Source-linked AI summary
Bayesian Networks in Healthcare: Distribution by Medical Condition
Scott McLachlan, Kudakwashe Dube, Graham A Hitman, Norman E Fenton, Evangelia Kyrimi
TL;DR
Existing reviews had not examined which medical conditions healthcare BNs model or whether their applications differ across conditions. This paper uses a scoping-review collection and structured classification process to identify and quantify those conditions and compare application patterns. It finds that cardiac, cancer, psychological, and lung disorders receive almost two-thirds of BN-modelling attention, while BN research interest remains unmatched by practice adoption.
Problem
Earlier reviews had not investigated the medical conditions modelled by healthcare BNs or differences in how and why BNs are applied across conditions.
Method
The study analyzes a scoping-review literature collection using an expert-informed condition list, dual review, and consensus resolution.
Results
Almost two-thirds of BN-modelling attention focuses on cardiac, cancer, psychological, and lung disorders.
Takeaways & Limitations
The paper identifies concentrated BN research attention across four conditions and acute differences in how BNs are applied to them.
Takeaways & Limitations
The review may not capture the entire healthcare-BN literature because relevant papers could lack the selected abstract keywords.
Abstract
from arXiv · showhide
Bayesian networks (BNs) have received increasing research attention that is not matched by adoption in practice and yet have potential to significantly benefit healthcare. Hitherto, research works have not investigated the types of medical conditions being modelled with BNs, nor whether any differences exist in how and why they are applied to different conditions. This research seeks to identify and quantify the range of medical conditions for which healthcare-related BN models have been proposed, and the differences in approach between the most common medical conditions to which they have been applied. We found that almost two-thirds of all healthcare BNs are focused on four conditions: cardiac, cancer, psychological and lung disorders. We believe that a lack of understanding regarding how BNs work and what they are capable of exists, and that it is only with greater understanding and promotion that we may ever realise the full potential of BNs to effect positive change in daily healthcare practice.
1. Introduction
This paper addresses the gap between growing research interest in Bayesian networks (BNs) and their limited adoption in healthcare by examining which medical conditions they model and how applications differ. It frames BNs as tools for representing probability distributions and reasoning under uncertainty, then defines the study’s aim of clarifying their healthcare use.
- 1. Introduction: The paper responds to a gap between increasing BN research enthusiasm and limited adoption in healthcare.The broader effort seeks to understand how BN potential might be harnessed in daily healthcare practice.
- 1. Introduction: BNs represent multivariate probability distributions compactly and support efficient reasoning under uncertainty.They are based on Bayes’ theorem, which updates belief as knowledge about related phenomena increases.
- 1. Introduction: BN development can be data-driven, expert-driven, or hybrid, combining data with knowledge.The paper notes that expert-driven and hybrid approaches could support Learning Health Systems, precision medicine, and personalised clinical decision-making.
- 1. Introduction: The study identifies and quantifies the medical conditions targeted by healthcare-related BN models.It also examines differences in application between the most common medical conditions.
- 1. Introduction: The paper contributes to a wider scoping review and is organised around method, results, discussion, and conclusion sections.Its specific focus is the distribution and application of BNs across medical conditions.
2. Method
The method draws on a wider scoping-review literature collection and uses broad healthcare-related search terms before applying publication, language, content, and model-type exclusions. Reviewers then classify target conditions using an expert-informed list, with disagreements resolved through author consensus.
- 2. Method: The study reused a literature collection from a scoping review of BNs in healthcare.Its search combined Bayesian or Bayes terms with network or probabilistic graphical model terms and medical or clinical terms.
- 2. Method: The search used broad medical and clinical terms because papers described conditions inconsistently.Searching separately for every specific medical condition was considered impractical.
- 2. Method: 3810 papers were initially identified before exclusions for publication date, language, healthcare relevance, and model type.The review excluded papers published before 2013, non-English papers, non-healthcare studies, Bayesian statistics or meta-analyses without BNs, naive BNs, and other graphical computational approaches.
- 2. Method: The paper examines Objective 5: the distribution and frequency of medical conditions targeted by BN models.This objective was one of six primary objectives in the larger review plan.
- 2. Method: Two clinical experts informed the medical-condition list, which was refined inductively during preliminary review.Two reviewers classified each paper, and disagreements were resolved collaboratively by two authors; free-text entries were allowed when classification was unclear.
3. Results and Discussion
BN healthcare literature concentrates on four condition groups, while models differ in their clinical focus and researcher orientation. The review also identifies a sampling limitation that may exclude relevant literature.
- Most common medical conditions: 59% of medical-condition BN literature concerned cardiac, cancer, psychological or psychiatric, and lung or breathing disorders.The remaining 41% covered a diverse collection of topics.
- Most common medical conditions: Cardiac BNs mainly addressed acute diagnosis or disease progression, often using electronic patient data despite reported EHR quality limitations.Models included diagnosis or severity classification, acute-event risk, disease progression, and comorbidity interactions.
- Most common medical conditions: Cancer BNs focused on breast cancer and gene expression, supporting diagnosis, tumour classification, treatment judgment, and discovery of previously unknown relationships.These models were applied to mammography, tumour classification, complex treatment decisions, symptoms and syndromes, and gene–metastasis relationships.
- Most common medical conditions: Psychological and psychiatric models most often addressed depression, dementia, and Alzheimer’s disease, frequently combining ontologies, symptom hierarchies, and clinical expertise.The review reports stronger reliance on clinical expertise in these models than on data alone.
- Most common medical conditions: Half of lung and breathing disorder models predicted exacerbation risk for diagnosed chronic conditions such as COPD and asthma.Other models assessed acute diagnoses, future severity, disease subtype, or patient status through smartphone-supported questionnaires.
- Researcher and Content Classification: Healthcare BN papers were classified as method-driven, problem-driven, or hybrid-driven according to whether technical methodology, clinical context, or both received primary emphasis.Method-driven works used medical applications mainly as case studies for evaluating BN methodology.
- Strengths and Limitations: The review may not represent the entire healthcare BN literature because keyword-based abstract screening could omit papers using only specific condition names.The authors nevertheless judged the large number of selected papers sufficient for drawing conclusions.
4. Summary
The paper addresses the lack of BN-focused medical-condition reviews through a systematic review that classifies and quantifies healthcare BN models. It identifies four dominant conditions, finds differences in application, and highlights the continuing gap between research interest and clinical adoption.
- Research problem: Other reviews had examined medical conditions addressed by AI and ML methods but had generally excluded Bayesian networks.The paper also identifies difficulties in determining prior reviews’ search terms, condition frameworks, and complete paper counts.
- Method: The review quantified healthcare BN models by medical condition using a clinician-informed classification list.Reviewers worked with clinicians to develop the list used to identify target conditions.
- Method: The study provided its search process, medical-condition classification list, and a link to the complete review dataset.
- Findings: Almost two-thirds of BN-modelling attention focused on cardiac, cancer, psychological, and lung disorders.The paper also found acute differences in how BNs were applied across these conditions.
- Implications: Strong healthcare BN research interest has not translated into comparable adoption in practice.The authors suggest that concentration on already well-funded conditions and limited understanding of BN capabilities may contribute to this gap.