Small-Cell Carcinoma of the Lung: Derivation of a Prognostic Staging System By U. Sagman, E. Maki, W.K. Evans, D. Warr, F.A. Shepherd, J.P. Sculier, R. Haddad, D. Payne, J.F. Pringle, J.L. Yeoh, A. Ciampi, G. DeBoer, S. McKinney, R. Ginsberg, and R. Feld Retrospective data on 22 pretreatment attributes were evaluated in 614 patients with small-cell carcinoma of the lung (SCCL). The series included 284 patients with limited disease (LD) and 328 patients with extensive disease (ED) managed between 1974 and 1986. Prognostic factors were evaluated by univariate analysis and by the Cox multivariate regression model. Recursive partition and amalgamation algorithm (RECPAM), two clustering methods well suited for obtaining strata and adapted for censoring survival data, were developed and used in the formulation of a new prognostic staging system. In univariate analysis, prognosis was significantly influenced by extent of disease (DE), the number of metastatic sites, and the detection of mediastinal spread in LD. Poor performance status (PS), male sex, and advanced age were negatively correlated with survival, as were increased serum levels of alkaline phosphates (AP), lactate dehydrogenase (LDH), carcinoembryonic antigen (CEA), total WBC count (WBCC), and low platelet count and low serum
CANCER
OF THE lung remains the leading cause of cancer deaths in adult men and recently has been established as the primary cause of cancer mortality in women.' Small-cell carcinoma of the lung (SCCL) constitutes between 20% and 25% of all types of lung cancer. To date, clinical studies have led to significant though limited progress in the management of patients with SCCL.2 Historically, surgery was considered the optimal treatment modality.3 Subsequent randomized trials in Great Britain established radiation therapy as a superior approach to treatment.4 The demonstration that a single chemotherapeutic agent, when combined with radiation therapy, improves response rates and survival5 ushered in an era of extensive experience with combination chemotherapy. Incorporation of novel drugs such as etoposide (VP-16) and cisplatin to clinical trial investigations6 further expanded the range of the therapeutic armamentarium. With modest improvement in response rates and in long-term survival, a suitable framework for the proper design of clinical trials and their interpretation has become essential. In particular, it is increasingly recognized that factors other than therapeutic
sodium. The Cox model identified plasma LDH and mediastinal spread as the only significant factors in LD; the influence of PS, number of metastatic sites, bone metastasis, brain metastasis, and platelet count were identified as significant in ED. The RECPAM model identified four distinct risk groups defined in a classification tree by the following eight attributes: DE, PS, serum AP, serum LDH, mediastinal spread, sex, WBCC, and liver metastasis. The four groups were distinguished by median survival times of 59, 49, 35, and 24 weeks, respectively (P = .0001). Interactions among prognostic factors are emphasized in the RECPAM classification model as evidenced by reassignment of patients across conventional staging barriers into alternate prognostic groups. The advantages of using RECPAM over the more conventional Cox regression techniques for a new staging system are discussed. J Clin Oncol 9:1639-1649. © 1991 by American Society of Clinical Oncology.
interventions determine response rates and the overall potential for cure. Virtually, all studies of prognostic factors in SCCL identify extent of disease (DE) and clinical performance status (PS) as the two most powerful prognostic factors.7'8 Although a plethora of other significant prognostic factors have been identified, 9-11 they have seldom been incorporated into disease staging or patient stratification in clinical trials. Recent studies indi-
From The Ontario CancerInstitute incorporatingThe Princess MargaretHospital, Toronto; The Toronto GeneralHospital, Toronto; The Mount SinaiHospital, Toronto; The Ontario Cancer Treatment and ResearchFoundation,Ottawa Regional Cancer Centre, Ottawa; Montreal Children's Hospital, Montreal;McGill University ResearchInstitute, Montreal, Canada; and InstituteJules Bordet, Brussels,Belgium. Submitted May 29, 1990; acceptedMarch 4, 1991. Presented in part at the 24th Annual Meeting of the American Society of Clinical Oncology, May 1988, New Orleans, LA; and at the Fifth World Conference on Lung Cancer,August 1988, Interlaken, Switzerland. Address reprint requests to Ronald Feld, MD, The Ontario Cancer Institute, 500 Sherbourne St, Toronto, Ontario, CanadaM4X1K9. © 1991 by American Society of Clinical Oncology. 0732-183X/91/0909-0022$3.00/0
Journal of Clinical Oncology, Vol 9, No 9 (September), 1991: pp 1639-1649
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1639
1640
SAGMAN ET AL
cate that the current staging system for SCCL may not accurately reflect segregation of patients into homogeneous risk groups. 12-41 Hence, reevaluation of available prognostic data aimed towards the formation of a new staging system may be desirable. Ideally, objectives embodied in this process would seek to accommodate three goals: to provide more accurate prognosis for individual patients, to define optimal risk groups for patient stratification in prospective clinical trials, and to standardize the assessment of diverse clinical trials. At the University of Toronto teaching hospitals we accrued data on 614 patients with SCCL entered onto sequential therapeutic studies during a 10-year interval. We evaluated 22 potentially significant prognostic factors by univariate analysis and by the Cox multivariate regression models. Based on the analysis of prognostic factors, we introduce a novel statistical methodology using recursive partitioning and amalgamation algorithms (RECPAM) as the vehicle for the formulation of a new staging classification.
