An Initiative to Reduce Insulin-Related Adverse Drug Events in a Children’s Hospital

Our institution is a tertiary medical center with 670 inpatient beds and 1.2 million patient encounters yearly. The endocrinology service admits to a dedicated unit with nursing staff highly familiar with insulin therapy. Patients hospitalized on other medical and surgical services are admitted to specialty specific units throughout the hospital. There is variability in the degree of endocrine involvement on each respective hospital subspecialty unit. On the endocrine and surgery units, endocrinology acts as the primary service managing insulin. On all other units, including the off-site psychiatry unit, the endocrinology unit provides a consulting service. The hospital’s primary mode of insulin delivery is basal bolus therapy by using a combination of short and long acting insulins. Short acting insulin doses are calculated and administered for hyperglycemia correction and with meals. Long acting insulin is used for basal glucose control.¹² 

Our former method for insulin ordering and administration was a single-order process. When a patient needed insulin to cover food or hyperglycemia, the bedside nurse would contact the prescribing provider with the blood glucose reading and carbohydrate count. This occurred before every meal and 3 hours after a short acting insulin dose had been given. The provider then wrote a 1-time insulin order for each individual dose of short acting insulin. Bedside nursing acted on each individual insulin order. This occurred up to 5 times per day, per patient. Using lean and 6 σ principles, with process mapping, we identified opportunities to eliminate time waste, introduce automation and redundancy, and standardize processes to promote reliability. Implementation of a standing order with pro re nata (PRN) parameters, defined the process for appropriate insulin dose calculation and administration by bedside nursing. These parameters included insulin dosing specific to the actual blood glucose and carbohydrate intake for the patient at that moment, plus specific criteria for hyperglycemia correction within an established range of acceptable blood glucose values. This sought to reduce human errors, maximize nursing autonomy in patient care, and set up specific criteria for when verbal contact should occur between the order writer (provider) and order enactor (bedside nurse). Approval was obtained from the Institutional Review Board at Cincinnati Children’s Hospital Medical Center.

Standardization of insulin ordering, documentation, and administration were chosen as a key intervention, providing an avenue for reproducibility in a hospital setting.¹³  With automation, like standardization, we also promoted safety by embedding a “double check” on all insulin dose calculations in the EHR. Both standardization and automation are keys to ensuring consistent, stable practices with the use of high-risk medications. We adopted a new PRN insulin ordering and administration process to replace the former single-entry order process. The new process used a standing order to facilitate efficient and accurate calculations at the point of care (ie, at the bedside) and ensured appropriate insulin dosing was based on each individual blood glucose value. This approach required ordering providers to update or electronically acknowledge the PRN standing order a minimum of every 48 hours. Nursing staff would continue performing real-time insulin dose calculations, with second nurse verification. The standing PRN insulin order was applicable to all patients receiving short acting insulin, provided the following criteria were met: (1) patient specific blood glucose targets were clearly identified (ie, “insulin can be administered if patient’s blood glucose is within X to Z”), (2) it had been at least 3 hours from the last short acting insulin dose (to prevent insulin stacking), (3) the patient has negative or small ketones (checked no sooner than every 3 hours, reducing the risk for entrance into diabetic ketoacidosis), and (4) the hospital’s standard hypoglycemia correction plan was in place. If these criteria were not met, bedside nurses returned to the former single-entry process and were required to contact providers to create an individualized intervention plan (ie, obtain a 1-time insulin order or revised PRN order). As demonstrated in the key driver diagram, initial testing of this process was focused on communication between ordering providers and bedside nurses. Several adjustments were required before the process was fully functional. For instance, determining the best route of communication (secure texting versus phone call versus paging), best way to communicate timely needs (an insulin order required immediately for active mealtime versus insulin required for upcoming meals), and best mode of order updating in the EHR.

Education is a crucial intervention that ensures all individuals, regardless of inpatient unit assignment, undergo uniform education before practice change implementation. Uniform education is essential to hospital-wide safety, given the vast difference across hospital units in exposure to and experience with insulin ordering and administration. To ensure hospital staff (resident physicians, fellow physicians, attending physicians, nurse practitioners, and bedside nurses) had adequate knowledge of insulin use, we created a month-long education cycle. Internal electronic and paper educational materials were created and distributed. Educational materials were used to explain the central concepts of insulin safety, and we included real-life vignettes to encompass various insulin-related duties, objectives, and problems. Education was finalized when individuals completed online testing and passed with a 100% score. Real-time educational tools and videos were embedded within the EHR for all insulin order sets and documentation sheets. Videos were intentionally short in duration to accommodate “real-time learning,” averaging 2 to 5 minutes in length. These were intended to be available for staff to refresh their knowledge of insulin at any given point.

