An estimated 6.6 million people in low-income and middle-income countries received antiretroviral treatment (ART) in 2010.1 The increase in ART coverage over time is, in part, related to an increase in the number of ART service sites and the ability of sites to handle more patients, which reflects 2 following related trends: decentralization and task shifting of ART services.1
Within ART clinics, task shifting consists of modifying the scopes of practice for health care workers to allow nonphysician clinicians (NPCs) to deliver ART services.2–4 A recent randomized control trial showed that task shifting could have similar outcomes to physician-led care.5 Limited evidence also suggests that task shifting can save physician time for non-ART activities and may lower costs.3,6 ART tasks that may be shifted include diagnosis of opportunistic infections, initiation of ART, switching antiretroviral (ARV) drug regimens, managing certain clinical conditions and drug side effects, and referring for additional care.7,8
Decentralization of ART from higher to lower level health facilities generally requires task shifting because many lower level facilities lack physicians. Often, hospitals staffed with physicians serve as a referral point for ART patients with complications or adverse reactions. If higher level facilities have already adopted task shifting, then decentralization may not change the cadres of workers delivering ART, but the specific tasks shifted may differ in higher and lower level facilities.
Since the introduction of ART in Ethiopia in 2003, the delivery model has shifted from a hospital-based program run by physicians to a decentralized program including the health center level.9 This decentralizing and task shifting contributed to rapid expansion of ART services from 3 facilities in 2004/2005 to 743 in 2010/2011,10 with the number of ART patients increasing from 8276 to 247,805.10 The decentralized model emphasizes health networks, which comprise linked health facilities at all levels of care serving a catchment area.11 Within networks, health centers refer ART patients with severe complications to the hospital within their network, although a health center may refer patients to a hospital in another network if there are accessibility problems. NPCs, including health officers and clinical nurses, staff most health centers providing ART.12
Ethiopia has also begun to employ task shifting within many of its hospitals. Having NPCs handle most of the routine ART care allows physicians to handle more challenging cases and cases referred from health centers, and spend more time on non-ART services. However, there remain a high percentage of people needing ART not accessing care and treatment.9 As policymakers look to further scale up services, they need a better understanding of the effectiveness and costs of clinician-led care, task-shifted care, and decentralized delivery of ART services.
This understanding is especially important at a time when growth in international and donor resources for HIV/AIDS and ART has slowed.13,14 In light of growing resource constraints, there have been calls to deliver ART more efficiently,13,15 highlighting the importance of evidence on the costs and impacts of different implementation approaches.
Previous assessments of the effectiveness and costs of task shifting have worked primarily with single models of task shifting, and analyses did not focus on what particular tasks are most suitable for task shifting.3 In Ethiopia, multiple implementing partners have overseen task shifting and decentralization, each using a different model of task shifting. Thus, the scale and scope of tasks shifted varies according to site, with NPCs at some sites delivering almost all ART services and at other sites limited to routine monitoring and follow-up.
In this study, we compare the probability that patients will be actively on ART after 2 years and the costs of ART delivery across categories of task shifting. The 2 categories compared are (1) minimal and moderate task shifting and (2) maximal task shifting. Facilities where NPCs monitor (minimal) and initiate ART are classified as having moderate task shifting. Facilities where NPCs initiate and monitor ART, treat severe drug side effects, and switch ARVs due to clinical failure are considered to have maximal task shifting. We use these classifications to determine the incremental cost-effectiveness of increasing the degree of task shifting.
We secondarily examine the costs and effects of ART delivery at hospitals and health centers. Incremental cost-effectiveness analysis is not possible for this comparison because, in Ethiopia, health centers cannot implement ART independently from an associated hospital. Data from before the decentralization were not available to make a comparison between “hospital-only” and “hospital plus health center” care.
This is a retrospective observational study using existing data and ART service delivery patterns. First, we interviewed facility managers about the responsibilities of different staff in the delivery of ART. These data are used to classify facilities by degree of task shifting. Second, we extracted data for a retrospective cohort of ART patients from facility records. Third, we collected costs of resources used for ART.
We used a stratified random sample of health networks across 4 following regions of Ethiopia: Addis Ababa, Amhara, Benishangul Gumuz, and Oromiya. We included a health network if regional staff reported that a physician was involved in ART care at the hospital; this captured all networks in Benishangul Gumuz. We randomly sampled 4 additional health networks in remaining regions. For each health network, we selected the hospital and 2 or 3 health centers, of an average of 5.7 (range: 2–15) health centers per network; for patient outcome and facility manager interviews, although the hospital and 1 health center, selected at random, were selected for costing. Supplemental Digital Content (see Annex 1, http://links.lww.com/QAI/A488) provides more details on the sampling.
