Modeling the Cost of Management Options for Stage I Nonseminomatous Germ Cell Tumors: A Decision Tree Analysis
http://www.100md.com
《临床肿瘤学》
the James Buchanan Brady Urological Institute and the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins, Johns Hopkins Medical Institutions, Baltimore, MD
ABSTRACT
PURPOSE: Patients with clinical stage I nonseminomatous germ cell tumors (NSGCTs) have been managed with surveillance, chemotherapy, or retroperitoneal lymphadenectomy (RPLND) with similar survival outcomes. Cost factors influencing the choice of therapy were evaluated using computer-based decision analysis.
METHODS: A detailed model was developed that integrates projected costs for more than 60 possible treatment outcomes. It incorporates primary, adjuvant, and salvage chemotherapy, primary and postchemotherapy RPLND, and both laparoscopic and open surgical approaches. Starting values and probabilities were derived from a comprehensive meta-analysis of the last 25 years of testes cancer literature. Hypothesis testing was performed using sensitivity analysis.
RESULTS: The model predicts a cost premium for both primary chemotherapy (18.7%) and RPLND (51.7%) compared with surveillance. If laparoscopic RPLND was practiced, the cost premium for primary surgery (29.1%) approached that of chemotherapy (26.4%). Open RPLND was 1.25x as costly as laparoscopic RPLND, primarily because of longer hospitalization. The choice of open RPLND yielded a 6.9% cost premium for a surveillance program in this model. For such a program, primary chemotherapy became cost advantageous when the probability of recurrence during surveillance was more than 46%.
CONCLUSION: This model allows a variety of treatment cost hypotheses to be tested. Primary RPLND is never cost advantageous over surveillance or primary chemotherapy. Surgical costs can significantly increase the overall cost of a surveillance program. In stage I patients with high-risk tumor characteristics, primary chemotherapy may have a cost advantage over surveillance.
INTRODUCTION
Despite several decades of clinical investigation, the management of patients with stage I nonseminomatous germ cell tumors (NSGCT) after radical orchiectomy remains controversial. Primary surveillance, retroperitoneal lymphadenectomy (RPLND), and two chemotherapy cycles (bleomycin, etoposide and cisplatin [BEP]) all have similarly excellent survival outcomes and low morbidity. Cost concerns, therefore, may play an important role in determining the choice of management in this disease.
In 1996, Baniel et al1 published an elegant cost analysis comparing surveillance and primary RPLND at Indiana University. Two years later, Lashley and Lowe2 reported a similar study comparing surveillance, chemotherapy, and RPLND. In both cases, the authors collected information on medical charges from 100 patients in each treatment group. They concluded that the cost difference was relatively small (< 13.5%) and should not alter clinical decision making. Although these studies significantly advanced our understanding of the socioeconomic impact of stage I NSGCT, they suffer from three limitations. No attempt was made to derive a mathematical cost model from these results that could be used to test hypotheses with sensitivity analysis. Samples might include, "How quickly must I discharge my RPLND patients to achieve cost equivalence with chemotherapy?" "At what surveillance failure rate does chemotherapy become cost advantageous over surveillance?" The answers to these questions cannot easily be determined from the data presented. These results also reflect an experience at two high-volume testes cancer programs and may not easily be generalized to lower-volume institutions. Finally, laparoscopic RPLND, popularized during the current decade, was not factored into these analyses from the 1990s.
Here we develop a detailed mathematical decision analysis model comparing the projected management costs for stage I NSGCT. It is built on a meta-analysis of the testes cancer literature and reflects a multi-institutional clinical experience. The model incorporates a variety of clinical outcomes downstream from the original therapy decision, given that these events would potentially influence costs. In addition, laparoscopic RPLND and updated reimbursement are factored into the analysis.
We focused on four critical questions as a measure of model utility: How do the various options for managing stage I NSGCT compare in cost, and are the differences significant enough to drive clinical decision making? How do laparoscopic and open RPLND compare in perioperative costs, and can we identify targets for operative time and length of stay (LOS) that result in cost equivalence? How much impact does surgery have on the cost of a surveillance program, and does the choice of the laparoscopic or open approach significantly alter that cost? Does the most cost-effective management option change when patients with low- and high-risk tumor characteristics are compared?
METHODS
A mathematical decision-tree model was developed to estimate the cost of surveillance, primary chemotherapy, and RPLND (Table 1). Because of space constraints, some details of model development have been included as an Appendix. Downstream of the original treatment decision, the model incorporates more than 60 potential clinical outcomes (Fig A1). It includes terms for RPLND, postchemotherapy RPLND, and BEP and salvage chemotherapy (ifosfamide, vinblastine, and cisplatin). It does not incorporate high-dose chemotherapy salvage treatment with stem-cell support. Perioperative surgical costs were expressed in a submodel incorporating costs centers for operative time, surgical consumables, surgeon and anesthesiologist professional fees, hospital room charges, and blood transfusions. Both laparoscopic (RPLNDlaparoscopic) and open (RPLNDopen) surgical therapy were independently modeled. For simplification, if laparoscopic surgery was used for RPLND, it was also used for postchemotherapy RPLND in a given option. The model incorporates 51 variables (Table A7) including constants and terms for the probabilities of various clinical events. Starting values are derived from a 25-year Medline search of the testes cancer literature as documented in Tables A2 to A6. In some cases, information was not available in the literature and had to be estimated. This was particularly true for postchemotherapy RPLNDopen, for which published information on operative times and transfusion rates were lacking. For postchemotherapy RPLND, we added 90 minutes to the operative time of standard RPLND, and transfusion rates were set as identical to standard RPLND.
