Antiviral Therapy 11:

Transcription

1 Antiviral Therapy 11: A longitudinal analysis of healthcare costs after treatment optimization following genotypic antiretroviral resistance testing: does resistance testing pay off? Mathew Simcock 1,2, Pedram Sendi 1,2 *, Bruno Ledergerber 3, Tamara Keller 3, Jörg Schüpbach 4, Manuel Battegay 1, Huldrych F Günthard 3 and the Swiss HIV Cohort Study 1 Division of Infectious Diseases, University Hospital Basel, Basel, Switzerland 2 Basel Institute for Clinical Epidemiology, University Hospital, Basel, Switzerland 3 Division of Infectious Diseases and Hospital Epidemiology, Zurich University Hospital, Zurich, Switzerland 4 Swiss National Center for Retroviruses, University of Zurich, Zurich, Switzerland *Corresponding author: Tel: +41 (0) ; Fax: +41 (0) ; pedram.sendi@unibas.ch Objective: To assess the impact of antiretroviral therapy optimized by genotypic antiretroviral resistance testing (GRT) on healthcare costs over a 2-year period in patients after antiretroviral treatment failure. Study design: Non-randomized, prospective, tertiary care, clinic-based study. Patients: One-hundred and forty-two HIV patients enrolled in the ZIEL study and the Swiss HIV Cohort Study who experienced virological treatment failure. Methods: For all patients GRT was used to optimize the antiretroviral treatment regimen. All healthcare costs during 2 years following GRT were assessed using microcosting. Costs were separated into ART medication costs and healthcare costs other than ART medication (that is, non-art medication costs, in-patient costs and ambulatory [out-patient] costs). These cost estimates were then split into four consecutive 6-month periods (period 1 4) and the accumulated cost for each period was calculated. Univariate and multivariate regression modelling techniques for repeated measurements were applied to assess the changes of healthcare costs over time and factors associated with healthcare costs following GRT. Results: Overall healthcare costs after GRT decreased over time and were significantly higher in period 1 (32%; 95% confidence interval [CI]: 18 47) compared with period 4. ART medication costs significantly increased by 1,017 (95% CI: 22 2,014) Swiss francs (CHF) from period 1 4, whereas healthcare costs other than ART medication costs decreased substantially by a factor of 3.1 (95% CI: ) from period 1 to period 4. Factors mostly influencing healthcare costs following GRT were AIDS status, costs being 15% (95% CI: 6 24) higher in patients with AIDS compared with patients without AIDS, and baseline viral load, costs being 12% (95% CI: 6 17) higher in patients with each log increase in plasma RNA. Conclusions: Optimized antiretroviral treatment regimens following GRT lead to a reduction of healthcare costs in patients with treatment failure over 2 years. Patients in a worse health state (that is, a positive AIDS status and high baseline viral load) will experience higher overall costs. Introduction Antiretroviral resistance testing has become standard practice in guiding the optimal treatment of HIVinfected patients with antiretroviral drugs. Current guidelines suggest routine use of this test to guide the antiretroviral therapy (ART) choice with evidence showing virological response to therapy increases when results of the test are available compared with a regimen based on drug history and expert opinion alone [1,2]. The use of genotypic resistance testing (GRT) has become an integral part of AIDS patient management [3 6]. Use of the GRT over the phenotypic resistance test (PRT) has been recommended due to the more rapid turnaround between blood sampling and the availability of the test results [1,7]. However, in countries where the health service provider often relies on health insurers to cover the cost of the test, some find that third-party payers have often been reluctant to do so and economic considerations are therefore becoming ever more important [8 11]. Despite this, experts predict that such a test will become a major tool for guiding therapy choice [5,12,13]. The efficacy of antiretroviral resistance testing has been investigated in several studies, including a metaanalysis that contrasted different resistance tests [14]. The meta-analysis considered 10 randomized 2006 International Medical Press

2 M Simcock et al. controlled trials and concluded GRT reduced RNA levels by an additional 11% and 10% of patients after 3 and 6 months, respectively, compared with the control group. In a recent study Haupts et al. [15] investigated the impact of different drug regimens based on the GRT result and expert opinion on virological outcome through the ZIEL (Zurich HIV Genotyping and Drug Level) study. The ZIEL study prospectively followed a subset of 145 patients in the Swiss HIV Cohort Study (SHCS) [16] with virological treatment failure. The outcome was based on different follow-up viral load and CD4 + T-cell count thresholds. The study showed treatment choice guided by the GRT was virologically superior to that based on drug history alone. The economic impact of antiretroviral resistance testing has been addressed by several studies. Weinstein et al. [11] and Lauria et al. [10] showed GRT was likely to be cost effective in patients with virological treatment failure. Corzillius et al. [9] also found GRT to be cost effective, with GRT prior to the first highly active antiretroviral therapy (HAART) also being cost effective if it could lower the probability of first HAART failure by approximately onethird. However, these studies were based mainly on disease modelling [17,18] and were not conducted alongside a prospective clinical study using patientlevel