Reducing the question burden of patient reported outcome measures using Bayesian networks.

Patient Reported Outcome Measures (PROMs) are questionnaires completed by patients about aspects of their health status. They are a vital part of learning health systems as they are the primary source of information about important outcomes that are best assessed by patients such as pain, disability, anxiety and depression. The volume of questions can easily become burdensome. Previous techniques reduced this burden by dynamically selecting questions from question item banks which are specifically built for different latent constructs being measured. These techniques analyzed the information function between each question in the item bank and the measured construct based on item response theory then used this information function to dynamically select questions by computerized adaptive testing. Here we extend those ideas by using Bayesian Networks (BNs) to enable Computerized Adaptive Testing (CAT) for efficient and accurate question selection on widely-used existing PROMs. BNs offer more comprehensive probabilistic models of the connections between different PROM questions, allowing the use of information theoretic techniques to select the most informative questions. We tested our methods using five clinical PROM datasets, demonstrating that answering a small subset of questions selected with CAT has similar predictions and error to answering all questions in the PROM BN. Our results show that answering 30% - 75% questions selected with CAT had an average area under the receiver operating characteristic curve (AUC) of 0.92 (min: 0.8 - max: 0.98) for predicting the measured constructs. BNs outperformed alternative CAT approaches with a 5% (min: 0.01% - max: 9%) average increase in the accuracy of predicting the responses to unanswered question items.

Early Identification of Trauma-induced Coagulopathy: Development and Validation of a Multivariable Risk Prediction Model.

OBJECTIVE: The aim of this study was to develop and validate a risk prediction tool for trauma-induced coagulopathy (TIC), to support early therapeutic decision-making. BACKGROUND: TIC exacerbates hemorrhage and is associated with higher morbidity and mortality. Early and aggressive treatment of TIC improves outcome. However, injured patients that develop TIC can be difficult to identify, which may compromise effective treatment. METHODS: A Bayesian Network (BN) prediction model was developed using domain knowledge of the causal mechanisms of TIC, and trained using data from 600 patients recruited into the Activation of Coagulation and Inflammation in Trauma (ACIT) study. Performance (discrimination, calibration, and accuracy) was tested using 10-fold cross-validation and externally validated on data from new patients recruited at 3 trauma centers. RESULTS: Rates of TIC in the derivation and validation cohorts were 11.8% and 11.0%, respectively. Patients who developed TIC were significantly more likely to die (54.0% vs 5.5%, P < 0.0001), require a massive blood transfusion (43.5% vs 1.1%, P < 0.0001), or require damage control surgery (55.8% vs 3.4%, P < 0.0001), than those with normal coagulation. In the development dataset, the 14-predictor BN accurately predicted this high-risk patient group: area under the receiver operating characteristic curve (AUROC) 0.93, calibration slope (CS) 0.96, brier score (BS) 0.06, and brier skill score (BSS) 0.40. The model maintained excellent performance in the validation population: AUROC 0.95, CS 1.22, BS 0.05, and BSS 0.46. CONCLUSIONS: A BN (http://www.traumamodels.com) can accurately predict the risk of TIC in an individual patient from standard admission clinical variables. This information may support early, accurate, and efficient activation of hemostatic resuscitation protocols.

Effectiveness of intubation devices in patients with cervical spine immobilisation: a systematic review and network meta-analysis.

BACKGROUND: Cervical spine immobilisation increases the difficulty of tracheal intubation. Many intubation devices have been evaluated in this setting, but their relative performance remains uncertain. METHODS: MEDLINE, EMBASE, and the Cochrane Library were searched to identify randomised trials comparing two or more intubation devices in adults with cervical spine immobilisation. After critical appraisal, a random-effects network meta-analysis was used to pool and compare device performance. The primary outcome was the probability of first-attempt intubation success (first-pass success). For relative performance, the Macintosh direct laryngoscopy blade was chosen as the reference device. RESULTS: We included 80 trials (8039 subjects) comparing 26 devices. Compared with the Macintosh, McGrath™ (odds ratio [OR]=11.5; 95% credible interval [CrI] 3.19-46.20), C-MAC D Blade™ (OR=7.44; 95% CrI, 1.06-52.50), Airtraq™ (OR=5.43; 95% CrI, 2.15-14.2), King Vision™ (OR=4.54; 95% CrI, 1.28-16.30), and C-MAC™ (OR=4.20; 95% CrI=1.28-15.10) had a greater probability of first-pass success. This was also true for the GlideScope™ when a tube guide was used (OR=3.54; 95% CrI, 1.05-12.50). Only the Airway Scope™ had a better probability of first-pass success compared with the Macintosh when manual-in-line stabilisation (MILS) was used as the immobilisation technique (OR=7.98; 95% CrI, 1.06-73.00). CONCLUSIONS: For intubation performed with cervical immobilisation, seven devices had a better probability of first-pass success compared with the Macintosh. However, more studies using MILS (rather than a cervical collar or other alternative) are needed, which more accurately represent clinical practice. CLINICAL TRIAL REGISTRATION: PROSPERO 2019 CRD42019158067 (https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=158067).