For surveillance, the follow-up protocol was modeled as described in the 2004 National Comprehensive Cancer Network Clinical Practice Guidelines in Oncology3 for a duration of 7 years. Costs for surveillance failures were determined by scaling costs based on the average time to failure in the literature. Pharmaceutical costs represent the lowest wholesale price for a particular drug published in the 2003 Drug Topics Red Book scaled to 2004 dollars using a web-based Consumer Price Index inflation calculator.4 Professional fees as well as imaging and laboratory costs were derived from 2004 Medicare reimbursement rates. Operative consumables were calculated from a list of items included on the surgical field and reflect purchase price. Hospital room and board and operative time charges were acquired from the administration at the Johns Hopkins Hospital (Baltimore, MD).
The cost model was programmed into TreeAge Pro Healthcare software (TreeAge Software Inc, Williamstown, MA), which was used to perform sensitivity analyses and probability calculations. In one-way analysis, a single cost factor is altered and its impact on total cost is determined by keeping all other factors fixed. In two-way analysis, two factors are altered simultaneously.
RESULTS
How Do the Various Options for Managing Stage I NSGCT Compare in Cost, and Are the Differences Significant Enough to Drive Clinical Decision Making?
Figure 1 shows a summary of baseline model predictions. Primary surveillance was the least costly option followed by chemotherapy (18.7% to 26.4% premium) and primary RPLND (29.1% to 51.7% premium), depending on the choice of surgical approach. Options including laparoscopic surgery were less costly than those incorporating open surgery. Primary RPLNDopen was 1.6 times more costly than the least-expensive option, surveillance withlaparoscopic therapy for retroperitoneal recurrence. Using two-way sensitivity analysis, operative time and LOS for RPLNDopen and RPLNDlaparoscopic were varied to evaluate how these factors affected the choice of the least costly therapy option (Fig 2). For the majority of operative times and LOS, surveillance shows a clear cost advantage.
How Do Laparoscopic and Open RPLND Compare in Perioperative Costs and Can We Identify Targets for Operative Time and LOS That Result in Cost Equivalence?
A submodel was developed that compared only the perioperative costs of RPLNDopen and RPLNDlaparoscopic (Table 2). This model predicts that RPLNDopen will be 1.25x as costly as RPLNDlaparoscopic, primarily based on longer inpatient hospitalization after RPLNDopen. The costs of hospitalization and operative time were the two most significant factors differentiating the two procedures. Transfusion rate was a negligible factor, contributing less than 0.5% to the total perioperative cost.
To identify operative time and LOS targets resulting in cost equivalence between RPLNDopen and RPLNDlaparoscopic, we performed a one- and two-way sensitivity analysis (Fig 3). This analysis demonstrated that cost equivalence could be met with a 5-day hospital stay after RPLNDopen, keeping other factors constant. Cost equivalence could also be reached if the room rate for inpatient hospitalization was less than $305 per day. In contrast, even unrealistic operative times (< 30 minutes) could not overcome the cost impact of a 7-day hospitalization for RPLNDopen.
How Much Impact Does Surgery Have on the Cost of a Surveillance Program, and Does the Choice of the Laparoscopic or Open Approach Significantly Alter That Cost?
The model not only allows projection of costs but also the probability that an individual patient will experience a particular outcome. Figure 4 shows a breakdown of predicted outcomes for patients entering surveillance: 99.6% of patients entering the program will ultimately be cured by some combination of observation with or without therapy for recurrence.
For a surveillance program, the probability of a patient undergoing an RPLND for disease recurrence is 0.195. The probability of a patient undergoing a postchemotherapy RPLND is much lower (0.017). Therefore, the maximal impact of surgical therapy costs on a surveillance program would be approximately 21%.
In practice, a comparison of laparoscopic and open surgical therapy is significantly more complex because the outcome probabilities are derived from a meta-analysis of the literature and differ slightly for the two approaches. For the purposes of this analysis, we ignored the impact of postchemotherapy surgery because it can only contribute 1.7% to the model. The submodel of perioperative RPLND cost predicts a difference of $3,244.86 between RPLNDlaparoscopic and RPLNDopen. If we multiply this by 0.195, it yields a predicted cost premium for RPLNDopen of 5.2%. This percentage represents the cost premium if all outcomes for the laparoscopic and open approaches were modeled identically. Given that the complete model also factors in postchemotherapy surgery and the slight variations in outcomes published for the two approaches, it predicts a cost premium of 6.9% for a surveillance program using open surgery compared with laparoscopy.
Does the Most Cost-Effective Management Option Change When Patients With Low- and High-Risk Tumor Characteristics Are Compared?
