Article 0020.R1/903122

Transcription

1 Journal of Medical Economics Vol. 17, No. 5, 2014, Article 0020.R1/ doi: / All rights reserved: reproduction in whole or part not permitted Original article Quality-adjusted survival with combination nab-paclitaxel þ gemcitabine vs gemcitabine alone in metastatic pancreatic cancer: a Q-TWiST analysis Michele Reni San Raffaele Scientific Institute, Milano, Italy Yin Wan Caitlyn Solem Pharmerit International, Bethesda, MD, USA Scott Whiting Celgene Corporation, Summit, NJ, USA Xiang Ji Marc Botteman Pharmerit International, Bethesda, MD, USA Address for correspondence: Marc Botteman, 4350 East West Highway, Suite 430, Bethesda, MD 20814, USA. Tel: ; Fax: ; mbotteman@pharmerit.com Keywords: Albumin-bound (nab)-paclitaxel Metastatic pancreatic adenocarcinoma Q-TWiST Quality-adjusted survival Pancreatic neoplasms Accepted: 6 March 2014; published online: 21 March 2014 Citation: J Med Econ 2014; 17: Abstract Objectives: To use the Quality-Adjusted Time Without Symptoms or Toxicities (Q-TWiST) methodology to compare the quality-of-life and survival benefits associated with the combination of albumin-bound (nab)-paclitaxel and gemcitabine vs gemcitabine alone in the first-line treatment of metastatic pancreatic adenocarcinoma. Methods: Total survival time through 45 months was partitioned into time before disease progression without toxicity grade 3 (TWiST), time with adverse event grade 3 (TOX), and time of disease progression (REL). Mean Q-TWiST was calculated by multiplying time spent in each health state by its respective utility (i.e., TWiST ¼ 1.00; TOX/REL ¼ 0, 0 1 in sensitivity analyses). Non-parametric bootstrap 95% confidence intervals (CI) were derived to assess the significance of between-treatment differences in TOX, TWiST, REL, and Q-TWiST. A relative gain in Q-TWiST (vs mean overall survival of gemcitabine) of 10% and 15% was defined as clinically important and clearly clinically important, respectively. Results: Patients on nab-paclitaxel þ gemcitabine spent a significantly longer time in every state and experienced significantly greater overall Q-TWiST (þ1.7 months [95% CI ¼ 0.8, 2.7]) than those receiving gemcitabine alone (8.2 months [95% CI ¼ 7.5, 8.9] vs 6.5 months [95% CI ¼ 5.8, 7.0]), assuming base-case utilities of TOX/REL ¼ 0. This Q-TWiST gain ranged from 1.0 month (95% CI ¼, 1.9), when REL/TOX utilities were both 0, to 2.5 months (95% CI ¼ 1.3, 3.7), when REL/TOX utilities were both 1. Relative gains in Q-TWiST were 21% in favor of nab-paclitaxel þ gemcitabine in the base case, and ranged from 12 30% in Copyright 2014 Informa UK Limited Not for Sale or Commercial Distribution sensitivity analyses. Unauthorized use prohibited. Authorised users can download, display, view and print a single copy for personal use Conclusions: There are limitations to Q-TWiST analyses, e.g., imprecision when defining duration/severity of TOX and lack of prospective collection of utilities. This analysis addressed these issues via sensitivity analyses and conservative assumptions to show that nab-paclitaxel þ gemcitabine results in statistically significant and clinically important gains in quality-adjusted survival, when compared to gemcitabine alone, in treatment-naive metastatic pancreatic adenocarcinoma patients. Background Pancreatic adenocarcinoma is the fourth leading cause of death among cancer patients in the US, with an estimated 37,390 deaths from the disease and an estimated 43,920 new cases in The prognosis of pancreatic adenocarcinoma is poor, with a 5-year survival rate of 6% in Europe and the US Quality-adjusted survival in metastatic pancreatic cancer Reni et al Informa UK Ltd

2 Gemcitabine (difluorodeoxycytidine) is indicated as a first-line treatment of patients with unresectable locally advanced or metastatic pancreatic adenocarcinoma 3. Since its approval, efforts have been focused on developing gemcitabine combination therapy to improve the clinical outcomes for patients with this cancer. Nab-paclitaxel is an albumin-bound formulation of paclitaxel, utilizing albumin binding proteins to increase intra-tumoral paclitaxel accumulation 4. The use of nab-paclitaxel eliminates the need for toxic solvents (e.g., cremophor), allowing for the delivery of higher doses and shorter infusion times compared with solvent-based paclitaxel. In the Phase 3 Metastatic Pancreatic Adenocarcinoma Clinical Trial (MPACT), the addition of nab-paclitaxel to gemcitabine as the first-line treatment significantly improved median overall survival (OS; 8.5 months vs 6.7 months, hazard ratio for death ¼ 0.72; 95% confidence interval [CI] ¼ 2, 0.83; p50.001) and progression-free survival (PFS) (5.5 months vs 3.7 months, hazard ratio for disease progression or death ¼ 9; 95% CI ¼ 8, 0.82; p50.001) in first-line metastatic pancreatic adenocarcinoma patients compared with gemcitabine alone 5. Although this median OS gain of 1.8 months appears modest in absolute terms, this gain corresponds to a 27% relative improvement due to the poor prognosis in this disease. A more