Radiotherapy may improve the outcome of patients with pancreatic cancer but at an increased cost. In this study, the authors evaluated the cost-effectiveness of modern radiotherapy techniques in the treatment of locally advanced pancreatic cancer.
A Markov decision-analytic model was constructed to compare the cost-effectiveness of 4 treatment regimens: gemcitabine alone, gemcitabine plus conventional radiotherapy, gemcitabine plus intensity-modulated radiotherapy (IMRT); and gemcitabine with stereotactic body radiotherapy (SBRT). Patients transitioned between the following 5 health states: stable disease, local progression, distant failure, local and distant failure, and death. Health utility tolls were assessed for radiotherapy and chemotherapy treatments and for radiation toxicity.
SBRT increased life expectancy by 0.20 quality-adjusted life years (QALY) at an increased cost of $13,700 compared with gemcitabine alone (incremental cost-effectiveness ratio [ICER] = $69,500 per QALY). SBRT was more effective and less costly than conventional radiotherapy and IMRT. An analysis that excluded SBRT demonstrated that conventional radiotherapy had an ICER of $126,800 per QALY compared with gemcitabine alone, and IMRT had an ICER of $1,584,100 per QALY compared with conventional radiotherapy. A probabilistic sensitivity analysis demonstrated that the probability of cost-effectiveness at a willingness to pay of $50,000 per QALY was 78% for gemcitabine alone, 21% for SBRT, 1.4% for conventional radiotherapy, and 0.01% for IMRT. At a willingness to pay of $200,000 per QALY, the probability of cost-effectiveness was 73% for SBRT, 20% for conventional radiotherapy, 7% for gemcitabine alone, and 0.7% for IMRT.
Controversy surrounds radiotherapy in the management of locally advanced, unresectable pancreatic cancer. Although studies by the Gastrointestinal Tumor Study Group (GITSG) that were completed 2 decades ago indicated a benefit from 5-fluorouracil (5FU)-based chemoradiotherapy,1, 2 the chemotherapy regimens and radiotherapy techniques in those trials are obsolete by today's standards. More recently, the chemotherapeutic agent gemcitabine has demonstrated activity in pancreatic cancer when used alone3 and when combined with radiotherapy.4 In addition, the recently reported Eastern Cooperative Oncology Group study E4201, which compared gemcitabine alone with gemcitabine plus radiotherapy, demonstrated a survival advantage in the combined-modality arm.5
Modern chemoradiotherapy regimens often suffer from poor rates of local control. In pancreatic cancer, local disease progression can lead to pain and obstruction and may impact survival.6 At Stanford, we have attempted to intensify local therapy and decrease treatment duration with the technique of single-fraction stereotactic body radiotherapy (SBRT).7-9 At our institution, a regimen that sandwiches a single, large, daily fraction of radiation in between cycles of gemcitabine has resulted in excellent local control and has yielded survival rates comparable to other chemoradiotherapy regimens. Despite the benefits of radiotherapy demonstrated in the E4201 study and in our SBRT experience, other clinical studies comparing chemotherapy with chemoradiotherapy have produced mixed results,10, 11 and the optimal management of locally advanced pancreatic cancer remains unclear.
Although radiotherapy has the potential to improve outcome in pancreatic cancer, the absolute incremental benefit of radiotherapy probably is small because of the limited survival inherent with this disease. This small, incremental benefit compounded by the high cost of radiation brings about the question of cost-effectiveness. The purpose of the current study was to evaluate the cost-effectiveness of gemcitabine combined with modern radiotherapy techniques compared with gemcitabine as a single agent in the management of locally advanced pancreatic cancer.
MATERIALS AND METHODS
We compared the cost-effectiveness of 4 different regimens that are used to treat locally advanced pancreatic cancer: 1) gemcitabine alone (gem-alone), 2) gemcitabine with conventionally fractionated radiotherapy (gem-RT), 3) gemcitabine with intensity-modulated radiotherapy (gem-IMRT), and 4) gemcitabine with stereotactic body radiotherapy (gem-SBRT). A decision-analytic Markov model was constructed to determine the cost-effectiveness of these 4 regimens. The 5 main health states in this model included the following: stable disease, local progression, distant metastatic failure, both local and distant failure, and death (for the model schema, see Fig. 1; for the decision tree, see Fig. 2). We assumed that all patients entered the model with stable disease and received either chemotherapy alone or chemotherapy and radiation according to their prespecified treatment regimen, as described in detail below (see Treatment). Although, from a clinical standpoint, patients can develop distant failure after local progression, and vice versa, our model did not allow these transitions. Death from natural causes could occur from any health state and was estimated from age-specific US mortality rates.12 Death from cancer was assumed to occur after disease progression. Patients incurred utility tolls and costs for receiving radiotherapy, chemotherapy or from experiencing treatment-related toxicity.
