Predictive Accuracy of Fixed and Random Survival Models: A case of TB-LAM Clinical Trial
Loading...
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The University of Dodoma
Abstract
The current study compared the prediction accuracy and covariate selection
performance of the fitted fixed and random models in covariates without validating the
proportional hazard assumption. Given the complexities of the covariate structures and
the higher-dimensional interactions between the explanatory covariates, predicting
clinical trial results and treatment responses remains challenging As a result, effective
survival prediction models that can manage and understand complex structures and
variable interactions are definitely required. The Cox proportional hazard (Cox PH)
model is the most widely used to predict survival time, despite its limitations due to
complicated interactions and non-linearity in covariates. Random survival model,
particularly random survival forest (RSF), has gained popularity in recent years due to
its capacity to understand complicated interactions among explanatory covariates and
nonlinearity effects among predictors. Accordingly, the tuberculosis and human
immunodeficiency virus (TB-HIV) co-infection dataset was obtained from Kibong'oto
Infectious Diseases Hospital in Tanzania using a retrospective cohort design. The
concordance index(C-index), Brier score (BS), and Integrated Brier score (IBS) were
used to calculate prediction accuracy. Furthermore, binary logistic regression were
used to examine variables related to the outcome status, which signified survival or
death, and was compared to predictors related to time-to-event using the Cox
regression model. Consequently, the RSF model outperformed the Cox PH model in
both prediction accuracy and covariate selection performance in C-index, BS, and IBS
outcomes. Similarly, the covariates chosen by the binary logistic regression model
based on the outcome status differed slightly from the variables picked by the Cox PH
based on the time to event. The RSF model outperformed Cox PH in prediction
accuracy in the TB-HIV co-infection dataset without verifying the PH assumption of
the covariates, suggesting that the RSF model should be used to analyze survival
datasets regardless of whether the variables' proportional hazard assumption is
satisfied.
Description
Master's Dissertation (Mathematical Sciences)
Keywords
Predictive accuracy, Fixed survival models, TB-LAM, Random survival model, Covariate structure, Covariate selection, RSF model, Random Survival Forest
Citation
Oscar, L. (2023). Predictive accuracy of fixed and random survival models: A aase of TB-LAM clinical trial (Master's dissertation). The University of Dodoma. Dodoma.