Task-shifting alcohol interventions for HIV+ persons in Kenya: a cost-benefit analysis

Alcohol use is responsible for 13.5% of global deaths due to infectious diseases including HIV, and 5.1% of disability-adjusted life years (DALYs) [1, 2]. The high worldwide alcohol use related morbidity and mortality suggests an urgent need for a global focus on treatment of alcohol use disorders [3‐5]. Alcohol consumption negatively impacts the effectiveness of prevention efforts by facilitating HIV transmission through increased sexual risk behavior as well as HIV treatment non-adherence [6‐11]. Although dramatic progress against HIV/AIDS has been made in sub-Saharan Africa [12] –with disease incidence decreasing by 25% in 22 countries from 1990 to 2009 and treatment access expanding from 50,000 to over 5 million persons from 2002 to 2012 [13] – the progress towards “zero new infections and zero AIDS-related deaths” promoted by UNAIDS [14] is hampered by several challenges. These include the growing HIV prevalence due to expanded lifespans, limited accessibility to antiretrovirals (ARVs), and continued suboptimal adherence to treatment [13, 14]. While increased access to ARVs improves overall health by lowering viral load and acting as a protective factor [15, 16], the predictive ability of adherence [17, 18] is strongly mediated by alcohol use [8, 19‐23]. This suggests that alcohol use can present a significant barrier to the Zero Campaign by increasing HIV infectivity. The evidence suggests a linear dose-response relationship between alcohol use and risk for comorbid and AIDS-defining illnesses [24]. That is, although high levels of alcohol use are associated with higher rates of AIDS-related complications and higher health care costs, any level of alcohol use places a patient at higher risk for such complications and higher costs—thus suggesting that there is no “safe” level of alcohol consumption for HIV-infected patients. For this reason, we consider alcohol use of all levels as harmful for HIV+ persons throughout this paper.

In sub-Saharan Africa, Kenya is among countries with high and hazardous alcohol consumption, 22.7% of adults (age 15+) using alcohol within the past 12 months, and an estimated 3.2% of adults (age 15+) exhibiting alcohol use disorders [2]. Furthermore in a survey among school-going Kenyan youth, 48.9% have drunk alcohol [25]. Because harmful alcohol use behavior is classified as a mental health disorder, known as alcohol use disorder (or AUD), and because such behavior can be associated with spread of HIV [9, 26‐28], it is important that those with AUD receive appropriate mental health treatment. Unfortunately, throughout much of the world, there exists a large treatment gap between psychiatric service availability and the high burden of mental disorders including substance abuse [29]. In Kenya in particular, the Ministry of Health reports that 20–30% outpatient visits are for mental disorders [30], and only 15% of mental health patients receive treatment [31]. Furthermore, despite an estimated 25% of patients in general health clinics suffering from alcohol and substance abuse disorders, only 0.1% of patients in general health clinics had their alcohol abuse problems picked up by clinicians [31]. This is likely because of the dearth of mental health care workers. In 2010, Kenya employed approximately 75 psychiatrists, only 12 of whom work across the eight provincial and 250 district hospital system – one psychiatrist for every province of 3–5 million people [32]. There were 250 trained psychiatric nurses deployed in psychiatry in Kenya, so approximately one psychiatric nurse per 160,000 Kenyans [32]. In the current context of acute shortages of specialist health workers, high alcohol use, and high HIV prevalence [33‐35], there are calls for innovative cost-effective strategies to reduce alcohol use in sub-Saharan Africa, and in Kenya in particular [36, 37].

One promising model to accelerate delivery is task-shifting of services in which tasks performed by professionals are delegated to those with less formal education or training called paraprofessionals [38‐41]. Shifting of mental health services through training additional personnel could be critical to meet demand [32, 42]. Task-shifting to scale-up other HIV prevention interventions in Kenya offers a practical model for healthcare delivery [43]. Task-shifting HIV and alcohol interventions may promote behavior change among larger numbers of people living with HIV, including men who generally utilize less HIV testing, less therapy and display lower adherence to care [44‐46].

