Prognostic factors differ by tumor stage for small cell lung cancer^†

Abstract

BACKGROUND:

An analysis of 14 small cell lung cancer (SCLC) trials was performed to improve the current understanding of potential prognostic factors for overall survival (OS) and progression-free survival (PFS) in groups of patients with limited-stage disease SCLC (LD-SCLC) and extensive-stage disease SCLC (ED-SCLC) separately.

METHODS:

Data on 688 patients with LD-SCLC and 910 patients with ED-SCLC were included. Clinical and laboratory factors were tested for their prognostic significance using Cox regression models that were stratified by protocol. Recursive partitioning and amalgamation (RPA) analyses were used to identify prognostic subgroups.

RESULTS:

Poorer performance status (PS) led to worse OS and PFS in the ED-SCLC group but not in the LD-SCLC group. The prognostic impact of PS was strong for men but weak for women in the ED-SCLC group (interaction P value <.012 for OS and PFS). Other negative prognostic factors included increased age and men for the LD-SCLC group and increased age, men, increased number of metastatic sites at baseline, and increased creatinine levels for the ED-SCLC group. In patients with the ED-SCLC, RPA analyses identified 5 subgroups with different prognosis based on baseline PS, creatinine levels, sex, and the number of metastatic sites.

CONCLUSIONS:

The current pooled analysis identified baseline creatinine levels and the number of metastatic sites as important prognostic factors in patients with ED-SCLC in addition to the well established factors of sex, age, and PS. There was a significant interaction between sex and PS within the ED-SCLC group, suggesting that PS is highly prognostic in men but has no significant impact in women. Within the LD-SCLC group, only age and sex were identified as important prognostic factors. RPA analyses confirmed many of these findings. Cancer 2009. © 2009 American Cancer Society.

In 2008, it is expected that lung cancer will have caused 161,840 deaths in the US.1 Approximately 16% of these lung cancer patients are expected to have small cell lung cancer (SCLC).2 Without treatment, SCLC is considered the most aggressive of the lung tumors with a median survival ranging from 2 months to 4 months.3 The long-term prognosis for patients with SCLC is relatively poor, and only 5% to 10% are expected to live for at least 5 years after diagnosis.4-6

At the time of diagnosis, approximately 30% of patients with SCLC will have a tumor that is confined to the following areas: hemithorax of origin, the mediastinum, or the supraclavicular lymph nodes.3 The patients who have disease that is limited to these areas and that can be encompassed within reasonable radiotherapy fields have limited-stage disease SCLC (LD-SCLC).3 With currently available treatments, the median survival for patients with LD-SCLC ranges from 16 months to 26 months.7-9 Patients with tumors that have metastasized beyond the supraclavicular areas have extensive-stage disease SCLC (ED-SCLC).3 With currently available treatment options, the median survival varies from 6 months to 12 months for patients with ED-SCLC.3

It was demonstrated previously that good performance status (PS), young age, being a woman, and limited-stage disease are associated with an improved prognosis.6, 10-15 In addition, other variables, including elevated serum lactate dehydrogenase (LDH) levels, liver metastases, low albumin levels, and low sodium levels, have been associated with a poor prognosis.6, 10-14 Risk categories that include stage, PS, and several laboratory tests, such as LDH, alkaline phosphatase, and sodium level, also have been developed.11, 14 In addition, a recursive partitioning and amalgamation (RPA) analysis was used to establish 4 prognostic subgroups based on stage, PS, age, and sex using a large database of patients with SCLC.13

We set out to investigate and improve our understanding of the impact of several baseline patient and tumor characteristics on overall survival (OS) and progression-free survival (PFS) separately for patients with LD-SCLC and patients with ED-SCLC. Specifically, we investigated the impact of several pretreatment factors above and beyond the previously explored factors of age, sex, and PS. A large pooled analysis such as this allowed us to assess the consistency of such correlations between prognostic factors and outcomes across a large number of trials rather than from 1 or 2 large studies.