Predicting the Outcome of Limb Revascularization in Patients With Lower-extremity Arterial Trauma: Development and External Validation of a Supervised Machine-learning Algorithm to Support Surgical Decisions.

OBJECTIVES: Estimating the likely success of limb revascularization in patients with lower-extremity arterial trauma is central to decisions between attempting limb salvage and amputation. However, the projected outcome is often unclear at the time these decisions need to be made, making them difficult and threatening sound judgement. The objective of this study was to develop and validate a prediction model that can quantify an individual patient's risk of failed revascularization. METHODS: A BN prognostic model was developed using domain knowledge and data from the US joint trauma system. Performance (discrimination, calibration, and accuracy) was tested using ten-fold cross validation and externally validated on data from the UK Joint Theatre Trauma Registry. BN performance was compared to the mangled extremity severity score. RESULTS: Rates of amputation performed because of nonviable limb tissue were 12.2% and 19.6% in the US joint trauma system (n = 508) and UK Joint Theatre Trauma Registry (n = 51) populations respectively. A 10-predictor BN accurately predicted failed revascularization: area under the receiver operating characteristic curve (AUROC) 0.95, calibration slope 1.96, Brier score (BS) 0.05, and Brier skill score 0.50. The model maintained excellent performance in an external validation population: AUROC 0.97, calibration slope 1.72, Brier score 0.08, Brier skill score 0.58, and had significantly better performance than mangled extremity severity score at predicting the need for amputation [AUROC 0.95 (0.92-0.98) vs 0.74 (0.67-0.80); P < 0.0001]. CONCLUSIONS: A BN (https://www.traumamodels.com) can accurately predict the outcome of limb revascularization at the time of initial wound evaluation. This information may complement clinical judgement, support rational and shared treatment decisions, and establish sensible treatment expectations.

Not just data: a method for improving prediction with knowledge.

Many medical conditions are only indirectly observed through symptoms and tests. Developing predictive models for such conditions is challenging since they can be thought of as 'latent' variables. They are not present in the data and often get confused with measurements. As a result, building a model that fits data well is not the same as making a prediction that is useful for decision makers. In this paper, we present a methodology for developing Bayesian network (BN) models that predict and reason with latent variables, using a combination of expert knowledge and available data. The method is illustrated by a case study into the prediction of acute traumatic coagulopathy (ATC), a disorder of blood clotting that significantly increases the risk of death following traumatic injuries. There are several measurements for ATC and previous models have predicted one of these measurements instead of the state of ATC itself. Our case study illustrates the advantages of models that distinguish between an underlying latent condition and its measurements, and of a continuing dialogue between the modeller and the domain experts as the model is developed using knowledge as well as data.

Combining data and meta-analysis to build Bayesian networks for clinical decision support.

Complex clinical decisions require the decision maker to evaluate multiple factors that may interact with each other. Many clinical studies, however, report 'univariate' relations between a single factor and outcome. Such univariate statistics are often insufficient to provide useful support for complex clinical decisions even when they are pooled using meta-analysis. More useful decision support could be provided by evidence-based models that take the interaction between factors into account. In this paper, we propose a method of integrating the univariate results of a meta-analysis with a clinical dataset and expert knowledge to construct multivariate Bayesian network (BN) models. The technique reduces the size of the dataset needed to learn the parameters of a model of a given complexity. Supplementing the data with the meta-analysis results avoids the need to either simplify the model - ignoring some complexities of the problem - or to gather more data. The method is illustrated by a clinical case study into the prediction of the viability of severely injured lower extremities. The case study illustrates the advantages of integrating combined evidence into BN development: the BN developed using our method outperformed four different data-driven structure learning methods, and a well-known scoring model (MESS) in this domain.