MATERIALS AND METHODS The present study includes 614 patients entered into clinical trials from 1976 and followed for response to 1986 at the Toronto General and The Princess Margaret Hospitals (Toronto, Canada). Only patients satisfying World Health 1 Organization (WHO) diagnostic criteria for SCCL 5 were included. Patients who had undergone prior surgical resection were not included in the study. The treatment of patients consisted of clinical trials incorporating combination chemotherapy and, as dictated by specific protocols, prophylactic cranial irradiation and/or thoracic irradiation. A summary of these trials, including the median and overall survival of patients, is outlined in Table 1 and described in detail elsewhere.1- 20 Sixty-five patients not on study protocols (NOS) constituted those not meeting eligibility criteria for concurrently active trials. Although management of such patients was based on the investigator's option, most received a chemotherapeutic regimen that included cyclophosphamide, doxorubicin, and vincristine (a modified CAV protocol; Feld and Sagman, personal communication, May 1990). The characteristics of patients included in the different trials are shown in Table 2. All patients had a complete medical history, physical examination, evaluation of PS (Eastern Cooperative Oncology Group [ECOG]),21 and investigations to define the DE. These included chest x-ray, bone, liver and brain scintiscans or computed tomographic (CT) scans, bone marrow evaluation (aspiration
Table 1. Trials of Chemotherapy and Radiotherapy for 614 Patients With SCCL Median Survival (weeks) Regimen
N
CAV1,' sequential
3 agents* plus irradiationt followed by 3 agentsJ 3 agents plus irradiationhl¶ 3 agents# plus irradiationll and ¶ or ** followed by or alternating with 2 agents-it 3 agents#f plus irradiationtS or 3 agents# alternating with 2 agentstt plus irradiationt¶ 3 agents§§ -- TPNII IIfollowed by 3 agents¶¶ and irradiationtli followed by 3 agents## 3 agents***
148
48 (60)
36 (88)
43
294 52
49 (129) 61 (52)
34 (165)
40 61
35 (12)
35
6
1
CAV2, , BR3 ,'" randomized BR,, randomized
TPN,19 randomized 2
NOSO
LD
ED
Overall
Trial
12
43
63 (23)
63 (20)
63
65
41 (22)
32 (43)
36
*Cyclophosphamide (750 mg/m2) + doxorubicin (50 mg/m') + vincristine (2 mg) every 3 weeks x 3.
"t25Gy/2 weeks in 10 fractions to primary tumor + mediastinum after 3 cycles of chemotherapy. :Lomustine (50 mg/m') every 6 weeks + procarbazine (100 mg/m') x 10 days every 3 weeks for 1 year + methotrexate (10 mg/m') 2 x per week. §As per *, substitute cyclophosphamide (900 mg/m') and doxorubicin (45 mg/m' ) every 3 weeks x 6.
JIAs per t, but after 6 cycles
of chemotherapy. 120 Gy/1 week in 5 fractions to cranium after 3 cycles of chemotherapy. #As per *, substitute cyclophosphamide (1,000 mg/m') every 3 weeks x 3. Gy/3 weeks in 15 fractions to primary tumor and mediastinum. ttEtoposide (100 mg/m') + cisplatin (25 mg/m2) daily x 3 every 3 weeks x 3. #iAs per #, but for 6 cycles. §§Cyclophosphamide (1,200 mg/m') + doxorubicin (70 mg/m') + vincristine (2 mg).
