Content uploaded by Matthieu Roustit
Author content
All content in this area was uploaded by Matthieu Roustit on Oct 17, 2017
Content may be subject to copyright.
Feature Articles
Automated drug dispensing system reduces medication errors in
an intensive care setting
Claire Chapuis, PharmD, MSc; Matthieu Roustit, PharmD, MSc; Gae¨lle Bal, MSc; Carole Schwebel, MD, PhD;
Pascal Pansu, PhD; Sandra David-Tchouda, MD, PhD; Luc Foroni, PharmD; Jean Calop, PharmD, PhD;
Jean-Franc¸ois Timsit, MD, PhD; Benoît Allenet, PharmD, PhD; Jean-Luc Bosson, MD, PhD;
Pierrick Bedouch, PharmD, PhD
Drug therapy safety relies on
many parameters, making
drug complications a major
cause of medical injury. Ad-
verse drug events, defined as injuries
caused by medical treatment, represent a
major health issue (1). Certain adverse
drug events are considered as medication
errors (MEs), defined as any preventable
event that may lead to inappropriate med-
ication use or patient harm (2). According
to available data, MEs harm more than 1.5
million people and cause 7,000 deaths an-
nually in the United States (3).
In the past 15 years, many studies
have described MEs in drug preparation
and administration in medical and surgi-
cal units (4 –7). Others have specifically
assessed the incidence of MEs in medical
intensive care units (MICUs) (8 –14).
Highly unstable critically ill patients
are more vulnerable to MEs, and the
risk of errors is increased in these pa-
tients because of the number of drugs
they receive and the way they are ad-
ministered (i.e., intravenous infusions)
(1). In most cases, MEs reached the
patient (8, 9, 12, 13).
Therefore, reducing errors is crucial to
improving patients’ outcomes. Information
technology and automated systems have
been introduced to improve the medication
use process: computerized physician order
entry systems (15), unit dose drug distribu-
tion (5), bar-coded medication administration
(16), and automated dispensing systems
(ADSs), which are computer-controlled dis-
pensing units providing secure storage and
drug distribution in care units. ADSs have
improved medication use in surgical and
medical units, with an impact on administra-
tion time errors, omissions, and work activi-
ties (17–20).
A recent report showed that ward-based
ADSs can reduce costs while reducing error
rates (21). However, their clinical impact in
intensive care units (ICUs) remains to be
determined.
From the Pharmacy Department (CC, LF, JC, BA, PB),
Grenoble University Hospital, France; Clinical Research Cen-
ter—Inserm CIC03 (MR, GB, SDT, JLB), Grenoble University
Hospital, Grenoble, France; Medical Intensive Care Depart-
ment (CS, JFT), Grenoble University Hospital, Grenoble,
France; Educational Science Laboratory (PP), Pierre Mende`s-
France University, Grenoble, France; ThEMAS TIMC UMR
CNRS 5525 (SDT, JC, BA, JLB, PB), Joseph Fourier Univer-
sity, Grenoble, France.
Financial support provided by De´le´gation Re´ gion-
ale a` la Recherche Clinique, Grenoble University Hos-
pital, Grenoble, France.
The authors have not disclosed any potential con-
flicts of interest.
For information regarding this article, E-mail:
PBedouch@chu-grenoble.fr
Copyright © 2010 by the Society of Critical Care
Medicine and Lippincott Williams & Wilkins
DOI: 10.1097/CCM.0b013e3181f8569b
Objectives: We aimed to assess the impact of an automated
dispensing system on the incidence of medication errors related
to picking, preparation, and administration of drugs in a medical
intensive care unit. We also evaluated the clinical significance of
such errors and user satisfaction.
Design: Preintervention and postintervention study involving a
control and an intervention medical intensive care unit.
Setting: Two medical intensive care units in the same depart-
ment of a 2,000-bed university hospital.
Patients: Adult medical intensive care patients.
Interventions: After a 2-month observation period, we imple-
mented an automated dispensing system in one of the units (study
unit) chosen randomly, with the other unit being the control.
Measurements and Main Results: The overall error rate was
expressed as a percentage of total opportunities for error. The
severity of errors was classified according to National Coordinat-
ing Council for Medication Error Reporting and Prevention cate-
gories by an expert committee. User satisfaction was assessed
through self-administered questionnaires completed by nurses. A
total of 1,476 medications for 115 patients were observed. After
automated dispensing system implementation, we observed a
reduced percentage of total opportunities for error in the study
compared to the control unit (13.5% and 18.6%, respectively; p<
.05); however, no significant difference was observed before
automated dispensing system implementation (20.4% and 19.3%,
respectively; not significant). Before-and-after comparisons in the
study unit also showed a significantly reduced percentage of total
opportunities for error (20.4% and 13.5%; p<.01). An analysis of
detailed opportunities for error showed a significant impact of the
automated dispensing system in reducing preparation errors (p<
.05). Most errors caused no harm (National Coordinating Council
for Medication Error Reporting and Prevention category C). The
automated dispensing system did not reduce errors causing
harm. Finally, the mean for working conditions improved from
1.0 ⴞ0.8 to 2.5 ⴞ0.8 on the four-point Likert scale.
