Although both radical cystectomy and intravesical immunotherapy are initial treatment options for high-risk, T1, grade 3 (T1G3) bladder cancer, controversy regarding the optimal strategy persists. Because bladder cancer is the most expensive malignancy to treat per patient, decisions regarding the optimal treatment strategy should consider costs.
A Markov Monte-Carlo cost-effectiveness model was created to simulate the outcomes of a cohort of patients with incident, high-risk, T1G3 bladder cancer. Treatment options included immediate cystectomy and conservative therapy with intravesical Bacillus Calmette-Guerin (BCG). The base case was a man aged 60 years. Parameter uncertainty was assessed with probabilistic sensitivity analyses. Scenario analyses were used to explore the 2 strategies among patients stratified by age and comorbidity.
The quality-adjusted survival with immediate cystectomy and BCG therapy was 9.46 quality-adjusted life years (QALYs) and 9.39 QALYs, respectively. The corresponding mean per-patient discounted lifetime costs (in 2005 Canadian dollars) were $37,600 and $42,400, respectively. At a willingness-to-pay threshold of $50,000 per QALY, the probability that immediate cystectomy was cost-effective was 67%. Immediate cystectomy was the dominant (more effective and less expensive) therapy for patients aged <60 years, whereas BCG therapy was dominant for patients aged >75 years. With increasing comorbidity, BCG therapy was dominant at lower age thresholds.
In North America, more than 60,000 cases of bladder cancer are diagnosed every year.1, 2 Approximately 75% of patients have superficial (nonmuscle-invasive) disease at first presentation.3 Such tumors carry a high risk of recurrence, mandating intensive follow-up. However, repeated office visits, cystoscopic and cytologic monitoring, and the treatment of recurrent lesions, often with intravesical therapy, can be costly. Approximately 60% of the lifetime cost affiliated with bladder cancer is attributable to tumor surveillance.4 Because bladder cancer is the most expensive malignancy to treat on a case-by-case basis,5 it is important to identify opportunities to reduce treatment and surveillance costs while preserving survival and quality of life.
The treatment of high-risk, high-grade, superficial (T1, grade G3 [T1G3]) bladder cancer remains controversial.6 These cancers invade the lamina propria, frequently recur, and often progress to invade underlying muscle, which is associated with decreased survival. The risk of disease progression is particularly high if cancers are multifocal or are associated with carcinoma in situ. For such high-risk cancers, some surgeons advocate immediate radical cystectomy to maximize the chance of a cure,7 whereas others have promoted conservative therapy with weekly intravesical Bacillus Calmette-Guerin (BCG) immunotherapy and delayed cystectomy if indicated as an effective therapy with the potential for bladder preservation.8 Although different studies have supported the effectiveness of each regimen, to the best of our knowledge, no study to date has assessed the costs associated with the treatment of high-risk T1G3 bladder cancer.
In a previous analysis, we observed that the preferred treatment option (assessing survival and quality of life but not costs) between these 2 strategies depended on patient characteristics, such as age and comorbidities, as well as individual preferences.9 In the current analysis, we updated our model to meet 2 additional objectives. First, we analyzed the costs and cost-effectiveness of each strategy. Second, we explored whether additional research in this area would be a wise investment of resources.
MATERIALS AND METHODS
We compared 2 treatment strategies for the treatment of potent man aged 60 years with a newly diagnosed, high-risk T1G3 carcinoma of the bladder: 1) immediate nerve-sparing radical cystectomy with creation of an orthotopic ileal neobladder and 2) initial conservative therapy, which consisted of intravesical BCG with possible delayed cystectomy. We described health effects in terms of life expectancy with and without quality adjustment and calculated the incremental cost-effectiveness ratio of the 2 interventions, defined as the ratio of marginal costs to marginal health effects (see the equation below).
We discounted both health effects and costs by 3% per annum.10 Our perspective was that of a healthcare payer because of the high direct costs incurred by health payers for bladder cancer treatment.5 TreeAge modeling software was used to build the model (TreeAge Software Inc., Williamstown, Mass).
