Lung cancer is the leading cause of cancer death in the U.S. for both. Cost-Effectiveness and Lung Cancer Clinical Trials

Transcription

1 1491 Cost-Effectiveness and Lung Cancer Clinical Trials Wei Du, Ph.D. 1,2 Jaxk H. Reeves, Ph.D. 3 Shirish Gadgeel, M.D. 1,2 Judith Abrams, Ph.D. 1,2 William P. Peters, M.D., Ph.D. 1,2 1 Center for Cancer Economics, Technology Assessment, Innovation and Development (CETAID), Karmanos Cancer Institute, Detroit, Michigan. 2 Department of Internal Medicine, Wayne State University, Detroit, Michigan. 3 Department of Statistics, University of Georgia, Athens, Georgia. BACKGROUND. Lung cancer is the leading cause of cancer death in the U.S., with an estimated annual economic burden of $5 billion. Clinical trials offer innovative therapeutic options with potentially better outcomes, but their effects on health care costs are disputed. METHODS. The authors analyzed the 1-year facility-based treatment cost and survival of 336 newly diagnosed nonsmall cell lung cancer patients who were deemed eligible for clinical trials between 1994 and 1998 at the Karmanos Cancer Institute. The incremental cost-effectiveness ratio (ICER) of clinical trial treatments with adjustment for confounders was calculated along with its 95% confidence interval (CI) using the bootstrap resampling method. RESULTS. Of the 336 patients, 76 (22.6%) were treated on clinical trials. Trial participation was associated significantly with race (P 0.01), gender (P 0.01), age (P 0.02), and insurance type (P 0.02). The average 1-year cost for trial enrollees was $41,734 with a median survival of 1.3 years, whereas the average 1-year cost for nonenrollees was $34,191 with a median survival period of 0.9 years. Differences in survival and 1-year cost between enrollees and nonenrollees were significant when controlling for age, race, gender, insurance, stage, performance status, and comorbidities. The ICER for trial participation after adjustment for confounders was $9741 per life year saved (95% CI, $3089 $19,149). CONCLUSIONS. Enrollment in lung cancer clinical trials was found to be associated with improved survival at a moderate incremental cost. Cancer 2003;98: American Cancer Society. Supported by Grant TURPG PBP from the American Cancer Society. The authors thank Mr. Daryn Smith for the contribution he has made to this project. This article could not have been completed without his valuable statistical and database programming support. Address for reprints: Wei Du, Ph.D., Center for Cancer Economics, Technology Assessment, Innovation and Development (CETAID), Karmanos Cancer Institute, 110 E. Warren, Detroit, MI 48201; Fax: (313) ; duw@karmanos.org Received March 12, 2003; revision received June 30, 2003; accepted July 6, American Cancer Society DOI /cncr KEYWORDS: nonsmall cell lung cancer (NSCLC), clinical trials, cost-effectiveness, incremental cost-effectiveness ratio (ICER). Lung cancer is the leading cause of cancer death in the U.S. for both men and women and it is estimated to have an annual economic burden of approximately $5 billion. 1,2 Five-year survival rates of only 14% highlight the limitations of current therapies for lung cancer as well as the need for significant improvement in treatment of the disease. Most major advances in the treatment of multiple cancer types have come from well conducted clinical trials. Participation in a high-quality clinical trial is the best treatment option for patients with cancer. 3 7 However, despite the large number of available studies, only 3 5% of new cancer patients are reported to participate in clinical trials. 8,9 Low enrollment lengthens a trial s duration and delays the delivery of potentially beneficial treatments to cancer patients. Many barriers contribute to low clinical trial enrollment, and one of the major barriers is cost. 10 According to a survey of 6000 cancer patients conducted by Harris Interactive, concern regarding insurance coverage is one of the most common reasons why cancer patients refused to participate in a clinical trial. 9 Many insurers are reluctant to pay for medical care associated with clinical trials due to the experimental or investigational nature of clinical trial treatments. In a previous study,

