Severe Community-acquired Pneumonia

Use of Intensive Care Services and Evaluation of American and British Thoracic
Society Diagnostic Criteria
Derek C. Angus, Thomas J. Marrie, D. Scott Obrosky, Gilles Clermont, Tony T. Dremsizov, Christopher Coley,
Michael J. Fine, Daniel E. Singer, and Wishwa N. Kapoor

The CRISMA (Clinical Research, Investigation, and Systems Modeling of Acute Illness) Laboratory, Department of Critical Care Medicine, and
Division of General Internal Medicine, Department of Medicine, University of Pittsburgh School of Medicine; Center for Research on Health
Care, University of Pittsburgh Medical Center; and Division of Health Policy and Management, Department of Health Services Administration,
Graduate School of Public Health, University of Pittsburgh, Pittsburgh, Pennsylvania; Department of Medicine, University of Alberta,
Edmonton, Alberta, Canada; Harvard University Health Services, Cambridge; and Division of General Internal Medicine, Department
of Medicine, Massachusetts General Hospital, Boston, Massachusetts

Despite careful evaluation of changes in hospital care for commu- The appropriate management of community-acquired pneu-
nity-acquired pneumonia (CAP), little is known about intensive care monia (CAP) has received close attention in the current era
unit (ICU) use in the treatment of this disease. There are criteria of rising health care costs (1, 2). Considerable efforts have
that define CAP as “severe,” but evaluation of their predictive value been made to shorten unnecessary hospital length of stay
is limited. We compared characteristics, course, and outcome of
(LOS) and optimize the initial decision to hospitalize (3–7).
inpatients who did (n ⫽ 170) and did not (n ⫽ 1,169) receive ICU
However, most of these efforts were designed to reduce un-
care in the Pneumonia Patient Outcomes Research Team prospec-
tive cohort. We also assessed the predictive characteristics of four necessary care for less sick patients. Less attention has been
prediction rules (the original and revised American Thoracic Society paid to patients with severe CAP, such as those requiring
criteria, the British Thoracic Society criteria, and the Pneumonia care in an intensive care unit (ICU). Several authors studied
Severity Index [PSI]) for ICU admission, mechanical ventilation, CAP in ICU patients but focused mainly on microbiologic
medical complications, and death (as proxies for severe CAP). ICU etiology (8, 9) or short-term mortality (8–12). Few studies
patients were more likely to be admitted from home and had more compared patients managed with and without ICU care (11,
comorbid conditions. Reasons for ICU admission included respira- 13–20), and those that did were generally of small sample
tory failure (57%), hemodynamic monitoring (32%), and shock size, were not recent, were from outside North America, or
(16%). ICU patients incurred longer hospital stays (23.2 vs. 9.1 days, provided few data comparing the two groups.
p ⬍ 0.001), higher hospital costs ($21,144 vs. $5,785, p ⬍ 0.001),
In a 1993 Consensus Statement designed to standardize
more nonpulmonary organ dysfunction, and higher hospital mortal-
ity (18.2 vs. 5.0%, p ⬍ 0.001). Although ICU patients were sicker,
and improve care, the American Thoracic Society (ATS)
27% were of low risk (PSI Risk Classes I–III). Severity-adjusted ICU defined a subset of CAP as “severe” on the basis of the
admission rates varied across institutions, but mechanical ventila- presence of specific risk factors, or criteria, and recommended
tion rates did not. The revised American Thoracic Society criteria that ICU admission be considered for these patients (21).
rule was the best discriminator of ICU admission and mechanical These criteria were evaluated in one study from Spain (22)
ventilation (area under the receiver operating characteristic curve, and reported in an abstract from one North American study
0.68 and 0.74, respectively) but none of the prediction rules were (23). Both studies suggested the definition of severe CAP
particularly good. The PSI was the best predictor of medical compli- was overly sensitive and nonspecific. In response, a second
cations and death (area under the receiver operating characteristic ATS consensus panel modified the definition of severe CAP,
curve, 0.65 and 0.75, respectively), but again, none of the prediction based in part on the Spanish classification of risk factors as
rules were particularly good. In conclusion, ICU use for CAP is com-
major or minor (22), and recommended evaluation of these
mon and expensive but admission rates are variable. Clinical predic-
revised criteria (24). There are two other clinical prediction
tion rules for severe CAP do not appear adequately robust to guide
clinical care at the current time.
rules for CAP: the British Thoracic Society (BTS) criteria
(25) and the Pneumonia Severity Index (PSI) (6). The relative
Keywords: artificial ventilation; community-acquired infections; inten- merits of these different rules have not been assessed in a
sive care; outcomes assessment; pneumonia common data set.
The goal of this article is twofold: first, to provide a descrip-
tion of differences in baseline characteristics, processes of
care, and medical outcomes between hospitalized patients
(Received in original form February 21, 2001; accepted in final form May 1, 2002) who do and do not receive ICU care in the Pneumonia Patient
Supported by the Agency for Health Care Policy and Research (R01 HSO 6468),
Outcomes Research Team (PORT) prospective cohort, a
the National Institute of General Medical Sciences (R01 GM61992-01), and an North American cohort enrolled from 1991 to 1994; and
unrestricted educational grant from Amgen (Thousand Oaks, CA). M.J.F. was also second, to evaluate the predictive characteristics of the origi-
supported as a Robert Wood Johnson Foundation Generalist Physician Faculty nal and revised ATS criteria, the BTS criteria, and the PSI
Scholar. for ICU admission, mechanical ventilation, medical compli-
Correspondence and requests for reprints should be addressed to Derek C. Angus, cations, and death—four proxies for “severe” CAP.
M.D., M.P.H., Room 604, Scaife Hall, Department of Critical Care Medicine,
University of Pittsburgh, 200 Lothrop Street, Pittsburgh, PA 15213. E-mail: angusdc@
ccm.upmc.edu METHODS
This article has an online data supplement, which is accessible from this issue’s Patient Characteristics
table of contents online at www.atsjournals.org
We studied the inpatients of the Pneumonia PORT cohort study
Am J Respir Crit Care Med Vol 166. pp 717–723, 2002
DOI: 10.1164/rccm.2102084 (4) at three U.S. and one Canadian sites. Patients were ⭓ 18
Internet address: www.atsjournals.org years of age, had clinical and radiographic evidence of pneumo-
718 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 166 2002