Given the limited number of mental health professionals in Kenya, we developed a cognitive behavioral therapy (CBT) to reduce alcohol consumption, and a framework of paraprofessional training and supervision to task-shift this CBT. CBT has been shown to positively influence health behavior by engaging participants in a deliberate exploration of thoughts, actions and feelings by which participants learn coping skills to handle high-risk substance use situations [47]. CBT was selected for the Kenyan adaptation because of its strong empirical support in both individual and group formats to reduce substance abuse [48, 49], durability of treatment effects, and prior successful applications in sub-Saharan Africa to reduce risky sexual behaviors among HIV-infected Zambian couples [50] and to improve mood among Nigerian surgical patients [51]. Furthermore, because of its highly-structured format, CBT was feasible for training paraprofessionals and for delivery to those with limited formal education. On the contrary, using Medication Assisted Therapy (MAT) was not feasible in this resource-limited setting due to extremely limited health care professionals to deliver or monitor the medication. Additionally, MAT typically requires longer delivery to curb alcohol use, recommended from 4 months [52] up to 12 months or longer [53].

In a previous pilot study, we found task-shifted CBT to be effective at reducing alcohol use, with a reported alcohol abstinence rate of 69% at a 90 day follow-up (vs. 38% usual care) [47, 54]. As mentioned earlier, all levels of alcohol use—even low levels—have been shown to have a negative impact on HIV+ persons [24]. Therefore, our pilot study included any HIV+ patients who had any amount of alcohol in the previous month, as well as a score of 3 on AUDIT-C or endorsement of binge drinking on a monthly basis. Furthermore, the level of competency of this task-shifted CBT was independently rated to be equivalent to therapy delivered by college-educated therapists in the U.S. [47, 54].

The present economic evaluation study has three aims. First, we describe field-based costs of a feasible rollout of CBT to reduce alcohol use among 13,440 HIV-infected outpatients at 12 sites in Kenya. Our assumptions are based on two CBT trials in western Kenya [47, 54]. Second, we estimate the potential economic benefits of the rollout in medium- and long-term based on simulation methods. Third, we calculate the potential medium to long-term benefit-to-cost ratio (BCR).

We chose to conduct a cost-benefit analysis for two reasons. First, previous research suggests that CBT may be cost-effective in the Kenyan context, but cost-saving only under very strict assumptions [37]. We hypothesized that including the additional potential economic impact of the CBT intervention in the Kenyan setting would provide a more comprehensive assessment. Thus, cost-benefit analysis (comparing both costs and all benefits in monetary terms) would be better suited for illustrating this point than cost-effectiveness analysis (which compares costs in monetary terms, and effects given only as health outcomes) or cost-utility analysis (which uses cost per utility effect such as dollar per quality-adjusted life year) [55, 56]. To show that CBT is effective and cost-saving is not a trivial task because only a few interventions are highly-effective and also save costs [57]. Second, modelling based on effectiveness outcomes alone has already been completed using our CBT estimates [36], making cost-benefit analysis (that includes monetary evaluation of health and economic benefits) the next logical step in this line of research. Our cost-benefit analysis offers useful novel information regarding the health and economic benefits of CBT.

A cost-benefit analysis is an economic evaluation technique which places a monetary valuation on a health program’s costs and benefits, allowing for the comparison of a health program’s incremental cost to its incremental benefits in corresponding monetary units [58]. The model presented in this paper estimates the costs and benefits of rolling out a task-shifted CBT intervention to reduce alcohol use among persons living with HIV in Kenya from a societal perspective, including beneficial economic impact over and above reductions in mortality and morbidity. Benefits included those associated with lowered HIV incidence and improvements in household as well as labor force productivity (Table 1). Costs included all costs associated with training and administering task-shifted CBT (Tables 2 and 3). There is inherent uncertainty when placing a monetary valuation on a program’s potential future costs and benefits. To help account for this uncertainty we created sensitivity analyses which vary our model’s key assumptions across different ranges as described in the Sensitivity analyses section of this paper.

Our program implementation model, based on field operations and expert consultation, contemplates a program that services 13,440 participants in 12 sites in Kenya across five years. In our base case, we assume the CBT treatment effect is maintained for two years, meaning that for every year which the intervention is rolled out, the participants of that year gain benefits associated with CBT for two years. Thus in our base case, we assume a 6-year time horizon. The program’s rollout was constructed from parameters based on our pilot experience in Eldoret, Kenya among 75 HIV-infected outpatients who reported hazardous or binge drinking, and our on-going randomized controlled efficacy trial with 614 randomized participants. The results, a full description of the pilot study, and an on-going trial are described elsewhere [47, 54, 59].

