In nations with generalized HIV epidemics, the burden of HIV-positive individuals on the public health care system is considerable, often accounting for more than half of all inpatient care provided.1,2 Provision of combination antiretroviral therapy (ART) in these settings is likely to have a significant impact on the annual, and particularly lifetime, cost of care, as it did when introduced to high-income countries.3-5 Most comprehensive studies of health care costs to date have been conducted in North America and Europe.6 Evidence from Africa is important, because the cost of providing ART to all who need it is likely to dwarf existing total health care expenditures on the continent; the cost of antiretroviral medicines alone is several times most African nations' current annual health care expenditures per capita.5
Given the variation in care delivery methods observed worldwide, it is not clear whether evidence from studies in non-African settings are comparable to those conducted on the continent. For example, antiretroviral medicines have typically comprised most of all treatment costs in high- and middle-income countries,3,7-10 but existing South African studies have found them to contribute only a third of all costs.11
Evidence on the total cost of public health care for individuals on ART in Africa is limited.11 In South Africa, one study calculated the program cost of running a clinic to provide ART to health care workers in KwaZulu-Natal to be between $1054 and $1382 per patient per year.12 Three cost-effectiveness studies, 2 of hospital-based ART programs and 1 of a primary care ART clinic, also produced estimates of cost per patient-year.13-15 All 3 showed that annual costs for those on ART were lower than for those not on ART, although lifetime costs were higher. In stratified analyses, those with low CD4+ T-cell count or a history of AIDS had higher health care costs. One of these studies considered factors associated with hospitalization; however, in multivariate analysis, this study found neither CD4 cell count nor HIV RNA viral load level to predict use of services.13 A costing study of the Western Cape provided a total cost per patient in a public sector clinic and total provincial expenditure requirements but not annual health care costs per patient.16
The only other study the authors are aware of that considered predictors of health care costs on ART outside a high-income setting was conducted in Mexico and found that patient health status measured by Centers for Disease Control and Prevention (CDC) stage, patient education level, and the specialty of the attending physician was associated with total cost.10
The first objective of this study was to determine the total annual cost to the health care system of a patient accessing a public sector ART clinic in South Africa according to time on ART, World Health Organization (WHO) stage, and CD4 cell count. The second objective was to analyze which of these factors were associated with significantly higher health care expenditures before treatment and in the first and second years on ART. It is hoped that this provides data for cost-effectiveness analyses and aids administrators in planning the provision of ART services.
This study followed all eligible patients enrolled at the Hannan Crusaid Treatment Center (HCTC) between September 1, 2002 and December 31, 2003. The HCTC is a dedicated ART clinic based at the Gugulethu Day Hospital in Nyanga district, a periurban settlement close to Cape Town. During the period of study, it was jointly run by the Desmond Tutu HIV Center (a not-for-profit HIV research organization), the UK-based charity Crusaid, and the Provincial Government of the Western Cape (PGWC). Eligibility criteria for the clinic at the time of the study included a CD4 cell count of <200 cells/μL or a history of an AIDS-defining illness. The HCTC acts as the primary care provider to all individuals enrolled at the center, providing ART and non-ART-related care. Its work has previously been described.17
Patients are referred by local primary care providers. They are screened at the HCTC and seen a second time 14 days later, before commencing ART. Patients attend scheduled visits at weeks 0, 4, 8, and 16 after commencement and every 16 weeks thereafter. Patients may make unlimited unscheduled visits to the HCTC as required. First-line ART comprises stavudine, lamivudine, and efavirenz unless the patient is pregnant or a woman of child-bearing age not taking birth control medication; in such cases, zidovudine and nevirapine replace stavudine and efavirenz. The second-line regimen comprises lopinavir/ritonavir (Kaletra®), zidovudine, and didanosine. HIV RNA load, CD4 cell counts, and alanine transaminase tests are conducted before treatment and every 16 weeks during ART. Other safety tests are performed according to the provincial ART protocol and vary according to the treatment regimen.18
Subjects were followed from enrollment at the HCTC until their final scheduled 4-month visit in 2004, a maximum of 112 weeks after ART commencement. Patients were right-censored for death, loss to follow-up (LTFU), or transfer to another ART site. The baseline characteristics of this study population have previously been described.19 Ethical approval for this study was gained from the University of Cape Town Research Ethics Committee, and all patients gave informed consent.
