Article, Pediatrics

Comparing errors in ED computer-assisted vs conventional pediatric drug dosing and administration

Original Contribution

Comparing errors in ED computer-assisted vs conventional pediatric Drug dosing and administration

Loren Yamamoto MD, MPH?, Joan Kanemori RN

Kapiolani Medical Center For Women And Children, Honolulu, HI 96826, USA University of Hawaii John A. Burns School of Medicine, Honolulu, HI 96826, USA

Received 25 December 2008; accepted 10 February 2009

Abstract

Background: Compared to fixed-dose single-vial drug administration in adults, pediatric drug dosing and administration requires a series of calculations, all of which are potentially error prone. The purpose of this study is to compare error rates and task completion times for common pediatric medication scenarios using computer program assistance vs conventional methods.

Methods: Two versions of a 4-part paper-based test were developed. Each part consisted of a set of medication administration and/or dosing tasks. Emergency department and pediatric intensive care unit nurse volunteers completed these tasks using both methods (sequence assigned to start with a conventional or a computer-assisted approach). Completion times, errors, and the reason for the error were recorded.

Results: Thirty-eight nurses completed the study. Summing the completion of all 4 parts, the mean conventional total time was 1243 seconds vs the mean computer program total time of 879 seconds (P b

.001). The conventional Manual method had a mean of 1.8 errors vs the computer program with a mean of 0.7 errors (P b .001). Of the 97 total errors, 36 were due to misreading the drug concentration on the label, 34 were due to calculation errors, and 8 were due to misplaced decimals. Of the 36 label interpretation errors, 18 (50%) occurred with digoxin or insulin.

Conclusions: Computerized assistance reduced errors and the time required for drug administration calculations. A pattern of errors emerged, noting that reading/interpreting certain drug labels were more error prone. Optimizing the layout of drug labels could reduce the error rate for error-prone labels.

(C) 2010

Introduction

Compared to fixed-dose single-vial drug administration in adults, typical pediatric drug dosing and administration requires a series of calculations listed below, all of which are potentially error prone.

* Corresponding author. Honolulu, HI 96828, USA. Tel.: +1 808 983

8387; fax: +1 413 208 2795.

E-mail address: [email protected] (L. Yamamoto).

    1. The patient is weighed, preferably in kilograms.
    2. The weight is recorded.
    3. A physician selects an appropriate drug.
    4. The physician calculates the dose in milligrams per kilogram (usually).
    5. The physician orders this dose in milligrams (usually).
    6. A nurse sees the order and retrieves a medication vial.
    7. The nurse interprets the vial label to determine the drug concentration and the name of the drug within the vial.

0735-6757/$ – see front matter (C) 2010 doi:10.1016/j.ajem.2009.02.009

    1. The nurse calculates the volume of administration by dividing the dose (step 4) by the drug concentration (step 7).
    2. The nurse draws up this volume in a syringe.
    3. The nurse administers the drug to the patient.

Although there are many variations on this, the above is what typically occurs many times per day at a typical patient care unit rendering care to children. Each of the above steps is potentially error prone as listed below:

  1. The patient is weighed in pounds instead of kilograms. The scale weighs the patient incorrectly. An analog scale is interpreted incorrectly.
  2. The recording of the weight is entered incorrectly (decimal in the wrong place, 65 instead of 6.5, key stroke error, or illegible writing) or in the wrong units (pounds instead of kilograms).
  3. The physician orders the wrong medication, or a medication that the patient is allergic to, or a medication that has an interaction with a medication that the patient is already taking.
  4. The physician calculates the dose incorrectly. The wrong weight results in the wrong dose calculated.
  5. The physician enters the dose or the medication incorrectly. The physician uses a computer drop- down menu to select the intended medication but clicks on the medication next to the intended medication. The physician uses a computer drop- down menu to select the medication dose or the units but clicks on the dose or unit next to the choice that is correct. For example, 250 mg is clicked because it is just below the intended dose of 125 mg. For example, micrograms or milliliters is clicked because it is just above or below the intended unit of milligrams.
  6. The nurse interprets the medication name incorrectly due to poor handwriting. The nurse obtains the wrong medication because it is in the wrong dispensing drawer or the label resembles the intended medication. Pharmacy sends the wrong medication to the nurse.
  7. The nurse or pharmacist interprets the label incorrectly. The name of the drug resembles the name of the intended drug, but it is the wrong drug. The concentration of the drug is misinterpreted.
  8. The nurse calculates the wrong volume of drug to administer by entering the wrong concentration, the wrong dose, or incorrectly keystroking the calculator.
  9. The wrong volume is drawn up.
  10. The nurse administers the medication to the wrong patient. The wrong syringe is picked up in the medication preparation station resulting in the patient getting the wrong medication.

