Improving Clinical Decision Support with Advanced Data Analytics
The University of California Davis Health System won the HIMSS Davies Award of Excellence in 2013, in part because of a system they developed to alert physicians through the Electronic Health Record (EHR) if a patient presented with symptoms indicative of sepsis. The system was developed by analyzing the electronic health records of 741 patients. The researchers used machine learning and determined that vital signs, combined with serum white blood cell count could be used to accurately predict hyperlactatemia. Hyperlactatemia in turn is associated with severe sepsis and septic shock. UC Davis’ system was designed to flag vital signs and white blood cell counts indicative of sepsis and suggest an appropriate treatment regimen, leading to a reduction in the mortality rate due to sepsis from 47% to 21% and saving more than 200 lives over a 3-year period at UC Davis. (UC Davis Health, 2014)
UC Davis’ sepsis warning system is an example of a Clinical Decision Support System (CDSS). UC Davis’ research is also an example of how big data and machine learning can be used to derive important correlations and make predictions designed to improve patient care. The combination of CDS Systems, big data and data analytics promises to revolutionize healthcare, by improving patient care while also potentially reducing the cost of healthcare.
Clinical Decision Support Systems
A Clinical Decision Support System (CDSS) is designed to assist physicians in making clinical decisions. In the case of U.C. Davis, the CDSS was designed to flag vital signs and White Blood Cell counts indicative of sepsis and provide a treatment recommended. According to Robert Hayward of the Centre for Health Evidence, “Clinical decision support systems link health observations with health knowledge to influence health choices by clinicians for improved health care”. (Karthigeyan et al, 2014)
There are 2 classes of CDSS:
Knowledge Based-Knowledge based systems consist of a knowledge repository, an interface engine and a user interface. The knowledge repository holds information from various sources along with a ruleset, typically in the form of what-if scenarios. The interface engine functions as a search engine to sift through the knowledge base for the correct data. The user interface allows the user to interact with the CDSS and review the results.
Non-Knowledge Based Systems-Non-Knowledge based systems on the other hand, develop knowledge through machine learning. Non-Knowledge based systems can start with relatively little data and learn from experience to define patterns in clinical data. There are primarily 3 types of Non-Knowledge based systems, based on function:
Support Vector Machines-Support Vector Machines simulate learning by dividing data into two categories. Support Vector systems are provided with examples, then use those examples to refine and learn from the available information.
Artificial Neural Networks (ANN)-Artificial Neural Networks mimic the function of the human brain. They consist of a variety of interconnected nodes which function as neurons. The neurons and the connections between them are weighted. A ANN is initially trained to solve a certain type of problem. It then learns by comparing the solution it produces from a given type of data, against the expected results, producing a margin of error. The ANN then adjusts the weight of its nodes and connections to optimize the results. This reevaluation of the weights is how the ANN learns.
Genetic Algorithms-Genetic algorithms learn by using a natural selection process to solve a problem. The algorithm evaluates a series of solutions to a problem. The solutions closest to optimal are then reevaluated and recombined, until the optimal solution is found.
According to Dr. Jerome Osheroff, author of Improving Outcomes with CDS: An Implementers Guide, Clinical Decision Support (CDS) systems should be designed to provide support by intervening in the patient care process at the most efficacious time. According to Dr. Osheroff, CDS intervention falls into 4 different categories:
Data Entry-A data entry intervention, involves the CDS providing data at an appropriate time. An example of a data entry CDS intervention is the order set tool where a collection of preformed orders are released in response to a specific procedure or disease state. UC Davis’ Sepsis CDS system is an example of data entry intervention because it provides an order set in response to vital signs and CBC readings indicating the occurrence of sepsis.
Data Review-A data review systems allows healthcare professionals to monitor a patient’s vital signs remotely in real time and provides physicians with an appropriate lifesaving regimen if vital signs and other readings reach a critical level.
