Related Experiment Video
Updated: Jun 9, 2026
Establishing a Competing Risk Regression Nomogram Model for Survival Data
Published on: October 23, 2020
Bayesian networks as prognostic models in oncology: a systematic review and recommendations for clinical practice
Casper Reijnen1, Johanna M A Pijnenborg2, Peter Hoskin3,4,5
1Department of Radiation Oncology, Radboud University Medical Center, Nijmegen, The Netherlands.
Objectives:
In the last decades, artificial intelligence has made tremendous steps in supporting decision-making in healthcare. Bayesian networks (BNs) have emerged as powerful tools for probabilistic modelling and offer important advantages compared with regression-based models. This systematic review identifies the available literature on BNs for prognostication in oncology, evaluates their performance and discusses how BNs can overcome limitations of traditional prediction models.
Design Setting Participants:
The systematic review was performed according to the Cochrane guidance and reported according to the guidelines for the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA). The protocol was registered in PROSPERO. Studies identified in MEDLINE and EMBASE were included.
Main Outcome Measures:
Eligible studies investigated BNs as prognostication or predictive tools within oncology (solid or haematological malignancies).
Results:
A total of 52 studies were included, with a median construction cohort size of 438 patients. More than half of studies harboured unclear or high concerns about bias. Most studies used score-based structure learning approaches to construct the BNs. External validation was applied in a minority of studies and showed excellent performance metrics when hybrid techniques (integration of machine-learning and expert knowledge) were applied. Studies comparing BN validation with Cox regression validation showed better performance for BNs in most studies.
Conclusions:
When constructed with hybrid techniques, BNs offer great potential to support decision-making in clinical oncology. Advantages include their ability to estimate treatment effects with non-randomised data, to model causal relationships, to apply counterfactual reasoning and to work with missing variables. To be implemented into clinical practice, hybrid construction techniques, high-quality external validation, prospective evaluation with a health technology assessment and adequate postmarketing surveillance and maintenance are essential.
Prospero Registration Number:
CRD420251140130.