The potential for an error in this process is substantial. Ideally, each of these steps should have automated preventive measures and checks resulting in a completely

error proof system. Partial solutions that reduce the potential for serious error include the removal of high-risk drugs from nursing units (eg, concentrated potassium and heparin) to avoid procuring the wrong vial of a dangerous drug, automated warnings about out of range drug doses in computer Order entry systems, and length-based resuscitation systems (eg, Broselow tape, Armstrong Medical Industries, Lincolnshire, IL) that eliminate the need to weigh the patient and calculate doses.

The purpose of this study is to compare error rates and task completion times for pediatric medication scenarios using computer program assistance vs conventional meth- ods, to determine if the computerization of drug dosing and volume of administration calculations can reduce errors and reduce the time required for these calculations.

Methods

This is a paired unblinded sequenced trial. A computer program was developed using Microsoft Excel macros and Visual Basic commands (Microsoft, Redmond, Wash). Nurses completed the tasks described below using a computer method (using this computer program) and using conventional methods that they would normally use, which includes pen/paper, calculators, dosing tools, handbooks, smart infusion pumps, and consulting with a colleague.

Two versions of a 4-part paper-based test were developed. Part 1 (routine drug administration) consisted of 10 drug doses requiring the nurse to calculate the drug administration volume. Part 2 (check the doctor’s dose) consisted of 5 drug doses given the patient’s weight, and the recommended dosing range, requiring the nurse to calculate the drug administration volume, the dose per kilogram, and confirm that the dose per kilogram is within the dosing range. Part 3 (calculate a dose) consisted of 5 problems given the patient’s weight and the dose per kilogram, requiring the nurse to calculate the drug dose, and the drug administration volume. Part 4 (drug infusions) consisted of 3 drug infusion cases given the dose in units per kilogram per hour (insulin) or micrograms per kilogram per minute (inotropes), and the concentration of the drug in the solution bag prepared by the pharmacy, requiring the nurse to calculate the infusion rate. Photographs of actual vial labels provided drug concen- tration information for parts 1, 2, and 3. Current and former pediatric emergency department and pediatric intensive care unit nurse volunteers were informed of the study, and a written informed consent was obtained if they were willing to participate. The institutional review board of this children’s hospital reviewed and approved this study. Nurse volunteers completed these tasks using both methods (sequenced to alternate the starting method with a conventional or a computer-assisted approach). The version of the test taken first was also sequenced resulting in 4 possible combina- tions. These 4 possible combinations (1-Test A done via computer first, 2-Test A done via conventional method first,

3-Test B done via computer first, 4-Test B done via conventional method first) were sequenced, and new subjects were assigned the next combination in the sequence. Nurses study volunteers completed the initial assignment and subsequently the remaining assignment (eg, if they did Test A done via computer first, then their remaining assignment was Test B done via conventional method done second). Test subjects were timed and visually observed during the study by one of the study investigators. Completion times, errors, and the reason for the error were recorded.

Study subjects were oriented to the computer application before this portion of the study. They were permitted to practice with the computer until they felt comfortable with the use of the computer application. The 4 test parts for 2 separate tests resulted in 8 total parts done in the sequence determined by the sequence assignment. Each part could be done independently. Some parts were done while nurses were on duty (during low census periods if time permitted). In most instances, nurses completed all 8 parts in a single day. However, in some instances, nurses completed parts of the study on 1 day, then completed the other parts on one or more other days.

Paired t tests were used to compare the mean number of errors and the completion times. 95% Confidence intervals (CIs) of the mean of the differences between the conven- tional value and the computer value determined clinical significance if the range excluded zero.

