Decreasing Prescribing Errors in Antimicrobial Stewardship Program-Restricted Medications Free

This study was conducted at an academic, metropolitan, tertiary care children’s hospital. Patients admitted to the PICU (26 beds) and medical/surgical floors (112 beds) were included. Services from pediatric hospital medicine, pediatric surgery, gastroenterology/hepatology, oncology/hematology, cardiology, nephrology, neurology, endocrinology, and adolescent medicine comprised the medical/surgical floor teams. Service teams were led by attending physicians and ran by house staff (residents or physician assistants). House staff were responsible for ordering medications and obtaining approvals.

The existing ASP consisted of a pediatric infectious diseases physician and clinical pharmacist. An alert in the electronic health record (EHR), EPIC, notified the prescribing provider to obtain approval of select antimicrobial agents through a prior approval stewardship model. Before our study, there was no standard way of obtaining approval. Providers contacted ASP in various ways and did not regularly communicate with the pharmacy. If approved, the ASP team provided documentation indicating the medication name, dosing instructions, and duration. Then, pharmacy verified the order accuracy against the ASP approval note before dispensing it.

A 12-month baseline period (January 2020–December 2020) was followed by a 14-month intervention period (January 2021–February 2022). During the 3-month sustainability period (March 2022–May 2022), active interventions were discontinued while data collection continued.

Inclusion criteria were orders for a course of ASP-restricted medications for patients <22 years old. ASP-restricted medications were: Acyclovir, amikacin, aztreonam, cefepime, ciprofloxacin, daptomycin, levofloxacin, linezolid, meropenem, vancomycin, valacyclovir, and valganciclovir.

Exceptions to requiring ASP approval were:

1. order placed during ASP off hours (11 pm–8 am); and

2. patients presenting with severe/critical disease (eg, sepsis pathway activation, code).

For exceptions, pharmacy dispensed 1 to 2 doses to grant providers sufficient time to obtain ASP approval without delaying timely administration. Orders verified under these exceptions were excluded because of workflow differences.

We formed an interprofessional QI team, including stakeholders from PICU, pediatric hospital medicine, gastroenterology/hepatology, hematology/oncology, ASP, nursing, house staff, and pharmacy. We developed an ideal process map for ordering, approving, and dispensing ASP-restricted medications (Supplemental Fig 5) and identified key drivers (Fig 1). A Pareto chart of all medication errors identified in the baseline period confirmed prescribing errors were the most common error type (Supplemental Fig 6). Therefore, our interventions targeted prescribing errors. Prescribing errors were defined as: Wrong start time/absent order and wrong dose/frequency. The most prevalent error was wrong start time/absent order. A wrong start time/absent order error occurred if the order had an incorrect start time and/or missing order when an ASP approval note or history of a previous dose indicated a dose was due (data collection process described below).

The QI team met monthly and used the Model for Improvement and Plan–Do–Study–Act cycles to develop and test interventions.²⁵  Interventions focused on achieving our key drivers: Adequate and timely interprofessional team communication, protocols for standardized workflow, and effective EHR safety nets (Fig 1).

Our main educational sessions for house staff occurred at the intervention period onset and were repeated at intern orientation (6 months later). These sessions introduced the ASP team, defined roles to share responsibilities, and reviewed the ideal communication workflow (Supplemental Fig 5). Posters describing this workflow were displayed.

One of our major interventions was the creation and implementation of a standard approach to daily medication review using the mnemonic IMED: Indication, medication name, end date, and dose. We taught the meaning and utility of IMED through house staff conferences, monthly PICU orientations, weekly secure chat reminders to inpatient residents, computer station signage, and badge buddy cards. IMED was incorporated into interdisciplinary teaching rounds to formalize and organize daily orders review. The intern verbalized the IMED components for all ordered antimicrobial agents during management plan presentation. Another resident verified order accuracy. House staff also used IMED during formal EHR medication reconciliation for patient transfers between units. In August 2022, we performed deliberate, 1-on-1 practice with new interns. They received targeted feedback after practicing placing orders, requesting ASP-restricted medication approval, and using IMED.

Because missed doses from absent orders were common prescribing errors, we developed a pharmacy standard operating procedure (SOP) to share ordering responsibilities with pharmacy. This SOP allowed pharmacy to pend a second order for a standing antimicrobial when an order was placed under an ASP exception. Pended orders remained in the order verification queue until ASP approval was obtained as a reminder to pharmacists and prescribing providers. Pharmacists had this ability before our study, but it was not universally known or used by pharmacists or house staff. The SOP formalized the process, decreased confusion among pharmacists and house staff, and reduced the risk of accidentally discontinuing medications placed during ASP exceptions.

Requesting ASP approval became an official, EHR-based consult. The consult order was linked to all ASP-restricted medications and was automatically placed upon medication order entry. A best practice advisory linked to ASP-restricted medication orders was modified to notify the prescriber to obtain ASP approval and methods for contact. House staff were trained to include specific consult information: Patient name, location, brief medical history, antibiotic requested, indication, current status, and disposition. The onus for contacting ASP fell on the prescribing provider. This was especially useful for overnight orders if the subsequent day team forgot to request approval. As backup, the consult order generated an active patient list for the ASP team to monitor.

