Continuing Medical Education Article

Patient-initiated device removal in intensive care units: A national

prevalence study*
Lorraine C. Mion, PhD, RN, FAAN; Ann F. Minnick, PhD, RN, FAAN; Rosanne M. Leipzig, MD, PhD;
Catherine D. Catrambone, PhD, RN; Mary E. Johnson, PhD, RN

LEARNING OBJECTIVES
On completion of this article, the reader should be able to:
1. Define device removal rates.
2. Explain consequences of device removal.
3. Use this information in a clinical setting.
All authors have disclosed that they have no financial relationships with or interests in any commercial companies pertaining
to this educational activity.
Lippincott CME Institute, Inc., has identified and resolved all faculty conflicts of interest regarding this educational activity.
Visit the Critical Care Medicine Web site (www.ccmjournal.org) for information on obtaining continuing medical education credit.

Objective: Information is needed about patient-initiated device
removal to guide quality initiatives addressing regulations aimed at
minimizing physical restraint use. Research objectives were to de-
termine the prevalence of device removal, describe patient contexts,
examine unit-level adjusted risk factors, and describe consequences.
Design: Prospective prevalence.
Setting: Total of 49 adult intensive care units (ICUs) from a
random sample of 39 hospitals in five states.
Methods: Data were collected daily for 49,482 patient-days by
trained nurses and included unit census, ventilator days, restraint days,
and days accounted for by men and by elderly. For each device removal
episode, data were collected on demographic and clinical variables.
Results: Patients removed 1,623 devices on 1,097 occasions:
overall rate, 22.1 episodes/1000 patient-days; range, 0 –102.4. Sur-
gical ICUs had lower rates (16.1 episodes) than general (23.6 epi-
sodes) and medical (23.4 episodes) ICUs. ICUs with fewer resources
had fewer all-type device removal relative to ICUs with greater resources
(relative risk, 0.76; 95% confidence interval, 0.66–0.87) but higher self-
extubation rates (relative risk, 1.27; 95% confidence interval, 1.07–1.52).
Men accounted for 57% of the episodes, 44% were restrained at the
time, and 30% had not received any sedation, narcotic, or psychotropic
drug in the previous 24 hrs. There was no association between rates of
device removal with restraint rates, proportion of men, or elderly. Self-
extubation rates were inversely associated with ventilator days (rs ⴝ
ⴚ0.31, p ⴝ .03). Patient harm occurred in 250 (23%) episodes; ten
incurred major harm. No deaths occurred. Reinsertion rates varied by
device: 23.5% of surgical drains to 88.9% of monitor leads. Additional
resources (e.g., radiography) were used in 58% of the episodes.
Conclusion: Device removal by ICU patients is common, resulting
in harm in one fourth of patients and significant resource expendi-
ture. Further examination of patient-, unit-, and practitioner-level
variables may help explain variation in rates and provide direction for
further targeted interventions. (Crit Care Med 2007; 35:2714–2720)
KEY WORDS: therapy disruption; device removal; treatment in-
terference; self-extubation; physical restraint

P rotection of therapeutic de-
vices from patient-initiated re-
moval is a major reason for the
use of physical restraints in in-
tensive care units (ICUs) (1). Premature
discontinuation of technologically complex
therapies (e.g., endotracheal tubes for me-
chanical ventilation, central venous cathe-
ters) may result in serious harm, injury, or
death (1, 2). A major challenge for ICU
clinicians is balancing the need to prevent
patients from removing devices and meet-
ing the federal regulations of the Centers
for Medicare and Medicaid and the national
accrediting standards of the Joint Commis-

*See also p. 2859.
Director, Nursing Research and Geriatric Nursing, Metro-
Health Medical Center, Cleveland, OH (LCM); Senior Associ-
ate Dean—Research, Julia Eleanor Chenault Professor of
Nursing, Vanderbilt University School of Nursing, Nashville,
TN (AFM); Vice Chair for Education, Gerald and May Ellen
Ritter Professor of Geriatrics, Professor, Brookdale Depart-
ment of Geriatrics and Adult Development, Mount Sinai
School of Medicine, New York City, NY (RML); Assistant
Professor, Adult Health Nursing (CDC), Associate Professor
(MEJ), College of Nursing, Rush University, Chicago, IL.
Supported by grant 1R01AG19715-01 from the National
Institute on Aging, National Institutes of Health.
The contents herein are solely the responsibility of the
authors and do not necessarily represent the official views of
the National Institute on Aging, National Institutes of Health.
For information regarding this article, E-mail:
lmion@metrohealth.org
Copyright © 2007 by the Society of Critical Care
Medicine and Lippincott Williams & Wilkins
DOI: 10.1097/01.CCM.0000291651.12767.52

2714 Crit Care Med 2007 Vol. 35, No. 12

sion of Healthcare Organizations restrict-
ing the use of physical restraints (3, 4).
Although maintenance of therapeutic
devices is a primary reason for use of
physical restraints in ICUs, little is known
regarding the rate of patient-initiated de-
vice removal or subsequent harm. Since
the 1970s, a number of investigators have
focused on patients’ self-extubation from
mechanical ventilation. In the past decade,
studies from the United States (5–7), Eu-
rope (8 –12), and Asia (13, 14) have re-
ported incidence rates of self-extubation
ranging from 0.3% to 14.3% and preva-
lence rates of 2.0 to 25.6/1000 ventilator-
days. All but one study (9) was limited to a
single hospital. One third or more of the
self-extubation events occurred despite use
of wrist restraints (5, 9, 12).
