Full text of DOI:10.1592/phco.22.11.803.33634

O RIG IN A L R ES EA RC H A RTIC LES
Risk Markers for Thrombocytopenia in Critically Ill
Patients: A Prospective Analysis
Stephen J. Shalansky, Pharm.D., FCSHP, Arun K. Verma, M.Sc., Marc Levine, Ph.D.,
John J. Spinelli, Ph.D., and Peter M. Dodek, M.D., M.H.Sc.
Study Objective. To identify independent risk markers for thrombocytopenia
in critically ill patients.
Design. Prospective, observational study.
Setting. Eleven-bed intensive care unit–coronary care unit (ICU-CCU) in a
community hospital.
Patients. Three hundred sixty-two consecutive patients meeting inclusion
criteria during 1 year.
Intervention. Potential risk marker data were collected on admission to the
ICU-CCU and for the period before development of thrombocytopenia
(defined as two or more consecutive platelet counts < 150 x 103/mm3
obtained at least 12 hours apart), or for the duration of ICU-CCU stay if
thrombocytopenia did not develop.
Measurements and Main Results. Thrombocytopenia developed in 68
patients (18.8%). Multivariate logistic regression analyses identified
patients at risk on admission, but the predictive potential of the regression
model improved when all risk marker exposures during the ICU-CCU stay
were considered. Independent risk markers included fresh frozen plasma
administration, sepsis, musculoskeletal diagnosis, pulmonary artery
catheter insertion, gastrointestinal diagnosis, packed red blood cell
administration, and nonsurgical respiratory diagnosis. Higher admission
platelet count and aspirin administration were associated with a lower risk
of thrombocytopenia. Heparin administration was not identified as a risk
marker, and no patient developed heparin-induced thrombocytopenia with
thrombosis. Patients with thrombocytopenia had longer ICU-CCU and
hospital stays, and higher ICU-CCU and hospital mortality than those
without thrombocytopenia.
Conclusions. Development of thrombocytopenia in critically ill patients is
associated with specific diagnoses, packed red cell and fresh frozen plasma
transfusions, pulmonary artery catheter insertion, and admission platelet count.
(Pharmacotherapy 2002;22(7):803–813)
Thrombocytopenia is common in critically ill
patients and is associated with increased duration
of hospital stay and increased mortality.1–6
Although a moderate decline in platelet count
does not present a serious risk to most patients,
clinicians often intervene in an attempt to
prevent life-threatening outcomes associated with
severe thrombocytopenia or heparin-induced
thrombocytopenia (HIT) with thrombosis. A
wide range of causes of thrombocytopenia such
as drug therapies, diagnoses, and procedures
have been reported in the literature7, 8; but study
results have been inconsistent, making treatment
decisions difficult. Drug therapies, especially

804
PHARMACOTHERAPY Volume 22, Number 7, 2002
heparin, often are discontinued when thrombo-
cytopenia develops, despite evidence indicating
that most patients are not at risk for drug-related
adverse events.2, 4, 9, 10
evaluated by using a prospective, observational
study design. Thrombocytopenia was defined as
two or more consecutive platelet counts less than
150 x 103/mm3 obtained at least 12 hours apart.
All patients older than 18 years were included
in the study if they had two or more platelet
counts obtained at least 12 hours apart during
their ICU-CCU stay. For patients with more than
one admission to the ICU-CCU, only the first
admission during the study period was
considered. Exclusion criteria included a platelet
count less than 150 x 103/mm3 on admission to
the unit, concomitant participation in another
study, and an admitting diagnosis strongly
associated with thrombocytopenia including
congenital thrombocytopenia, hypersplenism,
mechanical heart valve, disseminated intra-
vascular coagulation, idiopathic thrombo-
cytopenic purpura, or thrombotic thrombo-
cytopenic purpura. Patients who developed
these diagnoses during their ICU-CCU stay were
included in the analysis.
Several investigators have attempted to identify
the most important risk markers for
thrombocytopenia in critically ill patients, but
analyses generally have been limited by small
samples, retrospective design, and a limited scope
of variables assessed.1–5 In addition, the use of
various definitions for thrombocytopenia has
made it difficult to compare results. A clear
understanding of the most important risk
markers for thrombocytopenia in critically ill
patients would provide a basis for evidence-based
treatment decisions in this area. Hence, we
assessed the relative contribution of risk markers
suspected to be associated with the development
of thrombocytopenia in critically ill patients.
Methods
Protocol
Potential risk marker data were collected on
admission to the ICU-CCU and for the period
before development of thrombocytopenia, or for
the duration of the ICU-CCU stay if thrombo-
cytopenia did not develop. Data were collected
for variables previously identified or suspected to
be associated with the development of
thrombocytopenia, including patient demo-
graphics, history of alcohol abuse, diagnoses,
laboratory indexes, organ dysfunction, trans-
fusions, and procedures.1–5, 7 A comprehensive
record of administration of any of 73 drugs
previously identified as potential risk markers for
thrombocytopenia (list available from the
authors) was maintained for all study patients.1, 2,
This study was carried out at Lions Gate
Hospital, a 350-bed community institution.
Approval was obtained from the Ethics
Committees for Human Experimentation at Lions
Gate Hospital and the University of British
Columbia. The need for informed consent was
waived since this was an observational study that
did not influence the care of patients involved.
Consecutive patients admitted to the 11-bed
intensive care unit–coronary care unit (ICU-
CCU) from June 11, 1997–June 11, 1998, were
From the Pharmacy Department, Lions Gate Hospital,
North Vancouver, British Columbia, Canada (Dr.
Shalansky); and the Faculty of Pharmaceutical Sciences,
University of British Columbia (Drs. Shalansky and Levine,
and Mr. Verma), the British Columbia Cancer Agency,
Cancer Control Research (Dr. Spinelli), the Program in
Critical Care Medicine, St. Paul’s Hospital (Dr. Dodek), and
the Centre for Health Evaluation and Outcome Sciences,
University of British Columbia (Dr. Dodek), Vancouver,
British Columbia, Canada.