MATERIALS AND METHODS

Trial Characteristics

Individual patient data were pooled from 14 North Central Cancer Treatment Group (NCCTG) first-line SCLC therapy trials that opened between 1987 and 2001. Of those 14 SCLC trials, 9 trails enrolled only patients with ED-SCLC, 3 trials enrolled only patients with LD-SCLC, and 2 trials enrolled both patients with ED-SCLC and patients with LD-SCLC. All patients who received no study treatment or who were ineligible for trial participation were excluded from these analyses. In addition, if a patient was enrolled in more than 1 trial, then only data from the first trial was analyzed for that patient. The current analysis included a total of 910 patients with ED-SCLC and 688 patients with LD-SCLC. For a detailed listing of the individual trial characteristics, see Tables 1 and 2.

Table 1. Patients With Extensive-Stage Small Cell Lung Cancer (Total, N=910)

	Protocol (Reference)
Variable	862051 (Maksymiuk 199425)	882002 (Frytak 199626)	892051 (Rowland 199627)	912052 (Kirschling 199928)	922051 (Marschke 199929)	932051*	932053 (Schaefer 200330)	952052 (Perez 200431)	972052 (Jett 200332)	982052 (Molina 200633)	N0027*
No. eligible	299	31	237	42	14	66	60	54	43	38	26
Date opened	Mar 1987	Mar 1988	Aug 1990	Sept 1992	Oct 1993	Mar 1994	Oct 1994	Mar 1996	Mar 1998	Nov 1999	Nov 2001
Date closed	Aug 1990	Dec 1991	Jul/1993	Oct 1993	Dec 1994	May 1996	Feb 1998	May 2000	Jan 1999	Nov 2000	Jan 2003
Phase	3	2	3	2	2	2	2	2	2	2	2
PS	0-2	0-2	0-2	0-2	0-2	0-2	0-2	0-2	0-2	1-2	0-2
Agents†	A-D: Etoposide and cisplatin, CAV; patients with CR outside the chest, TRT and PCI; patients in CR only, recombinant γ-IFN or observation	A: Cytoxan, etoposide, and cisplatin; for responders, hyperfractionated TRT and PCI (or brain RT, or nothing) plus cytoxan and etoposide	A: Etoposide, cisplatin, and MEGACE ± whole-brain RT; B: Etoposide, cisplatin, and placebo±whole- brain RT	A: Paclitaxel and G-CSF; B: Paclitaxel and G-CSF, cisplatin and etoposide	A: Somatostatin	A: Paclitaxel and cisplatin	A: Topotecan, cisplatin, etoposide	A: Etoposide and paclitaxel	A: Etoposide and cisplatin, topotecan, paclitaxel, and G-CSF	A: Topotecan, paclitaxel, and G-CSF	A: Topotecan, carboplatin, and G-CSF
Factors collected
Hgb	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	No
WBC	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	No
PLT	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Bilirubin	Yes	No	Yes	Yes	Yes	Yes	Yes	No	No	Yes	Yes
Creatinine	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	No	Yes	Yes
BMI	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

• PS indicates performance status; CAV, cyclophosphamide, doxorubicin, and vincristine; CR, complete response; TRT, thoracic radiation therapy (physician discretion); PCI, prophylactic cranial irradiation (physician discretion); IFN, interferon; RT, radiotherapy; MEGACE, megestrol acetate; G-CSF, granulocyte–colony-stimulating factor (filgrastim or Neupogen); Hgb, hemoglobin; WBC, white blood cell count; PLT, platelets; bilirubin, total bilirubin; BMI, body mass index.
• * Unpublished data.
• † Patients with central nervous system progression only were generally allowed whole-brain radiation therapy.

Table 2. Patients With Limited-Stage Small Cell Lung Cancer (Total, N=688)