Patient removal of devices other than
endotracheal tubes (e.g., intraaortic bal-
loon pumps) may have similar levels of
life-threatening harm. On the other
hand, devices such as peripheral intrave-
nous catheters may be more likely to
cause minor to no harm in patients but
can consume significant staff time or
costly resources (6). Three studies re-
ported on patient-initiated removal of ad-
ditional therapeutic devices in ICU set-
tings (6, 10, 15). As with self-extubation
studies, rates varied greatly, from ⬍50 to
125 episodes/1000 patient-days. These
studies of patient-initiated removal of
multiple devices were limited to one or
two hospitals (6, 10, 15) or short obser-
vational periods (6). Information remains
sparse on the extent, circumstances, or
outcomes of patient-initiated therapy dis-
ruption events in ICUs. Thus, the ability
to develop appropriate prescriptive guide-
lines addressing physical restraint use is
hampered (16).
Given the regulatory and accrediting is-
sues related to physical restraints, larger
studies are needed in the United States to 1)
determine the extent and consequences of
the problem and 2) identify characteristics
of patients and of units that are associated
with removal of all types of devices. This
information would be useful to healthcare
professionals, administrators, and organi-
zations as they address Centers for Medi-
care and Medicaid regulations and Joint
Commission of Healthcare Organizations
standards to minimize physical restraint
use in their ICUs. Accordingly, we con-
ducted a multisite prospective prevalence
study to 1) describe the rate of patient-
Crit Care Med 2007 Vol. 35, No. 12
initiated removal of all devices and self- census of ⬎99. Hospitals were selected at ran-
extubation episodes in ICUs, 2) describe dom from six metropolitan areas that repre-
patient contexts related to device removal sented the Western, Southern, Midwestern,
episodes, 3) examine patient-adjusted risk and Northeastern regions of the United States.
The random selection was stratified to include
factors at the unit level, and 4) describe the
at least one large teaching hospital. Metropol-
consequences of patient-initiated device re- itan areas were used because of the concen-
moval to the patient and staff. tration of the population in these areas; the
hospital volume level was chosen because al-
most 90% of the U.S. population receives care
METHODS in hospitals of this size. The sampling process
Sample and Settings. The sample con- is described elsewhere (17).
sisted of 49 adult ICUs from 39 nonfederal, Variables. Therapeutic devices were de-
nonpsychiatric hospitals with an average daily fined as those used to monitor or treat dys-
function, disease, or injury (Table 1). A ther-
Table 1. Variables assessed for description and outcomes associated with patient-initiated device
removal episodes
Therapeutic devices Patient risk-adjusted variables
Peripherally inserted intravenous or central Physical restraint days
catheters Ventilator days
Central venous or pulmonary artery Proportion of days by men
flotation catheters Proportion of days by elderly
Indwelling bladder catheters ICU type: general, medical, surgical
Oxygen mask/nasal cannula ICU cluster category (0 or 1)a
CPAP/BiPAP
Endotracheal tube/tracheostomy tube
Percutaneous endoscopic gastrostomy tube
Surgical drains
Dressings
Arterial catheters
External ventricular drains/ICP bolts
Monitor leads
Traction devices
Nasogastric/feeding tubes
Patient variablesb Patient consequencesc
Demographics: age, sex Increased agitation or confusion
Cognition: alert, anxious/agitated, oriented Delay in therapy (e.g., missed medication)
Medications Previous 24 hrs Vocal cord damage/paralysis
Sedative-hypnotic Respiratory distress/failure
Benzodiazepine Hypotension (SBP ⬍ 90)
Neuromuscular blocking agents Hemorrhage/exsanguinations
Neuroleptic Bleeding, non-life-threatening
Narcotic Aspiration
Physically restrained at time Pneumothorax
Sedation/analgesia protocol in use at time Urinary retention
Additional resources/procedure as Soft-tissue injury
consequence of device removalc Laceration
Radiographs Death
New or more sedation Staff consequences
Sutures Unprotected exposure to blood
Surgical or other procedure Exposed to violence (hitting, kicking, etc.)
Cast/splinting Physically harmed
Increased monitoring/surveillance
New or more physical restraint
Laboratory work
New dressings, compression, skin barrier
Blood transfusion
Transfer to another unit
Specialty consultation
CPAP/BiPAP, continuous positive airway pressure/bilevel positive airway pressure; ICP, intracranial
pressure; ICU, intensive care unit, SBP, systolic blood pressure.
a
ICUs were measured on ⬎100 variables examining environmental, administrative, and organiza-
tional factors. Using cluster analysis, ICUs clustered into one of two groups. Cluster 1 had greater
personnel resources, technology, supplies, and environmental layout conductive to heightened sur-
veillance; bdata collected only on those patients who removed one or more devices; cadverse conse-
quences as direct result of the patient-initiated device removal or disruption.
2715
apy disruption episode was defined as any
patient-initiated removal of one or more de-
vices. The act could be purposeful (i.e., patient
actively sought to rid himself/herself of the
device) or accidental, witnessed or nonwit-
nessed. Device removals due to staff care or
manipulations were not included. Rate of
therapy disruption was measured as the
(number of therapy disruption episodes/total
patient-days) ⫻ 1000.
