Journal Pre-proof

Bayesian Networks in Healthcare: Distribution by Medical Condition

Scott McLachlan, Kudakwashe Dube, Graham A Hitman, Norman
Fenton, Evangelia Kyrimi

PII: S0933-3657(20)30077-4

DOI: https://doi.org/10.1016/j.artmed.2020.101912

Reference: ARTMED 101912

To appear in: Artificial Intelligence In Medicine

Received Date: 16 January 2020

Revised Date: 27 April 2020

Accepted Date: 9 June 2020

Please cite this article as: McLachlan S, Dube K, A Hitman G, Fenton N, Kyrimi E, Bayesian
Networks in Healthcare: Distribution by Medical Condition, Artificial Intelligence In Medicine
(2020), doi: https://doi.org/10.1016/j.artmed.2020.101912

This is a PDF file of an article that has undergone enhancements after acceptance, such as
the addition of a cover page and metadata, and formatting for readability, but it is not yet the
definitive version of record. This version will undergo additional copyediting, typesetting and
review before it is published in its final form, but we are providing this version to give early
visibility of the article. Please note that, during the production process, errors may be
discovered which could affect the content, and all legal disclaimers that apply to the journal
pertain.

© 2020 Published by Elsevier.

https://doi.org/10.1016/j.artmed.2020.101912
https://doi.org/10.1016/j.artmed.2020.101912


Bayesian Networks in Healthcare: 

Distribution by Medical Condition 
 Scott McLachlan1, 2, Kudakwashe Dube2,3, Graham A Hitman4, Norman Fenton1, Evangelia 

Kyrimi1 

1 Risk and Information Management, Queen Mary University of London, United Kingdom 
2 Health informatics and Knowledge Engineering Research (HiKER) Group  
3 School of Fundamental Sciences, Massey University, New Zealand 
4 Centre for Genomics and Child Health, Blizard Institute, Queen Mary University of London, London, 

United Kingdom 

 

Highlights 
 

 A review which identifies the medical conditions for which 

contemporary literature on Bayesian networks (BNs) have been 

focused. 

 Finds that most attention has gone towards four primary health 

conditions. 

 Identifies differences in the approaches used by authors between each 

of the four primary health conditions. 

 Contributes to our understanding of how and what BNs are being 

considered for in healthcare. 

  

 

 

Jo
ur

na
l P

re
-p

ro
of



Abstract 

 
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 there are any 

differences 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 there is a lack of 

understanding regarding how BNs work and what they are capable of, 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. 

 

 

Keywords: Bayesian networks, Healthcare, Medical conditions 

 

1. Introduction 

Bayesian Networks (BNs) have received increasing attention during the last two decades 

[1, 2] for their particular ability to be applied to challenging issues and aid those making decisions 

to reason about cause and outcome under conditions of uncertainty [3-5]. In 2016, the journal 

Machine Learning ran a special issue on Machine Learning for Healthcare and Medicine [6]. In 

explaining their motivation for that issue, the editors discussed how ever-increasing volumes of 

health data created potential for developing new knowledge that could improve the practice of 

patient care. Machine learning (ML) and artificial intelligence (AI) approaches have been 

proposed for a diverse range of health topics from genomics [7, 8] to treatment selection [9, 10] 

and outcome, prognosis, prediction [11]. A significant benefit of BNs over methods of pure ML 

from data is that BNs do not explicitly require massively large datasets. BNs can incorporate the 

accumulated knowledge of experts in circumstances where data are limited, and still produce 

meaningful and accurate decision-support systems. This paper is part of a larger effort to address 

the wide chasm between research enthusiasm for BNs and their lack of adoption in healthcare. 

The paper seeks to understand the medical conditions BNs are being used to model and make 

predictions for, and any apparent differences in their application between clinical domains, in 

order to shed light on how we may harness the potential of BNs in daily healthcare practice. 

BNs are based on Bayes’ theorem and use a graphical approach for compact 

representation of multivariate probability distributions and efficient reasoning under uncertainty 

[2]. Bayes’ theorem is a formula for understanding how belief in the probability of phenomenon 

under observation evolves as knowledge of related influential phenomena increases [12, 13]. For 

instance, where multiple diseases present with similar symptoms making differential diagnosis 

challenging, clinicians must update their degree of belief about which illness is causing the 

patient’s poor health as new information becomes available from diagnostic tests and physical 

examination of the patient.   

There are three approaches for developing a BN; (1) using only data (data-driven BNs) 

(2) using only knowledge (expert-driven BNs), and (3) using a combination of both data and 

Jo
ur

na
l P

re
-p

ro
of



knowledge (hybrid BNs). Expert-driven and hybrid BNs, which supplement data with knowledge, 

could potentially be the most capable approach for supporting Learning Health Systems (LHS), 

precision medicine, and thus enabling personalised clinical decision-making from large collections 

of aggregated health data. LHS, which include such well-known types as clinical decision support 

systems (CDSS) [14], have experienced slow adoption and are not routinely observed in clinical 

practice [15-17]. This is particularly true for those LHS based on BNs [2]. While others have 

reviewed the scope for AI and ML in healthcare [18-20], in each case BNs have been overlooked.  

Authors who have reviewed medical decision support models tend to provide a brief but 

broad brushstrokes view of either the ML or AI domain before focusing the bulk of their work on 

whichever method was their particular area of research interest. An example is [21] where the 

authors provide non-exhaustive lists and a brief glossary of domain-wide terms before focusing 

solely on learning approaches for neural networks, which are only one of a vast number of AI and 

ML types. In [18] the authors describe a list of ML algorithms but only expound on two from their 

research: support vector machines and neural networks. Attempts had been made to identify [19, 

22] and even quantify [18] the broad scope of medical conditions and clinical questions or 

activities to which ML and AI respectively are being applied. However, it was difficult from each 

author’s method and descriptions to identify: (a) the search terms they had used; (b) a list of the 

medical conditions or clinical questions; or (c) an accurate accounting of the papers identified by 

and used in each study.  