	Protocol (Reference)
Variable	862051 (Maksymiuk 199425)	882002 (Frytak 199626)	892052 (Bonner 199934)	952053 (Schild 200735)	N9923 (Garces 200736)
No. eligible	253	29	307	76	23
Date opened	Mar 1987	Mar 1988	Sept 1990	Oct 1996	Feb 2001
Date closed	Aug 1990	Dec 1991	Nov 1996	Mar 1999	Oct 2005
Phase	3	2	3	2	1-2 (MTD patients only)
PS	0-2	0-2	0-2	0-2	0-2
Agents	A-D: Etoposide and cisplatin, CAV; for stable or responding patients, TRT and PCI; for CR patients only, recombinant γ-IFN or observation	A: Cytoxan, etoposide, and cisplatin; for responders, hyperfractionated TRT and PCI (or brain RT, or nothing) plus cytoxan and etoposide	A: Etoposide and cisplatin, standard TRT±PCI plus etoposide and cisplatin; B: Etoposide and cisplatin, AHSCTRT±PCI plus etoposide and cisplatin	A: Etoposide and cisplatin, PCI; if regression on chest x-ray, then AHSCTRT plus etoposide and cisplatin, followed by etoposide and cisplatin alone	A: Topotecan, paclitaxel, and G-CSF. TRT plus etoposide, cisplatin, amifostine, topotecan, paclitaxel, and G-CSF, PCI
Factors collected
Hgb	Yes	Yes	Yes	Yes	No
WBC	Yes	Yes	Yes	Yes	No
PLT	Yes	Yes	Yes	Yes	Yes
Bilirubin	Yes	No	Yes	No	Yes
Creatinine	Yes	Yes	Yes	No	Yes
BMI	Yes	Yes	Yes	Yes	Yes

• MTD indicates maximum tolerated dose; PS, performance status; CAV, cyclophosphamide, doxorubicin, and vincristine; TRT, thoracic radiation therapy; PCI, prophylactic cranial irradiation; CR, complete response; IFN, interferon; RT, radiation therapy; ±, with or without; AHSCTRT, accelerated, hyperfractionated, split-course thoracic radiation therapy; G-CSF, granulocyte–colony-stimulating factor (filgrastim); Hgb, hemoglobin; WBC, white blood cell count; PLT, platelets; bilirubin, total bilirubin; BMI, body mass index.

Statistical Analysis

All analyses were performed separately in patients with LD-SCLC and patients with ED-SCLC. OS was defined as the time from registration to death from any cause. PFS was defined as the time from registration to the first of either death from any cause or disease progression. OS and PFS endpoints were censored at 5 years for analysis. A Cox proportional hazards model16 was used for both univariate and multivariate analyses. Score P values and likelihood-ratio P values were reported for the univariate and multivariate models, respectively. The models were stratified by protocol, and analyses were carried out on the data available based on the selected covariates for the respective endpoint. An RPA analyses with 10-fold cross validation was used to identify different prognostic subgroups for OS and PFS. Sample sizes of 688 patients and 910 patients, as a general guide, provided at least 80% and 90% power, respectively, to detect an effect reflected by a hazards ratio (HR) of 1.25 for a 2-level factor with a prevalence of 40% versus 60% (2-sided log-rank test; alpha level = .05) using the actual accrual rates from 1987 through 2005 and assuming exponential survival with 2 years of minimum follow-up on each patient.

The pretreatment variables identified at the time of the current analysis included factors that were collected across most trials that either were reported in the literature previously as prognostic for SCLC or were included in our previous NCCTG-led, pooled analysis in nonsmall cell lung cancer.17 The following pretreatment factors were assessed in univariate models and were considered for inclusion in multivariate models: age, sex, PS, body mass index (BMI), creatinine level (mg/dL), hemoglobin level (g/dL), white blood cell count (WBC) (×10⁹/L), platelet count (×10⁹/L), total bilirubin levels (mg/dL), and the number of metastatic sites at baseline for patients with ED-SCLC. Because the objective of this study was to identify additional prognostic factors for OS and PFS above and beyond the known clinical factors of age, sex, and PS, we explored 2 different multivariate models for each endpoint (OS and PFS) and for disease stage. The first model was considered the base clinical model and included only age, sex, and PS. The second model included the base model along with additional factors that were identified as potential predictors in the univariate setting (P < .20). For the factors that were included in the multivariate models, all 2-way interactions were assessed using stepwise regression modeling techniques.

The continuous factors that deviated substantially from linearity were modeled by using clinically meaningful cutoff points. The factors that were included in the multivariate models were tested for the appropriateness of the proportional hazards assumption18 and for between-trial heterogeneity. Model discrimination (ie, the ability to discriminate patients with different survival) was evaluated using the concordance index (C-Index).19 This is a measure of predictive accuracy for time to event data that contain censored observations. All tests were 2-sided, with P values <.05 for main effects and P values <.02 for 2-way interaction terms denoting statistical significance. In addition, HRs and the associated 95% confidence intervals (95% CIs) are reported for the univariate and multivariate results.