recent, updated analysis showed the median OS gain to have increased to 2.1 months, or a 32% relative improvement 6. In order to more comprehensively weigh the trade-offs of increased PFS and OS in the presence of more adverse events (AEs), which are to be expected with combination chemotherapy, a quality-adjusted time without symptoms of disease progression or toxicity (Quality-adjusted Time Without Symptoms of disease progression or Toxicity of treatment [Q-TWiST]) may be utilized 7. By partitioning survival time into a series of health states, the Q-TWiST method combines quantity and quality-of-life into a single measure. Since its first application in breast cancer, the Q-TWiST method has been widely applied to other cancers and disease conditions over the past decade 7,8. However, the Q-TWiST methodology has not been applied in metastatic pancreatic cancer (MPC). Thus, this study aims to examine the quality-adjusted survival benefits associated with nab-paclitaxel þ gemcitabine vs gemcitabine alone by applying the Q-TWiST approach to the MPACT trial data. Method Study design This study was based on the data from the open-label, randomized, international, multi-center, Phase 3 study (MPACT) designed to compare nab-paclitaxel in combination with gemcitabine vs gemcitabine monotherapy. In this trial, 861 patients with MPC were randomized at a 1:1 ratio to receive either nab-paclitaxel þ gemcitabine or gemcitabine alone. Key patient eligibility criteria included no radiotherapy, surgery, chemotherapy, or investigational therapy for the treatment of metastatic disease before enrollment and no history of interstitial lung disease 5. Treatments of nab-paclitaxel (125 mg/m 2 ) þ gemcitabine (1000 mg/m 2 ) were intravenously administered every 3 weeks followed by 1 week off treatments over a 28-day cycle, while gemcitabine treatments (1000 mg/m 2 alone) were given weekly for 7 weeks followed with 1 week off treatment for Cycle 1 only, and subsequently administered weekly for 3 weeks over a 28-day cycle in Cycle 2 and onwards 5. The primary efficacy end-point was the OS of patients treated with nab-paclitaxel in combination with gemcitabine compared with the OS of patients treated with gemcitabine alone. Other end-points include PFS and tumor response 5. Patients were followed for survival until death or study closure (up to 45 months from randomization). Statistical analyses To apply the Q-TWiST methodology, health states were partitioned into the following periods: time without symptoms of disease progression or toxicities (TWiST) for non- TOX time prior to disease progression, toxicity (TOX) during which patients experienced grade 3 (NCI Common Terminology Criteria for Adverse Events [CTCAE] grade) 9 AEs due to any cause, and relapse (REL) after disease progression. For grade 3 AEs that were classified as ongoing at patients last follow-up observation or disease progression, the period from AE onset through disease progression was used as a surrogate measure of TOX duration. This strategy was designed to be conservative and to capture all time spent with toxicity. To test the sensitivity of our analysis to this assumption, we also conducted an analysis in which the ongoing grade 3 AEs at last follow-up observation or at disease progression were assumed to be resolved when the next chemotherapy cycle was initiated. For all patients, the time spent with AEs was summed for each patient and a day with multiple events was only counted once. The duration of TOX was always capped at disease progression, and any AEs that occurred after disease progression were not counted towards TOX time. The total number of TOX days was grouped together during a patient s treatment, regardless of when TOX days prior to relapse actually occurred. For each treatment group, PFS, time with toxicities, and OS were estimated using the Kaplan-Meier method 10. Since the cumulative area under a Kaplan- Meier survival curve represents the mean time of the responding health state, taking into account censoring and differential follow-up, the mean duration of TWiST! 2014 Informa UK Ltd Quality-adjusted survival in metastatic pancreatic cancer Reni et al. 339

3 was calculated as the difference in area under PFS and TOX curves, and the mean duration of REL was calculated as the difference in area under OS and PFS curves. Within the base case, mean durations were measured through month 45. The Q-TWiST was calculated as the sum of the means for the three health states mentioned above, with each state weighted by its respective level of utility (Q-TWiST ¼ U TWiST TWiST þ U TOX TOX þ U REL REL), where U TWIST,U TOX, and U REL represent the utilities of TWiST, TOX, and REL, respectively. TWiST was considered to have utility equal to 1.00, representing the best possible quality-of-life for a patient with MPC. While a threshold analysis was used to vary the U TOX and U REL values between 0 1, the base case considered a utility score of 0 for both TOX and REL (Q-TWiST ¼ 1.00 TWiST þ 0 TOX þ 0 REL), which is customary in these analyses