The aim of this paper is to establish the medical conditions to which BNs are being applied 

as part of our wider effort [23] to determine the potential benefits and challenges being faced by 

those seeking to integrate BNs into daily practice. This paper seeks to achieve its aim by 

identifying and quantifying the range of medical conditions for which the healthcare-related BN 

models are being considered. This paper contributes to our overall understanding of the purposes 

for which BNs are being considered as a predictive or decision-making tool in the healthcare 

domain, and the differences in approach between the most popular medical conditions to which 

they have been applied. It is only through continued research to understand the challenges and 

issues that we may in future harness the potential of BNs to affect positive change, and improve 

patient outcomes. 

This paper is organised as follows: Section 3 presents the method used to achieve the 

objectives of this paper as part of our larger scoping review. Section 4 presents the results, 

followed by a discussion of the general findings and their significance in Section 5. The paper 

concludes with a summary in Section 6. 

2. Method 

We used the literature collection from our scoping review on BNs in healthcare [23]. The 

search term used to derive that collection was: 

 
“(((Bayes OR Bayesian) AND network) OR (probabilistic AND graphical AND 

model)) AND (medical OR clinical)” 

 

Terms such as “Bayesian networks” or “graphical probabilistic models” were used here 

because they are widely used in the targeted literature. Different ways for explaining the medical 

condition do occur: in some papers an exact condition is identified, while in others broader terms 

Jo
ur

na
l P

re
-p

ro
of



such as “medical or clinical application”, “medical or clinical condition”, or “medical or clinical 

setting” are used. Our scope review settled on the broader terms “medical” and “clinical” as they 

were found in a wider collection of papers. There are many thousands of possible medical 

conditions that might be targeted by some form of clinical decision support, artificial intelligence 

or machine learning system. It would be impossible to perform separate searches for each 

medical condition and each variation of clinical terminology, and such a search would impose an 

unintended restriction on the papers returned by excluding the opportunity for a natural and 

broadly representative sample. The terms chosen also allowed papers to be returned from clinical 

and non-clinical sources describing solutions that may be in clinical use, or which may have been 

used in a clinical trial. This was important for assessing adoption, which is the topic of a future 

paper in our series on BNs in Healthcare. 

Our initial and unrestricted search returned almost 100,000 papers which we contend was 

an impossible number to review, and therefore required some focus that we established not to 

affect the aim of this paper, nor reduce the efficacy of the search outcome. While using the same 

keywords, we limited the keyword search to identifying instances only in the title and abstract, 

reducing the initial number of papers returned to 3810. Through the process described in Figure 

1 we then excluded papers published before 2012, as the aim of this paper is to review papers 

providing the state-of-the-art, rather than to provide a complete archaeology of BNs in healthcare. 

We also removed those not published in English, and any whose primary content was not 

healthcare-related. In addition, papers that did not provide BNs and instead were focused merely 

on Bayesian statistics or meta-analyses were excluded. Graphical models, such as naive BNs, 

which are the simplest BNs and structurally assume all variables are independent, and other 

graphical computational approaches used in ML or AI, including neural networks, have also been 

excluded. This review focused on published works, which means commercial products not 

associated with published academic or research papers may not have been included. Generally, 

works that fall outside the identified body of existing literature are naturally excluded from a review. 

As a rule, this limitation is inherent in the majority of published academic literature reviews. 

   

 

 

Jo
ur

na
l P

re
-p

ro
of



 
Figure 1: PRISMA diagram (from [23]) 

 

Our initial review plan from [23] identified six primary objectives that would be investigated 

during review of the literature collection. Given the interrelationship between components of 

Objectives 1-3, they are reported together in our main review in [23]. Standalone Objectives 4 

and 6 will be the topic of works yet to be presented. Objective 5, also a standalone topic, is the 

focus of research presented in this paper which reports the distribution and frequency of medical 

conditions that have been the target for BN models in the same literature collection. Objective 5 

is identified in purple in our review concept plan [23] presented here as Figure 2. 

 

Jo
ur

na
l P

re
-p

ro
of



 
Figure 2: Concept map for the Scoping Review of Bayesian Networks in Healthcare (from [23]) 

 

The list of medical conditions in Table 1 was developed with the input of two clinical 

experts and refined inductively during a small-scale preliminary review described in [2]. Each 

paper was reviewed by two reviewers who recorded the target medical condition against the list 

which was presented to reviewers as part of a secure online survey used to manage and conduct 

the overall review. Where the medical condition was not listed, or the reviewer felt unable to make 

an explicit classification, it was possible to manually enter details or references to the condition 

from the literature into a free-text field exposed below the list. Where the two responses for a 

given paper differed, two authors (SM and EK) reviewed the paper collaboratively to achieve 

consensus. In rare circumstances where consensus could not be achieved, a clinician was 

available to assist with classification.  