4, 7, 11, 12
All heparin exposure was accounted for
including heparin administered for maintaining
patency of intravenous catheters and heparin
coating on pulmonary artery catheters. Heparin
exposure up to 8 weeks before ICU-CCU
admission was determined from patient
interviews, previous health records, and the
provincial prescription database. If heparin was
discontinued in a patient with thrombocytopenia,
the prescribing physician was interviewed to
determine whether thrombocytopenia was the
reason for discontinuation.
Supported by a grant from the Lions Gate Healthcare
Research Foundation.
Preliminary results presented at the 52nd annual meeting
of the Canadian Cardiovascular Society, Quebec City,
Canada, October 19–23, 1999.
Manuscript received December 14, 2001. Accepted
pending revisions January 29, 2002. Accepted for
publication in final form April 22, 2002.
Diagnostic categories were based on the most
common International Classification of Diseases,
Ninth Revision13 codes reported by the hospital’s
medical records department for the ICU-CCU
during the 2-year period immediately before
beginning data collection. These categories were
Address reprint requests to Stephen J. Shalansky,
Pharm.D., FCSHP, Pharmacy Department, St. Paul’s
Hospital, 1081 Burrard Street, Vancouver, BC, Canada V6Z
1Y6; e-mail: sshalansky@providencehealth.bc.ca; or Marc
Levine, Ph.D., Faculty of Pharmaceutical Sciences,
University of British Columbia, Vancouver, BC, Canada
V6T 1Z3; e-mail: levine@interchange.ubc.ca.

RISK MARKERS FOR THROMBOCYTOPENIA Shalansky et al
805
reported according to published guidelines.18, 19
Data were stored and analyzed by using SPSS
version 9.0 (SPSS Inc., Chicago, IL). Univariate
analyses were carried out to identify candidate
variables for the multivariate regression analysis.
nervous system disorders, respiratory surgery,
nonsurgical respiratory disorders (including
acute respiratory distress syndrome), vascular
surgery, acute myocardial infarction, unstable
angina, other nonsurgical cardiovascular
disorders (including heart failure and
arrhythmias), gastrointestinal disorders,
gastrointestinal bleeding, musculoskeletal
disorders (including trauma), endocrine
disorders, diabetes mellitus (including diabetic
ketoacidosis), genitourinary disorders, infection
(temperature > 38.5°C and white blood cell
count > 11 x 103/mm3, or if the patient was given
antibiotics for infection, excluding sepsis), sepsis
(as defined by the American College of Chest
Physicians and Society of Critical Care
Medicine14), malignancy, and drug overdose.
Detailed definitions of each diagnosis are
available from the authors.
History of alcohol consumption was determined
through discussion with the patient, family,
and/or the attending physician. Moderate-to-
high alcohol consumption—three or more drinks
daily—was evaluated as a risk marker in view of
previous reports suggesting that this level of
alcohol intake may be associated with
thrombocytopenia and platelet dysfunction.7, 15, 16
Renal dysfunction was defined as an estimated
creatinine clearance of less than 30 ml/minute/72
kg or a 50% decrease in creatinine clearance
(from first serum creatinine measurement in the
ICU-CCU) as calculated by the modified
Cockcroft and Gault equation normalized for
weight.17 Patients were considered to have
hepatic dysfunction if liver function test results
were elevated to a magnitude previously
associated with thrombocytopenia2: greater than
5 times the hospital’s upper normal limit for liver
aminotransferase levels or alkaline phosphatase,
or greater than 3 times the hospital’s upper
normal limit of total or direct bilirubin. If these
tests were not performed, hepatic function was
assumed to be normal.
2
Dichotomous variables were analyzed with the ␹
test, whereas continuous variables were analyzed
with the Wald statistic for the logistic regression
model. Variables with an exposure frequency of
less than 5% were excluded from analysis. Since
the univariate tests were used to screen variables,
rather than to test a specific hypothesis, a p value
of 0.25 was used as the threshold for entry into
the multivariate model. Tests were performed for
colinearity among variables selected for
multivariate analysis and for linearity of selected
continuous variables. Candidate variables were
entered into a backward stepwise multivariate
logistic regression model. Variables were selected
for removal by using a p value of 0.10 as the
significance level. Interactions between variables
within the models were explored, and regression
diagnostics were done to identify outliers and
influential observations based on standard
definitions.18 For any outliers or influential
observations identified, data were reviewed to
confirm accurate and complete data entry.
Two regression models were developed. The
first model identified risk markers associated
with the development of thrombocytopenia
during ICU-CCU stay attributed to patient
variables present on admission to the ICU-CCU
(admission model). For the second model, the
diagnoses most responsible for ICU-CCU stay
were used and all potential risk markers were
considered including those present on admission
and those encountered during the ICU-CCU stay
(comprehensive model). Clinical outcomes were
compared between patients with and those
without thrombocytopenia by using the
independent t test for continuous variables and
␹2 analysis for dichotomous variables
Clinical outcomes occurring during the index
hospital stay were recorded, including the
occurrence of hemorrhage (at least a 1-g/dl
decrease in hemoglobin level over 24 hrs plus
clinical evidence of bleeding), thrombosis (deep
vein thrombosis, pulmonary embolism, myocardial
infarction, or ischemic stroke), duration of ICU-
CCU stay, duration of hospital stay, ICU-CCU
mortality, and hospital mortality.