nia within 24 hours of presentation, and provided informed con- TABLE 1. SELECTED BASELINE CHARACTERISTICS
sent. We assessed characteristics through chart review and inter- ICU Admissions
views, using standardized data collection instruments (6, 26, 27). Characteristic n (%, n)* p Value
We quantified severity of illness by using the PSI (6).
Cohort 1,339 12.7, 170
Hospital Course Living arrangements
Private residence, alone 275 13.1, 36 0.011
We assessed hospital length of stay (LOS) and cost (U.S. sites only; Private residence, with others 815 14.5, 118
determined from detailed billing records [28]), ICU use, mechanical Nursing home/chronic care facility 184 6.5, 12
ventilation use, laboratory investigations, and antibiotic therapy. We Other† 64 6.3, 4
collected data on all medical complications within 30 days of presenta- Employment status
tion. We considered worsening of chronic conditions as a complication. Employed 218 7.8, 17 0.017
We defined shock as a systolic blood pressure ⬍ 90 mm Hg despite fluid Not employed 1,118 13.7, 153
resuscitation or vasopressor requirement. We noted the development of Significant comorbid conditions
acute organ dysfunction for each of six organ systems as defined pre- Chronic pulmonary disease‡ 451 15.5, 70 0.029
viously (see expanded Methods in the online data supplement) (29). Coronary artery disease 349 16.6, 58 0.010
Alcohol or intravenous drug abuse 260 16.5, 43 ⬍ 0.001
Outcomes Congestive heart failure 225 19.1, 43 0.002
Renal disease 139 18.0, 25 0.049
We determined survival at 30 and 90 days. Two investigators indepen- Dementia 133 3.8, 5 0.001
dently reviewed detailed case summaries of all deaths, based on medical None of the above 327 5.5, 18 ⬍ 0.001
record review and interviews with caregivers and family members, and Number of comorbid conditions
assigned the cause of death (30) according to World Health Organiza- 0 224 7.1, 16 ⬍ 0.001
tion criteria (31). We recorded return to work for those previously 1 301 11.0, 33
employed and return to usual activities. 2 or 3 568 14.1, 80
⭓4 246 16.7, 41
Criteria for Severe CAP Do-not-resuscitate orders at presentation 199 5.5, 11 0.001
We determined the presence of each of the seven original ATS risk Severity of illness (PSI)
factors (tachypnea, respiratory failure, mechanical ventilation, bilateral Risk Class I 184 6.0, 11 ⬍ 0.001
Risk Class II 233 5.6, 13
or multilobar pneumonia by chest radiograph, shock, vasopressor ther-
Risk Class III 253 8.7, 22
apy, and renal impairment, as previously defined; see expanded Meth-
Risk Class IV 446 15.9, 71
ods in the online data supplement [21]) and the three BTS risk factors
Risk Class V 223 23.8, 53
(respiratory rate ⭓ 30/minute, diastolic blood pressure ⬍ 60 mm Hg,
and serum urea ⬎ 7 mM) (25) at baseline. Definition of abbreviations: ICU ⫽ intensive care unit; PSI ⫽ pneumonia severity
To define severe CAP by original ATS criteria, any one of the seven index.
risk factors must be present. To define severe CAP by revised ATS * Missing data were excluded from the denominator; data were missing for ⬍
criteria, two of three minor criteria (systolic blood pressure ⭐ 90 mm Hg, 1% of patients for all variables except alcohol or intravenous drug abuse (22.6%).
multilobar disease, or PaO2/FiO2 ⬍ 250) or one of two major criteria †
Other living arrangements include group settings and the homeless.
‡
(mechanical ventilation or shock) must be present (22). To define severe Chronic pulmonary disease was defined as chronic obstructive pulmonary
CAP by the BTS criteria, any two of the three risk factors must be disease, interstitial or restrictive lung disease, or asthma.
present (25).