data. It is therefore unclear how antiretroviral therapy optimized by resistance testing affects healthcare costs in the period after treatment adjustment in a real-world setting. We therefore present a cost analysis alongside the ZIEL study using patient-level data assessing healthcare costs in patients with GRT after virological treatment failure. Methods This economic evaluation was conducted alongside a previously published study (ZIEL study) that investigated the impact of GRT on the selection of salvage regimens in patients with virological treatment failure [15]. This study used a subset of patients enrolled in the SHCS who attended the HIV outpatients clinic at University Hospital Zurich, Switzerland. The ZIEL study design was a non-randomized, prospective, tertiary care, centre-based study. Patients were eligible for entry into the study if they were considered to be experiencing virological treatment failure. Two new treatment regimens were then recommended in replacement of the failing ART. The two treatment regimens were those as recommended by a clinician/expert (expert ART) and the GRT (genotype ART). The expert ART was recommended by the treating physician and was based on the patients previous drug history. The genotypic ART was generated by software designed to rank antiretroviral drugs according to their level of resistance provided by the GRT using the decision support software Retrogram (Virology Networks, Utrecht, The Netherlands) version 1.2 (April 2000), version 1.4 (August 2001) or version 1.6 (since August 2002). The resistance test, ViroSeq HIV-1 genotyping system [19], relied on a sample of plasma HIV-1 RNA for the direct sequencing. Once the genotype ART regimen had been made available to the treating physician, the final treatment regimen was then agreed upon with the patient (new ART). A resistance score (RS) for each of the failing and new ART regimens was calculated (failing RS and new RS). To calculate the RS, each drug in an ART regimen was given an alphabetic ranking from A to D. Those ranked with an A indicated absence of drug resistance. B and C reflected that a particular drug was associated with a detected mutation of the HIV strain, each with an increasing resistance level respectively. D finally indicated a high level resistance. The letters A to D were then assigned a defined exponential weight of 1, 2, 4 and 8 respectively [15,20,21]. The sum of the weights for each drug in the ART was then calculated and the average taken to produce the RS. Patients that recorded a failing RS equal to 1 were classified as being non-compliant with the failing treatment regimen, as no HIV resistance was detected with their failing ART. Clinical information collected included the CD4 + T-cell counts, viral load, the amount of time the patient had been receiving ART, and where appropriate, the amount of time since receiving a HAART, both recorded in years. Demographic information included the patient s age, sex, AIDS status and whether they were infected by injecting drugs (IDU). The number of genotypic resistance tests each patient had in the 2-year follow-up period, including the initial (baseline) test, was also recorded. Costs All healthcare costs that accrued after resistance testing during the follow-up period of 2 years were extracted from patient charts using microcosting [22]. These costs were split into four main units. These units consisted of observed ART medication costs, non-art medication costs, in-patient costs and ambulatory (outpatient) costs. Each cost unit was then split into four consecutive 6-month periods and the accumulated cost for each period calculated. ART medication costs included the costs of all antiretrovial drugs used during the follow-up period. The official drug prices according to the Swiss compendium for pharmaceutical products (Arzneimittelkompendium: were used to estimate the costs of the drug regimens for each patient International Medical Press

3 Analysis of healthcare costs after genotypic antiretroviral resistance testing Non-ART medication costs included drugs other than antiretroviral drugs such as HIV related drugs for the prevention of opportunistic diseases (for example, antibiotic, antifungal and antiviral drugs), and non-hiv related drugs such as drugs to treat cardiovascular disease or diabetes (for example, antihypertensive, lipid-lowering and anti-diabetic agents). Furthermore, pain relievers, antacids, antidepressants, neuroleptics and other drugs were also included. In-patient costs reflected resource use that accrued when a patient was admitted to the hospital for at least one stay overnight. The amount that was charged to a health insurer was used to approximate the healthcare costs. Ambulatory costs included the cost for doctor s appointment, viral load, CD4 + T-cell counts, drug levels, blood chemistry, blood count and all other diagnostic procedures such as serological, radiological, endoscopic and/or cardiological examinations. Ambulatory costs also included costs for patient treatment that did not require hospitalization, excluding non-art medication costs as described above. Also included was the cost of each GRT carried out; each test costing Swiss francs (CHF) 800. The cost estimates were based on official tariffs used in Switzerland (Analyseliste and Spitalleistungskatalog: Federal Office of Insurance). In addition to these cost units, the daily ART medication cost attributed to each of the four ART regimens was estimated (that is, failing ART, expert ART, genotypic ART and new ART). It should be noted that estimated failing ART regimen costs represent the daily ART costs of the failing ART before resistance