Goksuluk Merve Basol, Goksuluk Dincer, Sipahioglu Murat Hayri
Department of Biostatistics, Faculty of Medicine, Sakarya University, Sakarya, Turkey.
Department of Biostatistics, Faculty of Medicine, Erciyes University, Kayseri, Turkey.
PLoS One. 2025 Jul 28;20(7):e0320385. doi: 10.1371/journal.pone.0320385. eCollection 2025.
This study investigates mortality risk prediction in peritoneal dialysis (PD) patients through longitudinal biomarker analysis, comparing traditional and advanced statistical approaches. A retrospective cohort of 417 PD patients followed up between 1995 and 2016 at Erciyes University was analyzed, with serum albumin, creatinine, calcium, blood urea nitrogen (BUN), and phosphorus assessed as predictors of all-cause mortality. Statistical methods included Cox proportional hazards models, time-dependent covariates, and joint modeling (univariate and multivariate) for longitudinal-survival data integration. Joint models outperformed baseline, averaged, and time-dependent methods, with multivariate joint modeling yielding the highest predictive accuracy by incorporating inter-biomarker relationships. Serum albumin emerged as the most consistent mortality predictor, while creatinine and phosphorus showed significance in specific contexts. Other biomarkers, such as calcium and BUN, were less predictive. Dynamic prediction capabilities of joint models demonstrated enhanced alignment with patient outcomes, underscoring their utility in personalized medicine. This study highlights the importance of integrating temporal changes and biomarker interdependencies into survival analysis to improve risk stratification and clinical decision-making in PD patients. Future research should explore the broader applicability of these methods across diverse chronic disease populations.
本研究通过纵向生物标志物分析，比较传统和先进的统计方法，调查腹膜透析（PD）患者的死亡风险预测。对1995年至2016年在埃尔西耶斯大学随访的417例PD患者的回顾性队列进行了分析，评估血清白蛋白、肌酐、钙、血尿素氮（BUN）和磷作为全因死亡率的预测指标。统计方法包括Cox比例风险模型、时间依赖协变量以及用于纵向生存数据整合的联合建模（单变量和多变量）。联合模型优于基线、平均和时间依赖方法，多变量联合建模通过纳入生物标志物间的关系产生了最高的预测准确性。血清白蛋白成为最一致的死亡预测指标，而肌酐和磷在特定情况下具有显著性。其他生物标志物，如钙和BUN，预测性较差。联合模型的动态预测能力显示出与患者预后的更强一致性，突出了它们在个性化医疗中的效用。本研究强调了将时间变化和生物标志物相互依赖性纳入生存分析以改善PD患者风险分层和临床决策的重要性。未来的研究应探索这些方法在不同慢性疾病人群中的更广泛适用性。