Conclusions: The implementation of an automated dispensing
system reduced overall medication errors related to picking,
preparation, and administration of drugs in the intensive care
unit. Furthermore, most nurses favored the new drug dispensation
organization. (Crit Care Med 2010; 38:2275–2281)
KEY WORDS: hospital drug distribution systems; intensive care
units; medication errors; drug safety; clinical pharmacy informa-
tion systems; job satisfaction
2275Crit Care Med 2010 Vol. 38, No. 12
Besides technological advances, the
quality of medication use relies on inter-
actions between care providers, patients,
information, and technology. The major-
ity of MEs are a direct consequence of the
intrinsic complexity of these interactions
(22). To be successful, users must be in-
volved in the implementation of new
technologies and be able to express per-
ceived benefits, or disadvantages, of the
new system.
Consequently, we aimed to assess the
impact of ADS implementation on the
incidence of MEs related to picking, prep-
aration, and administration of drugs in
the MICU. We also evaluated the clinical
impact of errors and user satisfaction as
secondary objectives.
PATIENTS AND METHODS
Population and Setting
This study was conducted over the course
of 4 months in two MICUs of the same depart-
ment in a 2,000-bed university hospital. Both
units (8 and 10 beds) had comparable activi-
ties and shared the same staff.
Medications were delivered daily with a
floor stock drug distribution system. Nurses
picked drugs directly from the classic medi-
cine cabinet. There was no onsite pharmaceu-
tical review of medication orders. Study ethics
approval was obtained on November 26, 2007
(Institutional Review Board 5891).
Intervention
An OmniRx ADS (Omnicell, Mountain
View, CA) was implemented in one unit (study
unit), chosen randomly, while the former dis-
tribution system was maintained in the other
unit (control unit). The computer-controlled
ADS stores most of the medications directly in
the nursing unit and records medication pick-
ing. The compartments of the ADS are re-
loaded daily by a pharmacy technician, except
on weekends, when drugs are dispensed con-
ventionally by an on-call pharmacist.
Before implementing the ADS, nurses at-
tended a training program. A 2-wk run-in pe-
riod was allowed before data collection to per-
mit nurses to become familiar with the
system.
Study Design and Data
Collection
Errors were collected by direct observation
(23), which is previously shown to be reliable
for identifying MEs (24). A pharmacist ob-
served picking, preparation, and administra-
tion of drugs by nurses. Two-month observa-
tion periods were performed: before (phase I)
and after (phase II) implementation of the
ADS in the study unit, each preceded by a
15-day run-in period.
Observation sessions were 3– 4 hrs long
and occurred 4 days per week, including
nights and weekends. They were unannounced
and concerned all the nurses present in the
unit (two or three patients per nurse), and the
unit and nurse were randomly assigned by a
methodologist. Preparations for several pa-
tients could be observed simultaneously (they
took place in the same workstation), whereas
administration was observed for only one pa-
tient at a time.
The data collection procedure had been
tested during the run-in period to lessen the
so-called Hawthorne effect (people may be-
have differently when they know they are be-
ing observed) (25). For ethical reasons, the
observer intervened whenever an error was
considered as possibly harmful (the error was
then included in the analysis).
ME Assessment
Any discrepancy between the medication
use process (picking, preparation, and admin-
istration) and prescriptions or recommenda-
tions was considered as an error.
Our primary outcome was the overall error
rate during picking, preparation, and admin-
istration. Overall error rate was defined as the
percentage of total opportunities for error
(%TOE), calculated by dividing the number of
drugs associated with one or more errors by
the number of drugs ordered (whether
picked), as previously described (11, 25).
As secondary outcomes, we expressed MEs
as the percentage of detailed opportunities for
error (%DOE) to calculate distinct error rates
for picking, preparation, and administration
(13). We defined as DOE any gesture/action by
a nurse that could result in the following types
of error: name, dosage, and pharmaceutical
form for picking; dose, solvent type/volume for
reconstitution, and mixtures for preparation;
technique (13), route, rate, time more than 1
hour before or 1 hour after expected time, and
physicochemical incompatibility for adminis-
tration. Thus, each prescribed dose was asso-
ciated with up to 12 DOE. For each type of
error we calculated %DOE by dividing all ob-
served errors by the number of DOE. Omis-
sions and extra doses were analyzed separately.
