Lung (2014) 192:191–195 DOI 10.1007/s00408-013-9516-y

PLEURAL DISEASE

Stratification of Malignant Pleural Mesothelioma Prognosis Using Recursive Partitioning Analysis Hidekazu Suzuki • Kazuhiro Asami • Tomonori Hirashima • Norio Okamoto • Tadahiro Yamadori • Motohiro Tamiya • Naoko Morishita • Takayuki Shiroyama • Sawa Takeoka • Akio Osa • Yuichiro Azuma • Kyoichi Okishio • Tomoya Kawaguchi Shinji Atagi • Ichiro Kawase



Received: 8 July 2013 / Accepted: 2 October 2013 / Published online: 20 October 2013 Ó Springer Science+Business Media New York 2013

Abstract Purpose Prognostic factors and complicated prognostic models have been proposed for malignant pleural mesothelioma (MPM). This study was designed to stratify MPM prognosis by using a simple model. Methods Patients diagnosed with MPM in the past 10 years (n = 122) were examined retrospectively. Data on the presence of chest pain, performance status (PS), asbestos exposure, smoking status, white blood cell count (WBC), haemoglobin (Hb) concentration, platelet count (PLT), lactate dehydronate (LD), histology, stage, and date of death or censored status were collected. After the factors were examined in the univariate analysis, recursive partitioning analysis was performed. Results Statistically significant factors related to survival were the type of histology, stage, PS, WBC, PLT, Hb concentration, and LD. Histology, stage, PS, and Hb concentration were used in multivariate analysis. Stage and Hb concentration showed good statistical significance, whereas PS was borderline significant. The survival analyses were stratified into five groups by PS, stage, Hb concentration, and chest pain using recursive partitioning analysis. Group A comprised patients showing the most favourable prognoses (PS 0–2 and Hb concentration [12.1 g dL-1 or PS

H. Suzuki (&)  T. Hirashima  N. Okamoto  T. Yamadori  M. Tamiya  N. Morishita  T. Shiroyama  S. Takeoka  A. Osa  Y. Azuma  I. Kawase Department of Thoracic Malignancy, Osaka Prefectural Medical Center for Respiratory and Allergic Diseases, 3-7-1 Habikino Habikino-shi, Osaka 583-8588, Japan e-mail: [email protected] K. Asami  K. Okishio  T. Kawaguchi  S. Atagi Department of Clinical Oncology, National Hospital Organization Kinki-chuo Chest Medical Center, Osaka, Japan

0–2 and Hb concentration B12.1 g dL-1 without pain), and group B comprised the remaining patients. The median overall survival in groups A and B was 563 days (95 % confidence interval [CI] 502–779) and 157 days (95 % CI 115–224), respectively (hazard ratio of 5.44 [3.46–8.53, P \ 0.0001]). Conclusions The MPM patients with PS 0–2 and Hb concentration [12.1 or B12.1 g dL-1 without chest pain had favourable prognoses. Keywords Haemoglobins  Mesothelioma  Multivariate analysis  Prognosis  Retrospective studies

Introduction Malignant pleural mesothelioma (MPM) is associated with asbestos exposure and is known as a fatal disease with an aggressive course. The prevalence of MPM is expected to increase, because the peak period when asbestos volumes were imported into Japan was the 1970s [1]. MPM shows a poor prognosis with a median survival time of less than 2 years. The median survival time in Japanese patients was estimated to be 308 days [2], and a recent report in Europe showed a 1-year survival rate of 47 % [3]. There are two well-known prognostic scoring systems for MPM. Herndon et al. advocated the use of six complex factors and dividing patients into six groups [4]; the six factors included performance status (PS), age, white blood cell (WBC) count, haemoglobin (Hb) concentration, chest pain, and body weight loss. Curran et al. [5] proposed the use of five factors and dividing patients into two groups; the five factors included WBC count, definite or probable histology, PS, sarcomatoid histology, and gender. More

123

192

than 10 years have passed since these scoring systems were established. Effective chemotherapy, such as that involving pemetrexed [6] and raltitrexed [7], proved beneficial for survival in phase III trials and were introduced in the past decade. Nowadays, this information is used to select the best treatment option for MPM. This study was designed to stratify the prognosis of MPM using simple factors from multicentre data.