Results

Thirty-eight nurses completed both parts (conventional and computer) of the study. Three nurses completed one half of the study but did not complete the other half (two completed computer half only, one completed conventional half only) at the time that the study was closed. These 3 nurses were excluded from the analysis. Task completion times and error means are summarized in Table 1. To complete all 4 parts (for a given test A or B), the mean conventional method total time was 1243 seconds vs the mean computer program total time of 879 seconds (95% CI about mean difference is 224 to 502, P b .001). Summing the number of errors for all 4 parts, the conventional manual

Table 1 Mean task completion times and errors (+-SD) for computer vs conventional methods

Completion times (s)

Conventional

Computer

95% CI of the mean paired differences

No. of errors

Conventional

Computer

95% CI of the mean paired differences

Fig. 1 Distribution of error types.

method had a mean of 1.8 errors vs the computer program with a mean of 0.7 errors (95% CI about mean difference is 0.6 to 1.7, P b .001). Seventy total errors occurred with conventional methods and 27 errors occurred with computer assistance. The error-type distribution is shown in Fig. 1. Of these 97 total errors, 36 were due to misreading the drug concentration on the label, 34 were due to calculation errors (all in the conventional method), 8 were due to misplaced decimals. Twelve errors occurred with digoxin, and 21 occurred with insulin. Of the 36 label interpretation errors, 18 (50%) occurred with digoxin or insulin. See these labels in Figs. 2 and 3. Nine errors occurred with conventional calculations of drug infusions. The computer saved the most time with the drug infusions (131 vs 399 seconds, P b .001). These drug infusions also had a relatively higher number of errors with the conventional method considering that there were only 3 problems to complete. At 0.29 mean errors per nurse, this is an error rate of about 10%, indicating that roughly 1 in 10 of these calculations resulted in the wrong infusion rate.

In reviewing the data forms, we noticed that some nurses did not adhere to the regulatory recommendation that a decimal amount less than 1 should have a leading zero. For example, 0.35 is preferred over .35.

We noticed that of the 38 nurses who completed the study, 29 always used leading zeros, 7 never used leading zeros, and 2 used both formats. There was no significant difference

Part 1

277

+- 135

308 +- 83

-74 to 11

0.53 +- 0.65

0.29 +- 0.52

-0.03 to 0.51

Part 2

320

+- 185

230 +- 81

33 to 148 a

0.61 +- 1.08

0.08 +- 0.27

0.15 to 0.91 a

Part 3

246

+- 87

210 +- 66

8 to 65 a

0.42 +- 0.68

0.29 +- 0.57

-0.12 to 0.39

Part 4

399

+- 241

131 +- 76

190 to 346 a

0.29 +- 0.57

0.05 +- 0.23

0.03 to 0.45 a

All (sum)

1243

+- 474

879 +- 211

224 to 502 a

1.84 +- 1.53

0.71 +- 0.96

0.61 to 1.65 a

a A 95% CI that excludes zero means that the conventional and computer-paired values are significantly different.

Fig. 2 Digoxin vial. The word layout makes this label error prone because the medication concentration (250 ug/mL) is near the total amount of medication in the vial (500 ug).

in error rates and task completion times between those who did use leading zeros vs those who did not.

Discussion

Computer applications and tools require a learning period. Unfamiliarity with the computer application requires addi- tional time to learn how to use the application. In a study such as this, we compared common nursing tasks done in the conventional manner with which nurses are highly familiar and skilled vs a computer application that they have never used. This computer application is simple enough such that it did not take much time for them to feel comfortable with the computer application. Nurses had longer mean task comple- tion times with the computer for part 1 (the first test portion that the computer application was used), but this difference was not statistically significant. Failure to demonstrate computer time superiority for part 1 probably occurred for

2 reasons. (1) This was their first experience with the computer application. (2) Part 1 required the entry of the patient’s weight data for the computer method, yet the weight was not required for manual calculation of the medication administration volume. The computer application yields more clinical information by requiring the weight data, but it was unnecessary to complete the task in part 1 resulting in more time to complete the task.

Despite the above handicap, summing the task completion times for parts 1, 2, 3, and 4 still resulted in faster times for the computer method compared to the conventional method because gaining experience resulted in faster completion times for parts 2, 3, and 4. The computer times for part 4 were substantially shorter than that for the manual method. Emergency department nurses do not frequently calculate continuous drug infusion rates. It is something they have learned, but it is not something that they do frequently. Pediatric intensive care unit nurses perform these calcula- tions more frequently. Most nurses do not calculate these infusion rates but rather use tools such as handbooks and smart infusion pumps to determine these infusion rates.

