Modeling Survival Outcomes in Eye Cancer Patients Using Advanced Statistical and Machine Learning Approaches II

Abstract

Eye cancer remains one of the less frequent but clinically serious malignancies, with survival outcomes that are difficult to predict given the heterogeneity of tumor types and patient profiles. This study examines a dataset of 5,000 patients diagnosed with various ocular malignancies and applies three computational models. Survival Support Vector Machines (SVCR), Random Survival Forests (RSF), and the DeepSurv deep learning model to estimate time-to-event outcomes. Model accuracy was compared using the concordance index, time-dependent AUC, and Brier score. DeepSurv returned the strongest result with a C-index of 0.912, with tumor stage, patient age, and BRAF mutation status emerging as the most influential prognostic variables. These findings point to the practical utility of deep learning in ocular oncology, particularly for stratifying patients by risk prior to treatment decisions.

Introduction

The text focuses on improving survival prediction for rare but serious eye cancers such as uveal melanoma, retinoblastoma, and intraocular lymphoma. These cancers have diverse behaviors, making accurate survival prediction challenging. Traditional statistical methods like Kaplan–Meier estimator and Cox regression are widely used but struggle with complex, high-dimensional clinical data.

To overcome these limitations, the study applies advanced machine learning models, including Support Vector Machine for Survival, Random Survival Forest, and DeepSurv. These models can handle nonlinear relationships, large datasets, and interactions between clinical and genetic variables, enabling more accurate and individualized survival predictions.

The methodology involves cleaning and preprocessing clinical data, encoding variables, and splitting datasets for training and testing. Models are evaluated using metrics like the concordance index (C-index), time-dependent AUC, and Brier score to measure prediction accuracy and reliability.

Results show that DeepSurv performed best, significantly outperforming traditional Cox models. Key factors influencing survival include tumor stage, age, and genetic markers such as BRAF mutations, while treatments like radiation therapy showed protective effects.

Overall, the study demonstrates that machine learning provides more precise and clinically useful survival predictions, helping improve risk stratification and supporting better decision-making in ocular cancer treatment.

Conclusion

This study tested three machine learning models against a Cox regression baseline for survival prediction in eye cancer patients. DeepSurv outperformed the alternatives on all three evaluation metrics, which is consistent with its theoretical advantage on datasets where predictor effects are non-linear and interactive. RSF also performed well and has the added benefit of producing variable importance estimates that are straightforward to interpret clinically. That said, this analysis has limitations worth acknowledging. The dataset was sourced from Kaggle and may not reflect the full demographic and clinical diversity of real-world ocular oncology populations. The models were not validated on an external cohort, which is the standard requirement before any clinical deployment. BRAF mutation status, while influential in this dataset, is not consistently recorded across registries, which could limit reproducibility. Future work should focus on external validation using registry data such as SEER, incorporation of imaging-derived features, and exploration of time-varying covariates to better capture disease progression. Prospective validation in a clinical setting would be the necessary next step before these models could support actual treatment decisions.

References

[1] Kaplan EL, Meier P. Nonparametric Estimation from Incomplete Observations. Journal of the American Statistical Association. 1958;53(282):457–481. [2] Cox DR. Regression Models and Life-Tables. Journal of the Royal Statistical Society Series B. 1972;34(2):187–220. [3] Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random Survival Forests. The Annals of Applied Statistics. 2008;2(3):841–860. [4] Harrell FE Jr, Califf RM, Pryor DB, Lee KL, Rosati RA. Evaluating the Yield of Medical Tests. JAMA. 1982;247(18):2543–2546. [5] Uno H, Cai T, Pencina MJ, D’Agostino RB Sr, Wei LJ. On the C-statistics for Evaluating Overall Adequacy of Risk Prediction Procedures with Censored Survival Data. Statistics in Medicine. 2011;30(10):1105–1117. [6] Huang Z, Li W, Xu G, et al. A Nomogram to Predict Cancer-Specific Survival of Ocular Melanoma. Frontiers in Medicine. 2025;12:1638733. [7] Stålhammar G. Long-term Relative Survival in Uveal Melanoma: A Meta-Analysis. Communications Medicine. 2022;2:00082. [8] He X, Wang L, Chen Y, et al. Epidemiology and Survival Outcomes of Patients with Primary Intraocular Lymphoma. BMC Cancer. 2022;22:6183. [9] Bilbeisi SC, Brichko T, Chakraborty R, et al. Causes of Death and Survival Analysis for Patients with Retinoblastoma. Frontiers in Medicine. 2023;10:1244308. [10] Witzenhausen R, Damato B, Charteris D, et al. Survival Benefit of Primary Tumor Treatment in Uveal Melanoma. Cancers (Basel). 2024;16(11):638514. [11] Singh AD, Turell ME, Topham AK. Uveal melanoma: trends in incidence, treatment, and survival. Ophthalmology. 2011;118(9):1881–1885. [12] The Global Retinoblastoma Outcome Study Group. Global retinoblastoma survival and globe preservation: analysis of 4,064 patients from 153 countries. Lancet Global Health. 2022;10(6):e807–e815. [13] Eye cancer patient data set: Kaggle, “Eye Cancer Survival Analysis Dataset,” access year 2025. [14] SEER Program: Surveillance, Epidemiology, and End Results (SEER) Program. National Cancer Institute, US. [15] Frontiers in Medicine, BMC Cancer, Communications Medicine, and Cancers (Basel) journals for peer-reviewed survival analysis papers. [16] Official lifelines and scikit-survival documentation for survival modeling methodology. [17] The Global Retinoblastoma Outcome Study Group, 2022; Demkowicz M, et al., 2023; Zhang N, et al., 2025; Lee J, et al., 2024 for advanced eye cancer analytics and machine learning in survival prediction. [18] Katzman JL, Shaham U, Cloninger A, Bates J, Jiang T, Kluger Y. DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network. BMC Medical Research Methodology. 2018;18(1):24.

Copyright

Copyright © 2026 Rohit Bhuse, Shaurya Sawant. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.