During monthly stakeholder meetings in the active intervention period, QI champions received feedback using run charts. Data were displayed overall and stratified by unit and service team. High and low performers shared ideas for successful interventions and lessons learned.

The pharmacy team provided a monthly report of all ASP-restricted medication orders. This report included: Patient name, sex, date of birth, medical record number, unique order number, medication name, order date/time, and location. Author K.T. evaluated all orders and removed any meeting exclusion criteria. Every month, 75 eligible orders were randomly selected for inclusion through a Web-based random number generator. Additional variables were collected through manual chart review: Race/ethnicity, primary service team, admission date, medication dose/frequency/route, order-designated start time, medication verification, dispense and administration date/time, ASP approval note date/time, infectious diseases consult, and prescribing error presence. To pilot this process, K.T. performed detailed chart review to evaluate and understand order documentation. Findings were reviewed and clarified with pharmacy, nursing, and house staff champions. An accurate and reliable data extraction plan was developed with senior author (K.P.). This process continued until no discrepancies were found. For consistency, author K.T. completed all chart reviews.

The primary outcome was the proportion of orders with a prescribing error (orders with prescribing error/total orders reviewed). A prescribing error was defined as incorrect dose, frequency, or start time of the index order based on ASP note, or absent order when the medication was due. An error in start time occurred if the order was placed >60 minutes after a dose was due, which was calculated using the last dose administered and frequency. Sixty minutes was chosen on the basis of our hospital’s nursing protocol for timely medication delivery. The secondary outcome was the time in days between orders with a prescribing error, which was evaluated on the same 75 charts per month.

The main process measure was proportion of weekdays IMED was performed during rounds by the medical/surgical floor teams as reported by senior residents (sum of weekdays IMED performed for patients on antibiotics per sum of weekdays each team rounded). Reports were obtained 1 to 2 times per month through verbal surveys of senior residents at the end of each week. This measure evaluated the uptake of the project’s main intervention influencing behavior change. Responses varied each month depending on the QI team’s ability to administer the in-person, verbal survey and each team’s senior resident availability.

The balancing measure was time to administration of the first medication dose, defined as the absolute value of minutes from ordered start time to administration time. We were concerned the interventions may increase the time for approval and inadvertently prolong the medication delivery process.

We reviewed 75 random orders per month throughout the study. This was chosen as the minimum subgroup size necessary to have a lower control limit (LCL) >0 on the basis of preliminary data revealing an error rate of 10%.²⁶  During the sustainability period, 35 orders were reviewed per month.

Primary and secondary outcomes and the balancing measure were plotted on statistical process control charts using QI Charts 2.0.23 (Process Improvement Products, Austin, TX) for Microsoft Excel 2010 (Microsoft Corporation, Redmond, WA). The primary outcome of proportion of orders with a prescribing error was recorded on a P chart. The time between orders with a prescribing error was recorded on a T chart. The balancing measure was recorded on an X bar S chart. Control charts were analyzed for special cause variation.^25,26  The process measure was plotted on a run chart because there was no baseline data for comparison.

Patient demographics and order characteristics were compared between baseline and intervention periods using χ² or Mann Whitney U tests as appropriate. Data were analyzed using Stata software, version 16.1 (StataCorp, College Station, TX). The study was approved by our University’s institutional review board.

There were 1935 ASP-restricted medication orders included in the study: 900 in the baseline and 1035 in the intervention period. An additional 105 orders were included in the sustainability period. Only 60 orders were eligible for inclusion in February 2021.

Table 1 compares patient demographics and order characteristics between periods. There were no significant differences in patient sex, age, and race/ethnicity. There were significant differences in primary team and type of antimicrobial ordered, with more primary gastroenterology/hepatology patients and more orders for vancomycin and meropenem in the baseline period. The baseline period had more orders with a pediatric infectious diseases consult (56.3% vs 49.9%, P < .01).

The mean monthly rate of prescribing errors decreased from 10.9% in the baseline period to 4.6% in the intervention period, exceeding our goal of a 25% decrease. Special cause variation occurred and the center line shifted in February 2021 after 8 points fell below the baseline centerline (Fig 2A). The shifted centerline was extended through the sustainability period and the improvement was sustained.

In the baseline period, the mean time between prescribing errors was 2.9 days, increasing to 8.5 days after interventions. In May 2021, special cause variation occurred with >8 points above the center line and 1 point outside the upper control limit (Fig 2B).

As a process measure, the median proportion of times the team performed IMED on weekday rounds was 49.2% (Fig 3). Because this intervention was developed during this QI initiative, there was no baseline comparison.

For the balancing measure, the baseline mean monthly time to administration was 53.9 minutes (Fig 4A). No special cause variation resulting in a centerline shift was observed. The S chart has a baseline SD of 87.7 minutes and demonstrated an unstable baseline process with many months plotting outside of the upper and LCLs (Fig 4B). During the intervention period, the stability of the process improved with more consistent monthly SDs.

