GENERAL ARTICLES
Independent Risk Factors for Postoperative Shivering
Leopold H. J. Eberhart, MD, Friederike Döderlein, MD, Gudrun Eisenhardt, RN,
Peter Kranke, MD, Daniel I. Sessler, MD, Alexander Torossian, MD, Hinnerk Wulf,
Astrid M. Morin, MD
Department of Anesthesiology and Critical Care, Philipps-University Marburg, Germany; Department of Anesthesiology,
University of Würzburg, Germany; Outcomes Research™ Institute and Departments of Anesthesiology and
Pharmacology, University of Louisville, Louisville, Kentucky
Postoperative shivering (PAS) is uncomfortable for pa-
tients and potentially risky. In this observational trial
we sought to identify independent risk factors for PAS
after general anesthesia. Potential risk factors for PAS
were recorded in 1340 consecutive patients. Signs of
shivering, peripheral and core temperature, and ther-
mal comfort were recorded in the postanesthetic care
unit. The data were split into an evaluation data set (n ⫽
1000) and a validation data set (n ⫽ 340). The first was
used to identify independent risk factors for PAS and to
formulate a risk score using backward-elimination lo-
gistic regression analysis. The proposed model was
subsequently tested for its discrimination and calibra-
tion properties using receiver operating characteristic
(ROC)-curve analysis and linear correlation between
the predicted and the actual incidences of PAS in the
P
ostoperative shivering-like tremor (PAS) is a rel-
atively frequent complication of anesthesia and
surgery that can be distressing to patients and is
occasionally associated with deleterious sequelae. Pa-
tients with compromised cardiopulmonary systems
(e.g., coronary artery disease) are presumably at great-
est risk because shivering increases cardiac output and
causes tachycardia. Furthermore, shivering occasion-
ally impedes monitoring techniques such as pulse
oximetry (1).
Most PAS is simply normal thermoregulatory shiv-
ering that is triggered by hypothermia and preceded
by arteriovenous shunts vasoconstriction. Shivering
Supported, in part, by NIH Grant GM 061655 (Bethesda, MD), the
Gheens Foundation (Louisville, KY), the Joseph Drown Foundation
(Los Angeles, CA), and the Commonwealth of Kentucky Research
Challenge Trust Fund (Louisville, KY) to Daniel I. Sessler.
Accepted for publication June 29, 2005.
Address correspondence and reprint requests to Leopold Eber-
hart, MD, Department of Anesthesiology and Critical Care,
Philipps-University Marburg, Baldingerstr. 1, D–35033 Marburg,
Germany. Address e-mail to eberhart@mailer.uni-marburg.de.
DOI: 10.1213/01.ANE.0000184128.41795.FE
©2005 by the International Anesthesia Research Society
0003-2999/05
MD, and
validation group. The incidence of PAS was 11.6%.
There were three major risk factors: young age, endo-
prosthetic surgery, and core hypothermia, with age be-
ing the most important. The risk score derived from this
analysis had a reasonable discriminating power, with
an area under the ROC-curve of 0.69 (95% confidence
interval, 0.60 – 0.78; P ⬍ 0.0001). Furthermore the equa-
tion of the calibration curve (y ⫽ 0.69x ⫹ 6; R2 ⫽ 0.82; P
⬍ 0.05) indicated a good and statistically significant
agreement between predicted and actual PAS inci-
dence. Postoperative shivering can be predicted with
acceptable accuracy using the four risk factors identi-
fied in the present study. The presented model may
serve as a clinical tool to help clinicians to rationally
administer prophylactic antishivering drugs.
(Anesth Analg 2005;101:1849 –57)
can also occur in normothermic patients who are de-
veloping fever (2). However, some shivering-like
tremor during labor (3) and after general anesthesia
(4) is not thermoregulatory. Although the etiology of
this tremor remains incompletely understood, it is
aggravated by inadequate pain control (5). Further-
more, some patients who are distinctly hypothermic
do not shiver (6).
A recent meta-analysis of pharmacological interven-
tions to prevent PAS suggested that in a population at
risk for PAS, roughly four patients need to receive
prophylactic clonidine or meperidine to prevent shiv-
ering in one patient (7). Even effective prophylactic
interventions may be associated with large numbers-
needed-to-treat if the incidence of the outcome of in-
terest is infrequent. An accurate predictive model
might reduce the number-needed-to-treat by obviat-
ing treatment in patients who are relatively unlikely to
shiver. This would mean that fewer patients would
need to be exposed to the cost and risk of treatment.
Furthermore, a predictive tool can be used for scien-
tific work, e.g., to ensure a homogenous risk for PAS
between comparative groups in future studies. The
Anesth Analg 2005;101:1849–57 1849
1850 EBERHART ET AL.
RISK FACTORS FOR POSTOPERATIVE SHIVERING
purpose of our study was to develop an algorithm,
considering other plausible contributors, for predict-
ing PAS.