For patients who removed a therapeutic de-
vice, information was collected on demographic
characteristics, cognition, sedating medications,
presence of physical restraint, and episode time.
Patient consequences were any documented ad-
verse events that occurred as a direct result of
the device removal or displacement, such as
bleeding (Table 1). Severity of harm as a direct
result of the episode was characterized as none,
mild, moderate, severe, or death. Resource con-
sequences were any additional resources, such
as procedures or treatments, that were utilized
as a direct consequence of the device removal or
displacement. Staff consequences included un-
protected exposure to body fluids or blood, ex-
posure to violence (e.g., hitting, kicking), and
sustaining physical harm.
Unit-level risk adjusted variables included
the proportion of days accounted for by men,
by older adults (age ⱖ65), and by physically
restrained patients. Age and sex have been
identified as potential risk factors for agitation
or physical restraint in previous studies (1, 12,
18, 19). To adjust for self-extubation risk, we
collected the number of ventilator days for each
ICU. Physical restraint was defined as any device
attached to the patient to limit voluntary move-
ment and included wrist, chest, and waist re-
straints, mitts, elbow splints, and sheets. Bedside
rails were excluded. Units were classified into
two categories based on cluster analysis of ⬎100
variables examining environmental factors (e.g.,
distances walked from bedside to supplies, com-
puter support) and organizational factors (e.g.,
presence of intensivists, staffing ratios, use of
sedation and analgesic protocols). In brief, ICUs
in cluster 1 had greater opportunity for patient
surveillance in terms of unit design, placement
of supplies, number of available interdisciplinary
personnel, arrangements for family visitation,
and technology for monitoring as compared
with ICUs in cluster 0. There were no differences
in nurse staffing ratios between clusters; all ICUs
maintained a 1:1.5 to 1:2 nurse-to-patient ratio.
Data Collection Procedures. Data collec-
tion occurred from mid-2003 to early 2005. A
total of 31 ICUs collected data for 90 days, and
the remainder collected for ⬍90 days, for a
total of 49,482 observed patient-days. Length
of participation varied because of institutional
burden, accreditation preparation, and the
conduct of other research on the study units.
A standardized data collection instrument that
was devised, tested, and implemented in an
earlier study (15) was used by trained ICU
nurses, who collected data daily at preset es-
tablished times (e.g., 1000 –1200). At each site,
a co-investigator conducted training using
2716
standardized procedural manuals, and interra-
ter reliability of ⬎95% was established before
data collection. One to two subsequent inter-
rater reliability checks were conducted at each
site, and only data from periods in which in-
terrater reliability remained at ⱖ95% were
utilized in the analysis.
Nurse-initiated notification of patient re-
moval of therapeutic devices was ascertained
through specially designed, 5- by 8-inch, bright
yellow notification cards that included the pa-
tient’s name, date, time, and a check-off list of
therapeutic devices and staff consequences. Staff
nurses were instructed to complete a card for
any patient with a therapy disruption episode
and to drop it in a designated locked box. During
the regular data collection rounds, the ICU
nurse data collectors checked the locked box for
notification cards, briefly interviewed the pa-
tient’s nurse, and conducted a chart audit of
nursing and medical notes for the previous 24
hrs. Because of feasibility issues, ICUs with ⬎20
beds conducted daily chart audits on 25–50% of
the charts to ascertain additional terminations.
A pilot study examining proportion of events
captured by each of the three methods (notifica-
tion card, chart documentation, nurse inter-
view) revealed the notification card captured
97% of events. Presence of patient-initiated de-
vice removal or displacement by any one of the
three methods was considered a positive identi-
fication. A patient could have removed more
than one device per episode. Only two types of
devices, monitor leads and oxygen mask/nasal
cannula, were counted as one episode per shift,
even if the patient removed these devices mul-
tiple times. This was done to minimize reporting
burden.
Unit-level data collected included the daily
ICU census, number of men, number of pa-
tients aged ⱖ65 yrs, number of patients me-
chanically ventilated, and number of patients
in physical restraints. Chart audits were con-
ducted on patients who removed or disrupted
devices to determine the patient characteris-
tics at the time of the episode, patient conse-
quences, and subsequent healthcare resources
utilized. Completed data instruments were
mailed to a central data collection office for
data entry.
Administratively mediated variables of staff-
ing levels, medical and nursing care models,
environmental design and layout, resources and
supplies, and ICU practices of sedation and an-
algesia were collected by one of the study co-
investigators in conjunction with a designated
nurse administrator at each hospital site. A stan-
dardized instrument and procedural manual
from previous studies by one of the co-investi-
gators were used (20). Interrater reliability
checks demonstrated 98% agreement between
this co-investigator and the other co-investiga-
tors.
The study procedures were reviewed and
approved by the institutional review boards at
each participating site and at each of the in-
vestigators’ institutions; the need for written
informed consent was waived.
Analyses. Data were analyzed using Latent
Gold 4.0 for cluster analysis and SAS for Win-
dows version 9.1 (Cary, NC) for all other anal-
yses. Missing data for the unit-level risk-
adjusted variables, which accounted for
⬍0.1%, were addressed using median values.