2 1492 CANCER October 1, 2003 / Volume 98 / Number 7 Peters and Rogers 11 demonstrated that predetermination decisions by insurers for patients entering clinical trials are arbitrary and capricious and are barriers to clinical trial accrual. To our knowledge, until recently, empirical data regarding the costs of clinical trials had been lacking. The Mayo Clinic estimated that during the years 1988 through 1994, the costs for 61 cancer patients treated in clinical trials were 3 13% greater than those for a matched control sample. 12 At 1-year follow-up, the average cost for trial enrollees was $24,645 compared with $23,964 for control patients. The analysis used direct medical costs that include hospital, physician, and ancillary services. Bennett et al. 13 analyzed 6-month direct medical costs of 70 cancer patients with varying diagnoses. They found that the mean cost for clinical trials patients was $57,542 whereas the mean cost for control patients was $63,721. In another study, Bennett et al. 14 summarized the treatment costs of 377 patients from 5 studies. They concluded that clinical trials result in a 10 14% increase in costs over standard treatment at 12-month follow-up. Fireman et al. 15 evaluated the direct medical costs of 270 cancer patients at a large health maintenance organization (HMO) and found that cancer clinical trials did not result in substantial increases in the direct costs of medical care. The mean 1-year cost for the enrollees was 10% higher than that for the control subjects. These studies indicate that clinical trials participants in general incur a modest increase in cancer treatment costs over existing standard treatments However, to our knowledge, there is no mention of whether the increase in costs is accompanied by an improvement in treatment outcomes such as survival or health-related quality of life. Currently, little is known regarding the cost-effectiveness of clinical trials treatments and the estimates of the increased cost from many studies are hard to interpret because patients with varying diagnoses were included. In the current study, we evaluated the effect of participation in lung cancer clinical trials on direct medical costs as well as on treatment outcome (i.e., patient survival) in a group of 336 patients with nonsmall cell lung cancer (NSCLC) who were eligible for clinical trials (260 control patients and 76 trials enrollees). The incremental cost-effectiveness ratio (ICER) of clinical trials treatment and its 95% confidence interval (95% CI) were calculated after adjusting for age, race, gender, stage, performance status, comorbidities, and insurance coverage. MATERIALS AND METHODS Study Subjects The clinicodemographic characteristics of 510 patients with NSCLC who received full initial treatment at the Karmanos Cancer Institute (Detroit, MI) between January 1994 and December 1998 were evaluated. Of these patients, 76 were treated on a clinical trial and 434 were not. In the current study, instead of using a one-to-one matching design, which would limit significantly the number of controls we could include, we established comparable study groups by removing control subjects who were unlikely to have been eligible for trials. Then, we used multivariable analyses to further adjust for any important confounders. To establish comparable study groups, 174 nonenrollees were excluded. These were patients who had TNM Stage I disease, more than 2 comorbidities, liver failure, renal failure, performance status greater than or equal to 2, age at diagnosis older than 80 years, who received no cancer treatment such as chemotherapy, radiation, or surgery, or who died within 2 weeks of diagnosis. After these exclusions, a total of 336 patients (76 enrollees and 260 controls) were available for the analysis of the cost-effectiveness of clinical trials participation. NSCLC Clinical Trials Twenty-five therapeutic trials were available during the period of analysis. Eight were Radiation Therapy Oncology Group (RTOG) trials, nine were Southwest Oncology Group (SWOG) trials, five were local investigator-initiated trials, and three were industry-sponsored trials. There were 9 Phase III, 12 Phase II, 1 Phase II/III, and 3 Phase I/II trials. All trials included patients with Stage II or higher NSCLC. Study End Points One-year cumulative treatment cost and survival duration of patients with NSCLC were the primary end points of the current study. Any costs associated with inpatient and outpatient visits (including chemotherapy drugs) between the date of diagnosis and 1-year follow-up or death (whichever came first) were included in the calculation of 1-year treatment cost, but hospice care costs were not included. The 1-year time frame was selected to allow comparison of standard treatment to experimental treatment options during the initial treatment period. All cost data were inflation adjusted to Year 2000 dollars using the consumer price index. For the current analysis, we assume the health care providers (hospital systems) perspective. Thus, only fully allocated facility costs from a cost accounting database, TSI (Transition Systems, Inc., Detroit Medical Center), were used to calculate treatment costs. The professional costs component, which generally is stable at 10 15% of facility-based costs, is not included in the current analysis. Clinical characteristics such as stage, performance status, comorbidities, and survival duration were ob-

