Impact of an Evidence-Based Computerized Decision Support System on Primary Care Prescription Costs

This paper joins a long literature demonstrating the effectiveness of "smart" prescription systems, dating back to the late 1970's. As the authors note, there is also a literature of failure of prescription decision support systems, and as our group has worked in a related area we have become aware of a large number of unpublished failures as well.

The key variable does indeed seem to be the degree to which the decision support system fits into workflow. Fit is a complex concept however, and very underappreciated. Its power is such that physicians will not only not use, but will actively disable, decision support systems that fit poorly. For example, we have observed that the majority of physicians in the 42 VA centers we are working with have disabled the clinical reminders for hypertension built into their electronic medical record. Those reminders are clinically very appropriate, but from a human factors engineering standpoint are designed inappropriately.

In this study the authors have insured use of the prescribing information because a diagnosis must be entered before a prescription can be written. Human factors engineers refer to that approach as a forcing function. In essence, it achieves fit by changing the workflow rather than the reminder, though of course it is materially aided by having the reminder not require a very large change in the workflow (the prescription system was apparently in use with or without the reminders). Given the low rate of adoption of free-choice reminders, forcing functions are a worthy avenue of further research.

There is one other key variable in the success of online decision support systems, whether built into prescribing modules or electronic medical records. That is the stage of training of the users. Results are consistently better for residents still in the formative stages of their training, than for experienced clinicians whose decision patterns are well established. I did not find where the current paper discussed whether or to what extent residents provided the care under study. Perhaps the authors can comment on that.

Competing interests:   None declared

We appreciate Dr. Fink’s comments and agree with him that successful implementation of a computerized decision support system (CDSS) hinges on the ability of the system to fit seamlessly into physicians’ workflow. We also agree that CDSS should add meaningful value to the patient care experience. In an independent survey conducted by a sponsoring health plan, 100% of surveyed physicians agreed that our system added value to their practice, and most (83%) agreed with the highest emphasis (i.e., felt strongly). Furthermore, the system was so readily accepted at the Affinity Health System study sites, that all 200 of their physicians are now using the CDSS.

One of the features that makes this CDSS so unique is that the information is continuously updated. Every evidence-based review in the system is a living document that is updated by our editorial staff each time an important new study is published. On average, these topics are updated every 2 to 4 months and a complete Medline search is performed a minimum of twice per year to ensure completeness. In addition, the evidence-based “Messages of the Day” are almost exclusively targeted toward recently published literature. In fact, 99% of the daily messages in 2004 have involved studies or treatment guidelines that were published this year. Most of these messages run within a few days or weeks of publication, depending on their importance. Because of the timeliness and quality of their content, our messages and evidence-based reviews are also accredited for continuing medical education by the Washington University School of Medicine in St. Louis, Missouri.

Dr. Fink’s mention of the recent Vioxx (rofecoxib) withdrawal represents yet another excellent example of why CDSS must remain current, as well as another important advantage of electronic health records. Within hours of the first FDA recall alerts, WELLINX had provided every medical office with the contact information for each patient who had an active prescription for Vioxx, as well as information on how patients could receive reimbursement for unused portions of their prescription. Not surprisingly, one of our daily messages in the prior month had summarized five recent reports suggesting that Vioxx was associated with a significantly increased risk of cardiovascular events.

We also agree with Dr. Fink that more comprehensive evaluations are needed to fully assess the cost-effectiveness of CDSS. For that reason, we believe it is important to note that the six-month savings reported in our study were estimated based only on the data for new prescriptions and their refills. These prescriptions represented less than one-fourth of all pharmacy claims during the study period. A 12-month follow-up analysis that assessed differences in the average cost per prescription for all pharmacy claims data (> 75,000 prescriptions in each group) suggested that the group using the CDSS produced an average savings of $843 per physician per month during the first year of use. When changes in the cost per member per month (PMPM) were analyzed, the average savings over the 12-month period were estimated to be $914 per physician per month. In both of these analyses, the average savings in the second half of the year were approximately twice as large as those observed in the first six months. Interestingly, similar results have now been observed in a second controlled trial of our system. Because each of these studies found that savings increase substantially over time, we expect that savings will be higher in subsequent years.

It is also important to note that these estimates represent the savings achieved at a single health plan. Approximately 35% of the patients treated by physicians in our study were covered by Network Health Plan, which supplied the pharmacy claims data. If similar savings were realized in other patients, the total prescription savings in the first year of use would be approximately $2500 per physician per month.

Finally, the savings reported from our trial are limited to prescription costs, and do not include other potential benefits that could be achieved through the use of CDSS (e.g., prevention of medical errors, avoidance of drug-drug interactions, improvements in quality of care, or enhanced physician office efficiencies related to the automated refill process and reduced pharmacy call backs). It is possible that the savings in drug expenditures only represent a small portion of the overall benefits to patients and the health system. Unfortunately, we only had access to pharmacy claims data and could not assess these other important outcome measures. Nevertheless, it is interesting that the annual savings from medications alone appear to be more than ten times higher than what is needed to cover the cost of this system.

Competing interests:   Dr. McMullin is employed by WELLINX, a company that markets an electronic prescribing system with integrated evidence-based decision support