 

 

 

 

Jo
ur

na
l P

re
-p

ro
of



 Medical Condition Exemplar 
Citations 

1 Blood disorders [61, 63] 

2 Brain or spine (including CNS injury) [64] 

3 Chronic Conditions (e.g.: diabetes and arthritic conditions) [65, 66] 

4 Cancer [30-32] 

5 Cardiac conditions [25-27] 

6 Fatigue [67] 

7 Genetic [68] 

8 Infectious diseases [69] 

9 Liver disease [70, 71] 

10 Lung and breathing disorders [46-48] 

11 Medication (concentration or reaction to) [72] 

12 Musculoskeletal Conditions [73] 

13 Oesophageal disorders (swallowing or speech) [74] 

14 Organ disorders or failure [75] 

15 Pregnancy disorders [76] 

16 Psychological or psychiatric disorders [41-43] 

17 Skin (burns or disorders) [77, 78] 

18 Sleep disorders [79] 

19 Surgery related infection [80] 

20 Utility of or experience of healthcare [81] 

21 Unclassified or Other [52, 82] 

 Table 1: list of Medical Conditions  

3. Results and Discussion 

Even though the quantity of BN literature is rapidly increasing, it was possible to identify a 

distinct focus on a small group of conditions. Specifically, as shown in Figure 3, cardiac conditions, 

cancer, psychological and psychiatric disorders, and lung and breathing disorders made up 59% 

of the medical conditions in the literature. This result was not surprising given that other research 

identified similar disease foci for AI generally [18], and the notoriety of these conditions in the 

mainstream media as those most likely to kill you [24]. The remaining 41% of papers were spread 

across a diverse collection of seemingly random topics. Figure 3 also illustrates the general 

distribution of BN literature against the entire list of medical conditions. The complete review 

Jo
ur

na
l P

re
-p

ro
of



dataset is available as a supplemental material spreadsheet on the PamBayesian website at 

(http://www.pambayesian.org/interim-data/).  

 

 
Figure 3: Distribution of medical conditions in BN literature 

3.1 Most Common Medical Conditions  

While in the dataset we have made available as supplementary material to the paper we identify 

every paper and distinct sub-classification for each, in this section we briefly examine only the 

most common or frequent topics discussed for each of the most common medical conditions. 

 

Cardiac conditions  

A strong theme for cardiac models was a restriction of focus to either acute diagnosis or prediction 

of disease progression. Common to many was the authors’ use of electronic patient data as the 

primary source for elements of BN structure and observations necessary for prediction. This, while 

simultaneously admonishing the poor quality they observed of most available electronic health 

record (EHR) data and the limitations that arose from this quality issue [25-27]. Proposed 

approaches ranged from models for classifying the diagnosis [28] or severity [29] of a patient’s 

condition, to prediction of risk for acute clinical events [25] and the likely progression of the 

individual’s disease [26]. It was not unusual for cardiac BNs to incorporate known comorbidities 

and validate their interaction with the patient’s cardiac condition [26, 27]. 

 

Cancer  

Jo
ur

na
l P

re
-p

ro
of



Breast cancer [30-32] and (onco)gene expression [33] separately, and sometimes together [34, 

35], were the targets receiving the greatest attention for the topic of cancer. BNs were applied to 

supporting and enhancing expert knowledge in diagnosing breast cancer from mammography 

images [36], classifying tumours [30], and as a potential method to improve expert judgement in 

situations where increasingly complex treatment options have made clinical decisions difficult 

[37]. Cancer BNs were more likely to seek previously unknown relationships either between 

symptoms and syndromes [38], or expressed genes and target metastatic cancers [34]. 

 

Psychological and psychiatric disorders  

The most prevalent conditions modelled in the literature were depression [39-41] and the age-

related cognitive degeneration diagnoses: dementia and Alzheimer's Disease (AD) [41-43]. 

Incorporating an organisational structure from ontologies and merging this with BNs was a popular 

approach in these medical models [39, 44], along with a stronger focus on identifying or relying 

on a hierarchy of symptoms [39, 44]. While medical models were generally seen to be more likely 

to draw on experts’ knowledge than data alone, the nature and mystery of psychological and 

psychiatric disorders may be the reason they more frequently drew on clinical expertise during 

BN development [39-41, 43, 45].  

 

Lung and breathing disorders  

Half of all models for lung and breathing disorders sought to predict the risk of exacerbation of an 

already diagnosed chronic clinical condition, including: chronic obstructive pulmonary disease 

(COPD) [46, 47] and asthma [48]. Others sought to use clinical signs and symptoms to either 

assess the probability of a particular acute pulmonary diagnosis [49], predict future severity [50], 

or classify the subtype of the diagnosed disease [51]. One approach proposed a mobile 

smartphone app that used questionnaires and a BN model to ascertain the patient’s current 

condition and provide contextually relevant advice, while delivering data on the patient’s health 

status to clinicians via an internet connection and a central server [46]. This is similar to the 

approach proposed by the authors of this paper for their PamBayesian project 

(www.pambayesian.org).  

3.2 Researcher and Content Classification 

Overall, papers could generally be classified as one of three types. First, there were 

papers written entirely by computer scientists and mathematicians that whilst being dense on 

technical detail and description for the math of BNs, were sparse on clinical detail for the 

conditions being modelled - these we describe as method-driven [28, 39, 52, 53]. Second, there 

were those that were written entirely by clinicians who included comprehensive discussion and 

contextualisation of the medical condition and its symptomatology, but significantly less technical 

discussion of the theory and development of their BN - these we felt were problem-driven [54, 

55]. Finally, there were those presented by a mix of both clinicians and computing or decision 

scientists - these we describe as hybrid-driven.  

In method-driven works, the focus is mainly on the BN methodology. A medical application 

is used simply as a case study to evaluate the proposed methodology.  

For problem-driven works, it was rare to find a paper that whilst appearing to be written 

entirely by clinicians, was methodological in both its regard for the BN’s development process and 

Jo
ur

na
l P

re
-p

ro
of



comprehensive in its technical description of the BN generally. One such unicorn paper drew the 

attention and comments of both reviewers [56]. The reviewers also generally felt that radiologists 

came across better than most other clinical disciplines at describing the mathematical and 

technical aspects of BN development, possibly resulting from their medical physics training [57, 

58].  