"**37
S11111i -1.5 g of protein/kg + 40 Kcal/kg body weight. ¶¶As per #, but for 2 cycles. 2 ##Etoposide (100 mg/m ) days 1, 2, and 3 + methotrexate (10 mg/m2) day 1 + leucovorin 15 mg/m CAV2 as per physician discretion (see Materials and Methods).
2
x 6 on every 3 weeks for 1 year.
"***Modified
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1641
STAGING FOR SMALL-CELL LUNG CANCER Table 2. Patient Characteristics
No. of patients Male/female Age (years) Median Range DE Limited ED ECOG PS 0 1 2 3
CAVI
CAV 2
BR 3
BR 4
TPN
NOS
Total
148 116/32
294 209/85
52 35/17
12 6/6
43 29/14
65 41/24
614 436/178
59 34-81
60 32-81
62 35-79
57 41-67
60 45-75
61 32-84
60 32-84
60 88
129 165
52 0
0 12
23 20
22 43
286 328
25 109 11 3
37 211 34 12
1 32 17 2
0 1 8 3
4 35 4 0
0 24 32 9
67 412 106 29
NOTE. See Table I for protocol definitions.
and biopsy), and, in select patients with limited disease (LD), mediastinoscopy. Laboratory determinations included hemoglobin, total and differential WBC count (WBCC), platelet count, serum alkaline phosphatase (AP), lactate dehydrogenase (LDH), carcinoembryonic antigen (CEA), and serum electrolytes (including serum sodium). Values for some variables were not available for all patients. Following appropriate staging, those patients with disease confined to a hemithorax, including ipsilateral supraclavicular adenopathy, were considered to have LD, and all others were considered to have extensive disease (ED). Based on tumor-node-metastasis (TNM) staging principles, we in the present study and others in the literature 7'",12 have recognized staging subcategories that bear prognostic information. Therefore, in addition to using the LD and ED stage classification we have also identified other staging categories related to the TNM system. Based on our earlier studies,22 the LD stage (LDS) was further divided into two prognostic categories according to the absence (LDS-1) or presence (LDS-2) of mediastinal disease. Regardless of mediastinal involvement, patients with ipsilateral supraclavicular adenopathy and/or pleural effusion were assigned to LDS-2. Similarly, the ED stage was subdivided by comparing patients with a single metastatic site (EDS-1) with those with more than one site of distant involvement (EDS-2). A complete response (CR) was defined as complete disappearance of all clinically and radiographically apparent tumor for at least 3 weeks after treatment. Partial response (PR) was defined as a decrease of 50% in the sum of the estimated products of the greatest cross-sectional diameters of well-outlined lesions, or greater than 50% decrease in the estimated area of poorly outlined lesions for at least 3 weeks in the absence of progressive disease or the occurrence of new lesions elsewhere. Progression of disease was defined as an increase of greater than 25% in the product of cross-sectional diameters of one or more outlined lesions or the occurrence of new lesions irrespective of response elsewhere. Survival time was measured from the time of first treatment to the date of death or last follow-up assessment.
Statistical Methods UnivariateAnalysis Product-limit survival estimates were obtained for all levels of each covariate using the method of Kaplan and Meier.' Statistical significance between survival curves was assessed using the log-rank test. Where appropriate, continuous variables were categorized before analysis.