Methods
During a 3-mo period we initially recruited 1385 adult
patients (approximately 50% of the patients operated
on during this time period) with IRB approval and
informed consent in this observational, one-center
trial. The patients were screened preoperatively; mor-
phometric data, type of medication taken, and medical
history were recorded. Patients who were hyperther-
mic (⬎37.8°C) before surgery or receiving medication
with the potential to influence thermoregulation (e.g.,
clonidine, phenothiazines, meperidine) were excluded
from the trial. General anesthesia and the surgical
procedure were not standardized, but patients receiv-
ing regional anesthesia and those bypassing the recov-
ery room (minor surgery with fast-tracking protocols
or extensive surgical procedures directly transferred
to an intensive care unit) were not included; the type
and duration of surgery and administration of anes-
thetics, fluids, and adjuvant medication were recorded
for each patient. Our protocol did not restrict active
perioperative warming and this was used in 145 pa-
tients at the discretion of the attending anesthesiolo-
gist. All procedures were performed at the university
surgical department of the corresponding author. All
operating rooms were climatized by an automatic sys-
tem ensuring a constant ambient temperature of
20.0°C–21.0°C and relative humidity of 35%– 40%.
The patients were assessed immediately after ar-
rival in the postoperative care unit (PACU) with re-
spect to the patient’s alertness, motor activity, hemo-
dynamic and respiratory stability, pain, and any
nausea or vomiting using the modified Aldrete recov-
ery score (7,8). Core and peripheral temperatures were
then recorded using a First Temp Genius Model 3000A
aural canal thermometer (Sherwood Medical Com-
pany, St. Louis, MO). Tympanic membrane tempera-
ture (mean of measurements in both ears) was consid-
ered core temperature (Tcore). Mean peripheral skin
temperature (Tskin) was derived from 8 measurements
of skin temperature at 4 standardized points meas-
ured bilaterally using a formula presented by Ra-
manathan (9): Tskin ⫽ 0.3 (chest ⫹ deltoid) ⫹ 0.2 (thigh
⫹ leg).
In the postanesthetic holding area, patients were
covered with a blanket but were not actively warmed.
Patients were continuously observed for the occur-
rence of PAS for the first 15 min before and then at
3-min intervals for the following hour. All measure-
ments were performed by the same specially trained
observer (FD) to minimize observer bias. The intensity
of PAS was graded using the scale described by Cross-
ley and Mahajan (10): 0 ⫽ no shivering; 1 ⫽ no visible
ANESTH ANALG
2005;101:1849 –57
Figure 1. Flowchart to illustrate patient enrollment and subsequent
exclusion of the patients.
muscle activity but piloerection, peripheral vasocon-
striction, or both are present (other causes excluded); 2
⫽ muscular activity in only one muscle group; 3 ⫽
moderate muscular activity in more than one muscle
group but no generalized shaking; 4 ⫽ violent mus-
cular activity that involves the whole body. Patients
were judged to have PAS when they displayed grade
3 or 4 activity for at least 3 min. These patients were
asked to rate how cold they felt using a simple verbal
rating scale (none, mild, moderate, severe). After this
evaluation, PAS was treated at the discretion of the
nursing staff with 25 mg meperidine (pethidine) or 75
to 150 ␮g IV clonidine.
Our main outcome was the occurrence of PAS
(grade 3 or 4) during the first postoperative hour.
Among the 1203 patients who were included in the
final analysis, 8 had to be withdrawn from the analysis
because of incomplete observational recordings. The
data from the remaining 1340 patients were randomly
split into an evaluation data set (n ⫽ 1000) and a
validation data set (n ⫽ 340). Figure 1 shows the trial
design.
To reduce the number of variables to be included in
the multifactorial model, a univariate statistic (␹2 or
Fisher’s exact tests for nominal or dichotomous data
and Mann-Whitney U-test for continuous data) was
calculated for each variable using the data of the eval-
uation data set. Variables with a P value of 0.20 or less
or those identified in previous studies were defined as
potentially relevant risk factors and were further sub-
jected to a stepwise backward logistic regression anal-
ysis using the maximum likelihood function. Type of
surgery was classified mainly according to the fre-
quency, the anatomical location (peripheral, abdomi-
nal, urological, surface, head and neck, neurosurgical),
the technique of the surgery (laparoscopic, endo-
scopic), and the complexity of the procedure (minor
versus major surgery) based on the average opioid
consumption of previous patients. The validity of the
model was verified by comparing it with the results of
ANESTH ANALG
2005;101:1849 –57
Figure 2. Receiver-operating characteristic (ROC) curve (solid line)
of the 340 patients of the validation data set. The area under the
curve of 0.69 was higher than 0.5 (P ⬍ 0.0001), which indicated that
postoperative shivering (POS) can be predicted when the four risk
factors are applied to an independent set of patients. The dotted line
indicates the decision criterion that leads to the sensitivity and
specificity, respectively.
a forward and a mixed forward-backward procedure.