Results
During the 12-month study period, 935
patients were admitted to the Lions Gate Hospital
ICU-CCU. Of these, 362 met the criteria for
inclusion in the study. Five hundred seventy-
three patients were excluded for the following
reasons: fewer than two platelet counts (432
patients), admission platelet count less the 150 x
103/mm3 (78), repeat admission (24), measurement
of only two platelet counts within 12 hours (21),
enrollment in another study that randomly
Statistical Analysis
Logistic regression analysis was carried out and

806
PHARMACOTHERAPY Volume 22, Number 7, 2002
Table 1. Characteristics of the 362 Study Patients
stay were distributed evenly between cardiac and
intensive care diagnoses. The most common
admission diagnoses were myocardial infarction,
other nonsurgical cardiovascular disease (e.g.,
heart failure, rhythm disturbance), and
nonsurgical respiratory disease. Mean admission
platelet count was well within the normal range,
and mean acute physiology and chronic health
evaluation (APACHE) II score (15.4 ± 9.4)
indicated moderate disease severity (Table 1). In
general, APACHE II scores range from 0–55, with
scores of 10–20 representing moderate disease
severity.20
Characteristic
Value
Mean ± SD
Age (yrs)
Weight (kg)
63.2 ± 15.4
76.2 ± 17.2
15.4 ± 9.4
246 ± 79
APACHE II score
Admission platelet count (x 103/mm3)
No. (%)
Gender
Men
Women
Caucasian
229 (63.2)
133 (36.7)
317 (87.6)
42 (11.6)
History of moderate-to-high alcohol intake
Location immediately before ICU-CCU
Emergency department
Ward
242 (66.9)
101 (27.9)
19 (5.2)
190 (52.5)
98 (27.1)
56 (15.5)
36 (9.9)
172 (47.5)
52 (14.4)
17 (4.7)
15 (4.1)
14 (3.9)
13 (3.6)
13 (3.6)
12 (3.3)
11 (3.0)
10 (2.8)
7 (1.9)
Occurrence of Thrombocytopenia
Other hospital
Cardiac diagnosisa
Acute myocardial infarction
Other nonsurgical cardiovascular
Unstable angina
During the data collection period, 68 patients
(18.8%, 95% confidence interval [CI]
14.8–22.8%) developed thrombocytopenia based
on the a priori definition. Mean ± SD onset of
thrombocytopenia was 2.9 ± 4.0 days after
admission to the unit (range 1–34 days, median
2.0 days). Frequency of thrombocytopenia for
patients in the CCU was 8.9% (95% CI
4.9–12.9%) and for patients in the ICU, 29.7%
(95% CI 22.9–36.5%). By using different
thresholds, some of which were used in
previously published research on this topic,1–5 the
frequency of thrombocytopenia for the entire
cohort was 3.6–26.5% (Table 2). Platelet counts
reported by the Lions Gate Hospital laboratory
had an inter- and intraday variability of 11% and
3%, respectively.
Intensive care diagnosisa
Respiratory, nonsurgical
Infection
Gastrointestinal
Nervous system
Sepsis
Drug overdose
Musculoskeletal
Respiratory, surgical
Gastrointestinal bleed
Vascular surgery
Diabetes mellitus
Malignancy
4 (1.1)
4 (1.1)
APACHE = acute physiology and chronic health evaluation; ICU =
intensive care unit; CCU = cardiac care unit.
aDiagnosis most responsible for stay in CCU or ICU, as defined in
Methods.
Risk Markers for Thrombocytopenia in the
Admission Model
assigned patients to receive heparin or hirudin
(11), age younger than 18 years (5), and
admission with disseminated intravascular
coagulation (2).
Thirteen admission variables were selected for
multivariate analysis based on univariate test
results. These 13 variables were age, moderate-
to-high alcohol consumption, APACHE II score,
The diagnoses most responsible for ICU-CCU
Table 2. Occurrence of Thrombocytopenia by Various Definitions
No. (%) of Patients
Cardiac
Diagnosis
(n=190)
Intensive Care
Diagnosis
Entire Cohort
(N=362)
Definition
(n=172)
At least one platelet count
< 150 x 103/mm3
96 (26.5)
28 (7.7)
13 (3.6)
33 (17.4)
1 (0.5)
1 (0.5)
63 (36.6)
27 (15.7)
12 (7.0)
< 100 x 103/mm3
< 50 x 103/mm3
At least two consecutive
platelet counts
< 150 x 103/mm3
< 100 x 103/mm3
< 50 x 103/mm3
68 (18.8)
23 (6.4)
12 (3.3)
17 (8.9)
1 (0.5)
1 (0.5)
51 (29.7)
22 (12.8)
11 (6.4)

RISK MARKERS FOR THROMBOCYTOPENIA Shalansky et al
807
Table 3. Risk Markers Independently Associated with Thrombocytopenia in the
Admission Model
Variable
Odds Ratio
95% CI
p Value
Sepsis
14.1
7.6
6.7
5.9
2.3
1.1
0.9
0.5
2.9–67.9
2.1–27.9
1.6–29.3
1.6–21.4
1.0–5.1
1.1–1.1
0.8–1.0
0.4–0.7
0.001
0.002
0.011
0.007
0.005
<0.001
0.01
Gastrointestinal diagnosis
Gastrointestinal bleed
Musculoskeletal diagnosis
Nonsurgical respiratory diagnosis
APACHE II scorea
Ageb
Admission platelet countc
<0.001
CI = confidence interval.
aPer 1-unit increment.
bPer 5-year increment.
cPer 50 x 103/mm3 increment.
admission platelet count, surgery in the previous
24 hours, and admission diagnoses of myocardial
infarction, gastrointestinal disease, gastro-
intestinal bleeding, musculoskeletal disease,
nervous system disease, respiratory disease,
sepsis, and unstable angina.
indication of how well the model discriminates
between those who did and those who did not
develop thrombocytopenia.21 The Hosmer-
Lemeshow goodness-of-fit-test p value equaled
0.52, indicating the model was a reasonable fit of
the observed data (p>0.05 indicates acceptance of
the null hypothesis that there is no significant
difference between estimated and observed
frequencies).18
Following multivariate modeling, eight
independent admission risk markers for the
development of thrombocytopenia were
identified (Table 3). An odds ratio (OR) greater
than 1 indicates that, controlling for the other
variables in the model, the odds of developing
thrombocytopenia were greater for patients with
that characteristic, diagnosis, or exposure. The
degree to which the risk is increased is the value
of the OR. For example, for patients with a
diagnosis of sepsis on admission, the odds of
developing thrombocytopenia during their ICU-
CCU stay was 14.1 times higher than those who
did not have a diagnosis of sepsis on admission.
Some ORs are expressed per increment (i.e.,
APACHE II, age, and admission platelet count in
this model). In these cases, for every unit
increase in that variable, the odds of developing
thrombocytopenia increased by the value of the
OR. For example, for each 1 unit increase in
APACHE II score, the odds of developing
thrombocytopenia increased by 1.1. Odds ratios
less that 1 indicate a decreased risk of thrombo-
cytopenia by the same principles just described.
