Innovative estimation of survival using log-normal survival modelling on ACCENT database

Area of Science:

• Oncology
• Biostatistics
• Clinical Trials

Background:

• The ACCENT database, comprising 20,898 patients from 18 colon cancer trials, supported FDA approval of 3-year disease-free survival (DFS) as a surrogate for 5-year overall survival (OS).
• Hypothesized substantive differences in survival estimation between log-normal modeling and standard Kaplan-Meier or Cox approaches.

Purpose of the Study:

• To compare survival estimations using log-normal, Kaplan-Meier, and Cox models in a specific colon cancer cohort.
• To evaluate the clinical relevance of differences in survival estimates derived from different statistical models.

Main Methods:

• Estimated time to relapse, DFS, and OS using Kaplan-Meier, Cox, and log-normal approaches.
• Analyzed data for male subjects aged 60-65 years with stage III colon cancer, comparing 5-fluorouracil-based chemotherapy (5FU) with surgery alone.

Main Results:

• Log-normal model yielded higher estimates for 3-year time to relapse (5.8-6.3%), 3-year DFS (4.8-5.1%), and 5-year OS (3.2-2.2%) compared to the Cox model.
• Model checking favored the log-normal model, with Cox and Kaplan-Meier estimates showing greater similarity.
• All models indicated a treatment benefit for 5FU, showing higher 3-year DFS (5.0-6.7%) and 5-year OS (5.3-6.8%).

Conclusions:

• Significant absolute differences exist between log-normal and Cox model estimates for 3-year DFS and 5-year OS.
• These differences are clinically relevant and warrant further consideration in survival analysis for colon cancer.