Learning T2D Evolving Complexity from EMR and Administrative Data by Means of Continuous Time Bayesian Networks

Predicting the complexity level (i.e. the number of complications and their related hospitalizations) in a T2D cohort is a critical step in prevention, resource optimization and overall patient management. Our data set was obtained by monitoring a T2D diabetic cohort along up to 10 years through electronic medical records of a local healthcare agency data warehouse. In order to conveniently handle temporarily sparse data, we designed a model describing the cohort evolution with Continuous Time Bayesian Networks (CTBN). The network structure and its parameters are entirely data driven. Compared to traditional Bayesian Networks, CTBNs admit cycles. As consequence, CTBNs fit the complexity of chronic metabolic syndromes where variables show a reciprocal influence. Network nodes represent metabolic (glycated hemoglobin, lipid profile (cholesterol, triglycerides), and biometric (BMI) data. We observed how these variables directly or indirectly affect the disease level of complexity, and how the variables influence the cumulative adverse events a patient undergoes.

[1]  R. Goldberg,et al.  Lipid Management in Type 2 Diabetes , 2006 .

[2]  Daphne Koller,et al.  Continuous Time Bayesian Networks , 2012, UAI.

[3]  Aesha Drozdowski,et al.  Standards of medical care in diabetes. , 2004, Diabetes care.

[4]  Ron Goeree,et al.  A Review of Methods Used in Long-Term Cost-Effectiveness Models of Diabetes Mellitus Treatment , 2012, PharmacoEconomics.

[5]  David Hsu,et al.  Probabilistic Approximations of Signaling Pathway Dynamics , 2009, CMSB.

[6]  Xiang Wang,et al.  Unsupervised learning of disease progression models , 2014, KDD.

[7]  Konstantina S. Nikita,et al.  A data gathering framework to collect Type 2 diabetes patients data , 2014, IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI).

[8]  Claudio Cobelli,et al.  A Dynamic Bayesian Network model for long-term simulation of clinical complications in type 1 diabetes , 2015, J. Biomed. Informatics.

[9]  Riccardo Bellazzi,et al.  Temporal data mining and process mining techniques to identify cardiovascular risk-associated clinical pathways in Type 2 diabetes patients , 2014, IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI).

[10]  Jing Xu,et al.  Continuous Time Bayesian Network Reasoning and Learning Engine , 2010, J. Mach. Learn. Res..

[11]  Daphne Koller,et al.  Learning Continuous Time Bayesian Networks , 2002, UAI.

[12]  Fabio Stella,et al.  A continuous time Bayesian network model for cardiogenic heart failure , 2012 .

[13]  Fabio Stella,et al.  Continuous Time Bayesian Networks for Gene Network Reconstruction: A Comparative Study on Time Course Data , 2014, ISBRA.

[14]  V. Foos,et al.  Validation of the IMS CORE Diabetes Model. , 2014, Value in health : the journal of the International Society for Pharmacoeconomics and Outcomes Research.