The Future CPOE Workflow: Augmenting Clinical Decision Support With Pharmacist Expertise

The advent of computerized provider order entry (CPOE) brought substantial changes to the medication order processing workflow. Much of the justification for these changes included improvements in medication safety. CPOE for medications has reduced errors contributing to improvements in patient safety.^1-5 The benefits of CPOE include a reduction in prescribing and transcription errors and the ability to manage medication-related problems in real time through alerts to the user.³ Benefits to patient care have been noted with many alert types including the following: drug-allergy, drug-drug interaction, drug-dose, drug-duplication, and formulary compliance alerts.⁶

The many different categories of alerts generated by computer decision support are designed to cast a wide net to catch potential medication-related problems.⁶ Computer decision support may be thought of as a nuisance, until it catches a medication-related problem important for the care of the patient. How many, and what type, of alerts to provide can be viewed from different perspectives. First, there is the risk management and hospital administration perspective. Risk managers and health care administrators would like to alert health care providers to as many risks as possible to protect patients. In this way, there is a “perception” that all alerts have clinical utility. Clinicians would like to review ONLY alerts that have “actual” clinical utility at the time they are making a decision (ie, initiating a medication order, changing a medication order, etc.). Many clinicians find the signal-to-noise relationship generated by the alerts overwhelming. Health systems must often make decisions on whether an alert is to be interruptive, noninterruptive, or not to fire at all.^6,7 Due to considerations of risk, some alerts may be designated to be interruptive simply to ensure that issues are addressed before an order is processed. Unfortunately, the volume of alerts may de-sensitize providers to their potential impact resulting in “alert fatigue” and automatic overriding of alerts.^8-14 Optimization of the alerts presented in the medication order entry workflow has been the subject of considerable discussion, activity, and publication.¹⁵ Specifically, there has been much work to decrease alert fatigue and provide more relevant and contextual information to the provider during the CPOE process.^16-19

Pharmacists are well equipped to manage medication-related problems.^2,20 This ability combined with computer system experience places pharmacists in an optimal position to augment computer-driven decision support during medication order entry. To accomplish safer and more efficacious drug therapy during CPOE, we propose a longitudinal context and priority structure to provide significant advantages over current medication workflows.

Efforts to integrate pharmacists more prominently in the automated decision support alert workflow in a seamless, longitudinal strategy can add incremental benefit to work being considered on the generation and contextual nature of alerts. Alert data can be accurate, but not useful. It becomes useful when complete information is available in the correct context and at the appropriate time. For example, it is virtually impossible to establish whether some drug interactions are clinically significant without monitoring them over time. They may have a delayed onset or an unpredictable association with dose. Some medication-related problems are not apparent until laboratory data or microbiology results are available. This is not at the time of initial ordering. In addition, consider the following scenario: a patient initiated on a potassium supplement who is receiving an angiotensin-converting enzyme (ACE) inhibitor prompts an alert. The alert has far more meaning if the context of the alert notes the patient has a serum potassium level which is above the therapeutic range. If the serum potassium level is within the normal range or below the normal laboratory range, the alert should not generate at the time of CPOE.

If an appropriate and reliable alert workflow were established, much of the need to address alerts and medication information related to medication-related problems can be accomplished by pharmacists. As discussed above, much effort has been placed on improving drug alerts, their specificity and information provided with the alerts, but the same effort has not been directed toward how pharmacists can assist with drug alerts. Pharmacists can support alert management based on the clinical condition of the patient and when new data become available. This workflow can have a profound impact on the number of alerts presented to the provider at the time of initiation, without compromising the assurance that specific issues would be addressed. This is accomplished through the suppression of some alerts during the prescribing action when it is not possible to determine the clinical validity of the alert, or there is no acuity to the alert (ie, alerting a physician to a potential drug interaction with the instructions to “monitor” is not a useful alert at the time of prescribing). In order to ensure the appropriate attention to the alerts, these types of alerts will be sent to a pharmacist for surveillance. Depending on the nature of the alert, this may occur prior to implementation of the order, or in the case of delayed events, this will be added to a monitoring queue for the pharmacist to follow as additional information is available. For example, a drug interaction with a moderate severity and a delayed onset may be “pushed” into a surveillance queue, to be evaluated as evidence of the drug interaction can be assessed. This type of longitudinal approach makes more sense and relieves providers of responding to irrelevant alerts.