By implementing communication and workflow improvements for ASP-restricted medications, we successfully decreased the proportion of ASP-restricted medication orders with a prescribing error from 10.9% to 4.6%. The mean time between errors increased from 2.9 to 8.5 days and improvements were sustained. There was no special cause variation leading to a centerline shift in time to administration, so inadvertent delays in medication delivery as a result of our improvement efforts were unlikely. QI methods can be used to achieve a sustained reduction in the proportion of ASP-restricted medication prescribing errors throughout a children’s hospital.

ASPs are successful in reducing antibiotic overuse, decreasing dosing errors, and improving patient outcomes.^{7,14,21,23,27–31}  However, incorporating ASPs into the ordering process adds complexity because common tools used by ASPs have unintended consequences and increase the prescribing error risk.¹⁴  We found an ASP prior approval process increases the prescribing error risk through the disruption of standard workflows. Our study expands upon QI literature on medication error reduction by specifically addressing medication errors associated with ASP workflows.^6,32–35 

We aimed to reduce ASP-restricted medication prescribing errors by developing interventions focused on our key drivers. Our initial decline in errors occurred after our main educational sessions focused on optimizing communication between the prescribing providers, ASP, and pharmacy, disseminating standardized workflows, and defining roles. IMED was developed as a standard approach to daily medication review to mirror high-quality medication reconciliation performed in care transitions for order accuracy.^36–39  We studied the utilization of IMED as our main process measure to evaluate changes in behavior over time. House staff reported using IMED a median of 50% of weekday rounds, which was expected given time restraints and staff availability. Toward the end of the study period, reported use of IMED increased, which likely helped drive our reduction in prescribing errors.⁴⁰  Although educational interventions are helpful in developing cultural change, the sustainability of the effects tend to fade with staff changes and passage of time.^36,41,42 

To address sustainability, we developed high-reliability interventions of EHR-based ASP consult order, best practice advisory, and pharmacy SOP.^43,44  The safety nets and standardized protocols helped facilitate a shared responsibility between primary, pharmacy, and ASP teams to prevent prescribing errors.⁶  Although we were unable to directly measure the impact of these interventions, we attribute some of our sustainability to the EHR changes and pharmacy SOP because they were high-reliability interventions compared with education.^{19,41,42,45–47} 

Comparing baseline and intervention periods, the types of ASP-restricted medications prescribed and the primary admitting team differed significantly. However, because the ordering process for ASP-restricted medications is the same throughout the hospital and the prescribing providers consist of house staff who rotate through all primary services, these differences likely were not the main drivers in decreasing prescribing errors. Furthermore, our study may have inadvertently impacted the types of antimicrobial agents ordered, resulting in the difference between study periods; with improved communication, the ASP team may have felt more empowered to make recommendations, and awareness of prescribing errors may have influenced antibiotic choice.^17,48  Additionally, differences in the baseline and intervention periods may reflect the impact of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic, where in the early stages, there was higher antibiotic usage and infectious diseases consults.^49,50  During the initial onset of the pandemic, our ASP team had to navigate the added stress of regulating SARS-CoV-2 therapeutics with limited resources and manpower. The SARS-CoV-2 pandemic surges may also be responsible for variability in time to administration in the baseline period and astronomical data point in January 2021. During surges, nurses would batch their cares for personal protective equipment preservation.

There are several limitations to our study. Mainly, our primary outcome focuses on reduction of errors but cannot be directly extrapolated to the impact on patient outcomes. Although serious injury and death from medication errors are rare, medication errors are known to increase patient morbidity and mortality, and should be minimized.^1,4  Additionally, the generalizability of our results is limited to academic hospitals with inpatient ASPs who use prior approval for a large proportion of antibiotics. As with any QI study, this is subjected to observer effect, because providers may have changed their practice while the study was ongoing and the process measure was limited by recall bias. The LCL of our primary outcome was 0.09% in the baseline period and dropped to 0 after the centerline shift because of restrictions from the small proportion of medications with a prescribing error and our goal of a 25% error reduction. Although we chose our sample size on the basis of the minimum subgroup size for an effective P chart with LCL >0,²⁶  the addition of more charts to raise the LCL was limited by the capacity of the single reviewer to manually review charts in a timely manner. Finally, on the basis of these chart review limitations, the results are a collective sample, representing the sample error rate rather than the true error rate, and our secondary measure is a sample time between events. Therefore, to minimize bias in our sampling, we used a random number generator to select included orders.

Our interprofessional team used QI methodology to achieve a 58% improvement in the proportion of ASP-restricted medication orders with prescribing errors by standardizing the process of reliable ordering. Shared responsibility between the prescribing provider, pharmacy, and ASP team was key to ensure appropriate and timely medication ordering and a sustained decrease in prescribing errors. Future studies should focus on reducing other causes of error, tracking of the sustainability of our efforts, and measuring the impact of our interventions on patient outcomes directly.

We thank the house staff, pharmacists, nurses, physician assistants, and attending physicians who participated in this project.