Methods

Lung (2014) 192:191–195

Recursive Partitioning Analysis We used the conditional inference method for tree-structured regression models that split the data into groups by the largest log-rank statistics. This algorithm consistently uses the most suitable factor to stratify the groups [9, 10]. Factors examined in the univariate analysis were used to establish groups with unbiased recursive partitioning by using R (R Core Team (2012). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria) and an additional package ‘‘Party.’’

Patients We examined retrospectively patients who were diagnosed with MPM from January 2001 to December 2009 at either the Osaka Prefectural Medical Center for Respiratory and Allergic Diseases or the National Hospital Organization Kinki-Chuo Chest Medical Center in Japan. This study was approved by the institutional review boards of the participating institutions. Data on age, gender, chest pain, PS, asbestos exposure, smoking status (packs/year), WBC count, Hb concentration, platelet count (PLT), lactate dehydronate (LD), histology, diagnosis day, stage according to the International Mesothelioma Interest Group [8], treatment type, intervention by palliative care team, and date of death or censored status were collected. Statistical Analysis Variables were divided into binary categories as follows: age \70 years or C70 years, PS 0–1 or 2–4, histology as epithelioid or nonepithelioid, stage I–II or III–IV, WBC count \8,300 or C8300 lL-1, PLT \409104 or C409104 lL-1, Hb concentration\12 or C12 g dL-1, and packs/year \40 or C40. The European Organization for Research and Treatment of Cancer prognostic score (EPS) is calculated as follows: EPS = 0.55 (if WBC count [8,300 lL-1) ? 0.6 (if PS is 1 or 2) ? 0.52 (if histology was probable) ? 0.67 (if sarcomatous type) ? 0.6 (if male) [5]. EPS B 1.27 is considered a good prognosis; therefore, we assigned patients to two groups as per their cutoff values. The sample size was decided upon without any specific intention. Patients were followed up for at least 2 years to secure a sufficient number of events. The last date included for analysis was the end of February 2012. Survival time was measured from the date of diagnosis until death or censoring. Kaplan–Meier curves were used to estimate the survival time, and log-rank tests were used to analyse differences in survival. Univariate and multivariate Cox proportional hazard models were used to calculate hazard ratios and 95 % confidential intervals (CIs) between groups. P \ 0.05 was considered statically significant.

123

Results Patient Background Data on 122 patients were collected (median age, 69.5 years; 107 men and 15 women). Table 1 shows the clinical background of the patients. A variety of mesothelioma was diagnosed, including 64 epithelioid mesotheliomas, 24 sarcomatoid mesotheliomas, 12 biphasic mesotheliomas, and 22 other or undivided mesotheliomas. Median smoking status was 40 pack-years, including 28 never-smokers. Cases of advanced stage were marginally more frequent; 32 patients had stage I or II cancer. Good PS also was noted with 4, 63, 34, 18, and 3 patients with a performance score of 0, 1, 2, 3, and 4, respectively. Forty-two patients reported known asbestos exposure. Regarding treatment type, 85 patients received chemotherapy and 24 patients underwent surgery. In addition, the palliative care team consulted 24 patients. Univariate and Multivariate Analyses Approximately 80 % of the patients’ events (97/122) were observed during the study period. The median overall survival was 408 days (95 % CI 307–532). The results of the univariate Cox proportional hazard model are shown in Table 2. The factors that were statistically significant in the univariate model include type of histology, stage, PS, WBC count, PLT, Hb concentration, and LD. Considering confounding factors and P values, we selected four factors for multivariate analysis: type of histology, stage, PS, and Hb concentration. Consequently, stage and Hb concentration were found to be statistically independent factors, whereas PS was borderline significant. The prognostic scores were divided into two groups and hazard ratios were calculated. The hazard ratio was 1.67 (95 % CI 1.11–2.47, P = 0.014) for the EPS B 1.27 group compared with the EPS [ 1.27 group (Fig. 1).

Lung (2014) 192:191–195

193

Table 1 Patients’ background and univariate analysis Factors

Category

Age (year)

\70

Gender

Hazard ratio [95 % CI]

P value

Factors

Category

N

Hazard ratio [95% CI]

P value

61

1.41

0.099

Histology

Epithelial

64

1.34

0.217

C70

61

[0.94–2.11]

Nonepithelial

58

[0.84–2.12]

Female

15

0.635

0.167

Stage

I or II

33

2.22

III or IV

89

[1.22–4.04]

Male Histology Smoking (Pack-years) Stage PS Chest pain WBC (ll-1) Hb (g dl-1)

N

Table 2 Multivariate analysis

107

[0.35–1.17]