Statistical modelling of determinants of child stunting using secondary data and Bayesian networks: a UKRI Global Challenges Research Fund (GCRF) Action Against Stunting Hub protocol paper.

INTRODUCTION: Several factors have been implicated in child stunting, but the precise determinants, mechanisms of action and causal pathways remain poorly understood. The objective of this study is to explore causal relationships between the various determinants of child stunting. METHODS AND ANALYSIS: The study will use data compiled from national health surveys in India, Indonesia and Senegal, and reviews of published evidence on determinants of child stunting. The data will be analysed using a causal Bayesian network (BN)-an approach suitable for modelling interdependent networks of causal relationships. The model's structure will be defined in a directed acyclic graph and illustrate causal relationship between the variables (determinants) and outcome (child stunting). Conditional probability distributions will be generated to show the strength of direct causality between variables and outcome. BN will provide evidence of the causal role of the various determinants of child stunning, identify evidence gaps and support in-depth interrogation of the evidence base. Furthermore, the method will support integration of expert opinion/assumptions, allowing for inclusion of the many factors implicated in child stunting. The development of the BN model and its outputs will represent an ideal opportunity for transdisciplinary research on the determinants of stunting. ETHICS AND DISSEMINATION: Not applicable/no human participants included.

Celecoxib Nanoformulations with Enhanced Solubility, Dissolution Rate, and Oral Bioavailability: Experimental Approaches over In Vitro/In Vivo Evaluation.

Celecoxib (CXB) is a Biopharmaceutical Classification System (BCS) Class II molecule with high permeability that is practically insoluble in water. Because of the poor water solubility, there is a wide range of absorption and limited bioavailability following oral administration. These unfavorable properties can be improved using dry co-milling technology, which is an industrial applicable technology. The purpose of this study was to develop and optimize CXB nanoformulations prepared by dry co-milling technology, with a quality by design approach to maintain enhanced solubility, dissolution rate, and oral bioavailability. The resulting co-milled CXB composition using povidone (PVP), mannitol (MAN) and sodium lauryl sulfate (SLS) showed the maximum solubility and dissolution rate in physiologically relevant media. Potential risk factors were determined with an Ishikawa diagram, important risk factors were selected with Plackett-Burman experimental design, and CXB compositions were optimized with Central Composite design (CCD) and Bayesian optimization (BO). Physical characterization, intrinsic dissolution rate, solubility, and stability experiments were used to evaluate the optimized co-milled CXB compositions. Dissolution and permeability studies were carried out for the resulting CXB nanoformulation. Oral pharmacokinetic studies of the CXB nanoformulation and reference product were performed in rats. The results of in vitro and in vivo studies show that the CXB nanoformulations have enhanced solubility (over 4.8-fold (8.6 ± 1.06 µg/mL vs. 1.8 ± 0.33 µg/mL) in water when compared with celecoxib pure powder), and dissolution rate (at least 85% of celecoxib is dissolved in 20 min), and improved oral pharmacokinetic profile (the relative bioavailability was 145.2%, compared to that of Celebrex®, and faster tmax 3.80 ± 2.28 h vs. 6.00 ± 3.67 h, indicating a more rapid absorption rate).

Assessing Serious Spinal Pathology Using Bayesian Network Decision Support: Development and Validation Study.