RESULTS

Data were frozen on November 19, 2007, and a total of 1598 eligible patients received first-line treatment, including 688 patients with LD-SCLC and 910 patients with ED-SCLC. Data are complete, with 98% of patients who had ED-SCLC and 81% of patients who had LD-SCLC followed until death. The median OS ranged from 17.2 months to 26.4 months for LD-SCLC trials and from 2.6 months to 12.3 months for ED-SCLC trials. The median PFS ranged from 10.9 months to 18.2 months for LD-SCLC trials and from 1.1 months to 8.4 months for ED-SCLC trials. Across all patients, disease stage was highly prognostic for both OS and PFS, where ED-SCLC patients had a worse prognosis (P < .0001).

Baseline Patient Characteristics

Table 3 provides a description of the patient characteristics for the full cohort. The LD-SCLC and ED-SCLC cohorts had similar patient characteristics, except for PS. Fifty-two patients (8%) with LD-SCLC had a PS of 2, where 222 patients (24%) with ED-SCLC had a PS of 2. The patient characteristics for the full cohort of 910 ED-SCLC patients was similar to the cohort of 716 patients used in the multivariate models (data not shown). Likewise, the full cohort of 688 patients with LD-SCLC was similar to the cohort of 574 patients used in the multivariate models (data not shown).

Table 3. Baseline Patient Characteristics

	No. of Patients (%)
Factor	ED-SCLC, N=910	LD-SCLC, N=688
Median age [range], y	64 [30-85]	63 [36-81]
PS
0	200 (22)	294 (42.7)
1	488 (53.6)	342 (49.7)
2	222 (24.4)	52 (7.6)
Sex
Women	359 (39.5)	307 (44.6)
Men	551 (60.6)	381 (55.4)
Metastatic sites at baseline
Missing	31 (3.4)	—
0/1	480 (52.8)	—
≥2	399 (43.9)	—
Median BMI [range], kg/m^2	25.9 [15.3-57.3]	26.1 [14.6-48.6]
Missing	4 (0.4)	0 (0)
Median WBC [range], ×10^9/L	9.0 [3.1-30.6]	8.2 [4.0-24.9]
Missing	65 (7.1)	24 (3.5)
Median Hgb [range], g/dL	13.7 [8.7-19.0]	13.8 [9.8-17.6]
Missing	65 (7.1)	25 (3.6)
Median total bilirubin [range], mg/dL	0.5 [0-2.4]	0.5 [0-2.6]
Missing	130 (14.3)	114 (16.6)
Creatinine level*
Missing	103 (11.3)	85 (12.4)
≤UNL	735 (80.8)	535 (77.8)
>UNL	72 (7.9)	68 (9.9)
Median creatinine [range], mg/dL	1.0 [0.1-2.0]	1.0 [0-2.3]
Missing	103 (11.3)	85 (12.4)
Median platelets, 10^9/L	305 [85-883]	317 [109-971]
Missing	5 (0.5)	3 (0.4)

• ED indicates extensive-stage disease; SCLC, small cell lung cancer; LD, limited-stage disease; PS, performance status; BMI, body mass index; WBC, white blood cell count; Hgb, hemoglobin; UNL, upper normal limit.
• * Creatinine was categorized as follows: for men, >1.3 mg/dL (UNL) versus ≤1.3 mg/dL; for women, >1.1 mg/dL (UNL) versus ≤1.1 mg/dL.

Univariate Analysis

For patients with LD-SCLC, total bilirubin, creatinine, sex, and age were promising prognostic factors for OS (P < .20); and total bilirubin, sex, and age were potential prognostic factors for PFS (P < .20). Although PS was not significant in the univariate models, it was explored in the multivariate models based on its prognostic importance from previous studies.6, 10-15 Table 4 provides the univariate results for LD-SCLC.