Ninety-five percent CIs around each health state and Q-TWiST were calculated using a non-parametric bootstrap method 16. In threshold analyses, the Q-TWiST was calculated over a wide range of U TOX and U REL, from a minimum of 0.00 (days were not counted towards total Q-TWiST) to a maximum of 1.00 (days fully counted). Consistent with other Q-TWiST studies, the utility of TWiST was assumed as 1.00 for both base-case scenario and threshold analysis. However, previous research indicated other values that patients assigned to the time in TWiST, and the relative values in TOX and REL varied widely as well 17,18. To further understand how the utility weight of TWiST affects the study results, the following utility weights were used in two alternate sensitivity analysis scenarios: (1) U TWiST ¼ U TOX ¼ U REL ¼ 0; (2) U TWiST ¼ 0.80, U TOX ¼ 5, and U REL ¼ The first case represented a very conservative scenario where the utility of TWiST is only half of what was used in the main analysis. In the second case, the utilities of TWiST and REL were obtained from prospectively collected trial data of patients with advanced pancreatic cancer 19. The utility of TOX was set as 5 according to several studies evaluating the decrement in the utility of a series of AEs 4,6,15,20,21. In the base case, all outcomes were measures using cumulative data through month 45. In additional sensitivity analyses, the difference in Q-TWiST between treatments was assessed using data truncated at 3, 6, 9, 12, 24, and 36 months to assess how this difference in Q-TWiST changed under different analytical horizons. Results Study population As previously reported, baseline characteristics of patients enrolled were well-balanced between treatment arms (nab-paclitaxel þ gemcitabine (n ¼ 431) vs gemcitabine alone (n ¼ 430)) in the intent-to-treat (ITT) population (Table 1) 5. In the ITT population, the mean age was 62.2 (range ¼ 27 88) years, 58.3% were male, and 39.6% had a Karnofsky Performance Score (KPS) of 70 80, while 60% had a KPS of The majority (84.2%) of the patients had liver metastases, and 43.1% of the patients had head as the primary pancreatic location. Among the treated population (nab-paclitaxel þ gemcitabine: n ¼ 421; gemcitabine: n ¼ 402), patients treated with nab-paclitaxel þ gemcitabine had more all-cause grade 3 AEs compared to gemcitabine alone (374 [88.8%] vs 303 [75.4%], p50.001), as well as treatment-related and treatment-emergent grade 3 AEs (325 [77.2%] vs 203 [5%], p50.001). The most frequent AEs in both treatment arms were fatigue (54% of 823 treated patients), alopecia (50%), and nausea (49%). Patients treated with nab-paclitaxel þ gemcitabine had more grade 3 neutropenia (37.8% vs 26.5%) and leukopenia (3% vs 16.2%). The incidences of grade 3 thrombocytopenia, anemia, and febrile neutropenia were similar between the two groups. The proportions of serious or fatal AEs were also similar between the treatment groups 5. Additional details of the safety data for the MPACT trial have been published elsewhere 5. Duration of time spent in each health state Kaplan-Meier survival curves for PFS, TOX, and OS and the respective partitioning of survival time into TWiST, TOX, and REL are shown in Figure 1. As Table 2 indicates, the mean duration of time spent in TWiST, TOX, and REL was significantly longer among patients receiving nab-paclitaxel þ gemcitabine therapy vs gemcitabine alone. Of note, mean PFS and OS were also significantly longer for those receiving nab-paclitaxel þ gemcitabine therapy vs gemcitabine alone. Q-TWiST When utility weights for the TOX and REL health states were set equal to 0 (base case), there was a statistically significant 1.7-month (95% CI ¼ 0.8, 2.7) difference in quality-adjusted survival favoring nab-paclitaxel þ gemcitabine over gemcitabine alone (Table 3). As expected, this number is lower than the OS differential due to a utility rating of less than 1 given to TOX and REL. When varying the utility for TOX and REL within the utility threshold analysis, the difference in Q-TWiST (nab-paclitaxel þ gemcitabine vs gemcitabine alone) is always positive and statistically significant, ranging from months (Figure 2). In other words, nab-paclitaxel þ gemcitabine had higher quality-adjusted survival times (as measured by Q-TWiST) than gemcitabine alone across 340 Quality-adjusted survival in metastatic pancreatic cancer Reni et al Informa UK Ltd

