original article
Annals of Oncology 21: 556–561, 2010 doi:10.1093/annonc/mdp392 Published online 14 October 2009
Prognostic and predictive factors in a randomized phase III trial comparing cisplatin–pemetrexed versus cisplatin–gemcitabine in advanced non-small-cell lung cancer K. N. Syrigos1*, J. Vansteenkiste2, P. Parikh3, J. von Pawel4, C. Manegold5, R. G. Martins6, L. Simms7, K. P. Sugarman8, C. Visseren-Grul9 & G. V. Scagliotti10 1
Oncology Unit, Department of Medicine, Athens School of Medicine, Athens, Greece; 2Respiratory Oncology Unit, University Hospital Gasthuisberg, Leuven, Belgium; Department of Medical Oncology, BSES G Municipal Hospital and Railway Hospital, Mumbai, India; 4Department of Oncology, Asklepios-Fachkliniken Munchen Gauting, Gauting; 5Department of Surgery–Interdisciplinary Thoracic Oncology, Heidelberg University Medical Center, Mannheim, Germany; 6Division of Medical Oncology, University of Washington Medical Center, Seattle, WA, USA; 7Department of Statistics and Information Sciences, Eli Lilly Canada, Toronto, Canada; 8 Medical Oncology, Eli Lilly and Company, Indianapolis, IN, USA; 9Medical Oncology, Eli Lilly Netherlands, Utrecht, The Netherlands and 10Department of Clinical and Biological Sciences, Thoracic Oncology Unit, San Luigi Hospital, University of Turin, Turin, Italy 3
Received 14 April 2009; revised 8 June 2009; accepted 9 June 2009
original article
Background: Baseline patient and disease characteristics are investigated in non-small-cell lung cancer (NSCLC) in an effort to predict response to treatment and optimize patients’ outcomes. Histology has recently been identified in multiple NSCLC phase III trials as a predictive factor for survival in patients receiving pemetrexed regimens. Methods: Cox-adjusted models were used to further analyze a randomized phase III study in 1725 chemonaive patients with stage IIIB or IV NSCLC and Eastern Cooperative Oncology Group performance status (PS) of zero or one who received cisplatin plus pemetrexed (CP; C, 75 mg/m2 and P, 500 mg/m2) or cisplatin plus gemcitabine (CG; C, 75 mg/m2 and G, 1250 mg/m2) every 21 days. Results: Histology was confirmed to be predictive of CP efficacy and may also be prognostic. Gender, ethnicity, disease stage, smoking status, and PS were not predictive in either treatment arm but were shown to be prognostic in the nonsquamous population, consistent with the results in the overall NSCLC population. Conclusions: NSCLC histology significantly predicts efficacy outcomes for patients receiving pemetrexed. Several other factors are prognostic for the overall study population as well as a subset of patients with advanced nonsquamous NSCLC. Key words: histology, markers, non-small-cell lung cancer, predictive, prognostic
introduction To optimize treatment choices, clinical trial results frequently include information about potential predictive and prognostic factors. Prognostic factors are disease and patient characteristics that are associated with efficacy outcomes (usually survival) independent of the treatment administered. Predictive factors are clinical, cellular, and molecular markers that predict response to treatment (tumor shrinkage or extended survival). Hence, prognostic factors define the effects of patient and disease characteristics on efficacy outcomes, whereas predictive factors associate these characteristics with the effect of treatment on efficacy outcomes. In lung cancer, disease stage, performance status (PS), gender, and weight loss have traditionally been identified as strong prognostic factors, whereas histology has not been clearly or consistently identified *Correspondence to: Dr K. N. Syrigos, Oncology Unit GPP, Athens School of Medicine, Sotiria General Hospital, Mesogion 152 Avenue, 115 27 Athens, Greece. Tel: +302107700220; Fax: +30-210-7781035; E-mail:
[email protected]
as a prognostic or predictive variable in advanced non-small-cell lung cancer (NSCLC) studies [1–4]. A recent phase III, randomized, noninferiority study compared overall survival (OS) of the standard regimen cisplatin plus gemcitabine (CG) versus cisplatin plus pemetrexed (CP) in 1725 chemonaive patients with advanced NSCLC [5]. The primary end point of the trial was met, showing that OS for CP was noninferior to CG. In addition, disease stage, PS, and gender were found to be prognostic for survival. Ethnicity, smoking status, and NSCLC histology were also identified as prognostic factors [5]. Prespecified analyses demonstrated that histology was also predictive for CP efficacy, but not CG. Patients with nonsquamous histology had statistically superior OS with CP compared with CG, whereas patients with squamous cell carcinoma had shorter survival with CP compared with CG. In the CP arm, patients with nonsquamous histology had longer survival than patients with squamous cell carcinoma, whereas in the CG arm, survival differences by histology were not observed [5].