The Markov model was based on a payer's perspective and ran with a 1-month cycle length. Although only a small percentage of patients with locally advanced pancreatic cancer survive for >2 years, the model was run over a 5-year time horizon to avoid excluding the minority of longer-term survivors. Costs were adjusted to 2009 dollars using the medical component of the Consumer Price Index. Costs and health outcomes were discounted at 3% per year. The Markov model was built with TreeAge Pro 2009 Suite (release 1.0.2; TreeAge Inc., Williamstown, Mass).
The cost-effectiveness of treatment regimens was measured with the incremental cost-effectiveness ratio (ICER). The base-case analysis was defined as the results of the analysis using the data and methods that we believed best characterized each treatment option.13 A treatment option was considered dominated if it was more costly and less effective than another treatment. Two cost-effectiveness analyses were conducted: one with all 4 treatment options and a second comparing gem-alone with conventionally fractionated radiotherapy (gem-RT and gem-IMRT). Model transition probabilities, costs, and utilities are described below and are included in Table 1.
The Markov model relied on monthly transition probabilities (for details see Materials and Methods). For simplicity, this table presents the average of the monthly probabilities from months 1 through 6, 7 through 12, 12 through 18 and 19 through 24.
Cost of grade 3-4 radiation toxicity (per event), $
Probability of grade 3-4 RT toxicity per mo (Loehrer 20085; Chang 20097; Murphy 201017)
Local disease progression or distant metastases alone
Local disease progression and distant metastases
On treatment (chemotherapy and/or RT)
With gem-alone and gem-RT, we used the treatment schema and outcome data from the preliminary report of the clinical trial E4201.5 That trial randomized patients to receive monthly cycles of gemcitabine alone (1000 mg/m2 weekly every 3 of 4 weeks), or gemcitabine plus radiotherapy (600 mg/m2 weekly with 50.4 grays [Gy] in 28 fractions) followed by monthly gemcitabine (1000 mg/m2 every 3-4 weeks). As in the E4201 trial, we assumed gem-RT included conventionally fractionated, 3-dimensional, conformal radiotherapy.
We assumed the gem-IMRT arm was treated with an IMRT technique designed to spare surrounding normal tissues without sacrificing tumor control. Compared with gem-RT, gem-IMRT received the same chemotherapy regimen and the same total radiation dose, experienced the same patterns of failure, and had the same survival. Because of the normal tissue sparing, we assumed that toxicity from radiation was lower for gem-IMRT compared with gem-RT.
With gem-SBRT, we used the treatment schema developed here at Stanford.7-9 We assumed that this group received a single cycle of gemcitabine (1000 mg/m2 weekly every 3-4 weeks), followed by SBRT (25 Gy in a single fraction), then monthly gemcitabine (1000 mg/m2 weekly every 3 of 4 weeks). We assumed that all treatment arms received gemcitabine for a maximum of 5 cycles or until disease progression. After any disease progression (local, distant or both), all treatment groups received salvage chemotherapy, which consisted of oxaliplatin, leucovorin, and 5-FU18 for 2 cycles or until death.
Key probabilities for this model included the probability of disease progression, the probability of death, and the probability of toxicity. The gem-alone, gem-RT, and gem-IMRT probabilities were based on the E42015 clinical trial; whereas the gem-SBRT probabilities were based on our Stanford SBRT experience. Although the E4201 trial included only patients with locally advanced disease, our previously published reports on SBRT included patients with locally advanced and metastatic disease who did and did not receive preradiation chemotherapy.7-9, 19 For the current analysis, we isolated the patients with nonmetastatic, locally advanced pancreatic cancer who received preradiation gemcitabine followed by a single fraction of 25 Gy (n = 48). These patients were treated on protocol (n = 31) and off protocol (n = 17), and both groups were treated identically. Progression and survival were determined with the Kaplan-Meier method and was calculated from the start of treatment. This reanalysis of our SBRT data set enabled a more precise comparison with the E4201 results; however, we were unable to directly compare our data set with E4201. Because information on true disease progression rates often are lacking, we used disease-free survival as a surrogate.