Costs were calculated from a public health care provider perspective. Data on patient-specific service utilization were collected prospectively from files at the HCTC and retrospectively from computer- and paper-based medical records at hospitals to which subjects were referred. The costs of antiretroviral medicines were taken from the 2004 national public sector tender, and those of other medicines were taken from provincial hospital tender prices. Medical test prices were taken from public sector tariffs charged by the National Health Laboratory Service and medical procedure costs from cost recovery charges made to private patients attending public hospitals.20
Non-patient-specific costs at the HCTC for the financial year 2005 to 2006 were calculated per patient-month in the program. Methods and results for this have been detailed previously.21 Non-patient-specific costs for hospital visits were calculated from financial year 2004 to 2005 expenditure data, excluding all patient-specific line items,22 on a per patient-day-equivalent basis using step-down accounting methods. Each inpatient day was weighted as 3.77 outpatient visits.15 The cost of providing tuberculosis directly observed therapy (DOTS) was based on data collected in the same health district in 2000, calculated on a per-day-on-DOTS basis.23 Unit costs for key inputs are shown in Table 1.
All costs were standardized to average 2004 prices using South Africa's consumer price index.24 They were then converted into US dollars using the average exchange rate for 2004 of US $1 to R 6.4347.25 Observation time was separated into 3 periods: between enrollment and ART commencement (Pre-ART), the first 48 weeks on ART (Year One), and any time after 48 weeks (Year Two). Costs were calculated per eligible patient enrolled in the program and per patient-year of observation (PYO; 365.26 days) in each period.
A multivariate regression model was used to determine which clinical factors were associated with demand for health care in each period. Utilization, and therefore cost, data are frequently noted to be right-skewed in distribution, with a large number of individuals consuming few resources and a few subjects using many. In this situation, ordinary least squares regression is not appropriate, because the data are not normally distributed and do not have constant variance.26 To take these features into account, a γ-distribution with a log link was used.27 The ability of this model to account for the nonnormal distribution was evaluated using deviance residuals.28
Explanatory variables used included age, gender, WHO stage at enrollment, CD4 cell count at the beginning of each period, change in CD4 cell count in the first 4 months on treatment, and whether the subject was censored for death or LTFU in the period. In all models, the dependent variable was the natural logarithm of costs. The primary outcome measure used was total cost per PYO. This is preferable to total cost per patient observed because it takes into account the intensity with which resources were consumed. The per-patient measure was used as a secondary outcome. Statistical analyses were performed using STATA 9.2 (Stata Corporation, College Station, TX). Differences between regression coefficients were tested using a joint Wald χ2 test. All statistical tests were 2-sided at α = 0.05.
Demographic and clinical characteristics of the sample are shown in Table 2. Just less than three quarters (72%) of the patients were female, and most (79%) were younger than 40 years of age. Most enrollees had late-stage HIV at program entry, with almost half (44%) having a previous AIDS-defining illness and more than half (58%) having a CD4 count of less than 100 cells/μL. After 16 weeks on the program, surviving patients had a median rise in CD4 count of 97 cells/μL, and after 48 weeks on the program, two thirds (66%) of patients had CD4 cell counts greater than 200 cells/μL. Most (78%) censoring events were attributable to death. For analytic purposes, all censoring events were assumed to be homogeneous and presented as a single group. Total length of follow-up was 25.2 patient-years in the Pre-ART period, 164.5 patient-years in Year One, and 88.6 patient-years in Year Two. During Year One, 5 patients moved from first-line to second-line therapy; during Year Two, a further 4 did so.
The mean number of hospital outpatient visits (excluding the HCTC) per patient in the cohort was 0.26 Pre-ART, 0.77 in Year One, and 1.05 in Year Two; the mean number of inpatient days was 3.04 Pre-ART, 3.73 in Year One, and 4.04 in Year Two. The mean number of hospital outpatient visits per PYO fell from 2.15 Pre-ART to 0.98 in Year One and 0.87 in Year Two; the mean number of inpatient days fell more sharply from 25.52 Pre-ART to 4.75 in Year One and 3.35 in Year Two.