Model Structure and Validation
We built a Markov model in which the clinical course of a hypothetical cohort of patients was described by discrete health states and transitions between these states.11 The model's time horizon was the lifetime of the patients. The details and structure of the original Markov model have been described elsewhere.9 Briefly, patients in the hypothetical cohort who underwent immediate cystectomy were subject to a risk of postoperative mortality (Fig. 1). Those who survived surgery entered 1 of 8 mutually exclusive health states defined by distinct combinations of the 3 major long-term postcystectomy complications (sexual, genitourinary, and gastrointestinal). Patients with lymph node-positive disease were deemed eligible for adjuvant platinum-based chemotherapy.12
Patients receiving initial intravescial management received 6 weekly instillations of induction BCG followed by 3 weekly maintenance instillations at 3 months, 6 months, 12 months, 18 months, 24 months, and 30 months after diagnosis according to the Southwest Oncology Group protocol.13 We used a cycle length of 6 months, which reflected the intervals when treatment decisions are made for patients with bladder cancer. After induction or maintenance BCG, patients were at risk for disease progression or recurrence. For progression, patients underwent radical cystectomy. Individuals with recurrent T1G3 lesions received a second cycle of BCG induction and maintenance. Those who experienced a T1G3 recurrence either during or after the second cycle of BCG therapy were deemed BCG failures and underwent radical cystectomy. After a successful complete BCG cycle (30 months), patients underwent annual cystoscopic and cytologic surveillance to monitor for recurrence.
At any point in the model, regardless of treatment strategy, patients could develop metastatic disease or could die of unrelated causes based on age-specific mortality rates obtained from national life tables.14 We assumed that all patients who presented with metastases were eligible for platinum-based chemotherapy15 and that the likelihood of developing metastatic disease was based on the tumor's local and lymph node stage.
The survival estimates of our model have been validated previously.9 Briefly, for patients who underwent radical cystectomy, the 5-year overall and disease-specific survival rates generated by the model were 80% and 91%, respectively. The corresponding rates for patients who received intravesical BCG were 81% and 90%, respectively. These values agree with estimates from published bladder cancer studies.13, 16-21
We made several simplifying assumptions in the construction of the model. We did not explicitly model nonadherence during BCG therapy, because published effectiveness values (ie, recurrence and progression rates) already account for nonadherent patients. We also assumed that all recurrences during BCG therapy would be detected at surveillance cystoscopy. Finally, we did not account for extremely rare complications, such as bladder contracture (<0.1%) after BCG,22 because their inclusion would not be likely to influence the model results.
Probabilities and Utilities
Probabilities were obtained from the published medical literature; where possible, probabilities from multiple studies were pooled using the inverse variance method (Table 1). We assumed that transition probabilities (probabilities of moving from 1 health state to another) occurred at a constant rate. Such rates, expressed as the number of events per unit of time, are positive, continuous, and not limited by an upper bound. Accordingly, we used a gamma distribution to describe these rates and calculated the distribution parameters using the method of moments.23 We pooled and described transition rates in the model accordingly. Similarly, we used a beta distribution to pool and describe the probabilities of 1-time events.
Table 1. Model Parameters
Base Case Value (SE)
SE indicates standard error; T1G3, T1, grade 3; GI, gastrointestinal; GU, genitourinary; BCG, Bacillus Calmette-Guerin; TURBT, transurethral resection of bladder tumor.
*Semiannual transition probabilities were modeled assuming that underlying transition rates were constant and followed a gamma distribution. All other parameters were modeled using a beta distribution (for details, see the text). References for model parameters have been published elsewhere (see Kulkami 20079).
Baseline rates of recurrence and progression for low-risk T1G3 tumors during surveillance cystoscopy after 1 cycle of BCG.
Baseline rate of metastatic disease for invasive tumors with lymph node involvement.
Probability of a nonlethal postoperative cystectomy complication
Utilities are quality-of-life weights attached to each model health state. Utilities are anchored at 0 (death) and 1 (best possible health). We calculated quality-adjusted survival in each cycle by multiplying the duration of the cycle, the utility of each state, and the proportion in each state. Overall quality-adjusted survival was calculated by integrating all cycles in the model. Because the medical literature lacks detailed utility data24 for bladder cancer, we extrapolated utility scores from other conditions in which similar health states could be expected. Disutilities (penalties to subtract from the utility of a health state) were used to model short-term procedural complications. We assumed that utility (and disutility) scores were bounded by 0 and 1 and, thus, could be described using a beta distribution (also bounded by 0 and 1).23 When the uncertainty around these estimates was not defined, we applied a relatively large standard error of 20% to each utility and disutility value.