4 Table 1. Characteristics of the patients at baseline*. Variables Nab-paclitaxel þ gemcitabine (n ¼ 431) Gemcitabine (n ¼ 430) All patients (n ¼ 861) Age (years), mean SD Sex, n (%), male 245 (56.8) 257 (59.8) 502 (58.3) Region, n (%) Australia 61 (14.2) 59 (13.7) 120 (13.9) Eastern Europe 64 (14.8) 62 (14.4) 126 (14.6) Western Europe 38 (8.8) 38 (8.8) 76 (8.8) North America 268 (62.2) 271 (63.0) 539 (62.6) Race or ethnic group, n (%) Asian 8 (1.9) 9 (2.1) 17 (2.0) Black 16 (3.7) 16 (3.7) 32 (3.7) White 378 (87.7) 375 (87.2) 753 (87.5) Hispanic 25 (5.8) 26 (6.0) 51 (5.9) Other 4 () 4 () 8 () Karnofsky performance score, n/total n (%) /429 (57.8) 268/429 (62.5) 516/858 (6) /429 (41.7) 161/429 (37.5) 340/858 (39.6) 570 2/429 () 0/429 2/858 (0.2) Time from primary diagnosis to randomization (months), mean SD Pancreatic tumor location, n (%) Head 191 (44.3) 180 (41.9) 371 (43.1) Body 132 (3) 136 (31.6) 268 (31.1) Tail 105 (24.4) 110 (25.6) 215 (25.0) Unknown 3 (0.7) 4 () 7 (0.8) Site of metastatic disease, n (%) Liver 365 (84.7) 360 (83.7) 725 (84.2) Lung 153 (35.5) 184 (42.8) 337 (39.1) Peritoneum 19 (4.4) 10 (2.3) 29 (3.4) No. of metastatic sites, n (%) 1 33 (7.7) 21 (4.9) 54 (6.3) (46.9) 206 (47.9) 408 (47.4) (31.6) 140 (32.6) 276 (32.1) (13.9) 63 (14.7) 123 (14.3) Level of carbohydrate antigen 19-9, n/total n (%) Normal 60/379 (15.8) 56/371 (15.1) 116/750 (15.5) ULN to559 ULN 122/379 (32.2) 120/371 (32.3) 242/371 (32.3) 59 ULN 197/379 (52.0) 195/371 (52.6) 392/371 (52.3) *There were no significant between-group differences at baseline. SD, standard deviation; ULN, upper limit of normal. the entire range of utility combinations for TOX and REL. Relative improvement in Q-TWiST associated with nab-paclitaxel þ gemcitabine was 20.7% using base case utilities, and ranged from % improvement over the gemcitabine group (Figure 3). Sub-group analyses The difference in Q-TWiST favored nab-paclitaxel þ gemcitabine (vs gemcitabine alone) across most of the pre-specified sub-groups (Figure 4). As compared to gemcitabine, patients treated with nab-paclitaxel þ gemcitabine had a significantly greater Q-TWiST in the following sub-groups: patients 565 years age, male patients, patients with KPS 70 80, KPS , patients with head as primary tumor location, patients with liver metastases, patients with 1 or43 metastatic sites, patients with CA19-9 level 59-times the upper limit of the normal range, and patients in North America or Australia. Sensitivity analyses When alternative utility values for TWiST were used, the difference in Q-TWiST for nab-paclitaxel þ gemcitabine vs gemcitabine alone was 1.2 (95% CI ¼ 0.7, 1.8) months when the utilities of TWiST, TOX, and REL were set at 0. When utilities from pancreatic cancer trial data of 0.80 for TWiST, 0.75 for REL, and 5 for TOX were applied, nab-paclitaxel þ gemcitabine was associated with a statistically significant improvement in Q-TWiST compared to gemcitabine alone of 1.9 (95% CI ¼, 2.8) months. When the TOX time was capped at the initiation of the next chemotherapy cycle for those patients whose grade 3 end date was not provided (which affected 55 of 194 patients without grade 3 end date), the Q-TWiST benefit of combination therapy increased slightly to 1.8 (95% CI ¼, 2.7) months at the 45-month follow-up. Figure 5 plots the differences in base case Q-TWiST when truncating the data at various time points.! 2014 Informa UK Ltd Quality-adjusted survival in metastatic pancreatic cancer Reni et al. 341

5 (a) Proportion surviving (b) Proportion surviving TWiST TOX TWiST TOX REL REL TOX PFS OS Months from Randomization TOX PFS OS Months from Randomization Figure 1. Kaplan-Meier survival curves showing the mean times in TOX, TWiST, and REL states. (a) Nab-paclitaxel þ gemcitabine; (b) Gemcitabine. Table 2. Mean duration of health states through 45 months, by treatment. Survival time in months Nab-paclitaxel þ gemcitabine, mean time (95% CI) Gemcitabine, mean time (95% CI) Difference, mean time (95% CI) TOX 1.2 (1.0, 1.4) 0.7 (, 0.8) (0.4, 0.8) TWiST 5.5 (4.8, 6.1) 4.5 (3.9, 5.1) 1.0 (, 1.9) REL 4.2 (3.4, 4.9) 3.2 (2.7, 3.8) (0.01, 1.8) PFS 6.7 (6.0, 7.4) 5.2 (4.6, 5.8) 1.6 (, 2.5) OS 1 (9.9, 11.8) 8.4 (7.7, 9.0) 2.5 (1.3, 3.7) CI, confidence interval; OS, overall survival; PFS, progression-free survival; REL, time to disease progression/relapse; TOX, time during toxicity; TWiST, time without symptoms of disease progression or toxicity of treatment. While the primary analysis used all data through 45 months, the Q-TWiST was initially statistically significantly better for the gemcitabine-alone arm when only data through 3 months was considered (difference in Q-TWiST of months, 95% CI ¼ 0.2, 0.04). However, the direction of this effect changed when including all data through 6 months. Using data through month 9, the difference in favor of nab-paclitaxel þ gemcitabine was significant. At month 12, the difference was months (95% CI ¼ 0.4, 1.4). Using cumulative data through 24 and 45 months of follow-up, the differences in the base case Q-TWiST increased slightly when including more person-time, with Q-TWiST differences of 1.5 (95% CI ¼ 0.7, 2.3) through 24 months, 1.6 (95% CI ¼ 0.8, 2.5) through 36 months and 1.7 (95% CI ¼ 0.8, 2.7) through 45 months. As illustrated in Figure 5, the 95% CIs widened over time, indicating greater variability as more time was included for measurement of follow-up, but more patients were censored, leading to less precision associated with the Q-TWiST estimate. The small differences in Q-TWiST estimates at later time points are reflective of the fact that the majority of patients had died by 24 months (93.0% overall) and, therefore, were not contributing any more REL or TWiST