MultivariateAnalysis Multiple regression model. Multivariate analyses were performed to assess the importance of each pretreatment variable in the presence of all other variables. The proportional hazards model of Cox 24 was used to assess the importance of pretreatment variables on survival, and to divide the patients into subgroups that were homogeneous with respect to prognosis. The hazards estimates were calculated as follows: Kpretreatment variables were labeled asZ,,,Z ... , ZK. The relationship between the time to death and the vector variable Z was assessed by a model in which the hazard (instantaneous death rate) is proportional to exp (P1Z, + 32Z 2 + ... PIKZ)," where Pi, i = 1, 2, ... K are regression coefficients. The proportionality assumption for death hazards was examined for each variable in the model by visual inspection of the log minus log survival plot. 5 The plot for males versus females indicated nonproportional hazards for these two groups; therefore, sex was used as a stratifying variable in the regression analysis. A score was calculated for each patient using 3Z, where p is the vector of estimated regression parameters, and Z is the vector of covariates. The model was selected by first considering all variables and then eliminating one at a time those variables failing to meet criteria of preassigned levels of significance (backward exclusion). Prognostic groupings were obtained by dividing the range of index scores. Log-rank analysis for survival differences and 2- and 5-year survival rates were calculated for patients in the different groups. The BMDP and SAS statistical software packages were used to perform 262 7 the analyses. , The classificationapproach. A second multivariate analysis was performed using the recursive partitioning method-
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1642
SAGMAN ET AL
ology. This methodology has been successfully used in the 2 analysis of medical data when there is no censoring, " as well as in the case of censored survival data.""' The purpose of this analysis is to obtain a classification rule that uses pretreatment variables to group patients with similar prognosis. Briefly, the classification proceeds as follows. Initially, the entire patient population is partitioned into two subgroups according to the variable that produces the most significant survival difference. Each of these subgroups is again partitioned into two subgroups in the same manner. The partitioning process stops when no available variable produces a significant survival difference in a given subgroup or when the size of a given subgroup is too small. By definition, any two clusters of patients arising from the same subgroup must be significantly different with respect to survival. The same is not necessarily true for clusters of patients arising from different parent subgroups. An amalgamation process, the second component of RECPAM, joins subgroups of patients who do not differ significantly from each other with regard to survival. This latter step produces the final prognostic classes. The classification rule is represented as a tree, an example of which is outlined in Fig 1. The software program used is known as RECPAM. It has been developed in an effort to integrate clustering and amalgamation techniques in the analysis of survival data. Further details regarding this statistical methodology are 0
available.3 ,32,33
Yes
AA
No
Root
.
Fig 1. Prototype RECPAM model. In this hypothetical example, the first split in the tree allocates patients in the first node (the root) to two groups (internal nodes 2 and 3), based on the prognostic factor age. Splitting of nodes 2 and 3 based on the prognostic factors hemoglobin and respectively, and 7 nodes (4 to 7). sex, Terminal nodes 4 culminates in four terminal are distinctly different with regard to survival (the outcome criteria) and, by definition, are designated as classes from (A anddif-e C, nodes 5 and 6 originating respectively). Terminal
Classes A, B, and C are homogeneous and by definition
significantly different from each other. Recursive partitioning and 6and originating from difrepaectientsly). the amalgamation (RP) steps are Terminal indicated nodes by solid5 lines algorithm (AA) by the broken lines. Circles represent internal the size of a nodates; squares, terminal nodes(4 to 7). venTermina, a particular (ie, cohort of population is indiffcatedrent within a node patients).
RESULTS
UnivariateAnalysis Of the 22 pretreatment variables evaluated, 18 were determined to have prognostic significance. Results of the univariate analysis for factors influencing clinical features are summarized in Table 3 and those for pretreatment laboratory values in Table 4. Not all values were available for each prognostic factor. Categorization of pretreatment laboratory values was carried out to simplify comparison with results from centers using other activity units. The series was divided into two categories when the log-rank test was applied. Values for LDH were divided into three groupings (test for trend: x2 = 23.11, df = 1, P = .0001). Similarly, values for CEA were divided into three categories (test for trend: X2 = 8.07, df= 1, P = .0045). As shown in Table 3, raised serum activities of LDH, CEA, and AP were associated with reduced survival, with the hazards ratio being most pronounced for serum LDH. Hematologic parameters revealed decreased survival for patients with either reduced platelet counts or elevated WBCCs. Levels of hemoglobin above or below 12 g/L did not significantly affect survival. Of the clinical attributes (Table 4), the degree of burden of disease-DE (ie, LD v ED), number of ED sites, LDS, and disease in specific metastatic sites-were all correlated with poorer survival. Male sex, age _ 70 years, and poor PS (ECOG 2 to 3) predicted for poor survival. Weight loss, the presence of a pleural effusion, and evidence of superior vena caval (SVC) obstruction did not contribute significantly to survival differences. Multiple RegressionAnalysis None of the treatment regimens resulted in significant survival differences for patients in this study. The model developed in the multiple regression analysis included all patients. All the factors that were found to have prognostic significance in univariate analysis were used in the multiple regression analysis. A second regression analysis including all the prognostic factors was performed. Those factors found not to be significant in univariate analysis were nonsignificant in this analysis as well. Since the number of factors that can be accommodated by the RECPAM software is at most 20, it was decided to exclude this same set of