The cost of providing treatment and care, to all patients and to those who were not censored, in each period is shown in Table 3. Total costs were $404 per patient between enrollment and ART commencement; the cost per patient for those who died in this period ($976) was more than twice this figure. Cost per PYO fell from $2502 in Year One to $1372 in Year Two. Censored patients in the Year One period cost an average of $16,645 per PYO, almost 5 times the average for that period. Per patient, this figure was $21,810. Censored patients in Year Two cost 4% less per PYO than the average for the period.
Figure 1 illustrates the division of costs in each period. In the Pre-ART period, one quarter (25%) of all health care costs were attributable to primary care visits to the HCTC, and almost all other costs (69%) were attributable to hospital-based care. In the Year One period, the cost of primary care and hospital care fell as a proportion of total health care costs, to 16% and 49%, respectively. The cost of providing ART amounted to 31% of total costs, with medicines accounting for 59% of these costs and monitoring tests for the remaining 41%. In the Year Two period, the absolute cost of ART fell slightly, with higher medicine costs being offset by lower monitoring test costs, but rose as a proportion of total health care costs to 55%. The proportion of costs attributable to primary care visits changed little but fell by 37% in absolute terms. The cost of hospital-based care fell rapidly, by 73% in absolute terms and by 50% in proportional terms. The cost of censored patients in the Pre-ART and Year One periods (not shown) comprised predominantly hospital care costs (88% and 85%, respectively), which were far higher in absolute and percentage terms than for other patients.
Results of the multivariate regression analysis are presented in Table 4. In the Pre-ART period, patients in WHO stage 1 or 2 cost one fifth as much per PYO as those with an AIDS diagnosis, whereas those in WHO stage 3 cost 61% as much per PYO as those in stage 4 (see Table 4, upper panel). These differences were jointly significant. In contrast, a patient's baseline CD4 cell count, age, and gender were not significantly associated with health care costs. Costs per PYO for patients who were censored for death in the Pre-ART period were 7 times higher than for other patients. The odds ratio for censored patients dropped to 2.02 when the dependent variable was changed to cost per patient (see Table 4, lower panel), but all other coefficients remained roughly similar to those for cost per PYO.
In the first year on ART, significant predictors of health care costs continued to be WHO stage at program entry and censoring, although the odds ratios were closer to the null than in the Pre-ART period in both cases. The effect of a patient's baseline CD4 cell count continued to be nonsignificant. The change in CD4 cell count between baseline and week 16 on ART, however, was a significant predictor of costs, with each 50-cell/μL increase being associated with a 12% decrease in cost per PYO. Furthermore, men had a 25% lower cost per PYO after adjusting for all other variables, although this difference was not significant. Once again, the cost per patient analysis found similar results for all explanatory variables with the exception of censored patients, who nevertheless had significantly higher total costs than other members of the sample. The results for Year One shown in Table 4 do not include 19 individuals who did not have a week 16 CD4 cell count because of having been censored (15 deaths, 3 LTFU, and 1 transfer out). A regression containing all 200 observations and excluding week 16 CD4 cell count change as an explanatory variable (not shown) found similar results.
In the Year Two period, predictors of health care changed. Patients with a baseline WHO stage of 1, 2, or 3 still had lower costs per PYO than those with an AIDS diagnosis but not significantly so. Furthermore, censored patients no longer had significantly higher costs per PYO. Instead, an individual's CD4 cell count at week 48 was a significant predictor, with each 50-cell/μL increment being associated with a 5% lower cost per PYO. Men cost 32% less per PYO than women.
Model checking using deviance residuals suggested that 1 patient, who had consumed a large proportion of all tertiary inpatient services, might have had a significant influence on the results in the Pre-ART and Year One periods. Dropping this observation from the regressions attenuated the coefficient on CD4 cell count rise in the Year One regression to 0.97 (95% confidence interval [CI]: 0.91 to 1.04) but did not significantly affect any other results.