Another secondary objective was to com-
pare errors in the conditions of drug storage
(storage errors) in both units before and after
ADS introduction. Errors concerned inappro-
priate storage temperature and/or protection
from light and humidity. They were expressed
as the percentage of storage errors, calculated
by dividing the number of storage errors by
the number of drugs. To ensure the quality of
data collection, 10% of the case report forms
were monitored by a research assistant.
ME Severity
An independent multidisciplinary commit-
tee (intensive care specialist, methodologist,
pharmacologist, clinical pharmacist), not in-
volved in the observation process, retrospec-
tively reviewed all errors. The committee had
access to the whole description of the error
but was blinded to the unit and the phase (i.e.,
ADS or conventional system) and to the nurse
and the patient.
Each error was classified according to the
following criteria: stage in the process (pick-
ing, preparation, and administration), type of
error, and severity using the National Coordi-
nating Council for Medication Error Report-
ing and Prevention (NCC MERP) method. We
distinguished among errors reaching the pa-
tient but causing no harm (NCC MERP cate-
gories C and D), those causing harm (NCC
MERP categories E to H), and those causing
death (NCC MERP category I) (2).
User Satisfaction
Perceived usefulness is related to the in-
tention of use and usage behavior (26). Self-
administered questionnaires were used to
evaluate nurses’ perceptions of the new sys-
tem. A four-point Likert scale was used to
avoid inconclusive neutral answers (27). Sur-
veys were organized 2 wks before, 6 wks after,
and 8 months after ADS implementation to
highlight any “resistance-to-change” phenom-
ena (28).
Sample Size Calculation and
Statistical Analysis
According to literature, the ME rate was
estimated at 7% DOE in the MICU (13, 29). A
39% decrease in errors after implementing the
ADS was described, including organizational
errors (19). Because our methodology did not
allow their detection, we hypothesized a 33%
decrease to calculate the number of observa-
tions needed. Assuming an error rate of 7%
DOE in the control group, 1,785 observations
would be needed in each group to detect sta-
tistical significance (chi-square test or an ex-
act Fisher test if required) with a risk of 0.05
and a power of 0.85 (nQuery Advisor 6.01 for
Windows). Anticipating 10% missing data,
2,000 observations per unit would be needed
before and after ADS introduction.
Categorical data were reported as fre-
quency and percentage and continuous data as
mean and standard deviation or median and
interquartile range (i.e., 25th and 75th per-
centiles) when appropriate. Exact 95% confi-
dence intervals were computed from the bino-
mial distribution for binary outcomes.
Characteristics were compared between study
groups using the Kruskal-Wallis test for quan-
titative data and the chi-square test, or Fish-
2276 Crit Care Med 2010 Vol. 38, No. 12
er’s exact test when appropriate, for categori-
cal data.
We then modeled binary outcomes using
logistic regression models that comprised the
study group, the study period, and a first-order
interaction between the study group and pe-
riod. We estimated the absolute difference in
change between study groups from the pre-
dicted probabilities derived from the logistic
regression models. This is the same strategy as
used previously (30).
We considered p⬍.05 as significant. Sta-
tistical analyses were performed using Stata
version 10.2 (Stata Corporation, College Sta-
tion, TX).
RESULTS
Descriptive Data
Sixty-eight nurses were involved in
this 4-month study. They picked, pre-
pared, and administered 1,476 medica-
tions (involving 8,753 DOE) to 115 pa-
tients. Patient characteristics are given in
Table 1. The median (interquartile range)
length of stay was 10 days (4 –23 days).
The most frequent principal diagnoses
were respiratory failure and septic shock.
During the whole study, we identified
295 errors related to picking, prepara-
tion, or administration. The observer in-
tercepted seven errors (four before and
three after ADS implementation). The
type of electrolytes (e.g., potassium, so-
dium, calcium, magnesium) was involved
in the highest number of errors (26%).
The second most common error con-
cerned insulin (6%). Most medications
(88%) were intravenous drugs.
Impact of the ADS on MEs
After ADS implementation, we ob-
served a difference in the %TOE between
control and study units (18.6% and
13.5% TOE, respectively; p⬍.05),
whereas no difference was observed be-
fore ADS implementation (19.3% and
20.4% TOE, respectively; not significant).
The %TOE also significantly decreased in
the study unit between phase I and phase
II (p⬍.01) (Fig. 1). The %TOE decrease
in the intervention unit was 6.9% and
0.7% in the control unit, with a 6.2%
absolute difference in change (95% con-
fidence interval, ⫺1.8% to 14.2%; p⫽
.13). Table 2 shows errors expressed as
%DOE in both units for both phases. In
the study unit, the incidence of prepara-
tion dose errors significantly decreased
from 3.8% to 0.5% DOE (p⫽.017). How-
ever, the ADS did not decrease picking
and administration errors. The latter
mostly concerned the rate of infusion
(47%) and administration time (34.2%).
