The joint modeling approach with a simulation study for evaluating the association between the trajectory of serum albumin levels and mortality in peritoneal dialysis patients


Creative Commons License

BAŞOL GÖKSÜLÜK M., GÖKSÜLÜK D., SİPAHİOĞLU M. H., Karaagaoglu E.

HACETTEPE JOURNAL OF MATHEMATICS AND STATISTICS, vol.51, no.3, pp.900-913, 2022 (SCI-Expanded) identifier identifier identifier

  • Publication Type: Article / Article
  • Volume: 51 Issue: 3
  • Publication Date: 2022
  • Doi Number: 10.15672/hujms.884326
  • Journal Name: HACETTEPE JOURNAL OF MATHEMATICS AND STATISTICS
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, zbMATH, TR DİZİN (ULAKBİM)
  • Page Numbers: pp.900-913
  • Keywords: joint modeling, longitudinal model, survival model, personalized medicine, simulation, CENSORED SURVIVAL-DATA, EXPERIENCE, OUTCOMES
  • Hacettepe University Affiliated: Yes

Abstract

We aimed to study the association between mortality and trajectory of serum albumin levels (g/dL) in peritoneal dialysis patients via a joint modeling approach. Joint modeling is a statistical method used to evaluate the relationship between longitudinal and time-to event processes by fitting both sub-models simultaneously. A comprehensive simulation study was conducted to evaluate model performances and generalize the findings to more general scenarios. Model performances and prediction accuracies were evaluated using the time-dependent ROC area under the curve (AUC) and Brier score (BS). According to the real-life dataset results, the trajectory of serum albumin levels was inversely associated with mortality increasing the risk of death 2.21 times (p=0.003). The simulation results showed that the model performances increased with sample size. However, the model complexity had increased as more repeated measurements were taken from patients and resulted in lower prediction accuracy unless the sample size was increased. In conclusion, using the trajectory of risk predictors rather than baseline (or averaged) values provided better predictive accuracy and prevented biased results. Finally, the study design (e.g., number of samples and repeated measurements) should be carefully defined since it played an important role in model performances.