The patients seen in this study were of similar age and clinical features as those seen in other early public sector ART cohorts in sub-Saharan Africa.29-31 This analysis found that the average cost per patient-year for those enrolled in an ART program fell rapidly from the immediate Pre-ART period to the first 48 weeks on treatment and then fell by almost 50% over the following 64 weeks. It further found that predictors of cost shifted from WHO stage at program entry to CD4 cell count change in the first 16 weeks on treatment, and then current CD4 cell count at 48 weeks on ART.
The strengths of this study include the comprehensive data collection process, which included all persons enrolled at the clinic studied, and all care received by these patients at public health care facilities during the follow-up period. Because the clinical inclusion criteria at this facility were the same as those used in the national treatment program and the WHO's treatment criteria recommendations, it is likely that the health care cost patterns observed here are representative of those elsewhere in the country, and possibly Africa as a whole.
There were 2 important limitations to the study. First, the costing of the tertiary hospital included all costs at the institution, despite it being an academic institution. The inclusion of these nonclinical costs may have skewed the data. Concerns on this front are allayed by a sensitivity analysis where per diem personnel costs were reduced by 45% to reflect staff costs calculated in a microcosting study conducted in 1999/2000 that excluded academic components, adjusted for inflation to 2004.32 This adjustment reduced the per diem cost of a visit to the tertiary hospital by 21%, but reduced the annualized cost of care by far less: 6.7% in the Pre-ART period, 6.7% in Year One, and 3.0% in Year Two.
Second, this study may be limited by the relatively small sample size. In particular, only 25 years of Pre-ART follow-up were available for analysis. The results may thus have been influenced by a few outlying large values, although the model specification reduced this concern in multivariate modeling. Nevertheless, the influence of a single observation was sufficient to attenuate the result seen for CD4 cell change by 75%, and although there is no theoretic justification for this patient's exclusion from the sample, this finding reinforces the preliminary nature of these cost data.
Total average annualized costs in this population, particularly in the Pre-ART and Year One periods, are higher than those previously reported in the South African public sector, although they are lower than private sector program annual costs (Table 5). Part of this disparity may be attributable to this study's focus on patients initiating treatment rather than on total on-ART lifetime costs: costs were far lower for the Year Two period than for Year One. High pre- and early on-ART morbidity is likely to make direct comparisons between average lifetime costs and the Pre-ART and Year One costs in this study unreasonable, particularly given the short period of Pre-ART follow-up (median = 30 days).
The high average cost of care suggests that at least part of the gap is attributable to higher resource consumption in this population than in other public sector settings. Contributing factors may include the relatively high cost of the HCTC compared with other primary care clinics21 or a differential pattern of service utilization. Utilization patterns may differ because of the availability of services or of variation in health need. The high cost of the HCTC reflects its high ratio of doctors to nurses relative to comparable public sector sites in the Western Cape.36 This ratio arose early in the program's life, when it was one of only a handful of public sector ART clinics in country, and is now falling. The ratio may reflect the heavy use of privately funded adherence counselors, who perform many of the roles traditionally played by nursing staff in South Africa.17 Given these observations, further data are needed to evaluate fully which level of costs is most representative of the South African public ART program.
One cost factor that has changed rapidly in recent years is the cost of antiretrovirals themselves. A recent review of published sources on ART costs in Africa found annual drug costs that ranged between $278 and $1540.11 The levels reported in this study (range: $454-$491) are similar to those of other South African studies although higher than those seen elsewhere in Africa. In line with other studies from South Africa, these drug costs are lower in absolute and relative terms than those seen in higher income settings.