Descriptive statistics were conducted on all
variables. Spearman’s correlation coefficients
were used between therapy disruption rates
and unit-level risk-adjusted variables of pro-
portion of restraint days, ventilator days, men,
and elderly.
Cluster analysis was conducted using en-
vironmental, capital, and labor input variables
and produced a two-cluster solution for ICU
units. In 80% of the ICUs, the cluster assign-
ment was exactly 0 or 1 (statistical identifica-
tion of a unit in/not in a cluster). In the
remaining cases, the estimation of assignment
varied ⬍0.04 from either 0 or 1, indicating
clear cluster assignment. Fifty-one percent of
the ICUs were assigned to cluster 1 (i.e., ICUs
with more resources, geographical layout con-
ducive to surveillance) and the remainder to
cluster 0. Between-group comparisons were
made using Wilcoxon’s rank-sum test and
Kruskal-Wallis test.
Poisson regression was conducted with the
therapy disruption episode rate as the out-
come variable, and with cluster type, ICU type
(surgical, nonsurgical), proportion of restraint
days, ventilator days, men, and elderly as the
independent variables. Subgroup analyses
were conducted on ventilator self-extubation,
as described above, for comparison with pre-
vious studies.
To determine whether length of time of
data collection influenced unit rates, we exam-
ined the monthly rates of those with 90 days of
data collection (n ⫽ 31). Only three units
demonstrated a pattern, each having a sus-
tained decrease in therapy disruption rates
from month 1 to month 3. All three units had
used their monthly rates as feedback for qual-
ity improvement purposes. To determine
whether the ICUs that collected data for ⬍90
days (n ⫽ 18) had rates different from those
that conducted the full 3 months of data col-
lection, the first month’s rates were compared
between the two groups. There were no sig-
nificant differences. Given these findings, we
used the entire data set in calculating the
overall therapy disruption rate.
RESULTS
ICU Profile
The 49 ICUs ranged in size from eight
to 42 beds, 26 (53%) were general adult
ICUs that provided care to a mixed pop-
ulation (medical, neurologic, surgical,
cardiac, and pulmonary), 12 (24%) were
medical ICUs, and 11 (22%) were surgical
ICUs. The study hospitals ranged in size
(⬍150 to ⬎1,000 beds), and 69% were
Crit Care Med 2007 Vol. 35, No. 12
nonprofit, 18% were for-profit, and 13% Table 2. Patient therapy disruption episode for all devices and self-extubation ratesa by intensive care
were governmentally controlled (state, unit (ICU) type and cluster
county, city).
Mean ⫾ SD Range Median

Rates of Device Removal Therapy disruption episodes:

Episodes all devices
By type of ICU
General ICUs (n ⫽ 26) 24.6 ⫾ 22.6 0–102.1 18.7
There were 1,097 episodes of patients Medical ICUs (n ⫽ 12) 24.0 ⫾ 22.1 0–71.0 14.0
removing one or more therapeutic de- Surgical ICUs (n ⫽ 11) 14.9 ⫾ 13.7 0–43.3 13.3
vices, yielding an overall prevalence of Kruskal-Wallis Test: chi-square 2.61, df ⫽ 2, p ⫽ .27
22.1 therapy disruption episodes/1000 pa- By clusteringb of ICUs
tient-days (range, 0 –102.4) (Table 2). Cluster 0 (n ⫽ 23) 14.9 ⫾ 12.7 0–43.3 11.9
Cluster 1 (n ⫽ 24) 28.2 ⫾ 24.8 3.9–102.1 18.8
Surgical ICUs had an overall rate of 16.1, (2 ICUs with missing data) Wilcoxon’s rank-sum statistic 449.00, Z ⫽ ⫺2.18, p ⫽ .029
medical ICUs of 23.4, and general ICUs of Ventilator self-extubation
23.6. There were no significant differ- episodes
ences in rates of therapy disruption by By type of ICU
type of ICU. ICUs in the cluster with General ICUs (n ⫽ 26) 11.0 ⫾ 7.7 0–26.0 10.3
Medical ICUs (n ⫽ 12) 18.2 ⫾ 18.1 0.98–60.0 11.6
fewer resources had significantly lower Surgical ICUs (n ⫽ 11) 6.6 ⫾ 9.5 0–30.1 2.3
rates of therapy disruption (overall rate of Kruskal-Wallis Test: chi-square 5.78, df ⫽ 2, p ⫽ .056
16.6) as compared with ICUs in the clus- By clusteringb of ICUs
ter with more resources (overall rate of Cluster 0 (n ⫽ 23) 10.8 ⫾ 13.0 0–60.0 8.1
Cluster 1 (n ⫽ 24) 12.0 ⫾ 11.3 0–48.3 9.8
25.7) (p ⫽ .029). (2 ICUs with missing data) Wilcoxon’s rank-sum test statistic 524.0, Z ⫽ ⫺0.59, p ⫽ .56

Rates of Self-Extubation from

a
Prevalence rates calculated as (no. of episodes/total patient days) ⫻ 1000; bclustering done on
⬎100 variables (environmental, administrative, organizational). ICUs in cluster 1 had greater numbers
Ventilators of interdisciplinary personnel, more supplies and computer technological support, and greater sur-
veillance of patients based on unit design.