We also observed 16 omission errors and
two extra doses during the whole obser-
vation period, with no difference between
the two phases.
Besides the 295 errors observed dur-
ing picking, preparation, and administra-
tion, we identified 145 storage errors. The
ADS had a major impact on this type of
error with 51 (27.7%) and 65 (34.9%)
errors (study and control units, respec-
tively) in phase I compared to 2 (0.7%)
and 27 (14.4%; p⬍.01) errors after ADS
implementation (phase II, study and con-
trol units, respectively). Storage errors
were drastically decreased in the study
unit (96% reduction; p⬍.01). It is of
note that they were also significantly de-
creased in the control unit (58% de-
crease; p⬍.01). Table 3 provides exam-
ples of observed errors and propositions
to avoid them.
Severity of Errors
The severity of observed errors, with
and without ADS, is summarized in Table
4. Most errors (84%) were classified as
errors causing no harm (NCC MERP cat-
egories C and D). The ADS decreased
NCC MERP category C errors by 35%
(2.6% [2.2%–3%] before and 1.7%
[1.3%–2.3%] after implementation).
However, errors in categories D to H
were not affected. No error resulted in
patient death. We observed 10 picking
errors causing patient harm, all in the
control unit.
User Satisfaction
The median age of the 68 nurses in-
volved in this study was 27 (4.2) years;
76.2% were female and they had worked
in the MICU for a median of 1.1 (1.8)
years. A total of 64 questionnaires were
returned by the nurses. Response rates
were 36% (n ⫽18) before ADS imple-
Figure 1. Overall medication error rate expressed as a percentage of total opportunities for error
(%TOE) in the study and the control units before (phase I) and after (phase II) implementation of the
automated dispensing system. *p⬍.05 (chi-squared test). §p⬍.01 (chi-squared test). #p⫽.10
(interaction test).
Table 1. Characteristics of the studied population in the control and the study units before (phase I)
and after (phase II) implementation of the automated dispensing system
Control Unit Study Unit
Total
(n ⫽115)
Phase I
(n ⫽31)
Phase II
(n ⫽25)
Phase I
(n ⫽32)
Phase II
(n ⫽27)
Age (yrs) 63 (54–73) 61 (53–74) 63 (56–71) 62 (49–76) 62 (53–73)
Male 17 (63) 21 (67.7) 20 (62.5) 18 (72) 76 (66.1)
Length of stay (days) 11 (5–25) 6 (3–12) 11.5 (6–25) 12 (4–34) 10 (4–23)
Simplified Acute Physiology
Score II
41 (37–53) 44 (30–57) 44 (36–61) 48 (37–68) 44 (34–60)
Deaths 8 (29.6) 5 (16.1) 5 (15.6) 6 (24) 24 (20.9)
Quantitative data are expressed as median (interquartile range). Qualitative data are expressed as
number (percentage).
2277Crit Care Med 2010 Vol. 38, No. 12
mentation, 31% (n ⫽14) at 6 wks, and
57% (n ⫽32) at 8 months. Only 10
nurses answered all three surveys, prob-
ably because of high staff turnover. De-
scriptive data about nurse satisfaction re-
vealed a tendency to greater satisfaction
with time, especially regarding time
saved and working conditions (Fig. 2).
Finally, the majority of nurses wished to
continue using ADS (96.7%).
DISCUSSION
Using a direct observation approach,
we showed a decrease in the incidence of
errors after implementing ADS. In addi-
tion, storage errors were drastically de-
creased by ADS. The most frequent errors
concerned preparation (especially the use
of wrong diluents) and administration
(especially rate and time) processes. Dose
errors were significantly reduced (p⬍
.05) by ADS. Furthermore, ADS was well
perceived by nurses.
To date, several studies have assessed
MEs in ICUs, showing highly variable
rates from one study to another (3.3%–
44.6%) (8, 10, 11, 13, 14). These discrep-
ancies may be explained by heteroge-
neous drug distribution systems and ways
of data expression (%TOE or %DOE).