Statistical Analysis
We compared categorical data using the ␹2 statistic or the Fisher exact and patients with dementia (Table 1). Of those admitted to
test (32) and continuous data using the Student t test (33) or Mantel–Cox an ICU, 68% were admitted on Day 1 and 79% by Day 3 of
log-rank test (34). We compared time to return to work and time to hospitalization. The principal reasons for ICU admission were
return to usual activities by Kaplan–Meier estimation (35). We built respiratory failure (57%), hemodynamic monitoring (32%), and
logistic regression models to compare severity-adjusted ICU admission
shock (16%). There were no differences in age (63.2 vs. 64.5,
rates and mechanical ventilation rates across centers (34).
To determine how well the prediction rules predicted an episode p ⫽ 0.46) between ICU and non-ICU patients, but males were
of CAP that was “severe,” we determined the relative risk (as a measure admitted more frequently to the ICU (14.0 vs. 10.3%; p ⫽ 0.04).
of association between the risk factor and outcome), and sensitivity, Patients admitted to the ICU were more likely to complain
specificity, positive and negative predictive values, and receiver op- of dyspnea and have tachypnea, tachycardia, hypothermia, or
erating characteristic (ROC) curves (as measures of discrimination) for altered mental status at presentation than non-ICU patients (see
four events: ICU admission, mechanical ventilation, development of a Table E1 in the online data supplement). There were no differ-
medical complication, and death. We dichotomized the PSI as low ences, however, in the total number of symptoms, symptom
(Classes I–III) or high (Classes IV and V) risk for these analyses. We
bother, or severity of symptom scores between the two groups
used the values at hospital admission for each criteria assessment. We
assumed statistical significance at p ⬍ 0.05 and conducted analyses in
(see Table E2 in the online data supplement). Abnormal labora-
SAS (SAS Institute, Cary, NC) and SPSS (SPSS, Chicago, IL). tory values were also more common and the chest radiograph
was more likely to show extensive disease (see Table E1 in the
RESULTS online data supplement). The etiologic pattern was similar in
both ICU and non-ICU patients (p ⫽ 0.19). Only Streptococcus
Baseline Characteristics pneumoniae (14.7%), Haemophilus influenzae (4.7%), and Staph-
Of the 1,339 inpatients in the study cohort, 12.7% (n ⫽ 170) ylococcus aureus (4.1%) were reported at a rate of ⬎ 2% in
were admitted to the ICU, with ICU admission rates ranging ICU patients. A specific organism was more commonly identified
from 8.8 to 26.1% across participating centers (p ⫽ 0.005). We in ICU patients but less than half of either group had a microbio-
found higher ICU admission rates for patients admitted from logic etiology (44.7 vs. 33.3%, p ⫽ 0.002). Patients with high risk
home, patients who were unemployed, patients with a history of death (PSI Risk Classes IV and V) were more likely to be
of substance abuse, and patients with underlying disease. We admitted to the ICU. However, 27% (n ⫽ 46) of the ICU admis-
found lower ICU admission rates for patients admitted from sions were for patients classified as low risk at presentation (Risk
nursing homes, patients with prior “do not resuscitate” orders, Classes I–III) (Table 1).
Angus, Marrie, Obrosky, et al.: Community-acquired Pneumonia and the ICU 719

Figure 1. Median daily hospital costs (U.S. dollars) by survi-

vor and ICU cohort. Costs were available for 846 U.S. pa-
tients. See METHODS in the online data supplement for details
regarding calculation of hospital costs.

Processes of Care Medical Complications and Organ Dysfunction

As might be expected, ICU patients received a more aggressive The ICU patients had a severalfold increase in most pulmonary
diagnostic work-up than the non-ICU patients, including signifi- and nonpulmonary complications compared with non-ICU pa-
cantly more frequent gram stains, sputum culture, pleural taps, tients (see Table E5 in the online data supplement). The cardiac
bronchoscopies, and serologic studies (data not shown) (p ⬍ complications were particularly notable, with half of all patients
0.05 for each comparison). Antibiotic management was also con- developing shock (47.6%), half showing signs of worsened con-
siderably more intense for ICU patients, with ICU patients re- gestive heart failure (51.2%), and a quarter developing atrial
ceiving twice the number of antibiotics as non-ICU patients arrhythmias (26.5%). Anemia (28.8%), abnormal liver function
(4.2 ⫾ 2.2 vs. 2.6 ⫾ 1.3, p ⬍ 0.001). Virtually all classes of tests (30.0%), and renal impairment (32.4%) were also common
in ICU patients. This trend to higher complications in those
antibiotics were prescribed more commonly in the ICU (see
admitted to the ICU is further reflected in the distribution of
Table E3 in the online data supplement).
acute nonpulmonary organ dysfunction, as shown in Figure 2.
ICU patients incurred longer overall hospital LOS than non-
ICU patients (23.2 ⫾ 26.5 vs. 9.1 ⫾ 9.3 days, p ⬍ 0.001), with Medical Outcomes
a mean ICU LOS of 7.1 days (median, 3 days). Hospital costs Medical outcomes are detailed in Table 2. Mortality was almost
(available for 846 U.S. patients) were also significantly higher, four times higher in ICU patients than in non-ICU patients.
with median costs of $21,144 versus $5,785 for ICU and non- Mortality rose with increasing risk class. Pneumonia was deter-
ICU patients, respectively (p ⬍ 0.001). This difference was due mined as the primary cause of hospital death in roughly three-
to increased LOS (see above), higher daily costs (see Table E4 quarters of both ICU (19 of 26 deaths, 73.1%) and non-ICU
in the online data supplement), and differences in survival (see patients (46 of 62 deaths, 74.2%). After discharge, both groups
below). The higher daily costs were seen across all cost centers
and did not simply reflect the increased costs of an ICU bed
(Figure 1). In the ICU, nonsurvivors had an LOS similar to that
of survivors but much higher daily costs (median daily hospital
costs for ICU patients: $2,168 vs. $1,343 for nonsurvivors and
survivors, respectively; p ⬍ 0.001). Total hospital costs were
$35,346 for ICU nonsurvivors and $20,347 for ICU survivors.
Although only 13.5% of patients received ICU care, they ac-
counted for 42.9% of total hospital costs.
In a multivariate regression model for ICU admission, which
had good fit (C statistic of 2.24 with 5 degrees of freedom,
p ⫽ 0.81), mechanical ventilation before admission, respiratory
failure, tachypnea, renal impairment, and vasopressor require-
ment were independently predictive. Using this model to control
for differences in case mix, significant differences in ICU admis-
sion rates persisted across sites. We constructed a similar model
to predict the use of mechanical ventilation, again with good fit
(C statistic of 2.19 with 4 degrees of freedom, p ⫽ 0.7). Indepen-
dent predictors of mechanical ventilation were respiratory fail-
ure, tachypnea, and vasopressor requirement. Of interest, al-
though there was a twofold variation in unadjusted mechanical Figure 2. Distribution of nonrespiratory acute organ dysfunction in pa-
ventilation rates across sites (range, 28–56%; p ⫽ 0.03), there tients with and without ICU care. See METHODS in the online data supple-
was no difference after controlling for case mix. ment for definitions of acute organ dysfunction.
720 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 166 2002