Epithelial

64

1.99

Nonepithelial

58

[1.32–3]

\40 C40

55 67

0.826 [0.55–1.24]

0.359

Hb (g dl-1)

0.000067*

N number of patients, CI confidential interval, PS performance status, WBC white blood cell, Hb haemoglobin, PLT platelet cell count, LD lactate dehydrogenase

I or II

33

2.76

III or IV

89

[1.59–4.8]

1

67

2.23

C2

55

[1.49–3.34]

No

71

1.05

Yes

51

[0.70–1.59]

\8300

91

1.81

C8300

31

[1.13–2.89]

\11

27

2.15 [1.33–3.48]

C11

95

PLT (9104 ll-1) \40

100

C40

22

1.96

0.001*

0.0001*

PS

1

67

1.56

C2

51

[0.999–2.44]

\11 C11

27 95

1.94 [1.17–3.22]

0.009* 0.05007 0.01*

* Statistically significant 0.811 0.013* 0.0018* 0.011*

[1.17–3.3]

LD (IU l-1)

\230

31

1.49

0.009*

EORTC score

C230 \1.27

91 69

[0.95–2.33] 1.67

0.0014*

C1.27

53

[1.11–2.498]

N number of patients, CI confidential interval, PS performance status, WBC white blood cell count, Hb haemoglobin, PLT platelet cell count, LD lactate dehydrogenase, EORTC score European organization for research and treatment of cancer prognostic score * Statistically significant

Recursive Partitioning Analysis We used all available variables (WBC, Hb concentration, PLT, LD, age, gender, chest pain, PS, smoking status, type of histology, and stage) to construct conditional inference trees (Fig. 2). The survival curves were stratified by four factors: PS, stage, Hb concentration, and pain. The strongest factor was PS (0–2 vs. C3); therefore, the patients were divided on the basis of PS. Those in the poor PS groups were further divided by stage of mesothelioma. Those in the good PS groups were divided according to Hb concentration. Subsequently, those with an Hb concentration B12.1 g dL-1 were further divided by the presence or absence of pain. The outcome of the conditional inference was consistent with the multivariate analysis. The patients with a PS of 0–2 and Hb concentration [12.1 g dL-1 showed the most favourable prognosis of 591 days (95 % CI 512–840 days), whereas patients with a PS of 3–4 and stage IV showed the worst prognosis of 103 days (95 % CI

Fig. 1 Survival curves of patients divided into two groups according to the prognostic score from the European Organization for Research and Treatment of Cancer. The solid line shows patients with a prognostic score B1.27 (low-risk group) and the broken line shows patients with a prognostic score [1.27 (high-risk group). HR hazard ratio, CI confidence interval

86–234 days). Moreover, we were able to divide patients into two groups according to prognosis: the patients who demonstrated the most favourable prognoses (PS 0–2 and Hb concentration [12.1 g dL-1 or PS 0–2 and Hb concentration B12.1 g dL-1 along with no chest pain) were included in group A, and the remaining patients were included in group B; this shows that stratification could make a great difference in predicting the prognosis. Median overall survival in group A and group B was 563 days (95 % CI 502–779 days) and 157 days (95 % CI 115–224 days), respectively, with a hazard ratio of 5.44 (3.46–8.53, P \ 0.0001; Fig. 3).

123

194

Lung (2014) 192:191–195

Fig. 2 Recursive partitioning analysis for survival prediction. Each nodes’ median survival time (days) and 95 % confidence interval was as follows: Node 3 = 103 (86–234), Node 4 = 160 (115– not available), Node 9 = 591 (512–840), Node 7 = 195 (127–473), Node 8 = 483 (329–744)

Fig. 3 Survival curves of patients divided according to prognoses. Solid line indicates group A (nodes 8 and 9 in Fig. 2), and broken line indicates group B (Nodes 3, 4, and 7 in Fig. 2)

Discussion This study showed that prognosis of MPM could be stratified by four simple factors: PS, stage, Hb concentration, and chest pain. The strongest factor was PS. A patient with a PS of 0–2 and Hb concentration [12.1 g dL-1 or Hb concentration B12.1 g dL-1 had the most favourable prognosis. These factors were previously reported to predict MPM. PS is used in not only the prediction of MPM but also survival models for many other malignancies. Herndon