As a first step in spreading our safety work off the endocrine and diabetes floor and onto all hospital units, we strategically chose a unit with high volume insulin use (hematology oncology) and one with low volume insulin use (surgery). Using this approach, we hoped to gain insights on the use of a high-risk drug on units with various levels of exposure. Over a 6-month period, we carefully tracked insulin safety measures and engaged in scheduled face-to-face meetings with unit leadership and front-line personnel every 2 to 3 months. Hospital-wide education of all staff, involving didactic online modules and posttest education, rolled out from July 2016 to August 2016. Hospital-wide implementation of the new process then occurred in September 2016.

Our primary process measure on the new method of insulin delivery was monitoring the time from a blood glucose check to insulin administration. Limiting the time between a blood glucose check and insulin administration to ≤30 minutes has been identified as a safety priority in clinical settings.¹⁴  If effective education, efficiency, automation, and redundancy were achieved through the new process, a reduction in the time from blood glucose measurement to insulin administration would be seen. For the purposes of this work, we considered episodes of insulin-related events (hypoglycemia and hyperglycemia) as the balancing measure. We defined rapid-acting insulin-associated hypoglycemia as a blood glucose measurement of <65 mg/dL, collected within 3 hours after receipt of a rapid-acting insulin dose. This time frame was used to account for the onset and duration of rapid-acting insulin. We defined rapid-acting insulin-associated hyperglycemia as a blood glucose measurement of >250 mg/dL to assess overall glycemic control while on rapid-acting insulin. Rapid-acting insulins available during this work included insulin lispro and insulin aspart only. Insulin glulisine was not available at our institution during the relevant time period.

Insulin use and exposure across hospital units are variable. Therefore, we considered the total number of insulin days per unit per year to assess the level of experience by calculating the number of RAIDs per unit. For example, if 1 patient on insulin was admitted to a unit for 5 days or 5 patients on insulin were admitted to a unit for 1 day, the unit’s overall number of RAIDs would be 5 in both instances. We excluded data collection in the (1) emergency department because of the short average length of patient stay, resulting in rare use of injectable insulin, and (2) ICU because insulin drips are primarily used in this area.

We collected the timeliness of insulin administration and hyperglycemic and hypoglycemic rates on each unit of the hospital electronically through the EHR. This was done by normalizing hyperglycemic and hypoglycemic event rates by using the number of RAIDs we had generated for each respective unit. We then divided the number of hyperglycemic and hypoglycemic events by the number of RAIDs and multiplied by 100 to calculate the number of events per 100 RAIDs. This process allowed for a comparison during the baseline period and after the new PRN process was gradually implemented.

U-charts were created as part of the statistical process control analysis to track progress with hypo- and hyperglycemia rates by using standard rules for determining special cause variation. Data were analyzed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC). Because of the distribution of variables, continuous data were summarized as medians with interquartile ranges (IQRs) (25th to 75th percentile), whereas categorical data were summarized as frequency counts with percentages. We dichotomized RAIDs into 2 groups (<50 vs ≥50 RAIDs per year) because of the limited number of units. Nonparametric Mann–Whitney U tests were used to compare hyperglycemic and hypoglycemic events between groups as well as the timeliness of insulin administration. P < .05 was considered statistically significant for analyses.

The new PRN process was used more than twice as often as the former single-entry process during the first year (13 335 uses and 6095 uses, respectively). Insulin use in the hospital remained consistent on each unit. The highest volume of injectable insulin occurred on the diabetes floor (>11 000 RAIDs per year), the pulmonary floor (700–800 RAIDs per year), and the oncology floor (500–600 RAIDs per year). The lowest volume of injectable insulin occurred on the general medicine floors (250 RAIDs per year) and off-site psychiatry floors (70 RAIDs per year).

A total of 12 months after the implementation of the new PRN process, hospital-wide hyperglycemia rates decreased from a baseline hospital-wide median rate of 55 events per 100 RAIDs to a median rate of 45 events per 100 RAIDs (Fig 1; P = .08). Hypoglycemia rates decreased from a baseline hospital-wide median rate of 6.9 events per 100 RAIDs to a median rate of 3.8 events per 100 RAIDs (Fig 2; P < .0001).

