov Model Structure and Cost effectiveness Analysis

For this study, we designed a novel Markov state-transition model of alcohol behavior and viral suppression to estimate the incremental change in cost over the incremental change in quality adjusted life years of the two interventions relative to the standard of care. We chose to build a Markov model because it allows for representation of how behavior changes over a specified period of time and is more easily interpretable when that particular health behavior, such as drinking, can fluctuate in short periods of time. A Markov Model typically consists of multiple “disease states,” distinct categories of individuals for a certain health issue or disease who can “transition” in and out of these disease states. The Markov model consisted of six states, with two states for viral suppression (suppressed and not suppressed), and three drinking states, based on % days alcohol abstinent in past 30 days (primary trial outcome), defined as: abstinent (100% abstinent), semi-frequent use (>=50% abstinent), and frequent use (<50% abstinent). The decision to define three drinking states reflects the distribution of drinking among trial participants and the literature on applying Markov models to alcohol use (16, 34). We derived transition probabilities from the primary trial, in which data were collected at four time points over a one-year follow-up period (baseline, 3 months, 6 months, and 12 months). Transition probabilities were calculated during each interval (0-3 months, 3-6 months, 6-12 months) using data from those participants who completed the start and end questionnaires for each interval (See Table I). We converted these data into monthly transition probabilities by assuming a constant transition probability in each contributing month such that the cumulative probability at the end of each interval matched the observed data distribution. We also used trial data – conditional on drinking status – to calculate monthly probabilities of viral suppression in the same fashion. The annual cost of ART was the same across all three arms. We assumed that ART-related costs were independent of viral suppression status, reflecting the literature from Vietnam showing that full-panel testing and in-person clinic visits remain the same for all patients even when becoming virally suppressed (35). We did not include an additional cost of routine monitoring of viral load as this is not reflective of the approach in Vietnam (36). These assumptions were evaluated in sensitivity analyses (see below). We applied the cost of the combined intervention as an incremental cost in the combined intervention arm and the cost of the brief intervention as an incremental cost in the brief intervention arm. The incremental cost-effectiveness ratio (ICER) was calculated by dividing the incremental cost by the incremental QALYs, with the result of the cost per QALY gained. Standard ICER thresholds in Vietnam do not currently exist, but in order to align our sensitivity analyses with approaches taken in other published economic analyses from Vietnam we determined an ICER threshold of $2000 USD per QALY gained (approximately the 2015 per capita GDP of Vietnam). This approach reflects the current literature on setting ICER thresholds and other published cost-effectiveness studies from Vietnam (37).

Table I.

Transition probabilities of the drinking states for the three trial arms

Standard of Care
	Frequent 3-months	Semi-Frequent 3-months	Abstinent 3-months
Baseline
Frequent baseline	0.77	0.19	0.04
Semi-Frequent Baseline	0.28	0.69	0.03
Abstinent Baseline	0	1	0
	Frequent 6-months	Semi-Frequent 6-months	Abstinent 6-months
3-Months
Frequent 3-months	0.74	0.25	0.01
Semi-Frequent 3-months	0.17	0.80	0.04
Abstinent 3-months	0.20	0.40	0.40
	Frequent 12-months	Semi-Frequent 12-months	Abstinent 12-months
6-months
Frequent 6-months	0.77	0.23	0.00
Semi-Frequent 6-months	0.22	0.71	0.06
Abstinent 6-months	0.00	0.50	0.50
Brief Intervention
	Frequent 3-months	Semi-Frequent 3-months	Abstinent 3-months
Baseline
Frequent baseline	0.41	0.50	0.09
Semi-Frequent Baseline	0.12	0.66	0.22
Abstinent Baseline	0.00	1.00	0.00
	Frequent 6-months	Semi-Frequent 6-months	Abstinent 6-months
3-Months
Frequent 3-months	0.68	0.32	0
Semi-Frequent 3-months	0.18	0.76	0.07
Abstinent 3-months	0	0.56	0.44
	Frequent 12-months	Semi-Frequent 12-months	Abstinent 12-months
6-Month
Frequent 6-months	0.68	0.29	0.03
Semi-Frequent 6-months	0.08	0.83	0.09
Abstinent 6-months	0.23	0.38	0.38
Combined Intervention
	Frequent 3-months	Semi-Frequent 3-months	Abstinent 3-months
Baseline
Frequent baseline	0.48	0.51	0.01
Semi-Frequent Baseline	0.13	0.78	0.09
Abstinent Baseline	0.00	0.50	0.50
	Frequent 6-months	Semi-Frequent 6-months	Abstinent 6-months
3-Months
Frequent 3-months	0.82	0.16	0.02
Semi-Frequent 3-months	0.71	0.29	0.00
Abstinent 3-months	0.00	0.29	0.71
	Frequent 12-months	Semi-Frequent 12-months	Abstinent 12-months
6-months
Frequent 6-months	0.73	0.27	0.00
Semi-Frequent 6-months	0.10	0.85	0.05
Abstinent 6-months	0.07	0.53	0.40

