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.
In the absence of an established gold standard, an understanding of the testing cycle from individual exposure to test outcome report is required to guide the correct interpretation of severe acute respiratory syndrome-coronavirus-2 reverse transcriptase real-time polymerase chain reaction (RT-PCR) results and optimise the testing processes. Bayesian network models have been used within healthcare to bring clarity to complex problems. We use this modelling approach to construct a comprehensive framework for understanding the real-world predictive value of individual RT-PCR results.We elicited knowledge from domain experts to describe the test process through a facilitated group workshop. A preliminary model was derived based on the elicited knowledge, then subsequently refined, parameterised and validated with a second workshop and one-on-one discussions.Causal relationships elicited describe the interactions of pre-testing, specimen collection and laboratory procedures and RT-PCR platform factors, and their impact on the presence and quantity of virus and thus the test result and its interpretation. By setting the input variables as ‘evidence’ for a given subject and preliminary parameterisation, four scenarios were simulated to demonstrate potential uses of the model.The core value of this model is a deep understanding of the total testing cycle, bridging the gap between a person's true infection status and their test outcome. This model can be adapted to different settings, testing modalities and pathogens, adding much needed nuance to the interpretations of results.
Reproducibility is a key measure of the veracity of a modelling result or finding. In other research areas, notably in medicine, reproducibility is supported by mandating the inclusion of an agreed set of details into every research publication, facilitating systematic reviews, transparency and reproducibility. Governments and international organisations are increasingly turning to modelling approaches in the development and decision-making for policy and have begun asking questions about accountability in model-based decision making. The ethical issues of relying on modelling that is biased, poorly constructed, constrained by heroic assumptions and not reproducible are multiplied when such models are used to underpin decisions impacting human and planetary well-being. Bayesian Network modelling is used in policy development and decision support across a wide range of domains. In light of the recent trend for governments and other organisations to demand accountability and transparency, we have compiled and tested a reporting checklist for Bayesian Network modelling which will bring the desirable level of transparency and reproducibility to enable models to support decision making and allow the robust comparison and combination of models. The use of this checklist would support the ethical use of Bayesian network modelling for impactful decision making and research.
Despite the potential of evidence-based medical innovations to improve patient outcomes, their integration remains difficult. Implementation science aims to assist by identifying and deploying effective implementation strategies within complex health care settings. Determinant frameworks, such as the Consolidated Framework for Implementation Research (CFIR), help identify factors influencing implementation success but do not specify mechanisms or methods for selecting optimal strategies. Selection methods are largely empirical, highlighting the need for objective, quantifiable approaches. We developed causal Bayesian networks (BNs) to model the interdependencies amongst contextual factors, determinants and outcomes with a specific example: the detection and management of chonic wet cough in Indigenous Australian children in primary health care settings. The BNs, informed by CFIR domains and prior qualitative research, quantifies the impact of barriers and enablers on implementation outcomes. The BNs enable predictions of intervention effects, and the assessment and quantification of potential implementation strategies, or a combination of strategies. The BNs are linked to a simple survey that allows implementation strategies to be tailored for each setting and that was administered at several sites across Australia to validate the models. The overall process, including the BNs and surveys, constitutes a generalisable structured workflow for selecting the most promising strategies. We describe the model development and validation, and the broader applicability of our BN-based workflow in implementation science.
Bayesian Networks (BNs) are an important tool for assisting probabilistic reasoning, but despite being considered transparent models, people have trouble understanding them. Further, current User Interfaces (UIs) still do not clarify the reasoning of BNs. To address this problem, we have designed verbal and visual extensions to the standard BN UI, which can guide users through common inference patterns. We conducted a user study to compare our verbal, visual and combined UI extensions, and a baseline UI. Our main findings are: (1) users did better with all three types of extensions than with the baseline UI for questions about the impact of an observation, the paths that enable this impact, and the way in which an observation influences the impact of other observations; and (2) using verbal and visual modalities together is better than using either modality alone for some of these question types.
Motivated by the ambiguity of operational case definitions for long COVID and the impact of the lack of a common causal language on long COVID research, in early 2023 we began developing a research framework on this post-acute infection syndrome. We used directed acyclic graphs (DAGs) and Bayesian networks (BNs) to depict the hypothesised mechanisms of long COVID in an agnostic fashion. The DAGs were informed by the evolving literature and subsequently refined following elicitation workshops with domain experts. The workshops were structured online sessions guided by an experienced facilitator. The causal DAGs aim to summarise the hypothesised pathobiological pathways from mild or severe COVID-19 disease to the development of pulmonary symptoms and fatigue over four different time points. The DAG was converted into a BN using qualitative parametrisation. These causal models aim to assist the identification of disease endotypes, as well as the design of randomised controlled trials and observational studies. The framework can also be extended to a range of other post-acute infection syndromes.