Invited commentary

Computerized physician order entry (CPOE) has been found to reduce medication error rates, although most of the published data come from a relatively small number of centers (1–3). In this issue of BUMC Proceedings, Seeley et al report an evaluation of their medication error rate before the implementation of CPOE, which brings up a number of important issues.

One is how best to measure medication safety. The answer varies, depending upon the resources available and the intent (4). In this instance, Seeley et al elected to use medication prescribing errors intercepted by pharmacists. This is an extremely useful approach, which has been used by Lesar to assess many important problems (5), and was probably the best one in this instance, because it is inexpensive and has a high yield of errors that can be prevented by using CPOE. But depending upon the circumstances, other methods may be better; for example, Barker et al have demonstrated that the direct observation approach is reliable and valid for administration errors (6), and before implementing an intervention to improve administration safety, this would represent a better choice.

Both these approaches have an important shortcoming in that they do not measure the frequency of harm caused by these errors. Other approaches are better for this: chart review has been the standard, although it is too expensive for routine use. Computerized adverse drug event monitoring, in which monitors look for signals suggesting that an adverse drug event has occurred, are probably the main way that adverse drug events will be detected in the future (2).

While it is widely believed that reducing medication error rates will improve safety, the extent to which this is true is untested, and it is clear that some errors have much greater potential for harm than others. For example, an order for a hundredfold overdose may actually be less likely to cause harm than a fivefold overdose, because the former would almost always be intercepted even without CPOE, while the latter might not.

The evidence on CPOE clearly demonstrates that it reduces the overall medication error rate and the serious medication error rate—serious medication errors are those that either result in harm or have the potential to do so (1, 2). While some have expressed concern about whether CPOE reduces the preventable adverse drug event rate, in fact, a study with adequate power to ad-dress this has not been performed, and it would be very expensive to do one because of the infrequency of the outcomes. The Adverse Drug Event Prevention Study used the serious medication error as the outcome, and it cost >$1 million. No such study is planned, but in the interim, organizations need to make decisions about what safety interventions to implement.

Another issue is whether it is necessary to measure the medi—cation error rate before implementing CPOE. Given the evidence to date that CPOE improves medication safety, my opinion is that it is not essential, although there are a number of reasons that it may be very useful to do so. Perhaps the most important is that much of the benefit of CPOE depends on the associated decision support, and assessing the types of medication errors can help with prioritization of implementation of checks, as well as providing a check to see to what extent important checks are in place. In the current study, as is usually the case (1, 2), dosing problems were the most frequent type of error. Implementation of default dos-ing—in which the computer suggests an initial dose—and renal dose checking (7) probably represent two of the most important types of safety checks. Another reason to check the medication error rate is to show that the problem exists at that particular institution; although it always does, leadership is often not aware of the magnitude of the problem.

Finally, there are many types of errors in the medication use process and a wide array of strategies for improving medication safety, so CPOE is only one piece of the puzzle. Among the specific interventions that organizations will need to consider are barcod—ing, “smart” pumps (pumps that can be told what medications are being given and warn the nurse if they attempt to enter too high a dose), and dispensing robots, as well as nontechnological strategies such as having a pharmacist round with teams in the intensive care unit (8). While this may stretch hospitals' budgets, ultimately a combination of strategies will almost certainly be needed to improve medication safety by two or more orders of magnitude.