Table 4. Univariate Results for Patients With Limited-Stage Small Cell Lung Cancer

	OS			PFS
Factor	HR	95% CL	P*	HR	95% CL	P*
Total bilirubin, mg/dL†	1.37	0.97-1.95	.08	1.39	0.96-2.00	.08
Age (10-y increase)	1.20	1.09-1.33	.0004‡	1.19	1.07-1.31	.0007‡
PS
1 vs 0	1.07	0.90-1.28	.39	0.97	0.81-1.15	.82
2 vs 0	1.25	0.89-1.75		1.06	0.76-1.49
Hgb, g/dL†	0.98	0.92-1.04	.44	1.00	0.94-1.06	.97
PLT, ×10^9/L†	0.98	0.90-1.06	.57	0.99	0.91-1.08	.80
WBC, ×10^9/L†	1.01	0.98-1.04	.57	1.01	0.97-1.04	.66
BMI, kg/m^2†	1.01	0.99-1.02	.53	1.00	0.99-1.02	.67
Sex
Men vs women	1.24	1.05-1.47	.01‡	1.22	1.03-1.44	.02‡
Creatinine, mg/dL†	1.29	0.93-1.80	.13	1.11	0.80-1.52	.54

• OS indicates overall survival; PFS, progression-free survival; HR, hazard ratio; 95% CL, 95% confidence limits; PS, performance status; Hgb, hemoglobin; PLT, platelets; WBC, white blood cell count, BMI, body mass index.
• * The score P value derived from a univariate Cox model stratified by study.
• † HR for a 1-unit increase.
• ‡ Significant P value.

For patients with ED-SCLC, possible predictors for OS and PFS included age, PS, WBC, sex, creatinine level, and the number of metastatic sites at baseline (all P < .20). Table 5 provides the univariate results for ED-SCLC.

Table 5. Univariate Results for Patients With Extensive-Stage Small Cell Lung Cancer

	OS			PFS
Factor	HR	95% CL	P*	HR	95% CL	P*
Total bilirubin, mg/dL†	1.11	0.87-1.42	.41	1.03	0.78-1.32	.84
Age (10-y increase)	1.15	1.06-1.25	.001‡	1.07	0.99-1.17	.09
PS
1 vs 0	1.29	1.08-1.53	<.0001‡	1.24	1.05-1.47	.0004‡
2 vs 0	1.76	1.44-2.15		1.49	1.22-1.81
Hgb, g/dL†	0.99	0.95-1.04	.76	0.97	0.93-1.02	.28
PLT, ×10^9/L†	0.97	0.91-1.03	.28	1.01	0.95-1.07	.86
WBC, ×10^9/L†	1.02	1.00-1.04	.08	1.02	1.00-1.05	.02‡
BMI, kg/m^2†	1.00	0.98-1.01	.57	1.00	0.99-1.01	.92
Sex
Men vs women	1.20	1.05-1.38	.01‡	1.18	1.02-1.35	.02‡
Creatinine level§
>UNL vs ≤UNL	1.50	1.17-1.92	.001‡	1.23	0.96-1.58	.10
No. of metastatic sites at baseline
≥2 vs 0 or 1	1.41	1.23-1.62	<.0001	1.42	1.24-1.63	<.0001‡

• OS indicates overall survival; PFS, progression-free survival; HR, hazard ratio; 95% CL, 95% confidence limits; PS, performance status, Hgb, hemoglobin; PLT, platelets; WBC, white blood cell count; BMI, body mass index; UNL, upper normal limit.
• * The score P value derived from a univariate Cox model stratified by study.
• † HR a for 1-unit increase.
• ‡, † Significant P value.
• § Creatinine was categorized as follows: for men, >1.3 mg/dL (UNL) vs ≤1.3 mg/dL and for women >1.1 mg/dL (UNL) vs ≤1.1 mg/dL.

Multivariate Analysis

For patients with LD-SCLC, age and sex were highly prognostic for OS and PFS in both the base model (age, sex, and PS) and the full model (age, sex, PS, total bilirubin, and creatinine), where older patients and men had a worse prognosis. Table 6 provides the multivariate results for LD-SCLC. The RPA analyses validated age as a strong predictor in patients with LD-SCLC for both OS and PFS. No other factors were identified as important.

