A fundamental question about any (Clinical Decision Support System) CDS is just how good is it, i.e. does it get the right answer for generic and specific patients? If it doesn’t this may be the result of one or more issues such as flawed information having been used to build the system, flawed programming, or the patient being outside of an often undefined or ill defined population when for another population the CDS does actually provide the right answer
A common CDS disclaimer is that it is always up to the practitioner to second guess the CDS as necessary, or in other words, the CDS is not actually supposed to be relied upon. Depending on the complexity of the underlying theory and data, the practitioner may or may not in reality have the ability to do this, or they may not have a more rational basis for doing so than “I don’t think that is right”. Such a conclusion would put the practitioner outside of what might be considered a practice guideline. On the other hand if a CDS is easy to second guess, then it might not be very valuable in the first place.
In this context comes the recent controversy over the new cholesterol and statin on line “risk calculator”. As first reported in the New York Times, it was determined that an online risk calculator overestimated patient specific risk by an average of 100% (100 here is not a typo). If action were based on this erroneous calculator, statin therapy would be substantially over-prescribed. In this regard the Times cites a statement from the organizations that published the guidelines that will continue to be a CDS classic: patients and doctors should discuss treatment options rather than blindly follow a calculator. Or, in other words, it is not to be relied on.
Apparently the problem with the risk calculator is at least in part that the risk data on which it was based was decades old and therefore did not apply to the current US population which in at least some ways has actually gotten healthier. In addition the mathematical model used was one of linearly increasing risk which has not been demonstrated to be correct. Thus the flaws in the calculator were a result of the inappropriateness and lack of justification of the knowledge bases used to build it. Despite these fundamental issues, no plans to remove or revise the calculator were identified.
This risk calculator was not imbedded within an EHR, and it requires manual input of multiple patient parameters. And of course there are additional potentially relevant patient parameters that are not part of the calculation. However something like this certainly could be part of an EHR either by direct integration, or by pointing the EHR user to it and perhaps automatically using relevant patient information that might already be in the EHR.