We calculated costs for each health state, including relevant side effects and complications (Table 2). Because healthcare costs have a lower bound of 0 but have no defined upper bound, we assumed that variable costs followed a gamma distribution. For cases in which variances were unavailable, we applied a standard error equivalent to 20% of the base cost estimate. All costs were inflated to 2005 Canadian dollars using an average rate of 3% per annum, which approximates the historic annual Canadian inflation rate.25
SE indicates standard error; GI, gastrointestinal; GU, genitourinary; TURBT, transurethral resection of bladder tumor; BCG, Bacillus Calmette-Guerin.
Costs were incorporated into the probabilistic sensitivity analysis by fitting gamma distributions to reported means and SEs.
The costs for cystectomy and TURBT reflect all inpatient costs incurred during the index surgical admission, including overhead and complications.
Follow-up costs were estimated based on the follow-up strategy postcystectomy, as suggested by Solsona et al.47 In Years 1 through 3, physician visits occur every 3 months; in Years 4 and 5, visit frequency decrease to semiannual visits; and, in Year 6 and beyond, visits occur annually. Urinary cytology and routine blood work are performed at each visit. Abdominal/pelvic computed tomography scans and chest x-rays are obtained semiannually in the first 2 years and annually thereafter.
These are the assumed outpatient medication costs based on a fixed dose.
It was assumed that the cost for adjuvant therapy (4 cycles) would be two-thirds the cost of curative therapy (6 cycles).
Cost data on inpatient and procedure-related visits, including short-term postoperative complications, were obtained from the University Health Network Case Costing Center, a large, high bladder cancer volume tertiary teaching hospital in Toronto, Canada. We similarly calculated costs of long-term operative complications (such as bowel obstruction, ureteral stenosis, and incisional hernia) and incorporated these costs into the model based on their unit cost and the chance that they would occur in the postcystectomy state (Table 2).
Chemotherapy cost data for patients with bladder cancer are not currently available.5 Consequently, costs for chemotherapy were extrapolated from a multinational costing study that assessed the costs of gemcitabine/cisplatin in nonsmall cell lung cancer.26 Because the administration, dosage, and number of cycles of chemotherapy is similar for the treatment of advanced lung cancer and bladder cancer,27-29 we assumed that the chemotherapy-related complications and medical oncology workload between the 2 disease sites would be similar.
The cost of dying from cancer or from other causes was extrapolated from an ongoing comprehensive case-costing study for prostate cancer (unpublished data). In that study, costs for 41,803 patients dying from prostate cancer and an equivalent number of controls dying from other causes were analyzed. We included costs incurred in the final 6 months of life.
Fixed costs included physician service fees and drug costs. These were obtained from the 2005 Ontario Schedule of Benefits30 and the Ontario Drug Benefits Formulary/Comparative Drug Index, 39th edition (2005),31 and from the University Health Network outpatient pharmacy. Although we assumed that unit drug costs were fixed, we allowed for variability in the frequency and duration with which drugs were used, such that overall drug costs were associated with some uncertainty. We applied standard errors of 20% to such estimates. (Table 2).
Probabilistic Sensitivity Analyses
We performed probabilistic sensitivity analysis to address the joint uncertainty of all model parameters simultaneously and to provide a more accurate estimate of the average incremental cost-effectiveness ratio. We performed 1000 second-order Monte-Carlo simulations and reported results in 3 ways. First, we reported mean health outcomes and costs with 95% credible intervals (CrIs), that is, the range covered by 95% of our simulations. Second, we graphed results on a cost-effectiveness acceptability curve, which starts from the assumption that cost-effectiveness ratios are considered attractive if the estimates are less than the amount society would be willing to pay for an additional unit gain in health. We graphed the proportion of simulations that fell below a range of societal willingness-to-pay thresholds.32 Third, we estimated the value of obtaining additional information to help guide this treatment decision using Expected Value of Perfect Information analysis.33 These calculations were performed using a threshold of $50,000 per quality-adjusted life year (QALY) both in the presence and in the absence of quality-of-life weights and were expressed as an average, per-patient amount.