BACKGROUND: Identifying and managing serious spinal pathology (SSP) such as cauda equina syndrome or spinal infection in patients presenting with low back pain is challenging. Traditional red flag questioning is increasingly criticized, and previous studies show that many clinicians lack confidence in managing patients presenting with red flags. Improving decision-making and reducing the variability of care for these patients is a key priority for clinicians and researchers. OBJECTIVE: We aimed to improve SSP identification by constructing and validating a decision support tool using a Bayesian network (BN), which is an artificial intelligence technique that combines current evidence and expert knowledge. METHODS: A modified RAND appropriateness procedure was undertaken with 16 experts over 3 rounds, designed to elicit the variables, structure, and conditional probabilities necessary to build a causal BN. The BN predicts the likelihood of a patient with a particular presentation having an SSP. The second part of this study used an established framework to direct a 4-part validation that included comparison of the BN with consensus statements, practice guidelines, and recent research. Clinical cases were entered into the model and the results were compared with clinical judgment from spinal experts who were not involved in the elicitation. Receiver operating characteristic curves were plotted and area under the curve were calculated for accuracy statistics. RESULTS: The RAND appropriateness procedure elicited a model including 38 variables in 3 domains: risk factors (10 variables), signs and symptoms (17 variables), and judgment factors (11 variables). Clear consensus was found in the risk factors and signs and symptoms for SSP conditions. The 4-part BN validation demonstrated good performance overall and identified areas for further development. Comparison with available clinical literature showed good overall agreement but suggested certain improvements required to, for example, 2 of the 11 judgment factors. Case analysis showed that cauda equina syndrome, space-occupying lesion/cancer, and inflammatory condition identification performed well across the validation domains. Fracture identification performed less well, but the reasons for the erroneous results are well understood. A review of the content by independent spinal experts backed up the issues with the fracture node, but the BN was otherwise deemed acceptable. CONCLUSIONS: The RAND appropriateness procedure and validation framework were successfully implemented to develop the BN for SSP. In comparison with other expert-elicited BN studies, this work goes a step further in validating the output before attempting implementation. Using a framework for model validation, the BN showed encouraging validity and has provided avenues for further developing the outputs that demonstrated poor accuracy. This study provides the vital first step of improving our ability to predict outcomes in low back pain by first considering the problem of SSP. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): RR2-10.2196/21804.

A Bayesian Network Decision Support Tool for Low Back Pain Using a RAND Appropriateness Procedure: Proposal and Internal Pilot Study.

BACKGROUND: Low back pain (LBP) is an increasingly burdensome condition for patients and health professionals alike, with consistent demonstration of increasing persistent pain and disability. Previous decision support tools for LBP management have focused on a subset of factors owing to time constraints and ease of use for the clinician. With the explosion of interest in machine learning tools and the commitment from Western governments to introduce this technology, there are opportunities to develop intelligent decision support tools. We will do this for LBP using a Bayesian network, which will entail constructing a clinical reasoning model elicited from experts. OBJECTIVE: This paper proposes a method for conducting a modified RAND appropriateness procedure to elicit the knowledge required to construct a Bayesian network from a group of domain experts in LBP, and reports the lessons learned from the internal pilot of the procedure. METHODS: We propose to recruit expert clinicians with a special interest in LBP from across a range of medical specialties, such as orthopedics, rheumatology, and sports medicine. The procedure will consist of four stages. Stage 1 is an online elicitation of variables to be considered by the model, followed by a face-to-face workshop. Stage 2 is an online elicitation of the structure of the model, followed by a face-to-face workshop. Stage 3 consists of an online phase to elicit probabilities to populate the Bayesian network. Stage 4 is a rudimentary validation of the Bayesian network. RESULTS: Ethical approval has been obtained from the Research Ethics Committee at Queen Mary University of London. An internal pilot of the procedure has been run with clinical colleagues from the research team. This showed that an alternating process of three remote activities and two in-person meetings was required to complete the elicitation without overburdening participants. Lessons learned have included the need for a bespoke online elicitation tool to run between face-to-face meetings and for careful operational definition of descriptive terms, even if widely clinically used. Further, tools are required to remotely deliver training about self-identification of various forms of cognitive bias and explain the underlying principles of a Bayesian network. The use of the internal pilot was recognized as being a methodological necessity. CONCLUSIONS: We have proposed a method to construct Bayesian networks that are representative of expert clinical reasoning for a musculoskeletal condition in this case. We have tested the method with an internal pilot to refine the process prior to deployment, which indicates the process can be successful. The internal pilot has also revealed the software support requirements for the elicitation process to model clinical reasoning for a range of conditions. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/21804.

Analyzing the Simonshaven Case Using Bayesian Networks.

