Department of Health Services Administration, China Medical University, 91 Hsueh-Shih Road, Taichung 40402, Taiwan, ROC.
Institute of Information System and Applications, National Tsing Hua University, Hsinchu, Taiwan.
Comput Methods Programs Biomed. 2020 Jul;191:105409. doi: 10.1016/j.cmpb.2020.105409. Epub 2020 Feb 25.
Many studies regarding health analysis request structured datasets but the legacy resources provide scattered data. This study aims to establish a health informatics transformation model (HITM) based upon intelligent cloud computing with the self-developed analytics modules by open source technique. The model was exemplified by the open data of type 2 diabetes mellitus (DM2) with related cardiovascular diseases.
The Apache-SPARK framework was employed to generate the infrastructure of the HITM, which enables the machine learning (ML) algorithms including random forest, multi-layer perceptron classifier, support vector machine, and naïve Bayes classifier as well as the regression analysis for intelligent cloud computing. The modeling applied the MIMIC-III open database as an example to design the health informatics data warehouse, which embeds the PL/SQL-based modules to extract the analytical data for the training processes. A coupling analysis flow can drive the ML modules to train the sample data and validate the results.
The four modes of cloud computation were compared to evaluate the feasibility of the cloud platform in accordance with its system performance for more than 11,500 datasets. Then, the modeling adaptability was validated by simulating the featured datasets of obesity and cardiovascular-related diseases for patients with DM2 and its complications. The results showed that the run-time efficiency of the platform performed in around one minute and the prediction accuracy of the featured datasets reached 90%.
This study helped contribute the modeling for efficient transformation of health informatics. The HITM can be customized for the actual clinical database, which provides big data for training, with the proper ML modules for a predictable process in the cloud platform. The feedback of intelligent computing can be referred to risk assessment in health promotion.
许多关于健康分析的研究都需要结构化的数据集，但传统资源提供的是分散的数据。本研究旨在建立一个基于智能云计算的健康信息学转换模型（HITM），并利用开源技术开发的分析模块。该模型以 2 型糖尿病（DM2）相关心血管疾病的开放数据为例进行说明。
采用 Apache-SPARK 框架构建 HITM 的基础设施，支持机器学习（ML）算法，包括随机森林、多层感知机分类器、支持向量机和朴素贝叶斯分类器以及智能云计算的回归分析。该模型应用 MIMIC-III 开放数据库作为示例，设计健康信息学数据仓库，其中嵌入基于 PL/SQL 的模块，用于提取分析数据进行训练过程。耦合分析流程可驱动 ML 模块训练样本数据并验证结果。
比较了四种云计算模式，以评估云平台根据其系统性能在超过 11500 个数据集上的可行性。然后，通过模拟 DM2 及其并发症患者的肥胖和心血管相关疾病的特征数据集，验证了建模的适应性。结果表明，平台的运行效率约为一分钟，特征数据集的预测精度达到 90%。
本研究有助于促进健康信息学的高效转换建模。HITM 可以针对实际的临床数据库进行定制，为训练提供大数据，并在云平台上使用适当的 ML 模块进行可预测的过程。智能计算的反馈可用于健康促进中的风险评估。