Table 6. Multivariate Results for Patients With Limited-Stage Small Cell Lung Cancer

	OS						PFS
	Base Model			Full Model			Base Model			Full Model
Factor	HR	95% CL	P*	HR	95% CL	P*	HR	95% CL	P*	HR	95% CL	P*
Age (10-y increase)	1.20	1.08-1.35	.001†	1.20	1.07-1.34	.002†	1.17	1.05-1.31	.003†	1.17	1.05-1.30	.004†
Sex
Men vs women	1.30	1.08-1.57	.005†	1.28	1.05-1.57	.02†	1.26	1.05-1.52	.01†	1.23	1.01-1.49	.04†
PS
1 vs 0	1.06	0.88-1.29	.057	1.06	0.88-1.29	.053	.96	0.79-1.15	.19	0.96	0.79-1.16	.17
2 vs 0	1.57	1.10-2.22		1.58	1.11-2.24		1.33	0.94-1.89		1.36	0.95-1.93
Total bilirubin, mg/dL‡				1.18	0.82-1.68	.39				1.23	0.83-1.80	.11
Creatinine, mg/dL‡§				0.99	0.69-1.43	.96

• OS indicates overall survival; PFS, progression-free survival; HR, hazard ratio; 95% CL, 95% confidence limits; PS, performance status.
• * Likelihood-ratio P value.
• † Significant P value.
• ‡ HR for a 1-unit increase.
• § Not significant at P = .20 in the PFS univariate model.

For patients with ED-SCLC, age, sex, and PS were highly prognostic for OS in both the base model (age, sex, and PS) and the full model (age, sex, PS, WBC, creatinine level, and the number of metastatic sites at baseline). Specifically, patients who were older, men, and those with a PS >0 had significantly worse OS. In addition, the number of metastatic sites at baseline and the creatinine level also were significant prognostic factors for OS. Specifically, patients who had ≥2 metastatic sites at baseline had a significantly worse OS (HR, 1.27; 95% CI, 1.09-1.48 [P = .002]), and patients who had increased creatinine levels at baseline (above the upper normal limit [UNL]) had a significantly worse OS (HR, 1.33; 95% CI, 1.03-1.71 [P = .04]). For PFS, being a man, a PS >0, and increased number of metastatic sites at baseline were associated significantly with a worse outcome. Table 7 provides the multivariate results for ED-SCLC.

Table 7. Multivariate Results for Patients With Extensive-Stage Small Cell Lung Cancer

	OS						PFS
	Base Model			Full Model			Base Model			Full Model
Factor	HR	95% CL	P*	HR	95% CL	P*	HR	95% CL	P*	HR	95% CL	P*
Age (10-y increase)	1.15	1.05-1.27	.003†	1.14	1.04-1.26	.006†	1.06	0.97-1.17	.21	1.06	0.96-1.17	.22
Sex
Men vs women	1.27	1.08-1.48	.003†	1.26	1.08-1.48	.003†	1.23	1.05-1.44	.009†	1.20	1.03-1.41	.02†
PS
1 vs 0	1.21	1.00-1.47	<.0001†	1.19	0.99-1.44	.0001†	1.19	0.99-1.43	.005†	1.15	0.95-1.39	.007†
2 vs 0	1.68	1.34-2.10		1.61	1.29-2.02		1.56	1.25-1.94		1.43	1.14-1.80
WBC, ×10^9/L‡				1.02	0.99-1.04	.17				1.02	1.00-1.04	.06
Creatinine level§
>UNL vs ≤UNL				1.33	1.03-1.71	.04†				1.12	0.86-1.44	.41
No. of metastatic sites
≥2 vs 0 or 1				1.27	1.09-1.48	.002†				1.34	1.15-1.57	.0002†

• OS indicates overall survival; PFS, progression-free survival; HR, hazard ratio; 95% CL, 95% confidence limits; PS, performance status, WBC, white blood cell count; UNL, upper normal limit.
• * Likelihood-ratio P value.
• † Significant P value.
• ‡ HR for a 1-unit increase.
• § Creatinine was categorized as follows: for men, >1.3 mg/dL (UNL) vs ≤1.3 mg/dL and for women >1.1 mg/dL (UNL) vs ≤1.1 mg/dL.