time, and, hence, had little effect on the Q-TWiST for either group. Discussion To the best of our knowledge, this study is the first to utilize Q-TWiST methodology in a metastatic pancreatic adenocarcinoma phase 3 study. In the current analysis, patients treated with nab-paclitaxel þ gemcitabine had significantly greater Q-TWiST than those receiving gemcitabine alone. The difference in Q-TWiST for patients treated with nab-paclitaxel þ gemcitabine represented a statistically significant quality-adjusted gain of 1.7 months at base case. Additionally, nab-paclitaxel þ gemcitabine was Table 3. Mean duration of Q-TWiST through 45 months, by treatment. Q-TWiST in months Nab-paclitaxel þ gemcitabine, mean time (95% CI) Gemcitabine, mean time (95% CI) Difference, mean time (95% CI) Improvement, %* U TOX ¼ 0, U REL ¼ (4.8, 6.1) 4.5 (3.9, 5.1) 1.0 (, 1.9) 11.8 U TOX ¼ 0, U REL ¼ 7.6 (6.9, 8.3) 6.1 (5.5, 6.7) 1.5 (, 2.4) 17.3 U TOX ¼ 0, U REL ¼ (8.7, 1) 7.7 (7.0, 8.4) 1.9 (0.7, 3.0) 22.8 U TOX ¼, U REL ¼ (5.4, 6.7) 4.8 (4.2, 5.4) 1.3 (0.3, 2.2) 15.2 U TOX ¼, U REL ¼ 8.2 (7.5, 8.9) 6.5 (5.8, 7.0) 1.7 (0.8, 2.7) 20.7 U TOX ¼, U REL ¼ (9.3, 11.2) 8.1 (7.4, 8.7) 2.2 (1.0, 3.4) 26.1 U TOX ¼ 1.0, U REL ¼ (6.0, 7.4) 5.2 (4.6, 5.8) 1.6 (, 2.5) 18.5 U TOX ¼ 1.0, U REL ¼ 8.8 (8.0, 9.5) 6.8 (6.2, 7.3) 2.0 (1.1, 3.0) 24.0 U TOX ¼ 1.0, U REL ¼ (9.9, 11.8) 8.4 (7.7, 9.0) 2.5 (1.3, 3.7) 29.5 *Percentage improvement ¼ Difference, Mean Time/Gemcitabine Mean OS of 8.4 months as described in Revicki et al. 8 CI, confidence interval; U TOX, utility of time with toxicity; U REL, utility of time to relapse/disease progression; TWiST, time without symptoms of disease progression or toxicity of treatment. 342 Quality-adjusted survival in metastatic pancreatic cancer Reni et al Informa UK Ltd

6 Utility of TOX Utility of REL Figure 2. Utility threshold plot through 45 months. The y-axis represents the utility for toxicity (U TOX ) time and the x-axis represents the utility for time after disease progression (U REL ). Both vary from 0 1, while the utility of TWiST is fixed at 1. The diagonal bands of different colors or shading represent increasing Q-TWiST gains (from the lower left corner to the upper right corner of the plot). To understand the Q-TWiST gain associated with a given combination of U TOX and U REL, one must select the corresponding values of U TOX and U REL on the y-axis and x-axis, respectively. The intersection of these two values inside the plot indicates to which band of Q- TWiST gain the results from this combination belong to. For instance, the base case use U TOX ¼ and U REL ¼, which intersect in the Q- TWiST gain and are consistent with the point estimate of 1.7 months. Note that all Q-TWiST gains in this plot are statistically significant. found to be statistically superior in improving Q-TWiST vs gemcitabine alone across most of the pre-specified sub-groups. While relatively modest in absolute terms, the statistically significant average Q-TWiST gain of 1.7 months constituted a 20.7% relative improvement in quality-of-life adjusted OS as compared to treatment with gemcitabine alone. While a comparison of Q-TWiST gains across different cancers has inherent limitations related to obvious difference in tumor types, the above relative improvement of 20.7% may be placed in the broader context of other published Q-TWiST results across other cancers. Specifically, a systematic review of Q-TWiST studies in oncology has indicated that 85% of estimates of published Q-TWiST gains are below 19% 8. This review also concluded that relative gains in Q-TWiST 10% and 15% should be considered to represent a clinically important and clearly clinically important difference, respectively. Thus, the benefits in Q-TWiST reported herein nearly 21% on average, and ranging from a minimum of 12% to a maximum of 30% in the sensitivity analysis can be considered at least clearly important 8. Of note, these Q-TWiST gains in favor of combination nab-paclitaxel þ gemcitabine therapy are Utility of TOX Utility of REL 27%-30% 24%-27% 21%-24% 18%-21% 15%-18% 12%-15% Figure 3. Plot of relative Q-TWiST gain through 45 months. The y-axis represents the utility for toxicity (U TOX ) time and the x-axis represents the utility for time after disease progression (U REL ). Both vary from 0 1, while the utility of TWiST is fixed at 1. The diagonal bands of different color represent increasing relative Q-TWiST gains (from the lower left corner to the upper right corner of the plot). The relative Q-TWiST gain is the percentage obtained by dividing the absolute Q-TWiST gain by the mean survival in the gemcitabine group. To understand relative Q-TWiST gain associated with a given combination of U TOX and U REL, one must select the corresponding values of U TOX and U REL on the y-axis and x-axis, respectively. The intersection of these two values inside the plot indicates to which band of relative Q-TWiST gain the results from this combination belong to. For instance, the base case use U TOX ¼ and U REL ¼, which intersect in the % relative Q-TWiST gain and