There were 181 patient-initiated self-
extubation episodes (endotracheal tudes or
tracheostomy tubes) during 18,308 ventila-
tor days, yielding an overall rate of 9.9 self-
Table 3. Frequency of removed devices with reinsertion rates (n ⫽ 1,097 episodes)a
extubations/1000 ventilator-days (Table 2).
Five units reported no self-extubation epi- Device Frequency (%)a Reinsertion Rate, %b
sodes. Rates ranged from 0 to 60.0 self-
extubations/1000 ventilator days. Self- Nasogastric tubes 317 (28.9) 65.3
extubation rates varied by type of ICU with Oxygen masks/nasal cannulac 258 (23.5) 85.7
highest rates found among medical ICUs Peripheral intravenous catheters 228 (20.8) 80.3
Monitor leadsc 217 (19.8) 88.9
(overall rate of 11.8), followed by general Endotracheal tubes 181 (16.5) 48.9
ICUs (overall rate of 10.4) and lowest rates Central venous catheters/pulmonary 78 (7.1) 43.6
among surgical ICUs (overall rate of 6.8) artery flotation catheters
(p ⫽ .056). No significant differences Indwelling bladder catheters 67 (6.1) 76.1
occurred by cluster assignment. Dressings 56 (5.1) 69.6
Arterial catheters 45 (4.1) 28.9
CPAP/BiPAP 33 (3.0) 72.7
Types and Frequency of Surgical drains 17 (1.5) 23.5
Percutaneous endoscopic 10 (0.9) 40
Devices gastrostomy tubes
External ventricular drains 6 (0.5) 33
There were 1,623 devices disrupted dur-
ing the 1,097 episodes. Table 3 displays the CPAP/BiPAP, continuous positive airway pressure/bilevel positive airway pressure.
frequency of disrupted devices in descend- a
More than one device could be removed during an episode; percentages do not add to 100;
ing order with their corresponding reinser- b
reinsertion rate ⫽ (number of reinsertions of a device/frequency of the device) ⫻ 100; cdevice removal
tion rates. The most frequently disrupted counted once per shift, even if patient removed numerous times during a shift.
devices were nasogastric tubes, oxygen
masks/nasal cannula, and peripheral intra-
venous catheters (each occurring in ⱖ20% (mean, 63.2 yrs). Men accounted for 625 anxiety at the time of the episode,
of the episodes). Reinsertion rates varied (57%) of the episodes. At the time of the whereas 27% were noted to be lethar-
from 23.5% for surgical drains to 88.9% for device removal, 494 patients (45%) gic. Within 24 hrs before the episode,
monitor leads. were in physical restraints, and 505 786 (70%) had received at least one of
(46%) of the episodes occurred during the following types of medication: nar-
Patient Characteristics the day shift. Nineteen (1.7%) of the cotic (n ⫽ 480), benzodiazepine (n ⫽
therapy disruptions occurred during a 380), other sedating agent (n ⫽ 208),
Patients with one or more device re- patient fall. More than half (58%) the neuroleptic (n ⫽ 152), or neuromuscu-
movals ranged in age from 11 to 98 yrs patients had documented agitation or lar blocking agent (n ⫽ 18).

Crit Care Med 2007 Vol. 35, No. 12 2717

Association of Unit-Level
Risk-Adjusted Factors
All Device Removal Episodes. A signif-
icant inverse relationship was found be-
tween units’ therapy disruption rates and
proportion of ventilator days (rs ⫽ ⫺0.31,
p ⫽ .03); no significant univariate corre-
lations were found with proportion of re-
straint days, days accounted for by men,
or elderly patients. Poisson regression
was conducted with therapy disruption
rate as the outcome variable and ICU type
(surgical, 0; medical/general, 1), ICU
cluster (fewer resources, 0; more re-
sources, 1), proportion of physical re-
straint days, ventilator days, and days ac-
counted for by men and elderly patients
as the predictor variables. After control-
ling for the effects of multiple variables,
only cluster assignment remained signif-
icant. Among ICUs with fewer resources
(cluster 0), therapy disruption rates de-
creased relative to ICUs with more re-
sources (cluster 1) (estimate of ⫺0.28, SE
0.07, chi-square ⫽ 15.31, p ⬍ .0001, rel-
ative risk ⫽ 0.76 with 95% confidence
interval: 0.66 – 0.87). To test the assump-
tion that ICU resources may have influ-
enced nurses’ reporting or documenta-
tion of episodes, we ran the same Poisson
model but excluded any episode that
solely involved removal of telemetry leads
or removal of oxygen mask or nasal can-
nula. In this latter model, cluster assign-
ment was nonsignificant.
Self-Extubation Rates. No significant
univariate associations were found be-
tween ICU self-extubation rates and unit-
level adjusted variables of proportion of
physical restraints, ventilator days, days
accounted for by men, or by elderly pa-
tients. A Poisson regression model dem-
onstrated that after controlling for the
multiple variables, only cluster assign-
ment and type of ICU remained signifi-
cant independent factors associated with
rates of self-extubation. ICUs with fewer
resources (cluster 0) had a self-extuba-
tion rate that increased relative to ICUs
with greater resources (cluster 1) (esti-
mate of 0.24, SE 0.09, chi-square ⫽ 7.08,
p ⫽ .008, relative risk ⫽ 1.27 with 95%
confidence interval: 1.07–1.52). Surgical
ICUs had self-extubation rates that de-
creased relative to nonsurgical ICUs (es-
timate of ⫺0.60, SE 0.14, chi-square ⫽
17.18, p ⬍ .0001, relative risk ⫽ 0.55
with 95% confidence interval: 0.42–
0.73).