Assessment and Understanding-This type of intervention is designed to provide information when the physician is formulating a treatment plan. An example of an Assessment and Understanding intervention is the Health Level 7 (HL7) info button, which appears in the EHR next to a patient list of conditions or medications. If the physician wants to know more about a specific disease in the list of patient conditions, they can click on the Infobutton and read a synopsis of the disease.
Triggered by User Tasks-This type of intervention occurs for events outside of the normal patient care workflow. For example, if an abnormal test result appears in the EHR, a text message or email or audible alert might be triggered.
Dr. Osheroff suggested a framework for designing CDS systems to be as effective as possible. This framework consists of 5 requirements or “rights” that a CDS should meet to provide timely and effective support.
The Right Information-The information presented to the end user should be evidence based and derived from recognized guidelines or national performance measures.
The Right Person-A CDS should provide information to a person or team who can act upon the information.
The Right Intervention Format-A CDS may use a variety of interventions to inform patients, care givers and healthcare professionals. These interventions may include alerts, order sets, Infobuttons and others. It is necessary therefore to design the CDS with the appropriate intervention format to perform its task. For example, it is appropriate for a CDS to initiate an order set if conditions indicative of sepsis is detected, while an alert is more appropriate if vital signs drop below a certain threshold.
The Right Channel-CDS interventions may be delivered over a variety of channels, including the EHR, the Order Entry System or a smartphone app. The correct channel depends primarily on who the information is being delivered to. For example, if an alert is intended for the physician, the EHR is an appropriate channel, while an alert intended for the patient may go to a smart phone app.
The Right Time in the Workflow-CDS intervention at the correct time in the patient workflow is critical. For example, a physician is ordering an H2 Receptor Antagonist, but the patient is taking warfarin (a blood thinner). The physician should be alerted immediately as they begin typing in the name of the H2 Receptor Antagonist, when they can select a better option, rather than at the very end of the ordering process when the medication order is ready to be transmitted to the pharmacy.
Dr. Osheroffs list of CDS rights provides a framework that a properly designed CDS should follow to ensure that the proper information is provided to the right people, at the right time and through the proper channel and format.
Benefits of CDSS
The benefits of a CDSS are numerous and include reduced medication errors, a reduction in misdiagnosis and improved efficiency and patient throughput. According to the Institute of Medicine, between $17-29 billion is wasted each year due to misdiagnosis and unnecessary or inaccurate patient care. (Hitchcock, 2011) Clinical Decision Support Systems help to reduce this cost and improve efficiencies by assisting the healthcare professionals to calculate the correct dosage of medication and reduce duplicate or unnecessary tests thereby saving time and eliminating wasteful expenditures.
A 2005 study indicated that CDS systems were most effective when they met the following criteria:
The CDSS is integrated into the clinical workflow.
The CDSS is electronic based, rather than paper based.
The CDSS provides clinical support during patient care, rather than before or after the patient encounter has occurred.
The CDSS provides recommendations for care, rather than simply assessments.
Federal Regulations and CDS
In 2009 President Barack Obama enacted the Health Information Technology for Economic and Clinical Health Act (HITECH) and the American Recovery and Reinvestment Act (ARRA). These laws were designed to increase the use of interoperable electronic health records throughout the United States. HITECH is designed to increase the use of EHR and interoperative technology by incentivizing the adoption and use of EHR and related systems through Meaningful Use. Healthcare organizations and providers who can demonstrate meaningful use of certified EHR technology are provided with an incentive payment. The Meaningful Use program is based on the 5 pillars of health outcomes policy priorities: (CDC, 2017)
Improving quality, safety, efficiency, and reducing health disparities
Engage patients and families in their health
Improve care coordination
Improve population and public health
Ensure adequate privacy and security protection for personal health information
Meaningful Use is divided into 3 phases, each with a series of core and menu goals. Phase 1 was launched in 2011 and focused on data capturing and sharing using a certified EHR. Phase 2 was launched in 2014 and focused on extending the use of EHRs including the adoption of CDS Systems to improve performance. Phase 3 began in 2017 and focuses on driving interoperability between EHR systems and improving patient outcomes.