The RPA analyses for OS in patients with ED-SCLC, identified the number of metastatic sites at baseline as the most important predictor, where patients who had ≥2 metastatic sites had an increased risk of a poor outcome. Additional factors that were identified as important included sex, PS, and creatinine levels at baseline. Five groups with differing prognoses were identified (see Fig. 1): Group 1 (median OS, 10.6 months), patients who had 0 or 1 metastatic sites at baseline and creatinine values ≤UNL; Group 2 (median OS, 9.7 months), women who had ≥2 metastatic sites at baseline; Group 3 (median OS, 8.8 months), men who had ≥2 metastatic sites at baseline and a PS of 0 or 1; Group 4 (median OS, 7.5 months), patients who had 0 or 1 metastatic sites at baseline and creatinine values >UNL; and Group 5 (median OS, 5.8 months), men who had ≥2 metastatic sites at baseline and a PS of 2. Except age, the RPA analyses identified the same significant prognostic factors as the Cox model for OS. For PFS, in patients with ED-SCLC, the RPA analyses only identified the number of metastatic sites at baseline as an important prognostic factor.

Figure 1

Recursive partitioning and amalgamation analyses identified 5 groups for overall survival (OS) among patients with small cell lung cancer who had extensive-stage disease: Group 1 (blue), 0 or 1 metastatic sites at baseline and creatinine value equal to or less than the upper normal limit (UNL); Group 2 (red), women with ≥2 metastatic sites at baseline; Group 3 (green), men with ≥2 metastatic sites at baseline and a performance status (PS) of 0 or 1; Group 4 (yellow), 0 or 1 metastatic sites at baseline and creatinine value >UNL; and Group 5 (black), men with ≥2 metastatic sites at baseline and a PS of 2.

Two-way Interactions

No 2-way interactions were identified as significant for LD-SCLC. In patients with ED-SCLC, the interaction between sex and PS was significant for both OS and PFS (P < .012) (Figs. 2, 3), suggesting that the effect of PS on outcome differed by sex. Among men, for both OS and PFS, patients who had a PS of 0 had the best prognosis, patients who had a PS of 2 had the worst prognosis, and patients who had a PS of 1 were in the middle (P < .0001). Among women, PS was not associated significantly with OS (P = .12) or PFS (P = .21).

Figure 2

Interaction plot for sex by performance status (PS) for overall survival (OS) in patients with small cell lung cancer who had extensive-stage disease.

Figure 3

Interaction plot for sex by performance status (PS) for progression-free survival (PFS) in patients with small cell lung cancer who had extensive-stage disease.

Model Diagnostics

All continuous factors met the linearity assumption except the creatinine level in patients with ED-SCLC. The creatinine level was grouped using clinically meaningful cutoff points for men (>1.3 mg/dL [UNL] vs ≤1.3 mg/dL) and for women (>1.1 mg/dL [UNL] vs ≤1.1 mg/dL). These UNL values were taken from the Mayo Clinic 2008 calendar, which contains the 2008 Mayo Clinic normal values by sex. All factors met the proportional hazards assumption. For the homogeneity assumptions, the minor deviations were quantitative with the HRs for each factor and each study mostly in the same direction as the overall effect. For factors and studies that demonstrated the reverse effect, the individual CIs contained 1 (suggesting a nonsignificant effect).

The additional factors that were added to the base model for LD-SCLC did not result in an increase in the C-Index. For OS in the ED-SCLC population, the full multivariate model had a 3.4% increase in the C-Index compared with the base model (increased from 0.58 to 0.60). For PFS, the full model had a 3.6% increase in the C-Index compared with the base model (increased from 0.56 to 0.58). Including the significant PS by sex interaction in the full ED-SCLC models increased the C-Index by an additional 2% for OS and PFS (increased from 0.60 to 0.61 for OS and from 0.58 to 0.59 for PFS).