Frequent – at last a drink every day in the past 30 days, Semi-frequent – fewer than 4 drinks a week in the past 30 days, Abstinent –no drinks in the past 30 days

Transition probabilities calculated from those who have provided start and end follow-up period: Standard of Care, baseline to 3-months (n=134), 3-months to 6-months (n=128), 6-months to 12-months (n=124); Brief Intervention, baseline to 3-months (n=135), 3-months to 6-months (n=130), 6-months to 12-months (n=126); Combined Intervention, baseline to 3-months (n=135), 3-months to 6-months (n=134), 6-months to 12-months (n=134)

Furthermore, we conducted an alternative scenario analysis in which we assumed equivalent transition probabilities in all arms, starting in month 12, such that the populations in each arm equalized over time (with any differences beyond 12 months reflecting pre-existing differences in the population at 12 months). We undertook this approach to evaluate the long-term impacts of one-time delivery of the intervention despite not having trial data beyond 12 months. We assumed in this scenario that after 12 months all persons would behave as if they had received the standard of care (no intervention effects). We evaluated this scenario over a 10-year time horizon, with the economic theory standard of 3% annual discounting of costs and effects to reflect decreasing values over time.

Uncertainty and Sensitivity Analyses

We conducted one-way sensitivity analyses to identify the key determinants of cost-effectiveness and probabilistic uncertainty analysis to characterize uncertainty in our estimates. We varied key model input parameters that had uncertainty in their values across reasonable ranges and evaluated the corresponding changes in cost-effectiveness estimates. Upper and lower bounds of these ranges were based on trial data when available, and on expert opinion and the literature otherwise. Details of the key input parameters, including sources, are included in Table II. We also conducted a sensitivity analysis in which we assumed a 25% reduction in ART costs for those not virally suppressed, based on findings from Gardner et al (38).

Table II.

Key inputs for the model of the brief intervention and combined intervention compared to the standard of care

Parameter	Value	Range	Source
Age	40	36, 44	REDART
% Virally Suppressed	84%	73, 95	REDART
Brief Intervention cost per person (Patient and Health Care System)	$33	$25, $41	REDART
Combined Intervention cost per person (Patient and Health Care System)	$79	$69, $89	REDART
Probability of transitioning out of frequent drinking by three months, no intervention	0.23	0.21, 0.24	REDART
Odds Ratio (OR) of Frequent to non-Frequent Drinking, Brief Intervention vs. Standard of Care	2.57	2.25, 2.90	REDART
Health Utility of Frequent Drinking	0.94	0.91, 0.96	REDART
Health Utility of Semi-Frequent Drinking	0.94	0.91, 0.96	REDART
Health Utility of Abstinent Drinking	0.95	0.92, 0.97	REDART
Annual ART Cost of Established Care (per patient)	$335	$330, $414	Duong 2014

A probabilistic sensitivity analysis was done to assess uncertainty in the values in which we used beta distributions for health utilities and proportions, lognormal distributions for odds ratio, and gamma distributions for cost parameters to run a Monte Carlo simulation (39).