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1643
STAGING FOR SMALL-CELL LUNG CANCER Table 3. Median Duration of Survival: Influence of Clinical Features
Variable Sex (female/male) Age ( 70 years/> 70 years) Weight loss >2 kg (no/yes) ECOG PS 0-1 2-3 DE LD ED LDS 1 2 No. sites involved (ED) 1 >1 Liver metostases No Yes Brain metastases No Yes Bone metastases No Yes Bone marrow metastases No Yes Nodes No Yes Pleural effusion No Yes SVC obstruction No Yes
No. of Patients Examined
Median Survival (weeks)
178/436 526/88 336/227
49/41 45/37 44/43
479 135
46 33
P (log-rank test)
.010
Relative Risk
.023 .060
1.29 1.32 1.17
.0001
1.49
286 328
CANCER
OF THE lung remains the leading cause of cancer deaths in adult men and recently has been established as the primary cause of cancer mortality in women.' Small-cell carcinoma of the lung (SCCL) constitutes between 20% and 25% of all types of lung cancer. To date, clinical studies have led to significant though limited progress in the management of patients with SCCL.2 Historically, surgery was considered the optimal treatment modality.3 Subsequent randomized trials in Great Britain established radiation therapy as a superior approach to treatment.4 The demonstration that a single chemotherapeutic agent, when combined with radiation therapy, improves response rates and survival5 ushered in an era of extensive experience with combination chemotherapy. Incorporation of novel drugs such as etoposide (VP-16) and cisplatin to clinical trial investigations6 further expanded the range of the therapeutic armamentarium. With modest improvement in response rates and in long-term survival, a suitable framework for the proper design of clinical trials and their interpretation has become essential. In particular, it is increasingly recognized that factors other than therapeutic
sodium. The Cox model identified plasma LDH and mediastinal spread as the only significant factors in LD; the influence of PS, number of metastatic sites, bone metastasis, brain metastasis, and platelet count were identified as significant in ED. The RECPAM model identified four distinct risk groups defined in a classification tree by the following eight attributes: DE, PS, serum AP, serum LDH, mediastinal spread, sex, WBCC, and liver metastasis. The four groups were distinguished by median survival times of 59, 49, 35, and 24 weeks, respectively (P = .0001). Interactions among prognostic factors are emphasized in the RECPAM classification model as evidenced by reassignment of patients across conventional staging barriers into alternate prognostic groups. The advantages of using RECPAM over the more conventional Cox regression techniques for a new staging system are discussed. J Clin Oncol 9:1639-1649. © 1991 by American Society of Clinical Oncology.
interventions determine response rates and the overall potential for cure. Virtually, all studies of prognostic factors in SCCL identify extent of disease (DE) and clinical performance status (PS) as the two most powerful prognostic factors.7'8 Although a plethora of other significant prognostic factors have been identified, 9-11 they have seldom been incorporated into disease staging or patient stratification in clinical trials. Recent studies indi-
From The Ontario CancerInstitute incorporatingThe Princess MargaretHospital, Toronto; The Toronto GeneralHospital, Toronto; The Mount SinaiHospital, Toronto; The Ontario Cancer Treatment and ResearchFoundation,Ottawa Regional Cancer Centre, Ottawa; Montreal Children's Hospital, Montreal;McGill University ResearchInstitute, Montreal, Canada; and InstituteJules Bordet, Brussels,Belgium. Submitted May 29, 1990; acceptedMarch 4, 1991. Presented in part at the 24th Annual Meeting of the American Society of Clinical Oncology, May 1988, New Orleans, LA; and at the Fifth World Conference on Lung Cancer,August 1988, Interlaken, Switzerland. Address reprint requests to Ronald Feld, MD, The Ontario Cancer Institute, 500 Sherbourne St, Toronto, Ontario, CanadaM4X1K9. © 1991 by American Society of Clinical Oncology. 0732-183X/91/0909-0022$3.00/0
Journal of Clinical Oncology, Vol 9, No 9 (September), 1991: pp 1639-1649
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1639
1640
SAGMAN ET AL
cate that the current staging system for SCCL may not accurately reflect segregation of patients into homogeneous risk groups. 12-41 Hence, reevaluation of available prognostic data aimed towards the formation of a new staging system may be desirable. Ideally, objectives embodied in this process would seek to accommodate three goals: to provide more accurate prognosis for individual patients, to define optimal risk groups for patient stratification in prospective clinical trials, and to standardize the assessment of diverse clinical trials. At the University of Toronto teaching hospitals we accrued data on 614 patients with SCCL entered onto sequential therapeutic studies during a 10-year interval. We evaluated 22 potentially significant prognostic factors by univariate analysis and by the Cox multivariate regression models. Based on the analysis of prognostic factors, we introduce a novel statistical methodology using recursive partitioning and amalgamation algorithms (RECPAM) as the vehicle for the formulation of a new staging classification.
