Personalized treatment in glycemic control (GC) is a visibly promising research area that requires improved mechanisms providing patient-specific procedures to enable complicated decision support. Available per-patient data must be more than written records, and be fully integrated in this personalization process. This article presents a process for relating the intensive care unit patients' demographic and admission data to their GC performance. With this objective, a probabilistic Bayesian network was chosen to provide more personalized decisions. As a case study, average daily blood glucose measurements were chosen as the interest target node in order to weigh GC that provides a reduced nursing workload. To test the idea, data from 482 patients, with nine variables from four Malaysian intensive care units with different controls were exploited. The identified steps crucial in building a dependable model are variable selection, continuous state discretization, and unsupervised structure learning. Using a multi-target node evaluation, a network with 80% mean overall classification precision was obtained with a normalized equal distance discretization algorithm and a maximum weight spanning tree technique. Meanwhile, the interest target node scored 90.39% precision. The results from this study, which are complemented with an evaluation of missing data, are proposed as a benchmark for using Bayesian networks in this type of application.
|Number of pages||8|
|Journal||IEIE Transactions on Smart Processing and Computing|
|Publication status||Published - 01 Jan 2019|
All Science Journal Classification (ASJC) codes
- Signal Processing
- Electrical and Electronic Engineering