testing. Expert and genotypic ART costs are hypothetical daily cost estimates based on the respective recommended regimen. New ART represents the final antiretroviral regimen cost that was actually provided to the patient after resistance testing. All monetary values presented in this paper are in 2002 CHF. One-hundred Swiss francs corresponded to $US on July 1 st 2002 ( Imputing missing data The importance of imputing missing cost data has been highlighted in many health economics papers, and it has been shown that a complete case analysis (that is, excluding patients with missing values from the analysis) may lead to biased results [23,24]. We therefore imputed missing cost data where appropriate. The percentages of missing values in the study data set were 11.62% for ART medication costs, 29.05% for non-art medication costs, 7.75% for ambulatory costs and 0.00% for in-patient costs. The reason for the missing data is simply down to the information not being available for collection or extraction. For this reason Rubin s missing data mechanism [25] can be ignored, allowing a wider range of possible imputation techniques to be considered. Many forms of imputation techniques were used and a standard simple repeated measures (RM) linear regression analysis was applied to each of the imputed data sets. The RM model was used to allow for the repeated cost measurements from one period to another for each patient, thus reducing within patient variation between the RM (that is, allowing for the correlation between time points). The random regression [23] technique was found to be the best method of imputation as it produced the lowest Akaike Information Criteria (AIC) value and also reduced the standard error (SE) surrounding most of the variables in the RM model the most. This result was found to exist across all the RM analyses for each of the different cost units. The random regression technique involved producing a mixed effects RM model that included most of the patients clinical measurements and demographics in the fixed effects part of the model with a random effect based on the error surrounding the intercept for each patient. Imputed values that fell below 0 were set to 0 to avoid negative monetary values. Statistical analysis Once missing cost data had been imputed, summary and descriptive statistics were produced. The four different cost units were then summarized into three cost groups of main interest: ART medication costs; all costs excluding ART medication costs (other costs); and all costs combined (total costs). In the cases where the cost data was not normally distributed (other costs and total costs) a natural log transformation was applied to allow the model residuals to be normally distributed. The respective cost group, using the log transformation if appropriate, was modelled as the response variable in the RM models for the univariate and multivariate analyses. The univariate analysis consisted of simple RM linear regression assessing the relationship between all available clinical and demographic variables against each cost group. In the same way as the univariate analysis, the multivariate analysis was carried out to produce a final model of the variables found to be predictors collectively of the cost groups. To avoid collinearity problems with the RS difference between the failing and new RS in the multivariate analysis, the failing RS and new RS were left out of the analysis. Significant and important interactions that were considered included the variable gender, but with such different percentages of males against females, an interaction that includes gender would not be appropriate. To understand yet further the effect that GRT has on long-term costs, the period effect was also investigated with the cost of each GRT removed from the costs. Antiviral Therapy 11:3 307

4 M Simcock et al. The models based on the untransformed costs produced estimates based on the mean of the arithmetic scale, whereas estimates modelled on the log scale are based on the mean of the log transformation. These estimates were converted to the arithmetic scale from the log scale by exponentiating them all. These exponentiated estimates produced means based on the geometric scale which are lower than the arithmetic mean. Therefore the exponentiated estimates should be interpreted as a rate of the median instead of an increase of the mean. Significant predictors using the untransformed arithmetic response show corresponding 95% confidence intervals (CIs) that do not include 0, whereas those based on the geometric scale have corresponding 95% CIs that do not contain 1. To understand how the costs alter over time, the period effect was included in the multivariate models. All analyses were carried out using the Statistical Analysis Systems (SAS) package (SAS Institute, Cary, NC, USA, version 8.2). Results Patient characteristics and summary statistics A total of 145 patients with a mean age of 42 years were recruited into the clinical study between August 2000 and November Cost data for 142 patients out of the 145 patients was available, 13 of which it was decided that treatment should be stopped. The baseline patient characteristics are shown in Table 1. Approximately 80% of those enrolled were male. Patients classified as being non-compliant accounted for 24 (16.90%) of the patients in the study. The average estimated daily ART medication costs associated with each of the four ART regimens (Table 2), show the lowest estimated average daily costs are associated with the failing ART, with all the other estimated daily ART medication costs being much higher. The failing ART was significantly different from the other