The goodness of fit of a model was judged using
Nagelkerkes’s R2. Potential interactions between the
independent variables were analyzed using the graph-
ical tools provided by the JMP statistical software
package (JMP 5.1; SAS Institute Inc., Cary, NC).
The factors included in the initial model were used
to calculate the probability of shivering for each pa-
tient of the validation data set. The discriminating
properties of the predictive model were judged by
calculating the area under a receiver operating char-
acteristic (ROC) curve, which was constructed by cor-
relating true-positive and false-positive rates (sensitiv-
ity plotted against 1 ⫺ specificity) for a series of cut-off
points defined as the predicted risk (Fig. 2). The area
under the ROC curve represents the probability that a
patient experiences PAS has a higher value than one
who does not experience it (11). Theoretically a 45°
bisector would yield a prediction score that was no
better than a random guess. Thus, the area under this
“random score” would be 0.5. A score performing
significantly better than chance has an area under the
ROC curve more than 0.5 with the lower limit of the
95% confidence interval (CI) exceeding the value of 0.5
(12).
Calibration was judged by plotting predicted
against the actual incidences. For this purpose, PAS
risk was calculated for each patient in the validation
data set. The patients were then divided into quintiles
(39 patients each) of increasing risk. Predicted risk in
EBERHART ET AL. 1851
RISK FACTORS FOR POSTOPERATIVE SHIVERING
each quintile was then plotted against observed risk in
each group. Calibration characteristics were expressed
as the slope (ideal: 1.0) and the offset (ideal: no offset)
of the regression line. The Spearman rank correlation
procedure was used to test the significance of agree-
ment between the predicted and the actual incidences.
To set up a risk model and to extract variables with
significant impact on PAS, an arbitrary number of
1000 patients was chosen. Based on results from a pilot
study, we assumed there would be a 12% incidence of
PAS. Thus, if 300 patients were used for validation of
the risk model the power would be 98% to achieve an
area under a ROC curve of 0.7 or higher (two-sided
alternative hypothesis). We thus intended to include
approximately 1300 patients in this survey in order to
have 1000 for evaluation and about 300 for validation.
Results
The incidence of moderate or severe shivering (grade
3 and 4) was 11.6% (95% CI, 9.7%–13.7%) in the 1000
patients of the evaluation data set. The overall inci-
dence of any shivering (including grade 2) was 14.4%
(12.3%–16.7%). Thermoregulatory work (including
vasoconstriction/piloerection) was observed in 17.6%
(95% CI, 15.3%–20.1%). Among these patients, only
33% (58 of 176) complained of feeling cold.
Based on the univariate statistical analysis for all
recorded data, we determined that 26 factors were
potentially related to the occurrence of PAS. These,
and some basic biometric data of the patients, are
listed in Table 1. Results of the postoperative assess-
ment using the modified Aldrete score are presented
in Table 2. Among the 26 potentially relevant factors
that were included in the stepwise logistic regression
analysis, 23 were removed because they proved to be
no significant predictors at the 5% level. An identical
model was achieved using a forward and a mixed
forward-backward logistic regression analysis.
The remaining 3 factors accounted for approxi-
mately 20% of the observed variation of the data
(Nagelkerkes’s R2 ⫽ 0.192). These 3 variables are pre-
sented in Table 3, along with their odds ratios and 95%
CIs. The ␤-coefficient and the constant allow the cal-
culation of a linear term z of which the negative value
is used as the exponent of the logit equation: predicted
risk ⫽ 100%/(1⫹e-z). Variables with a negative ␤ or an
odds-ratio ⬍1.0 are associated with reduced risk of
PAS; these were older age and higher core tempera-
ture at PACU admission. Factors with a positive sign
and an odds ratio more than 1.0, respectively, are
associated with an increased risk of PAS. In the final
model, endoprosthetic surgery compared with all
other kinds of procedures (Table 1) increases the inci-
dence of PAS: z ⫽ 17.5 ⫺ (0.531 · age in decades) ⫺
(0.462 · core temperature in °C) ⫹ [1.23 · (1 for endo-
prosthetic surgery or 0 for other surgery)].
