A Comparative Analysis of the TopLE Survival Model Using Lung Cancer Data

1Na’awurti William Nyandaiti, 2Isaac Esbond Gongsin 2Yusuf Abbakar Mohammed
1School of Health Information Management, University of Maiduguri Teaching Hospital, Nigeria
2Department of Statistics, University of Maiduguri

*Corresponding author’s Email: nnyandaiti@yahoo.com, doi.org/10.55639/607.02010077


ABSTRACT

This study introduces the Topp-Leone Epsilon (TopLE) distribution as a flexible parametric model for lung cancer survival data from the North Central Cancer Treatment Group (NCCTG) trial (n = 228). Compared to the exponential, log-logistic, and log-normal models, the TopLE model achieved the lowest Akaike Information Criterion (AIC = 2234.9) and highest log likelihood (−1107.4), making it the best among the standard models used in survival analysis. The TopLE model is also shown to have the highest discriminative ability (C-index = 0.742), and the lowest error metrics (IBS = 0.118; RMSE = 0.054). When extended with clinical covariates, male patients exhibited 48% higher mortality risk (HR = 1.48; 95% CI: 1.12 – 1.97), while age showed a marginal effect (HR = 0.98; p-value ≈ 0.05). Despite relatively wide confidence intervals for certain parameters, the TopLE model provided improved fit and visual agreement with Kaplan-Meier estimates. These findings suggest that the TopLE model is a robust, yet computationally sensitive, alternative for modelling complex hazard shapes in oncology survival data.

Keywords:

Censoring,
Clinical covariates,
Flexsurv,
Kaplan-Meier,
Oncology.