average estimated daily ART medication costs (P<0.0001) as shown in Figure 1. The box-plots in Figure 2 depict the total costs accumulated over the 2-year follow-up period split for each 6-month period with the cost of the GRT included. High and low leverage points are shown mainly in the first period. These are plausible points due to the potentially great variations in costs of different treatments. This plot also shows the data to be highly skewed. Period one was found to be statistically significantly different from the other periods (P<0.0001) with a similar statistical result found when the cost of GRT was excluded from total costs. The highest costs incurred are those based on ART medication costs, which increased from the first Table 1. Patient characteristics and descriptive at baseline (n=142) Descriptive Mean (SD)* Range Age 42 (9.10) Male, n [%] 113 [79.58] Defined resistance score Failing resistance score 4.22 (2.07) Expert resistance score 2.42 (1.55) Genotype resistance score 2.03 (1.27) New resistance score 2.88 (1.95) CD4 + T-cell counts (cells/µl) (182.57) , Viral load, HIV-1 RNA log 10 copies/ml 3.90 (0.80) Time since start of ART, years 4.62 (2.44) Time since start of HAART, years 3.59 (1.52) n [%] 137 [96.48] Injected drug user, n [%] 36 [25.35] AIDS CDC stage C, n [%] 55 [38.73] Non-compliant patients, n [%] 24 [16.90] Treatment type: followed genotypic 71 (50.00) test recommendation Period 1: number of genotypic resistance tests carried out Test number=0, n [%] 0 [0.00] Test number=1, n [%] 135 [95.23] Test number=2, n [%] 6 [4.23] Test number=23, n [%] 1 [0.70] Period 2: number of genotypic resistance tests carried out Test number=0, n [%] 136 [95.77] Test number=1, n [%] 6 [4.23] Period 3: number of genotypic resistance tests carried out Test number=0, n [%] 134 [94.37] Test number=1, n [%] 8 [5.63] Period 4: number of genotypic resistance tests carried out Test number=0, n [%] 133 [93.66] Test number=1, n [%] 8 [5.63] Test number=2, n [%] 1 [0.70] *Values are means (SD), unless otherwise indicated. The resistance score for the new treatment as expressed for the whole study population rose again when compared with the genotypic resistance score due to the fact that a fraction of patients stopped or continued their failing treatment when physicians felt that there were no treatment options. However, if only patients who effectively went on the new treatment were considered the improvement of the resistance score as obtained by the genotypic resistance score could be maintained as described before [15]. CDC, Centers for Disease Control and Prevention; HAART, highly active antiretroviral therapy. period to the second and third period before decreasing slightly in the fourth period. Other cost units show an opposite trend, with ambulatory, International Medical Press

5 Analysis of healthcare costs after genotypic antiretroviral resistance testing Table 2. Summary statistics for each type of cost over the four 6-month follow-up periods (n=142) Type of Cost Mean (SD) ART medication Period 1 10, (3,829.02) Period 2 11, (4,673.92) Period 3 11, (4,911.94) Period 4 11, (4,940.90) Combined 11, (4,631.03) Non-ART medication Period 1 3, (7,722.70) Period (2,518.44) Period (378.47) Period (2,476.09) Combined 1, (4,466.45) Ambulatory (without GRT) Period 1 3, (3,534.97) Period 2 2, (3,328.37) Period 3 2, (5,078.12) Period 4 1, (1,253.88) Combined 2, (3,598.59) In-patient Period 1 1, (5,351.97) Period (3,064.84) Period 3 2, (7,104.84) Period (3,474.82) Combined 1, (5,026.49) Cost of GRT Period (208.19) Period (161.50) Period (185.11) Period (226.43) Combined (398.51) Total costs (with GRT) Period 1 19, (12,486.05) Period 2 15, (7,733.01) Period 3 16, (10,431.70) Period 4 14, (7,258.71) Combined 16, (9,847.83) Total costs (without GRT) Period 1 18, (12,451.11) Period 2 15, (7,724.31) Period 3 16, (10,424.93) Period 4 14, (7,253.44) Combined 16, (9,804.77) Estimated daily costs Failing ART (11.45) Expert ART (16.34) Genotype ART (17.27) New ART (18.54) ART, antiretroviral therapy; GRT, genotypic resistance testing. in-patient and non-art medical costs decreasing from the first to the fourth period, although inpatient costs increased slightly again during the third Figure 1. Box-plots for each estimated daily costs for the failing ART and the different types of salvage ART regimens (genotypic, expert and actual prescribed new regimen) Daily ART costs Failing Genotype Expert New ART regimen Reference line set to the mean of the failing antiretroviral therapy (ART). Figure 2. Box-plots for the total costs over the four follow-up periods, each 6 months Total costs 100,000 75,000 50,000 25, Period period before decreasing during the fourth period (Table 2 and Figure 3). Drugs used in the new ART regimes are shown in Table 3 along with the associated mean costs. The table shows the number of patients to take each type of drug for each period. The most expensive drugs, based on the mean, were protease inhibitors during Reference line set to the mean total cost of period one. Antiviral Therapy 11:3 309

