A trial of automated safety alerts for inpatient digoxin use with computerized physician order entry

This article attempts to evaluate the utility of the introduction of a clinical decision support(CDS) algorithm, designed to “identify potential medication errors associated with digoxin”, to the CPOE component of the hospital EMR in a university teaching facility. The study design is a retrospective comparison of two patient cohorts each of 310 patients, identified by a computer search of the EMR and a manual review of timed clinician responses, which was the primary metric of interest. One cohort consisted of patients treated in the six months prior to the implementation of the digoxin CDS rule and the second cohort consisted of patients treated in the six months after implementation. The cohorts were not significantly different when compared for age, gender, total number of alerts or alerting situations, or electrolyte concentrations generating the alerts. The number of alerts for critical levels of hypokalemia(<3.0meq/L) was similar in both groups and not considered in the analyses because of separate reporting mechanisms for critical values at this institution.

The decision rule evaluated presents a real-time information only popup if digoxin is ordered and: K+ is < 3.5meq/L or Mg2+ is < 1.8meq/L without an order for electrolyte replacement, there is no recent recorded digoxin level in patients on digoxin, there is a digoxin level > 2.2mg/dL in the past 30 days, or the patient is concurrently on medications known to increase digoxin levels(eg – amiodarone, quinidine). The rule parameters identified in the paper are shown in the following table:

Table 1 j Alerting Situations and Expected Actions

Alerting Situation Expected Clinician Action

K+ ,<3.5 mEq/L, patient Order for K+ supplementation receiving digoxin Mg2+ ,<1.8 mEq/L, patient Order for Mg2+ supplementation receiving digoxin No recent Mg2+ level, patient Order for an Mg2+ level receiving digoxin No recent K+ level, patient Order for a K+ level receiving digoxin No recent digoxin level, Order for a digoxin level patient receiving digoxin currently and in the past

These parameters were chosen because they occurred with enough frequency and are associated with specific expected responses to allow evaluation. The authors also differentiate between synchronous alerts, those occurring at the time of the ordering process and asynchronous alerts, occurring later in response to abnormal results. Rates of expected compliance with alerts at 1 and 24 hours were tabulated for the 5 synchronous alerts listed in the table plus 2 asynchronous alerts, K+ < 3.5meq/L or Mg++ < 1.8meq/L.

The results of the study showed that prompts that recent K+, Mg++ or digoxin levels were not documented on the EMR at the time digoxin was ordered(synchronous activity) resulted in significantly improved rates of expected physician action – ordering of the respective level, compared with pre CDS action. Additionally, the two evaluated asynchronous alerts, low K+ or Mg++ levels generated on patients already on digoxin, also resulted in improved levels of physician compliance with expected actions, in this case the addition of K+ or Mg++ supplementation to the treatment regimen. Comparatively, synchronous notification of low electrolyte levels at the time of the initial digoxin order had no identifiable affect on the ordering of electrolyte supplementation in the two cohorts. The specific results are given as follows:

                             Compliance  1 hour                 Compliance  24 hours
                         Control %         Study %           Control %           Study %

Alert Synchronous No Dig Level 6 19 22 38 No K+ Level 9 57 49 81 No Mg++ Level 12 40 44 66 Low K+ Level 6 23 47 56(NS) Low Mg++ Level 22 39(NS) 74 65(NS) Asynchronous Low K+ Level 5 49 70 87 Low Mg++ Level 6 35 77 93 NS = Not Significant

The author’s conclude that the results of this study support that CDS alerts improve the speed and magnitude of clinician responses. They cite the collective sevenfold increase in orders for electrolyte supplementation at one hour as a specific example. They also emphasize the need to present clinicians with all relevant information on the same visual interface used for medication ordering. They attribute the difference in response to synchronous and asynchronous reporting of low electrolyte levels to the nature of house-staff interaction with nursing staff and the mechanism by which the results are presented to the physician.

Comment It seems to me that the data can be looked at as representing diagnostic versus therapeutic responses. Prompts to order a test were much more effective at generating a response than prompts to change therapy. While it is true that clinician response, at least to the asynchronous prompts, was statistically improved after CDS implementation the ultimate difference at 24 hours between the groups was relatively small – 70% vs 87% in the low K+ group and 77% vs 93% in the low Mg++ group. In the comparative synchronously prompted groups, there was no statistically significant change at all. Clinician behavior is inherently conservative. The first rule of medicine is to not make things worse. I suspect that the reason there was not a more noticeable effect on physician ordering is because, since critical levels were eliminated from consideration, there was little sense of urgency to respond rapidly or, at all, to prompts, it being perceived to be much easier and perhaps much safer to just order another test. As noted previously, this particular CDS rule was presented as information only and did not require a response or provide for an automatic order for the expected action. My conclusion is that if the designers of rules for therapeutic interventions want to improve compliance, this would best be obtained by requiring physician action on the rule at time of presentation