123

et al. [4] reported PS as the most important prognostic factor in combined data of seven phase II studies by using recursive partitioning algorithms. Similarly, this study confirmed that PS is the strongest prognostic factor. Previous studies demonstrated that WBC, Hb concentration, and PLT are prognostic factors of MPM. Stage, PS, and LD were identified as independent prognostic factors by multivariate analysis in our previous report [11]. However, the staging system needs revision, because T factor and N factor have not been confirmed [12], and validation using large prospective data is required. Stage seems to be an important reflection of prognosis in recent surgical cases [13, 14]. In addition, stage and Hb concentration were reported as prognostic factors in an analysis of MPM cases from 2007 to 2009 [15]. These findings are in agreement with those of our study. Nevertheless, determining an appropriate cutoff value is difficult. Analysis of a randomized trial of raltitrexed showed that the cutoff value of Hb concentration was lower than the normal limit (men = 16 g dL-1; women = 14 g dL-1) [7]. Haemoglobin remained a significant prognostic factor even after multivariate analysis. In our study, the cutoff value of Hb concentration was 12.1 g dL-1, as derived from our statistical analysis. We believe this value is feasible because it is similar to the lower normal Hb concentration limit in our hospital (12.0 g dL-1). Through the recursive partitioning analysis, Hb concentration proved to be of superior prognostic value than WBC count, PLT, and LD. The results of this analysis suggest a simple model for the prognosis of MPM. EPS could be reliable for some validation studies [16, 17]. Interestingly, the prognostic stratification of our study was more clearly performed compared to EPS; however, a

Lung (2014) 192:191–195

validation study of our model is required before further conclusions can be made. The nonepithelioid-type mesothelioma had previously shown a poor prognosis [18]. The histological type is still a prognostic factor, as mentioned in the latest research [15, 19, 20]. Our study showed that the nonepithelioid-type mesothelioma had a poor prognosis by univariate analysis, but this finding did not hold true in a final tree model, because the nonepithelioid-type group included many patients with poor PS or anaemia. The histological type was finally not considered in the CALBG prognosis model as mentioned above. Our study has some limitations. This was a retrospective study of two institutions and had a small sample size compared with past studies. The introduction of positronemission tomography might change the staging process. In addition, the presence of an effective chemotherapeutic agent, such as pemetrexed, might improve survival time when its widespread use is established among Japanese patients. We had compared the regimens retrospectively, including PEM or non-PEM in a previous study [21], but PEM did not seem to have led to a clear improvement in the prognosis until now. Lastly, our statistical models may have overfitted our analysis by combining them into two groups. In conclusion, this study showed that MPM prognosis is based on four simple factors: PS, stage, Hb concentration, and chest pain. Patients with PS 0–2 and Hb concentration [12.1 or B12.1 g dL-1 without chest pain had a favourable prognosis. Further studies are required to validate these findings, and investigation into accurate biomarkers would further enhance our understanding of MPM. Conflict of Interest

195

7.

8.

9.

10.

11.

12.

13.

14.

15.

None. 16.

References 1. Kanazawa N, Ioka A, Tsukuma H, Ajiki W, Oshima A (2006) Incidence and survival of mesothelioma in Osaka, Japan. Jpn J Clin Oncol 36:254–257 2. Nojiri S, Gemba K, Aoe K, Kato K, Yamaguchi T, Sato T, Kubota K, Kishimoto T (2011) Survival and prognostic factors in malignant pleural mesothelioma: a retrospective study of 314 patients in the west part of Japan. Jpn J Clin Oncol 41:32–39 3. van der Bij S, Koffijberg H, Burgers JA, Baas P, van de Vijver MJ, de Mol BA, Moons KG (2012) Prognosis and prognostic factors of patients with mesothelioma: a population-based study. Br J Cancer 107:161–164 4. Herndon JE, Green MR, Chahinian AP, Corson JM, Suzuki Y, Vogelzang NJ (1998) Factors predictive of survival among 337 patients with mesothelioma treated between 1984 and 1994 by the Cancer and Leukemia Group B. Chest 113:723–731 5. Curran D, Sahmoud T, Therasse P, van Meerbeeck J, Postmus PE, Giaccone G (1998) Prognostic factors in patients with pleural mesothelioma: the European Organization for Research and Treatment of Cancer experience. J Clin Oncol 16:145–152 6. Vogelzang NJ, Rusthoven JJ, Symanowski J, Denham C, Kaukel E, Ruffie P, Gatzemeier U, Boyer M, Emri S, Manegold C,

17.

18.

19.

20.

21.