is consistent with the point estimate of 20.2%. not surprising given the statistically significant gains in OS (8.5 months vs 6.7 months, hazard ratio for death ¼ 0.72; 95% CI ¼ 2, 0.83; p50.001) and PFS (5.5 months vs 3.7 months, hazard ratio for disease progression or death ¼ 9; 95% CI ¼ 8, 0.82; p50.001) in the presence of limited increases in toxicity associated with the use of combination nab-paclitaxel þ gemcitabine relative to gemcitabine alone 5. Quality-of-life end-points often have not been reported in clinical trials comparing treatments for patients with advanced pancreatic cancer and, thus, the knowledge regarding quality-of-life outcomes data in patients with MPC is limited. One study 22 comparing FOLFIRINOX vs gemcitabine found a significant improvement in quality-of-life outcome measured by Global Health Status and Quality of Life scale 23. However, this study population was selected based on a series of restrictive selection criteria, such as Eastern Cooperative Oncology Group performance status 0 1, metastatic cancer mainly of the pancreatic body, and age 576 years, which limits the ability to generalize these results to different patient groups. Another study evaluated quality-adjusted life months (survival in months average monthly KPS/100) among 46 patients! 2014 Informa UK Ltd Quality-adjusted survival in metastatic pancreatic cancer Reni et al. 343

7 Figure 4. Differences in Q-TWiST (base case) among pre-specified sub-groups through 45 months. *The percentage for the level of CA19-9 was calculated based on the number of total patients with CA19-9 data: 116/750 ¼ 15%, 242/750 ¼ 32%, and 392/750 ¼ 52%. No. of patients (%), number of patients (%); Mean diff (95% CI), mean difference (95% confidence interval); GEM, gemcitabine; Nab-Paclitaxel þ GEM, albumin-bound paclitaxel þ gemcitabine; CA 19-9, carbohydrate antigen Difference in Q -TWiST (mean, 95% C I) with MPC, and found that the combination therapy (gemcitabine plus cisplatin) had shorter quality-adjusted life months vs gemcitabine alone (3.8 vs 5.6, p50.001) 24. However, the quality-of-life outcome was restricted to Cut-off points of follow-up duration in months Figure 5. Difference in the Q-TWiST at various follow-up durations. CI, confidence interval. Note: the survival rate at 3-, 6-, 9-, 12-, 24-, 36, and 45-months between treatment groups (nab-paclitaxel-gemcitabine vs gemcitabine alone) were 83% vs 80%, p ¼ 0.324, 66% vs 56%, p ¼ 0.001, 49% vs 37%, p50.001, 36% vs 23%, p50.001, 15% vs 10%, p50.001, 12% vs 8%, p ¼ 0.074, 12% vs 8%, p ¼ 0.092, respectively. 1.7 the information contained in the KPS, a single-scale measurement of physical functions. It should be noted that, when non-q-twist quality-of-life analyses are conducted, they are often conducted on the sub-set of patients who are progression-free survivors. As such, these analyses do not integrate the experience of all patients, including those who have died. In contrast, the Q-TWiST analysis uses a single measure across all patient types, with the utility of death implicitly valued at 0 on a scale from 0 1. Hence, the Q-TWiST provides a more comprehensive overview of the quality-of-life impact of cancer and its therapies. The Q-TWiST analysis provides a way to examine progression, OS, and toxicity as a single metric 25. The valuation of utility weights is a challenging issue in defining quality-adjusted life, as many clinical trials do not collect utility values prospectively. However, one key advantage of the Q-TWiST is precisely the ability to choose utility values for REL and TOX that are specific to the individual (often a clinician) who is valuing the relative trade-off between time in TOX and time in REL. While an exhaustive review of the Q-TWiST literature on this topic is beyond the scope of the present 344 Quality-adjusted survival in metastatic pancreatic cancer Reni et al Informa UK Ltd

8 research, it is clear that the REL ¼ TOX ¼ 0 is very frequently utilized as a base case, not because there is a belief that these should always be the values used, but to provide a mid-point for interpretation of Q-TWiST values We chose to follow this convention within our manuscript for the base case to allow for ease of comparability to other Q-TWiST studies. In addition, the present analysis also reported the Q-TWiST via a threshold analysis (Figures 2 and 3) as a means to visually assess the Q-TWiST gains associated with a full range of utilities for REL and TOX. Thus, by providing results over a range of utilities, the threshold analysis overcomes the inherent uncertainty associated with the selection of utility weights and allows for the application of these results given different preferences for time spent with post-progression and time with AEs. Of note, in the present threshold analysis, for all values of U TOX and U REL between 0 1, the combination therapy yielded better quality-adjusted survival than monotherapy, which confirmed the robustness of our study conclusion. Furthermore, the sensitivity analysis using alternative utility weights (less