This paper is one in a series of analyses of the Dutch Simonshaven murder case, each using a different modeling approach. We adopted a Bayesian network (BN)-based approach which requires us to determine the relevant hypotheses and evidence in the case and their relationships (captured as a directed acyclic graph) along with explicit prior conditional probabilities. This means that both the graph structure and probabilities had to be defined using subjective judgments about the causal, and other, connections between variables and the strength and nature of the evidence. Determining if a useful BN could be quickly constructed by a small group using the previously established idioms-based approach which provides a generic method for translating legal cases into BNs, was a key aim. The model described was built by the authors during a workshop dedicated to the case at the Isaac Newton Institute, Cambridge, in September 2016. The total effort involved was approximately 26 h (i.e., an average of 6 h per author). With the basic assumptions described in the paper, the posterior probability of guilt once all the evidence is entered is 74%. The paper describes a formal evaluation of the model, using sensitivity analysis, to determine how robust the model conclusions are to key subjective prior probabilities over a full range of what may be deemed "reasonable" from both defense and prosecution perspectives. The results show that the model is reasonably robust-pointing not only generally to a reasonably high posterior probability of guilt but also generally below the 95% threshold expected in criminal law. Given the constraints on building a complex model so quickly, there are inevitably weaknesses; hence, the paper describes these and how they might be addressed, including how to take account of supplementary case information not known at the time of the workshop.

Evidence-based investment selection: Prioritizing agricultural development investments under climatic and socio-political risk using Bayesian networks.

Agricultural development projects have a poor track record of success mainly due to risks and uncertainty involved in implementation. Cost-benefit analysis can help allocate resources more effectively, but scarcity of data and high uncertainty makes it difficult to use standard approaches. Bayesian Networks (BN) offer a suitable modelling technology for this domain as they can combine expert knowledge and data. This paper proposes a systematic methodology for creating a general BN model for evaluating agricultural development projects. Our approach adapts the BN model to specific projects by using systematic review of published evidence and relevant data repositories under the guidance of domain experts. We evaluate a large-scale agricultural investment in Africa to provide a proof of concept for this approach. The BN model provides decision support for project evaluation by predicting the value-measured as net present value and return on investment-of the project under different risk scenarios.

Cyber-physical-based PAT (CPbPAT) framework for Pharma 4.0.

Industry 4.0 aims to integrate manufacturing operations into a seamless digital whole by incorporating flexibility, agility, re-configurability, and sustainability. The result of this integration is a "smart factory" that is more lean, agile, and flexible in operations. There are valid reasons, and perhaps requirements, for pharmaceutical industries to embrace smart factory and to "borrow" the concept of Industry 4.0 to give rise to "Pharma 4.0" (i.e., the pharmaceutical version of Industry 4.0). This paper proposes a cyber-physical-based PAT framework called CPbPAT for implementing smart manufacturing systems in the pharmaceutical industry. The framework has been developed using an agent-based system and is presented by a standard system modeling language called the Unified Modeling Language (UML). The pharmaceutical manufacturing system shown in "Quality by Design for ANDAs" is used as a case study to illustrate the application of the proposed framework.

Towards an Evidence-Based Decision Support Tool for Management of Musculoskeletal Conditions.

Musculoskeletal (MSK) problems present an increasing burden for the healthcare sector, particularly in ageing populations. Advances in evidence are often slow to influence clinical decisions, suggesting decision support would be beneficial. We propose a Bayesian network (BN) for providing evidence-based decision support as it can explicitly represent domain knowledge as causal relations and allows both domain knowledge and clinical data to be combined to create a usable decision model. We make a preliminary evaluation of the model's performance.

Long-term, patient-centered outcomes of lower-extremity vascular trauma.

OBJECTIVE: To describe the long-term outcomes of military lower-extremity vascular injuries, and the decision making of surgeons treating these injuries. BACKGROUND: Lower-extremity vascular trauma is an important cause of preventable death and severe disability, and decisions on amputation or limb salvage can be difficult. Additionally, the complexity of the condition is not amenable to controlled study, and there is limited data to guide clinical decision making and establish sensible treatment expectations during rehabilitation. METHODS: A cohort study of 554 US service members who sustained lower-extremity vascular injury in Iraq or Afghanistan (March 2003 to February 2012) was performed using the military's trauma registry, its electronic health record, patient interviews, and quality-of-life surveys. Long-term surgical and functional outcomes, and the timing and rationale of surgical decisions, were analyzed. RESULTS: Of 579 injured extremities, 49 (8.5%) underwent primary amputation and 530 (91.5%) an initial attempt at salvage. Ninety extremities underwent secondary amputation, occurring in the early (n = 60; <30 days) or late (n = 30; >30 days) phases after injury. For salvage attempts, freedom from amputation 10 years after injury was 82.7% (79.1%-85.7%). Long-term physical and mental health outcomes were similar between service members who underwent reconstruction and those who underwent amputation. CONCLUSION: This military experience provides data that will inform an array of military and civilian providers who care for patients with severe lower-extremity injury. While the majority salvage attempts endure, success is hindered by ischemia and necrosis during the acute stage and pain, dysfunction and infection in the later phases of recovery. LEVEL OF EVIDENCE: Therapeutic/prognostic, level III.