We evaluated several additional scenarios to explore how results might change for different patient subgroups. Specifically, we varied the age and comorbid status of the base case, defining comorbidity as absent (base case), mild, or moderate, corresponding to Charlson comorbidity index scores of 0, 1, or >1, respectively.34 On the basis of published bladder cancer studies, we modeled a proportional increase in overall and postoperative mortality of 1.11 and 1.37, respectively, for patients with mild comorbidity and 1.48 and 1.73, respectively, for patients with moderate comorbidity.34, 35 Because patients with more comorbidities are likely to have longer hospital stays and experience more postoperative complications, the cost associated with major surgery also will increase,36 although these estimates are not known with certainty. We modified the cost of radical cystectomy in the presence of comorbidity by adding costs equivalent to 1 standard deviation ($6124) for mild comorbidity and 2 standard deviations for moderate comorbidity. These values correspond to published costs for systemic-related complications after cystectomy.37
For the base case, the mean survival with immediate cystectomy was 14.61 years (95% CrI, 11.27-16.78 years), and the mean survival with conservative BCG therapy was 13.89 years (95% CrI, 10.33-16.28 years) (Table 3). After accounting for quality of life and discounting, immediate radical cystectomy yielded a quality-adjusted survival of 9.46 QALYs (95% CrI, 3.98-11.90 QALYs), and BCG management yielded 9.39 QALYs (95% CrI, 6.16-11.49 QALYs). Immediate cystectomy was associated with improved survival in 89% of simulations and with improved, discounted, quality-adjusted survival in 65% of simulations. The estimates of incremental health gain were 0.72 life years (CrI, from −0.42 to 2.24 life years) and 0.08 discounted QALYs (95% CrI, from −2.58 to 1.43 discounted QALYs) in favor of immediate cystectomy.
Immediate cystectomy also was associated with lower lifetime discounted costs ($37,600; 95% CrI, $35,000-$40,200) compared with conservative BCG therapy ($42,400; 95% CrI, $34,800-$47,500). Immediate cystectomy was associated with decreased costs in 91% of simulations. The estimates of cost savings were $4800 (95% CrI, $3000-$9600) in favor of immediate cystectomy. Thus, immediate cystectomy was associated with both better health outcomes and lower costs than conservative BCG therapy. In economic terms, this means that such strategies dominate the alternatives and are strongly preferred.
We constructed a cost-effectiveness acceptability curve to illustrate the proportions of simulations (quality-adjusted and discounted) that were cost-effective at various willingness-to-pay thresholds (Fig. 2). At a cost-per-QALY threshold of $50,000, immediate cystectomy was cost-effective in 67% of simulations. At thresholds of $20,000 per and $100,000 per QALY, the probabilities that immediate cystectomy was cost-effective were 70% and 66%, respectively.
Expected Value of Perfect Information Analysis
In our base-case analysis, the expected value of perfect information at a willingness-to-pay threshold of $50,000 per life year gained was $1877 per patient. At a threshold of $50,000 per QALY (ie, with quality-of-life adjustment), the expected value of perfect information was $28,220 per patient. With approximately 65,000 incident bladder tumors diagnosed annually in North America, of which 15%38 (n = 9750) are high-risk T1G3 lesions, perfect information regarding all variables in the model would be valued between $18.3 million and $275 million.
We conducted several scenario analyses in which we modified the age and comorbid status of the base case (Table 4). Regardless of the comorbid status, immediate cystectomy was the dominant strategy for patients aged ≤55 years. At aged ≥70 years, conservative therapy either was dominant or was associated with incremental cost-effectiveness ratios less than $32,700 per QALY. Between ages 60 years and 70 years, the optimal choice depended on the presence of comorbidities, with increasing comorbid burden making conservative therapy more attractive economically.
Incremental Health Gain With Conservative Therapy per QALY
QALY indicates quality adjusted life year; ICER, incremental cost-effectiveness ratio.
Each cell in this table reports the discounted incremental cost and health gains (QALY) comparing conservative therapy with cystectomy. When conservative therapy is associated with increased costs and decreased health gains, cystectomy is dominant. Conversely, conservative therapy is dominant when it is associated with lower costs and improved health effects. ICERs are calculated when both the incremental costs and effects are positive.
All costs were rounded to the nearest $100 and are stated in 2005 Canadian dollars.
