The objective of this research was to develop methodologies for early warning of the abnormal conditions from Intensive Care Unit (ICU bedmaster and electronic health record (EHR) data that can lead to outcomes such as cardiovascular events, mortality, cardiac arrest and long stay in critical care units. The technical framework includes data quality control and sequential temporal data for prototyping a Clinical Decision Support System (CDS).
Due to HIPAA compliance issue at Georgia Tech, accessing CHOA data at Georgia Tech became a challenge. Thus, we have used MIMIC-II data (MIMIC II (Multiparameter Intelligent Monitoring in Intensive Care II) that consists of more than 40,000 ICU stay records of more than 30,000 patients) to prototype the methods needed for this work, For missing data in EMR, we assign the missing data into three categories: missing completely at random (MCAR), missing at random (MAR) and missing not at random (MNAR). For each missing data type we propose a novel imputation scheme.
We then evaluate this process using Random Forests and using Matthew’s correlation coefficient as the evaluating metric. Our results indicate that novel imputation techniques outperformed no filling and standard mean filling techniques with a statistical significance p =.01.
Then on ICU Clinical Decision Support, we used Association Rule Mining (ARM) with five metrics to report associations among various clinical data under the direct guidance of ICU physicians. icuARM features user-friendly interfaces that enable real-time mining of these clinical data in ICU setting. Applying our newly designed icuARM, we have investigated the associations between prolonged ICU stay possibility and the patient conditions such as demographics, and pre-existing comorbidities etc.
Venugopalan J, Chanani N, Maher K, and Wang MD (2017) “Combination of static and temporal data analysis to predict mortality and readmission in intensive care unit” in “ Conf Proc IEEE Eng Med Biol Soc, EMBC. 2017
Cheng C, Sha Y, and Wang MD (2016) InterVisAR: An Interactive Visualization for Association Rule Search,” in ACM Conference on Bioinformatics, Computational Biology and Health Informatics (ACM-BCB) 2016 Annual International Conference of ACM 2016.
Sha Y, Venugopalan J, Chanani N, Maher K, and Wang MD (2016), ” A Novel Temporal Similarity Measure for Patients Based on Irregularly Measured Data in Electronic Health Records,” in ACM Conference on Bioinformatics, Computational Biology and Health Informatics (ACM-BCB) 2016 Annual International Conference of ACM 2016.
Sha Y, and Wang MD (2017) , ” An Attention-based Recurrent Neural Network for Interpretable Predictions of Clinical Outcomes,” in ACM Conference on Bioinformatics, Computational Biology and Health Informatics (ACM-BCB) 2017 Annual International Conference of ACM 2017.
Venugopalan J, Hoffman R, Cheng C, and Wang MD, (2014) “Time-series data analysis to predict mortality and cardiac arrest in pediatric populations,” in 2014 Pediatric Healthcare Innovation Conference
Venugopalan J, Chanani N, Maher K, and Wang MD. (2015) “Data quality control for the improved prediction of readmission in intensive care unit” in Annual Critical Care Congress,SCCM, 2015
Changdae L, Cheng C, and Wang MD (2014) icuARMmobile – An Mobile interface of Clinical Decision Support using Association Rule Mining. 2014 Wireless Health Conference.