1852 EBERHART ET AL. ANESTH ANALG
RISK FACTORS FOR POSTOPERATIVE SHIVERING 2005;101:1849 –57

Table 1. Biometric Data of Patients in the Validation Dataset

Patients without
All patients shivering Patients with
(n ⫽ 1000) (n ⫽ 884) shivering (n ⫽ 116) P value
0.69
Male 552 490 (55) 62 (53)
Female 448 394 (45) 54 (47)
Age (yr) 53 ⫾ 19 55 ⫾ 18 40 ⫾ 18 ⬍0.001
Body mass index (kg 䡠 m⫺2) 26.5 ⫾ 6.4 26.7 ⫾ 6.6 24.9 ⫾ 4.1 0.003
Duration of surgery (min) 70 (45–105) 70 (45–100) 80 (50–110) 0.011
Duration of anesthesia (min) 115 (85–155) 115 (80–155) 125 (95–160) 0.010
Temperature at arrival in PACU (°C)
Core temperature 35.8 ⫾ 0.7 35.9 ⫾ 0.7 35.8 ⫾ 0.7 0.126
Mean skin temperature 32.3 ⫾ 0.9 32.3 ⫾ 0.9 32.5 ⫾ 0.9 0.090
Anesthesia induction ⬍0.001
Propofol 773 665 (75) 108 (93)
Etomidate 227 219 (25) 8 (7)
Maintenance of anesthesia
IV anesthetic 358 307 (35) 65 (56) 0.063
Volatile anesthetic 642 577 (65) 51 (44)
Remifentanil supplementation 187 158 (18) 29 (25) 0.076
Use of nitrous oxide 655 572 (65) 83 (72) 0.177
Nondepolarizing muscle ⬍0.001
relaxants
None 72 53 (6) 19 (16)
cistracurium 102 97 (11) 5 (4)
Mivacurium 146 131 (15) 15 (13)
Rocuronium 680 603 (68) 77 (66)
Suxamethonium 0.029
No 846 740 (84) 106 (91)
Yes 154 144 (16) 10 (9)
Vasopressor (norepinephrine) ⬍0.001
No 747 644 (73) 103 (89)
Yes 253 240 (27) 13 (11)
Prophylactic diclofenac ⬍0.001
No 739 676 (76) 63 (54)
Yes 261 208 (24) 53 (46)
Type of surgery ⫽0.0043
Endoprosthetic surgery 160 130 (15) 30 (26)
Other 840 754 (85) 86 (74)
Prophylactic non-opioid analgesics (metamizole or acetaminophen) 0.142
No 258 235 (27) 23 (20)
Yes 742 649 (73) 93 (80)
Chronic treatment
ACE-inhibitor 147 142 (16) 5 (4) ⬍0.001
Acetyl-salicylate acid 107 104 (12) 3 (3) 0.001
Beta adrenergic blocker 181 173 (17) 8 (4) ⬍0.001
Calcium-antagonist 95 92 (10) 3 (3) 0.004
Diuretic 106 102 (12) 4 (3) 0.006
H⫹ pump blocker 90 87 (10) 3 (3) 0.009
Oral antidiabetic 70 69 (8) 1 (1) 0.003
All variables that tend to exert influence on the occurrence of shivering (defined as a P ⱕ 0.20 in the univariate analysis) were included. The peripheral skin
temperature was calculated according to the formula by Ramanathan (9). Data are expressed as mean ⫾ sd, absolute (relative) incidences, or median (25th–75th
percentile).
PACU ⫽ postanesthesia care unit; ACE ⫽ angiotensin converting enzyme.

For example, when a 22-yr-old patient is admitted
normothermic (36.5°C core temperature) to the
recovery room after a nonendoprosthetic surgery
then the linear term z can be calculated as follows:
z ⫽ 17.5 ⫺ (0.531 · 3) ⫺ (0.462 · 36.5) ⫹ (1.23 · 0) ⫽
⫺ 0.956.
When the negative of z (⫹ 0.956) is inserted into the
logit equation then the predicted risk is 27.8%. A
72-yr-old patient under the same preconditions has a
risk for PAS of only 8.5%. If both patients are admitted
hypothermic (core temperature 35.0°C), then the risk
to develop PAS for the young patient is increased to
ANESTH ANALG EBERHART ET AL. 1853
2005;101:1849 –57 RISK FACTORS FOR POSTOPERATIVE SHIVERING

Table 2. Postoperative Clinical Status of the Patients Entering the Postanesthetic Recovery Area
Patients without Patients
All patients shivering with shivering
(n ⫽ 1000) (n ⫽ 884) (n ⫽ 116) P value
Vigilance 0.89
Awake and orientated 488 430 (49) 58 (50)
Rousable with minimal stimulation 460 407 (46) 53 (46)
Responsive to tactile stimulation 52 47 (5) 5 (4)
Activity 0.79
Moves all extremities on command 813 719 (82) 94 (81)
Some weakness in movement of extremities 162 144 (16) 18 (16)
Unable to move extremities 25 21 (2) 4 (3)
Hemodynamic stability 0.88
Blood pressure ⬍15% of preoperative baseline 769 689 (78) 90 (78)
Blood pressure 15–30% of preoperative baseline 180 149 (17) 21 (18)
Blood pressure ⬎30% of preoperative baseline 51 46 (5) 5 (4)
Respiratory stability 0.95
Able to breathe deeply 874 774 (88) 100 (86)
Tachypnea with good cough 108 94 (10) 14 (12)
Dyspneic with weak cough 18 16 (2) 2 (2)
Oxygen saturation status 0.82
Maintained ⬎90% with room air 649 571 (65) 78 (67)
Required supplemental oxygen 337 301 (34) 36 (31)
Saturation ⬍90% with supplemental oxygen 14 12 (1) 2 (2)
Postoperative pain assessment 0.95
None or mild discomfort 117 103 (12) 14 (12)
Moderate, controllable pain 779 690 (78) 89 (77)
Severe pain 104 91 (10) 13 (11)
Postoperative emetic symptoms 0.81
None or mild nausea, no vomiting 960 848 (96) 112 (97)
Transient vomiting or retching 37 33 (4) 4 (3)
Persistent nausea and vomiting 3 3 (⬍1) 0
All assessments were performed using a standardized postanesthesia recovery score (modified Aldrete score (8)). Neither a single item nor the sum score was
entered in the logistic regression model due to insignificant testing in the univariate statistics. Values are expressed as absolute (relative) incidences.