The area under the receiver operating
characteristic (ROC) curve (C statistic) was 0.85
(95% CI 0.80–0.90), indicating good association
between the predicted probability of developing
thrombocytopenia and the actual observed cases
of thrombocytopenia (closer to 1.0 indicates
better predictive probability). The ROC curve is
calculated by using the model sensitivity and
specificity at various cut points to provide an
Risk Markers for Thrombocytopenia in the
Comprehensive Model
When exposures during the ICU-CCU stay
were considered in addition to admission
characteristics, univariate analyses yielded 25
candidate variables for regression analysis:
admission platelet count; age; APACHE II score;
history of moderate-to-high alcohol consumption;
hepatic dysfunction; surgery within the previous
24 hours; pulmonary artery catheter insertion;
packed red blood cell transfusion; fresh frozen
plasma transfusion; mean heparin dose; exposure
to aspirin, imipenem, salbutamol, inotropes,
cephalo-sporins, or histamine H2-receptor
antagonists; and diagnoses most responsible for
ICU-CCU stay of myocardial infarction,
gastrointestinal disease, gastrointestinal bleed,
infection, musculoskeletal disease (11 of 12
patients had trauma), nervous system disease,
nonsurgical respiratory disease, sepsis, and
unstable angina.
Multivariate regression analysis identified nine
independent risk markers for thrombocytopenia
(Table 4). Interpretation of the ORs follows the
same principles described in the previous section.
The presence of each risk marker substantially
increased or decreased the probability of
developing thrombocytopenia (Figure 1). The

808
PHARMACOTHERAPY Volume 22, Number 7, 2002
Table 4. Risk Markers Independently Associated with Thrombocytopenia in the
Comprehensive Model
Variable
Odds Ratio
95% CI
p Value
Fresh frozen plasma transfusion
Sepsis
20.0
15.1
9.5
8.4
4.1
2.5
2.3
0.4
0.4
2.0–199.3
3.1–74.4
2.6–34.6
3.9–17.9
1.1–16.0
0.9–6.7
0.9–5.7
0.2–1.0
0.3–0.6
0.01
0.001
0.001
<0.001
0.04
0.07
0.07
0.04
<0.001
Musculoskeletal diagnosis
Pulmonary artery catheter
Gastrointestinal diagnosis
Packed red blood cell transfusion
Nonsurgical respiratory diagnosis
Aspirin
Admission platelet counta
CI = confidence interval.
aPer 50 x 103/mm3 increment.
presence of several risk markers increased the
predicted risk even more dramatically. For
example, the regression model predicted that a
patient admitted with a platelet count of 200 x
103/mm3 who developed sepsis and had a
pulmonary artery catheter placed had a 94%
predicted probability of developing thrombo-
cytopenia, based on our definition. No two
variables were highly correlated, including
packed red blood cell and fresh frozen plasma
administration (Pearson correlation coefficient
0.41). Although the quantity of packed red
blood cells and fresh frozen plasma transfusions
varied among patients with thrombocytopenia
(packed red blood cells: median 2 units, range
1–12 units; fresh frozen plasma: median 2 units,
range 2–8 units), these were analyzed as
dichotomous variables because of the limited
number of patients exposed. The Hosmer-
Lemeshow goodness-of-fit-test p value equaled
0.15, and the area under the ROC curve (C
statistic) improved slightly, compared with that
of the admission model, to 0.89 (95% CI
0.85–0.93).
To further test the risk of thrombocytopenia
associated with heparin, heparin was entered into
the regression model using several dichotomous
variables: any exposure, low-dosage exposure
(< 1000 U/day), medium-dosage exposure
(1000–16,000 U/day), and high-dosage exposure
(> 16,000 U/day). Heparin also was analyzed as a
continuous variable by using the mean dose/day.
In the resultant models, heparin was not a
statistically significant independent risk marker,
and the predictive ability of the model showed no
improvement.
60
56
48
50
40
30
20
10
0
Clinical Outcomes
37
Twenty-six patients (7.2%) had a hemorrhage
at some time during their ICU-CCU stay;
however, the frequency of hemorrhage subsequent
to thrombocytopenia was lower than the overall
frequency in patients without thrombocytopenia.
Patients with thrombocytopenia had longer mean
lengths of ICU-CCU stay (12.3 ± 13.5 vs 4.5 ±
4.8 days, p<0.001) and hospital stay (32.0 ± 39.0
vs 14.5 ± 18.3 days, p<0.001), as well as a higher
ICU-CCU mortality (17.6% vs 4.4%, p<0.001)
and hospital mortality (22.1% vs 7.8%, p=0.001)
than those without thrombocytopenia (Table 5).
Of the 57 patients receiving heparin when
thrombocytopenia developed, 10 (18%) had
heparin discontinued specifically because of
thrombocytopenia (according to the prescribing
physician) within 24 hours of the first platelet
count below 150 x 103/mm3. Four of these 10
34
20
14
13
6
3
2
Risk Markers
Figure 1. Predicted probability of developing thrombo-
cytopenia for patients with individual risk markers in the
comprehensive model. “No risk marker” assumes the
platelet count is equal to the sample mean (246 x 103/mm3).
For “High admission platelets,” the platelet count is 1 SD
above the sample mean (i.e., 325 x 103/mm3).

RISK MARKERS FOR THROMBOCYTOPENIA Shalansky et al
809
Table 5. Clinical Outcomes in Patients with and Those without Thrombocytopenia
No
Thrombocytopenia Thrombocytopenia
Outcome
Hemorrhage, no. (%)
Thrombosis, no (%)
(n=68)
(n=294)
11 (3.7)
4 (1.4)
4.5 ± 4.8
14.5 ± 18.3
13 (4.4)
23 (7.8)
p Value
2 (2.9)a
0.003
<0.001
<0.001
<0.001
<0.001
0.001
3 (4.4)a
Length of ICU-CCU stay, mean ± SD (days)
Length of hospital stay, mean ± SD (days)
ICU-CCU mortality, no. (%)
Hospital mortality, no. (%)
12.3 ± 13.5
32.0 ± 39.0
12 (17.6)
15 (22.1)
aAfter developing thrombocytopenia.
patients were bleeding before or on the day of
developing thrombocytopenia.
than for patients in the CCU, which is consistent
with previous studies in these populations.1–5, 23
The frequency reported in this study is difficult
to compare with those in previous reports
because of the wide range of definitions used;
however, several other studies of critically ill
patients have used a definition of one platelet
count less than 100 x 103/mm3.1–5 With this
definition, reported frequencies ranged from
13%3 to 41%,4 and in our study the frequency
was 15.7% in the ICU population (Table 2). The
wide range of reported frequencies likely stems
from the different mix of patients in different ICU
settings, varying exposure to risk factors, and
varying frequency of platelet count determination.