MATERIALS AND METHODS The present study includes 614 patients entered into clinical trials from 1976 and followed for response to 1986 at the Toronto General and The Princess Margaret Hospitals (Toronto, Canada). Only patients satisfying World Health 1 Organization (WHO) diagnostic criteria for SCCL 5 were included. Patients who had undergone prior surgical resection were not included in the study. The treatment of patients consisted of clinical trials incorporating combination chemotherapy and, as dictated by specific protocols, prophylactic cranial irradiation and/or thoracic irradiation. A summary of these trials, including the median and overall survival of patients, is outlined in Table 1 and described in detail elsewhere.1- 20 Sixty-five patients not on study protocols (NOS) constituted those not meeting eligibility criteria for concurrently active trials. Although management of such patients was based on the investigator's option, most received a chemotherapeutic regimen that included cyclophosphamide, doxorubicin, and vincristine (a modified CAV protocol; Feld and Sagman, personal communication, May 1990). The characteristics of patients included in the different trials are shown in Table 2. All patients had a complete medical history, physical examination, evaluation of PS (Eastern Cooperative Oncology Group [ECOG]),21 and investigations to define the DE. These included chest x-ray, bone, liver and brain scintiscans or computed tomographic (CT) scans, bone marrow evaluation (aspiration
Table 1. Trials of Chemotherapy and Radiotherapy for 614 Patients With SCCL Median Survival (weeks) Regimen
N
CAV1,' sequential
3 agents* plus irradiationt followed by 3 agentsJ 3 agents plus irradiationhl¶ 3 agents# plus irradiationll and ¶ or ** followed by or alternating with 2 agents-it 3 agents#f plus irradiationtS or 3 agents# alternating with 2 agentstt plus irradiationt¶ 3 agents§§ -- TPNII IIfollowed by 3 agents¶¶ and irradiationtli followed by 3 agents## 3 agents***
148
48 (60)
36 (88)
43
294 52
49 (129) 61 (52)
34 (165)
40 61
35 (12)
35
6
1
CAV2, , BR3 ,'" randomized BR,, randomized
TPN,19 randomized 2
NOSO
LD
ED
Overall
Trial
12
43
63 (23)
63 (20)
63
65
41 (22)
32 (43)
36
*Cyclophosphamide (750 mg/m2) + doxorubicin (50 mg/m') + vincristine (2 mg) every 3 weeks x 3.
"t25Gy/2 weeks in 10 fractions to primary tumor + mediastinum after 3 cycles of chemotherapy. :Lomustine (50 mg/m') every 6 weeks + procarbazine (100 mg/m') x 10 days every 3 weeks for 1 year + methotrexate (10 mg/m') 2 x per week. §As per *, substitute cyclophosphamide (900 mg/m') and doxorubicin (45 mg/m' ) every 3 weeks x 6.
JIAs per t, but after 6 cycles
of chemotherapy. 120 Gy/1 week in 5 fractions to cranium after 3 cycles of chemotherapy. #As per *, substitute cyclophosphamide (1,000 mg/m') every 3 weeks x 3. Gy/3 weeks in 15 fractions to primary tumor and mediastinum. ttEtoposide (100 mg/m') + cisplatin (25 mg/m2) daily x 3 every 3 weeks x 3. #iAs per #, but for 6 cycles. §§Cyclophosphamide (1,200 mg/m') + doxorubicin (70 mg/m') + vincristine (2 mg).
"**37
S11111i -1.5 g of protein/kg + 40 Kcal/kg body weight. ¶¶As per #, but for 2 cycles. 2 ##Etoposide (100 mg/m ) days 1, 2, and 3 + methotrexate (10 mg/m2) day 1 + leucovorin 15 mg/m CAV2 as per physician discretion (see Materials and Methods).
2
x 6 on every 3 weeks for 1 year.