Table 3. Odds Ratios and 95% Confidence Intervals of the Four Variables Identified as Independent Risk Factors for the
Occurrence of Postoperative Shivering
Beta-coefficient Odds-ratio *Nagelkerkes’s
⫾ se P (95% CI) R2
Age (per decade) ⫺0.531 ⫾ 0.055 ⬍0.0001 0.59 (0.53–0.66) 0.138
Endoprosthetic surgery 1.23 ⫾ 0.28 ⬍0.0001 3.40 (1.97–5.89) 0.172
Tcore (per °C) ⫺0.462 ⫾ 0.13 ⬍0.0001 0.63 (0.49–0.81) 0.192
Constant 17.5 ⫾ 4.7
se ⫽ Standard error. CI ⫽ confidence interval.
Results are from the stepwise backward logistic regression analysis using the maximum likelihood function. The beta-coefficient and the constant can be used
to calculate a linear equation z that can be used to compute the individual risk for a patient using the logit equation risk (%) ⫽ 100/(1 ⫹ e⫺z).
Nagelkerkes’s R2 is an incremental measure of the goodness of fit of the regression model as additional variables are included in the model.

43.5% and that for the older patient is increased to
15.6%.
This calculation was performed in each of the 340
patients of the validation data set. These patients did
not differ from those in the evaluation data set with
respect to biometric data or type and duration of
anesthesia and surgery. The incidence of PAS grade
3 and 4 in. this validation data set was 12.7% and
was thus very similar to the rate of the evaluation
data set (11.6%). Figure 2 shows the calculated ROC
curve. The area under the curve was 0.69 (95% CI,
0.60 – 0.78), indicating that PAS could be predicted
with a moderate accuracy in these patients (P ⬍
0.0001). The gray line indicates the decision criterion
that can be used to judge the sensitivity and speci-
ficity respectively for each cut-off point. For exam-
ple, if patients are classified as “shivering” even
when the expected risk is low (e.g., 10%) then the
score has a high sensitivity but a corresponding low
specificity (the ROC curve is in the upper right
corner of the diagram). Conversely, if patients are
classified as shivering only when a high cut-off level
1854 EBERHART ET AL.
RISK FACTORS FOR POSTOPERATIVE SHIVERING
Figure 3. Calibration characteristics of the risk score to predict
postanesthetic shivering. The predicted (calculated) incidences that
were grouped into an arbitrary number of 5 quintile (68 patients
each) were plotted on the x-axis against the observed incidences
(with their 95%-confidence intervals) on the y-axis. A linear regres-
sion analysis was used to derive an equation (y ⫽ 0.69 x ⫹ 6) used
to judge the correlation. (R2 ⫽ 0.82; P ⬍ 0.05 using the Spearman
rank correlation test).
(e.g., 40%) is exceeded, then the decision criterion
has a low sensitivity but a high specificity (ROC
curve runs in the lower left corner of the figure).
Figure 3 shows the results of the calibration analy-
sis. Here, the predicted incidences (x-axis) that were
grouped within five predicted risk quintiles are plotted
against the observed incidences PAS of these patients
(y-axis). Calibration characteristics are expressed as the
slope and the offset of the regression line. This equa-
tion can be described by y ⫽ 0.69x ⫹ 6. The correlation
coefficient indicates a moderate but statistically signif-
icant agreement between the predicted and the actual
incidences (R2 ⫽ 0.82; P ⬍ 0.05).
In Table 4 those 10 factors potentially influencing
the occurrence of PAS that were removed during the
last steps of the logistic regression analysis are presented.
For these variables the P value and the step at which they
were removed from the model are listed.
Discussion
More than a decade ago, Crossley (13) used logistic
regression analyses on data from 2595 patients and
found that several variables impact on PAS develop-
ment: longer surgery duration, male sex, anticholin-
ergic premedication (e.g., atropine), spontaneous ven-
tilation, higher ASA physical status, general versus
ANESTH ANALG
2005;101:1849 –57
orthopedic surgery, and administration of blood. He
also reported that older age and administration of
propofol, alfentanil, or morphine were the most im-
portant protective factors against PAS. However,
these results have not been further verified. It is also
unclear whether attempts were made to reduce the
number of factors in Crossley’s model by a stepwise
elimination of insignificant factors.
