Bayesian networks in healthcare: What is preventing their adoption?

Using staged tree models for health data: Investigating invasive fungal infections by aspergillus and other filamentous fungi

Machine learning models are increasingly used in the medical domain to study the association between risk factors and diseases to support practitioners in understanding health outcomes. In this paper, we showcase the use of machine-learned staged tree models for investigating complex asymmetric dependence structures in health data. Staged trees are a specific class of generative, probabilistic graphical models that formally model asymmetric conditional independence and non-regular sample spaces. An investigation of the risk factors in invasive fungal infections demonstrates the insights staged trees provide to support medical decision-making.

Unraveling atherosclerotic cardiovascular disease risk factors through conditional probability analysis with Bayesian networks: insights from the AZAR cohort study

This study aimed at modelling the underlying predictor of ASCVD through the Bayesian network (BN). Data for the AZAR Cohort Study, which evaluated 500 healthcare providers in Iran, was collected through examinations, and blood samples. Two BNs were used to explore a suitable causal model for analysing the underlying predictor of ASCVD; Bayesian search through an algorithmic approach and knowledge-based BNs. Results showed significant differences in ASCVD risk factors across background variables' levels. The diagnostic indices showed better performance for the knowledge-based BN (Area under ROC curve (AUC) = 0.78, Accuracy = 76.6, Sensitivity = 62.5, Negative predictive value (NPV) = 96.0, Negative Likelihood Ratio (LR-) = 0.48) compared to Bayesian search (AUC = 0.76, Accuracy = 72.4, Sensitivity = 17.5, NPV = 93.2, LR- = 0.83). In addition, we decided on knowledge-based BN because of the interpretability of the relationships. Based on this BN, being male (conditional probability = 63.7), age over 45 (36.3), overweight (51.5), Mets (23.8), diabetes (8.3), smoking (10.6), hypertension (12.1), high T-C (28.5), high LDL-C (23.9), FBS (12.1), and TG (25.9) levels were associated with higher ASCVD risk. Low and normal HDL-C levels also had higher ASCVD risk (35.3 and 37.4), while high HDL-C levels had lower risk (27.3). In conclusion, BN demonstrated that ASCVD was significantly associated with certain risk factors including being older and overweight male, having a history of Mets, diabetes, hypertension, having high levels of T-C, LDL-C, FBS, and TG, but Low and normal HDL-C and being a smoker. The study may provide valuable insights for developing effective prevention strategies for ASCVD in Iran.

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.

Outcome prediction of electroconvulsive therapy for depression

INTRODUCTION

We developed and tested a Bayesian network(BN) model to predict ECT remission for depression, with non-response as a secondary outcome.

METHODS

We performed a systematic literature search on clinically available predictors. We combined these predictors with variables from a dataset of clinical ECT trajectories (performed in the University Medical Center Utrecht) to create priors and train the BN. Temporal validation was performed in an independent sample.

RESULTS

The systematic literature search yielded three meta-analyses, which provided prior knowledge on outcome predictors. The clinical dataset consisted of 248 treatment trajectories in the training set and 44 trajectories in the test set at the same medical center. The AUC for the primary outcome remission estimated on an independent validation set was 0.686 (95%CI 0.513-0.859) (AUC values of 0.505 - 0.763 observed in 5-fold cross validation of the model within the train set). Accuracy 0.73 (balanced accuracy 0.67), sensitivity 0.55, specificity 0.79, after temporal validation in the independent sample. Prior literature information marginally reduced CI width.

DISCUSSION

A BN model comprised of prior knowledge and clinical data can predict remission of depression after ECT with reasonable performance. This approach can be used to make outcome predictions in psychiatry, and offers a methodological framework to weigh additional information, such as patient characteristics, symptoms and biomarkers. In time, it may be used to improve shared decision-making in clinical practice.

Predicting the causative pathogen among children with pneumonia using a causal Bayesian network

BACKGROUND

Pneumonia remains a leading cause of hospitalization and death among young children worldwide, and the diagnostic challenge of differentiating bacterial from non-bacterial pneumonia is the main driver of antibiotic use for treating pneumonia in children. Causal Bayesian networks (BNs) serve as powerful tools for this problem as they provide clear maps of probabilistic relationships between variables and produce results in an explainable way by incorporating both domain expert knowledge and numerical data.

