HYBRID GWBURR III SURVIVAL MODEL FOR PROSTATE CANCER: A COMPARATIVE MLE-BASED EVALUATION USING CLINICAL DATA FROM NIGERIA

Abstract

Accurate modeling of prostate cancer survival data requires flexible statistical distributions that
capture complex hazard structures and account for long-term survivors. In this study, we introduce
a hybrid Generalized Weibull–Burr III (GWBurr III) model and compare its performance with
three widely used flexible parametric alternatives: the Generalized Weibull (GW), Generalized
Log-normal (GL), and Generalized Exponential (GE) distributions. Parameter estimation across
all models was performed using Maximum Likelihood Estimation (MLE) to ensure statistical
efficiency and consistency. Model adequacy and comparative performance were assessed using a
comprehensive suite of evaluation metrics, including the Akaike Information Criterion (AIC),
Bayesian Information Criterion (BIC), log-likelihood values, goodness-of-fit test p-values, and
predictive accuracy measured by the area under the ROC curve. To demonstrate real-world
applicability, we analyzed survival data from oncology patients treated at the Ahmadu Bello
University Teaching Hospital (ABUTH), Zaria, Nigeria, a high-burden setting with substantial
prostate cancer mortality. The results show that the hybrid GWBurr III model provided a
substantially improved fit across all evaluation metrics and greater flexibility in modeling diverse
hazard trajectories. These findings highlight the value of hybrid distributional approaches for
oncology survival analysis and reinforce the methodological relevance of the GWBurr III model

for researchers working with complex, heterogeneous clinical data, particularly in resource-
limited and high-burden environments.