Using stepwise multivariate analysis, we identified
three independent risk predictors of PAS: younger
age, endoprosthetic surgery, and low core body tem-
perature. Age proved to be by far the most important
risk factor for PAS, accounting for more than 70% of
the predictive power of our entire model. This was not
surprising because the thermoregulatory responses to
cold and heat are attenuated in older patients (2). For
example, the vasoconstriction threshold during ni-
trous oxide/isoflurane anesthesia (14) and the shiver-
ing threshold during spinal anesthesia (15) are each
decreased by about 1°C in the elderly.
It is difficult to extract independent risk factors for
PAS in the perioperative setting, as numerous vari-
ables influence the postoperative course of the pa-
tients, perhaps most importantly, thermoregulatory
impairment resulting from residual volatile (16) or IV
anesthetics and sedatives (17). The situation is further
complicated by co-linearities between risk factors. Us-
ing a univariate analysis, almost all drugs taken reg-
ularly by the patients (e.g., antihypertensive and an-
tidiabetic drugs, cumarine derivates) seem to offer
significant protection against PAS. This observation is
not new. In fact, a significantly less frequent incidence
of PAS in patients taking propranolol was reported in
1986 (18). Obviously, this kind of chronic medication
treatment is much more frequent in older patients
than in younger ones. Another example for co-
linearity within the data set was the ASA physical
status of the patients. Higher ASA classification is
positively correlated with age and is thus supposed to
be a significant protective factor in a univariate
analysis, although it has only a minor impact in the
multivariate model. Thus, by appropriately including
age into the multifactorial model, numerous suppos-
edly protective factors were removed early during the
stepwise regression analysis.
The hypothermia that develops within the first hour
after induction of general (19) or neuraxial anesthesia
(20) results primarily from core-to-peripheral redistri-
bution of body heat. Shivering and vasoconstriction
are each 80% controlled by core temperature, with the
remaining 20% being derived from mean skin temper-
ature (21). It is thus reasonable to assume that hypo-
thermia contributes to development of PAS, and nor-
mothermia is indeed protective in the absence of
surgery (22). In the data analyzed by Crossley (13),
core temperature of most patients who did not shiver
ANESTH ANALG
2005;101:1849 –57
Table 4. Variables Included Into the Backward Logistic Regression Models That Were Removed at the Last 10 Steps of
the Procedure
Variable
Sex
Duration of anesthesia
Administration of fresh frozen plasma
Abdominal versus all other types of surgery
Administration of acetaminophen
Volatile anesthetics versus propofol for maintenance of anesthesia
Administration of packed red cells
Intraoperative administration of corticosteroids
Minor versus major surgery
Duration of surgery
Please note that P values must not necessarily decrease in the order the factors are removed from the model, because the regression model is recalculated at
each step.
postoperatively was not recorded and were thus un-
available for analysis. Most of our patients entered the
PACU slightly hypothermic (with a mean core tem-
perature of 35.8°C). However, 11.1% of the patients
had core temperatures ⬍35°C and core temperature
was ⬍34°C in 1% of the patients. When core temper-
ature of patients who developed shivering and those
who did not were compared, there was neither a
clinically relevant nor a statistically significant differ-
ence (Table 1). However, when included in the multi-
factorial regression model, a significant influence was
detected.
Our observation that core temperature has only a
slight influence on PAS development compared with
age as the most important determinate is consistent
with other studies that have shown that PAS is poorly
predicted by either core or peripheral body tempera-
ture (10,23). According to a multifactorial logistic re-
gression analysis, absolute postoperative core temper-
ature was not directly related to postoperative
shivering in children whereas relative perioperative
temperature change was one of three independent
predictors (24). However, patients with a body tem-
perature ⬍36°C shivered for a longer time than those
who were warmer (25), and at least one study found a
linear relationship between PAS and esophageal tem-
perature (26). A confounding factor in many studies,
including ours, is suboptimal measurement of core
temperature. Among infrared aural canal thermome-
ters, the one we used (Kendall GENIUS) is among the
most accurate (27). However, it is important to recog-
nize that any inaccuracy in core temperature meas-
urements degrades the apparent contribution of tem-
perature compared with factors when measurement
accuracy is minimal such as patient age.
Because simply covering patients with a blanket
was reported to reduce PAS without altering core
temperature (28), skin temperature has also been im-
plicated as a causative factor for PAS; our data refute
EBERHART ET AL. 1855
RISK FACTORS FOR POSTOPERATIVE SHIVERING
Removed P value
at step at removal
14 0.64
15 0.89
16 0.38
17 0.62
18 0.69
19 0.39
20 0.39
21 0.24
22 0.14
23 0.11
this assertion. In fact, only core temperature was re-
tained in our final model whereas peripheral skin
temperature and the derived mean body temperature
were removed as insignificant variables. However, it
is important to recognize that skin temperature dropped
out because skin and core temperature are usually well
correlated. There is no question that sufficiently increas-
ing skin temperature alone can stop PAS (29).