Within each facility, we randomly sampled 50 patient records from those in the eligible time frame. To avoid potential bias from a change in ART guidelines in 2007, only patients newly beginning ART after 2008 were eligible for analysis. Patients initiating ART between January 2008 and June 2010 were eligible for selection to ensure a 2-year follow-up was possible. However, if task shifting started at a facility after January 2008, then the eligibility period was from the start of task shifting until June 2010. We excluded patients who started care at another facility but transferred into the facility. Patients transferring out of a facility before completing 2 years of treatment were included in the sample but excluded from the analysis.
All data collectors underwent 3 days of training before the start of data collection and were accompanied by senior supervisors who checked the data collection daily. Data collection tools were piloted at 1 hospital and 1 health center, finalized, and then applied to all health facilities in the analysis.
At the national level, data collection included prices for nationally procured items such as ARVs, drugs for opportunistic infections, laboratories, and supplies. If prices were unavailable, we used prices from WHO's Global price reporting mechanisms,16 from the International Drug Price Indicator Guide,17 and from the published literature.18
Regional level data collection captured information on the costs for activities supporting task shifting including training, supervision, mentoring, and infrastructure upgrades. We interviewed primary stakeholders of the program, including the Federal Ministry of Health, Federal HIV/AIDS Prevention and Control Office, and implementing partners.
At the facility level, we collected data from facility managers on activities related to ART delivery and task shifting, supervision, decentralization, number of ART patients, and ARV drug usage. We interviewed staff to estimate the quantities of medical supplies used in an average ART visit. We also collected data on the number of staff working in the ART clinic, how much time they spend in the ART clinic, how much time they work overall, and staff salaries. A time motion study and task checklist assessed staff roles and time spent in ART care.
We extracted patient-level data on treatment follow-up, mortality, and resource usage from medical records. Exit interviews collected costs related to patients' payment for transportation, other costs for accessing care such as accommodation and food costs, costs incurred at the health facility but not elsewhere captured, and lost wages.
We assigned probability weights to each facility, using the Horvitz–Thompson method,19 and standard errors reflect the sampling design, with health networks as the primary sampling unit and regions as the strata. The primary outcome measure was the probability of a patient being “active” on ART at a health facility 2 years after initiation, with all patients who died, dropped out of treatment, or had not come to the facility in 3 months classified as inactive.
We risk adjusted outcomes with a regression model including CD4 count and WHO clinical stage at initiation as independent variables. We then calculated the mean residual for each facility from this regression. For example, a mean residual of 0.05 for a facility indicates that its patients had a 5% higher than average probability of being actively on ART therapy given the facility patients' CD4 counts and WHO clinical stages at initiation. Other potential predictors of treatment success, such as age, sex, or socioeconomic status, were not routinely available in medical records. We use the risk-adjusted outcome and the unadjusted outcome to assess differences between the comparison groups using t tests with an α = 0.05.
Because this study is not randomized, as an exploratory analysis, we use regression to compare the risk-adjusted outcome to various facility-level characteristics that might influence treatment outcomes; this applies also to the costs.
The cost metrics are the cost per patient per year for the first and second year of ART. The costs of training, mentorship, new equipment, and infrastructure upgrades associated with the start of the facility offering ART care were classified as capital or start-up costs and discounted over their assumed useful life to calculate an annual equivalent cost. We assumed that facility overhead costs, such as utilities and administrative support, would be similar across categories of task shifting and did not collect these costs. We also collected the costs of refresher or yearly training and mentorship.
Costs were further classified as variable or fixed. Variable costs, such as drugs or laboratory tests, were incurred at the visit or patient level. Fixed costs included infrastructure upgrades, training, mentoring, etc. Salary costs are defined as fixed at facilities where there is a stand-alone ART clinic because staff must work in the clinic however many patients attend. Salary costs at integrated clinics are defined as variable.
All costs are presented in 2011 US dollars, using a currency conversion rate of 15.997 Birr per dollar and health sector inflation rates when necessary.20,21 All costs incurred in the second year of patient treatment are discounted using a 3% discounting rate; the same rate used to calculate the annual equivalent cost.22 To determine difference, we use bootstrapped t tests with an α = 0.05 and 100 draws.