The role of splenic angioembolization as an adjunct to nonoperative management of blunt splenic injuries: A systematic review and meta-analysis.

BACKGROUND: Nonoperative management (NOM) of hemodynamically normal patients with blunt splenic injury (BSI) is the standard of care. Guidelines recommend additional splenic angioembolization (SAE) in patients with American Association for the Surgery of Trauma (AAST) Grade IV and Grade V BSI, but the role of SAE in Grade III injuries is unclear and controversial. The aim of this systematic review was to compare the safety and effectiveness of SAE as an adjunct to NOM versus NOM alone in adults with BSI. METHODS: A systematic literature search (Medline, Embase, and CINAHL) was performed to identify original studies that compared outcomes in adult BSI patients treated with SAE or NOM alone. Primary outcome was failure of NOM. Secondary outcomes included morbidity, mortality, hospital length of stay, and transfusion requirements. Bayesian meta-analyses were used to calculate an absolute (risk difference) and relative (risk ratio [RR]) measure of treatment effect for each outcome. RESULTS: Twenty-three studies (6,684 patients) were included. For Grades I to V combined, there was no difference in NOM failure rate (SAE, 8.6% vs NOM, 7.7%; RR, 1.09 [0.80-1.51]; p = 0.28), mortality (SAE, 4.8% vs NOM, 5.8%; RR, 0.82 [0.45-1.31]; p = 0.81), hospital length of stay (11.3 vs 9.5 days; p = 0.06), or blood transfusion requirements (1.8 vs 1.7 units; p = 0.47) between patients treated with SAE and those treated with NOM alone. However, morbidity was significantly higher in patients treated with SAE (SAE, 38.1% vs NOM, 18.6%; RR, 1.83 [1.20-2.66]; p < 0.01). When stratified by grade of splenic injury, SAE significantly reduced the failure rate of NOM in patients with Grade IV and Grade V splenic injuries but had minimal effect in those with Grade I to Grade III injuries. CONCLUSION: Splenic angioembolization should be strongly considered as an adjunct to NOM in patients with AAST Grade IV and Grade V BSI but should not be routinely recommended in patients with AAST Grade I to Grade III injuries. LEVEL OF EVIDENCE: Systematic review and meta-analysis, level III.

Value of information analysis for interventional and counterfactual Bayesian networks in forensic medical sciences.

OBJECTIVES: Inspired by real-world examples from the forensic medical sciences domain, we seek to determine whether a decision about an interventional action could be subject to amendments on the basis of some incomplete information within the model, and whether it would be worthwhile for the decision maker to seek further information prior to suggesting a decision. METHOD: The method is based on the underlying principle of Value of Information to enhance decision analysis in interventional and counterfactual Bayesian networks. RESULTS: The method is applied to two real-world Bayesian network models (previously developed for decision support in forensic medical sciences) to examine the average gain in terms of both Value of Information (average relative gain ranging from 11.45% and 59.91%) and decision making (potential amendments in decision making ranging from 0% to 86.8%). CONCLUSIONS: We have shown how the method becomes useful for decision makers, not only when decision making is subject to amendments on the basis of some unknown risk factors, but also when it is not. Knowing that a decision outcome is independent of one or more unknown risk factors saves us from the trouble of seeking information about the particular set of risk factors. Further, we have also extended the assessment of this implication to the counterfactual case and demonstrated how answers about interventional actions are expected to change when some unknown factors become known, and how useful this becomes in forensic medical science.

Explicit evidence for prognostic Bayesian network models.

Many prognostic models are not adopted in clinical practice regardless of their reported accuracy. Doubts about the basis of the model is considered to be a major reason for this as the evidence behind clinical models is often not clear to anyone other than their developers. We propose a framework for representing the evidence behind Bayesian networks (BN) developed for prognostic decision support. The aim of this evidence framework is to be able to present all the evidence alongside the BN itself. We illustrate this framework by a BN developed with clinical evidence to predict coagulation disorders in trauma care.