In hybrid-driven works, authors tended to introduce both the medical condition (the 

problem aspects) and BN theory (the methodology aspects) early [48, 59-62]. Having a 

multidisciplinary team offered the advantage that both aspects generally received equal space 

and descriptive detail in the paper. From the medical perspective, the research problem and 

benefit of the BN to clinical practice are better explained. From the computing point of view, the 

overall BN development process is described in greater detail. In papers where both clinicians 

and computer scientists are involved the BN development and validation process were found to 

possess greater accuracy, as clinical input was considered. 

3.3 Strengths and Limitations 

A strength of this review is that this is a component of the largest scoping review of BNs 

in healthcare [23]. Another strength is that tools, including a structured list of medical conditions 

and an online standardised survey form were developed to ensure a consistent review process. 

A limitation of this review is that, even if this is a representative sample of papers published 

in both medical and AI journals and conference proceedings, it may not reflect the entire range of 

the literature regarding BNs in healthcare. We looked for keywords such as “Bayesian Networks”, 

“probabilistic graphical models”, “medical”, “clinical” to appear in the abstract of each paper. While 

it would not constitute a significant number, it is possible that some relevant papers were not 

included because they did not use the selected keywords in their abstract. This is especially true 

in cases where the actual name of the medical condition is described without mentioning the 

words “medical” or “clinical”. However, we believe that the large number of selected papers was 

sufficient for drawing conclusions. 

Because of the very large number of published papers and the time needed for review 

and data extraction, we limited the study to the seven-year period 2012-2018. However, we 

acknowledge this has left out early implementations of BNs that included for instance HUGIN [83] 

and Pathfinder [84].  We also note that the most recent patents for Pathfinder [85, 86] make no 

mention of the use of Bayesian networks in the current Pathfinder software, which may suggest 

the authors have the BN approach in their most recent releases of the software tool. We believe 

this is a safe number of years to be able to draw conclusions for the purpose of this paper. As the 

review began in early 2019, we chose not to include papers published in those first few months 

of 2019. We do not believe this exclusion had any significant impact on our findings. 

It is possible that evidence of commercially available applications using BNs may not 

necessarily be in journal articles, appearing in other places, such as company reports, media 

(grey literature) and general commentaries. For example, based on limited marketing materials 

there are indications that Babylon have used BNs in their non-public research and application 

development [87], including what is claimed as potentially the largest BN in the world [88]. As a 

result, BNs like these may not be represented in this review. This is a limitation inherent in all 

literature reviews.  

Jo
ur

na
l P

re
-p

ro
of



4. Summary and Conclusions 

This paper began by recognising that while other authors had reviewed, and in some 

cases identified and quantified the medical conditions for which other AI and ML methods had 

been developed, in all cases they had eschewed BNs. This was our primary research problem. 

We also highlighted three secondary issues present in those reviews. That it was often difficult to 

identify: (a) the search terms they had used; (b) a framework or description for medical conditions; 

or (c) an accurate accounting of the papers identified by and used in each study. 

As a core component of our systematic review of BNs in healthcare, our reviewers worked 

with clinicians to develop a list for use in identifying the target medical conditions that had been 

the focus of BN modelling. To address the primary research problem, we quantified the BNs from 

our literature collection by medical condition, identifying four conditions that received the majority 

of attention from authors and reviewing papers from each of those conditions in greater detail. 

Also, to ensure we did not proliferate the three secondary issues observed of other works, we 

provided: (a) the search term and process used to arrive at our literature collection; (b) the list of 

medical conditions that reviewers were asked to use in classifying medical BNs; and (c) a URL 

link to the complete review dataset.  

BNs have and continue to receive significant interest for their ability to combine available 

evidence and accurately reason under conditions of uncertainty. This paper has contributed to 

our understanding of how and what BNs have been considered for in healthcare. We identified 

the four primary conditions that have received almost two-thirds of BN-modelling attention in the 

literature: cardiac, cancer, psychological and lung disorders. We also found a number of acute 

differences in how BNs were being applied to each.  

The wider observation is that despite the strong research interest in BN models in 

healthcare, this interest is not matched by adoption in practice. It is possible that with most BN in 

healthcare research effort going to these four primary conditions that already receive vast 

amounts of funding and attention from other solution areas, that BNs are being drowned out: 

effectively lost in the noise. It is also possible that a lack of understanding pervades with respect 

to how BNs work and what they are capable of. Either way, it is only through understanding and 

promotion, continued multidisciplinary research and adherence to the full disclosure required of 

the scientific method that we may in future harness the potential of BNs in daily healthcare practice 

and affect positive change, improving outcomes for clinicians and patients alike.  

 

Conflict of Interest 

AUTHOR DECLARATION TEMPLATE 

We wish to confirm that there are no known conflicts of interest associated with this 

publication and there has been no significant financial support for this work that could have 

influenced its outcome. 

We confirm that the manuscript has been read and approved by all named authors and 

that there are no other persons who satisfied the criteria for authorship but are not listed. We 

further confirm that the order of authors listed in the manuscript has been approved by all of us. 

We confirm that we have given due consideration to the protection of intellectual property 

associated with this work and that there are no impediments to publication, including the timing 

Jo
ur

na
l P

re
-p

ro
of



of publication, with respect to intellectual property. In so doing we confirm that we have followed 

the regulations of our institutions concerning intellectual property. 

Acknowledgement 

 

 
SM, EK, GAH and NF acknowledge support from the EPSRC under project EP/P009964/1: PAMBAYESIAN: Patient 
Managed decision-support using Bayes Networks. KD acknowledges financial support from Massey University for his 
study sabbatical and visits with the PamBayesian team. 
 

 

References 

[1] Pena, J. M., Björkegren, J., & Tegnér, J. (2005). Growing Bayesian network models of gene networks from seed 
genes. Bioinformatics, 21(suppl_2), ii224-ii229. 