Methods for rollout were based on our previous work, with some modifications to enhance future sustainability [47, 54]. For example, while the pilot study was conducted in a large town (Eldoret) with a tertiary care medical center, our costs and methods for the rollout are based on delivering the intervention at the level of district or local hospital administered by the Ministry of Health. While paraprofessionals in our previous work possessed varying degrees of formal training from high school diploma to a bachelor’s degree, salaries and training in the rollout are based on high school diploma only, to make the intervention more realistic. Though we condensed the training period from 4 to 2 weeks in this exercise, we also extended the paraprofessional counselor consultation up to one year. Whereas a psychiatrist managed safety concerns (i.e., psychiatric risk and alcohol withdrawal symptoms) in our previous studies, our rollout methods are based on training diploma nurses to manage safety issues. Finally, while our past groups were delivered by gender-matched paraprofessionals, gender of the counselor in same-gender rollout group may not be always matched. Our rollout estimates are conservatively based on two counselors at each of 12 sites delivering two groups per week with eight same-gender participants per group, amounting to an annual workload of two counselors at each site would be 160 participants per site in year one. In years 2–5, the counselors are projected to have gained the training and experience to handle three groups per week, amounting to 320 participants per site. We have added monthly “maintenance” groups of post-CBT patients who would like to maintain or sharpen behavioral skills. We have reduced the rate of transport reimbursement per participant because of the rollout to more proximal location to participants, and we examine a wide range of transport reimbursement rates in our sensitivity analyses. Program evaluation methods would be left to each site and could be accomplished by providing a brief measure of alcohol use (e.g., AUDIT) before and after intervention participation.

The overall benefit-cost ratio was 1.13. Figure 1 summarizes the costs and benefits for the base case CBT implementation (costs are in the first bar labeled in gray, benefits are shown in the second bar labeled in blue). Training costs in the first year were $158,000, which were higher than subsequent years, primarily due to personnel costs. However, costs declined across the final four years to about $94,000 in year 5. The total discounted cost over the five-year rollout was approximately $554,000. Over the five year period, the average cost-per-participant was $44. (Cost breakdowns are presented in Tables 2 and 3). Benefits totaled $49,000; $118,000; $137,000; $133,000; $129,000; and $62,000 in years 1, 2, 3, 4, 5, and 6 respectively. Note that benefits continued in year 6, despite the costs of the program ending in year 5. The average annual savings from decreased HIV incidence (benefit 1) was $11 per patient; and the annual value for increased productivity (benefit 2) was $14.50 per patient. The total value of benefit 1 and benefit 2 was estimated to be approximately $628,000.

Figure 2 shows the input variables that most affected the base case scenario in one-way sensitivity analyses. For nearly all inputs shown in Fig. 2, CBT remained cost-savings with a benefit-cost ratio greater than 1. Not pictured in Fig. 2 are values associated with non-drug related inputs associated with parameter 1.3. These four inputs had little effect on the benefit-cost ratio, and when varied across their ranges, yielded benefit-cost ratio ranges of: 1.138 to 1.140 (support services); 1.13 to 1.15 (lab tests); 1.13 to 1.15 (visits); and 1.11 to 1.18 (fixed costs). Also not shown in Fig. 2 is the uncertainty associated with the assumed treatment effect length. When we adjusted our assumption to the treatment effect lasting 1, 5 and 10 years, the benefit-cost ratio changed to 0.58; 2.72; and 5.05 respectively.

In Fig. 3, we further characterize the inherent uncertainty in the parameters by summarizing the probabilistic sensitivity analyses. The results were split into groups based on treatment effect duration; for each group we present separate Monte Carlo simulations with 10,000 repetitions, each with an independent sample of 13,440 participants. The mean benefit-cost ratio under assumptions that the CBT treatment effect lasted for 1, 2, 3, 4, 5 or 10 years was 0.65 (95% CI 0.64–0.66); 1.28 (95% CI: 1.26–1.31); 1.90 (95% CI 1.86–1.94); 2.49 (95% CI 2.44–2.55); 3.07 (95% CI 3.01–3.14); and 5.72 (95% CI: 5.61–5.84) respectively.