Furthermore, the ratio of intravenous
compared to oral doses prescribed may
influence the error rate. Our results ex-
pressed as %TOE are consistent with the
findings from Kopp et al (14) and Barker
et al (7), showing an incidence of approx-
imately one error for every five doses of
medication administered before ADS was
implemented. Besides, Tissot et al (13)
and Fahimi et al (10) found slightly
higher error rates with data expressed as
%DOE, probably because of variations in
Table 2. Distribution of medication errors related to picking, preparation, and administration, expressed as a percentage of detailed opportunities for error
before (phase I) and after (phase II) implementation of the automated dispensing system
Control Unit Study Unit
Phase I (n ⫽300) Phase II (n ⫽333) Phase I (n ⫽368) Phase II (n ⫽475)
Stage Type of Error %DOE Errors DOE %DOE Errors DOE %DOE Errors DOE %DOE Errors DOE
Picking Total 1.9 9 465 3.3 15 460 1.8 8 435 1.5 9 595
Brand name 0.8 2 241 1.7 4 240 1.3 3 236 0.6 2 331
Dosage 4.0 4 100 5.7 7 123 4.5 4 88 3.2 5 156
Form 2.4 3 124 4.1 4 97 2.4 1 111 1.9 2 108
Preparation Total 3.8 26 678 4.1 25 604 6 42 695 3.4 26 764
Dose 1.6 3 191 1.9 3 157 3.8 7 185 0.5 1 200
Diluent type 0.7 1 148 2.3 3 131 0.6 1 159 1.1 2 178
Diluent volume 11.0 18 163 10.5 16 153 15.7 26 166 11.6 22 189
Incompatibility 2.3 4 176 1.8 3 163 4.3 8 185 0.5 1 197
Administration Total 3.1 28 898 3.1 25 799 2.8 31 1120 2.7 33 1240
Technique 0 0 186 0.6 1 164 0.4 1 232 0.4 1 251
Route 0.5 1 187 0.6 1 165 0 0 232 0 0 249
Rate 10.6 18 170 7.3 11 150 4.3 9 211 6.8 17 250
Time 4.2 8 191 3.8 7 182 7.4 18 243 2.6 7 271
Incompatibility 0.6 1 164 3.6 5 138 1.5 3 202 3.7 8 219
Total DOE
(95%
confidence
interval)
3.1 (2.4–4.0) 63 2041 3.9 (3.1–4.9) 65 1863 3.8 (3.1–4.7) 81 2250 2.7 (2.1–3.4) 68 2599
DOE, detailed opportunities for error; N, number of medications observed.
Data are expressed as numbers (errors and DOE) or percentages (%DOE), with 95% confidence interval in parenthesis for total DOE.
Table 3. Examples of errors and how automated dispensing system can avoid them
Examples of Error Stage Type of Error Solution ADS
Magnesium chloride instead of calcium chloride Picking Brand name Write calcium on the screen of ADS Yes
Cefotaxime instead of ceftriaxone (rocephine) Picking Brand name Write Rocephine, the ADS recognizes it as
ceftriaxone
Yes
Oral piracetam administered by intravenous route Picking Form Write piracetam and the ADS will propose both
intravenous and oral
Yes
Mycophenolate mofetil 250 instead of 500 mg
(capsule opened)
Picking Dosage ADS proposes both dosages Yes
L-Dopa immediate release instead of prolonged
release
Picking Form ADS proposes both forms Yes
L-Thyroxin, wrong dosage Picking Dosage ADS proposes several dosages Yes
Insulin administered at wrong rate Administration Rate Follow protocol and adjust rate to glycemia No
Potassium chloride, wrong dose Preparation Dose Check quantity prescribed twice No
Dilution of liposomal amphotericin B in NaCl 0.9% Preparation Diluent type Check type of diluent in reference book before
preparation
No
Intravenous ofloxacin administered within ⬍5
mins
Administration Rate Check recommendation for infusion rate in reference
book
No
ADS, automated dispensing system.
2278 Crit Care Med 2010 Vol. 38, No. 12
the nature and number of observed DOE.
Because there is no consensus about data
expression, we used both methods. The
%TOE as defined by Barker et al (31) has
been widely used, making it easier to com-
pare our results with other works. The ap-
proach of Tissot et al (29) is more realistic,
calculating separate error rates for each
stage of the preparation and administration
process, which is the reason why we also
expressed our results as %DOE. Interest-
ingly, some recent works have expressed
MEs as the number of events per patient
ICU days. Valentin et al (12) showed an
incidence of 74.5 errors per 100 patient ICU
days. Previous data revealed a highly vari-
able rate of MEs, ranging from 1.2 to 947
errors per 1000 patient ICU days, with a
median of 106 errors per 1000 patient ICU
days (32). Because of this variability, and
because our study was not designed to in-
clude all patients admitted to the partici-
pating units, we did not express our results
as MEs per patient days.
