Clinical Decision Support Systems (CDSSs) comprise one of the fastest growing and most widely discussed areas of Health Information Technology (HIT) in existence today. CDSSs have been defined as “Active knowledge systems which use two or more items of patient data to generate case specific advice,” (Wyatt J, Spiegelhalter D, 1991); or there’s this from Robert Hayward of the Center for Health Evidence, “Clinical Decision Support Systems link health observations with health knowledge to influence health choices by clinicians for improved health care.”
Put more succinctly, a CDSS is a computerized system that assists health care professionals with some part of the process of providing patient care. This of course is a very broad definition, and intentionally so. CDSSs range from very specialized tools used to help determine diagnosis within a narrow spectrum of possibilities to far reaching systems integrated into Electronic Health Records (EHR) systems designed to seamlessly assist and alert healthcare professionals as records are being made and updated. One of the most broadly implemented types of CDSS at this time are those that monitor drug prescriptions and provide alerts for potential negative interactions between medications.
But CDSSs can do more than just monitor drug interaction, or provide specialized assistance in a narrow field. As technology develops, and particularly as EHR systems become the norm, the potential for CDSSs becomes great. On the one hand you have knowledge based CDSSs. These are the more common type, and consist of a knowledge database, an inference drawing system, and a means of receiving input. The user inputs specific data about a patient, the system integrates that data with knowledge from its database, and then, typically through a very complex series of “if, then” queries, the inference engine is able to provide information, whether probable/possible diagnosis, suggested treatment options, etc. These systems work very well, but require a great deal of information as a base.
Alternatively, non-knowledge based CDSSs may be even more impressive. While still less common and more often used in the narrower, more specialized applications mentioned above, non-knowledge based CDSSs use artificial intelligence technology that allows them to “learn” as they go. These systems use very sophisticated algorithms modeled on systems like neural networks and genomes to learn as they process information. The benefit of this is that they do not require the massive databases knowledge-based systems do; however, in some ways it is less transparent to the user how the system is arriving at its conclusion.
As we mentioned above, the functions CDSSs can serve are fairly broad. While the idea of machines that crank out diagnoses or treatments is perhaps the most likely idea to draw a headline, CDSSs are just as likely to be used for more mundane tasks that, far from taking the place of doctors, could free them up to focus more on patients. L. Perreault and J.A. Metzger identify four common uses.
- “Administrative: Supporting clinical coding and documentation, authorization of procedures, and referrals.
- “Managing clinical complexity and details: Keeping patients on research and chemotherapy protocols; tracking orders, referrals follow-up, and preventive care.
- “Cost control: Monitoring medication orders; avoiding duplicate or unnecessary tests.
- “Decision support: Supporting clinical diagnosis and treatment plan processes; and promoting use of best practices, condition-specific guidelines, and population-based management. “
So what about risk management? Legislators and policy makers in Washington clearly think that CDSSs are keys to reducing risk in the long run. The Health Information Technology for Economic and Clinical Health (HITECH) Act, which is part of the American Recovery and Reinvestment Act (ARRA) of 2009 requires not only the implementation and meaningful use of EHR systems, but also the implementation of at least one CDSS type rule in the first round. Failure to do so could mean reductions in Medicare and Medicaid payments.
But most of us want more than just what Washington thinks. Thankfully, we’re beginning to get more and more hard data on the impact that implementation of CDSSs have on risk management. For instance, a recent study that tracked malpractice claims over a seven year period found that over half of the substantiated claims “were potentially preventable with CDS[S].” This seems to be consistent with other studies, most of which indicate that, broadly speaking, the implementation of CDSSs leads to reduced malpractice.
However, specific types of CDSSs produce more mixed results. For instance, a known issue with one of the most common types of CDSS, that which tracks patient medications, is what is known as alert fatigue. This occurs when the systems provide alerts for every possible drug-drug interaction (DDI), including the most rare and/or least likely to be harmful. When this happens the tendency is for physicians to pay less attention to alerts in general, or even to turn the feature off. Unfortunately this leaves an electronic audit trail that could actually expose the physician to greater liability. The obvious solution to this seems to be to provide alerts only for clinically significant potential interactions; however, vendors that provide the systems have been unwilling to do this in many cases because they fear liability reverting to them. There are solutions to problems like this, and they are being actively pursued, but it will take time to work these kinds of kinks out.
Overall, we think there is room to be optimistic about the development and implementation of CDSSs. Medicine is such a complex and deeply human field that no fears of machines taking over the roles of doctors need worry us. There may be some back and forth between regulators, insurers, and doctors over what and how much should be automated, but ultimately we foresee CDSSs, like so many technologies, settling into a role that truly supports the more efficient, effective, and safe practice of medicine.
For Further Reading