3 Cost-Effectiveness and Lung Cancer/Du et al tained from chart review and from the Surveillance, Epidemiology and End Results registry (SEER). Performance status was recorded using the SWOG scale (0 4). Comorbidities include heart disease, diabetes, chronic obstructive pulmonary disease, renal failure, and liver failure. Patient demographics include age, race, gender, and insurance coverage. Insurance was classified as Medicare only, Medicaid, commercial insurance (conventional, HMO, or preferred provider organization), or Medicare plus supplemental insurance (Medigap or Medicaid). Statistical Methods Chi-square and Student t tests were used to compare the clinicodemographic characteristics of clinical trials enrollees and nonenrollees. Cox proportional hazards regression and linear regression with log-transformed dependent variables were used to assess the effect of various patients characteristics on survival and 1-year treatment costs. Variables included in the multivariable analyses are either factors that, generally, are considered important prognostic factors (e.g., age, stage, or performance status) or those that are significantly different between the two comparison groups. Treatment options (e.g., surgery, chemotherapy, and radiation) were not included because they are typically highly confounded with clinical trials enrollment. Before the estimation of the ICER, log-normal survival regression was used to impute the survival times for 35 patients whose survival times were censored. Then, two methods of calculating ICER were employed and compared. The most commonly used method is to simply take the difference in mean cost of the two patient groups (C 1 C 2 ) and divide it by the difference in mean survival (S 1 S 2 ). This method is valid only if the two comparison groups are comparable with respect to major confounders. In the current study, we propose calculating the ICER by estimating the effect of enrollment on cost and survival separately, using linear regression models with adjustment for confounders. In addition, because the distributions of cost and survival data typically are not normal, log transformation on the cost and survival data is performed before fitting the regression models. If we ignore the normality issue and calculate the adjusted ICER using the untransformed data, we need only to take the regression coefficient of cost on enrollment ( c ) and divide it by the coefficient of survival on enrollment ( s ), which yields ICER c / s. However, because log transformation of the data is necessary in our case, it is slightly more complicated to obtain the adjusted ICER. First, the adjusted mean of each of the two comparison groups and their common variance were computed separately for the transformed cost (m c1, m c2, and v c ) and survival (m s1, m s2, and v s ) using regression analyses. Next, by using the fact that the expectation of exp(y) exp (M [V/2]), when y is distributed normally with mean M and variance V, the four adjusted means (m c1, m c2, m s1, and m s2 ) were transformed back to their original units (M c1, M c2, M s1, and M s2 ). The estimated adjusted ICER becomes (M c1 M c2 )/(M s1 M s2 ). The bootstrap 17 method was used to calculate 95% CI bounds for the two estimated ICERs. Twenty thousand bootstrap samples of size n 336 were generated with replacement from the original sample (76 enrollees and 260 nonenrollees). Then, the ICER for each of these 20,000 samples was calculated using both methods described earlier. Finally, 95% CI bounds were calculated using the 2.5th and 97.5th percentiles of the 20,000 ICER bootstrap estimates. RESULTS Characteristics of all Patients with NSCLC Of the 510 patients with NSCLC who received full initial treatment between 1994 and 1998, 76 (15%) were treated on a clinical trial and 434 were not. Clinical trials enrollees differed significantly from nonenrollees in the distribution of race, gender, age, insurance coverage, stage, and performance status. Trial enrollees were