DIGITAL HEALTH

Digitalization of clinical data in the electronic health records (EHR) together with the road to value-based care, shifting reimbursement climate, lifetime cost of treating patients, unsustainability of hospital-based care in chronic disease management, and viability of remote patient monitoring makes the time right in the U.S. to utilize EHR data analytics and artificial intelligence (AI) approaches to identify high-risk, high cost patients.  Utilizing a combination of computer systems engineering, machine learning, and domain specific knowledge of disease and EHR, we are creating an “information solution package” combining EHR-based analytics with patient-specific treatment protocols powered by sensor technologies and big-data rationalized prediction algorithms to reduce morbidity and mortality.

PROJECT: DIGITAL HEALTH

Dmedic: a real time EHR Indexing, visualization, and analysis platform that empowers users to perform big data analytics of EHR

A distributed, JSON-based search and analytics engine that provides real-time data extraction among patient records including full-text search of free-text clinical reports, Dynamic online visualizations which provides visual exploration of the massive EHR data and supports template-based automatic generation of clinical reports of a patient, A driverless AI system for model training, which embeds multiple open-source machine learning platforms as well as GPU technologies for EHR mining.

Deep learning of EHR to identify characteristics associated with early hospital readmission or other costly healthcare encounters after heart failure

Through the analysis of the massive structured and unstructured clinical data in the healthcare, to identify and evaluate prognostic markers and precipitants of rehospitalization or other costly healthcare encounters in patients initially presenting to the hospital with a diagnosis of Acute decompensated heart failure(ADHF)

Unstructured Clinical notes normally contain more knowledge and patient detailed information, besides the traditional NLP strategy which limited by the incomplete ontology of clinical practice and the lack of human annotated labels, we utilize the AI algorithms to extract relevant unsupervised factors from the EHR of a patient and generate the multidimensional patient space based on them

Utilizing new AI algorithms to extract relevant clinical information from the EHR of a patient and develop a personalized management plan optimized for the patient based on the analysis of the massive EHR database.

Big Data-guided, sensor-monitored digital health system for personalized medicine in HF

To define the metrics for sensors monitoring of patients based upon markers predicted to be important in the treatment for ADHF, develop the sensor kit, and conduct pilot study, 

To continually improve in the algorithm(s) and evaluate which additional sensors might best monitor the predictive parameter for each type of HF, work to refine existing or develop new sensor technologies specifically for human chronic disease monitoring, and the initiation of efficacy testing in larger multicenter clinical trials.    

Unstructured Clinical notes normally contain more knowledge and patient detailed information, besides the traditional NLP strategy which limited by the incomplete ontology of clinical practice and the lack of human annotated labels, we utilize the AI algorithms to extract relevant unsupervised factors from the EHR of a patient and generate the multidimensional patient space based on them

Utilizing new AI algorithms to extract relevant clinical information from the EHR of a patient and develop a personalized management plan optimized for the patient based on the analysis of the massive EHR database.

 

Dr. Peng Li is in charge of this project, in collaboration of Dr. Chia-Jung Chang.