A major shortcoming of traditional clinical decision support (CDS) systems is that they operate on highly incomplete patient data, which sets the foundation of the usefulness of the tool. With incomplete data – prompts down the line are guaranteed to be inaccurate. Besides having access to all data, precision of this information is imperative for a CDS system to enhance care delivery and patient outcomes.
Traditional CDS systems can read and utilize only structured data entries in electronic medical records (EMRs) and ancillary systems. However, this portion of the patient’s clinical profile, represents somewhere between 30 and 40 percent of all medical information. If the system doesn’t have complete and accurate data, it’s going to error. Data and data comprehension is key.
Capturing all data and precise comprehension of this information requires the CDS system to function as closely as possible to how a physician thinks. Clinicians communicate patient data primarily through narrative reports, follow-up notes, and summaries of CT scans, X-rays and other imaging reports. In general, dictation, turned later in to free text notes, is how a significant data portion is entered into EMRs.
Approximately, 50 to 70 percent of data in healthcare, if not more, resides in non-retrievable, unusable information embedded in free text communication. This is an enormous amount of vital data that traditional systems simply can’t process because they don’t have the “intelligence”. Seeing only a small portion of the total clinical picture makes traditional CDS systems prone to error.
Why traditional CDS systems alerts are inaccurate
In traditional CDS systems, prompts, which are intended to help clinicians in “actionable” ways, are often ignored. They are ignored, because being based on incomplete information, they lead to alerts supplied and actions prompted that are frequently inaccurate. These inaccuracies, for example, may advise for an action already completed, or that an action should be delivered at a later stage in the clinician’s workflow and not at the point of care. Obviously, when these questionable prompts do occur, they are not useful.
Unreliable prompts not only frustrate clinicians, they cause alert fatigue, which further leads clinicians to mistrust the CDS system and ultimately ignore the prompts. The result for the tool, which should be highly actionable, then ceases to assist in a crucial way. This is unfortunate because CDS systems, like the computer system in an aircraft, could save lives. In fact, a recently published “PLOS One” study found doctors accepting only 40 percent of 24,849 drug interaction alerts due to inaccuracy and inappropriateness.
However, if clinicians had a truly trustworthy CDS system, one that suppressed “noisy” alerts, and delivered only precise prompts with accurate recommendations at the point of care, acceptance and response rates would significantly increase. The right technology can deliver that value to benefit clinicians and patients. This must start and end with complete and precise data that can be acted upon. The computer must see the total, clinical normalized patient profile and comprehend the information. A CDS system that is precise and understands in context, would contribute to the clinician’s expertise and actions and surely be an asset at point of care, enterprise-wide. But it must not interrupt workflow or require multiple steps and hours of training; it must seamlessly fit into the clinical workflow.
Technical innovations to increase value and reliability of CDS systems alerts
As explained above, the first requirement for precise and useful alerts is for the CDS system to be capable of reading and understanding in context all clinical information in the patient record. This is the starting point and makes it possible for the system to perform an accurate analysis of the total clinical situation. In order for the system to achieve this, it must be expertly maintained in terms of storing comprehensive content for customized best practices.
For the above to work, the system must be capable of intelligently analyzing information. For instance, the CDS system must be nimble enough to “think” and differentiate, say, between an obstetrics patient versus a surgical patient and apply a different set of standards of care and compliance based on unique clinical criteria.
The above is very different from If/then rules that cannot accomplish intuitive, evidence-based processes required for infinite clinical situations. In order for a CDS system to significantly help clinicians, it must continuously and in real-time precisely understand an enormous amount of data. Only in this way will the tool be capable of delivering meaningful recommendations – and do the job of heightening clinical response, improve care delivery protocols adherence and achieve better outcomes.
Why IT-programmed tools are problematic: the role of the CDS systems vendor
Because traditional CDS systems have been functioning with huge gaps, explained above, the tool has been very ineffective – resulting in a high margin of error. The advanced level of artificial, intuitive intelligence, needed to turn the effectiveness and results around, which requires programming by teams of expert clinicians and technologists. This is the vendor’s responsibility – not the client’s IT department.
CDS system programming must be an ongoing process; it is not a static operation. This includes the system being able to trace how clinicians use the solution – whether in outpatient or inpatient settings. This aspect is crucial since each care delivery situation is quite specific. The system must, without fail, identify lack of use, inappropriate or inefficient use, alert fatigue, and false positives – and automatically relay this information to the vendor. It is the vendor’s job to continuously improve its CDS system to create a successful, productive, and satisfying user experience.
The next wave of CDS systems
There is no question that CDS systems can and must become a useful healthcare tool. We have seen in industries as diverse as commercial aviation and manufacturing, how computer systems contribute to quality, efficiency, and safety. Clearly, CDS systems will never replace physicians any more than flight systems have replaced pilots, but the system can heighten care quality and outcomes exponentially.
The role of CDS systems in clinical situations is to assist physicians in making the most informed decisions. Evidence-based medicine and best practices guidelines require an advanced level of artificial intelligence, embedded into CDS capabilities that aligns with the particular organization’s needs. Medicine will always be a mixture of art and science, but with a precise, exceptionally intelligent system – care delivery and clinical outcomes can be enhanced.
 “Are We Heeding the Warning Signs? Examining Providers’ Overrides of Computerized Drug-Drug Interaction Alerts in Primary Care,” Sarah P. Slight et al, PLOS ONE Dec. 26, 2013.