"***Modified
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1641
STAGING FOR SMALL-CELL LUNG CANCER Table 2. Patient Characteristics
No. of patients Male/female Age (years) Median Range DE Limited ED ECOG PS 0 1 2 3
CAVI
CAV 2
BR 3
BR 4
TPN
NOS
Total
148 116/32
294 209/85
52 35/17
12 6/6
43 29/14
65 41/24
614 436/178
59 34-81
60 32-81
62 35-79
57 41-67
60 45-75
61 32-84
60 32-84
60 88
129 165
52 0
0 12
23 20
22 43
286 328
25 109 11 3
37 211 34 12
1 32 17 2
0 1 8 3
4 35 4 0
0 24 32 9
67 412 106 29
NOTE. See Table I for protocol definitions.
and biopsy), and, in select patients with limited disease (LD), mediastinoscopy. Laboratory determinations included hemoglobin, total and differential WBC count (WBCC), platelet count, serum alkaline phosphatase (AP), lactate dehydrogenase (LDH), carcinoembryonic antigen (CEA), and serum electrolytes (including serum sodium). Values for some variables were not available for all patients. Following appropriate staging, those patients with disease confined to a hemithorax, including ipsilateral supraclavicular adenopathy, were considered to have LD, and all others were considered to have extensive disease (ED). Based on tumor-node-metastasis (TNM) staging principles, we in the present study and others in the literature 7'",12 have recognized staging subcategories that bear prognostic information. Therefore, in addition to using the LD and ED stage classification we have also identified other staging categories related to the TNM system. Based on our earlier studies,22 the LD stage (LDS) was further divided into two prognostic categories according to the absence (LDS-1) or presence (LDS-2) of mediastinal disease. Regardless of mediastinal involvement, patients with ipsilateral supraclavicular adenopathy and/or pleural effusion were assigned to LDS-2. Similarly, the ED stage was subdivided by comparing patients with a single metastatic site (EDS-1) with those with more than one site of distant involvement (EDS-2). A complete response (CR) was defined as complete disappearance of all clinically and radiographically apparent tumor for at least 3 weeks after treatment. Partial response (PR) was defined as a decrease of 50% in the sum of the estimated products of the greatest cross-sectional diameters of well-outlined lesions, or greater than 50% decrease in the estimated area of poorly outlined lesions for at least 3 weeks in the absence of progressive disease or the occurrence of new lesions elsewhere. Progression of disease was defined as an increase of greater than 25% in the product of cross-sectional diameters of one or more outlined lesions or the occurrence of new lesions irrespective of response elsewhere. Survival time was measured from the time of first treatment to the date of death or last follow-up assessment.
Statistical Methods UnivariateAnalysis Product-limit survival estimates were obtained for all levels of each covariate using the method of Kaplan and Meier.' Statistical significance between survival curves was assessed using the log-rank test. Where appropriate, continuous variables were categorized before analysis.
MultivariateAnalysis Multiple regression model. Multivariate analyses were performed to assess the importance of each pretreatment variable in the presence of all other variables. The proportional hazards model of Cox 24 was used to assess the importance of pretreatment variables on survival, and to divide the patients into subgroups that were homogeneous with respect to prognosis. The hazards estimates were calculated as follows: Kpretreatment variables were labeled asZ,,,Z ... , ZK. The relationship between the time to death and the vector variable Z was assessed by a model in which the hazard (instantaneous death rate) is proportional to exp (P1Z, + 32Z 2 + ... PIKZ)," where Pi, i = 1, 2, ... K are regression coefficients. The proportionality assumption for death hazards was examined for each variable in the model by visual inspection of the log minus log survival plot. 5 The plot for males versus females indicated nonproportional hazards for these two groups; therefore, sex was used as a stratifying variable in the regression analysis. A score was calculated for each patient using 3Z, where p is the vector of estimated regression parameters, and Z is the vector of covariates. The model was selected by first considering all variables and then eliminating one at a time those variables failing to meet criteria of preassigned levels of significance (backward exclusion). Prognostic groupings were obtained by dividing the range of index scores. Log-rank analysis for survival differences and 2- and 5-year survival rates were calculated for patients in the different groups. The BMDP and SAS statistical software packages were used to perform 262 7 the analyses. , The classificationapproach. A second multivariate analysis was performed using the recursive partitioning method-
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1642
SAGMAN ET AL
ology. This methodology has been successfully used in the 2 analysis of medical data when there is no censoring, " as well as in the case of censored survival data.""' The purpose of this analysis is to obtain a classification rule that uses pretreatment variables to group patients with similar prognosis. Briefly, the classification proceeds as follows. Initially, the entire patient population is partitioned into two subgroups according to the variable that produces the most significant survival difference. Each of these subgroups is again partitioned into two subgroups in the same manner. The partitioning process stops when no available variable produces a significant survival difference in a given subgroup or when the size of a given subgroup is too small. By definition, any two clusters of patients arising from the same subgroup must be significantly different with respect to survival. The same is not necessarily true for clusters of patients arising from different parent subgroups. An amalgamation process, the second component of RECPAM, joins subgroups of patients who do not differ significantly from each other with regard to survival. This latter step produces the final prognostic classes. The classification rule is represented as a tree, an example of which is outlined in Fig 1. The software program used is known as RECPAM. It has been developed in an effort to integrate clustering and amalgamation techniques in the analysis of survival data. Further details regarding this statistical methodology are 0
available.3 ,32,33
Yes
AA
No
Root
.