The use of 100 x 103/mm3 versus 150 x 103/mm3
for the definition of thrombocytopenia is
somewhat arbitrary. The risk of spontaneous
bleeding is not significantly increased until the
platelet count decreases below 20 x 103/mm3,
although clinicians often intervene much
earlier.12, 24, 25 A relative decline in platelets is not
typically considered in studies evaluating the
occurrence of general thrombocytopenia.
Although a relative decrease in platelets may be a
sign of a specific immune-mediated thrombo-
cytopenia (e.g., HIT), the risk of bleeding is
associated with the absolute count. Previous
investigators generally have based their definition
of thrombocytopenia on one platelet count below
a specific threshold; however, we required two
consecutive platelet counts below 150 x
103/mm3. This definition limited the possibility
of overestimating the frequency of thrombo-
cytopenia based on one spurious platelet count
result. As identified in our results, the variability
of reported platelet counts is considerable.
Three patients with thrombocytopenia
developed thrombosis; however, none appeared
to have HIT. One patient had not received
heparin during his ICU-CCU stay or within the
previous 8 weeks. The other two patients had
received heparin but had negative results from
two assays recommended for the diagnosis of
HIT: the 14C-serotonin release assay and the
enzyme-linked immunosorbent assay for
heparin-dependent antiplatelet antibodies.22
The relatively high frequency of hemorrhage
before thrombocytopenia (19.1%) in this study
and the identification of bleeding episodes as a
risk marker for thrombocytopenia in another
study5 prompted further investigation of bleeding
as a potential risk marker. Logistic regression
analysis was carried out with the candidate
variables identified by the original univariate
analyses for the comprehensive model, with
bleeding episodes (hemorrhage plus gastro-
intestinal bleeding) replacing gastrointestinal
bleeding; however, the predictive performance of
the model was not improved.
Discussion
The risk markers identified in this study were
derived from the first, to our knowledge, large
prospective database specifically designed to
collect information relevant to thrombocytopenia
in this population. Variables associated with a
higher risk of thrombocytopenia in the compre-
hensive model included specific diagnoses
(sepsis, musculoskeletal, gastrointestinal, and
respiratory), fresh frozen plasma and packed red
blood cell transfusions, and pulmonary artery
catheter insertion. No drug, including unfrac-
tionated heparin, was identified as a positive
independent risk marker.
This is the first study, to our knowledge, to
evaluate risk markers for thrombocytopenia on
admission to an ICU or CCU. The predictive
potential of the admission model suggests that
information available on admission to the ICU or
The frequency of thrombocytopenia in this
study was much higher for patients in the ICU

810
PHARMACOTHERAPY Volume 22, Number 7, 2002
CCU can help identify patients who are more
likely to develop thrombocytopenia. However,
the improved predictive potential of the compre-
hensive model indicates that consideration of
potential risk markers encountered during the
ICU-CCU stay is also important. Although the
risk markers identified by the two models were
similar, the admission model included two
variables that did not appear in the compre-
hensive model: APACHE II score and gastro-
intestinal bleed. However, the comprehensive
model included variables associated with greater
severity of illness, including pulmonary artery
catheter insertion. In addition, transfusion of
packed red blood cells, identified as a risk marker
in the comprehensive model, may be a more
specific measure of blood loss than the diagnosis
of gastrointestinal bleed.
Many of the risk markers identified in this
study have logical pathophysiologic explanations,
and several have been associated with
thrombocytopenia in previous studies. However,
although these variables describe patients at
highest risk, regression analysis does not
establish causality. The strongest risk marker in
the comprehensive model was the administration
of fresh frozen plasma, which has been
demonstrated to reduce platelet count.26, 27
Whereas the conditions for which fresh frozen
plasma was administered may have contributed
to the development of thrombocytopenia (e.g.,
coagulation disorders), fresh frozen plasma has
been observed to increase platelet destruction by
macrophages through the effects of alloantibodies
directed against the glycoprotein IIIa platelet
receptor.26, 28–30
Packed red blood cell transfusions also were
identified as a risk marker in this study and were
associated with thrombocytopenia through
univariate analysis in two previous studies of
critically ill patients.1, 4 Patients receiving
massive red blood cell transfusions (more than a
patient’s normal blood volume within 24 hrs) can
experience significant dilutional thrombo-
cytopenia7, 31, 32; however, none of the patients
with thrombocytopenia in our study received a
massive blood transfusion. Nonetheless, since
the decline in platelet count is inversely related
to the number of units transfused,31, 33 even
moderate transfusions could have contributed to
the development of thrombocytopenia, based on
the definition used in our study. Significant
thrombocytopenia through splenic platelet
sequestration has been reported in patients
receiving fewer than five units of blood.34
Platelet consumption due to tissue injury also
may have contributed.31, 32
Sepsis has been identified as a risk marker for
thrombocytopenia in several previous studies1–3, 5
and has been associated with increased periph-
eral platelet destruction in the microvasculature,7,
disseminated intravascular coagulation,7, 36, 37
35
and toxic bone marrow suppression.7,
38
Musculoskeletal diagnoses (mostly trauma) and
gastrointestinal bleeding were associated with
thrombocytopenia in our study, likely owing to
blood loss and platelet consumption. Bleeding
has been identified as a risk marker for thrombo-
cytopenia5 and appeared to be a significant
explanatory variable in our exploratory
regression analysis. Respiratory failure and acute
respiratory distress syndrome were also risk
markers in our study and have been reported to
result in thrombocytopenia through reduced
platelet survival and increased sequestration in
the lungs and reticuloendothelial system.39–41
Pulmonary artery and central venous catheters
have been associated with a local or systemic
antiplatelet effect, as noted by many authors.2, 7,
42–46
Although no mechanism has been demon-
strated, presence of a foreign surface possibly
leads to direct platelet destruction.7 Pulmonary
artery catheters may be a marker for greater
disease severity, although these two variables
were not strongly correlated in our study.