2718
Consequences of Patient-
Initiated Device Removal
Harm or injury as a direct result of the
device removal episode occurred in 250
(23%) of the episodes. Of these, 174
(69.6%) were minor (e.g., bruising at the
intravenous insertion site), 66 (26.4%)
were moderate (e.g., required therapy but
harm not life threatening), and ten
(4.0%) were major (i.e., requiring major
medical therapy or surgical therapy). The
ten cases with major harm were re-
viewed. All involved either an endotra-
cheal tube extubation (n ⫽ 6) or a central
catheter discontinuation (n ⫽ 5). All re-
quired reinsertion of the devices; two pa-
tients required surgical procedures and
one required blood transfusions.
Staff consequences occurred in 110 of
the episodes (10%). The most common
consequence was unprotected exposure
to blood or bodily fluids (n ⫽ 74), fol-
lowed by exposure to violence (hit,
kicked; n ⫽ 30), physical injury (n ⫽ 15),
and other (n ⫽ 11).
Additional treatments or procedures
(other than reinsertion) as a direct con-
sequence of the device removal occurred
in 58% of the episodes. They were: new or
additional physical restraint (n ⫽ 320),
new or additional sedation (n ⫽ 219),
radiography (n ⫽ 140), increased moni-
toring/surveillance (n ⫽ 84), surgical
procedure (n ⫽ 38), new or additional
dressings such as compression (n ⫽ 69),
consultation with specialty service (n ⫽
42), or other (n ⫽ 133).
DISCUSSION
Prevention of patient-initiated removal
of therapeutic devices is a major ICU pa-
tient safety initiative. The main findings of
this multisite, national prevalence study
were: 1) patient-initiated device removal
and self-extubation events are not uncom-
mon but vary considerably among ICUs and
by types of ICUs; 2) the types of ICUs (e.g.,
surgical vs. medical) and the degree of ICU
resources are associated with therapy dis-
ruption and self-extubation rates; 3) device
disruption frequency is likely a result of
frequency of use, but reinsertion rates by
devices vary considerably; and 4) patient-
initiated removal of devices results in some
degree of harm in approximately one out of
four episodes, with most minor in nature.
Our rates are similar to a study involv-
ing six ICUs at two hospitals, (15) as well
as the study by Carrion et al (10). Fraser et
al. (6) reported a rate of 125.6 episodes/
1000 patient-days, a rate greater than our
highest ranking ICU. This may be a func-
tion of design differences; they followed 36
patients in one ICU for 1 month. Others
have observed similar differences between
surgical ICU and medical ICU populations
in self-extubation rates (8, 13, 18, 21–23).
The extent to which variation is a function
of patient type or processes (e.g., greater
predictability in admissions and transfers to
surgical ICUs as compared with medical
ICUs) remains to be determined (24).
Patient characteristics of agitation,
disorientation, or presence of physical re-
straint at the time of the device removal
have been associated with higher rates of
self-extubation (5, 6, 9, 12, 18, 19). In
30% of the episodes in this study, patients
had not received any sedation or analgesia
in the preceding 24 hrs. Several studies and
reviews have demonstrated the wide varia-
tion in ICU practices of sedation and anal-
gesia (25–29). Agitation or lethargy can be
a reflection of the patient’s illness but also
of inadequate or inappropriate dosing of
sedative and analgesic drugs (19, 30 –32).
This large-scale, observational preva-
lence study prohibited the collection of data
at the individual patient level; however, sev-
eral patient risk factors were adjusted at the
aggregate unit level. No correlation existed
between unit-level physical restraint rates
and rates of all types of device removal or of
self-extubation, a finding similar to some
studies (13, 15) but not others (7, 10, 33).
Preponderance of young or old patients or
of men also does not have an association
with overall device removal or self-extuba-
tion.
The methodologic approach we used
to measure unit-level resources (includ-
ing nurse staffing levels) yielded no de-
finitive data on whether these resources
affected therapy disruption rates. In a
multivariate analysis, we found that ICUs
with greater resources had lower self-
extubation rates but higher overall ther-
apy disruption when all minor and major
devices were included. Even when remov-
ing minor devices, such as telemetry
leads, results did not match those of self-
extubation only. These mixed results mir-
ror what has been occurring in studies
examining the influence of nursing re-
sources and staffing on patient outcomes.
That is, nursing staff levels and resources
sometimes explain outcomes and some-
times do not (34). Indeed, the effect of
nurse staffing levels on self-extubation
rates has been inconclusive (9, 12, 35,
36). Introducing new study designs and
further development of effective perfor-
Crit Care Med 2007 Vol. 35, No. 12
mance measurement systems are war-
ranted to better understand and monitor
the degree to which hospital resources,
including nurse staffing levels, affect
quality patient outcomes (34).
No deaths occurred during this study
as a direct result of the patient’s removal
of a device. The morbidity rate in our study
was 23% or five harmful outcomes/1000
patient-days, and the majority of harm was
minor. Significant harm was incurred by
ten patients, six of whom self-extubated and
required reintubation and five who pulled
out central venous catheters. These mor-
bidity findings are similar to other studies
(5, 7, 8, 15). Others have reported deaths
after self-extubation (9, 13, 14).