more likely to be white, male, younger in age, to have commercial insurance, and good performance status (all had either a 0 or 1).In addition, they were more likely to have higher stage disease when compared with nonenrollees. Comparisons of Clinical Trials Enrollees Versus Comparable Nonenrollees To establish comparable study groups, 174 of 434 control patients were excluded from the analyses of cost and outcome because they were unlikely to have been eligible for any trial, yielding 76 enrollees and 260 nonenrollees for the following analyses. Of the remaining 336 patients with NSCLC, 76 (22.6%) were treated on a therapeutic clinical trial protocol. Differences between enrollees and nonenrollees were narrowed significantly after the exclusion of these 174 patients (Table 1). The only factors that remained significantly different were race, gender, age, and insurance coverage. The data show that white patients were more likely to participate than African-American patients (P 0.01), men were more likely to participate than women (P 0.02), patients with commercial insurance coverage were more likely to participate than those with noncommercial insurance coverage (P 0.02), and patients younger than 70 years of age were more likely to participate than patients 70 years or older (P 0.01). None of the pretreatment clinical

4 1494 CANCER October 1, 2003 / Volume 98 / Number 7 TABLE 1 Characteristics of Enrollees versus Comparable Nonenrollees Characteristics (%) All patients (n 336) (%) Enrollees (n 76) (%) Nonenrollees (n 260) (%) P value (two-sided) African-American Male Age (mean/median) (yrs) 61/64 60/61 62/ Age 70 yrs Insurance 0.07 Commercial (0.02) a Medicaid Medicare Medicare/supplemental Stage 0.40 II IIIA IIIB IV SWOG performance status Diabetes Heart disease COPD No. of comorbidities COPD: Chronic obstructive pulmonary disease. SWOG: Southwest Oncology Group. a The P value was obtained by dichotomizing insurance into commercial or noncommercial. TABLE 2 Multivariable Analysis of 1-Year Treatment Cost and Survival Survival Log (1-yr cost) Characteristics (%) Hazard ratio P value (2-sided) Regression coefficient a P value (2-sided) African-American vs. others a 0.26 Male vs. female Age (yrs) Commercial insurance vs. others Stage IIIA vs. II IIIB vs. II IV vs. II SWOG performance status 0 vs Any comorbidities (yes/no) Trial participation (yes/no) SWOG: Southwest Oncology Group. a Because of the log transformation of Y, i.e., ln (Y) ( 0 1 X 1 2 X 2... k X k ), the model in its original unit is a multiplicative one: Y exp ( 0 1 X 1 2 X 2... k X k ). The coefficients in the table are regression coefficients transformed back to the original unit. characteristics were significantly different between the two comparison groups. Treatment Costs As expected, trial enrollment was associated highly with the type of treatment received by the patient. The data show that all clinical trials enrollees, compared with only 62% of nonenrollees, received chemotherapy (P 0.01), whereas no differences were found in the receipt of radiation or surgical treatment. The average 1-year cost for patients treated on protocols was $41,734 compared with $34,191 for patients treated with standard therapy. In a linear regression analysis of the log-transformed cost data (Table 2), stage, age, and clinical trials enrollment were significant predictors of 1-year treatment cost, whereas

5 Cost-Effectiveness and Lung Cancer/Du et al TABLE 3 The Incremental Cost-Effectiveness Ratio Characteristics Difference in mean survival (yrs) Difference in mean cost ICER Lower bound ICER 95% CI Upper bound Unadjusted $7543 $13,695 $3887 $43,052 Adjusted $8215 $9741 $3089 $19,149 ICER: incremental cost-effectiveness ratio; CI: 95% confidence interval. race, gender, socioeconomic status (SES), performance status, insurance coverage, and comorbidities were not significant predictors. Younger age was associated positively with higher costs (P 0.01). On average, clinical trials enrollment was associated with a 24% increase in cost (P 0.01). When Stage I patients were excluded from the analyses, Stage II and IV patients accrued the least costs whereas Stage IIIA patients accrued the most costs. On average, Stage IIIA patients accrued 61% higher costs than Stage II patients (P 0.01), whereas Stage IIIB patients accrued only 25% higher costs than Stage II patients (P 0.06). The latter difference is of borderline statistical significance. Patient Survival The median survival period of clinical trials enrollees was 1.3 years compared with 0.9 years