6 M Simcock et al. Figure 3. Different types of cost units observed during the four follow-up periods in the study, each lasting 6 months Mean cost, CHF 12,000 11,000 10,000 5,000 4,000 3,000 2,000 1, Period Cost unit: ART medication Ambulatory (excluding GRT) Ambulatory (including GRT) Non-ART medication In patient ART, antiretroviral therapy; GRT, genotypic resistance testing. the first period and fusion inhibitors for the second, third and fourth periods. However, the number of patients receiving fusion inhibitors was very low (one in period 1, two in the second period and four and six in the third and fourth period respectively). Univariate analysis The results of the univariate analysis are shown in Table 4. The period effect was statistically significant with an increase in ART medication cost between the first and the fourth period with the average difference from the first and fourth period being CHF 1,167. On the other hand, there was a significant decrease in the costs other than ART medication costs, these costs being 3.10 times higher in the first period than in the fourth when including the cost of GRT, excluding the cost of GRT and the difference between the first period compared with the fourth period was 2.68 times. Total costs in period one were 32% and 25% higher than in the fourth period when including and excluding the cost of GRT respectively. The common variables found to be predictors in all three cost groups besides the period effect were AIDS status, whether the treatment was stopped, and viral load. Of these common predictors, the one that had the biggest impact on cost was whether the patient s treatment was stopped. Patients whose treatment was stopped resulted in ART medication costs being reduced by CHF 3,017 compared with those whose treatment was not stopped. This cost reduction was also seen for all other costs other than ART medication costs and for all costs combined, this reduction being 32% of the median cost for each group. Other variables found to be predictors of the total cost were RS difference, gender, CD4 + T-cell count and both the time spent on ART and HAART. Females had a cost reduction of 19% compared with males. A wider difference between the failing RS and new RS was associated with higher overall healthcare costs. The longer the time a patient had been receiving ART or HAART resulted in total costs being raised as a result of more costly ART medications. Non-compliant patients had a reduced ART medication cost but an increase in the other costs, this result not being statistically significant. Additional GRT during the follow-up period did not affect the overall costs significantly. Although the costs estimated for each period altered when the cost of the GRT was excluded from the analysis compared with when it was included, the effect this had on the individual variables was minimal with little difference in the estimated means and medians. Multivariate analysis The results of the multivariate analysis are shown in Table 5. The period effect was found to be statistically significant across all three cost groups. ART medication costs were CHF 1,018 less in the first period compared with the fourth period. However, costs other than ART medication costs were 3.11 times higher in Table 3. Summary statistics of the drugs used to make up the salvage ART, representing the total number of types of drugs used with periods combined Mean cost of HIV drug type used in each study period Drug type Period 1 (n, SD) Period 2 (n, SD) Period 3 (n, SD) Period 4 (n, SD) PI 4, (135, ) 5, (127, ) 5, (119, ) 5, (115, 2,196.49) NRTI 4, (154, ) 4, (142, ) 5, (135, ) 4, (135, 1,837.77) NNRTI 1, (81, ) 2, (68, ) 2, (61, ) 2, (62, ) FI 7, (1, -) 11, (2, ) 16, (4, 0.00) 10, (6, 6,521.03) ART, antiretroviral therapy; FI, fusion inhibitor; NRTI, nucleotide/nucleoside reverse transcriptase inhibitor; NNRTI, non-nrti; PI, protease inhibitor International Medical Press

7 Analysis of healthcare costs after genotypic antiretroviral resistance testing Table 4. Univariate analysis of association between clinical and demographic variables against the different cost groups Cost group: estimate (95% CI) Factor Mean ART medication cost (CHF) Ratio of median other cost Ratio of median total cost RS difference, failing new (344.73, )* 0.98 (0.9385, ) 1.04 (1.0150, )* Gender, female vs male -2, (-3,271.03, -1,419.81)* 0.93 (0.7718, ) 0.81 (0.7283, ) AIDS status, positive vs negative 1, (588.97, )* 1.28 (1.0948, )* 1.19 (1.0949, )* IDU status, positive vs negative -1, (-2,096.61, )* 0.99 (0.8326, ) 0.92 (0.8337, ) Non-compliant -1, (-2,898.10, )* 1.23 (1.0002, )* 0.94 (0.8342, ) Stopped treatment -3, (-4,315.41, -1,718.99)* 0.68 (0.5191, )* 0.68 (0.5907, )* Number of genotypic ( , ) 1.22 (1.0420, )* 1.04 (0.9554, ) resistance tests Baseline CD4 + T-cell count ( , 80.56) 0.93 (0.8939, )* 0.96 (0.9379, )* (cells/µl) per 100 cells Baseline viral load, log RNA copies/ml (361.27, 1,305.75)* 1.26 (1.1502, )* 1.14 (1.0771, )* Time on ART, years (87.71, )* 1.03 (0.9959, ) 1.03 (1.0104, )* Time on HAART, years (286.38, )* 0.96 (0.9084, ) 1.02 (1.0071, )* Failing RS (78.13, )* 0.96 (0.9286, ) 1.01 (0.9906, ) New RS ( , 56.89) 0.97 (0.9354, ) 0.98 (0.9610, ) Period effect with GRT P-value * <0.0001* <0.0001* Period 1-1, (-2,239.30, )* 3.10 (2.7176, )* 1.32 (1.1736, )* Period (-1,346.69, ) 1.21 (1.0634, )* 1.03 (0.9120, ) Period ( , 1,360.39) 1.17 (1.0237, )* 1.11 (0.9856, ) Period Period effect without GRT P-value * <0.0001* < Period 1-1, (-2,239.30, )* 2.68 (2.2112, )* 1.25 (1.1108, )* Period (-1,346.69, ) 1.23 (1.0074, )* 1.03 (0.9095, ) Period ( , 1,360.39) 1.18 (0.9676, ) 1.11 (0.9846, ) Period *Statistically significantly associated at the 5% significance level. Interpreted as odds ratio of the estimated median. ART, antiretroviral therapy; CI, confidence interval; GRT, genotypic resistance testing; HAART, highly active antiretroviral therapy, RS, resistance score. period one compared with period four with total costs being 32% higher in the first period than the fourth period when including the cost of GRT, when excluding the cost of GRT the costs were 2.70 and 1.25 times more, respectively. The common variables collectively considered as predictors of the cost groups are the same as those found to be