Cost-Effectiveness Analysis and Uncertainty Analysis
We use a decision-tree framework to assess the incremental costs and effects of maximal versus minimal or moderate task shifting (see Annex 2, Supplemental Digital Content, http://links.lww.com/QAI/A488). This model includes 3 time points (3 months, 1 year, and 2 years) after the initiation of ART; we calculate the risk-adjusted probability and conditional probability that a patient was active at each time point. For costs, we assume that, on average, the discontinuing patients stop treatment half way through the time period. For uncertainty analysis, we use bootstrapping with 1000 draws for both the conditional probability of being actively on ART after 2 years and the 2-year costs. For sensitivity analysis, we use the results of regression adjusted outcomes and costs and employ bootstrapping of the regression to reflect uncertainty.
This study received ethical review and approval from the Research Ethical Review Office of the Ethiopian Health and Nutrition Research Institute and the Institutional Review Board of Abt Associates. All patients interviewed were given oral informed consent before being interviewed; no data identifying individual patients were collected in the course of this study.
The sample comprised 21 networks, including 21 hospitals, 57 health centers, and 3575 patient records with outcome information. Approximately 30% (1090) of patients sampled were receiving care at hospitals and 70% (2485) at health centers. We conducted 633 patient exit interviews.
We found no facilities where physicians handled all aspects of ART delivery, and only 3 facilities where only physicians were responsible for initiating ART (Table 1). In these 3 facilities, physicians also were responsible for all of the other ART tasks measured except for routine monitoring or adherence counseling. Because there were only 3 sites with minimal task shifting, for purposes of this article, we compare maximal task shifting to the combined group of minimal and moderate task shifting. Forty-seven (60%) of facilities visited were classified as having either minimal or moderate task shifting.
The average CD4 cell count at ART initiation was 180 cells per cubic millimeter (Table 2), with no statistically significant differences across comparison groups. More than 75% of facilities reported receiving financial support from a source other than the Ethiopian government, and just under 50% integrated ART care with non-HIV departments. The average hospital started offering ART in 2006 (Ethiopian fiscal year, 1998) while health centers started, on average, in 2008 (Ethiopian fiscal year, 2000) (P < 0.05). All health centers and more than 80% of hospitals reported that they referred ART patients to another facility; all hospitals and just over half of health centers received referrals. Half of hospitals reported providing mentoring, in collaboration with implementing partners, to other facilities.
Hospitals, on average, had more than 1000 patients actively on ART in the year before the survey took place, whereas health centers had about 250 (P < 0.01) patients. This translates into about 100 patients per full time equivalent staff member, and just more than 900 patients per full time equivalent clinical staff member (physician, nurse, or health officer) at both types of facilities.
Three months after initiation, 93% of patients were still actively on ART, with 94% active at facilities with minimal or moderate task shifting and 90% at facilities with maximal task shifting (P < 0.05) (Table 3). More than 90% of patients were still active 1 year after initiation, and more than 88% of patients were still active 2 years after initiation, with no statistically significant differences between comparison groups. After risk adjustment for CD4 cell count and WHO clinical stage at ART initiation, there were no statistically significant differences between the comparison groups at any time point.
Table 4 shows the cost for ART was about $206 per patient per year, with no statistically significant differences between categories. Costs for ARVs constituted, on average, more than 56% of costs, mentoring and supervision about 13%, salary cost about 8%, and laboratory tests about 7%. Fixed costs constituted about 27% of the total costs of delivering care. The costs for training and meetings and for mentoring and supervision were higher at health centers than at hospitals (P < 0.01). However, these differences are a small proportion of the total costs.
Combining the costs and effect estimates shows that maximal task shifting costs $36 (95% confidence interval: −$40 to $111) more over 2 years per patient than minimal or moderate task shifting, but results in 0.4% (95% confidence interval: −0.9% to 0.2%) fewer patients remaining active at the end of 2 years (see Annex 3, Supplemental Digital Content, http://links.lww.com/QAI/A488). Neither of these differences achieves statistical significance. Figure 1 shows the result of the uncertainty analysis by plotting incremental cost and effectiveness pairs estimated from bootstrapping. In just under 9% of draws, maximal task shifting had better outcomes than minimal or moderate task shifting, and in about 27% of draws maximal task shifting was less costly than minimal or moderate task shifting.
There remains no statistically significant difference between the task shifting categories after using regression to control for facility characteristics for either outcomes or costs (see Annex 4, Supplemental Digital Content, http://links.lww.com/QAI/A488). The regressions show that facilities integrating ART care with other non-HIV services, facilities referring patients out for care, and facilities with physicians on staff are associated with a lower treatment success probability (P < 0.05). Having more patients is associated with better treatment success probabilities at moderate levels, but this effect is attenuated with larger patient loads. Hospitals are associated with higher costs than health centers, showing that costs at hospitals may be up to $90 more for 2 years of treatment (P < 0.05).