[2] Kyrimi, E., Mclachlan, S., Dube, K., & Fenton, N. (2019). Bayesian Networks in Healthcare: The Chasm between 
Research Enthusiasm and Clinical Adoption. Manuscript submitted to BMJ Health and Care Informatics. 

[3] Kahn, C. E., Laur, J. J., & Carrera, G. F. (2001). A Bayesian network for diagnosis of primary bone tumors. Journal 
of digital imaging, 14(1), 56-57. 

[4] Constantinou, A. C., & Fenton, N. (2018). Things to know about Bayesian networks: Decisions under uncertainty, 
part 2. Significance, 15(2), 19-23. 

[5] Fenton, N., & Neil, M. (2011). The use of Bayes and causal modelling in decision making, uncertainty and risk. 
CEPIS Upgrade, 12(5), 10-21. 

[6] Weins, J., Wallace, B. (Eds) (2016) Editorial: Special Issue on Machine Learning for Health and Medicine, Machine 
Learning, 102(3), pp 305-307. 

[7] Libbrecht, M. W., & Noble, W. S. (2015). Machine learning applications in genetics and genomics. Nature Reviews 
Genetics, 16(6), 321. 

[8] Huang, S., Cai, N., Pacheco, P. P., Narrandes, S., Wang, Y., & Xu, W. (2018). Applications of support vector 
machine (SVM) learning in cancer genomics. Cancer Genomics-Proteomics, 15(1), 41-51. 

[9] Vitanovski, D., Ralovich, K., Ionasec, R., Zheng, Y., Suehling, M., Krawtschuk, W., ... & Comaniciu, D. (2012, May). 
Personalized learning-based segmentation of thoracic aorta and main branches for diagnosis and treatment 
planning. In 2012 9th IEEE International Symposium on Biomedical Imaging (ISBI) (pp. 836-839). IEEE. 

[10] Alizadehsani, R., Abdar, M., Jalali, S. M. J., Roshanzamir, M., Khosravi, A., & Nahavandi, S. (2018, January). 
Comparing the Performance of Feature Selection Algorithms for Wart Treatment Selection. In Proceedings of 
The International Workshop on Future Technology (pp. 6-18). 

[11] Parmar, C., Grossmann, P., Rietveld, D., Rietbergen, M. M., Lambin, P., & Aerts, H. J. (2015). Radiomic machine-
learning classifiers for prognostic biomarkers of head and neck cancer. Frontiers in oncology, 5, 272. 

[12] Spiegelhalter, D. J., Myles, J. P., Jones, D. R., & Abrams, K. R. (1999). An introduction to Bayesian methods in 
health technology assessment. Bmj, 319(7208), 508-512. 

[13] Neil, M., Fenton, N., & Nielson, L. (2000). Building large-scale Bayesian networks. The Knowledge Engineering 
Review, 15(3), 257-284. 

[14] McLachlan, S., Potts, H.W.W., Dube, K., Buchanan, D., Lean, S., Gallagher, T., Johnson, O., Daley, B., Marsh, 
W., & Fenton, N. (2018) The Heimdall Framework for supporting characterisation of Learning Health Systems. 
Journal of Innovation in Health Informatics, 25(2)  

[15] Kaplan, B. (2001). Evaluating informatics applications—some alternative approaches: theory, social interactionism, 
and call for methodological pluralism. International journal of medical informatics, 64(1), 39-56. 

[16] Kilsdonk, E., Peute, L. W., Riezebos, R. J., Kremer, L. C., & Jaspers, M. W. (2016). Uncovering healthcare 
practitioners’ information processing using the think-aloud method: From paper-based guideline to clinical 
decision support system. International journal of medical informatics, 86, 10-19. 

[17] McLachlan, S., Potts, H. W., Dube, K., Buchanan, D., Lean, S., Gallagher, T., ... & Fenton, N. (2018). The Heimdall 
Framework for Supporting Characterisation of Learning Health Systems. J Innov Health Inform. 

[18] Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., ... & Wang, Y. (2017). Artificial intelligence in healthcare: past, 
present and future. Stroke and vascular neurology, 2(4), 230-243. 

[19] Rajkomar, A., Dean, J., & Kohane, I. (2019). Machine learning in medicine. New England Journal of Medicine, 
380(14), 1347-1358. 

[20] Miotto, R., Wang, F., Wang, S., Jiang, X., & Dudley, J. T. (2017). Deep learning for healthcare: review, opportunities 
and challenges. Briefings in bioinformatics, 19(6), 1236-1246. 

Jo
ur

na
l P

re
-p

ro
of



[21] Yu, K. H., Beam, A. L., & Kohane, I. S. (2018). Artificial intelligence in healthcare. Nature biomedical engineering, 
2(10), 719-731. 

[22] Ngiam, K. Y., & Khor, W. (2019). Big data and machine learning algorithms for health-care delivery. The Lancet 
Oncology, 20(5), e262-e273. 

[23] Kyrimi, E., Mclachlan, S., Dube, K., & Fenton, N. (2019). Bayesian Networks in Healthcare: Past, Present and 
Future. Manuscript in preparation for submission to ACM Surveys 

[24] Belger, T. (2017). What is most likely to kill you: Study reveals what kills men and women of different ages. The 
Mirror, July 14 2017. Sourced from: <https://www.mirror.co.uk/news/uk-news/what-most-likely-kill-you-
10795316>  

[25] Bandyopadhyay, S., Wolfson, J., Vock, D. M., Vazquez-Benitez, G., Adomavicius, G., Elidrisi, M., ... & O’Connor, 
P. J. (2015). Data mining for censored time-to-event data: a Bayesian network model for predicting 
cardiovascular risk from electronic health record data. Data Mining and Knowledge Discovery, 29(4), 1033-
1069. 