statistically significant in the univariate analysis after the period effect (AIDS status, RS difference, whether the patients treatment was stopped and viral load). The greater the reduction of drugs against which the virus has developed resistance (high resistance), hence leading to a drug regimen with increased antiviral efficacy (low resistance), the average ART medication costs were being increased by CHF 434 for each unit increase between the old and new RS. The opposite was noted in the other costs, leading to total costs increasing by 3% on the median cost for each unit increase between the old and new RS. Patients with a positive AIDS status incurred costs 1.15 times higher than patients with a negative AIDS status. Patients whose antiretroviral treatment was stopped had costs reduced by 24% compared with those whose treatment was not stopped. A higher viral load resulted in a higher cost with each unit increase in the baseline viral load, the total costs increasing by 1.12 times. Again it was found that the means and medians estimated for each of the periods differed, although the same statistical significance remained, between whether the cost of GRT was included or excluded from the analysis. However, the estimates for the other variables did not change and the model remained the same. Antiviral Therapy 11:3 311

8 M Simcock et al. Table 5. Multivariate analysis of factors associated with different types of costs after antiretroviral resistance testing Cost group: estimate (95% CI) Factor Mean ART medication cost (CHF) OR of other cost* OR of total cost* RS difference, failing new (222.46, ) 0.96 (0.9220, ) 1.03 (1.0007, ) Stopped treatment -1, (-3,033.30, ) 0.64 (0.4955, ) 0.76 (0.6580, ) Gender, female vs male -1, (-2,808.59, ) 0.85 (0.7729, ) AIDS status, positive vs negative (10.81, 1,470.89) 1.24 (1.0867, ) 1.15 (1.0577, ) IDU status, positive vs negative -1, (-1,995.00, ) Baseline viral load, (228.68, 1,118.56) 1.22 (1.1234, ) 1.12 (1.0638, ) log RNA copies/ml Time on ART, years 1.08 (1.0425, ) 1.02 (1.0038, ) Time on HAART, years (254.53, ) 0.87 (0.8135, ) Period effect with GRT P-value < < Period 1-1, (-2,014.37, ) 3.11 (2.5871, ) 1.32 (1.1831, ) Period (-1,202.85, ) 1.21 (1.0055, ) 1.03 (0.9192, ) Period ( , 1,356.67) 1.18 (0.9775, ) 1.11 (0.9933, ) Period Period effect without GRT P-value < Period 1-1, (-2,014.37, ) 2.70 (2.2393, ) 1.25 (1.1197, ) Period (-1,202.85, ) 1.23 (1.0158, ) 1.03 (0.9168, ) Period ( , 1,356.67) 1.18 (0.9803, ) 1.11 (0.9925, ) Period *Interpreted as odds ratio (OR) of the estimated median. ART, antiretroviral therapy; CI, confidence interval; GRT, genotypic resistance testing; HAART, highly active antiretroviral therapy, RS, resistance score. Discussion In this paper we found that the additional cost of using more complex and potent antiretroviral drug regimens following antiretroviral GRT are offset by the reduction in other healthcare costs, with total overall healthcare costs significantly decreasing over 2 years (Table 4 and Table 5). One of the main points of interest to note in this study is how the costs change over the 2-year follow-up period from the initial GRT being carried out. Based on ART medication costs using only the multivariate model, the first period accumulated a mean costs of CHF 2,107 before increasing through the second period to the third period with costs averaged at CHF 2,919 and CHF 3,432 for these periods respectively. The accumulated cost in the fourth period dropped by CHF 360 from the third period to CHF 3,125 after adjustment for other variables. The costs of the ART medication accounted for most of the costs incurred during the study period with the other costs, in comparison, being much lower (Table 2 and Figure 3). The opposite in the trend of other costs incurred over time was noticed compared with the ART medication costs. These costs started high but then gradually reduced from the first period through to the fourth period. The greatest change in period costs was between the first and second period. The costs of the first, second and third period compared with the fourth period were 3.11, 1.21 and 1.18 times higher, respectively, when including the cost of GRT, and 1.25, 1.03 and 1.11 times higher, respectively, when excluding the cost of GRT (Table 5). This overall change over time of the total costs resulted in a statistically highly significant reduction in costs between the first and fourth period. Without modelling the data, the mean total cost over time from the first period excluding the cost of the GRT (Table 2) shows ART optimized by GRT may lead to a reduction in total costs over time. Of note, the cost of GRT itself only represented a minor proportion of the total healthcare costs. It must be clearly stated that our study design does not allow to answer the question does genotypic resistance testing results in a cost reduction overall? directly because this was a non-randomized trial. Nevertheless, several findings of our investigation suggest that genotyping does indeed reduce costs overall. In the ZIEL study we have shown that genotyping helps physicians choosing antiretroviral drug regimens with higher potency as expressed by lower International Medical Press