Data were analyzed in SAS 9.4 (SAS Institute, Cary, NC), the model was constructed in Excel 2016 (Microsoft, Redmond, WA) and sensitivity analyses were performed using Crystal Ball (Oracle, Redwood Shores, CA).

Ethics Statement

The study was approved by the University of North Carolina Chapel Hill Office of Human Research Ethics Institutional Review Board (IRB) (approval number 14-0225) and the IRB for Biomedical Research of the Thai Nguyen Centers for Disease Control. Registered with Clinicaltrials.gov, NCT02720237.

Results

All participants in the primary trial screened positive for at-risk alcohol use at entry using the AUDIT-C; by the start of the intervention drinking behavior varied as measured by the TLFB. As such, based on TLFB results from the REDART trial, we modeled a simulated cohort where at the start of analysis there were an estimated 55% of persons in the frequent drinking state (<50% days abstinent), 44% in the semi-frequent drinking state (50-99% days abstinent), and 1% in the abstinent drinking state (100% days abstinent). Reflecting data from the primary trial, the two modeled interventions were effective in reducing frequent drinking by one-year post-intervention. In the Markov model, persons who were in the frequent drinking category at the start of the analysis had a 0.77 probability of remaining in the frequent drinking category by three months post-intervention if they only received the standard of care, versus 0.44 probability of remaining in the frequent drinking category for the brief intervention and 0.48 probability of remaining in the frequent drinking category for the combined intervention.

We identified drinking-related health utility from the REDART trial results of the 6-month follow-up visit where participants completed the EQ-5D-5L; scores did not vary substantively by drinking status in this population and were reflective of other studies in HIV populations (40); health utility for those who were abstinent was slightly higher (0.95) as compared those who were frequent (0.94) or semi-frequent drinkers (0.94) (Table I).

When the transition probabilities derived from the primary trial were applied to the 1000-person simulated cohort per trial arm (for a total of 3000 persons), we determined that at 12-months post-intervention, 763 persons receiving the standard of care would be virally suppressed as compared to 911 persons receiving the brief intervention and 834 persons receiving the combined intervention (Appendix II). The total person years of viral suppression was lower for those receiving the standard of care (841 person-years) than among those receiving the brief intervention (869 person-years) or the combined intervention (852 person-years).

For a hypothetical cohort of 1000 persons receiving each of the interventions, the estimated cost for the standard of care consisted of the one year cost of anti-retroviral therapy (ART) alone was $335,000, the estimated total cost, including ART, of the brief intervention was $374,000, and the estimated cost of the combined intervention, including ART, was an estimated $430,000 (Table II). The incremental cost of the brief intervention as compared to the standard of care was $39,000 and the incremental cost of the combined intervention as compared to the brief intervention was an estimated $56,000.

Compared to the standard of care, the incremental cost-effectiveness of the brief intervention, the change in total cost divided by the change in total QALYs, was estimated at $525 per QALY gained. This figure increased to $692 per QALY gained under the assumption that those who were not virally suppressed incurred only 75% of the ART costs of those who were virally suppressed. The brief intervention was less costly and more effective than the combined intervention in the one-year time horizon.

When we conducted a scenario analysis of a 10-year time horizon, discounting at 3% both costs and QALYs gained from alcohol abstinence and incorporating the discounting for QALYs gained from viral suppression, the estimated cost-effectiveness of the brief intervention improved to $102 per QALY gained (Table III). The result was largely due to the one-time delivery of the brief intervention (in year one), discounting, and equal costs across all three arms attributed to HIV treatment. The combined intervention remained both less effective and more expensive than the brief intervention, and thus was dominated by the other two arms.

Table III.

