Methods/Statistics
Analysis of Kidney Cancer Patients Using the SICHEL Regression Model within GAMLSS FrameworkAIM Tiago Almeida de Oliveira* Tiago Almeida de Oliveira Cleanderson Romualdo Fidelis Maria Carmen Escalante- Rojas Alessandro Gouveia Nunes Ana Patricia Bastos Peixoto
AIM: To evaluate kidney cancer progression using INCA clinical data and assess the SICHEL regression model’s flexibility and robustness within the GAMLSS framework for handling asymmetric, heterogeneous, and overdispersed data. Kidney cancer is a significant public health concern due to its high mortality rates and the complex nature of its progression. Accurate modeling of clinical data is crucial for improving prognosis, guiding treatment decisions, and enhancing patient outcomes. This study introduces the SICHEL regression model, applied within the GAMLSS framework. By identifying key clinical factors that influence survival times, the study provides practical insights into disease progression, offering the potential to improve prognostic assessments and tailor treatments more effectively. The study focused on patients who experienced the event (death), and only positive times (in days) were considered. The number of days between diagnosis and death served as the dependent variable. Covariates included age, sex, diagnostic method, tumor extension, and morphology. To assess the model’s performance, comparisons were made with Poisson and Negative Binomial models. These models were evaluated using the GAIC criterion to identify the best fit for the dataset. Results: For mean (μ) Parameter: Male sex (p = 0.5185) did not significantly affect time until death compared to females, while the “Malignant Neoplasia” morphology (p < 0.001) significantly reduced time until death compared to "Carcinoma." The "Investigation" diagnostic method (p = 0.0756) and "Metastasis" extension (p = 0.0117) significantly influenced time until death. Dispersion (σ) Parameter: Variations in data dispersion were observed, reflecting the impact of heterogeneity in the analysis. Shape (ν) Parameter: Male sex (p = 0.0304) was significantly associated with greater asymmetry in the time until death distribution, while "Malignant Neoplasia" (p < 0.001). The SICHEL model achieved the lowest GAIC value, indicating superior fit compared to Poisson and Negative Binomial models, due to its ability to accommodate data asymmetry, overdispersion, and variability. The SICHEL regression model, integrated with the GAMLSS framework, demonstrated flexibility and robustness in modeling times of kidney cancer patients.