DISCUSSION

There has been little progress in the treatment of SCLC in the last decade. This disease is generally chemosensitive, and multiagent chemotherapy has been effective in increasing OS.10 In both LD-SCLC and ED-SCLC, patients who respond to systemic therapy appear to benefit in terms of OS with the addition of prophylactic cranial irradiation.20, 21 In addition, there is strong evidence that thoracic radiotherapy improves the OS of patients with LD-SCLC and of select patients with ED-SCLC who respond very favorably to the initial chemotherapy.22-24 Despite these therapeutic findings associated with improved OS, the vast majority of patients with SCLC do die from their disease.

In the current pooled analysis, we identified baseline creatinine levels and the number of metastatic sites as important prognostic factors in patients with ED-SCLC in addition to the well established factors of sex, age, and PS. In addition, there was a significant sex-by-PS interaction in patients with ED-SCLC. Specifically, among women, there was no significant difference in outcome by PS; where, among men, those with a PS of 0 had the best prognosis, followed by a PS of 1, and by a PS of 2. The RPA analyses also identified 5 subgroups of ED-SCLC patients with different prognoses based on baseline PS, creatinine levels, sex, and the number of metastatic sites. Among the patients with LD-SCLC, only age and sex were identified as important prognostic factors, where age also was confirmed in the RPA analyses.

Increased patient age and being a man were associated with significantly worse outcomes both in patients with ED-SCLC and in patients with LD-SCLC. This is consistent with previous reports.10, 13, 15 PS also was a significant prognostic factor in our ED-SCLC patients (P < .01 for OS and PFS) but had only borderline significance in our LD-SCLC patients (OS, P = .053; PFS, P = .17). This latter finding may have been caused in part by the lack of patients with a PS of 2 in the LD-SCLC series because of the eligibility criteria used in these trials and by possible hesitation on the part of treating oncologists to enroll patients with a PS of 2 in aggressive multimodality trials. PS as a prognostic factor also is consistent with the literature.6, 10-15

For patients with ED-SCLC, an increased number of metastatic sites at baseline also was a significant adverse prognostic factor. This also is consistent with some previous studies.6, 10 In addition, baseline creatinine levels were identified as prognostic for OS (independent of age), where patients who had increased creatinine levels (>UNL) had a significantly worse OS compared with patients who had lower levels (≤UNL). One possible reason for this is that patients with impaired renal function may have poorer tolerance to chemotherapy. These patients would have poorer renal clearance of the drug and decreased renal production of erythropoietin, both of which could cause more bone marrow suppression. To evaluate this hypothesis, we assessed the relation between the amount of chemotherapy administered and renal function and observed that those with compromised renal function (creatinine >UNL) at baseline received significantly fewer chemotherapy treatment cycles than those with normal renal function in the ED-SCLC population (P < .001), but not in the LD-SCLC population (P = .73). This may be a reason why increased creatinine was identified as an adverse prognostic factor in the ED-SCLC population. However, this observation should be verified in future studies.

The RPA analyses for OS in patients with ED-SCLC confirmed that increased number of metastatic sites at baseline yielded worse outcomes; and so did being a man, poorer PS, and elevated creatinine levels. Five subgroups with different prognoses were identified based on these factors. One limitation of this pooled analysis was that not all factors were collected consistently at baseline across all 14 trials. In addition, factors such as serum albumin and LDH, which have demonstrated prognostic importance in previous analyses, were not collected in these trials and, hence, were not evaluated.

Future progress will require better therapy. This likely will include improvements in systemic therapy (eg, targeted agents, immunotherapy) based on a better molecular understanding of this disease, better methods of irradiating patients, and possibly adding other treatment modalities. In each case, novel therapeutic combinations will need to be piloted and compared with standard therapies. Historic data become critical for these comparisons, thus making appropriate study design vital to answering these scientific questions. Proper study design, in turn, requires a clear understanding of the clinical factors associated with outcome. Without proper stratification for these prognostic factors, trial results can be misleading because of biases and imbalances within treatment arms. We have reviewed our recent experience treating SCLC in an attempt to clarify these factors further so that our future trials can yield useful data, and historic data can be put in perspective. In addition to the well established prognostic factors of age, sex, and PS, the current study has demonstrated that it may also be critical to stratify or adjust for baseline creatinine levels and the number of metastatic sites at baseline for patients with ED-SCLC.

Conflict of Interest Disclosures

Supported by Grant CA-25224.