Use of a CPOE system during patient care provides a unique opportunity to interject decision support features that improve clinical workflows, provide focused relevant educational materials, and influence how care is delivered to patients. It is somewhat of an art to be able to provide clinical decision support that is well accepted and used widely. Key considerations in the approach to provide decision support include: what content to provide (see below, Section 4); when to intervene in the clinical work-flow process (see below, Section 5); and, how to intervene, in terms of both degree of disruption of workflows and mechanism of interruption (see below, Section 6).
The nature of the clinical application domain determines what types of decision support content to provide. It is not appropriate to allow a clinician to spend 1–2 min constructing an intricate medication order only to later inform the clinician that the medication is contra-indicated due to a previously documented patient allergy. Allergy warnings should take place at the time the clinician first indicates the name of a new medication to be prescribed. Conversely, delivering a warning to a clinician to order a partial thromboplastin time (PTT) monitoring test—immediately as the clinician completes a heparin order—may cause both exasperation and a lost sense of autonomy when that is exactly what the clinician intends to order next. Checking whether a PTT monitoring test has been ordered at the end of an order entry session during which intravenous heparin therapy was initiated may be more appropriate, since the user is “done” entering orders at that stage. Oppenheim et al. [61
] observed that permitting the physician to enter an order with feedback provided only at the conclusion of order construction, and then only if the order is possibly incorrect, serves dual purposes. First, delayed warnings make clinicians first commit to a preferred course of action, thus discouraging reliance on CPOE systems to make clinical decision for the users. Second, delayed warnings give the clinician-user the opportunity to correct problems they detect spontaneously, whereas early warnings may impart negative reinforcement by underscoring clinicians’ errors.
In the authors’ experience, busy clinical users value CPOE system responsiveness and intuitiveness. A key aspect of responsiveness involves creating orders at an appropriate clinical level (both for users’ levels of training and for their knowledge of their patients). The physicians and nurses entering orders into a CPOE system typically have a different mindset than individuals who will carry out the orders in ancillary areas (e.g., Pharmacy, Radiology, and Dietary Departments). Problems in creating CPOE system “orderable item” names can occur when the technical terms used in ancillary departments are carried forward as the orderable items vocabulary for clinicians. So while radiology technicians might think in terms of “chest X-ray 2 views” and “knee X-ray 3 views,” clinicians are more comfortable ordering “chest X-ray PA and lateral,” and “knee X-ray AP, lateral and oblique.” Similarly, if the CPOE system asks the physician ordering a chest X-ray how the patient should be transported to the Radiology Department, the physician is unlikely to give an optimal response because physicians are rarely involved in determining a patient’s transport. Thus, CPOE systems should not ask clinicians to perform tasks that fall outside of their job responsibilities, or about which they have little knowledge. Structuring orderable items with the clinician in mind helps to overcome major barriers to adoption.
“Intelligent middleware” system interfaces can dramatically decrease the burden of ancillary departments in dealing with CPOE-system-generated orders. For example, pharmacists use the pharmacy system to fill and dispense the “clinical orderables” specified within the CPOE system; if a high-level order is issued by the physician, it may require more work on the pharmacy system side of the interface to specify all components of an order correctly. Once the physician specifies a medication order at a clinical level, an “intelligent interface” within the CPOE system can evaluate both the pharmacy’s formulary and the floor stock inventory on patient’s unit, and then automatically determine the correct “dispensable” within the pharmacy system. Currently, the “intelligent pharmacy interface” within WizOrder “guesses” the correct pharmacy-level dispensable item over 90% of the time. This allows the pharmacist to devote more time to evaluating each order’s clinical validity, safety, and efficacy.
As a frequently used clinician data entry tool, an institution’s CPOE system becomes a target for administrators and researchers wishing to capture additional data “at the point of care.” It is important to avoid overburdening clinicians with requests that interrupt their workflows, and when extra information is required, the system should only ask clinicians for information about which they are the definitive source. For example, at Vanderbilt, upon patient admission, the attending physician of record was originally input into the ADT system by an admitting clerk. However, the admitting clerks were not always informed of the specifics of physician group coverage schedules, and often did not know the correct name to enter. The problem was addressed by finding a more definitive data source—the admitting house staff team, who must discuss each admission with the attending physician—and having them enter the name into the CPOE system. Conversely, if one wants to record whether a patient received aspirin in the Emergency Department just prior to admission, asking an intern who is entering discharge orders for the patient several days later (and who did not admit the patient) could be viewed as a nuisance, and cause lower-than-optimal data quality.
While some decision support functions not directly related to order entry can be delivered during an order entry session, they will not be discussed in this article. For example, a laboratory system that generates alerts whenever abnormal patient results occur might notify clinicians responsible for the patient’s care by paging them, via e-mail, or via an asynchronous “pop-up” alarm that occurs when the clinician is currently “logged into” the CPOE application [62
]. Such alerts originate outside of the CPOE session context. Many CPOE systems, including WizOrder, display permanent “taskbars,” an array of useful links, continuously during the application session [45
]; however, such taskbars rarely provide context-specific decision support of the sort described here—they instead allow the user to access common CPOE functions. For instance, the BICS (Brigham Integrated Computer System, in Boston) toolbar allows the clinician to quickly view orders and search for patients, among other functions [64