[26] Exarchos, K. P., Carpegianni, C., Rigas, G., Exarchos, T. P., Vozzi, F., Sakellarios, A., ... & Fotiadis, D. I. (2014). 
A multiscale approach for modeling atherosclerosis progression. IEEE journal of biomedical and health 
informatics, 19(2), 709-719. 

[27] Bukhanov, N., Balakhontceva, M., Kovalchuk, S., Zvartau, N., & Konradi, A. (2017). Multiscale modeling of 
comorbidity relations in hypertensive outpatients. Procedia computer science, 121, 446-450. 

[28] Alobaidi, A. T. S., & Mahmood, N. T. (2013, December). Modified full bayesian networks classifiers for medical 
diagnosis. In 2013 International Conference on Advanced Computer Science Applications and Technologies 
(pp. 5-12). IEEE. 

[29] Gladence, L. M., Ravi, T., & Karthi, M. (2014, May). An enhanced method for detecting congestive heart failure-
Automatic Classifier. In 2014 IEEE International Conference on Advanced Communications, Control and 
Computing Technologies (pp. 586-590). IEEE. 

[30] Khademi, M., & Nedialkov, N. S. (2015, December). Probabilistic graphical models and deep belief networks for 
prognosis of breast cancer. In 2015 IEEE 14th International Conference on Machine Learning and Applications 
(ICMLA) (pp. 727-732). IEEE. 

[31] Aljawad, D. A., Alqahtani, E., Ghaidaa, A. K., Qamhan, N., Alghamdi, N., Alrashed, S., ... & Olatunji, S. O. (2017, 
February). Breast cancer surgery survivability prediction using bayesian network and support vector 
machines. In 2017 International Conference on Informatics, Health & Technology (ICIHT) (pp. 1-6). IEEE. 

[32] Refai, A., Merouani, H. F., & Aouras, H. (2016). Maintenance of a Bayesian network: application using medical 
diagnosis. Evolving Systems, 7(3), 187-196. 

[33] Luo, Y., McShan, D. L., Matuszak, M. M., Ray, D., Lawrence, T. S., Jolly, S., ... & El Naqa, I. (2018). A multiobjective 
Bayesian networks approach for joint prediction of tumor local control and radiation pneumonitis in nonsmall‐
cell lung cancer (NSCLC) for response‐ adapted radiotherapy. Medical physics, 45(8), 3980-3995. 

[34] Lee, S., & Jiang, X. (2017). Modeling miRNA-mRNA interactions that cause phenotypic abnormality in breast 
cancer patients. PloS one, 12(8), e0182666. 

[35] Gámez-Pozo, A., Trilla-Fuertes, L., Berges-Soria, J., Selevsek, N., López-Vacas, R., Díaz-Almirón, M., ... & 
Moreno, F. G. (2017). Functional proteomics outlines the complexity of breast cancer molecular subtypes. 
Scientific reports, 7(1), 10100. 

[36] Velikova, Marina, et al. "On the interplay of machine learning and background knowledge in image interpretation 
by Bayesian networks." Artificial intelligence in medicine 57.1 (2013): 73-86. 

[37] Cypko, M. A., Stoehr, M., Kozniewski, M., Druzdzel, M. J., Dietz, A., Berliner, L., & Lemke, H. U. (2017). Validation 
workflow for a clinical Bayesian network model in multidisciplinary decision making in head and neck oncology 
treatment. International journal of computer assisted radiology and surgery, 12(11), 1959-1970. 

[38] Ji, Z., & Wang, B. (2014). Identifying potential clinical syndromes of hepatocellular carcinoma using PSO-based 
hierarchical feature selection algorithm. BioMed research international, 2014. 

[39] Chang, Y. S., Fan, C. T., Lo, W. T., Hung, W. C., & Yuan, S. M. (2015). Mobile cloud-based depression diagnosis 
using an ontology and a Bayesian network. Future Generation Computer Systems, 43, 87-98. 

[40] McNally, R. J., Mair, P., Mugno, B. L., & Riemann, B. C. (2017). Co-morbid obsessive–compulsive disorder and 
depression: A Bayesian network approach. Psychological medicine, 47(7), 1204-1214. 

[41] Gross, T. J., Araújo, R. B., Vale, F. A. C., Bessani, M., & Maciel, C. D. (2018). Dependence between cognitive 
impairment and metabolic syndrome applied to a Brazilian elderly dataset. Artificial intelligence in medicine, 
90, 53-60. 

[42] Jin, Y., Su, Y., Zhou, X. H., Huang, S., & Alzheimer’s Disease Neuroimaging Initiative. (2016). Heterogeneous 
multimodal biomarkers analysis for Alzheimer’s disease via Bayesian network. EURASIP Journal on 
Bioinformatics and Systems Biology, 2016(1), 12. 

[43] Moreira, L. B., & Namen, A. A. (2018). A hybrid data mining model for diagnosis of patients with clinical suspicion 
of dementia. Computer methods and programs in biomedicine, 165, 139-149. 

[44] Hu, H., & Kerschberg, L. (2018, August). Evolving Medical Ontologies Based on Causal Inference. In 2018 
IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) (pp. 
954-957). IEEE. 

Jo
ur

na
l P

re
-p

ro
of



[45] Constantinou, A. C., Freestone, M., Marsh, W., & Coid, J. (2015). Causal inference for violence risk management 
and decision support in forensic psychiatry. Decision Support Systems, 80, 42-55. 

[46] van der Heijden, M., Velikova, M., & Lucas, P. J. (2014). Learning Bayesian networks for clinical time series 
analysis. Journal of biomedical informatics, 48, 94-105. 