9 Analysis of healthcare costs after genotypic antiretroviral resistance testing resistance scores [15]. Thus, resistance testing clearly helped to optimize potency of antiretroviral drug regimens [15]. Furthermore, we demonstrated a highly significant correlation between virological response and the resistance score of the new ART, suggesting that the gain in potency of drugs chosen can actually also be translated into virological response. In large cohort studies, viral load measurements have shown to be strong independent predictors of disease progression during the natural history of the HIV disease [26] and in patients receiving antiretroviral treatment [27]. Thus, viral load is a surrogate marker for disease progression, and disease progression is unambiguously associated with higher healthcare costs. This cause and effect relationship between disease status, viral load and costs is also evident by our own finding that higher costs are associated with AIDS status and higher viral load. We have shown that GRT does not statistically significantly impact on the total costs, although the costs in period one are increased slightly if the cost of the GRT is included in the analysis. Other studies have addressed the impact of resistance testing on medical costs [9 11]. However, those investigations have differed from the study presented here as they have used a cost-effectiveness analysis, whereas we performed a longitudinal analysis of cost data without an explicit comparator to GRT. As GRT is likely to increase survival by reducing the likelihood of disease progression, overall healthcare costs over the increased lifetime are usually higher and cost-effectiveness analyses are therefore needed to address the issue. However, our study nonetheless demonstrates that within a real-life setting using patient-level data healthcare costs decreased over a period of 2 years. Our study has several limitations. First, the model does not allow for the follow-up GRT past the initial test in terms of a re-calculated RS. This is particularly important since the difference between the failing RS and the new RS was found to be a main predictor of the total costs but does not consider further changes. However, the principle of showing that the more one was able to improve the RS of a new ART regimen when compared with the failing regimen that is, the greater the increase in more potent regimens resulting in higher ART medication costs was well illustrated in both the univariate and multivariate analysis. The number of patients in the study who had more than one GRT was small (n=28 [19.72%]) with a total of no more than three GRTs being performed on any one patient in the follow-up period (Table 1). The model, however, did take into account the cost of the extra GRT as well as the number of resistance tests carried out, but neither factor was found to be of statistical importance (Table 4 and Table 5). Second, there were missing cost values in the original data set before imputation. In any study, missing data is problematic, but not accounting for the missing values leads to a loss of valuable information and potentially biased results. Though imputing the missing values does not completely eliminate bias from the analysis, it does prevent the sample size from being reduced, thus helping to maintain the validity of analyses carried out. Many forms of data imputation were considered in this study, before the random regression technique was chosen as the most suitable. Again, this method is conditioned around not only patient s clinical measures, but also their demographics, making the generated values more realistic to each individual patient for that particular period in time with the use of the RM model. Third, due to the different scales being used for the analysis of the ART medication costs and the costs other than the ART medication costs, it is not possible to directly compare how much the cost of performing GRT and the expense of the consequent treatment is offset by a reduction in the other costs. However, it is possible to relate the increase in ART medication costs with the decrease in overall healthcare costs (including ART medications), showing a clear long-term cost reduction when all costs are combined (Table 4 and Table 5). Finally, cost data prior to study entry were not available for inclusion in our analysis. Such data would increase the precision of our cost estimates, as resource consumption within a patient is correlated over subsequent periods, but we would not expect that the point estimates in our regression models would change substantially. This study has set out to produce a longitudinal cost analysis without an explicit comparator, and some questions still remain. It is important to remember that costs collected were over a 2-year period, and the impact of GRT on long-term costs is not yet known. This issue would need to be addressed within a costeffectiveness analysis where costs and health effects are projected over a life-time. Finally, it is important to note that we only addressed the impact of GRT on costs; the clinical benefit of GRT within the contexts of the ZIEL study has been reported by Haupts et al. in a previous article in this journal [15]. In conclusion, we have shown that the use of GRT optimized treatment resulted in more effective antiretroviral therapy regimens with cost reductions over a period of 2 years. Acknowledgements We are grateful to Reto Nüesch for his valuable comments on an earlier draft of this article. We would like to thank the patients for their commitment and the clinical team of the Division of Infectious Diseases for excellent patient care and G Maag and H Jacober for Antiviral Therapy 11:3 313