As shown in Fig 1, there is a clear financial advantage to surveillance over chemotherapy and RPLND. To evaluate the impact of high- and low-risk tumor characteristics on this cost relationship, we performed one-way sensitivity analyses varying the probability of disease recurrence during primary surveillance (Fig 5). If RPLNDlaparoscopic were practiced for retroperitoneal recurrence, surveillance would remain the most cost-effective option compared with primary chemotherapy or primary RPLNDlaparoscopic until the probability of retroperitoneal recurrence increased above 60%. Likewise, primary RPLNDopen required a treatment failure probability of 80%. In contrast, if open surgery was practiced, primary chemotherapy became cost advantageous over surveillance once the probability of retroperitoneal recurrence increased to only 46%.
DISCUSSION
When facing stage I NSGCTs, patients and urologic oncologists must decide among surveillance, chemotherapy, or RPLND. This decision hinges on the critical issues of oncologic effectiveness, morbidity, and cost. All three options have similarly high survival outcomes (97% to 100%).5 Likewise, the morbidity of nerve-sparing modified-template RPLND or short-course chemotherapy is modest.6-15 Cost may, therefore, play an important role in directing clinical decisions in this disease. A detailed decision-tree model was developed for projecting the management costs for stage I NSGCTs. The model’s advantage lies in the ease with which various hypotheses can be rapidly tested through the use of sensitivity analysis.
There was a clear cost advantage for primary surveillance over chemotherapy or RPLND. The magnitude of this advantage (18.7% to 51.7%) was larger than that previously reported during the 1990s (< 13.5%).1,2 However, our methodology differs significantly from these earlier studies, which determined cost by reviewing hospital charges for 100 patients in each of the treatment arms. The current model is based on a mixture of hospital costs (ie, consumables, pharmaceuticals), Medicare reimbursement rates (ie, professional fees, imaging), and hospital charges (ie, room and board, operative time). For example, Lashley and Lowe2 report an abdominal computed tomography scan costs $1,152, whereas our model uses the current Medicare reimbursement rate of $419 (2.7x less). Other differences include variations in the surveillance protocol and the omission of salvage chemotherapy from earlier studies. Finally, outcome probabilities are derived from a meta-analysis and differ slightly from those estimated in the older reports.
The current model predicts that perioperative cost of RPLNDopen will be 1.25x that of RPLNDlaparoscopic. This conclusion differs significantly from the results of a similar perioperative modeling study by Ogan et al,16 which predicted that RPLNDlaparoscopic would be 1.1x as expensive as RPLNDopen. For the current model, starting values were derived from a more comprehensive meta-analysis of RPLND reports (n = 20) than used by Ogan et al (n = 5).16 In addition, calculations of operative time and hospital room charges are radically different between the Johns Hopkins and University of Texas Southwestern institutional models. For example, the daily room rate of $395 used in the study by Ogan et al16 approaches the target room rate for cost equivalence between laparoscopic and open surgery in the Johns Hopkins model ($305). It is also important to stress that the current study focuses specifically on cost and does not represent a comparison of outcomes between RPLNDopen and RPLNDlaparoscopic.
Another conclusion relates to the management of stage I NSGCT patients with high-risk tumor characteristics such as vascular or lymphatic invasion, presence of embryonal cell carcinoma, and absence of yoke sac components. In this high-risk group, the rate of retroperitoneal recurrence during surveillance has been estimated at approximately 50%.17 Our model predicts that primary chemotherapy may be cost advantageous over a surveillance program using RPLNDopen with this failure rate.
It is critical to appreciate the limitations as well as the strengths of a modeling exercise of this type. Any model depends on a set of critical assumptions, some of which can be supported by the literature, whereas others are simply estimates or reflect local clinical practice. Cost modeling is also inherently dependent on institution and region, and these results may not be valid for all institutions. Yet the structure of the model reflects general management principles for stage I NSGCT, and should be easily translated to other centers provided that institution-specific cost components are modified.
One potential criticism relates to our LOS estimate after open RPLND. This is important because hospital room cost was the most significant factor differentiating the two approaches. Unfortunately, most large series of RPLNDopen patients do not report LOS; hence, the model starting value of 7.1 days is derived from only four of the 12 evaluated studies. The RPLNDopen series also tend to be historically older or from international institutions, hence hospitalization may not reflect 2004 United States practice. Sensitivity analysis identified an LOS target of 5 days for cost equivalence between RPLNDlaparoscopic and RPLNDopen, which may be a more realistic estimate of current LOS. In that case, results reported here for management options using RPLNDlaparoscopic would also reflect options including RPLNDopen.
Consistent with previously published studies,2,16 the current model does not include treatment-related morbidity or patient compliance in the cost analysis. Although morbidity influences patient and physician decision making, we feel that its impact on the aggregate cost of each strategy will be small. The morbidity of nerve-sparing modified-template RPLND is modest, relating primarily to a 1% to 3% rate of bowel obstruction6,7 and a 1% to 7% rate of anejaculation.7-10 The morbidity of two BEP cycles also appears to be minimal. For 58 patients, Bohlen et al11 reported only transient neutropenia and thrombocytopenia; granulocyte stimulation support was required in only one patient. Numerous studies have confirmed the low acute morbidity of two cycles of BEP and documented no significant long-term pulmonary or cardiac toxicity from this regimen.12-15 The long-term risk of secondary leukemia after BEP also appears to be low (< 0.7%) for patients who receive less than 2 g/m2 of etoposide.18 The total etoposide dose given for two BEP cycles is less than 0.8 g/m2.12 For patients receiving more than two adjuvant BEP cycles, the risk of organ toxicity and secondary malignancies may be higher. However, the model predicts that no more than 10% of patients on any treatment regimen will receive more than two cycles. Even if morbidity doubled the cost of chemotherapy for this patient subset, our overall conclusions would not change.