We performed multiple pairwise com-
parisons between study groups (i.e., con-
trol vs. intervention) and periods (prein-
tervention vs. postintervention) as
described by Bates et al (15), which is a
common method, to compare our results
with those of other similar studies. Al-
though less common, a more appropriate
approach is to compute the absolute
change in TOE for each study group sep-
arately and then derive the difference in
changes from logistic regression models,
including a first-order interaction term
between study group and period (30). Us-
ing such a method, however, we did not
reach statistical significance, probably
because of a lack of power. Sample size
calculation was not based on this test.
Most studies have assessed MEs using
direct observation. Although this is con-
sidered as more efficient and accurate
than reviewing charts and incident re-
ports (24), it may underestimate MEs.
Our study design did not provide infor-
mation linking the MEs to a nurse. Thus,
we could not account for ME clustering
by nurse, with the potential for optimistic
95% confidence intervals.
The picking of incorrect medications
and the confusion of similar drug names
are problems that occur with open-shelf
storage of pharmaceuticals (22). It was
therefore a reasonable assumption that the
ADS would reduce picking errors. How-
ever, although we observed a significant
reduction in preparation dose errors, we
found no significant decrease in picking
errors with the ADS. This may be explained
by the scarcity of such events and a poten-
tial lack of power and should be tested in a
larger-scale study. Nonetheless, on close in-
spection, we realized that ADS could not
prevent all picking errors. For example,
when potassium chloride 2 g/L is pre-
scribed in a 500-mL volume, there is a risk
that 2 grams are added instead of 1 gram.
Another example we observed was picking
of fluconazole 400 mg instead of flucon-
azole 200 mg, because they appeared next
to one another on the screen. Our ADS
allowed users to select any drug available,
thus overriding the prescription, creating a
potential source of error. Optimal integra-
tion of a computerized physician order en-
try interface could help to achieve greater
error reduction, making medications avail-
able to nurses only after physician orders
are entered into the system (18 –20).
The lack of effect on administration
errors (time, omissions, rate) was ex-
pected because most of these are not di-
rectly preventable by ADS. For example,
Figure 2. User satisfaction assessed through self-administered questionnaires filled in by the nurses
2 wks before automated dispensing system implementation (n ⫽18), 6 wks after (n ⫽14), and 8
months after (n ⫽32). Satisfaction is expressed as mean score of a Likert scale ranging from 0 (not
satisfied) to 3 (very satisfied).
Table 4. Potential clinical significance of errors in the study unit before the automated dispensing
system implementation and in the control unit (without automated dispensing system) compared to
the study unit after automated dispensing system implementation (with automated dispensing system)
No Automated Dispensing System
(6,154 Detailed Opportunities for
Error) % (n)
Automated Dispensing
System (2,599
Detailed Opportunities
for Error) % (n)
Errors not causing harm 3.1 (190) 2.1 (54)
C (reached the patient) 2.6 (158) 1.7 (44)
D (reached the patient and resulted
in increased patient monitoring
and/or intervention)
0.5 (32) 0.4 (10)
Errors causing harm 0.6 (35) 0.7 (16)
E (resulted in need for therapy or
intervention, caused temporary
harm)
0.3 (17) 0.3 (7)
F (resulted in initial or prolonged
hospitalization and temporary
harm to patient)
0.2 (13) 0.3 (6)
H (resulted in near-death event) 0.1 (5) 0.1 (3)
Data from the control unit (phases I and II) and from the study unit (phase I) were pooled and
compared with data in the study unit after automated dispensing system implementation (phase II).
Data are expressed as a percentage of detailed opportunities for errors according to their severity
using the National Coordinating Council for Medication Error Reporting and Prevention method. We
distinguished errors reaching the patient causing no harm (National Coordinating Council for
Medication Error Reporting and Prevention categories C and D) and those causing harm (National
Coordinating Council for Medication Error Reporting and Prevention categories E to H).
2279Crit Care Med 2010 Vol. 38, No. 12
we frequently observed serious errors in
the glycemia control protocol, indepen-
dent of ADS use.
Another finding was the sharp reduction
in storage errors after ADS implementa-
tion, not only in the study unit but also in
the control unit. This may be explained by
the positive effect of ADS on drug storage
information and by a methodologic limita-
tion of our study: the same nurses worked
in both units. This limitation may have
improved the quality of medication use in
the control unit, thus underestimating the
impact of the ADS and explaining the non-
significant absolute difference in change.
However, we favored a comparison between
two units with similar activities and orga-
nization despite sharing the same staff,
rather than choosing a control MICU in
another institution or a surgical ICU in our
hospital.
The analysis of error severity revealed
that almost 85% of errors caused no
harm (NCC MERP category C). These er-
rors were reduced by the ADS. However,
we observed no impact on errors associ-
ated with harm, mostly concerning per-
fusion rates (e.g., insulin, propofol) and
omissions. Such errors cannot be pre-
vented by ADS.