METHODS

We used domain expert knowledge and data in combination and iteratively, to construct, parameterise and validate a causal BN to predict causative pathogens for childhood pneumonia. Expert knowledge elicitation occurred through a series of group workshops, surveys and one-on-one meetings involving 6-8 experts from diverse domain areas. The model performance was evaluated based on both quantitative metrics and qualitative expert validation. Sensitivity analyses were conducted to investigate how the target output is influenced by varying key assumptions of a particularly high degree of uncertainty around data or domain expert knowledge.

RESULTS

Designed to apply to a cohort of children with X-ray confirmed pneumonia who presented to a tertiary paediatric hospital in Australia, the resulting BN offers explainable and quantitative predictions on a range of variables of interest, including the diagnosis of bacterial pneumonia, detection of respiratory pathogens in the nasopharynx, and the clinical phenotype of a pneumonia episode. Satisfactory numeric performance has been achieved including an area under the receiver operating characteristic curve of 0.8 in predicting clinically-confirmed bacterial pneumonia with sensitivity 88% and specificity 66% given certain input scenarios (i.e., information that is available and entered into the model) and trade-off preferences (i.e., relative weightings of the consequences of false positive versus false negative predictions). We specifically highlight that a desirable model output threshold for practical use is very dependent upon different input scenarios and trade-off preferences. Three commonly encountered scenarios were presented to demonstrate the potential usefulness of the BN outputs in various clinical pictures.

CONCLUSIONS

To our knowledge, this is the first causal model developed to help determine the causative pathogen for paediatric pneumonia. We have shown how the method works and how it would help decision making on the use of antibiotics, providing insight into how computational model predictions may be translated to actionable decisions in practice. We discussed key next steps including external validation, adaptation and implementation. Our model framework and the methodological approach can be adapted beyond our context to broad respiratory infections and geographical and healthcare settings.

The feasibility of a Bayesian network model to assess the probability of simultaneous symptoms in patients with advanced cancer

Although patients with advanced cancer often experience multiple symptoms simultaneously, clinicians usually focus on symptoms that are volunteered by patients during regular history-taking. We aimed to evaluate the feasibility of a Bayesian network (BN) model to predict the presence of simultaneous symptoms, based on the presence of other symptoms. Our goal is to help clinicians prioritize which symptoms to assess. Patient-reported severity of 11 symptoms (scale 0-10) was measured using an adapted Edmonton Symptom Assessment Scale (ESAS) in a national cross-sectional survey among advanced cancer patients. Scores were dichotomized (< 4 and ≥ 4). Using fourfold cross validation, the prediction error of 9 BN algorithms was estimated (Akaike information criterion (AIC). The model with the highest AIC was evaluated. Model predictive performance was assessed per symptom; an area under curve (AUC) of ≥ 0.65 was considered satisfactory. Model calibration compared predicted and observed probabilities; > 10% difference was considered inaccurate. Symptom scores of 532 patients were collected. A symptom score ≥ 4 was most prevalent for fatigue (64.7%). AUCs varied between 0.60 and 0.78, with satisfactory AUCs for 8/11 symptoms. Calibration was accurate for 101/110 predicted conditional probabilities. Whether a patient experienced fatigue was directly associated with experiencing 7 other symptoms. For example, in the absence or presence of fatigue, the model predicted a 8.6% and 33.1% probability of experiencing anxiety, respectively. It is feasible to use BN development for prioritizing symptom assessment. Fatigue seems most eligble to serve as a starting symptom for predicting the probability of experiencing simultaneous symptoms.

Urinary tract infections in children: building a causal model-based decision support tool for diagnosis with domain knowledge and prospective data

BACKGROUND

Diagnosing urinary tract infections (UTIs) in children in the emergency department (ED) is challenging due to the variable clinical presentations and difficulties in obtaining a urine sample free from contamination. Clinicians need to weigh a range of observations to make timely diagnostic and management decisions, a difficult task to achieve without support due to the complex interactions among relevant factors. Directed acyclic graphs (DAG) and causal Bayesian networks (BN) offer a way to explicitly outline the underlying disease, contamination and diagnostic processes, and to further make quantitative inference on the event of interest thus serving as a tool for decision support.