[47] Mcheick, H., Saleh, L., Ajami, H., & Mili, H. (2017). Context Relevant Prediction Model for COPD Domain Using 
Bayesian Belief Network. Sensors, 17(7), 1486. 

[48] Spyroglou, I. I., Spöck, G., Rigas, A. G., & Paraskakis, E. N. (2018). Evaluation of Bayesian classifiers in asthma 
exacerbation prediction after medication discontinuation. BMC research notes, 11(1), 522. 

[49] Magrini, A., Luciani, D., & Stefanini, F. M. (2018). A probabilistic network for the diagnosis of acute cardiopulmonary 
diseases. Biometrical Journal, 60(1), 174-195. 

[50] Zarikas, V., Papageorgiou, E., & Regner, P. (2015). Bayesian network construction using a fuzzy rule based 
approach for medical decision support. Expert Systems, 32(3), 344-369. 

[51] Ong, M. S., Mullen, M. P., Austin, E. D., Szolovits, P., Natter, M. D., Geva, A., ... & Mandl, K. D. (2017). Learning 
a comorbidity-driven taxonomy of pediatric pulmonary hypertension. Circulation research, 121(4), 341-353. 

[52] Agarwal, P., Verma, R., & Mallik, A. (2016, August). Ontology based disease diagnosis system with probabilistic 
inference. In 2016 1st India International Conference on Information Processing (IICIP) (pp. 1-5). IEEE. 

[53] Aljawad, D. A., Alqahtani, E., Ghaidaa, A. K., Qamhan, N., Alghamdi, N., Alrashed, S., ... & Olatunji, S. O. (2017, 
February). Breast cancer surgery survivability prediction using Bayesian network and support vector 
machines. In 2017 International Conference on Informatics, Health & Technology (ICIHT) (pp. 1-6). IEEE. 

[54] Ducher, M., Mounier-Véhier, C., Lantelme, P., Vaisse, B., Baguet, J. P., & Fauvel, J. P. (2015). Reliability of a 
Bayesian network to predict an elevated aldosterone-to-renin ratio. Archives of cardiovascular diseases, 
108(5), 293-299. 

[55] Merli, M., Moscatelli, M., Mariotti, G., Pagliaro, U., Bernardelli, F., & Nieri, M. (2016). A minimally invasive technique 
for lateral maxillary sinus floor elevation: a Bayesian network study. Clinical oral implants research, 27(3), 
273-281. 

[56] Jochems, A., Deist, T. M., Van Soest, J., Eble, M., Bulens, P., Coucke, P., ... & Dekker, A. (2016). Distributed 
learning: developing a predictive model based on data from multiple hospitals without data leaving the 
hospital–a real life proof of concept. Radiotherapy and Oncology, 121(3), 459-467. 

[57] Luo, Y., McShan, D. L., Matuszak, M. M., Ray, D., Lawrence, T. S., Jolly, S., ... & El Naqa, I. (2018). A multiobjective 
Bayesian networks approach for joint prediction of tumor local control and radiation pneumonitis in non-small‐
cell lung cancer (NSCLC) for response‐ adapted radiotherapy. Medical physics, 45(8), 3980-3995. 

[58] Luo, Y., El Naqa, I., McShan, D. L., Ray, D., Lohse, I., Matuszak, M. M., ... & Ten Haken, R. K. (2017). Unraveling 
biophysical interactions of radiation pneumonitis in non-small-cell lung cancer via Bayesian network analysis. 
Radiotherapy and Oncology, 123(1), 85-92. 

[59] Sesen, M. B., Nicholson, A. E., Banares-Alcantara, R., Kadir, T., & Brady, M. (2013). Bayesian networks for clinical 
decision support in lung cancer care. PloS one, 8(12), e82349. 

[60] Wang, Z., Jenkins, M. W., Linderman, G. C., Bezerra, H. G., Fujino, Y., Costa, M. A., ... & Rollins, A. M. (2015). 3-
D stent detection in intravascular OCT using a Bayesian network and graph search. IEEE transactions on 
medical imaging, 34(7), 1549-1561. 

[61] Yet, B., Bastani, K., Raharjo, H., Lifvergren, S., Marsh, W., & Bergman, B. (2013). Decision support system for 
Warfarin therapy management using Bayesian networks. Decision Support Systems, 55(2), 488-498. 

[62] Constantinou, A. C., Freestone, M., Marsh, W., Fenton, N., & Coid, J. (2015). Risk assessment and risk 
management of violent reoffending among prisoners. Expert Systems with Applications, 42(21), 7511-7529. 

[63] Yet, B., Perkins, Z., Fenton, N., Tai, N., & Marsh, W. (2014). Not just data: A method for improving prediction with 
knowledge. Journal of biomedical informatics, 48, 28-37. 

[64] HeidarAbadi, N. N., Hakemi, L., Kolivand, P., Safdari, R., & Saeidi, M. G. (2017). Comparing performances of 
intelligent classifier algorithms for predicting type of pain in patients with spinal cord injury. Electronic 
physician, 9(7), 4847. 

[65] Orphanou, K., Stassopoulou, A., & Keravnou, E. (2014). Temporal abstraction and temporal Bayesian networks in 
clinical domains: A survey. Artificial intelligence in medicine, 60(3), 133-149. 

[66] Yousefi, L., Tucker, A., Al-luhaybi, M., Saachi, L., Bellazzi, R., & Chiovato, L. (2018, June). Predicting disease 
complications using a stepwise hidden variable approach for learning dynamic bayesian networks. In 2018 
IEEE 31st International Symposium on Computer-Based Medical Systems (CBMS) (pp. 106-111). IEEE. 

[67] Akhtar, M. J., & Utne, I. B. (2014). Human fatigue’s effect on the risk of maritime groundings–A Bayesian Network 
modeling approach. Safety science, 62, 427-440. 