It is important to differentiate the cost analysis presented here from a formal cost-effectiveness analysis. Given the small relative differences in mortality among surveillance, RPLND, and chemotherapy for stage I NSGCT, use of mortality as a measure of effectiveness is difficult. An alternative strategy would be to assign estimated quality-of-life utility values to all the outcome states within the model. As noted previously, the morbidity differences between chemotherapy and RPLND are also likely to be small. Estimates based on such small and controversial differences in morbidity would add an unacceptable degree of systematic error to results, and would cloud their interpretation. As more comprehensive morbidity data become available, however, extension of this analysis to include effectiveness would likely be valuable.
The cost model presented here allows a variety of economic hypotheses about the management of stage I NSGCTs to be tested readily. It emphasizes the cost advantage of surveillance, quantifies the difference between laparoscopic and open surgical options, and suggests that, under some circumstances, primary chemotherapy may be cost advantageous for patients with high-risk tumor characteristics.
Authors’ Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
Appendix
Click here for Fig A1http://www.jco.org/cgi/content/full/23/24/5762/DC1
NOTES
Presented in part at the 2004 Society for Urologic Oncology Meeting, Bethesda, MD, December 3-4, 2004.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Baniel J, Roth BJ, Foster RS, et al: Cost- and risk-benefit considerations in the management of clinical stage I nonseminomatous testicular tumors. Ann Surg Oncol 3:86-93, 1996
Lashley DB, Lowe BA: A rational approach to managing stage I nonseminomatous germ cell cancer. Urol Clin North Am 25:405-423, 1998
Clinical Practice Guidelines in Oncology: Testicular Cancer. Version 1.2004. http://www.nccn.org/professionals/physician_gls/PDF/testicular.pdf
Consumer Price Index Calculator: Inflation Calculator. http://data.bls.gov/cgi-bin/cpicalc.pl
Porter J, Obek C, Lange PH: The management of stage I nonseminomatous germ cell tumor. Campbells Urol Updates 2:1-12, 2004
Baniel J, Foster RS, Rowland RG, et al: Complications of primary retroperitoneal lymph node dissection. J Urol 152:424-427, 1994
Heidenreich A, Albers P, Hartmann M, et al: Complications of primary nerve sparing retroperitoneal lymph node dissection for clinical stage I nonseminomatous germ cell tumors of the testis: Experience of the German Testicular Cancer Study Group. J Urol 169:1710-1714, 2003
Donohue JP, Thornhill JA, Foster RS, et al: The role of retroperitoneal lymphadenectomy in clinical stage B testis cancer: The Indiana University experience (1965 to 1989). J Urol 153:85-89, 1995
Donohue JP, Foster RS, Rowland RG, et al: Nerve-sparing retroperitoneal lymphadenectomy with preservation of ejaculation. J Urol 144:287-292, 1990
Jewett MA, Kong YS, Goldberg SD, et al: Retroperitoneal lymphadenectomy for testis tumor with nerve sparing for ejaculation. J Urol 139:1220-1224, 1988
Bohlen D, Borner M, Sonntag RW, et al: Long-term results following adjuvant chemotherapy in patients with clinical stage I testicular nonseminomatous malignant germ cell tumors with high risk factors. J Urol 161:1148-1152, 1999
Studer UE, Burkhard FC, Sonntag RW: Risk adapted management with adjuvant chemotherapy in patients with high risk clinical stage I nonseminomatous germ cell tumor. J Urol 163:1785-1787, 2000
Cullen MH, Stenning SP, Parkinson MC, et al: Short-course adjuvant chemotherapy in high-risk stage I nonseminomatous germ cell tumors of the testis: A Medical Research Council report. J Clin Oncol 14:1106-1113, 1996
Pont J, Albrecht W, Postner G, et al: Adjuvant chemotherapy for high-risk clinical stage I nonseminomatous testicular germ cell cancer: Long-term results of a prospective trial. J Clin Oncol 14:441-448, 1996
Oliver RT, Raja MA, Ong J, et al: Pilot study to evaluate impact of a policy of adjuvant chemotherapy for high risk stage 1 malignant teratoma on overall relapse rate of stage 1 cancer patients. J Urol 148:1453-1456,1992
Ogan K, Lotan Y, Koeneman K, et al: Laparoscopic versus open retroperitoneal lymph node dissection: A cost analysis. J Urol 168:1945-1949, 2002
Read G, Stenning SP, Cullen MH, et al: Medical Research Council prospective study of surveillance for stage I testicular teratoma: Medical Research Council Testicular Tumors Working Party. J Clin Oncol 10:1762-1768, 1992
Boshoff C, Begent RH, Oliver RT, et al: Secondary tumours following etoposide containing therapy for germ cell cancer. Ann Oncol 6:35-40, 1995(Richard E. Link, Mohamad )
ABSTRACT
PURPOSE: Patients with clinical stage I nonseminomatous germ cell tumors (NSGCTs) have been managed with surveillance, chemotherapy, or retroperitoneal lymphadenectomy (RPLND) with similar survival outcomes. Cost factors influencing the choice of therapy were evaluated using computer-based decision analysis.
