Impact of Older Age on the Efficacy of Newer Adjuvant Chemotherapy Regimens in Colon Cancer, a Subgroup Analysis of a Meta-analysis: Practice Changing? Certainly Not; Hypothesis Generating? Perhaps

Abstract

With the overall aging US population and a median age at diagnosis of 72 years, colon cancer is a common disease in the elderly.1 Adjuvant therapy has been the standard of care for stage III disease since 1990.2 Population-based data derived from the Surveillance, Epidemiology, and End Results registry reveals a marked disparity in the administration of adjuvant chemotherapy between younger and older patients with colon cancer. In addition, when adjuvant chemotherapy is initiated, older individuals are more likely to have their treatment discontinued before completion, possibly decreasing its effectiveness.3 Randomized trials have demonstrated that adjuvant and palliative chemotherapies for all patients with colon cancer improve survival and quality of life.4 Unfortunately, due to the underrepresentation of elderly patients in clinical trials, robust evidence-based data for elderly patients with colon cancer is lacking. The existing literature mainly consists of pooled data sets from large studies. In most of these individual large trials, only 10%-20% of patients are elderly. However, at least half of the patients who are treated in the community setting are older than 70 years of age. Any increase in the effective use of adjuvant chemotherapy in this disease might further enhance the survival of stage III colon cancer.5 At this year’s meeting of the American Society of Clinical Oncology, McCleary et al6 presented data from the ACCENT (Adjuvant Colon Cancer End Points) database on 2170 patients with colon cancer aged 70 years or older (out of a total of 12,669 patients) who were treated in 6 phase III adjuvant colon cancer trials. They compared intravenous 5-fluorouracil (5-FU) monotherapy to combination regimens with irinotecan, oxaliplatin, or oral 5-FU. Based on their analysis, they concluded that patients aged 70 years or older did not receive the same clinical benefit compared with patients younger than age 70. The main endpoints of this study were overall survival (OS), disease-free survival, and time to recurrence. Cox models were stratified by age and adjusted for sex and stage. Interaction testing was used to explore the differential benefit by age. In other words, this was a subgroup analysis of a large meta-analysis data set from 6 previously conducted randomized phase III clinical trials. With all possible biases and shortcomings inherent to meta-analyses7 and subgroup analyses,8 how robust and conclusive is this data? Some immediate questions that come to mind include the following: • Were all relevant clinical trials identified and included in this meta-analysis? We know that the data from at least 2 large adjuvant trials—Avant and NSABP C-08—were not included. • Were all randomized patients included in the analysis in accordance with the treatment allocated at randomization? Was an “intention-to-treat” analysis performed? • How were the data for analyses gathered? Were they aggregate data or individual patient data in which the details for each participant in every trial were collected and analyzed centrally? • Were there additional follow-up data since the publication of each individual trial data? Or were the results frozen in time at the time of their initial publication? • Was the subgroup analysis in this presentation performed when randomized clinical trials were running or after they were completed? In other words, were these pre- or post-analysis subgroups? Subgroup analyses can be based on formal tests of interaction, but they should be interpreted with caution. Interaction testing might decrease the risk of finding false-positive subgroups, but its power is low.8 In a trial with 80% power for the overall treatment effect, the interaction test has, at most, a 29% power to detect an interaction effect of the same magnitude of the overall effect, thereby requiring a 4-fold increase in the sample size.9 Thus, is a meta-analysis data set using the ACCENT database sufficiently large to accomplish this goal despite including over 12,000 patients? Increasing age will almost always be associated with a poorer OS because of the probability of dying from noncancer causes. Although it may be viewed as a prognostic factor for OS, chronologic age alone should not be a predictive factor for the benefits of adjuvant chemotherapy in elderly patients with colon cancer. In an era of individualized medicine, with a burgeoning interest and recognition of various biologic and molecular markers in this disease, making a decision about adjuvant chemotherapy based on the chronologic age alone can be very misleading. Results of subgroup analysis of ACCENT database meta-analysis should not be over-interpreted. At this time, the findings from this study can and should not be viewed as practice changing; however, they may be viewed as hypothesis generating. The best way to directly