Predictive factors seen in the multivariate models are similar to those observed elsewhere and reflect predictors of mortality seen on ART. The importance of WHO stage has been noted in past cost models in various settings.9,10,13,14 This study adds nuance to this result, however, finding that after 1 year on treatment, baseline WHO stage ceased to be predictive of such costs. This is in line with evidence that the importance of a historic AIDS-defining illness in predicting death decreases dramatically after 6 months on treatment and is not significant after 3 years.37 After 12 months on ART, difference in CD4 cell count becomes the strongest clinical predictor of health care costs, as has been seen elsewhere.7,9
In addition to WHO stage, in the Year One period, the change in CD4 cell count by 16 weeks was a significant predictor of health care demand in this population. This reflects the importance of CD4 cell count response in the early months of treatment previously reported for mortality at this site.38 The result in the current study, however, is independent of mortality, because death is adjusted for in the model. Also, those dying in the first 4 months on treatment are not included in the Year One analysis shown because they cannot have a week 16 CD4 cell count result. This suggests that the effect of low response to ART is to increase mortality and morbidity, although the fragility of this result to the inclusion of a single observation must make this finding a cautious one that would benefit from further investigation.
In the Year One and Year Two periods, being male was associated with reduced health care expenditures, significantly so by Year Two. An analysis of the differences in categories of cost (not shown) found the most important difference in resource use to be at the primary care level, where men made one third fewer unscheduled annual clinic visits than women on average. This difference was, however, not statistically significant.
Some effects seen in past studies were not observed in this setting. Although age is a traditional risk factor for general mortality and morbidity, such an association is not seen in this population. This is likely to reflect the narrow range of ages covered in this population, with 50% falling between 29 and 39 years old and only 4 individuals being more than 50 years old. In Mexico, variation in health care demand arose from differences in supply-side factors such as income and access to services.10 This study is not able to speak to such matters, because the population seen here was economically and socially homogeneous, coming from a single district and having almost uniformly no earned income.
This study is among the first to analyze cost data for patients initiating ART in Africa. It shows a 31% decrease in cost per patient-year in the first year of ART provision compared with the immediate pretreatment period. It also shows another 45% fall in cost per patient-year between the first and second years of ART. This study suggests that factors predictive of costs change from clinical stage early in the treatment process to CD4 cell count after 12 months on treatment. These findings reinforce the importance of beginning ART before AIDS diagnosis in reducing the short-term burden of HIV on health care systems in countries with a high burden of HIV and limited resources to deal with it.
1. Arthur G, Bhatt SM, Muhindi D, et al. The changing impact of HIV/AIDS on Kenyatta National Hospital, Nairobi from 1988/89 through 1992 to 1997. AIDS
2. Colvin M, Dawood S, Kleinschmidt I, et al. Prevalence of HIV and HIV-related diseases in the adult medical wards of a tertiary hospital in Durban, South Africa
. Int J STD AIDS
3. Krentz HB, Auld MC, Gill MJ. The changing direct costs of medical care for patients with HIV/AIDS, 1995-2001. Can Med Assoc J
4. Beck EJ, Mandalia S. The cost
of HIV treatment and care in England since HAART-part 1. Br J Soc Med
5. Yazdanpanah Y. Costs associated with combination antiretroviral therapy in HIV-infected patients. J Antimicrob Chemother
6. Levy AR, James D, Johnston KM, et al. The direct costs of HIV/AIDS care. Lancet Infect Dis
7. Lowy A, Page J, Jaccard R. Costs of treatment of Swiss patients with HIV on antiretroviral therapy in hospital-based and general practice-based care: a prospective cohort study. AIDS Care
8. Hellinger FJ. Economic models of antiretroviral therapy: searching for the optimal strategy. Pharmacoeconomics
9. Chen RY, Accortt NA, Westfall AO, et al. Distribution of health care
expenditures for HIV-infected patients. Clin Infect Dis
10. Aracena B, Gutiérrez JP, Bertozzi SM, et al. Cost
of AIDS care in Mexico: what are its main individual predictors. Arch Med Res
11. Rosen S, Long L. How much does it cost
to provide antiretroviral therapy for HIV/AIDS in Africa? Center for International Health and Development discussion paper no. 9. Boston, MA: Boston University; 2006.