for patients treated with standard therapy. This difference was statistically significant (P 0.02). In a Cox proportional hazards regression (Table 2), clinical trial enrollment, stage, age, and performance status were significant predictors of survival, whereas race, gender, SES, comorbidities, and insurance type were not. Clinical trials enrollees had a 30% lower hazard of death than nonenrollees (P 0.01). Older age was associated positively with an increased risk of death (P 0.05). As expected, disease stage was the most significant predictor of survival (P 0.01), and patients with a performance status of 0 had a 36% lower risk of death than patients with a performance status of 1 (P 0.01). Estimation of Incremental Cost-Effectiveness Ratio Before the calculation of the ICER of clinical trials, the overall survival of 35 (10.4%) patients with a censored survival time was imputed using a log-normal survival regression model. There was no statistically significant difference in the proportion of subjects with censored survival time between enrollees and nonenrollees (13% vs. 10%, P 0.37). Linear regression analyses of log-transformed cost and survival time were performed using race, gender, stage, age, performance status, comorbidities, and insurance coverage as covariates. Table 3 shows the calculated ICERs and the 95% CIs for the unadjusted and adjusted methods. The ICER estimate without adjustment for covariates is $13,695 (95% CI, $3887 $43,052). However, once adjustments are made, the ICER under the log-transformed regression model is $9741 (95% CI, $3089 $19,149). DISCUSSION A major principle of evidence-based medicine is that formal evaluation of treatment approaches in clinical trials improves the general quality of medical care. Cancer clinical trials translate basic scientific research results into better ways to prevent, diagnose, and treat cancer. Previous studies indicated that, on average, the participation of cancer patients in clinical trials incurs a modest but real increase in cancer treatment costs over existing standard treatments. However, these studies did not attempt to correlate increased treatment costs of clinical trial participation with patient outcomes. The current study provides the first quantitative information demonstrating that participation in a lung cancer clinical trial is associated with improved survival and cost-effectiveness of care. Clinical trials offer cancer patients access to novel and innovative therapeutic options that are associated frequently with better treatment outcomes. 3 7 This is particularly relevant in patients with lung cancer because current therapies provide only modest benefits. Similar to several previous studies, we found that clinical trial enrollment was associated positively with a 24% higher 1-year cost (P 0.01), as were Stage III disease and younger age. The data from the current study also demonstrate that clinical trial participation, as well as early-stage disease, younger age, and good performance status, was associated significantly with longer survival. Clinical trials enrollees had a 30% lower hazard ratio of death than nonenrollees (P 0.01). Combining cost and survival data, we found that lung cancer clinical trials were associated with improved survival at a moderate incremental cost of $13,695 per life year saved (95% CI, $3887 $43,052)

6 1496 CANCER October 1, 2003 / Volume 98 / Number 7 without adjustment for confounders and at a cost of $9741 per life year saved (95% CI, $3080 $19,149) with adjustment for confounders. It is noteworthy that the adjusted ICER not only is $3954 lower than the unadjusted ICER, but it is also estimated much more precisely (narrower 95% CI). The 95% CI for the adjusted ICER is much narrower because variations between patients were removed by the multivariable adjustment. One limitation of the current study was that professional costs were not included in the analysis. One might thus argue that our calculation of the ICER is an underestimation of the true financial effects of lung cancer trials. It is our experience that the professional cost component is generally 10 15% of the hospitalbased costs. Thus, if one adds 15% to the estimated ICER as the increase in professional costs due to clinical trial enrollment, the total incremental direct medical costs during Year 1 would only be $11,202, which is still rather favorable when compared with the standard acceptable costs of $50,000 per life year saved. Another important point is that, unlike many previous studies that evaluated cost-effectiveness