than 1) of TWiST also showed favorable results for nab-paclitaxel þ gemcitabine over gemcitabine alone. The sensitivity analyses confirmed that the same conclusion can be achieved when the follow-up duration is restricted to no less than 9 months. In particular, a statistically significant Q-TWiST gain of 1.5 months was observed when the analysis is restricted to 24 months of followup. The Q-TWiST gains in favor of nab-paclitaxel þ gemcitabine improve over a longer follow-up period (and are actually statistically significantly in favor of gemcitabine alone at 3 months), because the longer survival (both in TWiST and REL) benefits of nabpaclitaxel þ gemcitabine accrued over time while all TOX time was counted at the beginning of follow-up. Since nab-paclitaxel þ gemcitabine had more TOX time, during shorter time intervals the relative amount of TWiST to TOX time would appear shorter. Thus, while gains continue to accrue until at least month 45, most of the gains are observed after 24 months of follow-up. Our study has several limitations. First, patient-derived data regarding the utility weights were not collected prospectively. For this reason, the base case utilities for TOX and REL were both set to 0, consistent with numerous previous studies. A threshold analysis was applied to provide results over a wide range of utilities. Additional sensitivity analyses were also conducted to assess how the Q-TWiST value would change under alternate assumptions. Second as demonstrated in a separate sensitivity analysis in which the TOX time was capped at the initiation of the next chemotherapy cycle for those patients whose grade 3 end date was not provided the main analysis likely over-estimated the total time spent in the TOX health state for AEs grade 3 because, for 23% of patients, the exact timing of improvement of the AEs from grade 3 to grade53 (which defines in the present analysis whether TOX continues) was generally not available. Hence, in this analysis, grade 3 AEs were assumed to continue if the event was not explicitly reported as resolved or progression occurred. As corollary, the TWiST time may have been correspondingly underestimated. Third, the same utility was assumed regardless of the severity and type of AE (provided grade was 3). This assumption may not be strictly accurate, as AEs of different severity or different types of AEs (e.g., thrombocytopenia and neutropenia) could be associated with different effects on utility. In the same manner, certain drugrelated effects may not be reportable as AEs but could affect patients quality-of-life and would not be captured by the Q-TWiST approach 26. Although there is a limitation on the accuracy of utilities assigned by patients, our study provided a range of the benefits in Q-TWiST between the extreme cases using the threshold analysis. Finally, it is important to recognize that the present analysis was designed to assess the quality-of-life and survival benefits associated with nab-paclitaxel þ gemcitabine and gemcitabine monotherapy using a Q-TWiST framework. As such, it did not consider any cost implications associated with the use of these therapies. Since nabpaclitaxel is an add-on therapy with a longer duration of treatment, it will have additional cost, but such an assessment should be the focus of separate analyses conducted according to accepted recommendations in the respective country of assessment. Conclusion Based on the findings of this Q-TWiST analysis, it may be concluded that the use of nab-paclitaxel þ gemcitabine results in a statistically significant benefit in Q-TWiST when compared to gemcitabine alone in patients with metastatic pancreatic adenocarcinoma. For patients treated with nab-paclitaxel þ gemcitabine, the longer duration of toxicity is offset by longer time spent in PFS. The magnitude of the clinical benefit associated with nab-paclitaxel þ gemcitabine depends on the value that an individual patient assigns to time with toxicity and time following the onset of disease progression. However, across all combinations of utility values for TOX and REL used in the threshold and sensitivity analyses, nabpaclitaxel þ gemcitabine resulted in significantly better Q-TWiST than gemcitabine treatment alone. Furthermore, these gains were relatively large when compared to other Q-TWiST assessments across other cancers, and may be considered at least clinically important if not clearly clinically important.! 2014 Informa UK Ltd Quality-adjusted survival in metastatic pancreatic cancer Reni et al. 345