The results of regression analysis attenuate the findings from the unadjusted comparison, with about 64% of bootstrapped pairs having a better outcome, and 42% of bootstrapped pairs having lower costs, for maximal task shifting (see Annex 5, Supplemental Digital Content, http://links.lww.com/QAI/A488).
Despite our efforts to seek out facilities least likely to employ task shifting, all surveyed facilities used some form of task shifting, our findings highlight that Ethiopia has already fully adopted task shifting as a strategy to deliver ART services. An important next step in Ethiopia, and likely many other high HIV burden low- and middle-income countries, will be understanding the best approaches to implement task shifting to optimize outcomes within a finite resource envelope.
To the extent that maximal task shifting improves patients' access to ART, the results of this study support further shifting the number of tasks NPCs perform. Although this validates previous research demonstrating task-shifted care has similar outcomes to physician-led care, our analysis looked at task shifting beyond NCP ART initiation and routine monitoring, which is relatively unreported in the literature.
Although previous studies suggest a cost difference between not task shifting and task shifting models of delivery,3 we did not find differences in costs between degrees of task shifting. Although greater task shifting conserve resources, because lower paid NPCs assume the duties of higher paid physicians, the lack of difference in costs in this study may indicate that incremental changes in the degree of task shifting do not greatly affect physician time. Further, labor costs are less than 8% of estimated total costs, so the potential cost savings may not be substantive enough to impact total costs. In areas where labor costs are higher relative to the costs of drugs, etc., this may be more important. Costs of supporting NPCs providing ART, like training and mentoring, could also be substantive enough to limit potential cost savings.
Although the comparison groups did not have statistically significant differences in important outcomes, regression analysis suggested some interesting trends. For example, only 2% of facilities with maximal task shifting had a physician on staff, compared with 16% of facilities minimal or moderate task shifting (P < 0.10), but facilities with physicians had a lower probability of 2-year treatment success. This may reflect a potential bias not entirely captured by risk adjustment, because facilities with physicians may attract more complicated patients. Furthermore, this study was limited to the first 2 years of ART, which may not be sufficient to capture differences between NPC and physician-managed ART treatment failure. Regardless of differences between facilities, overall 2 year retention rates (88%) in our sample reflect positively on the implementation approaches taken to improve access and quality of ART services in Ethiopia; a 24-month retrospective cohort study of patients starting ART from 2003 to 2006 in Ethiopia reported retention of 76% at health centers and 67% at hospitals.12
The results also suggest that health centers have similar treatment success probabilities to hospitals. Although exploratory in nature, the regression analysis shows an association between hospitals and higher costs than health centers. At least 2 interpretations of this finding are possible as follows: decentralization is an efficient means of increasing ART enrollment, or hospitals treat patients with more expensive needs.
Similar to previous studies,23 we find that facilities with fewer patients had lower probabilities of treatment success, up to a point, than facilities with more patients, and facilities with fewer patients tend to have a higher cost per patient, again up to a point. This suggests that if decentralization is expanded, there may be some decrease in cost-effectiveness if new facilities only enroll small numbers of patients. It may also suggest that facilities with lower patient loads may need more mentoring or supervision to ensure treatment success, again affecting costs.
This analysis has a variety of strengths and limitations. Data reported here include a large number of facilities and reflect implementation of ART programs under routine, rather than research, circumstances. Further, this is, to our knowledge, the first study to examine the degrees of task shifting, especially related to tasks beyond NPC ART initiation and routine monitoring.5,24
However, given that the study used observational data, the results may reflect some selection bias in where patients sought services, how frequently they utilized care, and why facilities developed patterns of task shifting (eg, location, existing staff skill sets). For example, if facilities with doctors attract sicker patients and our risk adjustment did not fully capture this selection, we may have over stated the effectiveness of maximal task shifting in comparison to minimal or moderate task shifting. This is possibly reinforced if staff at facilities with maximal task shifting provide better care in general; results may reflect staff ability rather than the task shifting model employed. Additionally, some data, such as patients' age, weight, height, and sex, were not routinely available in clinical records, limiting the depth of the analysis. Despite some of the limitations inherent with operational research studies like this one, these findings help build an evidence base for task shifting, help explore some of the nuances with operationalizing task shifting programs, and can provide program planners with useful information to optimize service delivery.