In predicting ICU admission, individual risk factors were

generally specific, with high negative predictive value, but insen-
sitive, with poor positive predictive value. This is because each
risk factor usually required ICU admission, but there were many
different risk factors. For example, most patients who require
vasopressors are admitted to the ICU, but there are many pa-
tients admitted to the ICU who are not receiving vasopressors.
In contrast to individual risk factors, the different clinical
prediction rules generally had better sensitivity for ICU admis-
sion, because they captured more than one possible cause for
ICU admission. The revised ATS criteria had the best overall
discrimination, as measured by ROC curves, but none of the
rules were particularly good (ranging from 0.58 to 0.68, where
0.5 occurs by random chance alone) because many non-ICU
patients met criteria. For example, 60% (n ⫽ 804) of all inpatients
met original ATS criteria for severe CAP, 83% of whom (n ⫽
665) were never admitted to an ICU. One-third (n ⫽ 440) of
all inpatients met the revised ATS criteria, 74% of whom (n ⫽
324) were never admitted to an ICU; 24% (n ⫽ 321) of all
inpatients met the BTS criteria, 80% of whom (n ⫽ 256) were
never admitted to an ICU. The revised ATS criteria were a good
Figure 3. Kaplan–Meier survival curves for ICU and non-ICU patients. discriminator for the need for mechanical ventilation, whereas
Survival was lower in ICU patients (p ⬍ 0.05). the PSI was a good discriminator for death. The performance
of the different rules was consistent across hospitals (Figure 4;
and see Table E6 in the online data supplement).
incurred further mortality but the rates were comparable (mor-
tality between discharge and 90 days was 6.6 vs. 8.3% for ICU
DISCUSSION
vs. non-ICU patients, respectively; p ⫽ 0.5) (Figure 3). Most
patients discharged alive returned home, but the rate was lower Despite the considerable attention to CAP, comparatively little
for ICU patients. In addition, ICU patients had a slower recovery is known about current ICU use in the treatment of this disease.
and a lower proportion of patients were either back at work or Our study demonstrated several important points. Not surpris-
usual activities by Day 30 (Table 2). ingly, ICU patients were sicker, as reflected by several baseline
criteria, and had poorer outcomes and greater resource use.
Predictive Characteristics of Different Clinical Prediction Rules Although the number of patients receiving ICU care was only
for Severe CAP a small proportion of all patients with CAP, they consumed
The predictive characteristics of baseline individual ATS risk more than one-third of all hospital costs for CAP. The likelihood
factors, the original ATS criteria (any one risk factor), the revised of receiving ICU care was poorly predicted by most measures
ATS criteria, the BTS criteria, and high-risk PSI scores (PSI of severity, raising the possibility that the ICU admission decision
Risk Classes IV and V) are presented in Table 3. may be rather discretionary and influenced by local practice

TABLE 2. OUTCOMES FOR PATIENTS WITH AND WITHOUT ICU CARE

Characteristic Non-ICU ICU p Value

Hospital mortality, %
All 5.0 18.2 ⬍ 0.001*
Risk Class I 0.0 0.0 NA
Risk Class II 0.9 7.7 0.16
Risk Class III 0.4 4.6 0.17
Risk Class IV 5.1 21.1 ⬍ 0.001
Risk Class V 21.2 26.4 0.43
Pneumonia as major cause of hospital death, % of deaths 74.2 73.1 0.91
Mortality by 30 days, % 6.9 15.3 ⬍ 0.001
Mortality by 90 days, % 13.0 24.7 ⬍ 0.001
Discharge location, % of hospital survivors
Home 83.1 71.0 ⬍ 0.001
Nursing home 16.7 22.9
Other institution 0.3 6.1
RTW by 30 days,† % of those who worked before onset of pneumonia 70.5 25.9 0.019
Time for patients to RTW, median (days) 21 —
RTUA by 30 days,‡ % 65.0 38.3 ⬍ 0.001
Time for patients to RTUA, median (days) 20 —
Hospital readmission within 30 days of presentation, % 10.5 6.9 0.25