Conservative therapy dominant
Conservative therapy dominant
Conservative therapy dominant
Conservative therapy dominant
Conservative therapy dominant
Conservative therapy dominant
Conservative Therapy dominant
Conservative therapy dominant
Conservative therapy dominant
The current results suggest that immediate resection is the preferred treatment for healthy patients aged ≤65 years who have high-risk T1G3 bladder cancer and is associated with better health outcomes and lower costs than conservative BCG management. Our model, which is the first to our knowledge that addresses this question from an economic perspective, has several strengths. First, we used probabilistic sensitivity analysis to reflect joint uncertainty in all input parameters. Second, our model was comprehensive, accounting for all major toxicities associated with both surgery and conservative BCG management. Finally, we were able to model the most cost-effective treatment option for various levels of patient age and comorbid status; thus, our findings are broadly applicable.
To the best of our knowledge, only a few studies published to date have analyzed economic outcomes associated with bladder cancer. These studies have evaluated the treatment of patients with transitional cell cancer,39, 40 analyzed options for screening for bladder cancer,41, 42 or quantified costs associated with bladder cancer follow-up and complications.37, 43, 44 Although those results are not directly comparable with our analysis, these and other45, 46 studies reported cost estimates that were similar to the estimates in our model, suggesting that our single institution's cost estimates may be broadly applicable to other settings.
At ages <60 years or >70 years, our model suggests possibilities for cost savings coupled with improved health outcomes. With increasing age and comorbidity, radical cystectomy yielded higher costs and lower health benefits. The greater costs are attributable to the increased risk of systemic-related complications with more comorbidities, whereas the lower health benefits are attributable to the greater risk of postoperative complications and death in older, sicker patients. For patients between ages 60 years and 70 years, our analyses suggest that cost savings with improved health outcomes, or, at the very least, cost-effective care, is possible when therapy is tailored according to patient age and comorbid status.
Both the cost-effectiveness acceptability curve and the calculation of the expected value of perfect information indicate considerable residual uncertainty regarding our cost-effectiveness estimates. We estimated that an upper limit for perfect information regarding model parameters would be at least $18.3 million (quality-unadjusted) and perhaps as high as $275 million (quality-adjusted). These limits suggest that further research aimed at yielding more precise model inputs would be a worthwhile investment.
We acknowledge several limitations to the current study. First, we made assumptions regarding the timing, treatment, and costs of nonperioperative complications after BCG therapy and cystectomy. However, our use of relatively wide, albeit arbitrary, standard errors for these parameters enabled us to incorporate the uncertainty introduced by this assumption into the model. Second, we used a third-party payer perspective rather than a societal perspective, which generally is preferred.10 The societal perspective may be particularly important for patients who still are working, for whom the incremental indirect costs associated with BCG instillation may be underestimated. Thus, our findings are likely to be conservative in this scenario. Third, we extrapolated the costs of dying of bladder cancer from the costs of dying of hormone-refractory prostate cancer because of similarities in metastatic landing sites (ie, lymph nodes, bone, lung, liver) and complications (such as bone and pelvic pain and bladder outlet obstruction) experienced by patients with both diseases. The costs of bladder cancer chemotherapy were extrapolated from costs of treating lung cancer with gemcitabine and cisplatin, because the regimens for each are quite similar. Although these assumptions have face validity, they have not been tested empirically. Fourth, we made assumptions regarding the treatment of T1G3 recurrences while receiving conservative BCG therapy, assuming a second cycle of induction and maintenance BCG therapy. Although our current model included the option of cystectomy after a T1G3 recurrence, which is a legitimate treatment approach in high-risk patients, sensitivity analyses in our previous work demonstrated that changes in model structure pertaining to the timing of cystectomy after T1G3 recurrence minimally influence health outcomes.9 Fifth, we extrapolated utilities from nonbladder cancer health states because of the absence of utility data in this regard. Although we believe that this approach is reasonable, we acknowledge that it may introduce error into our model. The uncertainty around this estimate can be appreciated by comparing the expected value of perfect information without and with quality adjustment ($18.3 million and $275 million, respectively). Hence, better utility estimates in bladder cancer constitute a priority topic for future research. Finally, the estimates of expected value of perfect information may be too high, because we incorporated wide standard errors for parameters that had uncertain values.
In conclusion, the results of the current study indicated that, for healthy patients aged 60 years with T1G3 bladder cancer, immediate radical cystectomy would provide improved survival and quality-adjusted survival at lower cost. However, for patients aged ≥70 years, the most cost-effective therapy is initial conservative treatment with intravesical BCG. Proper risk stratification can help optimize patient outcomes while decreasing costs to the healthcare system.
Conflict of Interest Disclosures
Dr. Kulkarni is supported by a Canadian Institutes for Health Research research fellowship.