10 M Simcock et al. their excellent support in managing the archive of the charts. Moreover, we are indebted to M Aebli and her team of the University Hospital Zurich and many hospitals from all over Switzerland for providing detailed ambulatory and stationary costs for this study. This cost analysis was supported by the Swiss National Science Foundation (SNF project # 3345CO /1 and SHCS project # 400). Pedram Sendi and Manuel Battegay are the principal investigators of these projects. Futhermore, the initial clinical trial was financed in the framework of the Swiss HIV Cohort Study (Grant ) and was furthermore supported by an unrestricted research grant from Roche Pharma Schweiz AG, Reinach, Switzerland (M Borer). Swiss HIV Cohort Study Members The Swiss HIV Cohort Study is supported by the Swiss National Science Foundation (Grant number ). The members of the Swiss HIV Cohort Study are S Backmann, M Battegay, E Bernasconi, H Bucher, Ph Bürgisser, M Egger, P Erb, W Fierz, M Fischer, M Flepp (Chairman of the Clinical and Laboratory Committee), P Francioli (President of the SHCS, Centre Hospitalier Universitaire Vaudois, CH-1011, Lausanne), HJ Furrer, M Gorgievski, H Günthard, P Grob, B Hirschel, L Kaiser, C Kind, Th Klimkait, B Ledergerber, U Lauper, D Nadal, M Opravil, F Paccaud, G Pantaleo, L Perrin, J-C Piffaretti, M Rickenbach (Head of Data Centre), C Rudin (Chairman of the Mother & Child Substudy), J Schüpbach, R Speck, A Telenti, A Trkola, P Vernazza (Chairman of the Scientific Borad), R Weber and S Yerly. Reference 1. Guidelines for the use of antiretroviral agents in HIVinfected adults and adolescents. U.S. Department of Health and Human Services (DHHS). BETA 1997; Aceti A, Carosi G. Genotypic resistance tests in the management of the HIV-infected patient at virological failure. Scand J Infect Dis Suppl 2003; 35 Suppl 106: Blum RA, Wylie N, England T, French C. HIV resistance testing in the USA a model for the application of pharmacogenomics in the clinical setting. Pharmacogenomics. 2005; 6: Clinical and laboratory guidelines for the use of HIV-1 drug resistance testing as part of treatment management: recommendations for the European setting. The EuroGUidelines Group for HIV resistance. AIDS 2001; 15: Hirsch MS, Brun-Vezinet F, D Aquila RT, et al. Antiretroviral drug resistance testing in adult HIV-1 infection: recommendations of an International AIDS Society-USA Panel. JAMA 2000; 283: Hirsch MS, Brun-Vezinet F, Clotet B, et al. Antiretroviral drug resistance testing in adults infected with human immunodeficiency virus type 1: 2003 recommendations of an International AIDS Society-USA Panel. Clin Infect Dis 2003; 37: Vandamme AM, Sonnerborg A, Ait-Khaled M, et al. Updated European recommendations for the clinical use of HIV drug resistance testing. Antivir Ther 2004; 9: Chaix C, Grenier-Sennelier C, Clevenbergh P, et al. Economic evaluation of drug resistance genotyping for the adaptation of treatment in HIV-infected patients in the VIRADAPT study. J Acquir Immune Defic Syndr 2000; 24: Corzillius M, Muhlberger N, Sroczynski G, Jaeger H, Wasem J, Siebert U. Cost effectiveness analysis of routine use of genotypic antiretroviral resistance testing after failure of antiretroviral treatment for HIV. Antivir Ther 2004; 9: Lauria FN, Angeletti C. Cost-effectiveness analysis of using antiretroviral drug resistance testing. Scand J Infect Dis Suppl 2003; 35 Suppl 106: Weinstein MC, Goldie SJ, Losina E, et al. Use of genotypic resistance testing to guide hiv therapy: clinical impact and cost-effectiveness. Ann Intern Med 2001; 134: Dean GL, Fisher M, Loveday C. HIV drug-resistance testing on archived samples to help current clinical decisions. AIDS 2000; 14: Stephenson J. HIV drug resistance testing shows promise. JAMA 1999; 281: Panidou ET, Trikalinos TA, Ioannidis JP. Limited benefit of antiretroviral resistance testing in treatment-experienced patients: a meta-analysis. AIDS 2004; 18: Haupts S, Ledergerber B, Boni J, et al. Impact of genotypic resistance testing on selection of salvage regimen in clinical practice. Antivir Ther 2003; 8: Ledergerber B, von Overbeck J, Egger M, Luthy R. The Swiss HIV Cohort Study: rationale, organization and selected baseline characteristics. Soz Praventivmed 1994; 39: Buxton MJ, Drummond MF, Van Hout BA, et al. Modelling in economic evaluation: an unavoidable fact of life. Health Econ 1997; 6: Sendi PP, Craig BA, Pfluger D, Gafni A, Bucher HC. Systematic validation of disease models for pharmacoeconomic evaluations. Swiss HIV Cohort Study. J Eval Clin Pract 1999; 5: Mracna M, Becker-Pergola G, Dileanis J, et al. Performance of Applied Biosystems ViroSeq HIV-1 Genotyping System for sequence-based analysis of nonsubtype B human immunodeficiency virus type 1 from Uganda. J Clin Microbiol 2001; 39: Condra JH, Holder DJ, Schleif WA, et al. Genetic correlates of in vivo viral resistance to indinavir, a human immunodeficiency virus type 1 protease inhibitor. J Virol 1996; 70: Kempf DJ, Isaacson JD, King MS, et al. Identification of genotypic changes in human immunodeficiency virus protease that correlate with reduced susceptibility to the protease inhibitor lopinavir among viral isolates from protease inhibitor-experienced patients. J Virol 2001; 75: Drummond MF, O Brian B, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Healthcare Programmes. New York: Oxford University Press USA; Briggs A, Clark T, Wolstenholme J, Clarke P. Missing... presumed at random: cost-analysis of incomplete data. Health Econ 2003; 12: Oostenbrink JB, Al MJ. The analysis of incomplete cost data due to dropout. Health Econ 2005; 14: Rubin DB. Inference and missing data. Biometrika 1976; 63: Mellors JW, Rinaldo CR, Jr., Gupta P, White RM, Todd JA, Kingsley LA. Prognosis in HIV-1 infection predicted by the quantity of virus in plasma. Science 1996; 272: Chene G, Sterne JA, May M, et al. Prognostic importance of initial response in HIV-1 infected patients starting potent antiretroviral therapy: analysis of prospective studies. Lancet 2003; 362: Accepted for publication 9 January International Medical Press