Disease Progression, Patterns of Failure, and Overall Survival
With the limited life span inherent in pancreatic cancer, we suspected that our Markov model would be sensitive to disease progression and overall survival. Consequently, we sought to model these events as accurately as possible by using transition probabilities that varied each month, thus creating cost-effectiveness model outputs that replicate Kaplan-Meier plots. Kaplan-Meier survival estimates lose precision at later time points, when few patients remain available for analysis; therefore, when survival estimates dropped below 10%, we assumed that survival followed SEER survival data in patients with locally advanced pancreatic cancer.20 Once a patient progressed, they fell into 1 of 3 states: local progression alone, distant progression alone, or local and distant progression. These patterns of failure were deduced from the E4201 trial and from our Stanford experience.
We considered only grade 3 or higher gastrointestinal toxicity attributable to radiation. We assumed that grade 1/2 toxicity and chemotherapy toxicity would be similar among the treatment arms and, thus, these explicitly were not included in the model. For gem-RT, the rate of toxicity from radiation was defined from E4201 as the difference in gastrointestinal toxicity rates between gem-alone (14%) and gem-RT (38%). To our knowledge, the relative decrease in toxicity from IMRT compared with conventional radiotherapy has not been reported in patients with locally advanced pancreatic cancer. However, a dosimetric study21 estimated that IMRT would reduce small bowel toxicity from 24.4% to 9.3%, for a 62% relative reduction; therefore, the rate of toxicity for gem-IMRT was assumed to be 38% that of gem-RT. The rate of toxicity from SBRT was determined from our Stanford experience.7, 17 The total rates of toxicity were converted to a constant monthly rate for each group (Table 1). Each toxicity event was assumed to last for 1 month.
Quality of Life
The health utility state for stable disease was based on expert opinion. The health utility state for local disease progression or distant metastatic failure was assumed to be identical and was estimated from the literature.22 The utility decrement between stable disease and local or distant progression was 0.06. We assumed the utility decrement between stable disease and patients with both local and distant progression was double, or 0.12. Therefore, the utility of stable disease was 0.68, the utility of local or distant progression was 0.62, and the utility of both local and distant progression was 0.56. Health utility tolls refer to a 1-time, absolute utility decrement in a given cycle. Utility tolls were assessed for treatment and toxicity, and these values were estimated from the literature23 (Table 1).
The costs of radiotherapy, IMRT, SBRT, and administering chemotherapy were derived from the 2009 Medicare Physician Fee Schedule adjusted to Santa Clara County (http://www.cms.hhs.gov/PhysicianFeeSched/ [access date February 6, 2010]). The costs of chemotherapeutic agents were determined from 2009 Medicare Part B reimbursements. Gem-SBRT patients had small gold fiducial markers implanted into their tumors before SBRT for radiation tracking purposes, and these costs were estimated from the literature.14
The average cost of a radiation toxicity event was estimated from a weighted average of the costs of individual toxicity events observed in our SBRT data set. The cost of each event was estimated from the mean cost identified in the 2007 US Agency for Healthcare Research and Quality national statistics on inpatient hospital stays.24 Specific toxicity details were not available from the E4201 study; therefore, we assumed a similar cost per toxicity event for gem-RT and gem-IMRT.
The monthly costs of additional nontreatment medical care, including office visits, medications, laboratory, and radiology costs, were estimated from the literature.15 The cost of end-of-life care was estimated from health care expenditures in cancer patients during their last month of life.16
One-way and probabilistic sensitivity analyses were performed to test the robustness of the Markov model. All costs, probabilities, and utilities were varied independently in 1-way sensitivity analyses. In addition, the model time horizon was varied (2-5 years) as was number of gemcitabine cycles (3-7 cycles), the number of salvage chemotherapy cycles (1-3 cycles), patient age (45-85 years), and annual discount rate (0%-5%).