We identified an inverse relationship between insulin error rates and RAIDs on a given patient care unit. Specifically, having at least 50 RAIDs per year on a unit was significantly associated with a reduction in the rate of hyperglycemic and hypoglycemic events: hyperglycemia (Fig 3A; P = .03); hypoglycemia (Fig 3B; P = .04).

After the new PRN process was implemented, the timeliness of insulin administration improved hospital-wide. Specifically, the time from blood glucose check to insulin administration improved from a median of 32 minutes (IQR: 20–51 minutes), in cases in which the former single-entry process was applied, to 18 minutes (IQR: 9–37 minutes) in cases in which the new PRN process was applied (Fig 4; P < .0001).

The improvement work outlined reveals that ADEs associated with the use of insulin, a high-risk injectable drug, may be successfully reduced in an inpatient setting including medical, surgical, and psychiatric care. In our tertiary care center, we showed a significant reduction in the duration of time between glucose checks and insulin administration and improved rates of hyperglycemic and hypoglycemic events. We achieved these successes through implementing a comprehensive insulin safety plan that educates all staff, standardizes and automates insulin ordering and administration, and empowers frontline staff to derive glucose-specific insulin doses in real-time using tools in the EHR. Additionally, the new PRN process helps to compensate for a unit’s lack of exposure to insulin, when it exists, regardless of the patient population at hand. After successful rounds of testing, we found that paging was the best mode of initiating communication when the PRN order parameters were not met (ie, blood glucose outside of stated target), order entry was the most accurate when done on rounds plus real-time (accommodating insulin needs with blood glucose variability), and order accuracy was best achieved when requiring updates and electronic order acknowledgment every 48 hours.

Our institution’s insulin-related ADEs were not directly related to the involvement of the endocrinology service or to a specific unit’s primary population (ie, medical or surgical). Instead, we found a given unit’s overall exposure to insulin was a key factor in safety. We noted a threshold effect, in which units with a rate of at least 50 RAIDs per year had fewer insulin-related errors. Through standardizing and automating insulin ordering, outlining the reasons for communication between insulin order writers and administrators, and providing multiple levels of real-time education and double checks, we reduced insulin-related ADEs, even when incorporating units with <50 RAIDs per year.

Given the high-risk nature of rapid-acting insulin use in the inpatient setting, insulin safety is an ongoing priority in children’s hospitals. In 1 recent study on children, researchers specifically addressed insulin safety and focused on hypoglycemia reduction through automated alerts in the EHR, maximizing communication, enhancing hospital policies, and improving staff members’ knowledge.¹⁵  With our study, we contribute further to the existing literature by addressing insulin safety in medical and surgical units, quantifying hyperglycemia reduction, following additional indicators of safe insulin use (ie, the timeliness of insulin administration), and automating insulin ordering.

The use of an automated system in the EHR at our hospital was vital in reducing insulin-related ADEs. In previous studies, researchers capitalized on automation in efforts to improve hospital safety, particularly to identify hypoglycemia^16–18  as well as other conditions (ie, renal function and coagulopathies).¹⁸  These studies revealed the successful use of automated systems to detect adverse events after they have occurred. With our work, we illustrate how automated systems may also be used proactively to aid in the prevention of ADEs. Ultimately, there is a likely benefit in applying automated EHR systems proactively and reactively to make in-hospital care of children safer with regard to ADEs.

There are limitations to this study. Given the nature of our large institution, our findings may not be generalizable to all inpatient settings. Institutions with limited resources dedicated to extensive education efforts or where the endocrinology unit is not readily available to help determine insulin dosing parameters may not benefit from this approach. Additionally, support from hospital leadership and EHR technical support were vital, and we recognize these resources are not universally available.

Our comprehensive approach to hospital-wide insulin safety is a practical method for standardizing insulin ordering and administration in a tertiary children’s hospital. All or individual portions may be adapted to any size hospital interested in addressing drug events related to injectable medications (ie, insulin).

ADEs related to high-risk injectable medications are unfortunately common. Standardizing approaches to inpatient ordering and administration prove to be helpful in reducing insulin-related ADEs.

ADE

adverse drug event

EHR

electronic health record

IQR

interquartile range

PRN

pro re nata

RAID

rapid-acting insulin day