We confirmed the previous finding (13) that ortho-
pedic surgery, particularly endoprosthetic surgery us-
ing bone cement is an independent risk factor for the
development of PAS. However the underlying biolog-
ical reasons for this remain unclear. One possible expla-
nation is that bone cement (polymethyl-methacrylate),
which is often used in arthroplastic surgery, stimulates
the release of cytokines such as ␣-tissue necrosis factor
and interleukin-6 (30), both of which can increase the set
point of the thermoregulatory system postoperatively.
The question that arises from these possibilities is
whether shivering in normothermic patients is caused
by unknown intrinsic mechanisms of certain anesthet-
ics or by an increase of the individual thermoregula-
tory set point that, in other words, might be described
as continuing “postoperative fever.” The first expla-
nation is unlikely because we found no association
between a certain anesthesia technique or anesthetic
drugs and an increased incidence of PAS. For exam-
ple, maintenance of anesthesia (volatile anesthetics
versus propofol) was removed at step 19 with a P
value of 0.39. However, Frank et al. (31) found that
even in the absence of clinical signs of infection, 50%
of postoperative patients reach core temperatures
⬎38°C–38.5°C within the first 24 hours postopera-
tively. Furthermore, in the same study there was a
positive association between the maximum tempera-
ture reached during the postoperative course and
younger age and, as already discussed, younger age
was the most important predictor for PAS in our
model. In this context, it is interesting that all variables
1856 EBERHART ET AL.
RISK FACTORS FOR POSTOPERATIVE SHIVERING
that were removed during the last steps of the regres-
sion analysis with P values close to 0.05 are indicators
for an invasive, more painful surgery. Longer dura-
tion of surgery was associated with more frequent
PAS, although this effect was corrected for a slightly
lower core temperature in these patients by having the
temperature variable still included at this step. After
more invasive procedures more frequent PAS was
observed, although this influence could not be proven
on the 0.05% level. One possible explanation as to why
duration of surgery predicts PAS is that these proce-
dures are usually more complex and invasive. Injured
tissue can release pyrogenic substances that, in turn,
increase the set point of the thermoregulatory system
postoperatively and, thus, induce PAS (31). In support
of this theory, increased plasma concentrations of
interleukin-6 are more often observed after longer and
more invasive procedures than shorter ones and are
associated with higher body temperatures during the
postoperative period (31).
Several potential risk factors for PAS identified in
previous studies were rejected by our analysis, includ-
ing pain, male sex, and type of anesthesia. One review
suggested that shivering, as an integral part of the
thermoregulatory system, is tightly linked to other
homeostatic systems, including pain control (2). The
authors argue that pain and temperature signals are
transmitted along similar fiber systems that both syn-
apse in dorsal root horn regions. Furthermore most
antishivering drugs have weak or moderate analgesic
action (7,32). However, we did not observe any asso-
ciation between postoperative pain intensity and PAS.
Pain, though, was well controlled in most of our pa-
tients, perhaps obscuring the importance of this factor
in our analysis.
Male sex was also thought to contribute to the
incidence of PAS (13), but we did not find that this
to be an important factor. It was removed at an early
step during the regression procedure. It may be that
shivering is not more common in males but is more
apparent to casual clinical observation because the larger
muscle mass of males makes severe PAS especially
impressive.
A limitation of our study is that we can not test the
tempting hypothesis that postoperative inflammation
may play an important role with our data because we
did not measure temperature or inflammatory mediator
concentrations as a function of postoperative time.
In conclusion, PAS can be predicted with moderate
discriminating power using four risk factors derived
from a logistic regression analysis. Age of the patient
was the variable with the most predictive power by
far. Other risk factors included low core temperature
at admission to the PACU, prolonged surgery, and
orthopedic surgery.
ANESTH ANALG
2005;101:1849 –57
References
1. Buggy DJ, Crossley AW. Thermoregulation, mild perioperative
hypothermia and postanaesthetic shivering. Br J Anaesth 2000;
84:615–28.
2. De Witte J, Sessler DI. Perioperative shivering: physiology and
pharmacology. Anesthesiology 2002;96:467– 84.
3. Panzer O, Ghazanfari N, Sessler DI, et al. Shivering and
shivering-like tremor during labor with and without epidural
analgesia. Anesthesiology 1999;90:1609 –16.
4. Horn EP, Sessler DI, Standl T, et al. Non-thermoregulatory
shivering in patients recovering from isoflurane or desflurane
anesthesia. Anesthesiology 1998;89:878 – 86.
5. Horn EP, Schroeder F, Wilhelm S, et al. Postoperative pain
facilitates nonthermoregulatory tremor. Anesthesiology 1999;
91:979 – 84.
6. Camus Y, Delva E, Just B, Lienhart A. Leg warming minimizes
core hypothermia during abdominal surgery. Anesth Analg
1993;77:995–9.