With disease progression and survival, we altered the monthly time-dependent transition probabilities to adjust the mean survival with the following transformation: psen(t) = 1 − (1 − p0[t])s, where p0(t) is the original monthly probability, psen(t) is the new monthly probability used in the sensitivity analysis, and s is the scale factor. For each treatment regimen, we determined the scale factor range that resulted in the desired change in mean survival, and this scale factor range was used in the sensitivity analysis. We assumed that the rates of disease progression and death were correlated and, thus, adjusted these rates simultaneously in the sensitivity analysis.
The costs of radiotherapy, IMRT, and SBRT depend on several overlapping Common Procedural Terminology (CPT) codes. To accurately represent this cost structure in our sensitivity analysis, we varied the individual cost CPT codes as well as a global radiation oncology reimbursement rate. This method reflects the uncertainty in individual billing codes as well as the uncertainty in overall reimbursement for radiation oncology.
Finally, a probabilistic sensitivity analysis was preformed with a Monte Carlo simulation of 10,000 repetitions.25 With the probabilistic sensitivity analysis, all costs were modeled with log-normal distributions. Beta distributions were chosen to model the probabilities of toxicity, all utilities, and utility tolls. Fractile distributions were used to model the number of cycles of gemcitabine and salvage chemotherapy. Finally, gamma distributions were used to vary the scale factor (defined above), which effectively varied the probability of disease progression and survival. A complete list of parameters is available online (http://188.8.131.52/AK/pancreas_CEA_PSA.pdf.
Figure 3 illustrates the predicted disease progression, patterns of failure, and survival for all 4 treatment regimens. Gem-SBRT demonstrated decreased rates of local progression and increased rates of distant metastatic failure compared with the other modalities. Gem-alone had slightly inferior survival compared to all of the radiation regimens.
The primary cost-effectiveness analysis, including all 4 treatment options, demonstrated that gem-SBRT had an ICER of $69,500 per quality-adjusted life-year (QALY) compared with gem-alone (Table 2). Gem-SBRT dominated the more costly and less effective options of gem-RT and gem-IMRT.
To understand the cost-effectiveness of conventionally fractionated radiotherapy, we conducted a secondary cost-effectiveness analysis excluding gem-SBRT. This demonstrated that gem-RT had an ICER of $126,800 per QALY compared with gem-alone. The decreased toxicity of IMRT increased the incremental effectiveness; however, the high cost of IMRT increased the ICER above $1 million per QALY compared with conventional radiotherapy (Table 2).
Table 3 demonstrates key results from the primary 1-way sensitivity analysis comparing gem-alone with gem-SBRT. The analysis indicates a high degree of sensitivity to changes in mean survival for the gem-alone and gem-SBRT arms. The ICER of gem-SBRT increased to $200,000 per QALY when the mean survival of gem-alone increased by 2.5 months or when the mean survival of gem-SBRT decreased by 2.6 months. Conversely, the ICER of gem-SBRT decreased below $50,000 per QALY when the mean survival of gem-alone decreased by 2.0 months or when the mean survival of gem-SBRT increased by 2.0 months.
Table 3. One-Way Sensitivity Analysis With Gemcitabine Alone Versus Gemcitabine Plus Stereotactic Body Radiotherapy
Incremental Cost-Effectiveness Ratio, $/QALY
Abbreviations: QALY, quality-adjusted life-year; SBRT, stereotactic body radiotherapy.
Represents the base-case value.
Holding constant the number of patients with distant progression.
One theoretical advantage of SBRT relates to its high rate of local control; however the data supporting this observation come from our single-institution experience. After 1 year, our model predicted that, among patients who received gem-SBRT, 16% would fail with local progression (8% local failure alone and 8% local + distant failure), and 63% would fail with distant progression (55% distant alone and 8% local + distant). When we assumed double the rate of local progression (32%; 16% local alone and 16% local + distant) after gem-SBRT, keeping distant failure constant (63%; 47% distant alone and 16% local + distant), the ICER increased to $130,000 per QALY compared with gem-alone. In addition, our model assumed the rate of death was identical after either local or distant progression; however, the relation between local progression and survival in locally advanced pancreatic cancer is debatable. If we halved the rate of death after local progression (holding the rate of death after distant failure constant), the ICER of gem-SBRT increased to $88,000 per QALY. When we reduced the rate of death after local progression to zero, then all radiation options added only cost and toxicity and, subsequently, were dominated by gem-alone.