Two variables, administration of aspirin and
higher admission platelet count, were associated
with a decreased risk of thrombocytopenia.
Patients with a high platelet count at admission
would require a large absolute decline in platelets
to be categorized as thrombocytopenic based on
our definition. Previous investigators who based
their definition on a threshold platelet level have
shown a lower admission platelet count to be a
strong risk marker for thrombocytopenia.5, 7 The
apparent protective effect of aspirin is more
difficult to explain. Although aspirin was
administered to a greater proportion of patients
in the CCU than in the ICU (85% vs 15%),
cardiac diagnosis did not displace aspirin from
the regression model, suggesting that aspirin
administration was not simply a surrogate
marker for cardiovascular conditions. Platelet
inhibitors such as aspirin may reduce platelet
consumption through inhibition of platelet
aggregation, although our literature search
revealed no published evidence that aspirin
reduces the risk of thrombocytopenia.
In accordance with earlier studies in this area,
thrombocytopenia was associated with higher

RISK MARKERS FOR THROMBOCYTOPENIA Shalansky et al
811
ICU-CCU and hospital mortality, as well as
longer ICU-CCU and hospital stay.1-5 This
observation was not adjusted for other deter-
minants of mortality or length of stay; however, a
recent study identified thrombocytopenia as an
independent marker for ICU mortality.6 Despite
this finding, thrombocytopenia is likely a risk
marker for, rather than a cause of, mortality.6
Previous studies investigating risk markers for
thrombocytopenia in critically ill patients
included fewer patients than our study, and most
were retrospective reviews.1–5 In these studies,
data were collected for a narrower range of
potential risk markers, and various definitions of
thrombocytopenia were used. These different
approaches may explain the inconsistent results,
although risk markers related to sepsis1, 5 and
severity of illness4, 5 have been identified by more
than one group of authors.
To our knowledge, this is the first study
published on this topic from a community
hospital. Therefore, the risk markers identified
in this study may not be applicable to critically ill
populations in other settings. However, the
patient mix appears comparable in some respects
with that of previous reports on this topic. The
mean APACHE II score was similar to the only
previous study that reported this variable (15.4
vs 17.5).5 A higher proportion of patients with
cardiovascular diagnoses were included in our
study owing to the combination of ICU-CCU,
although previous studies included up to 20% of
patients admitted for cardiovascular conditions.2
As among patients in the ICU in our study,
previous studies reported respiratory, infectious,
and gastrointestinal diagnoses as among the most
common in their study cohort.1, 2
There are several limitations to our study.
First, the low frequency of thrombocytopenia and
low prevalence of some risk markers do not allow
great precision in the estimation of ORs
(resulting in wide CIs), and some important risk
markers might have been missed. Therefore,
these results should be validated in a larger
cohort. Furthermore, bleeding was not assessed
prospectively as a potential independent risk
marker. Also, the definition of hemorrhage
included a moderate decline in hemoglobin (> 1
g/dl) that may not have resulted in substantial
bleeding. The influence of bleeding on the
probability of developing thrombocytopenia may
have been captured in other variables assessed
(e.g., gastrointestinal bleeding, packed red blood
cell transfusions); however, a more compre-
hensive assessment of bleeding as a risk marker
possibly would have resulted in an improved
and/or simpler model. We explored this
possibility by carrying out logistic regression
analysis using hemorrhage plus gastrointestinal
bleed as a single variable; however, the predictive
potential of the model was not improved.
Another limitation is that information for some
variables was obtained, at least in part, by word
of mouth (e.g., previous alcohol consumption,
previous heparin exposure). Whenever possible,
this information was confirmed through review
of the medical record. In addition, clinical
outcomes occurring after discharge from ICU-
CCU often were determined by retrospective
chart review, which may have compromised
accuracy. Finally, low-molecular-weight heparins
(LMWHs) were not administered frequently
during data collection, and administration of
LMWHs for venous thrombosis and acute
coronary syndromes has increased dramatically
in recent years.22 However, whereas LMWHs
may cause thrombocytopenia, the frequency
appears to be even lower than that for
unfractionated heparin.22
Our data can be used to improve clinical
decisions regarding thrombocytopenia in the
critical care setting. Some identified risk markers
are not amenable to change, including age,
diagnoses, and admission platelet count.
However, administration of transfusions and
insertion of pulmonary artery catheters could be
influenced. This is particularly relevant in light
of recent studies questioning the liberal use of
red blood cell transfusions and effectiveness of
pulmonary artery catheterization in critically ill
patients.47, 48
Whereas these results may help identify targets
for intervention in patients with thrombo-
cytopenia, they also may be used to prevent
unnecessary discontinuation of beneficial
therapies,10 particularly heparin. Patients in this
cohort had only a 6% predicted probability of
developing thrombocytopenia if they had the
mean admission platelet count (246 x 103/mm3)
and were not exposed to the identified risk
markers. The predicted risk of thrombocytopenia
increased to 56% if the patient was administered
fresh frozen plasma (Figure 1). As pointed out in
Results, the presence of several risk markers
increased the predicted risk even more
dramatically. A patient admitted with a platelet
count of 200 x 103/mm3 who developed sepsis
and had a pulmonary artery catheter placed had a
94% predicted probability of developing
thrombocytopenia. Although awareness of risk

812
PHARMACOTHERAPY Volume 22, Number 7, 2002
elsewhere.22, 53 Deciding whether to stop heparin
in critically ill patients is difficult given that most
patients with thrombocytopenia who are
receiving heparin do not have HIT. Whereas HIT
can be life-threatening, unnecessary discontinuation
of heparin can increase the risk of thrombosis.
Of course, in patients with active major bleeding,
heparin should be discontinued or the dosage
lowered. In our study, heparin was discontinued
in 10 (18%) patients with thrombocytopenia,
only four of whom had bleeding before or on the
day of developing thrombocytopenia. There were
no cases of HIT with thrombosis. Others have
reported frequent discontinuation of heparin in
patients with thrombocytopenia.2, 4, 9 The risk
markers for thrombocytopenia not induced by
heparin identified here and further data on the
occurrence of HIT in the critical care setting
could be used to identify patients for whom
discontinuation of heparin plus performance of
laboratory tests to confirm the diagnosis of HIT
should be performed.
factors may not lead to a specific intervention for
such a patient, other therapies could be
continued with the understanding that they are
unlikely to adversely affect platelet count.