New technologies may lead to misuse
or incorrect behaviors when not per-
ceived as appropriate (33, 34). In the
present study, evaluation of users’ satis-
faction showed an intermediate (short-
term) stage corresponding to a period of
reorganization and “resistance to
change” phenomena (28). The implemen-
tation of ADS induced a substantial reor-
ganization in the drug distribution sys-
tem. However, overall satisfaction was
clearly improved after 8 months when
nurses estimated that they were spending
less time receiving medication-related ac-
tivities. However, our quantitative data
about user satisfaction should be consid-
ered with caution because response rates
to questionnaires were low (from 31%–
57%) and only ten nurses answered to all
three surveys, thus lessening the validity
of such a tool. Nonetheless, overall feed-
back for ADS was positive, as described
previously (35). Fatigue, stress, heavy
workload, new staff, and personal neglect
are risk factors for MEs (33); thus, efforts
should focus on improving nurses’ work-
ing conditions and knowledge about
medications (34).
Finally, profound changes in medica-
tion use procedures could further im-
prove safety. Particularly, preparation of
doses by the pharmacy department could
decrease preparation errors (36). Fur-
thermore, remarkably low ME rates are
reported after combined ADS introduc-
tion and healthcare team reorganization
(8) involving physicians, nurses, pharma-
cists, unit directors, and managers (37).
CONCLUSIONS
The implementation of ADS reduced
overall MEs related to picking, prepara-
tion, and administration of drugs in the
MICU. Most nurses favored the new drug
dispensation organization. However, one
should bear in mind that any changes
may generate new error risks, justifying
continuous quality monitoring. A larger-
scale multicenter study is in preparation
to further assess the impact of ADS on
the detailed medication use process and
to determine its cost efficiency.
ACKNOWLEDGMENTS
We thank Marion Proust, research as-
sistant, for her valuable assistance; Jean-
Louis Quesada for help on statistical
methods; Dr. Ce´ line Villier, pharmacolo-
gist, Pharmacovigilance Centre Grenoble
University Hospital, for her expertise; and
Dr. Alison Foote, for editing the text. We
also thank all the nurses who took part in
the study.
REFERENCES
1. Leape LL, Brennan TA, Laird N, et al: The
nature of adverse events in hospitalized pa-
tients. Results of the Harvard Medical Prac-
tice Study II. N Engl J Med 1991; 324:
377–384
2. National Coordinating Council for Medica-
tion Error Reporting and Prevention (NCC
MERP): Taxonomy of medication errors.
Available at: http://www.NCCMERP.org. Ac-
cessed August 18, 2009
3. Aspden PW JA, Palugo R, Bastien T: Institute
of Medicine. Preventing Medication Errors.
Washington, DC, National Academies Press,
2006
4. Bates DW, Cullen DJ, Laird N, et al: Inci-
dence of adverse drug events and potential
adverse drug events. Implications for preven-
tion. ADE Prevention Study Group. JAMA
1995; 274:29 –34
5. Taxis K, Dean B, Barber N: Hospital drug dis-
tribution systems in the UK and Germany—a
study of medication errors. Pharm World Sci
1999; 21:25–31
6. Anselmi ML, Peduzzi M, Dos Santos CB: Er-
rors in the administration of intravenous
medication in Brazilian hospitals. J Clin
Nurs 2007; 16:1839 –1847
7. Barker KN, Flynn EA, Pepper GA, et al: Med-
ication errors observed in 36 health care fa-
cilities. Arch Intern Med 2002; 162:
1897–1903
8. Calabrese AD, Erstad BL, Brandl K, et al:
Medication administration errors in adult pa-
tients in the ICU. Intensive Care Med 2001;
27:1592–1598
9. Rothschild JM, Landrigan CP, Cronin JW, et
al: The Critical Care Safety Study: The inci-
dence and nature of adverse events and seri-
ous medical errors in intensive care. Crit
Care Med 2005; 33:1694 –1700
10. Fahimi F, Ariapanah P, Faizi M, et al: Errors
in preparation and administration of intrave-
nous medications in the intensive care unit
of a teaching hospital: An observational
study. Aust Crit Care 2008; 21:110 –116
11. van den Bemt PM, Fijn R, van der Voort PH,
et al: Frequency and determinants of drug
administration errors in the intensive care
unit. Crit Care Med 2002; 30:846 – 850
12. Valentin A, Capuzzo M, Guidet B, et al: Er-
rors in administration of parenteral drugs in
intensive care units: Multinational prospec-
tive study. BMJ 2009; 338:b814
13. Tissot E, Cornette C, Demoly P, et al: Medi-
cation errors at the administration stage in
an intensive care unit. Intensive Care Med
1999; 25:353–359
14. Kopp BJ, Erstad BL, Allen ME, et al: Medica-
tion errors and adverse drug events in an
intensive care unit: Direct observation ap-
proach for detection. Crit Care Med 2006;
34:415– 425
15. Bates DW, Leape LL, Cullen DJ, et al: Effect
of computerized physician order entry and a
team intervention on prevention of serious
medication errors. JAMA 1998; 280:
1311–1316
16. Paoletti RD, Suess TM, Lesko MG, et al: Us-
ing bar-code technology and medication ob-
servation methodology for safer medication
administration. Am J Health Syst Pharm
2007; 64:536 –543
17. Guerrero RM, Nickman NA, Jorgenson JA:
Work activities before and after implementa-
tion of an automated dispensing system.
