Akademik Veri Yönetim Sistemi
BMC cardiovascular disorders, cilt.25, sa.1, ss.673, 2025 (SCI-Expanded, Scopus)
Abstract Background Heart failure (HF) remains a major global health burden with high mortality rates. Accurate survival prediction is essential for clinical decision-making. This study investigates how follow-up adequacy, quantified by Person-Time Follow-up Rate (PTFR), impacts the performance of survival models—specifically, Cox proportional hazards regression(CPHR) and Random Survival Forest (RSF). Methods A routinely collected health dataset of 299 HF patients was analyzed. PTFR was calculated using the formal method by Xue et al. (2022), resulting in a PTFR of 45.6%. A simulated version of the dataset was generated by proportionally extending follow-up times to increase PTFR to 67.2%. Both CPHR and RSF models were applied to the original and simulated datasets. Model performance was assessed using C-index and Area Under Curve(AUC). Results In the original dataset, the CPHR model achieved a C-index of 0.754 and AUC of 0.959, while the RSF model achieved a C-index of 0.884 and AUC of 0.988. In the simulated dataset, model performance improved slightly, with the improvement being more pronounced in RSF. It also more effectively identified clinically relevant predictors such as ejection fraction and serum creatinine. Increased PTFR led to better model stability and predictive accuracy. Conclusion Improving PTFR enhances the validity and robustness of survival models. RSF outperformed Cox regression across both datasets, particularly under higher PTFR. Strategies such as extending follow-up duration and integrating data sources can help increase PTFR. These findings underscore the importance of adequate follow-up in predictive modeling and support the use of machine learning in clinical survival analysis. Keywords Person-time follow up rate, Random survival forest, Machine learning, Heart failure, Survival analysis