Definition of abbreviations: ICU ⫽ intensive care unit; NA ⫽ not available; RTUA ⫽ return to usual activities; RTW ⫽ return to
work.
* p Value remains ⬍ 0.001 after stratifying for risk class.
†
Complete RTW data available for 75.8% of the 215 inpatients employed at the time of presentation and alive at 30 days.
‡
Complete RTUA data available for 68.9% of the 1,232 inpatients alive at 30 days.
Angus, Marrie, Obrosky, et al.: Community-acquired Pneumonia and the ICU 721

TABLE 3. CHARACTERISTICS OF DIFFERENT CLINICAL PREDICTION RULES FOR SEVERE

COMMUNITY-ACQUIRED PNEUMONIA
Sensitivity Specificity PPV NPV
Event* (%) (%) ROC (95% CI) (%) (%) RR (95% CI)

ICU admission
Presence of ATS risk factor
Respiratory rate† 34.1 82.7 22.3 89.6 2.1 (1.5–3.1)
Respiratory failure‡ 56.5 69.5 21.2 91.6 2.5 (1.8–3.5)
Mechanical ventilation 6.5 100.0 100.0 88.0 8.4 (4.4–15.7)
Bilateral/multilobe X-ray§ 37.1 73.7 17.0 89.0 1.5 (1.1–2.2)
Shock 4.7 96.9 18.2 87.5 1.5 (0.7–3.2)
Vasopressor therapy
10.0 96.3 28.3 88.0 2.4 (1.3–4.3)
Renal impairment# 20.0 93.5 30.9 88.9 2.8 (1.8–4.3)
Original ATS criteria 81.8 43.1 0.61 (0.57–0.65) 17.3 94.2 3.0 (2.0–4.5)
Revised ATS criteria 70.7 72.4 0.68 (0.64–0.73) 26.4 94.7 4.9 (3.4–7.1)
BTS criteria 39.6 78.2 0.58 (0.53–0.63) 20.2 90.3 2.1 (1.5–2.9)
High PSI (Risk Class IV or V) 72.9 53.4 0.60 (0.56–0.65) 18.5 93.1 2.7 (1.9–3.9)
Mechanical ventilation**
Original ATS criteria 86.2 42.3 0.64 (0.58–0.69) 10.2 97.6 4.2 (2.3–7.6)
††
Revised ATS criteria 100.0 72.8 0.74 (0.69–0.79) 21.9 100.0
BTS criteria 51.1 78.0 0.64 (0.58–0.71) 15.0 95.4 3.3 (2.1–5.0)
High PSI (Risk Class IV or V) 53.8 50.5 0.63 (0.58–0.69) 7.6 93.6 1.2 (0.8–1.8)
Medical complication
Original ATS criteria 69.2 71.1 0.60 (0.57–0.64) 89.1 40.4 1.5 (1.1–2.0)
Revised ATS criteria 67.4 62.2 0.60 (0.57–0.63) 84.1 39.1 1.3 (1.0–1.7)
BTS criteria 28.3 86.6 0.57 (0.54–0.60) 83.8 33.1 1.3 (0.9–1.7)
High PSI (Risk Class IV or V) 58.0 77.3 0.65 (0.61–0.68) 89.7 35.1 1.4 (1.0–1.9)
Death**
Original ATS criteria 79.8 41.4 0.60 (0.54–0.65) 8.8 96.6 2.6 (1.5–4.5)
Revised ATS criteria 39.6 67.6 0.63 (0.57–0.69) 8.2 93.9 1.3 (0.9–2.1)
BTS criteria 56.0 78.4 0.62 (0.55–0.68) 15.9 96.1 4.0 (2.6–6.2)
High PSI (Risk Class IV or V) 94.4 53.2 0.75 (0.71–0.78) 12.6 99.3 16.8 (6.8–41.8)

Definition of abbreviations: ATS criteria ⫽ American Thoracic Society criteria for severe community-acquired pneumonia; BTS
criteria ⫽ British Thoracic Society criteria for severe community-acquired pneumonia; CI ⫽ confidence interval; NPV ⫽ negative
predictive value; PPV ⫽ positive predictive value; PSI ⫽ Pneumonia Severity Index; ROC ⫽ receiver operator characteristic area
under the curve; RR ⫽ relative risk of event.
All criteria are determined at least one day before the predicted event.
* We chose four events (two clinical decisions and two adverse outcomes) as proxies for “severe” community acquired pneumonia.
†
Respiratory rate ⬎ 30 breaths/minute.
‡
Respiratory failure defined as PaO2/FIO2 ratio ⬍ 250 mm Hg.
§
In addition to bilateral/multilobe involvement on the chest radiograph, an increase in the size of the opacity by at least 50%
within 48 hours of admission.