METHODS: A detailed model was developed that integrates projected costs for more than 60 possible treatment outcomes. It incorporates primary, adjuvant, and salvage chemotherapy, primary and postchemotherapy RPLND, and both laparoscopic and open surgical approaches. Starting values and probabilities were derived from a comprehensive meta-analysis of the last 25 years of testes cancer literature. Hypothesis testing was performed using sensitivity analysis.
RESULTS: The model predicts a cost premium for both primary chemotherapy (18.7%) and RPLND (51.7%) compared with surveillance. If laparoscopic RPLND was practiced, the cost premium for primary surgery (29.1%) approached that of chemotherapy (26.4%). Open RPLND was 1.25x as costly as laparoscopic RPLND, primarily because of longer hospitalization. The choice of open RPLND yielded a 6.9% cost premium for a surveillance program in this model. For such a program, primary chemotherapy became cost advantageous when the probability of recurrence during surveillance was more than 46%.
CONCLUSION: This model allows a variety of treatment cost hypotheses to be tested. Primary RPLND is never cost advantageous over surveillance or primary chemotherapy. Surgical costs can significantly increase the overall cost of a surveillance program. In stage I patients with high-risk tumor characteristics, primary chemotherapy may have a cost advantage over surveillance.
INTRODUCTION
Despite several decades of clinical investigation, the management of patients with stage I nonseminomatous germ cell tumors (NSGCT) after radical orchiectomy remains controversial. Primary surveillance, retroperitoneal lymphadenectomy (RPLND), and two chemotherapy cycles (bleomycin, etoposide and cisplatin [BEP]) all have similarly excellent survival outcomes and low morbidity. Cost concerns, therefore, may play an important role in determining the choice of management in this disease.
In 1996, Baniel et al1 published an elegant cost analysis comparing surveillance and primary RPLND at Indiana University. Two years later, Lashley and Lowe2 reported a similar study comparing surveillance, chemotherapy, and RPLND. In both cases, the authors collected information on medical charges from 100 patients in each treatment group. They concluded that the cost difference was relatively small (< 13.5%) and should not alter clinical decision making. Although these studies significantly advanced our understanding of the socioeconomic impact of stage I NSGCT, they suffer from three limitations. No attempt was made to derive a mathematical cost model from these results that could be used to test hypotheses with sensitivity analysis. Samples might include, "How quickly must I discharge my RPLND patients to achieve cost equivalence with chemotherapy?" "At what surveillance failure rate does chemotherapy become cost advantageous over surveillance?" The answers to these questions cannot easily be determined from the data presented. These results also reflect an experience at two high-volume testes cancer programs and may not easily be generalized to lower-volume institutions. Finally, laparoscopic RPLND, popularized during the current decade, was not factored into these analyses from the 1990s.
Here we develop a detailed mathematical decision analysis model comparing the projected management costs for stage I NSGCT. It is built on a meta-analysis of the testes cancer literature and reflects a multi-institutional clinical experience. The model incorporates a variety of clinical outcomes downstream from the original therapy decision, given that these events would potentially influence costs. In addition, laparoscopic RPLND and updated reimbursement are factored into the analysis.
We focused on four critical questions as a measure of model utility: How do the various options for managing stage I NSGCT compare in cost, and are the differences significant enough to drive clinical decision making? How do laparoscopic and open RPLND compare in perioperative costs, and can we identify targets for operative time and length of stay (LOS) that result in cost equivalence? How much impact does surgery have on the cost of a surveillance program, and does the choice of the laparoscopic or open approach significantly alter that cost? Does the most cost-effective management option change when patients with low- and high-risk tumor characteristics are compared?
METHODS
A mathematical decision-tree model was developed to estimate the cost of surveillance, primary chemotherapy, and RPLND (Table 1). Because of space constraints, some details of model development have been included as an Appendix. Downstream of the original treatment decision, the model incorporates more than 60 potential clinical outcomes (Fig A1). It includes terms for RPLND, postchemotherapy RPLND, and BEP and salvage chemotherapy (ifosfamide, vinblastine, and cisplatin). It does not incorporate high-dose chemotherapy salvage treatment with stem-cell support. Perioperative surgical costs were expressed in a submodel incorporating costs centers for operative time, surgical consumables, surgeon and anesthesiologist professional fees, hospital room charges, and blood transfusions. Both laparoscopic (RPLNDlaparoscopic) and open (RPLNDopen) surgical therapy were independently modeled. For simplification, if laparoscopic surgery was used for RPLND, it was also used for postchemotherapy RPLND in a given option. The model incorporates 51 variables (Table A7) including constants and terms for the probabilities of various clinical events. Starting values are derived from a 25-year Medline search of the testes cancer literature as documented in Tables A2 to A6. In some cases, information was not available in the literature and had to be estimated. This was particularly true for postchemotherapy RPLNDopen, for which published information on operative times and transfusion rates were lacking. For postchemotherapy RPLND, we added 90 minutes to the operative time of standard RPLND, and transfusion rates were set as identical to standard RPLND.