The proportion of the devices requiring
reinsertion varied by device type. Those
that are noncomplex and easily replaced
(e.g., nasal cannula or monitor leads) were
replaced more frequently than those that
are more difficult or complex to reinsert.
Others have postulated whether these de-
vices, such as endotracheal tubes, are uti-
lized too long (11, 12, 37). This may con-
tribute to physical restraint use, which in
turn contributes to delirium or agitation
and subsequent patient removal of the de-
vices.
There are important consequences to
the staff and the institution: 10% of the
episodes resulted in some degree of harm
to the nursing staff. In addition to rein-
sertion of devices, more than half of the
episodes resulted in further nursing or
medical treatments. Fraser et al. (6) dem-
onstrated that the cost to a single ICU
due to these episodes can exceed a quar-
ter million dollars a year.
Several study limitations warrant
mention. First, we relied on nurses’ re-
ports or clinician documentation for pa-
tient removal of devices. Both methods
have been shown to underreport events
(38, 39), which may have led to an esti-
mate lower than actual removal rates. As
with any study utilizing chart abstrac-
tion, we were limited by the quality of the
documentation. Second, ICUs either did
not use standardized sedation scales or
varied in the ones used. Thus, our ability
to rate specific patients’ behaviors at the
time of the disruption, an important
characteristic found in case control and
cohort studies on self-extubation, was
limited. Third, this was a multisite de-
scriptive study, not a cohort study; thus,
we could not determine characteristics at
the patient level predictive of device re-
moval but relied on unit-level character-
istics. Nor could we accurately assess the
Crit Care Med 2007 Vol. 35, No. 12
temporal relationship between sedation
and physical restraint use with therapy
disruption. Last, the study design of daily
surveillance may have affected practices
and subsequent rates of device removal.
In summary, this large, multisite study
provides data for benchmarking purposes.
Patient-initiated removal of therapeutic de-
vices is not uncommon and is an important
target for quality improvement and patient
safety initiatives. Indeed, rates of device re-
moval, either all types or specific types, may
serve as a surrogate marker for quality of
care for appropriate identification and
management of delirium, weaning proto-
cols, or sedation and analgesia protocols in
the ICU setting. Further examination of
patient-level, practitioner-level, and unit-
level (environmental and administrative)
variables may help explain variation in rates
of device removal and provide direction for
further targeted interventions.
REFERENCES
1. Minnick AF, Mion LC, Leipzig R, et al: Prev-
alence and patterns of physical restraint use
in the acute care setting. J Nurs Adm 1998;
28:19 –24
2. Mion LC, Minnick A, Palmer R, et al: Physical
restraint use in the hospital setting: Unre-
solved issues and directions. Milbank Q 1996;
74:411– 433
3. Centers for Medicare and Medicaid Services:
Medicare and Medicaid Programs; Hospital
Conditions of Participation: Patients’ Rights;
Final Rule. 42CFR Part 482. Federal Register
Volume 71, Number 236. Washington, DC,
US Government Printing Office, December 8,
2006
4. Joint Commission of Accreditation of Health-
care Organizations: The Joint Commission
2006 Comprehensive Accreditation Manual for
Hospitals: The Official Handbook (CAMH).
Oakbrook Terrace, IL, Joint Commission of Ac-
creditation of Healthcare Organizations, 2006
5. Christie JM, Dethlefsen M, Cane RD: Un-
planned endotracheal extubation in the in-
tensive care unit. J Clin Anesthesia 1996;
8:289 –293
6. Fraser GL, Riker RR, Prato BS, et al: The
frequency and cost of patient-initiated device
removal in the ICU. Pharmacotherapy 2001;
21:1– 6
7. Frezza EE. Carleton GL, Valenziano CP: A
quality improvement and risk management
initiative for surgical ICU patients: A study of
the effects of physical restraints and sedation
on the incidence of self-extubation. Am J Med
Qual 2000; 15:221–225
8. Moons P, Sels K, De Becker W, et al: Devel-
opment of a risk assessment tool for deliber-
ate self-extubation in intensive care patients.
Intensive Care Med 2004; 30:1348 –1355
9. Boulain T: Unplanned extubations in the
adult intensive care unit: A prospective mul-
ticenter study. Am J Respir Crit Care Med
1998; 157:1131–1137
10. Carrion MI, Ayuso D, Marcos M, et al: Acci-
dental removal of endotracheal and nasogas-
tric tubes and intravascular catheters. Crit
Care Med 2000; 28:63– 66
11. Betbese AJ, Perez M, Bak E, et al: A prospec-
tive study of unplanned endotracheal extuba-
tion in intensive care unit patients. Crit Care
Med 1998; 26:1180 –1186
12. Chevron V, Menard JF, Richard JC, et al:
Unplanned extubation: Risk factors of devel-
opment and predictive criteria for reintuba-
tion. Crit Care Med 1998; 26:1049 –1053
13. Kapadia FN, Bajan KB, Raje KV: Airway ac-
cidents in intubated intensive care unit pa-
tients: An epidemiologic study. Crit Care
Med 2000; 28:659 – 664
14. Chiang AA, Lee KC, Lee JC, et al: Effective-
ness of a continuous quality improvement
program aiming to reduce unplanned extu-
bations: A prospective study. Intensive Care
Med 1996; 22:1269 –1271
15. Mion LC, Fogel J, Sandhu S, et al: Outcomes
following physical restraint reduction pro-
grams in two acute care hospitals. Jt Comm
J Qual Improv 2001; 27:605– 618
16. Maccioli GA, Dorman T, Brown BR, et al:
Clinical practice guidelines for the mainte-
nance of patient physical safety in the inten-
sive care unit: Use of restraining therapies.