Previous research—which used only health outcomes— showed that CBT was generally cost-effective, but only cost-saving under strict assumptions (i.e., program costs of less than $1 per individual) [36, 37]. In this study, across a broad range of assumptions, task-shifted CBT was not only favorable, but even cost-saving from a societal perspective, when we included health and economic productivity benefits. Our base-case scenario showed a benefit-cost ratio of 1.13 when the treatment effects were assumed to last two years; and the benefit-cost ratio increased to 1.90 with a 3-year treatment maintenance scenario.

Our results are consistent with a related exercise which characterized task-shifted CBT for alcohol use reduction in Kenya as cost-effective [37]. However, the extant results go further, showing not only reductions in morbidity and mortality associated with the scale-up, but also monetary cost-savings for the system as a whole in Kenya. Some costs included in our analyses such as specific per-diem rates and housing allowances, are set by the Kenyan government. These are likely not required in other sub-Saharan African countries, so the benefit-cost ratio would likely be higher in those settings. The provision of transport reimbursement increases the cost of the intervention, but likely enhances the effectiveness of the intervention by facilitating better attendance.

A strength of this study is that the scale-up was constructed using in-depth knowledge and data of the training and implementation of CBT procedures from two clinical trials. The costing exercise presents a novel and realistic way of shifting the CBT tasks to paraprofessionals in an organized, feasible and scalable manner. The training, enrollment and compensation procedures have been implemented in our past and current trials. Furthermore, the potential economic benefits are based on rigorous analyses based in Kenya with the same or similar populations [74, 78], and have been recognized to be more widely applicable [80]. The effect of this approach could support the country’s 2010 Alcoholic Drinks Control Act which seeks to tighten legislation against alcohol abuse while also, for the first time, ensuring the right to access to treatment programs [81].

The input parameter most affecting the results was the duration of CBT-alcohol reduction treatment effects. It was difficult to choose a base case scenario because evidence regarding the long-term effect length of CBT for alcohol-use reduction is limited. One reason may be because the length of follow-up of CBT studies is often constrained by the time limits of grant cycles. Despite these limitations, there is reason to believe that CBT treatment effects could be maintained over a long period without major reductions in effectiveness. In a 128-person treatment sample, Kadden and colleagues observed only marginal reductions in treatment effects in alcohol use at 18-month post-treatment follow up; including a group CBT condition [82, 83]. Although we felt a base case of a two-year effect was reasonable, given the uncertainty surrounding the treatment effects, we also conducted sensitivity analyses where the base case treatment effect was adjusted to 1, 5, and 10 years. Our sensitivity analysis showed that our base case results held for treatment effect lengths greater than two years, but at a one-year duration assumption, CBT did not achieve cost-neutrality. This exercise in modeling reveals that while a task-shifted CBT has excellent potential for considerable cost-savings, if treatment effects are particularly short, it is unlikely that CBT will be cost-savings. It may still be highly cost-effective (i.e., averting a DALY for less than the per capita GDP), but it may not save money to the government in the long run. Again, very few interventions achieve the status of cost saving [84, 85].

This study has limitations. First, the CBT effects are based on a small pilot study. Thus, we are currently conducting a larger trial, and expect to update the results presented here in future research. Second, due to the limited information regarding the effects of various levels of adherence, we have dichotomized the treatment effect of our modeled CBT as “abstinent” vs. “non-abstinent. It is possible that if continuous data were available regarding the effect of reducing drinking by one drink, then our modeled CBT would be shown to have even greater levels of benefits in terms of ARV adherence and reduced HIV infectivity given the positively correlation between alcohol use and these modeled benefits [7, 31, 36, 37, 64‐67, 71‐73]. Third, cost-benefit models regarding future cost-savings carry inherent uncertainty as assumptions must be made regarding the consequences of an intervention. As discussed above, the input parameter most affecting CBT treatment cost-neutrality was the CBT treatment effect maintenance. However, to deal with the uncertainty of this and all other variables, we conducted various sensitivity analyses: assumptions regarding the parameters in this model were varied one at a time, as well as simultaneously across ranges presented in the literature.