The base-case analysis derived rates of gem-SBRT disease progression and survival from our institutional experience. If we assumed that gem-SBRT had rates of disease progression and survival identical to the rates of gem-RT in E4201, then the ICER of gem-SBRT would increase slightly to $92,800 per QALY. This lower ICER of gem-SBRT compared with gem-RT came from the lower cost of SBRT compared with radiotherapy.
We undertook a secondary 1-way sensitivity analysis between gem-RT and gem-IMRT to explore conditions in which IMRT might be considered cost-effective. The base-case analysis assumed identical survival for gem-RT and gem-IMRT. If we relaxed this assumption, then the mean survival of gem-IMRT would have to increase by 1.0 months over gem-RT to bring the ICER of gem-IMRT down from $1,584,100 per QALY to $200,000 per QALY, and it would have to increase by 4.8 months to bring the ICER of gem-IMRT to $50,000 per QALY. In addition, the base-case analysis assumed that IMRT reduced the rate of toxicity to 38% compared with gem-RT. If we assumed that IMRT eliminated radiation toxicity, then the ICER of gem-IMRT only decreased to $843,200 per QALY.
Finally, we conducted a probabilistic sensitivity analysis with all 4 treatment options (Fig. 4). The probability of cost-effectiveness at a willingness to pay of $50,000 per QALY was 78% for gem-alone, 21% for gem-SBRT, 1.4% for gem-RT, and 0.01% for gem-IMRT. Conversely, the probability of cost-effectiveness at a willingness to pay of $200,000 per QALY was 73% for gem-SBRT, 20% for gem-RT, 7% for gem-alone, and 0.7% for gem-IMRT. At willingness-to-pay levels above $70,900 per QALY, gem-SBRT surpassed gem-alone to become the most likely cost-effective treatment option. And, as the willingness-to-pay threshold increased above $124,600 per QALY, gem-RT became a more likely cost-effective option than gem-alone.
Advances in the field of radiation oncology have led to improvements in imaging and targeting as well as radiation treatment delivery. This gain in technology allows the delivery of an increased radiation dose to tumor and a decreased dose to surrounding normal tissues. Although these advances have the potential to increase tumor control and decrease toxicity, randomized clinical evidence supporting their widespread adoption does not exist. Plus, the potentially small absolute survival benefits of these advances come at a large cost. These technology innovations in radiation oncology naturally lead to the question of cost-effectiveness—an extremely relevant subject in today's health care economy.
Perhaps the most pertinent finding of this study relates to the cost-effectiveness of IMRT. In the current study, we observed that IMRT had an ICER >$1,500,000 per QALY, likely because of the small incremental survival benefit combined with the large increase in cost. Even when we assumed that IMRT improved survival, it would have to increase mean survival by 4.8 months over conventional radiotherapy before IMRT becomes cost-effective at a $50,000 per QALY threshold. These findings provide useful insight into the cost-effectiveness of IMRT and indicate that justifying this treatment from a cost-effectiveness standpoint would require a substantial improvement in survival or a large decrease in cost. The findings in this study emphasize the fact that this expensive technology will unlikely be cost-effective in diseases like pancreatic cancer that have such limited survival. Although IMRT was more expensive than conventional radiotherapy, we observed that SBRT was significantly less expensive than other forms of radiotherapy. Despite the increased technology demands with SBRT, the shorter treatment duration results in an overall decrease in cost compared with conventional radiotherapy.
The challenge of interpreting cost-effectiveness analyses in the United States arises from the lack of a definitive willingness-to-pay threshold. The United States does not have a set monetary threshold below which a treatment is considered “cost effective.”26, 27 The commonly cited value of $50,000 per QALY arose in 1982 as the estimated cost-effectiveness of dialysis in patients with chronic renal failure; however, many consider this value well below what society would currently consider acceptable.26 A recent report estimates that $183,000 to $264,000 per QALY more accurately reflects society's willingness to pay for health care.28 This current study found that gem-SBRT had an incremental cost-effectiveness ratio of $69,500 per QALY compared with gem-alone, which is below several willingness-to-pay thresholds but above the commonly cited $50,000 per QALY threshold. Other cost-effectiveness studies involving radiation in pancreatic cancer have produced similar results. Krzyzanowska et al,23 reported that 5-FU-based chemoradiotherapy had an ICER of $68,724 per QALY compared with no treatment. Krzyzanowska et al also observed that gemcitabine-based chemoradiotherapy would have an ICER <$100,000 per QALY compared with 5-FU chemoradiotherapy if the gemcitabine regimen improved survival by 5% at 1 year. Conversely, adding erlotinib to gemcitabine in advanced pancreatic cancer (improvement in median survival, <1.5 weeks) increased cost by $364,680 to $400,000 per life-year gained,29, 30 which far exceeds what we observed with radiotherapy or SBRT in the current study.