Am J Health Syst Pharm 1996; 53:548 –554
18. Barker KN, Pearson RE, Hepler CD, et al:
Effect of an automated bedside dispensing
machine on medication errors. Am J Hosp
Pharm 1984; 41:1352–1358
19. Borel JM, Rascati KL: Effect of an automated,
nursing unit-based drug-dispensing device
on medication errors. Am J Health Syst
Pharm 1995; 52:1875–1879
20. Shirley KL: Effect of an automated dispens-
ing system on medication administration
time. Am J Health Syst Pharm 1999; 56:
1542–1545
21. Perras C, Jacobs P, Boucher M, et al: Tech-
nologies to reduce errors in dispensing and
administration of medication in hospitals:
Clinical and economic analyses. Technology
report number 121. Ottawa, Ontario, Canada,
Canadian Agency for Drugs and Technologies
in Health, 2009
22. Kuiper SA, McCreadie SR, Mitchell JF, et al:
Medication errors in inpatient pharmacy op-
2280 Crit Care Med 2010 Vol. 38, No. 12
erations and technologies for improvement.
Am J Health Syst Pharm 2007; 64:955–959
23. Barker K, MacConnell W: The problems of
detecting errors in hospitals. Am J Hosp
Pharm 1962; 19:360 –369
24. Flynn EA, Barker KN, Pepper GA, et al: Com-
parison of methods for detecting medication
errors in 36 hospitals and skilled-nursing
facilities. Am J Health Syst Pharm 2002;
59:436 – 446
25. Allan EL, Barker KN: Fundamentals of med-
ication error research. Am J Hosp Pharm
1990; 47:555–571
26. Venkatesh V, Davis F: A theoretical extension
of the technology acceptance model: Four
longitudinal field studies. Manage Sci 2000;
46:186 –204
27. Day G, Minichiello V, Madison J: Self-
reported perceptions of Registered Nurses
working in Australian hospitals. J Nurs
Manag 2007; 15:403– 413
28. Schein E: Kurt Lewin’s change theory in the
field and in the classroom: Notes toward a
model of managed learning. System Pract
Action Res 1996; 9:27– 47
29. Tissot E, Cornette C, Capellier G, et al: As-
sessment of medication errors: Methodolog-
ical details. Intensive Care Med 1999; 25:
1478
30. Roy PM, Durieux P, Gillaizeau F, et al: A
computerized handheld decision-support
system to improve pulmonary embolism di-
agnosis: A randomized trial. Ann Intern Med
2009; 151:677– 686
31. Barker K, Heller W, Brennan J, et al: The
development of a centralized unit dose dis-
pensing system Part IV: The pilot study—
Medication errors and drug losses. Am J
Hosp Pharm 1964; 2:609 – 625
32. Moyen E, Camire´ E, Stelfox H: Clinical re-
view: Medication errors in critical care. Crit
Care 2008; 12:208
33. Tamblyn R, Huang A, Kawasumi Y, et al: The
development and evaluation of an integrated
electronic prescribing and drug management
system for primary care. J Am Med Inform
Assoc 13:148 –159
34. Horsky J, Kuperman G, Patel V: Comprehen-
sive analysis of a medication dosing error
related to CPOE. J Am Med Inform Assoc
2205; 12:377–382
35. Hufschmid Thurnherr E, Lanker Klossner B,
Oertle M: Potential benefits of an automated
medication distribution system combined
with computerized physician order entry.
EJHP 2003; 6:26 –30
36. Rudis MI, Brandl KM: Position paper on crit-
ical care pharmacy services. Society of Crit-
ical Care Medicine and American College of
Clinical Pharmacy Task Force on Critical
Care Pharmacy Services. Crit Care Med 2000;
28:3746 –3750
37. Camire´ E, Moyen E, Stelfox HT: Medication
errors in critical care: Risk factors, preven-
tion and disclosure. CMAJ 2009; 180:
936 –943
2281Crit Care Med 2010 Vol. 38, No. 12