Vasopressor administered for more than 4 hours.
#
Renal impairment defined by urine output lower than 20 ml/hour, or total urine output lower than 80 ml in 4 hours, or dialysis
for acute renal failure.
** n ⫽ 1,328. Analysis excludes 11 patients who already had been ventilated at hospital admission.
††
Hospital death by 30 days after presentation.

patterns. In contrast, mechanical ventilation rates did not vary but their discrimination appeared too low to guide individual
across institutions after adjusting for severity of illness, sug- decision-making. The biggest problem was the poor positive
gesting this decision is less discretionary and more closely linked predictive value. For example, three-quarters of the patients
to the patient’s severity of illness. who met any of the criteria were never admitted to the ICU. As
Given the high cost of ICU care and the considerable varia- reported previously, the original ATS criteria, although sensitive,
tion in ICU admission decisions, a closer examination of how had low specificity. Unfortunately, at least in this cohort, the
patients are admitted is warranted. Prior studies of CAP also improvement in predictive ability of the revised ATS criteria was
suggested a wide variation in ICU admission rates, ranging from modest. The BTS criteria, attractive because of their simplicity,
3 to 39% (10, 36). Suboptimal decision-making regarding ICU performed less well than the revised ATS criteria.
admission could result in under- or overuse of the ICU, with Other ICU risk prediction methods, such as APACHE III
potential consequences including adverse outcomes due to de- (37), do have good predictive characteristics for ICU course and
layed or inadequate care for some patients and excessive re- outcome but are generally not available at the time of the deci-
source use for other patients. For example, low risk of death as sion to admit a patient to the ICU. It may be worthwhile explor-
predicted by the PSI (Risk Classes I, II, and III) has been pro- ing whether elements from such scores, measured before admis-
posed as a reason to deny hospital admission, yet one-quarter sion, enhance the predictive accuracy of the PSI or ATS criteria.
of the ICU patients in our cohort were in these classes. Any attempt to study and improve the ICU admission decision
Our study focused on the relationship between the ATS crite- ought also to standardize the type and level of care offered in
ria and subsequent care decisions (ICU admission and mechani- the ICU and in the alternative to the ICU (e.g., the floor) if the
cal ventilation) and outcomes (medical complications and death) benefits of ICU care are to be best understood. Our data further
that might define CAP as “severe,” while comparing them with suggest that outcome should be assessed beyond hospital dis-
other prognostic instruments. Our results suggest that all the charge if the full economic and clinical burden of disease is to
rules were associated with the events suggestive of severe CAP, be captured.
722 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 166 2002

CAP (21), would be considered inappropriate today. On the

other hand, the extent to which current practice is compliant
with the latest recommendations is unclear. Also, there was
considerable variation in practice across institutions, yet the pre-
dictive characteristics of the different rules were robust to these
variations.
We determined only which factors were associated with a
higher likelihood of receiving ICU care, and not which factors
were associated with a higher likelihood of benefit from ICU
care. We also examined only patient characteristics that influence
the ICU admission decision, yet other factors, such as bed avail-
ability or family preferences, may affect the admission decision.
We did not have culture sensitivity data and could not therefore
analyze how management was influenced by the appropriateness
of initial antibiotic choice, an important variable affecting CAP
management (41). Finally, there is no gold standard for the term
“severe CAP.” We therefore presented results defining severe
CAP in four ways (i.e., CAP with one of four separate specific
events). We chose these events on the basis of clinical face
validity, but recognize that the definitions are arbitrary.
In summary, although overshadowed in numbers by patients
with low-risk CAP, patients who receive ICU care represent an
important subset, both in terms of cost and morbidity. The cur-
rent use of ICU services for CAP is expensive and may be
somewhat discretionary with outcomes that, although reason-
able, require measurement beyond hospital discharge to be fully
understood. Existing risk predictors will likely require modifica-
tion before they can be used to guide individual ICU admission
decisions, but such work is essential if ICU services are to be
Figure 4. Discriminative power of the revised ATS and BTS prediction
used optimally.
rules for severe CAP across hospitals. Point estimates and 95% confi-
dence intervals for the area under the receiver operating characteristic Acknowledgment : The authors thank research nurses Rhonda Grandy, R.N., Dawn
(ROC) curve of the revised ATS and the BTS rules in predicting ICU Menon, G.N., Jackie Cunning, R.N., Linda Kraft, R.N., and Maxine Young, R.N.,
in Halifax; Mary Walsh, R.N., Donna Polenik, R.N., M.P.H., and Kathryn Fine,
admission, death, and mechanical ventilation. Higher numbers for the
R.N. in Pittsburgh; Mary Ungaro, R.N., Leila Haddad, A.B., M.P.H., and Marian
area under the ROC curve indicate better discriminative power. Hendershot, R.N. in Boston. The authors also thank Walter T. Linde-Zwirble for
thoughtful review and comments.