After the 2017 election, the future of Meaningful Use has become somewhat unclear, as President Donald J. Trump seeks to increase profitability and reduce government intervention.
Another law pertinent to Clinical Decision Support Systems is the Protecting Access to Medicare Act which will require referring physicians to consult Appropriate Use Criteria (AUC) before ordering advanced diagnostic imaging services including MRI, CT, Nuclear Medicine and PET for Medicare patients starting in 2020. As a result, a AUC Clinical Decision Support system is being included in many EHRs.
The Future of CDSS
The future of Clinical Decision Support Systems lies in greater integration with the EHR, more sophisticated analytics and automation. According to Dr. Anil Jain, Chief Medical Information Officer with IBM Watson Health, EHR systems in the future will offer greater integration with CDS systems. CDS systems in the future will have fewer alerts because the system will be tailored to the needs of the of the physicians and patients. For example, more experienced physicians do not require as many interventions as a newer, less experienced physician.
CDS systems may also offer greater automation. For example, future CDS systems might note that a patient with an artificial knee needs a replacement. Instead of simply alerting the physician, the EHR would order and schedule the procedure and notify the physician.
Quartz Clinical is an advanced healthcare data analytics platform that uses an artificial intelligence engine to provide deep insight into all facets of your healthcare organization. In addition to financial and performance analysis, Quartz Clinical advanced analytics automates complex data analysis to assist in decision making. Quartz Clinical is also flexible and can integrate with any EHR as well as any database.
Clinical Decision Support Systems combine machine learning, analytics and big data to assist the physician in making clinical decisions. Clinical Decision Support Systems have demonstrated the ability to improve patient care, reduce errors and eliminate unnecessary orders. The adoption of CDS systems has been encouraged by government regulations such as HITECH and the Meaningful Use program, although the continuation of these programs is now somewhat in doubt. The future of CDS should involve more sophisticated analytics, automation and a more tailored integration with the EHR.
Abbasi, M M, and S Kashiyarndi. Clinical Decision Support Systems: A Discussion on Different Methodologies Used in Health Care. www.idt.mdh.se/kurser/ct3340/ht10/FinalPapers/15-Abbasi_Kashiyarndi.pdf+.
Baum, Stephanie, et al. “How Will EMRs Use Clinical Decision Support Tools in the Future?” MedCity News, Breaking Media Inc, 29 Mar. 2017, medcitynews.com/2012/07/how-will-emrs-of-the-future-use-clinical-decision-support-tools/.
“Case Studies and Current Trends.” HIMSS, HIMSS, 29 Dec. 2016, www.himss.org/library/clinical-decision-support/trends-case-studies.
Cortada, James W, et al. “The Value of Analytics in Healthcare.” IBM, www-01.ibm.com/common/ssi/cgi-bin/ssialias?subtype=XB&infotype=PM&appname=GBSE_GB_TI_USEN&htmlfid=GBE03473USEN&attachment=GBE03473US.PDF.
Dowding, Dawn. “Using Computerised Decision-Support Systems.” Nursing Times, Using computerised decision-support systems.
Hardin, J Michael, and David C Chhieng. “Data Mining and Clinical Decision Support Systems.” Data Mining and Clinical Decision Support Systems, www.cs.uky.edu/~goldsmit/cdm/DMCDSS.pdf+.
Hitchcock, Robert. “Top 5 Benefits of Clinical Decision Support in the ED.” Healthcare IT News, HIMSS Media, 28 Nov. 2011, www.healthcareitnews.com/blog/top-5-benefits-clinical-decision-support-ed.
“ .” IBM The Value of Analytics in Healthcare, IBM, www-935.ibm.com/services/us/gbs/thoughtleadership/ibv-healthcare-analytics.html.