Another important finding of this study pertains to the high degree of sensitivity our model exhibited to changes in survival. This observation reflects the small absolute incremental benefit of radiotherapy compounded by the large cost. The sensitivity to survival naturally leads to a tight correlation between clinical data used to inform the model and the results. Had we informed our model with other trials of gemcitabine-based treatments10 or different SBRT regimens,31 our results may have differed. Ultimately, we chose clinical studies for this model that provided the highest level of evidence while allowing direct comparison between treatment arms.
The main limitations of this analysis relate to the data used to inform the Markov model. The clinical trial E4201 that we used to inform the gem-alone and gem-RT groups in our cost-effectiveness model closed early because of poor accrual, and only preliminary results have been reported. Despite this, all eligible patients in the E4201 trial had died by the time of preliminary analysis, suggesting that the final toxicity, progression, and survival estimates will not change. In addition, a preliminary report from a separate, randomized phase 2 study comparing gemcitabine alone with gemcitabine plus radiotherapy indicated a survival benefit in the radiotherapy group.32 This second study agrees with the results from E4201 and adds support to our cost-effectiveness modeling assumption that gem-RT had improved outcomes compared with gem-alone.
Another limitation relates to the fact that the Markov model in our study compared the preliminary results from a phase 3 clinical trial (gem-alone and gem-RT in E4201) with phase 2 clinical data (gem-SBRT). We were unable to account for differences in study populations between our Stanford SBRT cohort and the E4201 cohorts. Differences in patient characteristics, treatment, or follow-up could bias the results of this study. Ideally, our cost-effective model would be informed with a randomized study comparing all 4 treatment regimens (gem-alone, gem-RT, gem-IMRT, and gem-SBRT). Realistically, such a trial would be challenging to complete given the inherent difficulty with patient accrual into multimodality trials.33 Plus, the large sample sizes required to detect the anticipated small differences in outcome make such a trial infeasible. Despite these limitations, our model predicted no significant survival advantage for gem-RT or gem-SBRT (see Fig. 3). In addition, when we assumed that gem-SBRT produced survival identical to that produced by gem-RT in the E4201 trial, the ICER for gem-SBRT remained relatively unchanged at $92,800 per QALY.
In addition to our primary data, other limitations in this study relate to the assumptions used to build the cost-effectiveness model. Similar to other cost-effectiveness analyses, this analysis excluded grade 1/2 toxicity. Differing rates of low-grade toxicity among the treatment arms may have biased our results. However, the sensitivity analysis indicated that assumptions about grade 3/4 toxicity did not alter our conclusions, suggesting that low-grade toxicity would be unlikely to affect the model. Another limitation concerns our assumptions about supportive care. In pancreatic cancer, supportive care is complex and includes components such as pain, jaundice, weight loss, gastric or biliary obstruction, diabetes, pancreatic insufficiency, paraneoplastic syndromes, chronic nausea, vomiting or diarrhea, and depression. Some of these costs were accounted for indirectly in our estimation of additional medical expenses and end-of-life care, and the decrement in quality of life was partially reflected in our health utility states. Precise accounting for the costs and quality-of-life outcomes with these events would be challenging, and our current study excluded these events. Differences in costs or benefits of supportive care between treatment arms could have biased our results in unpredictable ways. A final limitation pertains to the finding that health utility states are poorly defined in pancreatic cancer and are derived mostly from expert opinion. Despite this limitation, the model proved insensitive to changes in utility (Table 3), suggesting that more precise estimates would not alter our conclusions.
In summary, this study provides a useful framework for comparing the tradeoff between the benefits and burdens of radiotherapy in patients with locally advanced pancreatic cancer. IMRT in this population surpasses what society considers cost-effective. Conversely, combining gemcitabine with SBRT increases the clinical effectiveness beyond gemcitabine alone at a cost potentially acceptable by today's standards.
This work was supported in part by the Henry S. Kaplan Research Education Fund (J.D.M.).