Effect of Management Strategies and Clinical Status on Costs of Care for Advanced HIV
Published Online: May 19, 2014
Paul G. Barnett, PhD; Adam Chow, BA; Vilija R. Joyce, MS; Ahmed M. Bayoumi, MD, MSc; Susan C. Griffin, MSc, BSc, PhD; Huiying Sun, PhD; Mark Holodniy, MD; Sheldon T. Brown, MD; D. William Cameron, MD; Mark Sculpher, PhD; Mike Youle, MB, ChB; Aslam H. Anis, PhD; and Douglas K. Owens, MD, MS
Human immunodeficiency virus (HIV) care costs $20,000 per year in industrialized countries.1 Highly active antiretroviral treatment (ART) accounts for approximately half of this expense.2 The annual cost of caring for persons with HIV continues to grow and now exceeds $13 billion worldwide.3
Newer ART medications have markedly improved treatment outcomes.4 Models have been used to estimate the cost effectiveness of these new drugs,5-8 and the effect of patent expiration on the cost effectiveness of older medications.9 Cost-effectiveness models may be hindered by the lack of more contemporary data on the cost of HIV care.10
HIV care is more costly when the disease is more severe, as indicated by a low cluster of differentiation (CD4) level (<50 cells/μL),11-13 a higher level of virus,14 HIV-related symptoms, 15 or an AIDS-defining illness.15,16 Results of other studies of the cost of specific services have showed that hospital costs are lower for patients who receive more ART,18,19 and that the cost of medication for opportunistic infections is lower for patients with higher CD4 count.12
Cost is higher for patients with psychiatric comorbidities, 20-22 among male patients,23 in patients presenting later for antiretroviral treatment,24,25 and for those infected by injection drug use (IDU).14 Other studies have showed baseline characteristics to be unimportant predictors of cost after current health status, as reflected by CD4 count, is considered.12
Multivariate analyses have looked at the effect of Medicaid eligibility21 and compliance with ART guidelines23 on total cost while controlling for disease severity. These studies employed ordinary least squares regression, a method that requires assumptions that are usually inappropriate for cost data, which have a highly skewed distribution. The more robust general linear models have been used only in the most recent studies of HIV cost.20,26
We report the cost of care received by participants enrolled in OPTIMA (OPTions In Management with Antiretrovirals), the largest randomized clinical trial of treatment strategies in advanced HIV. We applied advanced multivariate methods to address gaps in the knowledge about the cost of advanced HIV care: whether preexisting factors have a sustained effect on cost, and the effect of these factors and current health status on the different types of cost, including hospital services, combination ART medication, medications for opportunistic infections, and outpatient care.
The OPTIMA trial studied 2 interventions: an intended 12-week ART-free period and intensification of ART to a target of 5 or more drugs.27 A 2-by-2 factorial design evaluated both interventions simultaneously. Participants were randomized to ART interruption, ART intensification, both interventions, or neither intervention.
The trial enrolled patients who had been taking their current ART regimen for at least 3 months, had failed 2 or more prior regimens involving all 3 available ART classes in use at the time, and had evidence of current treatment failure (HIV viral load ≥2500 copies/mL and CD4 count of ≤300 cells/μL), while excluding those with an ongoing opportunistic infection.28 Consent was obtained according to the human subjects regulations of the 3 countries where participants were enrolled: the United States, Canada, and the United Kingdom.
Participants were assessed every 12 weeks and a case report form was used to record inpatient stays, outpatient visits, and changes in antiretroviral medication regimens. Unit costs for HIV care were derived from analysis of costs of HIVpositive patients at the US Veterans Health Administration (VHA), where 78% of study participants were enrolled. Inpatient costs were derived from a regression method.29 Outpatient estimates were based on the Medicare reimbursement for the procedure codes used in visits of patients with HIV.30 Cost for all participants was determined by multiplying the utilization recorded on case report forms by these unit costs: $5004 per day of hospital stay in intensive care, $1782 per day of other medical-surgical care, $768 per day of nursing home care, $201 per visit to an infectious disease clinic, and $254 per visit to other medical clinics. All costs were adjusted to 2007 US dollars using the consumer price index.