For surveillance, the follow-up protocol was modeled as described in the 2004 National Comprehensive Cancer Network Clinical Practice Guidelines in Oncology3 for a duration of 7 years. Costs for surveillance failures were determined by scaling costs based on the average time to failure in the literature. Pharmaceutical costs represent the lowest wholesale price for a particular drug published in the 2003 Drug Topics Red Book scaled to 2004 dollars using a web-based Consumer Price Index inflation calculator.4 Professional fees as well as imaging and laboratory costs were derived from 2004 Medicare reimbursement rates. Operative consumables were calculated from a list of items included on the surgical field and reflect purchase price. Hospital room and board and operative time charges were acquired from the administration at the Johns Hopkins Hospital (Baltimore, MD).
The cost model was programmed into TreeAge Pro Healthcare software (TreeAge Software Inc, Williamstown, MA), which was used to perform sensitivity analyses and probability calculations. In one-way analysis, a single cost factor is altered and its impact on total cost is determined by keeping all other factors fixed. In two-way analysis, two factors are altered simultaneously.
RESULTS
How Do the Various Options for Managing Stage I NSGCT Compare in Cost, and Are the Differences Significant Enough to Drive Clinical Decision Making?
Figure 1 shows a summary of baseline model predictions. Primary surveillance was the least costly option followed by chemotherapy (18.7% to 26.4% premium) and primary RPLND (29.1% to 51.7% premium), depending on the choice of surgical approach. Options including laparoscopic surgery were less costly than those incorporating open surgery. Primary RPLNDopen was 1.6 times more costly than the least-expensive option, surveillance withlaparoscopic therapy for retroperitoneal recurrence. Using two-way sensitivity analysis, operative time and LOS for RPLNDopen and RPLNDlaparoscopic were varied to evaluate how these factors affected the choice of the least costly therapy option (Fig 2). For the majority of operative times and LOS, surveillance shows a clear cost advantage.
How Do Laparoscopic and Open RPLND Compare in Perioperative Costs and Can We Identify Targets for Operative Time and LOS That Result in Cost Equivalence?
A submodel was developed that compared only the perioperative costs of RPLNDopen and RPLNDlaparoscopic (Table 2). This model predicts that RPLNDopen will be 1.25x as costly as RPLNDlaparoscopic, primarily based on longer inpatient hospitalization after RPLNDopen. The costs of hospitalization and operative time were the two most significant factors differentiating the two procedures. Transfusion rate was a negligible factor, contributing less than 0.5% to the total perioperative cost.
To identify operative time and LOS targets resulting in cost equivalence between RPLNDopen and RPLNDlaparoscopic, we performed a one- and two-way sensitivity analysis (Fig 3). This analysis demonstrated that cost equivalence could be met with a 5-day hospital stay after RPLNDopen, keeping other factors constant. Cost equivalence could also be reached if the room rate for inpatient hospitalization was less than $305 per day. In contrast, even unrealistic operative times (< 30 minutes) could not overcome the cost impact of a 7-day hospitalization for RPLNDopen.
How Much Impact Does Surgery Have on the Cost of a Surveillance Program, and Does the Choice of the Laparoscopic or Open Approach Significantly Alter That Cost?
The model not only allows projection of costs but also the probability that an individual patient will experience a particular outcome. Figure 4 shows a breakdown of predicted outcomes for patients entering surveillance: 99.6% of patients entering the program will ultimately be cured by some combination of observation with or without therapy for recurrence.
For a surveillance program, the probability of a patient undergoing an RPLND for disease recurrence is 0.195. The probability of a patient undergoing a postchemotherapy RPLND is much lower (0.017). Therefore, the maximal impact of surgical therapy costs on a surveillance program would be approximately 21%.
In practice, a comparison of laparoscopic and open surgical therapy is significantly more complex because the outcome probabilities are derived from a meta-analysis of the literature and differ slightly for the two approaches. For the purposes of this analysis, we ignored the impact of postchemotherapy surgery because it can only contribute 1.7% to the model. The submodel of perioperative RPLND cost predicts a difference of $3,244.86 between RPLNDlaparoscopic and RPLNDopen. If we multiply this by 0.195, it yields a predicted cost premium for RPLNDopen of 5.2%. This percentage represents the cost premium if all outcomes for the laparoscopic and open approaches were modeled identically. Given that the complete model also factors in postchemotherapy surgery and the slight variations in outcomes published for the two approaches, it predicts a cost premium of 6.9% for a surveillance program using open surgery compared with laparoscopy.
Does the Most Cost-Effective Management Option Change When Patients With Low- and High-Risk Tumor Characteristics Are Compared?
