Automated Clinical Decision Support (CDS) using Pattern Recognition/Temporal Relationships
Clinical decision support (CDS) has come a long way, most notably when one thinks of decision support, we think of alerts, reminders, drug-drug interaction checking, order sets, and note templates. As “big data” only gets bigger on a daily basis, data warehouses fill with unstructured and structured data which provide a means for developing CDS. “One of the “grand challenges” in CDS is thus the automatic production of CDS from the bottom-up by data-mining clinical data sources” .
Clinical Decision Support has been defined by many authors, though simply put, its “clinical knowledge or patient-related information, filtered or presented at appropriate times to enhance patient care” . Many argue that our current CDS tools with the EHR are quite primitive, traditionally it has been broken down into the following 6 categories, none of which are automated:
Current types of CDS tools 
- Documentation forms/templates
- Relevant data presentation
- Order creation facilitators
- Time based checking and protocol/pathway support
- Reference information and guidance
- Reactive alerts and reminders
Temporal abstraction is an integral component within intelligent data analysis (IDA), which is defined as “encompassing statistical, pattern recognition, machine learning, data abstraction and visualization tools to support the analysis of data and discovery of principles that are encoded within the data” . This was described early on by Stacey et al., in the paper titled “Temporal abstraction in intelligent clinical data analysis: A survey”. Temporal abstraction provides the means to achieve precise, high level qualitative descriptions from low level quantitative patient data, which can then be used as input to a reasoning engine where they are evaluated against a knowledge base to arrive at possible clinical hypothesis .
Temporal reasoning was also described by J.C. Augusto in his paper, “Temporal Reasoning for Decision Support in Medicine” . “Clinicians in general need to know that some symptoms were recurrent with some particular temporal pattern in some specific context in order to diagnose correctly instead of having the rough data that, after possibly long consideration, would lead to the discovery of a certain condition” . The authors provide examples of using temporal reasoning by representing and retrieving time-oriented data for diagnosis, prognosis, and therapy/treatment . For example, using temporal reasoning for online monitoring and detecting trends in blood pressure behavior as an intelligent alarm system in an intensive care unit to provide real time, automated clinical decision support for clinicians .
Future of CDS
The future resides in knowledge discovery in pattern recognition and recommender systems using big data and analytics (i.e. temporal relationships) to develop automated CDS. Essentially, incorporating knowledge discovery in the form of CDS from the practice of others. For example, clinicians managing patients with rare disease/diagnostic uncertainty with limited “evidence” would be presented information to facilitate management based on other physicians and outcomes in real time using pattern recognition from a data warehouse. The future state will encompass data driven, automatically generated CDS content that can reproduce and optimize top-down CDS like order sets while largely avoiding inappropriate and irrelevant recommendations .
Veterans Like Mine – Support for therapeutic decision making. A novel idea, VLMine is planned to serve as a CDS tool when resources like PubMed or UpToDate are not adequately detailed or specific enough to answer questions in patients with clinical uncertainty. On average, clinicians accrue questions about patient care every two to three outpatient encounters, and yet more than half of these questions go unanswered . VLMine will retrieve data from data warehouses about other patients similar to the individual patient at hand and present information to clinicians to facilitate management. This will provide VA clinicians access to information from the collective experience of fellow clinicians using data processing from all facilities. VLMine will constitute a new kind of clinical decision support, a type that does not currently exist.
- Augusto, J. C. (2005). Temporal reasoning for decision support in medicine. Artificial Intelligence in Medicine, 33(1), 1-24.
- Chen, J. H., Podchiyska, T., & Altman, R. B. (2015). OrderRex: Clinical order decision support and outcome predictions by data-mining electronic medical records. Journal of the American Medical Informatics Association : JAMIA,
- Electronic Health Records. A Guide for Clinicians and Administrators; Jerome H. Carter, 2nd edition, 2008.
- Stacey, M., & McGregor, C. (2007). Temporal abstraction in intelligent clinical data analysis: A survey. Artificial Intelligence in Medicine, 39(1), 1-24.
- Veterans like Mine_Cognitive support for therapeutic decision making - matthew samore Retrieved 10/18/2015, 2015, from http://grantome.com/grant/NIH/I01-HX001169-01
Submitted by Umar Iqbal, MD