The Paper Feed
A feed of Bayesian network related papers, articles, books and research that we happen across and find of interest
Visual Analysis of Bayesian Networks for Electronic Health Records
Worldwide the amount of data generated by the medical community is staggering, and increasing dramatically. Using this data to improve patient care using analytics and machine learning is a huge and largely untapped opportunity. The most important medical data captured exist in patients' electronic health records (EHRs) which are maintained and utilized by health care providers. EHRs consist of rich and comprehensive patient-specific information from a large number of sources in different formats with heterogeneous data types. There are numerous challenges in attempting to apply existing analytic tools and methodologies to this data. Many features extracted from EHRs have dependent relationships - for example, “flu” and “high body temperature”. Bayesian networks, as one of the few modeling methodologies which capture feature dependence rather than assuming independence, provide a flexible foundation for modeling EHRs. However, existing Bayesian network learning methodologies produce models whose complexity makes them difficult for clinicians to utilize or even interpret. Therefore, better model visualization methodologies, as well as learning methods which produce models more amenable to simplification and summarization, are critical to making them interpretable and useful to clinicians, and therefore to improving patient care. In this dissertation, I present a framework for predictive analysis of patient clinical data, from feature extraction to model analysis. I first study straightforward machine learning approaches on extracted EHR features and find that incorporating diagnosis features improves area under ROC curve (AUC) by 10% compared to a baseline. Because of the many dependencies between features extracted from EHRs, I next investigate Bayesian network models, in which my clinician collaborators have identified known and suspected high pressure ulcer risk factors. The models also substantially increase sensitivity of the prediction - nearly three times higher comparing to logistical regression models - without sacrificing overall accuracy. However, interpreting these models involves a significant cognitive burden, motivating my investigation of visual analytic techniques. To this end, I develop an interactive tool for visualizing Bayesian networks to improve clinicians’ insight and interpretation of models. I perform a user study to assess the impact of the tool and its features. The results show quantitatively that users complete tasks more efficiently when using the tool, and qualitatively that they found it useful. Bayesian networks containing natural groupings or “clusters” are better suited to visualization and summarization. Since existing Bayesian network learning methods do not naturally yield such groupings, I alter the Bayesian network learning process to learn structures which optimize not just for representing dependency relationships, but additionally and simultaneously, for clusterability measures. My results show that the augmented Bayesian network process can find structures with much larger clusterability measures, with only a small decrease in their standard scoring measure. Visualizations of learned clustered Bayesian networks show that the algorithm cohesively groups related features, making the networks easier to interpret.
Sensitivity Analysis in a Bayesian Network for Modeling an Agent
Agent-based social simulation (ABSS) has become a popular method for simulating and visualizing a phenomenon while making it possible to decipher the system’s dynamism. When a large amount of data is used for an agent’s behavior, such as a questionnaire survey, a Bayesian network is often the preferred method for modeling an agent. Based on the data, a Bayesian network is used in ABSS. However, it is very difficult to learn the accurate structure of a Bayesian network from the raw data because there exist many variables and the search space is too wide. This study proposes a new method for obtaining an appropriate structure for a Bayesian network by using sensitivity analysis in a stepwise fashion. This method enables us to find a feature subset, which is good to explain objective variables without reducing the accuracy. A simple Bayesian network structure that maintains accuracy while indicating an agent’s behavior provides ABSS users with an intuitive understanding of the behavioral principle of an agent. To illustrate the effectiveness of the proposed method, data from a questionnaire survey about healthcare electronics was used.
Partial Least Squares Discriminant Analysis and Bayesian Networks for Metabolomic Prediction of Childhood Asthma
To explore novel methods for the analysis of metabolomics data, we compared the ability of Partial Least Squares Discriminant Analysis (PLS-DA) and Bayesian networks (BN) to build predictive plasma metabolite models of age three asthma status in 411 three year olds (n = 59 cases and 352 controls) from the Vitamin D Antenatal Asthma Reduction Trial (VDAART) study. The standard PLS-DA approach had impressive accuracy for the prediction of age three asthma with an Area Under the Curve Convex Hull (AUCCH) of 81%. However, a permutation test indicated the possibility of overfitting. In contrast, a predictive Bayesian network including 42 metabolites had a significantly higher AUCCH of 92.1% (p for difference < 0.001), with no evidence that this accuracy was due to overfitting. Both models provided biologically informative insights into asthma; in particular, a role for dysregulated arginine metabolism and several exogenous metabolites that deserve further investigation as potential causative agents. As the BN model outperformed the PLS-DA model in both accuracy and decreased risk of overfitting, it may therefore represent a viable alternative to typical analytical approaches for the investigation of metabolomics data.
Modeling interrelationships between health behaviors in overweight breast cancer survivors: Applying Bayesian networks
Obesity and its impact on health is a multifaceted phenomenon encompassing many factors, including demographics, environment, lifestyle, and psychosocial functioning. A systems science approach, investigating these many influences, is needed to capture the complexity and multidimensionality of obesity prevention to improve health. Leveraging baseline data from a unique clinical cohort comprising 333 postmenopausal overweight or obese breast cancer survivors participating in a weight-loss trial, we applied Bayesian networks, a machine learning approach, to infer interrelationships between lifestyle factors (e.g., sleep, physical activity), body mass index (BMI), and health outcomes (biomarkers and self-reported quality of life metrics). We used bootstrap resampling to assess network stability and accuracy, and Bayesian information criteria (BIC) to compare networks. Our results identified important behavioral subnetworks. BMI was the primary pathway linking behavioral factors to glucose regulation and inflammatory markers; the BMI-biomarker link was reproduced in 100% of resampled networks. Sleep quality was a hub impacting mental quality of life and physical health with > 95% resampling reproducibility. Omission of the BMI or sleep links significantly degraded the fit of the networks. Our findings suggest potential mechanistic pathways and useful intervention targets for future trials. Using our models, we can make quantitative predictions about health impacts that would result from targeted, weight loss and/or sleep improvement interventions. Importantly, this work highlights the utility of Bayesian networks in health behaviors research.
Reducing COPD Readmissions: A Causal Bayesian Network Model - IEEE Journals
This paper introduces a causal Bayesian network model to study readmissions reduction for chronic obstructive pulmonary disease (COPD) patients. The model employs a Bayesian network learning method and adopts domain knowledge. Using this model, we analyze the impacts of critical variables on a patient's readmission risk by manipulation of such variables. Through this analysis, effective intervention options to reduce readmission can be identified, which can provide a quantitative tool for designing personalized interventions to reduce COPD readmissions.