of care, we chose not to use a one-to-one matched pair design because such a design significantly limits the number of controls investigators could include and thus reduces the statistical power of analysis. Furthermore, the comparability of comparison groups using a oneto-one matching design is restricted to only a very small number of preselected factors used for matching. In the current study, all potentially eligible controls were included in the analysis and important confounders were included in the multivariable adjustment to ensure comparability. If a one-to-one matched pair design had been used in the current study, we would have had at most 152 (i.e., 76 2) patients available for analysis. The design we used allowed us to use a much larger sample size (n 336), thus resulting in a much more precise estimate for the ICER. The positive survival outcome in lung cancer trials may be due partly to the increase in new and effective medications being developed and tested. The increase in costs associated with clinical trials may, in part, reflect increased costs associated with these new agents, but these effects appear to be offset by the improved survival. In the current policy debates surrounding the escalating costs of medical care, it is worthwhile to emphasize that medication costs represent only 10% of total health care costs, whereas the improvement in outcome appears to be due mostly to the new therapeutics. In the current study, enrollment in lung cancer clinical trials was found to be associated with improved survival at a moderate incremental cost, regardless of which statistical method (with or without adjustment for confounders) one used to estimate ICER. Continued improvement in the outcome of patients with lung cancer would appear to be advanced by continued and enhanced emphasis on participation in clinical trials. This demonstration of cost-effectiveness in lung cancer trials should be evaluated in other disease settings. REFERENCES 1. Leighl NB, Shepherd FA, Kwong R, Burkes RL, Feld R, Goodwin PJ. Economic analysis of the TAX 317 trial: docetaxel versus best supportive care as second-line therapy of advanced non-small-cell lung cancer. J Clin Oncol. 2002;20: Ramsey SD, Moinpour CM, Lovato LC, et al. Economic analysis of vinorelbine plus cisplatin versus paclitaxel plus carboplatin for advanced non-small-cell lung cancer. J Natl Cancer Inst. 2002;94: Stiller C. Centralized treatment, entry to trials, and survival. Br J Cancer. 1994;70: Friedman MA, McCabe MS. Assigning care costs associated with therapeutic oncology research: a modest proposal. J Natl Cancer Inst. 1992;84: Antman K, Schnipper LE, Frei E III. The crisis in clinical cancer research: third-party insurance and investigational therapy. N Engl J Med. 1988;319: Farrar WB. Clinical trials: access and reimbursement. Cancer. 1991;67(6 Suppl): Cova JL. A swift response to a modest proposal. J Natl Cancer Inst. 1992;84: Lara PN, Higdon R, Lim N, et al. Prospective evaluation of cancer clinical trial accrual patterns: identifying potential barriers to enrollment. J Clin Oncol. 2001;19: Harris Interactive. Misconceptions and lack of awareness greatly reduce recruitment for cancer clinical trials [monograph online]. Available from URL: harrisinteractive.com/news/allnewsbydate.asp?newsid 222 [accessed 22 January 2001]. 10. Gotay CC. Accrual to cancer clinical trials: directions from the research literature. Soc Sci Med. 1991;33: Peters WP, Rogers MC. Variation in approval by insurance companies of coverage for autologous bone marrow transplantation for breast cancer. N Engl J Med. 1994;330: Wagner JL, Alberts SR, Sloan JA, et al. Incremental costs of enrolling patients in clinical trials: a population-based study. J Natl Cancer Inst. 1999;91: Bennett CL, Adams JR, Knox KS, Kelahan AM, Silver SM, Bailes JS. Clinical trials: are they a good buy? J Clin Oncol. 2001;19: Bennett CL, Stinson TJ, Vogel V, et al. Evaluating the financial impact of clinical trials in oncology: results from a pilot study from the Association of American Cancer Institutes/ Northwestern University Clinical Trials Costs and Charges Project. J Clin Oncol. 2000;18: Fireman BH, Fehrenbacher L, Gruskin EP, Ray T. Cost of care for patients in cancer clinical trials. J Natl Cancer Inst. 2000;92: Chirikos TN, Ruckdeschel JC, Krischer JP. Impact of clinical trials on the cost of cancer care. Med Care. 2001;39: Efron B, Tibshirani R. Bootstrap methods for standard errors, confidence intervals, and other measures of statistical accuracy. Stat Sci. 1986;1:54 7.