Fig 1. Prototype RECPAM model. In this hypothetical example, the first split in the tree allocates patients in the first node (the root) to two groups (internal nodes 2 and 3), based on the prognostic factor age. Splitting of nodes 2 and 3 based on the prognostic factors hemoglobin and respectively, and 7 nodes (4 to 7). sex, Terminal nodes 4 culminates in four terminal are distinctly different with regard to survival (the outcome criteria) and, by definition, are designated as classes from (A anddif-e C, nodes 5 and 6 originating respectively). Terminal
Classes A, B, and C are homogeneous and by definition
significantly different from each other. Recursive partitioning and 6and originating from difrepaectientsly). the amalgamation (RP) steps are Terminal indicated nodes by solid5 lines algorithm (AA) by the broken lines. Circles represent internal the size of a nodates; squares, terminal nodes(4 to 7). venTermina, a particular (ie, cohort of population is indiffcatedrent within a node patients).
RESULTS
UnivariateAnalysis Of the 22 pretreatment variables evaluated, 18 were determined to have prognostic significance. Results of the univariate analysis for factors influencing clinical features are summarized in Table 3 and those for pretreatment laboratory values in Table 4. Not all values were available for each prognostic factor. Categorization of pretreatment laboratory values was carried out to simplify comparison with results from centers using other activity units. The series was divided into two categories when the log-rank test was applied. Values for LDH were divided into three groupings (test for trend: x2 = 23.11, df = 1, P = .0001). Similarly, values for CEA were divided into three categories (test for trend: X2 = 8.07, df= 1, P = .0045). As shown in Table 3, raised serum activities of LDH, CEA, and AP were associated with reduced survival, with the hazards ratio being most pronounced for serum LDH. Hematologic parameters revealed decreased survival for patients with either reduced platelet counts or elevated WBCCs. Levels of hemoglobin above or below 12 g/L did not significantly affect survival. Of the clinical attributes (Table 4), the degree of burden of disease-DE (ie, LD v ED), number of ED sites, LDS, and disease in specific metastatic sites-were all correlated with poorer survival. Male sex, age _ 70 years, and poor PS (ECOG 2 to 3) predicted for poor survival. Weight loss, the presence of a pleural effusion, and evidence of superior vena caval (SVC) obstruction did not contribute significantly to survival differences. Multiple RegressionAnalysis None of the treatment regimens resulted in significant survival differences for patients in this study. The model developed in the multiple regression analysis included all patients. All the factors that were found to have prognostic significance in univariate analysis were used in the multiple regression analysis. A second regression analysis including all the prognostic factors was performed. Those factors found not to be significant in univariate analysis were nonsignificant in this analysis as well. Since the number of factors that can be accommodated by the RECPAM software is at most 20, it was decided to exclude this same set of
Downloaded from ascopubs.org by Boston University on April 8, 2019 from 128.197.229.194 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.
1643
STAGING FOR SMALL-CELL LUNG CANCER Table 3. Median Duration of Survival: Influence of Clinical Features
Variable Sex (female/male) Age ( 70 years/> 70 years) Weight loss >2 kg (no/yes) ECOG PS 0-1 2-3 DE LD ED LDS 1 2 No. sites involved (ED) 1 >1 Liver metostases No Yes Brain metastases No Yes Bone metastases No Yes Bone marrow metastases No Yes Nodes No Yes Pleural effusion No Yes SVC obstruction No Yes
No. of Patients Examined
Median Survival (weeks)
178/436 526/88 336/227
49/41 45/37 44/43
479 135
46 33
P (log-rank test)
.010
Relative Risk
.023 .060
1.29 1.32 1.17
.0001
1.49
286 328