Niyikiza C, Paoletti P (2003) Phase III study of pemetrexed in combination with cisplatin versus cisplatin alone in patients with malignant pleural mesothelioma. J Clin Oncol 21:2636–2644 van Meerbeeck JP, Gaafar R, Manegold C, Van Klaveren RJ, Van Marck EA, Vincent M, Legrand C, Bottomley A, Debruyne C, Giaccone G (2005) Randomized phase III study of cisplatin with or without raltitrexed in patients with malignant pleural mesothelioma: an intergroup study of the European Organisation for Research and Treatment of Cancer Lung Cancer Group and the National Cancer Institute of Canada. J Clin Oncol 23:6881–6889 Rusch VW (1995) A proposed new international TNM staging system for malignant pleural mesothelioma. From the International Mesothelioma Interest Group. Chest 108:1122–1128 Torsten H, Kurt H, Achim Z (2006) Unbiased recursive partitioning: a conditional inference framework. J Comput Graph Stat 15:651–674 John C, Antje H (2012) Prognostic groups by tree-based partitioning and data refinement methods. In: Handbook of statistics in clinical oncology, 3rd edn. CRC Press, Boca Raton, pp 505–528 Suzuki H, Hirashima T, Kobayashi M, Okamoto N, Matsuura Y, Tamiya M, Morishita N, Okafuji K, Shiroyama T, Morimura O, Morita S, Kawase I (2012) Prognostic factors in malignant pleural mesothelioma: a retrospective study. Intern Med 51:707–710 Kindler HL (2012) Robust data: the essential foundation of a revised staging system for pleural mesothelioma. J Thorac Oncol 7:1623–1624 Richards WG, Godleski JJ, Yeap BY, Corson JM, Chirieac LR, Zellos L, Mujoomdar A, Jaklitsch MT, Bueno R, Sugarbaker DJ (2010) Proposed adjustments to pathologic staging of epithelial malignant pleural mesothelioma based on analysis of 354 cases. Cancer 116:1510–1517 Rusch VW, Giroux D, Kennedy C, Ruffini E, Cangir AK, Rice D, Pass H, Asamura H, Waller D, Edwards J, Weder W, Hoffmann H, van Meerbeeck JP (2012) Initial analysis of the international association for the study of lung cancer mesothelioma database. J Thorac Oncol 7:1631–1639 Kao SC, Vardy J, Chatfield M, Corte P, Pavlakis N, Clarke C, van Zandwijk N, Clarke S (2013) Validation of prognostic factors in malignant pleural mesothelioma: a retrospective analysis of data from patients seeking compensation from the New South Wales dust diseases board. Clin Lung Cancer 14:70–77 Edwards JG, Abrams KR, Leverment JN, Spyt TJ, Waller DA, O’Byrne KJ (2000) Prognostic factors for malignant mesothelioma in 142 patients: validation of CALGB and EORTC prognostic scoring systems. Thorax 55:731–735 Fennell DA, Parmar A, Shamash J, Evans MT, Sheaff MT, Sylvester R, Dhaliwal K, Gower N, Steele J, Rudd R (2005) Statistical validation of the EORTC prognostic model for malignant pleural mesothelioma based on three consecutive phase II trials. J Clin Oncol 23:184–189 Mineo TC, Ambrogi V (2012) Malignant pleural mesothelioma: factors influencing the prognosis. Oncology (Williston Park) 26:1164–1175 Flores RM, Zakowski M, Venkatraman E, Krug L, Rosenzweig K, Dycoco J, Lee C, Yeoh C, Bains M, Rusch V (2007) Prognostic factors in the treatment of malignant pleural mesothelioma at a large tertiary referral center. J Thorac Oncol 2:957–965 Musk AW, Olsen N, Alfonso H, Reid A, Mina R, Franklin P, Sleith J, Hammond N, Threlfall T, Shilkin KB, de Klerk NH (2011) Predicting survival in malignant mesothelioma. Eur Respir J 38:1420–1424 Higashiguchi M, Suzuki H, Hirashima T, Kobayashi M, Goya S, Okamoto N, Matsuura Y, Tamiya M, Morishita N, Tsumori T, Kawase I (2012) A retrospective study of chemotherapy with and without pemetrexed in malignant pleural mesothelioma. Anticancer Res 32:609–613

123

Stratification of malignant pleural mesothelioma prognosis using recursive partitioning analysis.

Prognostic factors and complicated prognostic models have been proposed for malignant pleural mesothelioma (MPM). This study was designed to stratify ...
202KB Sizes 0 Downloads 0 Views