ª The Author 2009. Published by Oxford University Press on behalf of the European Society for Medical Oncology. All rights reserved. For permissions, please email:
[email protected]
original article
Annals of Oncology
In this report, we carry out additional statistical analyses on the data from the CG versus CP study [5] to determine whether factors other than histology might also be predictive of efficacy outcomes for CP or CG and to ascertain which baseline characteristics are prognostic for survival within histology subgroups.
methods patients Analyses were based on a study [5] in which patients were considered eligible according to the following criteria: chemonaive patients with histologically or cytologically confirmed stage IIIB/IV NSCLC, one or more measurable lesions [6], an Eastern Cooperative Oncology Group [7] PS of zero or one, adequate organ function, and age ‡18 years.
study design and treatment The noninferiority, phase III, randomized study compared OS between treatment arms using a fixed margin method for the hazard ratio (HR) of 1.17647. Patients received up to six cycles of cisplatin 75 mg/m2 on day 1 plus gemcitabine (GEMZAR; Eli Lilly and Company, Indianapolis, IN) 1250 mg/m2 on days 1 and 8 of a 21-day cycle or cisplatin 75 mg/m2 plus pemetrexed (ALIMTA; Eli Lilly and Company) 500 mg/m2 on day 1 of a 21-day cycle. Randomization [8] was done according to disease stage (IIIB versus IV), PS (0 versus 1), history of brain metastases (yes versus no), gender (male versus female), basis of diagnosis (histologic versus cytologic), and investigative center. Basis of diagnosis was included as a randomization factor to balance the enrollment of patients with available specimens for a companion pharmacogenomic study.
statistical methods Efficacy analyses incorporated all randomized patients on an intent-to-treat basis. Cox-adjusted analyses of OS were used, controlling for four randomization factors (PS, gender, disease stage, and basis of diagnosis) and treatment (CP versus CG). Median survival was estimated using the Kaplan–Meier method [9]. Predictive associations were concluded if there was a significant (P < 0.05) treatment-by-characteristic interaction. This interaction test determines whether the efficacy of a particular treatment varies according to the characteristic being considered. The interaction Cox model included treatment, the four randomization factors, the baseline characteristic under consideration, and a treatment-by-characteristic interaction term. A statistically significant treatment-by-characteristic interaction indicates that the survival benefit for CP relative to CG is significantly different for one subgroup of a given baseline characteristic (e.g. age ‡ 65) compared with its complementary subgroup (e.g. age < 65). To assess the prognostic role of the baseline characteristics, four additional variables (age, ethnicity, smoking status, and histology) were added separately to the Cox model. Assessment of histology as a prognostic factor required separate analyses for each treatment arm because of the significant treatment-by-histology interaction [5] that demonstrated that the efficacy of CP differed significantly between patients with nonsquamous and squamous tumors, whereas the efficacy of CG did not. Due to this interaction in which the effect of histology on survival is confounded by treatment, the evaluation of the prognostic effect of histology required that analyses be carried out separately for each arm. Statistical modeling for additional prognostic factors was also done, incorporating age, ethnicity, smoking status, and histology, one at a time, into the main Cox model (based on treatment assignment, PS, gender, disease stage, and basis of diagnosis). To assess robustness of the results, age was alternatively modeled as a continuous or dichotomous (