The following discussion further illustrates a structure for pharmacists to enhance CPOE for medications. The higher the acuity of an alert will determine if a provider should receive the alert upon initial CPOE. Directing the alert to the provider is based on the effect that the medication will have on the patient in a short period of time. Quick responses will be necessary for most severe, immediate issues, but longer term monitoring is essential in many cases. In Figure 1, the traditional decision support workflow is seen. It relies heavily on the provider to manage the clinical decision support workflow. Medication-related problems may be missed. In Figure 2, a more complete decision support workflow is provided. As in the traditional workflow, if the medication does not generate an alert, it will pass to the pharmacist for verification. If an alert is generated, the alert will require a response. The provider will need to respond immediately to drug allergy alerts that indicate an immunological basis (ie, penicillin allergy in which the patient suffered anaphylaxis, ACE inhibitor allergy where the patient exhibited angioedema), drug-drug interactions where the patient may suffer a fatal or significant clinical consequence quickly after the interacting drugs have been initiated (ie, meperidine and selegiline or enoxaparin and rivaroxaban), dose alerts for when single or daily doses are prescribed significantly over recommended doses or daily doses (ie, enoxaparin for deep vein thrombosis treatment prescribed great than 1.5 mg/kg SQ q12h), drug condition alerts (ie, prescribing metoprolol for a patient with second degree atrioventricular block [AV] block), and formulary management needs.

Figure 1.

Computer decision support workflow.

Note. CPOE = computerized provider order entry.

^aHigh acuity alerts-drug alerts-high probability for adverse effect, drug-allergy, drug-drug interactions-severity high, immediate, dose checking, drug condition, formulary management.

Figure 2.

Computer decision support pharmacist-augmented alert workflow.

^aHigh acuity alerts-drug alerts-high probability for adverse effect, drug-allergy, drug-drug interactions-severity high, immediate, dose checking, drug condition, formulary management.

^bDrug-allergy, drug-drug interactions: severity-medium,delayed onset, drug-drug duplicates.

^cDrug-drug interactions: delayed onset, drug-bug, drug-laboratory, intravenous to oral.

Along with review and approval of provider actions and resolution of incomplete orders, the pharmacist will be required to address some alerts immediately. Alerts that will be directed to the pharmacist include drug-drug interaction alerts that are not of immediate consequence (ie, HMG CoA reductase inhibitor and niacin), drug allergy alerts that are not immunologically based but may have consequence for the patient (ie, patient prescribed ceftriaxone that had a rash to penicillin) or allergies that are documented as intolerances or complete reaction information is not provided, and alerts that are triggered from drug-class or drug-drug duplicates (ie, patient is receiving amlodipine 5 mg and the provider increases the amlodipine to 10 mg but does not discontinue the amlodipine 5 mg order). Orders needing clarification from the provider can be triaged for provider approval.

Finally, ongoing review of medication-related problem alerts will be tracked by the pharmacist over time as data are available. These alerts will encourage the pharmacist to intervene on the patient’s medication therapy in real time. Examples of these surveillance alerts include drug-drug interactions that have a delayed effect, new microbiology data that support a different antimicrobial therapy or a de-escalation of therapy, laboratory values that display out of range (ie, supra-therapeutic glucose laboratory values, supra-therapeutic International Normalized Ratio [INR] values for a patient on warfarin, supra-therapeutic activated partial thromboplastin time [aPTT] values for a patient on unfractionated heparin), laboratory values that support adjustment of medications based on hepatic or renal function, and potential intravenous to oral recommendations. Based on data received, the pharmacist can interact with the provider to recommend appropriate medications and/or protocols can be approved through medical staff committees to support pharmacist-led medication modifications. This pharmacist surveillance phase (immediate attention and longitudinal attention alerts) enhances the patient experience by ensuring medication-related problems are addressed more completely.

The question is, how can systems be designed to support this initiative? The current alert strategy is complicated by a lack of integration in key systems. Many electronic medical record (EMR) systems, and some data vendors, have developed their systems to generate alerts and provide mechanisms to modify the level of alert triggers or even the classification of the data. Pharmacy-directed clinical surveillance programs permit the development of “rules,” which identify the situations where pharmacist-initiated monitoring will occur, and facilitate and automate collection of data. Unfortunately, the EMR systems do not communicate with the surveillance programs. In the future, it is recommended that EMR systems develop and integrate surveillance rules and algorithms, either with current Pharmacy-oriented software or through de novo development of these capabilities. This enhancement will make this type of integration and stratification a seamless part of decision support workflow.

CPOE has improved tremendously over the last 15 years and enhanced the patient care experience by improving safety and efficacy of medication therapy. Much effort has been expended to improve automated clinical decision support and the medication alerts that are generated. In addition to these efforts, the pharmacist-augmented workflow described above provides the opportunity to continue to promote effective alerting in CPOE and augment the value of clinical decision support by expanding the process to include the pharmacist: the medication expert.