9 Transparency Declaration of funding This study was sponsored in part by Celgene Corporation, Summit, NJ, which manufactures and commercializes nabpaclitaxel. Besides funding, Celgene Corporation provided access to the patient-level data of the MPACT trial to conduct the present analysis. Declaration of financial/other relationships Dr Michele Reni, MD has disclosed that he has received grants from and is an advisor to Celgene, and is a physician at the San Raffaele Scientific Institute, Milan. Yin Wan, Caitlyn Solem, Ji Xiang, and Marc Botteman are employees of Pharmerit International, an independent contract research organization that received research funding from Celgene. Scott Whiting is an employee of Celgene Corporation. JME peer reviewers on this manuscript have received an honorarium from JME for their review work, but have no other relevant financial relationships to disclose. Acknowledgments The authors wish to thank anonymous reviewers for providing valuable comments on this manuscript. The authors thank Joan Hudson from ProEd Communications Inc. for providing compensated editorial support on the manuscript. References 1. American Cancer Society. Cancer Facts & Figures Atlanta, GA: American Cancer Society, Garcia M, Jemal A, Ward E, et al. Global cancer facts & figures Atlanta, GA: American Cancer Society, Burris 3rd HA,, Moore MJ, Andersen J, et al. Improvements in survival and clinical benefit with gemcitabine as first-line therapy for patients with advanced pancreas cancer: a randomized trial. J Clin Oncol 1997;15: Tolley K, Goad C, Yi Y, et al. Utility elicitation study in the UK general public for late-stage chronic lymphocytic leukaemia. Eur J Health Econ 2010;14: Von Hoff DD, Ervin T, Arena FP, et al. Increased survival in pancreatic cancer with nab-paclitaxel plus gemcitabine. N Engl J Med 2013;369: Goldstein D, El Maraghi RH, Hammel P, et al. Updated survival from a randomized phase III trial (MPACT) of nab-paclitaxel plus gemcitabine versus gemcitabine alone for patients (pts) with metastatic adenocarcinoma of the pancreas (oral presentation at 2014 Gastrointestinal Cancers Symposium). J Clin Oncol 2014;32(3 Suppl):abstract Goldhirsch A, Gelber RD, Simes RJ, et al. Costs and benefits of adjuvant therapy in breast cancer: a quality-adjusted survival analysis. J Clin Oncol 1989;7: Revicki DA, Feeny D, Hunt TL, et al. Analyzing oncology clinical trial data using the Q-TWiST method: clinical importance and sources for health state preference data. Qual Life Res 2006;15: Trotti A, Colevas AD, Setser A, et al. CTCAE v3. 0: development of a comprehensive grading system for the adverse effects of cancer treatment. Sem Radiat Oncol 2003;13: Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Statist Assoc 1958;53: Cole BF, Gelber RD, Goldhirsch A. Cox regression models for quality adjusted survival analysis. Stat Med 1993;12: Caudill-Slosberg MA, Schwartz LM, et al. Office visits and analgesic prescriptions for musculoskeletal pain in US: 1980 vs Pain 2004;109: Sullivan MD, Edlund MJ, Fan M-Y, et al. Trends in use of opioids for noncancer pain conditions in commercial and Medicaid insurance plans: the TROUP study. Pain 2008;138: Farrar JT, Young Jr JP, LaMoreaux L, et al. Clinical importance of changes in chronic pain intensity measured on an 11-point numerical pain rating scale. Pain 2001;94: Nafees B, Stafford M, Gavriel S, et al. Health state utilities for non small cell lung cancer. Health Qual Life Outcomes 2008;6: Glasziou P, Simes R, Gelber R. Quality adjusted survival analysis. Stat Med 1990;9: Zee B, Cole B, Li T, et al. Quality-adjusted time without symptoms or toxicity analysis of interferon maintenance in multiple myeloma. J Clin Oncol 1998;16: Kilbridge KL, Cole BF, Kirkwood JM, et al. Quality-of-life adjusted survival analysis of high-dose adjuvant interferon alfa-2b for high-risk melanoma patients using Intergroup clinical trial data. J Clin Oncol 2002;20: Romanus D, Kindler HL, Archer L, et al. Does health-related quality of life improve for advanced pancreatic cancer patients who respond to gemcitabine? Analysis of a randomized phase III trial of the cancer and leukemia group B (CALGB 80303). J Pain Symptom Manage 2012;43: Lloyd A, Nafees B, Narewska J, et al. Health state utilities for metastatic breast cancer. Br J Cancer 2006;95: Swinburn P, Lloyd A, Nathan P, et al. Elicitation of health state utilities in metastatic renal cell carcinoma. Curr Med Res Opin 2010;26: Conroy T, Desseigne F, Ychou M, et al. FOLFIRINOX versus gemcitabine for metastatic pancreatic cancer. N Engl J Med 2011;364: Labianca R, Merelli B, Mosconi S. Treatment of advanced pancreatic cancer. Ann Oncol 2012;23(Suppl 10):x139-x Chao Y, Wu C-Y, Wang JP, et al. A randomized controlled trial of gemcitabine plus cisplatin versus gemcitabine alone in the treatment of metastatic pancreatic cancer. Cancer Chemother Pharmacol 2013;72: Fairclough DL. Design and analysis of quality of life studies in clinical trials. 2nd edn. Boca Raton, FL: Chapman and Hall/CRC Press, Irish W, Sherrill B, Cole B, et al. Quality-adjusted survival in a crossover trial of letrozole versus tamoxifen in postmenopausal women with advanced breast cancer. Ann Oncol 2005;16: Notice of Correction: The version of this article published online ahead of print on 21 March 2014 contained errors in Figure 1. The publishers apologise for these errors, which were introduced during the typesetting process. The errors have been corrected in this version of the article. 346 Quality-adjusted survival in metastatic pancreatic cancer Reni et al Informa UK Ltd