Table 1.
The prevalence of the five comorbidities in the final dataset based on the five years of care.
Table 2.
Demographics of the patients included in the study.
Fig 1.
General Scheme of the proposed method.
Fig 2.
Implementation of the Proposed Algorithm (Unsupervised).
Fig 3.
The network structure inferred from literature review.
Fig 4.
Learned BN structure from the proposed method (Unsupervised Method).
Table 3.
Similarity between the learned matrices.
Table 4.
The Area Under the Curve (AUC) performance of the competing methods for predicting future comorbidities, given comorbidity information of past years.
Table 5.
Conditional probabilities of the comorbidities in year two, given year one data, for a sample patients with the following risk factors, gender (male), marital status (unmarried), education (less than high school), race (white), and age (18-30).
Table 6.
The confusion matrices of the unsupervised, semi-supervised, and supervised MTBNs for prediction of comorbidities of year 5, given year 1, 2, 3, 4 data, based on a 50% threshold.
Fig 5.
The longest paths in the MTBNs. From left (a) Unsupervised Network (b) Semi-supervised Network and (c) Supervised Network.