The dates that each ART medication was initiated and stopped were recorded on a case report form. Estimates of medication use assumed the guideline-recommended dosage and were adjusted by participant-reported adherence. We characterize ART medications by categories defined by the year in which the drug was approved by the US Food and Drug Administration (FDA). We identified periods in which participants received 1 of the 5 new ART medications: etravirine, tipranavir, darunavir, maraviroc, or raltegravir.
The unit costs of medications were estimated as 64% of the average wholesale price, which is the net cost of brand name medications to Medicaid.31 Medicaid is an important sponsor of US HIV care, with costs that are similar to those of the US AIDS Drug Assistance Program, another important sponsor. Since we wanted study findings to be generalizable to the US healthcare system, we did not use the lower cost paid by VHA. Data on prescription medications used to prevent or treat opportunistic infections were obtained from the pharmacy benefit management system of VHA. Although data on ART medications was available for participants who enrolled at sites in Canada and the United Kingdom, information on these concomitant medications was not available at these sites. Cost and utilization were assigned to each 3-month period following randomization. These quarterly observations were matched to clinical data indicating if the patient had an AIDS-defining event (ADE).
We further characterized health using the mean CD4 count and viral load tests conducted during the quarter. If no laboratory test was conducted in the quarter, we used the weighted average of tests that preceded and followed the quarter. Weights were inversely proportional to the number of days the test preceded the beginning of the quarter and the number of days the test followed the end of the quarter.
The association of patient characteristics with cost was assessed with Generalized Linear Model (GLM). These Take-Away Points Healthcare costs were measured in a clinical trial of alternatives for advanced human immunodeficiency virus (HIV): n Antiretroviral drugs accounted for most (61.8%) of the $26,832 annual cost of care for advanced HIV. n Costs were higher during AIDS-defining illnesses. n Costs were especially high in patients infected with HIV from injection drug use and in patients coinfected with hepatitis C. n Antiretroviral drugs have replaced hospital stays as the most costly component of HIV care, but hospitalization remains the predominant cost in especially advanced disease. The cost of interventions to improve the use of antiretroviral treatments could be offset by reduced use of other services. regressions avoid the inappropriate assumption of normal distribution and homoscedastic errors.32 A log link function was determined appropriate based on a Box-Cox regression.33 The gamma distribution was selected based on a modified Park test. Standard errors were corrected for the correlation between multiple observations from each trial participant. Independent regressors were chosen based on significance in univariate tests. Final models were chosen based on predictive fit, with attention to the extremes of the distribution. To express the effect of current clinical status in the original units of analysis—dollars cost— we used the regression parameters to predict log cost for each observation given the participant’s baseline characteristics and a set of current status variables (indicators of CD4 count, viral load, and presence of AIDS-defining illness). We exponentiated these predicted log values without correction for retransformation bias. This correction is not needed with GLM regressions. We reported the mean cost in dollars of predicted values for each set of clinical status variables, and the associated standard error.
Participants were enrolled between 2001 and 2006 and followed until 2007. This maximized study power, but there were fewer observations representing periods more distant from randomization. Bias from this design was corrected by applying an inverse probability weight, the Kaplan-Meier estimator for the probability of administrative censoring.34
There were 367 participants with cost data (excluding 1 participant with no follow-up data). Table 1 describes participant characteristics at randomization. Participants were immunologically compromised, with 32.2% having baseline CD4 counts below 50 cells/μL. Most (78%) participants enrolled at VA sites, with the balance enrolling at sites in Canada (11%) and the United Kingdom (11%). HIV infection was attributed to men having sex with men (MSM) (47.6%), heterosexual transmission (23.4%), IDU (14.1%), and infection from blood products (9.5%). There were 7 participants (1.9%) who had both MSM and IDU risk factors.
PDF is available on the last page.