Incremental Cost-effectiveness of Behavioral Interventions to Reduce Alcohol Use in PWH over a one-year time period

	Cost of Intervention	Cost of ARTa	Total Cost	Incremental Cost	Total Person- Years of Viral Suppression	Total QALYsb	Incremental QALYs	ICERc
Standard of Care		$335,000	$335,000		837	902
Brief Intervention	$39,000	$335,000	$374,000	$39,000	869	976	74	$525
Combined Intervention	$95,000	$335,000	$430,000	$56,000	852	939	(38)	DOMINATED

^aAdjusted for 2015 USD

^bQuality-adjusted life years (QALYs)

^cIncremental cost-effectiveness ratio (ICER) is the incremental cost divided by the incremental QALYs

For the one-way sensitivity analysis, the parameters most impacting the model were the quality of life years gained from viral suppression and the cost of the brief intervention. The health utility of frequent drinking also had an effect on incremental cost-effectiveness ratio estimates given the effectiveness of the brief intervention impact on frequent alcohol use by one-year follow up (Figure I).

Figure 1.

One-way sensitivity analysis of the uncertainty of key inputs and their impact on the change in cost per quality-adjusted life year (QALY) gained from the brief intervention (BI) compared to the standard of care with a one-year time horizon. The base case incremental cost-effectiveness ratio (ICER) is $525 per QALY gained and is reflected in the bold vertical line. Each horizontal bar represents a key variable adjusted within reasonable ranges and the impact on the ICER. Viral suppression is defined as a viral load <20 copies/mL.

In the Monte Carlo simulation (probabilistic sensitivity analysis) we found that the brief intervention as compared to the standard of care fell below the cost-effectiveness threshold of $2000 per QALY gained in 93% of simulations (Figure II).

Figure II.

Probabilistic sensitivity analysis of the cost per quality-adjusted life year gained for the brief intervention compared to the standard of care using a 1000 trial Monte Carlo simulation

Discussion

We sought to determine the relative cost-effectiveness of two behavioral interventions designed to reduce frequent alcohol use among PWH on ART, one a combined MET and CBT approach and the other a brief intervention, compared to the current standard of care. In the Markov model, the hypothetical cohort of 1000 persons receiving the brief intervention gained 74 QALYs (from change in alcohol use and viral suppression) over the standard of care. We determined the incremental cost of the brief intervention above the standard of care was $39,000, resulting in approximately $525 dollars spent for each additional QALY gained. The combined intervention approach resulted in both higher cost and less impact on quality of life than the brief intervention, and thus was considered a “dominated” strategy in standard cost-effectiveness terms.

In this setting, northern Vietnam, there is no agreed upon metric for defining a “cost-effective program”. One approach, however, is setting a threshold. Thresholds of cost-effectiveness have been applied in Vietnam (37, 41, 42) despite ongoing debate in the literature of how useful they can be(43). The WHO recommends an ICER threshold of one to three times the per capita GDP to determine if an intervention is considered “cost-effective” for that setting. Therefore, we used the 2015 per capita GDP of Vietnam ($2085 USD, worldbank.org) to set a threshold of $2000 and found the brief intervention was a “cost-effective” approach to reducing frequent alcohol use with an incremental cost-effectiveness ratio of $525.

Limited literature exists on cost-effectiveness of behavioral alcohol interventions in resource-constrained settings. Given that this study is one of the first to model and analyze the cost-effectiveness of a behavioral intervention targeting persons who engage in frequent drinking and are living with HIV, we found that implementing a brief intervention in the ART clinic setting is a potentially cost-effective approach. Future studies should assess ways to reduce implementation costs given their impact on cost-effectiveness in our analyses. In particular, the impact of counselor costs on the cost-effectiveness suggest the need for exploring the cost-effectiveness of alternative modalities in this setting, such as task shifting to lay health workers or training HIV care nurses in counseling techniques(44). Furthermore, additional research should explore how adaptive models of HIV service delivery (including fewer tests and in-persons visits for persons virally suppressed) (45) may reduce costs and further emphasize the cost-effectiveness of a brief intervention to reduce alcohol use and improve viral suppression status.