[68] Zarringhalam, K., Enayetallah, A., Reddy, P., & Ziemek, D. (2014). Robust clinical outcome prediction based on 
Bayesian analysis of transcriptional profiles and prior causal networks. Bioinformatics, 30(12), i69-i77. 

[69] Sa-ngamuang, C., Haddawy, P., Luvira, V., Piyaphanee, W., Iamsirithaworn, S., & Lawpoolsri, S. (2018). Accuracy 
of dengue clinical diagnosis with and without NS1 antigen rapid test: Comparison between human and 
Bayesian network model decision. PLoS neglected tropical diseases, 12(6), e0006573. 

Jo
ur

na
l P

re
-p

ro
of



[70] Jiao, Y., Wang, X. H., Chen, R., Tang, T. Y., Zhu, X. Q., & Teng, G. J. (2017). Predictive models of minimal hepatic 
encephalopathy for cirrhotic patients based on large-scale brain intrinsic connectivity networks. Scientific 
reports, 7(1), 11512. 

[71] Zhang, Y., Zhang, T., Zhang, C., Tang, F., Zhong, N., Li, H., ... & Xue, F. (2015). Identification of reciprocal causality 
between non-alcoholic fatty liver disease and metabolic syndrome by a simplified Bayesian network in a 
Chinese population. BMJ open, 5(9), e008204. 

[72] Duan, L., Khoshneshin, M., Street, W. N., & Liu, M. (2012). Adverse drug effect detection. IEEE journal of 
biomedical and health informatics, 17(2), 305-311. 

[73] Farmer, N. (2014). An update and further testing of a knowledge‐ based diagnostic clinical decision support system 
for musculoskeletal disorders of the shoulder for use in a primary care setting. Journal of evaluation in clinical 
practice, 20(5), 589-595. 

[74] Solomon, N. P., Dietsch, A. M., Dietrich-Burns, K. E., Styrmisdottir, E. L., & Armao, C. S. (2016). Dysphagia 
management and research in an acute-care military treatment facility: The role of applied informatics. Military 
medicine, 181(suppl_5), 138-144. 

[75] Ducher, M., Kalbacher, E., Combarnous, F., Finaz de Vilaine, J., McGregor, B., Fouque, D., & Fauvel, J. P. (2013). 
Comparison of a Bayesian network with a logistic regression model to forecast IgA nephropathy. BioMed 
research international, 2013. 

[76] Moreira, M. W., Rodrigues, J. J., Oliveira, A. M., Ramos, R. F., & Saleem, K. (2016, May). A preeclampsia diagnosis 
approach using Bayesian networks. In 2016 IEEE International Conference on Communications (ICC) (pp. 1-

5). IEEE. 
[77] Berchialla, P., Gangemi, E. N., Foltran, F., Haxhiaj, A., Buja, A., Lazzarato, F., ... & Gregori, D. (2014). Predicting 

severity of pathological scarring due to burn injuries: a clinical decision making tool using Bayesian networks. 
International wound journal, 11(3), 246-252. 

[78] Kaewprag, P., Newton, C., Vermillion, B., Hyun, S., Huang, K., & Machiraju, R. (2015). Predictive modeling for 
pressure ulcers from intensive care unit electronic health records. AMIA Summits on Translational Science 
Proceedings, 2015, 82. 

[79] Yetton, B. D., McDevitt, E. A., Cellini, N., Shelton, C., & Mednick, S. C. (2018). Quantifying sleep architecture 
dynamics and individual differences using big data and Bayesian networks. PloS one, 13(4), e0194604. 

[80] Sohn, S., Larson, D. W., Habermann, E. B., Naessens, J. M., Alabbad, J. Y., & Liu, H. (2017). Detection of clinically 
important colorectal surgical site infection using Bayesian network. Journal of Surgical Research, 209, 168-
173. 

[81] Chao, Y. S., Wu, H. T., Scutari, M., Chen, T. S., Wu, C. J., Durand, M., & Boivin, A. (2017). A network perspective 
on patient experiences and health status: the Medical Expenditure Panel Survey 2004 to 2011. BMC health 
services research, 17(1), 579. 

[82] Dawson, P., Gailis, R., & Meehan, A. (2015). Detecting disease outbreaks using a combined Bayesian network 
and particle filter approach. Journal of theoretical biology, 370, 171-183. 

[83] Andersen, S. K., Olesen, K. G., Jensen, F. V., & Jensen, F. (1989, August). HUGIN-A Shell for Building Bayesian 

Belief Universes for Expert Systems. In IJCAI (Vol. 89, pp. 1080-1085). 

[84] Heckerman, D. E., Horvitz, E. J., & Nathwani, B. N. (1992). Toward normative expert systems: Part 1 the pathfinder 

project. Methods of information in medicine, 31(02), 90-105. 

[85] Justia Patents (2008). Method for detection of cardiac arrhythmias. Last accessed: 24th April, 2020. Sourced from: 

https://patents.justia.com/patent/8233973 

[86] Justia Patents (2012). Systems for detecting cardiac arrhythmias. Last accessed: 24th April, 2020. Sourced from: 

https://patents.justia.com/patent/8774908 

[87] Armstrong S. The apps attempting to transfer NHS 111 online. BMJ. 2018;360(January 2018):1-5. 

doi:10.1136/bmj.k156. 

[88] MMC Ventures. Beyond The Hype: Artificial Intelligence Podcast Interview - Babylon Health.; 2017. 

https://www.healthcare.digital/single-post/2018/03/29/Healing-Healthcare-with-Artificial-Intelligence-“Beyond-

The-Hype”-with-Dr-Ali-Parsa-Founder-and-CEO-of-Babylon-Health#! 

 

 

 

 
Jo

ur
na

l P
re

-p
ro

of