Girma Debela Geresu and Abiy Tsegaye Gebrekidan led data collection teams and contributed to the design of the data collection tools. Kimberly Kennedy (Abt Associates) helped train and oversee data collectors. Sarah Dominis (Abt Associates) contributed to the design of the data collection tools. The Federal Ministry of Health (FMOH) Human Resource Development and Administrative Directorate, Ethiopia, Federal HIV/AIDS Prevention and Control Office (FHAPCO), Addis Ababa City Administration Health Bureau, Amhara Regional Health Bureau, Oromiya Regional Health Bureau, and the Benishangul Gumuz Regional Health Bureau all facilitated data collection and contributed to the concept of the study. Management Sciences for Health, International Training and Education Center for Health, ICAP, and Johns Hopkins University generously provided us with their time, thoughts, and cost data. Nejmudin Bilal, Itamar Katz, and Hailu Zelelew contributed to the conceptual design of the study. USAID/Ethiopia also contributed their thoughts and time to make this study possible. The authors are grateful for their help and support. The views expressed in this article are those of the authors and do not necessarily reflect the official policy or position of the US Government.
1. UNAIDS. UNAIDS World AIDS Day Report 2011. Geneva, Switzerland: UNAIDS; 2011.
2. Callaghan M, Ford N, Schneider H. A systematic review of task-shifting for HIV treatment and care in Africa. Hum Resour Health. 2010;8:8.
3. Mdege ND, Chindove S, Ali S. The effectiveness and cost implications of task-shifting in the delivery of antiretroviral therapy to HIV-infected patients: a systematic review. Health Policy Plan. 2012;28:223–236.
4. World Health Organization. Task Shifting: Global Recommendations and Guidelines. Geneva, Switzerland: World Health Organization; 2008.
5. Fairall L, Bachmann MO, Lombard C, et al.. Task shifting of antiretroviral treatment from doctors to primary-care nurses in South Africa (STRETCH): a pragmatic, parallel, cluster-randomised trial. Lancet. 2012;380:889–898.
6. Babigumira JB, Castelnuovo B, Lamorde M, et al.. Potential impact of task-shifting on costs of antiretroviral therapy and physician supply in Uganda. BMC Health Serv Res. 2009;9:192.
7. Miles K, Clutterbuck D, Seitio O, et al.. Antiretroviral treatment roll-out in a resource-constrained setting: capitalizing on nursing resources in Botswana. Bull World Health Organ. 2007;85:555–560.
8. Harries A, Makombe S, Schouten E, et al.. Different delivery models for antiretroviral therapy in sub-Saharan Africa in the context of “universal access”. Trans R Soc Trop Med Hyg. 2008;102:310–311.
9. Koethe JR, Lukusa A, Giganti MJ, et al.. Association between weight gain and clinical outcomes among malnourished adults initiating antiretroviral therapy in Lusaka, Zambia. J Acquir Immune Defic Syndr. 2010;53:507–513. doi: 510.1097/QAI.1090b1013e3181b1032baf.
11. Ministry of Health of the Government of Ethiopia. HIV/AIDS Prevention and Control Office. Accelerating Access to HIV/AIDS Treatment in Ethiopia: Road Map 2007–2008/10. Addis Ababa, Ethiopia: Ministry of Health; 2007.
12. Assefa Y, Jerene D, Lulseged S, et al.. Rapid scale-up of antiretroviral treatment in Ethiopia: successes and system-wide effects. PLoS Med. 2009;6:e1000056.
13. Boyd MA, Cooper DA. Optimisation of HIV care and service delivery: doing more with less. Lancet. 2012;380:1860–1866.
14. Leach-Kernon K, Chou DP, Schneider MT, et al.. The global financial crisis has led to a slowdown in growth of funding to improve health in many developing countries. Health Aff (Millwood). 2012;31:1–8.
18. Bikilla AD, Jerene D, Robberstad B, et al.. Cost estimates of HIV care and treatment with and without anti-retroviral therapy at Arba Minch Hospital in southern Ethiopia. Cost Eff Resour Alloc. 2009;7:6.
19. Horvitz D, Thompson D. A generalization of sampling without replacement from a finite universe. J Am Stat Assoc. 1952;47:663–685.
21. The Federal Democratic Republic of Ethiopia Central Statistical Agency. Country and Regional Level Consumer Price Indices. Addis Ababa, Ethiopia: The Federal Democratic Republic of Ethiopia Central Statistical Agency; 2012.
22. Murray CJ, Evans DB, Acharya A, et al.. Development of WHO guidelines on generalized cost-effectiveness analysis. Health Econ. 2000;9:235–251.
23. Handford CD, Tynan AM, Rackal JM, et al.. Setting and organization of care for persons living with HIV/AIDS. Cochrane Database Syst Rev. 2006:CD004348.
24. Bemelmans M, van den Akker T, Ford N, et al.. Providing universal access to antiretroviral therapy in Thyolo, Malawi through task shifting and decentralization of HIV/AIDS care. Trop Med Int Health. 2010;15:1413–1420.