Given the changing funding structure and the loss of external funding such as the United States AIDS-related funding through the President’s Emergency Plan for AIDS Relief (PEPFAR), further consideration must be made about what is considered to be cost-effective in Vietnam. Some have noted that basing thresholds on per capita GDP may be inflated based on previous funding structures and that these metrics are no longer relevant (8). An additional piece to consider moving forward is that the cost shift from donors to government and individuals requires understanding of the individuals’ willingness to pay (46). We have yet to fully understand the impacts of this in Vietnam as the changes in insurance and cost sharing have only recently taken effect.

Screening for alcohol use is difficult given the time providers have to ask patients about their alcohol use, the complexities of an alcohol use disorder (47) and the cultural differences that exist in defining at-risk alcohol use in a setting such as Vietnam (10, 17). We assumed a 1000-person cohort in each intervention arm, with initial states informed by the timeline follow-back (TLFB) data from the primary REDART trial, assessed within one month of enrollment. Trial participants were screened for at-risk drinking using a brief measure (AUDIT-C) and completed the TLFB one month later at enrollment. Although the AUDIT-C was used to screen for enrollment, frequency of drinking (percentage of days abstinent) was the primary trial outcome and is also more readily applied to monthly health states in Markov models (16). As such, many participants were screened as “at-risk” drinkers but were not considered “frequent” drinkers at the start of the intervention. If we had instead modeled all 1000 persons in each arm as engaging in frequent alcohol use at baseline, then both the brief intervention and the combined intervention would have resulted in an even greater number of persons reducing frequent alcohol use compared to the standard of care. The brief intervention is an effective approach to reducing frequent alcohol use in HIV settings in Vietnam (12), but the AUDIT-C screening tool is designed to identify at-risk, rather than frequent, drinkers (48). Further research is needed to find the screening tools that would identify those in Vietnam that would most benefit from evidence-based interventions for frequent drinking.

There are other limitations to this study. First, data from the primary trial were collected for a 12-month period only. Therefore, we assumed that participants’ alcohol use at 12-month follow-up was predictive of their future use. History of frequent alcohol use can result in relapse, suggesting that changes in alcohol use may not be maintained after 12 months post-intervention. Given this, in our alternative scenario analysis of 10-year costs we used the conservative approach of assuming that after 12 months all persons would behave as if they had received only the standard of care.

The generalizability of our findings is limited given that the primary trial was conducted with PWH, mostly men, in northern Vietnam. For example, the cost of ART would not factor in to the cost-effectiveness of an alcohol intervention in the general population who did not require such treatment for HIV. Furthermore, it was not feasible to collect data from families of persons engaging in frequent drinking and other community-related data such as motorbike crashes, violent crime, HIV transmission to partners, and other factors that are oftentimes collected to reflect the impact on communities of frequent drinking among PWH. Such information could provide a more robust societal perspective and would likely make our results even more cost-effective given the gains.

Ultimately this analysis will help inform the health system perspective and policymakers in resource-constrained settings considering substance use treatment programs as an approach for improving the health of persons with HIV in need of alcohol reduction interventions. Resources for HIV prevention and treatment continue to tighten as the realities for competing health priorities shift. Thus, it is important to identify the most cost-effective program that may improve the health of persons with HIV and more broadly, the community.

Conclusion

We determined that a brief intervention to target frequent alcohol use can be cost-effective in the HIV clinic depending on the threshold set by local policymakers. Efforts to address implementation, counselor-related costs, and accurate alcohol screening tools may further improve the cost-effectiveness of this program. Scaling up alcohol use treatment within HIV clinic settings can be a good use of healthcare resources and improve HIV health outcomes in resource-constrained settings such as Vietnam.

Table IV.