Ideally, this model should be validated by using
data from other critical care settings before these
results are applied broadly for clinical use.
No specific drug or drug class was identified as
a positive risk marker for thrombocytopenia.
This finding is consistent with most studies in
this area, although one study did report
administration of inotropic agents and length of
histamine H2-receptor antagonist treatment as
independent markers.2 Intravenous heparin
frequently results in a reduction in platelet count
when administered to healthy volunteers.49–52 As
such, heparin often is implicated when
thrombocytopenia develops in critically ill
patients. Whereas previous studies have not
identified heparin as an independent risk marker
for thrombocytopenia,1–5 these reports did not
specify whether all heparin exposure was
accounted for, or if they relied on retrospective
chart review to calculate heparin administration.
In our study, all heparin exposure carefully was
documented prospectively, and heparin
administration was analyzed by using several
approaches. The results confirm that heparin’s
mild proaggregatory effect on platelets is
unimportant when other risk markers for
thrombocytopenia are present.
Heparin-induced thrombocytopenia, an
immune-mediated syndrome that carries
substantial risk of life-threatening thrombosis,
must be considered when patients receiving
heparin develop thrombocytopenia. Although
the frequency of HIT has been estimated at
1–3%,22 this estimate varies considerably among
different populations and has not been
established for patients in the ICU or CCU. The
diagnosis of HIT in the critical care setting is
further complicated by the high frequency of
thrombocytopenia not induced by heparin, and
the wide range of potential contributing factors.
Unfortunately, guidelines on diagnosis of HIT do
not identify factors, other than heparin, that
should be considered.22, 53 Guidelines suggest
that in patients who have clinical signs of HIT,
particularly those with thrombocytopenia and
new thrombosis, heparin should be discontinued
and alternate anticoagulation therapy begun.53
Assays to aid in the diagnosis of HIT are
recommended, but their expense and limited
availability make routine use difficult. A detailed
review of HIT diagnosis can be found
Conclusions
Critically ill patients with specific diagnoses
(sepsis, musculoskeletal, gastrointestinal, and
respiratory) and those who receive transfusions
(fresh frozen plasma or packed red blood cells)
or pulmonary artery catheters are more likely to
develop thrombocytopenia. Exposure to
unfractionated heparin is not a risk marker,
although discontinuation of heparin should be
considered if patients are bleeding and when
clinical signs of HIT are present. These results
narrow the extensive list of factors previously
reported to be associated with thrombocytopenia
and help define when intervention may or may
not be appropriate. Additional research on the
validity of this model in other critically ill
populations, and on the frequency of HIT in
critically ill patients, would further clarify
treatment decisions in patients with thrombo-
cytopenia.
References
1. Baughman RP, Lower EE, Flessa HC, Tollerud DJ. Thrombo-
cytopenia in the intensive care unit. Chest 1993;104:1243–7.
2. Bonfiglio MF, Traeger SM, Kier KL, Martin BR, Hulisz DT,
Verbeck SR. Thrombocytopenia in intensive care patients: a
comprehensive analysis of risk factors in 314 patients. Ann
Pharmacother 1995;29:835–42.
3. Cawley MJ, Wittbrodt ET, Boyce EG. Potential risk factors
associated with thrombocytopenia in a surgical intensive care
unit. Pharmacotherapy 1999;19:108–13.
4. Hanes SD, Quarles DA, Boucher BA. Incidence and risk factors
of thrombocytopenia in critically ill trauma patients. Ann
Pharmacother 1997;31:285–9.

RISK MARKERS FOR THROMBOCYTOPENIA Shalansky et al
813
5. Stephan F, Hollande J, Richard O, Cheffi A, Maier-
Redelsperger M, Flahault A. Thrombocytopenia in a surgical
ICU. Chest 1999;115:1363–70.
6. Vanderschueren S, De Weerdt A, Malbrain M, et al.
Thrombocytopenia and prognosis in intensive care. Crit Care
Med 2000;28:1871–6.
7. Bogdonoff DL, Williams ME, Stone DJ. Thrombocytopenia in
the critically ill patient. J Crit Care 1990;5:186–205.
8. Wazny LD, Ariano RE. Evaluation and management of drug-
induced thrombocytopenia in the acutely ill patient.
Pharmacotherapy 2000;20:292–307.
9. Kruse JA. Review: heparin reduces central venous and
pulmonary catheter clots. ACP J Club 1998;129:5.
10. Shalansky SJ, Verma AK, Levine M. Factors to consider before
discontinuing heparin in patients who develop thrombo-
cytopenia. Pharmacotherapy 1999;19:1011–12.
11. Williams WJ. Classification and clinical manifestations of
disorders of hemostasis. In: Beutler E, Lichtricn MA, Coller BS,
et al, eds. Williams hematology. 5th ed. Toronto: McGraw-Hill,
1995:1276–81.
12. Handin RI. Disorders of platelets and vessel wall. In:
Isselbacher KJ, Braunwald E, Wilson JD, et al, eds. Harrison’s
principles of internal medicine, 13th ed. Toronto: McGraw-Hill,
1994:1798–803.
thrombocytopenia following infusion of plasma containing
anti-PIA1. Am J Hematol 1987;25:219–21.
30. Scott EP, Moilan-Bergeland J, Dalmasso AP. Posttransfusion
thrombocytopenia associated with passive transfusion of a
platelet-specific antibody. Transfusion 1988;28:73–6.
31. McCullough J. Transfusion therapy in specific clinical
situations. In: McCullough J, ed. Transfusion medicine. New
York: McGraw-Hill, 1998:275–317.
32. Slichter SL. Platelet production, physiology, hemostasis and
transfusion therapy. In: Spiess BC, Counts RB, Gould SA, eds.
Perioperative transfusion medicine. Baltimore, MD: Williams &
Wilkins, 1998:61–77.
33. Counts RB, Haisch C, Simon TL, et al. Hemostasis in
massively transfused trauma patients. Ann Surg 1979;190:91–9.