References
We identified an etiologic agent in less than half of the ICU 1. Graves EJ, Gillum BS. 1994 Summary: National Hospital Discharge
patients. In other studies of patients with CAP requiring admis- Survey, advance data. Washington, DC: National Center for Health
sion to the ICU, a microbiologic etiology was determined in Statistics; 1996. Report No. 278.
58–72% of patients (8, 9, 38). The lower rate in our study may 2. Dahmash NS, Chowdhury MN. Re-evaluation of pneumonia requiring
admission to an intensive care unit: a prospective study. Thorax 1994;
have been because the ordering of microbiologic cultures and 49:71–76.
serologies was at the discretion of the clinical team. As with our 3. McCormick D, Fine MJ, Coley CM, Marrie TJ, Lave JR, Obrosky DS,
study, other investigators (8–12, 36, 38) have shown that the Kapoor WN, Singer DE. Variation in length of hospital stay in patients
spectrum of etiologic agents of pneumonia in patients admitted with community-acquired pneumonia: are shorter stays associated with
to ICU is similar to that in the general population of patients worse medical outcomes? Am J Med 1999;107:5–12.
4. Halm EA, Fine MJ, Marrie TJ, Coley CM, Kapoor WN, Obrosky DS,
with CAP, although the frequency of these pathogens varies Singer DE. Time to clinical stability in patients hospitalized with com-
across studies. munity-acquired pneumonia: implications for practice guidelines. JAMA
Nonpulmonary complications were common among the ICU- 1998;279:1452–1457.
treated patients. In a study of 299 patients with severe CAP, 5. Ramirez JA, Srinath L, Ahkee S, Huang A, Raff MJ. Early switch from
Leroy and coworkers (39) noted the development of similar intravenous to oral cephalosporins in the treatment of hospitalized
patients with community-acquired pneumonia. Arch Intern Med 1995;
nonpulmonary complications but generally at lower rates than 155:1273–1276.
those observed in our study. Torres and coworkers (40) studied 6. Fine MJ, Auble TE, Yealy DM, Hanusa BH, Weissfeld LA, Singer DE,
92 patients with severe CAP requiring ICU treatment and re- Coley CM, Marrie TJ, Kapoor WN. A prediction rule to identify low-
ported a high incidence of acute renal failure, septic shock, risk patients with community-acquired pneumonia. N Engl J Med 1997;
cardiac dysrhythmias, and abnormal liver function tests, similar 336:243–250.
7. Atlas SJ, Benzer TI, Borowsky LH, Chang Y, Burnham DC, Metlay JP,
to our study. These data support the need to provide multisystem Halm EA, Singer DE. Safely increasing the proportion of patients
care to patients with CAP requiring ICU care. with community-acquired pneumonia treated as outpatients: an inter-
There are limitations to our study. The Pneumonia PORT ventional trial. Arch Intern Med 1998;158:1350–1356.
cohort is one of the largest, most detailed, and most contempo- 8. British Thoracic Society Research Committee, Public Health Laboratory
rary studies of CAP. However, data were collected in the early Service. The aetiology, management and outcome of severe commu-
nity-acquired pneumonia in the intensive care unit. Respir Med 1992;
and mid 1990s at four institutions. Thus, care patterns may not
86:7–13.
be representative of current care at other North American sites. 9. Moine P, Vercken JB, Chevret S, Gajdos P. Severe community-acquired
For example, the high use of second-generation cephalosporins, pneumococcal pneumonia. French Study Group of Community-
although consistent with the 1993 ATS consensus statement on Acquired Pneumonia in ICU. Scand J Infect Dis 1995;27:201–206.
Angus, Marrie, Obrosky, et al.: Community-acquired Pneumonia and the ICU 723