Scenario Analysis of Incremental Cost-effectiveness of Behavioral Interventions to Reduce Alcohol Use in Vietnamese PWH over 10 years (Discounted)

	Cost of Intervention	Cost of ART	Total Cost	Incremental Cost	Total Person- Years of Viral Suppression	QALYs gained from Viral Suppression	Total QALYs	Incremental QALYs	ICER
Standard of Care		$2,670,187	$2,670,187		7494	388	8013
Brief Intervention	$33,000	$2,687,295	$2,720,295	$50,108	7638	463	8506	493	$102
Combined Intervention	$79,000	$2,687, 471	$2,766,471	$46,176	7535	424	8251	(256)	DOMINATEDa

^aDominated = more expensive and fewer QALYs gained as compared to the other two trial arms

Abbreviations

ART

Antiretroviral therapy

AUDIT-C

Alcohol Use Disorders Inventory Test

CoI

Combined Intervention

BI

Brief Intervention

CBT

Cognitive Behavioral Therapy

MI

Motivational Interviewing

HIV

Human immunodeficiency virus

PWH

Persons with HIV

RCT

randomized control trial

SOC

Standard of Care

VS

Viral Suppression

Appendix

Appendix I.

Details of REDART Intervention Content

	In-person Sessions	Other Sessions	Strategies	Content
Standard of Care	0 Sessions	None	Referral	No interaction between HIV provider and alcohol treatment provider
Brief Intervention	2 sessions	2 phone booster sessions	Cognitive Behavioral Motivational Enhancement	Didactic training on skills including: Problem solving, coping skills, and drinking refusal
Combined Intervention	6 sessions	3 group sessions	Cognitive Behavioral Motivational Enhancement	Role playing emphasized, participant-led for training on skills including: Problem solving, coping skills, and drinking refusal

Appendix II.

1000 Person Cohort of the Drinking and Viral Suppression Status when Comparing the three trial arms

Drinking Status										HIV Viral Suppression
Time Point*	Frequent			Semi-Frequent			Abstinent			Suppressed			Not Suppressed
	SOC	BI	CoI	SOC	BI	CoI	SOC	BI	CoI	SOC	BI	CoI	SOC	BI	CoI
Baseline	547	547	547	438	438	438	15	15	15	840	840	840	160	160	160
Month 1	547	547	547	438	438	438	15	15	15	840	840	840	160	160	160
Month 2	547	547	547	438	438	438	15	15	15	840	840	840	160	160	160
Month 3	542	276	320	421	575	625	37	149	52	871	866	822	129	134	178
Month 4	534	182	235	430	664	676	36	154	86	881	877	815	119	123	185
Month 5	530	154	201	434	680	689	36	165	106	884	882	812	116	118	188
Month 6	478	225	224	483	655	626	38	120	148	839	872	865	161	128	135
Month 7	473	269	238	488	633	589	38	97	172	842	870	878	158	130	122
Month 8	471	295	246	491	618	568	38	86	186	843	869	880	157	131	120
Month 9	442	311	251	518	609	555	40	80	194	828	869	881	172	131	119
Month 10	439	320	254	521	604	548	40	77	198	829	868	881	171	132	119
Month 11	438	325	255	522	600	544	40	75	201	830	868	881	170	132	119
1 year	456	280	256	491	624	636	53	96	108	763	911	834	237	89	166
Total Person-Months	5897	3737	3574	5676	7138	6945	427	1126	1481	10091	10432	10,229	1909	1568	1771
Total Person-Years	491	311	298	473	595	579	36	94	123	841	869	852	159	131	148

Definitions: Frequent – at last a drink every day in the past 30 days; semi-frequent – fewer than 4 drinks a week in the past 30 days; abstinent – no drinks in the past 30 days; BI – brief intervention; SOC – standard of care; CoI – combined intervention

^*Transitions for Months 7-12 are based on trial data collected at 6 months and 12 months post-intervention