METHODS

We prospectively collected data on children present to ED with suspected UTIs. Through knowledge elicitation workshops and one-on-one meetings, a DAG was co-developed with clinical domain experts (the Expert DAG) to describe the causal relationships among variables relevant to paediatric UTIs. The Expert DAG was combined with prospective data and further domain knowledge to inform the development of an application-oriented BN (the Applied BN), designed to support the diagnosis of UTI. We assessed the performance of the Applied BN using quantitative and qualitative methods.

RESULTS

We summarised patient background, clinical and laboratory characteristics of 431 episodes of suspected UTIs enrolled from May 2019 to November 2020. The Expert DAG was presented with a narrative description, elucidating how infection, specimen contamination and management pathways causally interact to form the complex picture of paediatric UTIs. Parameterised using prospective data and expert-elicited parameters, the Applied BN achieved an excellent and stable performance in predicting Escherichia coli culture results, with a mean area under the receiver operating characteristic curve of 0.86 and a mean log loss of 0.48 based on 10-fold cross-validation. The BN predictions were reviewed via a validation workshop, and we illustrate how they can be presented for decision support using three hypothetical clinical scenarios.

CONCLUSION

Causal BNs created from both expert knowledge and data can integrate case-specific information to provide individual decision support during the diagnosis of paediatric UTIs in ED. The model aids the interpretation of culture results and the diagnosis of UTIs, promising the prospect of improved patient care and judicious use of antibiotics.

mHealth apps for gestational diabetes mellitus that provide clinical decision support or artificial intelligence: A scoping review

AIMS

Gestational diabetes (GDM) is the most common metabolic disorder of pregnancy, requiring complex management and empowerment of those affected. Mobile health (mHealth) applications (apps) are proposed for streamlining healthcare service delivery, extending care relationships into the community, and empowering those affected by prolonged medical disorders to be equal collaborators in their healthcare. This review investigates mHealth apps intended for use with GDM; specifically those powered by artificial intelligence (AI) or providing decision support.

METHODS

A scoping review using the novel Survey Tool approach for collaborative literature Reviews (STaR) process was performed.

RESULTS

From 18 papers, 11 discrete GDM-based mHealth apps were identified, but only 3 were reasonably mature with only one currently in use in a clinical setting. Two-thirds of the apps provided condition-relevant contextual user feedback that could aid in patient self care. However, although each app targeted one or more components of the GDM clinical pathway, no app addressed the entirety from diagnosis to postpartum.

CONCLUSIONS

There are limited mHealth apps for GDM that incorporate AI or AI-based decision support. Many exist only to record patient information like blood glucose readings or diet, provide generic patient education or advice, or to reduce adverse events by providing medication or appointment alerts. Significant barriers remain that continue to limit the adoption of mHealth apps in clinical care settings. Further research and development are needed to deliver intelligent holistic mHealth apps using AI that can truly reduce healthcare resource use and improve outcomes by enabling patient self care in the community.

A comprehensive scoping review of Bayesian networks in healthcare: Past, present and future

No comprehensive review of Bayesian networks (BNs) in healthcare has been published in the past, making it difficult to organize the research contributions in the present and identify challenges and neglected areas that need to be addressed in the future. This unique and novel scoping review of BNs in healthcare provides an analytical framework for comprehensively characterizing the domain and its current state. A literature search of health and health informatics literature databases using relevant keywords found 3810 articles that were reduced to 123. This was after screening out those presenting Bayesian statistics, meta-analysis or neural networks, as opposed to BNs and those describing the predictive performance of multiple machine learning algorithms, of which BNs were simply one type. Using the novel analytical framework, we show that: (1) BNs in healthcare are not used to their full potential; (2) a generic BN development process is lacking; (3) limitations exist in the way BNs in healthcare are presented in the literature, which impacts understanding, consensus towards systematic methodologies, practice and adoption; and (4) a gap exists between having an accurate BN and a useful BN that impacts clinical practice. This review highlights several neglected issues, such as restricted aims of BNs, ad hoc BN development methods, and the lack of BN adoption in practice and reveals to researchers and clinicians the need to address these problems. To map the way forward, the paper proposes future research directions and makes recommendations regarding BN development methods and adoption in practice.