We analyzed the 2,580-patient Southwest Oncology Group (SWOG) small-cell lung cancer data base from 1976 to 1988 in order to (1) determine the prognostic value of favorable demographic and tumor-related factors and therapy programs using Cox multivariate analyses in limited- and extensive-stage disease (LD, ED), and (2) define patient subgroups with significantly different survivals using recursive partitioning and amalgamation (RPA) analysis to refine the current two-stage system. Cox multivariate models were applied to 1,363 patients in six LD trials: good performance status, female sex, age less than 70 years, white race, and normal lactate dehydrogenase (LDH) were significant favorable independent predictors. Concurrent chemoradiotherapy was also a strong independent predictor of survival. For 1,217 patients in four ED trials, a normal LDH, treatment with an intensive multidrug regimen, and a single metastatic lesion were favorable independent variables in the Cox model. RPA analysis of 1,137 patients in recent LD and ED trials resulted in a regression tree in which the most important prognostic split was LD versus ED. Normal or abnormal LDH, absence or presence of a pleural effusion, and age less than 70 or ≥70 years were important in LD, but only LDH was significant in ED. The terminal nodes of the regression tree were amalgamated to form four distinct prognostic subgroups with median survivals of 19.0, 12.5, 10.5, and 6.3 months (P < .0001). The best survival occurred for younger patients with 'true' LD: no effusion and normal LDH. The two intermediate patient subgroups had either LD or ED but still lived significantly longer than those patients with true ED (elevated LDH). This analysis suggests that although several factors were independent prognostic variables in LD in the Cox models, a smaller number of variables can be used to form important prognostic subgroups through RPA. The LDH emerged as a highly significant factor, but performance status and sex did not. A refinement of the current staging system should be made if our results can be confirmed with a combined-group data base analysis.
ASJC Scopus subject areas
- Cancer Research