As shown in Fig 1, there is a clear financial advantage to surveillance over chemotherapy and RPLND. To evaluate the impact of high- and low-risk tumor characteristics on this cost relationship, we performed one-way sensitivity analyses varying the probability of disease recurrence during primary surveillance (Fig 5). If RPLNDlaparoscopic were practiced for retroperitoneal recurrence, surveillance would remain the most cost-effective option compared with primary chemotherapy or primary RPLNDlaparoscopic until the probability of retroperitoneal recurrence increased above 60%. Likewise, primary RPLNDopen required a treatment failure probability of 80%. In contrast, if open surgery was practiced, primary chemotherapy became cost advantageous over surveillance once the probability of retroperitoneal recurrence increased to only 46%.
DISCUSSION
When facing stage I NSGCTs, patients and urologic oncologists must decide among surveillance, chemotherapy, or RPLND. This decision hinges on the critical issues of oncologic effectiveness, morbidity, and cost. All three options have similarly high survival outcomes (97% to 100%).5 Likewise, the morbidity of nerve-sparing modified-template RPLND or short-course chemotherapy is modest.6-15 Cost may, therefore, play an important role in directing clinical decisions in this disease. A detailed decision-tree model was developed for projecting the management costs for stage I NSGCTs. The model’s advantage lies in the ease with which various hypotheses can be rapidly tested through the use of sensitivity analysis.
There was a clear cost advantage for primary surveillance over chemotherapy or RPLND. The magnitude of this advantage (18.7% to 51.7%) was larger than that previously reported during the 1990s (< 13.5%).1,2 However, our methodology differs significantly from these earlier studies, which determined cost by reviewing hospital charges for 100 patients in each of the treatment arms. The current model is based on a mixture of hospital costs (ie, consumables, pharmaceuticals), Medicare reimbursement rates (ie, professional fees, imaging), and hospital charges (ie, room and board, operative time). For example, Lashley and Lowe2 report an abdominal computed tomography scan costs $1,152, whereas our model uses the current Medicare reimbursement rate of $419 (2.7x less). Other differences include variations in the surveillance protocol and the omission of salvage chemotherapy from earlier studies. Finally, outcome probabilities are derived from a meta-analysis and differ slightly from those estimated in the older reports.
The current model predicts that perioperative cost of RPLNDopen will be 1.25x that of RPLNDlaparoscopic. This conclusion differs significantly from the results of a similar perioperative modeling study by Ogan et al,16 which predicted that RPLNDlaparoscopic would be 1.1x as expensive as RPLNDopen. For the current model, starting values were derived from a more comprehensive meta-analysis of RPLND reports (n = 20) than used by Ogan et al (n = 5).16 In addition, calculations of operative time and hospital room charges are radically different between the Johns Hopkins and University of Texas Southwestern institutional models. For example, the daily room rate of $395 used in the study by Ogan et al16 approaches the target room rate for cost equivalence between laparoscopic and open surgery in the Johns Hopkins model ($305). It is also important to stress that the current study focuses specifically on cost and does not represent a comparison of outcomes between RPLNDopen and RPLNDlaparoscopic.
Another conclusion relates to the management of stage I NSGCT patients with high-risk tumor characteristics such as vascular or lymphatic invasion, presence of embryonal cell carcinoma, and absence of yoke sac components. In this high-risk group, the rate of retroperitoneal recurrence during surveillance has been estimated at approximately 50%.17 Our model predicts that primary chemotherapy may be cost advantageous over a surveillance program using RPLNDopen with this failure rate.
It is critical to appreciate the limitations as well as the strengths of a modeling exercise of this type. Any model depends on a set of critical assumptions, some of which can be supported by the literature, whereas others are simply estimates or reflect local clinical practice. Cost modeling is also inherently dependent on institution and region, and these results may not be valid for all institutions. Yet the structure of the model reflects general management principles for stage I NSGCT, and should be easily translated to other centers provided that institution-specific cost components are modified.
One potential criticism relates to our LOS estimate after open RPLND. This is important because hospital room cost was the most significant factor differentiating the two approaches. Unfortunately, most large series of RPLNDopen patients do not report LOS; hence, the model starting value of 7.1 days is derived from only four of the 12 evaluated studies. The RPLNDopen series also tend to be historically older or from international institutions, hence hospitalization may not reflect 2004 United States practice. Sensitivity analysis identified an LOS target of 5 days for cost equivalence between RPLNDlaparoscopic and RPLNDopen, which may be a more realistic estimate of current LOS. In that case, results reported here for management options using RPLNDlaparoscopic would also reflect options including RPLNDopen.
Consistent with previously published studies,2,16 the current model does not include treatment-related morbidity or patient compliance in the cost analysis. Although morbidity influences patient and physician decision making, we feel that its impact on the aggregate cost of each strategy will be small. The morbidity of nerve-sparing modified-template RPLND is modest, relating primarily to a 1% to 3% rate of bowel obstruction6,7 and a 1% to 7% rate of anejaculation.7-10 The morbidity of two BEP cycles also appears to be minimal. For 58 patients, Bohlen et al11 reported only transient neutropenia and thrombocytopenia; granulocyte stimulation support was required in only one patient. Numerous studies have confirmed the low acute morbidity of two cycles of BEP and documented no significant long-term pulmonary or cardiac toxicity from this regimen.12-15 The long-term risk of secondary leukemia after BEP also appears to be low (< 0.7%) for patients who receive less than 2 g/m2 of etoposide.18 The total etoposide dose given for two BEP cycles is less than 0.8 g/m2.12 For patients receiving more than two adjuvant BEP cycles, the risk of organ toxicity and secondary malignancies may be higher. However, the model predicts that no more than 10% of patients on any treatment regimen will receive more than two cycles. Even if morbidity doubled the cost of chemotherapy for this patient subset, our overall conclusions would not change.