10. Ortqvist AB, Sterner G, Nilsson JA. Severe community-acquired pneu- LJ, Obrosky DS, Schulz R, Ricci EM, et al. Processes and outcomes
monia: factors influencing need of intensive care treatment and prog- of care for patients with community-acquired pneumonia: results from
nosis. Scand J Infect Dis 1985;17:377–386. the Pneumonia Patient Outcomes Research Team (PORT) cohort
11. van Eeden SF, Coetzee AR, Joubert JR. Community-acquired pneumo- study. Arch Intern Med 1999;159:970–980.
nia: factors influencing intensive care admission. S Afr Med J 1988;73: 27. Metlay JP, Fine MJ, Schulz R, Marrie TJ, Coley CM, Kapoor WN, Singer
77–81. DE. Measuring symptomatic and functional recovery in patients with
12. Sorensen J, Cederholm I, Carlsson C. Pneumonia: a deadly disease de- community-acquired pneumonia. J Gen Intern Med 1997;12:423–430.
spite intensive care treatment. Scand J Infect Dis 1986;18:329–335. 28. Fine MJ, Pratt HM, Obrosky DS, Lave JR, McIntosh LJ, Singer DE,
13. Ortqvist AB, Hedlund J, Grillner L, Jalonen E, Kallings I, Leinonen M, Coley CM, Kapoor WN. Relation between length of hospital stay and
Kalin M. Aetiology, outcome and prognostic factors in community- costs of care for patients with community-acquired pneumonia. Am J
acquired pneumonia requiring hospitalization. Eur Respir J 1990;3:1105– Med 2000;109:378–385.
1113. 29. Clermont G, Angus DC, Linde-Zwirble WT, Griffin MF, Fine MJ, Pinsky
14. Ortqvist AB, Grepe A, Julander I, Kalin M. Bacteremic pneumococcal MR. Does acute organ dysfunction predict patient-centered outcomes?
pneumonia in Sweden: clinical course and outcome and comparison Chest 2002;121:1963–1971.
with non-bacteremic pneumococcal and mycoplasmal pneumonias. 30. Mortensen EM, Coley CM, Singer DE, Marrie TJ, Obrosky DS, Kapoor
Scand J Infect Dis 1988;20:163–171. WN, Fine MJ. Causes of death for patients with community-acquired
15. Gransden WR, Eykyn SJ, Phillips I. Pneumococcal bacteraemia: 325 pneumonia: results from the Pneumonia Patient Outcomes Research
episodes diagnosed at St Thomas’s Hospital. BMJ 1985;290:505–508. Team (PORT) cohort study. Arch Intern Med 2002;162:1059–1064.
16. Marrie TJ. Bacteraemic pneumococcal pneumonia: a continuously evolv- 31. World Health Organization. Manual of the international statistical classi-
ing disease. J Infect 1992;24:247–255. fication of diseases, injuries and causes of death. Geneva, Switzerland:
17. van Eeden SF, de Beer P. Community-acquired pneumonia: evidence World Health Organization; 1977.
of functional inactivation of ␣1 proteinase inhibitor. Crit Care Med 32. Kendall M, Stuart A. The advanced theory of statistics: design and analy-
1990;18:1204–1209. sis, and time-series. New York: Macmillan; 1979. p. 580–585.
18. Puren AJ, Feldman C, Savage N, Becker PJ, Smith C. Patterns of cytokine 33. Feinstein AR. Clinical epidemiology. Philadelphia, PA: WB Saunders;
expression in community-acquired pneumonia. Chest 1995;107:1342– 1985. p. 145–153.
1349. 34. Norman GR, Streiner DL. Biostatistics: the bare essentials. St. Louis,
19. Lippmann ML, Goldberg SK, Walkenstein MD, Herring W, Gordon M. MO: Mosby-Year Book; 1993. p. 190–191.
Bacteremic pneumococcal pneumonia: a community hospital experi- 35. Norman GR, Streiner DL. Biostatistics: the bare essentials. St. Louis,
ence. Chest 1995;108:1608–1613. MO: Mosby-Year Book; 1993. p. 186–187.
20. Marfin AA, Sporrer J, Moore PS, Siefkin AD. Risk factors for adverse 36. Fang GD, Fine MJ, Orloff JJ, Arisumi D, Yu VL, Kapoor WN, Grayston
outcome in persons with pneumococcal pneumonia. Chest 1995;107: JT, Wang SP, Kohler R, Muder RR, et al. New and emerging etiologies
457–462. for community-acquired pneumonia with implications for therapy: a
21. Niederman MS, Bass JBJ, Campbell GD, Fein AM, Grossman RF, Man- prospective multicenter study of 359 cases. Medicine (Baltimore) 1990;
dell LA, Marrie TJ, Sarosi GA, Torres A, Yu VL. Guidelines for the 69:307–316.
initial management of adults with community-acquired pneumonia: 37. Knaus WA, Wagner DP, Draper EA, Zimmerman JE, Bergner M, Bastos
diagnosis, assessment of severity, and initial antimicrobial therapy. PG, Sirio CA, Murphy DJ, Lotring T, Damiano AM, et al. The
American Thoracic Society. Am Rev Respir Dis 1993;148:1418–1426. APACHE III prognostic system: risk prediction of hospital mortality
22. Ewig S, Ruiz M, Mensa J, Marcos MA, Martinez JA, Arancibia F, Nieder- for critically ill hospitalized adults. Chest 1991;100:1619–1636.
man MS, Torres A. Severe community-acquired pneumonia: assess- 38. Arancibia F, Ewig S, Martinez JA, Ruiz M, Bauer T, Marcos MA, Mensa
ment of severity criteria. Am J Respir Crit Care Med 1998;158:1102– J, Torres A. Antimicrobial treatment failures in patients with commu-
1108. nity-acquired pneumonia: causes and prognostic implications. Am J
23. Gordon GS, Throop D, Berberian L, Niederman M, Bass JBJ, Alemay- Respir Crit Care Med 2000;162:154–160.
ehu D, Mellis S. Validation of the therapeutic recommendations of the 39. Leroy O, Santre C, Beuscart C, Georges H, Guery B, Jacquier JM,
American Thoracic Society (ATS) guidelines for community-acquired Beaucaire G. A five-year study of severe community-acquired pneu-
pneumonia in hospitalized patients. Chest 2000;110:55S. monia with emphasis on prognosis in patients admitted to an intensive
24. Niederman MS, Mandell LA, Anzueto A, Bass JB, Broughton WA, care unit. Intensive Care Med 1995;21:24–31.
Campbell GD, Dean N, File T, Fine MJ, Gross PA, et al. Guidelines 40. Torres A, Serra-Batlles J, Ferrer A, Jimenez P, Celis R, Cobo E, Rodri-
for the management of adults with community-acquired pneumonia: guez-Roisin R. Severe community-acquired pneumonia: epidemiology
diagnosis, assessment of severity, antimicrobial therapy, and preven- and prognostic factors. Am Rev Respir Dis 1991;144:312–318.
tion. Am J Respir Crit Care Med 2001;163:1730–1754. 41. Gleason PP, Kapoor WN, Stone RA, Lave JR, Obrosky DS, Schulz R,
25. British Thoracic Society. Guidelines for the management of community- Singer DE, Coley CM, Marrie TJ, Fine MJ. Medical outcomes and
acquired pneumonia in adults admitted to hospital. Br J Hosp Med antimicrobial costs with the use of the American Thoracic Society
1993;49:346–350. guidelines for outpatients with community-acquired pneumonia. JAMA
26. Fine MJ, Stone RA, Singer DE, Coley CM, Marrie TJ, Lave JR, Hough 1997;278:32–39.