BGU’s Medical Informatics Research Center has made two significant breakthroughs in proving the efficacy of computer-based decision support for human clinicians. In two papers recently published in the top journals in the field, The International Journal of Medical Informatics and The Journal of the American Medical Informatics Association, Prof. Yuval Shahar, Dr. Erez Shalom, and Mr. Denis Klimov, outline the research milestones of how medical informatics can impact the clinical workflow and medical research.
The first paper (part of the Ph.D. research, advised by Prof. Yuval Shahar and Prof. Eitan Lunenfeld, of Dr. Erez Shalom , who is now a research scientist at the Medical Informatics Research Center) presents the first rigorous evaluation of the effect of computer-based decision support on doctors’ decisions when managing patients continuously, over a significant period of time. The study has demonstrated that providing clinicians with a computer's series of recommendations, generated through an application of established formal clinical-guideline knowledge within a medical decision-support system that was built at the BGU Medical Informatics Research Center, dramatically enhanced their compliance with medical guidelines, increased the soundness of their actions, and prevented the performance of many redundant actions.
The IJMI paper by Erez Shalom, Prof. Yuval Shahar and their collaborators ( Prof. Eitan Lunenfeld of the Faculty of Health Sciences and the head, at the time of the study, of the Soroka University Medical Center’s Ob/Gyn Division, and Dr. Yisrael Parmet of the Department of Industrial Engineering and Management), describes the rigorous evaluation of the BGU Medical Informatics Research Center’s Picard clinical-guideline application framework in the pre-eclampsia (toxemia of pregnancy) domain, at Soroka University Medical Center. In this particular case, the researchers specified formally (i.e., digitally), within Picard’s knowledge base (which is in fact the BGU Medical Informatics research Centers’ Digital Guideline Library, DeGeL), the American College of Obstetrics and Gynecology (ACOG) guideline for management of pre-eclampsia. The ACOG guideline serves as
the basis for care at most obstetrics wards, including those of Soroka.
Dr. Shalom’s framework was assessed with the assistance of 36 obstetricians (24 residents and 12 board-certified experts); each clinician had to make 60 different decisions regarding six different highly realistic clinical pre-eclampsia complete longitudinal (continuous-care) simulated scenarios created by a team of senior obstetricians, led by Prof. Eitan Lunenfeld . (The experimental software could not be applied to real patient’s data; however, the longitudinal patient records representing each scenario were generated in realistic fashion, using a specialized software module developed at the Medical Informatics Research Center). Half of the decisions were made manually, the clinician being provided only with the longitudinal electronic medical record and the fetal growth chart; the other half of the same clinicians’ decisions were made using the same resources, but this time, they were augmented by the recommendation of the Picard system, which could be accepted or rejected; an explanation for the advice could be requested as well. The senior obstetricians also served as referees of the quality of all of the clinicians' actions, the ACOG guideline serving as the ultimate “gold standard” of care.
Compliance of the clinicians with the ACOG guideline-based recommendations rose from 41% when not using Picard , to 93% when provided by suggestions from the Picard system.
The percentage of the clinicians’ actions that were judged as correct and necessary per the ACOG guideline increased from 27% when not using Picard , to 91% when using it .
In parallel, the percentage of actions that were deemed redundant per the ACOG guideline [though not erroneous] dropped from 68% when not using Picard, to only 3% when using it -a fact that has a significant economic implications.
Recent data from Johns Hopkins University’s Medical Center supports this surprising redundancy rate. Thus, it is quite possible that 2/3 of clinicians' tests and related actions might well be redundant, possibly due to defensive medicine and/or ambiguity regarding the relevant evidence-based guideline that needs to be followed in each case. An intelligent decision-support system, which considers all of the patient’s data as well as the relevant clinical guideline, might well be one way of significantly reducing that rate.
Current State : The Picard guideline-application engine is currently the key module within the EU MobiGuide project coordinated by Prof. Mor Peleg of Haifa University, Israel. The objective of the MobiGuide project is to help chronic patients anytime, anywhere, and in particular, at their home, manage their care through a set of biosensors and the patient’s smart phone. The MobiGuide project includes 13 partners from five EU and Associated countries (Israel, Italy, Spain, the Netherlands, and Austria). It focuses on supporting the decisions of the patients and on empowering them to manage themselves, by providing them with evidence-based, guideline-based alerts and reminders through their mobile device. In parallel, the MobiGuide system provides their care providers with evidence-based recommendations, based on the best-practice guidelines. All of the core medical knowledge engineering, monitoring, and medical decision-support technology in the MobiGuide project is being developed by the BGU Medical Informatics Research Center.
------------------------------------------------------------
The second paper (part of the Ph.D. research of Prof. Shahar’s student, Denis Klimov, and of the M.Sc. research of another of Prof. Shahar’s students, Alexander Shknevsky) represents a major milestone in the progress of the BGU Medical Informatics Research Center towards a complete framework for supporting the semi-automated and fully automated discovery of new clinical knowledge from "big" clinical data.
The paper demonstrates that established medical knowledge can be used - through interactive and computational data mining and machine learning technologies to discover new and valuable clinical knowledge.
Denis Klimov's JAMIA paper demonstrates the value of using the BGU Medical Informatics Research Center’s Visual Temporal Analysis Laboratory (ViTA-Lab) framework, an integration of two systems (developed by Klimov and by another former Ph.D. student of Prof. Shahar, Dr. Robert Moskovitch), one for discovery of frequent temporal patterns in the data of patient populations that were collected over significant time stretches, and one for their interactive, visual exploration over time.
The researchers have demonstrated that the highly flexible ViTA-Lab framework enables researchers to discover temporal patterns in the data of 22,000 Type II Diabetes patients, which were predictive of the patients’ developing, within up to 5 years, micro- or macro-albuminuria (a significant secretion of protein by the kidneys, a sign of renal-function deterioration in diabetes patients).
It should be noted that renal dysfunction is one of the major problems in diabetes care. It often leads to end stage renal disease (ESRD), costing the health-care system billions of dollars. For example, in the USA alone, treatment of ESRD cost at least $66,000/Yr. per Medicare patient in 2010, with an increase of 8% per year in the following years (USRDS annual data report, 2012).
Current State : The ViTA-Lab engine is currently deployed to analyze clinical longitudinal data in additional clinical domains, such as oncology and intensive care. It is also (along with the Picard guideline-application engine) one of the modules offered by MediLogos Co., a subsidiary of BGU’s commercialization arm, BGN Technologies.
Shalom, E., Shahar, Y., Parmet, Y., and Lunenfeld, E. (2015). A multiple-scenario assessment of the effect of a continuous-care, guideline-based decision support system on clinicians' compliance to clinical guidelines. The International Journal of Medical Informatics 84 (4):248-262. DOI: 10.1016/j.ijmedinf.2015.01.004.
Klimov, D., Shknevsky, A., and Shahar, Y. (2015). Exploration of patterns predicting renal damage in diabetes type II patients using a visual temporal analysis laboratory. The Journal of the American Medical Informatics Association 22 (2):275-289. DOI:10.1136/amiajnl-2014-002927.