It is important to differentiate the cost analysis presented here from a formal cost-effectiveness analysis. Given the small relative differences in mortality among surveillance, RPLND, and chemotherapy for stage I NSGCT, use of mortality as a measure of effectiveness is difficult. An alternative strategy would be to assign estimated quality-of-life utility values to all the outcome states within the model. As noted previously, the morbidity differences between chemotherapy and RPLND are also likely to be small. Estimates based on such small and controversial differences in morbidity would add an unacceptable degree of systematic error to results, and would cloud their interpretation. As more comprehensive morbidity data become available, however, extension of this analysis to include effectiveness would likely be valuable.
The cost model presented here allows a variety of economic hypotheses about the management of stage I NSGCTs to be tested readily. It emphasizes the cost advantage of surveillance, quantifies the difference between laparoscopic and open surgical options, and suggests that, under some circumstances, primary chemotherapy may be cost advantageous for patients with high-risk tumor characteristics.
Authors’ Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
Appendix
Click here for Fig A1http://www.jco.org/cgi/content/full/23/24/5762/DC1
NOTES
Presented in part at the 2004 Society for Urologic Oncology Meeting, Bethesda, MD, December 3-4, 2004.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Baniel J, Roth BJ, Foster RS, et al: Cost- and risk-benefit considerations in the management of clinical stage I nonseminomatous testicular tumors. Ann Surg Oncol 3:86-93, 1996
Lashley DB, Lowe BA: A rational approach to managing stage I nonseminomatous germ cell cancer. Urol Clin North Am 25:405-423, 1998
Clinical Practice Guidelines in Oncology: Testicular Cancer. Version 1.2004. http://www.nccn.org/professionals/physician_gls/PDF/testicular.pdf
Consumer Price Index Calculator: Inflation Calculator. http://data.bls.gov/cgi-bin/cpicalc.pl
Porter J, Obek C, Lange PH: The management of stage I nonseminomatous germ cell tumor. Campbells Urol Updates 2:1-12, 2004
Baniel J, Foster RS, Rowland RG, et al: Complications of primary retroperitoneal lymph node dissection. J Urol 152:424-427, 1994
Heidenreich A, Albers P, Hartmann M, et al: Complications of primary nerve sparing retroperitoneal lymph node dissection for clinical stage I nonseminomatous germ cell tumors of the testis: Experience of the German Testicular Cancer Study Group. J Urol 169:1710-1714, 2003
Donohue JP, Thornhill JA, Foster RS, et al: The role of retroperitoneal lymphadenectomy in clinical stage B testis cancer: The Indiana University experience (1965 to 1989). J Urol 153:85-89, 1995
Donohue JP, Foster RS, Rowland RG, et al: Nerve-sparing retroperitoneal lymphadenectomy with preservation of ejaculation. J Urol 144:287-292, 1990
Jewett MA, Kong YS, Goldberg SD, et al: Retroperitoneal lymphadenectomy for testis tumor with nerve sparing for ejaculation. J Urol 139:1220-1224, 1988
Bohlen D, Borner M, Sonntag RW, et al: Long-term results following adjuvant chemotherapy in patients with clinical stage I testicular nonseminomatous malignant germ cell tumors with high risk factors. J Urol 161:1148-1152, 1999
Studer UE, Burkhard FC, Sonntag RW: Risk adapted management with adjuvant chemotherapy in patients with high risk clinical stage I nonseminomatous germ cell tumor. J Urol 163:1785-1787, 2000
Cullen MH, Stenning SP, Parkinson MC, et al: Short-course adjuvant chemotherapy in high-risk stage I nonseminomatous germ cell tumors of the testis: A Medical Research Council report. J Clin Oncol 14:1106-1113, 1996
Pont J, Albrecht W, Postner G, et al: Adjuvant chemotherapy for high-risk clinical stage I nonseminomatous testicular germ cell cancer: Long-term results of a prospective trial. J Clin Oncol 14:441-448, 1996
Oliver RT, Raja MA, Ong J, et al: Pilot study to evaluate impact of a policy of adjuvant chemotherapy for high risk stage 1 malignant teratoma on overall relapse rate of stage 1 cancer patients. J Urol 148:1453-1456,1992
Ogan K, Lotan Y, Koeneman K, et al: Laparoscopic versus open retroperitoneal lymph node dissection: A cost analysis. J Urol 168:1945-1949, 2002
Read G, Stenning SP, Cullen MH, et al: Medical Research Council prospective study of surveillance for stage I testicular teratoma: Medical Research Council Testicular Tumors Working Party. J Clin Oncol 10:1762-1768, 1992
Boshoff C, Begent RH, Oliver RT, et al: Secondary tumours following etoposide containing therapy for germ cell cancer. Ann Oncol 6:35-40, 1995(Richard E. Link, Mohamad )