American College of Critical Care Medicine
Task Force 2001–2002. Crit Care Med 2003;
31:2665–2676
17. Minnick AF, Mion LC, Johnson ME, et al:
Prevalence and variation of physical restraint
use in acute care settings in the US. J Nurs
Scholarsh 2007; 39:30 –37
18. Atkins PM, Mion LC, Mendelson W, et al:
Characteristics and outcomes of patients
who self-extubate from ventilatory support.
Chest 1997; 112:1317–1323
19. Woods JC, Mion LC, Connor JT, et al: Severe
agitation among ventilated medical intensive
care unit patients: Frequency, characteristics
and outcomes. Intensive Care Med 2004; 30:
1066 –1072
20. Minnick AF, Roberts MJ, Young WB, et al:
What influences patients’ reports of three
aspects of hospital services? Med Care 1997;
35:399 – 409
20a.Vermunt JK, Magidson J: Hierarchical mix-
ture models for nested data structures. In:
Classification: The Ubiquitous Challenge.
Weihs C, Gaul W [Eds]. Heidelberg, Springer,
2005.
21. Maguire GP, DeLorenzo LJ, Moggio RA: Un-
planned extubation in the intensive care
unit: A quality-of-care concern. Crit Care
Nurs Q 1994; 17:40 – 47
22. Taggart JA, Lind MA: Evaluating unplanned
endotracheal extubations. Dimens Crit Care
Nurs 1994; 13:114 –120
23. Grap MJ, Glass C, Lindamood MO: Factors
related to unplanned extubation of endotra-
cheal tubes. Crit Care Nurse 1995; 15:57– 65
24. Moons P, Ferdinande P: Reasons for the
higher incidence of unplanned extubation in
2719
 medical ICUs than in surgical ICUs. Letter.
Intensive Care Med 2004; 30:2290
25. Bair N, Bobek MB, Hoffman-Hogg L, et al:
Introduction of sedative, analgesic, and neuro-
muscular blocking agent guidelines in a med-
ical intensive care unit: Physician and nurse
adherence. Crit Care Med 2000; 28:707–713
26. Egerod I, Christensen BV, Johansen L:
Nurses’ and physicians’ sedation practices in
Danish ICUs in 2003: A national survey. In-
tensive Crit Care Nurs 2006; 22:22–31
27. Mehta S, Burry L, Fischer S, et al: Canadian
survey of the use of sedatives, analgesics, and
neuromuscular blocking agents in critically
ill patients. Crit Care Med 2006; 34:374 –380
28. Siegel MD: Management of agitation in the
intensive care unit. Clin Chest Med 2003;
24:713–725
29. Weinert CR, Chlan L, Gross C: Sedating crit-
ically ill patients: Factors affecting nurses’
2720
delivery of sedative therapy. Am J Crit Care
2001; 10:156 –165
30. Arroliga AC, Shehab N, McCarthy K, et al:
Relationship of continuous infusion loraz-
epam to serum propylene glycol concentra-
tion in critically ill adults. Crit Care Med
2004; 32:1709 –1714
31. Tuohy KA, Nicholson WJ, Schiffman F: Agi-
tation by sedation. Lancet 2003; 361:308
32. Tung A, Tadimeti L, Caruana-Montaldo B, et
al: The relationship of sedation to deliberate
self-extubation. J Clin Anesth 2001; 13:24 –29
33. Tominaga GT, Rudzwick H, Scannell G, et al:
Decreasing unplanned extubations in the
surgical intensive care unit. Am J Surg 1995;
170:586 –590
34. Needleman J, Kurtzman ET, Kizer KW: Per-
formance measurement of nursing care:
State of the science and the current consen-
sus. Med Care Res Rev 2007; 64(2 Suppl):
10S– 43S
35. Yeh SH, Lee LN, Ho TH, et al: Implications of
nursing care in the occurrence and conse-
quences of unplanned extubation in adult
intensive care units. Int J Nurs Stud 2004;
41:255–262
36. Marcin JP, Rutan E, Rapetti PM, et al: Nurse
staffing and unplanned extubation in the pe-
diatric intensive care unit. Pediatr Crit Care
Med 2005; 6:254 –257
37. Baer CL: Is there an answer to preventing
unplanned extubations? Crit Care Med 1998;
26:989 –990
38. Kanten DN, Mulrow CD, Gerety MB, et al:
Falls: An examination of three reporting
methods in nursing homes. J Am Geriatr Soc
1993; 41:662– 666
39. Weingart SN, Pagovich O, Sands DZ, et al:
What can hospitalized patients tell us about
adverse events? Learning from patient-
reported incidents. J Gen Intern Med 2005;
20:830 – 836
Crit Care Med 2007 Vol. 35, No. 12