12. Deghaye N, Pawinski RA, Desmond C. Financial and economic costs of scaling up the provision of HAART to HIV-infected health care
workers in KwaZulu-Natal. S Afr Med J
13. Badri M, Maartens G, Mandalia S, et al. Cost
-effectiveness of highly active antiretroviral therapy
in South Africa
. PLoS Med
14. Badri M, Cleary S, Maartens G, et al. When to initiate highly active antiretroviral therapy
in sub-Saharan Africa? A South African cost
-effectiveness study. Antivir Ther
15. Cleary SM, McIntyre D, Boulle AM. The cost
-effectiveness of antiretroviral treatment in Khayelitsha, South Africa
-a primary data analysis. Cost Eff Resour Alloc
16. Cleary S, Okorafor O, Chita W, et al. Financing antiretroviral treatment and primary health care
services. In: South African Health Review 2005
. Durban, South Africa
: Health Systems Trust; 2005:58-73.
17. Bekker LG, Myer L, Orrell C, et al. Rapid scale-up of a community-based HIV treatment service: programme performance over 3 consecutive years in Guguletu, South Africa
. S Afr Med J
19. Harling G, Orrell C, Wood R. Healthcare utilization by a cohort of late-stage HIV-positive patients commencing antiretroviral medication in South Africa
[abstract WePe0085]. Presented at: XVI International AIDS Conference; 2006; Toronto.
20. Department of Health, Republic of South Africa
. User Guide-UPFS (with revised 2004 tariffs): Uniform Patient Fee Schedule for Patients Attending Public Hospitals
. Pretoria, South Africa
: Department of Health, Republic of South Africa
21. Harling G, Bekker LG, Wood R. The economics of antiretroviral provision: the evolving cost
of running a dedicated ART clinic. S Afr Med J
. (In press).
22. Provincial Administration of the Western Cape. Department of Health Budget Review 2004/2005
. Cape Town, South Africa
23. Sinanovic E, Floyd K, Dudley L, et al. Cost
-effectiveness of community-based care for tuberculosis in Cape Town, South Africa
. Int J Tuberc Lung Dis
26. Diehr P, Yanez D, Ash A, et al. Methods for analyzing health care
utilization and costs. Annu Rev Public Health
27. Dodd S, Bassi A, Bodger K, et al. A comparison of multivariate regression models to analyse cost
data. J Eval Clin Pract
28. McCullagh P, Nelder JA. Generalized Linear Models
. London: Chapman and Hall; 1989.
29. Laurent C, Gueye NF, Ndour CT, et al. Long-term benefits of highly active antiretroviral therapy
in Senegalese HIV-1-infected adults. J Acquir Immune Defic Syndr
30. Coetzee D, Hildebrand K, Boulle A, et al. Outcomes after two years of providing antiretroviral therapy in Khayelitsha, South Africa
31. Djomand G, Roels T, Ellerbrock T, et al. Virologic and immunologic outcomes and programmatic challenges of an antiretroviral treatment pilot project in Abidjan, Côte d'Ivoire. AIDS
. 2003;17(Suppl 3):S5-S15.
32. Govender V, McIntyre D, Grimwood A, et al. The Costs and Perceived Quality of Care for People Living with HIV/AIDS in the Western Cape Province in South Africa
. Partnerships for Health Reform Small Applied Research no. 14. Bethesda, MD: Abt Associates; 2000.
33. Churchyard G. Light for Life Workplace Treatment and Care Programme: the Anglo American experience. Presented at: UCLA Business and AIDS Meeting; 2006; Zimbali.
35. Martinson S, Bakos D, Mohapi L, et al. Outpatient Provider Costs of Treating Adults with ARVs in Soweto
[abstract 502]. Presented at: PEPFAR HIV/AIDS Implementers Meeting; 2006; Durban.
36. Pienaar D, Myer L, Cleary S, et al. Models of Care for Antiretroviral Service Delivery
. Cape Town: University of Cape Town; 2006.
37. Van Sighem AI, Van de Wiel MA, Ghanic AC, et al. Mortality and progression to AIDS after starting highly active antiretroviral therapy
38. Lawn SD, Myer L, Harling G, et al. Determinants of mortality and nondeath losses from an antiretroviral treatment service in South Africa
: implications for program evaluation. Clin Infect Dis