Medication vial labels typically have 4 required compo- nents that the nurses review: (1) the medication name, (2) the total amount of drug contained in the vial, (3) the total volume of the vial contents, (4) the concentration of the drug

in the vial. For example, the clindamycin vial label included the following information: (1) clindamycin, (2) 300 mg, (3) 2 mL, (4) 150 mg/mL. The 300-mg total vial dose is more commonly reviewed by nurses administering a standard adult dose of 300 mg (one vial) because the entire vial is generally drawn up and administered. However, because pediatric doses are fractions of this, a nurse responsible for administering 80 mg of clindamycin would be most interested in the concentration of 150 mg/mL. There are other mathematical alternatives to derive the correct volume as well. However, it should be noted that nurses will often focus on different parts of the label for pediatric vs adult patients. The 4 informational components on the vial labels are not in a standard arrangement. This requires the nurse to read and interpret the vial in a nonstandardized fashion.

The digoxin and insulin labels in Figs. 2 and 3 are demonstrated in this study to be more error prone. This study did not compare label arrangements, but most of the other labels did not have the medication concentration near the total vial volume or total drug quantity in the vial. It appears that having the concentration near the total vial volume or the total drug quantity is the factor that makes the label more error prone.

Fully standardizing labels such that the key elements are in a prescribed location, with a standard font, and colors will make the drug vials look alike, which creates a different type of risk for error. Thus, fully standardizing labels is unwise as well. However, simply arranging the label so that the concentration is not near the total vial volume or total medication quantity is likely to reduce interpretation errors.

Fig. 3 Insulin vial. The word layout makes this label error prone because the medication concentration (100 U/mL) is near the volume of the vial (10 mL).

The Introduction section describes the multiple points of task completion where an error can occur. Computer support systems such as that described in this study can reduce errors and speed task completion times for only a few of these points in the sequence. Errors still occurred with computer use. It is difficult to imagine a system that is absolutely error proof, but it might look something like this:

    1. Instead of weighing the patient, the patient is scanned by a laser body dimensioning system.
    2. This weight estimate is automatically entered into the computer system without human intervention.
    3. Basic symptoms are entered from a check list, and a physician evaluates the patient and chooses the diagnostic category from a limited set of choices produced by the symptom check list, reducing the likelihood of a diagnostic error. Alternate Diagnostic categories that are rare could be entered but would require additional justification. Based on this diag- nostic category, a limited set of drug choices (allergic drugs or drugs with interaction potential are auto- matically excluded) are presented and one (or more) is selected by the physician.
    4. The computer calculates the dose based on dosing recommendations. This dose is confirmed by the physician.
    5. The dose is communicated within the hospital computer system in a standardized unit.
    6. A nurse sees the order, and a bin in an automated dispensing unit opens for the nurse. The vial was bar code scanned by the automated dispensing unit during the stocking of the dispensing unit to confirm that the correct medication vial with the expected medication concentration was placed in the bin.
    7. The automated dispensing unit calculates the volume of administration for the nurse. Medication vials that are shipped as powder requiring reconstitution just before administration will be shipped as units with the diluent self-contained within the unit to prevent volume of diluent errors.
    8. The nurse inserts the vial into a newly designed smart infusion pump that is connected to the hospital computer system. The infusion pump asks for the

patient’s name and ID from a list. It reads the bar code on the medication vial and confirms that the patient entered has that medication ordered and that the medication was recently removed from the automated drug dispensing system.

    1. The smart infusion pump computes the correct infusion volume. The nurse proceeds to the patient’s bedside.
    2. The smart infusion pump scans the patient’s ID wrist bracelet and/or thumbprint. It confirms the patient’s identity and confirms that this patient has the medication contained within it ordered for this time. The nurse attaches the infusion pump to the patient’s IV and the drug is delivered by the smart infusion pump. Oral, rectal, intramuscular, subcutaneous, inhaled, topical, ophthalmic, and so on, medications would require alternate smart delivery systems. The smart infusion pump then sends a signal to the hospital computer that the medication has been administered and an entry is made into the patient’s medication administration record.
    3. Additional checks and balances could be implemented. If the medication is not removed from the automated medication dispensing unit within 30 minutes of the medication order administration time, an alert will notify the nurse and the medication administration supervisor (eg, charge nurse, unit supervisor, pharma- cist). If medication administration in not confirmed within 1 hour of removal from the automated medication dispensing unit, an alert will notify the nurse and the medication administration supervisor.

It is clear that the implementation of high reliability systems have a long way to go. The computer program described in this study affects only 1 or 2 of these steps, and its reliability is still compromised by human data input, which this study confirms to have a potential for error.

In conclusion, computerized assistance reduced errors and the time required for drug administration calculations. A pattern of errors emerged, noting that drug infusion calculations and reading/interpreting certain drug labels (digoxin and insulin) were more error prone. Optimizing the layout of drug labels could reduce the error rate for error prone labels.