Frequency, Timing, and Types of Medication Ordering Errors Made by Residents in the Electronic Medical Records Era

Ari Garber, MD, EdD; Amy S. Nowacki, PhD; Alexander Chaitoff, MPH; Andrei Brateanu, MD; Colleen Y. Colbert, PhD; Seth R. Bauer, PharmD; Zubin Arora, MD; Ali Mehdi, MD; Simon Lam, PharmD; Abby Spencer, MD; Michael B. Rothberg, MD, MPH

Disclosures

South Med J. 2019;112(1):25-31. 

In This Article

Methods

We conducted a retrospective cohort study of all inpatient medication orders placed by internal medicine residents (postgraduate year [PGY] 1–3) between July 2011 and June 2015 at the Cleveland Clinic main campus, a not-for-profit multispecialty academic medical center located in Cleveland, Ohio. During this period, no specific quality improvement initiatives were aimed at decreasing order errors.

The Cleveland Clinic has used an EMR system (Epic Systems) since 2001. This EMR has some error detection functionality, such as the ability to warn users before filing orders for duplicate therapies or medications with allergy contraindications. Studies have suggested, however, that up to 96% of EMR alerts are ignored, which may result in adverse events.[21,22] To better identify potentially harmful medication order errors, pharmacists at Cleveland Clinic also review each order placed. Using the imbedded pharmacy interventions documentation and communication module (iVent, Epic Systems), pharmacists manually verify all medication orders, directly communicate with the healthcare team about suspected errors, and document their productivity. The EMR can be queried for order information captured in iVent communications, including the medication order number, when it was placed, and whether there is a possible error. All iVent communications are categorized. The majority of Cleveland Clinic's 33 predefined iVent communication categories are associated with interpharmacy productivity documentation, but 5 are associated exclusively with suspected medication order errors. After consultation with our interdisciplinary team, which includes pharmacists, internal medicine physicians and trainees, a biostatistician, and a PhD educator, we defined a medication order error as an order that was associated with one of these five categories: allergy detection/caution, drug interaction, duplicate therapy, order needing clarification, and renal dose monitoring/adjustment required. These categories remained constant during the study period, and detailed descriptions of the categories and examples of order errors appear in Supplemental Digital Content, Appendix Table 1 (http://links.lww.com/SMJ/A127).

The EMR was queried for the following elements: medication order number, date and time of order, resident entering the order, whether there was an associated iVent, and iVent category. The time of the medication order also was dichotomized into before or after the implementation of duty hour restrictions (January 1, 2015), which takes into account Cleveland Clinic's participation in the iCOMPARE duty hour trial.[23] This query was cross-referenced and combined with the files from the internal medicine residency program regarding provider name and training level (PGY 1, 2, or 3) at the time of the order. The institutional review board at Cleveland Clinic approved this research as a quality improvement study.

Statistical Analysis

Characteristics of order errors were summarized, with counts and percentages. The association among the order characteristics (training level, shift, month, therapeutic class) and the probability of having an error was investigated with multivariable regression modeling. Because there were multiple orders per provider, a generalized estimation equation model was fit to accommodate the order clustering within provider. A compound symmetry structure was assumed for this correlation, which means that each order of a provider is equally correlated with each other order from that provider.[24] Adjustment for year of order also was included to account for changes in the healthcare system. A significance level of 0.05 was assumed, and all of the analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....