34. Bareford D, Chandler ST, Hawker RJ, et al. Splenic platelet
sequestration following routine blood transfusion is reduced by
filtered/washed product. Br J Haematol 1987;67:177–80.
35. Gawaz M, Fateh-Moghadam S, Pilz G, et al. Platelet activation
and interaction with leucocytes in patients with sepsis or
multiple organ failure. Eur J Clin Invest 1995;25:843–51.
36. Neame PB, Kelton JG, Walker IR, et al. Thrombocytopenia in
septicemia: the role of disseminated intravascular coagulation.
Blood 1980;56:88–92.
37. Kelton JG, Neame PB, Gauldie J, et al. Elevated platelet-
associated IgG in the thrombocytopenia of septicemia. N Engl J
Med 1979;300:760–4.
13. Puckett CD. The educational annotation of ICD-9-CM. Reno,
NV: Channel Publishing, 1998.
14. American College of Chest Physicians and Society of Critical
Care Medicine. Definitions for sepsis and organ failure and
guidelines for the use of innovative therapies in sepsis. Crit
Care Med 1992;20:864–74.
15. Rubin R. Effect of ethanol on platelet function. Alcohol Clin
Exp Res 1999;23:1114–18.
16. Levine RF, Spivak JL, Meagher R, Sieber F. Effect of ethanol
on thrombopoiesis. Br J Haematol 1986;62:345–54.
17. Cockcroft DW, Gault MH. Prediction of creatinine clearance
from serum creatinine. Nephron 1976;16:31–41.
18. Hosmer DW, Lemeshow S. Applied logistic regression. New
York: John Wiley and Sons, 1989.
19. Concato J, Feinstein AR, Holford TR. The risk of determining
risk with multivariable models. Ann Intern Med
1993;118:201–10.
20. Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE
II: A severity of disease classification system. Crit Care Med
1985;13:818–27.
38. Bessman DJ, Gardner FH. Platelet size in thrombocytopenia
due to sepsis. Surg Gynecol Obstet 1983;156:177–80.
39. Schneider RC, Zapol WM, Carvalho A. Platelet consumption
and sequestration in severe acute respiratory failure. Am Rev
Respir Dis 1980;122:445–51.
40. Heffner JE, Sahn SA, Repine JE. The role of platelets in the
adult respiratory distress syndrome: culprit or bystanders? Am
Rev Respir Dis 1987;135:482–92.
41. Bone R, Francis P, Pierce A. Intravascular coagulation
associated with the adult respiratory distress syndrome. Am J
Med 1976;61:585–9.
42. Kim YL, Richman KA, Marshall BE. Thrombocytopenia
associated with Swan-Ganz catheterization in patients.
Anesthesiology 1980;53:261–2.
43. Miller JJ, Venus B, Mathru M. Comparison of the sterility of
long-term central venous catheterization using single lumen,
triple lumen, and pulmonary artery catheters. Crit Care Med
1984;12:634–7.
21. Hanley JA, McMeil BJ. The meaning and use of the area under
a receiver operating characteristic (ROC) curve. Radiology
1982;143:29–36.
22. Hirsh J, Warkentin TE, Shaughnessy SG, et al. Heparin and
low-molecular-weight heparin: mechanisms of action,
pharmacokinetics, dosing, monitoring, efficacy, and safety.
Chest 2001;119(suppl):S64–94.
23. McClure MW, Berkowitz SD, Sparapani R, et al. Clinical
significance of thrombocytopenia during a non-ST-elevation
acute coronary syndrome: the platelet glycoprotein IIb/IIIa in
unstable angina—receptor suppression using Integrilin therapy
(PURSUIT) trial experience. Circulation 1999;99:2892–900.
24. Davis GI. Quantitative and qualitative disorders of platelets. In:
Steine-Martin EH, Lotspeich-Steiner CA, Koepke JA, eds.
Clinical hematology: principles, procedures and correlations.
New York: Lippincott, 1998:717–34.
44. Layon AJ. The pulmonary artery catheter: nonexistential entity
or occasionally useful tool? Chest 1999;115:859–62.
45. McNulty SE, Maguire DP, Thomas RE. Effect of heparin-
bonded pulmonary artery catheters on the activated
coagulation time. J Cardiothorac Vasc Anesth 1998;12:533–5.
46. Rull JR, Aguirre JL, de la Puerta E, et al. Thrombocytopenia
induced by pulmonary artery flotation catheters: a prospective
study. Intensive Care Med 1984;10:29–31.
47. Connors AF, Speroff T, Dawson NV, et al. The effectiveness of
right heart catheterization in the initial care of critically ill
patients. JAMA 1996;276:889–97.
48. Hebert PC, Wells G, Blajchman MA, et al. A multicenter,
randomized, controlled clinical trial of transfusion
requirements in critical care. N Engl J Med 1999;340:409–17.
49. Gollub S, Ulin AW. Heparin-induced thrombocytopenia in
man. J Lab Clin Med 1962;59:430–5.
25. Lind SE. The hemostatic system. In: Handin RI, Lux SE, Stossel
TP, eds. Blood. Principles and practice of hematology.
Philadelphia: Lippincott, 1995:949–73.
26. Brunner-Bolliger S, Kiefel V, Horber FF, et al. Antibody studies
in a patient with acute thrombocytopenia following infusion of
plasma containing anti-PIA1. Am J Hematol 1997;56:119–21.
27. Noe DA, Graham SM, Luff R, et al. Platelet counts during
rapid massive transfusions. Transfusion 1982;22:392–5.
28. Nugent DJ. Alloimmunization to platelet antigens. Semin
Hematol 1992;29:83–8.
50. Davey MG, Lander H. Effect of infected heparin on platelet
levels in man. J Clin Path 1968;21:55–9.
51. Saffle JR, Russo J, Dukes GE, et al. The effect of low-dose
heparin therapy on serum platelet and transaminase level. J
Surg Res 1980;28:297–305.
52. Schwartz KA, Royer G, Kaufman DB, et al. Complications of
heparin administration in normal individuals. Am J Hematol
1985;19:355–63.
53. Warkentin TE, Barkin RL. Newer strategies for the treatment
of heparin-induced thrombocytopenia. Pharmacotherapy
1999;19:181–95.
29. Nijjar TS, Bonacosa IA, Israels LG. Severe acute