7. Kranke P, Eberhart LH, Roewer N, Tramèr MR. Single-dose
parenteral pharmacological interventions for the prevention of
postoperative shivering: a quantitative systematic review of
randomized controlled trials. Anesth Analg 2004;99:718 –27.
8. White PF, Song D. New criteria for fast-tracking after outpatient
anesthesia: a comparison with the modified Aldrete’s scoring
system. Anesth Analg 1999;88:1069 –72.
9. Ramanathan NL. A new weighting system for mean surface
temperature of the human body. J Appl Physiol 1964;19:531–3.
10. Crossley AWA, Mahajan RP. The intensity of postoperative
shivering is unrelated to axillary temperature. Anaesthesia 1994;
49:205–7.
11. Hanley JA, McNeil BJ. The meaning and use of the area under
a receiver operating characteristic (ROC) curve. Radiology 1982;
143:29 –36.
12. Eberhart LHJ, Högel J, Seeling W, et al. Evaluation of three risk
scores to predict postoperative nausea and vomiting. Acta An-
aesthesiol Scand 2000;44:480 – 8.
13. Crossley AW. Six months of shivering in a district general
hospital. Anaesthesia 1992;47:845– 8.
14. Kurz A, Plattner O, Sessler DI, et al. The threshold for thermo-
regulatory vasoconstriction during nitrous oxide/isoflurane an-
esthesia is lower in elderly than in young patients. Anesthesi-
ology 1993;79:465–9.
15. Vassilieff N, Rosencher N, Sessler DI, Conseiller C. Shivering
threshold during spinal anesthesia is reduced in elderly pa-
tients. Anesthesiology 1995;83:1162– 6.
16. Xiong J, Kurz A, Sessler DI, et al. Isoflurane produces marked
and nonlinear decreases in the vasoconstriction and shivering
thresholds. Anesthesiology 1996;85:240 –5.
17. Matsukawa T, Kurz A, Sessler DI, et al. Propofol linearly re-
duces the vasoconstriction and shivering thresholds. Anesthe-
siology 1995;82:1169 – 80.
18. Lee DS, Shaffer MJ. Low incidence of shivering with chronic
propranolol therapy. Lancet 1986;1:500.
19. Matsukawa T, Sessler DI, Sessler AM, et al. Heat flow and
distribution during induction of general anesthesia. Anesthesi-
ology 1995;82:662–73.
20. Matsukawa T, Sessler DI, Christensen R, et al. Heat flow and
distribution during epidural anesthesia. Anesthesiology 1995;
83:961–7.
21. Cheng C, Matsukawa T, Sessler DI, et al. Increasing mean skin
temperature linearly reduces the core-temperature thresholds
for vasoconstriction and shivering in humans. Anesthesiology
1995;82:1160 – 8.
22. Sessler DI, Rubinstein EH, Moayeri A. Physiologic responses to
mild perianesthetic hypothermia in humans. Anesthesiology
1991;75:594 – 610.
23. Holdcroft A, Hall GM. Heat loss during anaesthesia. Br J An-
aesth 1978;50:157– 64.
24. Lyons B, Taylor A, Power C, Casey W. Postanaesthetic shivering
in children. Anaesthesia 1996;51:442–5.
ANESTH ANALG
2005;101:1849 –57
25. Vaughan MS, Vaughan RW, Cork RC. Postoperative hypother-
mia in adults: relationship of age, anesthesia, and shivering to
rewarming. Anesth Analg 1981;60:746 –51.
26. Lienhart A, Fiez N, Deriaz H. Frisson postopératoire: analyse
des principaux facteurs associés. Ann Fr Anesth Réanim 1992;
11:488 –95.
27. Imamura M, Matsukawa T, Ozaki M, et al. The accuracy and
precision of four infrared aural canal thermometers during car-
diac surgery. Acta Anaesthesiol Scand 1998;42:1222– 6.
28. Buggy D, Hughes N. Pre-emptive use of the space blanket
reduces shivering after general anaesthesia. Br J Anaesth 1994;
72:393– 6.
EBERHART ET AL. 1857
RISK FACTORS FOR POSTOPERATIVE SHIVERING
29. Sharkey A, Lipton JM, Murphy MT, Giesecke AH. Inhibition of
postoperative shivering with radiant heat. Anesthesiology 1987;
66:249 –52.
30. Miyaguchi M, Kobayashi A, Iwaki H, et al. Human monocyte
response to retrieved polymethylmethacrylate particles.
J Biomed Mater Res 2002;62:331–7.
31. Frank SM, Kluger MJ, Kunkel SL. Elevated thermostatic setpoint
in postoperative patients. Anesthesiology 2000;93:1426 –31.
32. Kranke P, Eberhart LH, Roewer N, Tramer MR. Pharmacologi-
cal treatment of postoperative shivering: a quantitative system-
atic review of randomized controlled trials. Anesth Analg 2002;
94:453– 60.