| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
* From the Pulmonary and Critical Care Division, Bridgeport Hospital and Yale University School of Medicine, Bridgeport, CT.
Correspondence to: Constantine A. Manthous, MD, FCCP, Bridgeport Hospital, West Tower 6, 267 Grant St, Bridgeport, CT 06610; e-mail: pcmant{at}bpthosp.org
Abstract
Study objectives: To define risk factors, identifiable on initial presentation, that predict subsequent physiologic derangements that are consistent with critical illness in patients presenting to hospital with GI hemorrhage (GIH).
Design: Observational, cohort study.
Setting: Fourteen-bed medical ICU in a 300-bed community teaching hospital.
Patients: One hundred ninety-three patients were studied during 199 separate hospital admissions for GIH.
Methods and measurements: Demographic and physiologic variables were extracted from the medical records of patients admitted with GIH. Comprehensive data, from after 2 h in the emergency department to the time of discharge or death, were used to determine whether patients met established ICU admission criteria. Physiologic and demographic data from the initial 2-h period were then compared for patients who subsequently met and for those who did not meet ICU admission criteria. Independent predictors of meeting ICU admission criteria were identified using multiple logistic regression analyses. Sensitivity and specificity associated with the combined use of these predictors were assessed.
Results:
Thirty-four patients satisfied ICU admission criteria after the initial
2-h period in the emergency department. Sixty-five patients, including
29 of 34 patients who met ICU admission criteria, were actually
admitted to the ICU. Among those who never fulfilled ICU admission
criteria, the duration of hospital stay was longer for those admitted
to the ICU than for those not admitted to ICU (6.6 ± 0.6 days vs
5.2 ± 0.3 days; p = 0.04). The admission prothrombin time
(international normalized ratio > 1.2), hypotension (systolic BP
< 90 mm Hg), acute neurologic changes, and initial APACHE (acute
physiology and chronic health evaluation) II score ( 
15) were the
best independent predictors for meeting the defined criteria for
admission to ICU. The presence of one or more of these in the first
2 h of presentation was associated with a sensitivity of 88% and
specificity of 74% for predicting subsequent critical instability. The
area under the receiver operator characteristic curve for use of these
four variables was 86% for predicting whether patients met ICU
admission criteria.
Conclusions: Many patients with
GIH were admitted to the ICU who never met local criteria for
admission, and these patients experienced a significantly longer length
of hospital stay than other, similarly ill patients. Coagulopathy,
hypotension, neurologic dysfunction, and a higher ( 15) APACHE II
score in the first 2 h of hospitalization were the best
independent predictors of the subsequent development of critical
illness.
Key Words: bleeding • ICU • intensive care • risk
Several studies1 2 3 4 5 6 have assessed risk factors that predict morbidity and mortality in patients admitted to the hospital with GI hemorrhage (GIH). No
previous study has examined which factors, present on admission, predict the need for treatment in an ICU. In the absence of such data, caregivers and institutions rely on either unproven criteria or ad hoc physicians’ judgements to guide disposition decisions for this very common reason for hospitalization. In this study, we prospectively collected demographic and physiologic data on all patients admitted to our hospital with GIH and identified those patients who met physiologic criteria for ICU admission. We then compared variables associated with patients who met ICU admission criteria, obtained around the time of admission, with variables associated with patients who did not meet ICU admission criteria, to define parameters that predicted subsequent physiologic instability consistent with critical illness.
Materials and Methods
Our hospital investigational review committee waived formal review of this study. The study hospital is a 300-bed, university-affiliated, community teaching hospital with a 14-bed, medical-cardiac critical care unit, in which all medical attending physicians, guided by ICU admission criteria, have admitting privileges. Physicians are not prevented from admitting patients who do not strictly meet physiologic criteria for admission to the ICU.
Between March 1998 and February 1999, our admitting department provided a daily list of all patients admitted in the preceding 24 h with the primary diagnosis of GIH. Using a uniform data abstract form, designed specifically for this study, information was gathered on all patients from the time of admission to the time of discharge or death. The data extracted included age, gender, race, APACHE (acute physiology and chronic health evaluation) II score at admission, hematocrit at admission, heart rate, and BP; prothrombin time (PT), expressed as international normalized ratio (INR); platelet count; creatine phosphokinase levels, with isoenzymes, at admission; and mental status based on the Glasgow coma scale. The highest heart rates and lowest systolic BPs recorded for each patient during the first 4 h in the emergency department (ED), and then for each day thereafter, were also extracted. The lowest hematocrit levels for each patient on each of the first 3 days and all transfusions received by patients over the first 2 days were also recorded. Patients were classified as having experienced rebleeding if, after initial presentation, they experienced hematemesis, hematochezia, or endoscopically proven recurrent bleeding accompanied by a reduction in hematocrit of 3%. New onset neurologic dysfunction was defined as a change from baseline and a Glasgow coma score < 15. Additional data extracted from the records included endoscopic and radiologic evaluations for causes of bleeding, lengths of stay in the hospital and ICU, and outcomes.
Stratification of the Data: ICU Admission Criteria
Patients were categorized as having met ICU admission criteria
if they met any of the following after 2 h of treatment in the ED
until the time of discharge: systolic BP < 90 mm Hg for 15 min;
sustained tachyarrhythmia (with heart rate > 140 beats/min); clinical
evidence of evolving myocardial ischemia or infarction (based on
symptoms, ECG, or enzymatic evidence of myocardial injury); acute
respiratory failure, need for 50% inspired oxygen to maintain
PaO2 > 60 mm Hg; pH < 7.30,
unstable neurologic status; and/or Glasgow coma score 
10.
Statistical Analysis
The selected demographic variables and physiologic parameters
present within the first 2 h of treatment in the ED were compared
between patients who met ICU admission criteria and patients who did
not meet ICU admission criteria. Using risk ratios as the measure
of association, the ability of each of the physiologic parameters to
predict the subsequent fulfillment of the ICU admission criteria was
ascertained. A multiple logistic regression model was subsequently used
to adjust for confounding parameters, to assess effect modifiers, and
to identify parameters that independently predicted meeting ICU
admission criteria. The choice of parameters for inclusion in the
logistic model was determined by the biological plausibility of each
variable and/or evidence of an association in the univariate analysis.
Where appropriate, threshold levels were defined using standard norms
or empiric quartiles where no such norms exist. Sensitivities and
specificities were computed using combinations of those variables found
in the multiple logistic regression to be independent predictors of
meeting ICU admission criteria. Receiver operator characteristic (ROC)
curves were also constructed to assess the overall predictive
utilities of the combined predictors. Predictive characteristics of the
best model derived from our data set were then compared to those
applying the criteria of active Bleeding, hypotension
(Low BP), coagulopathy (Elevated PT),
Erratic mental status changes, and the presence of one or
more unstable comorbid Diseases (abbreviated as BLEED)
1
to the cohort. Statistical significance was signified by
a p value < 0.05.
Results
One hundred ninety-three patients, 104 men and 89 women, with a mean ( ± SE) age of 67.8 ± 1.2 years and mean APACHE II score of 9.6 ± 0.4, were studied over 199 admissions. For the overall cohort, the mean number of comorbid conditions was 0.6 ± 0.1; the mean admission hematocrit was 30.2 ± 0.6%; the mean admission systolic BP was 125.7 ± 2.2 mm Hg; and the mean heart rate was 91.6 ± 1.4 beats/min. Only 83 patients (43%) were checked for postural changes of BP and heart rate in the ED. Mean hospital stay was 6.2 ± 0.3 days, and 8 of 199 patients (4%) did not survive hospitalization. With one exception, patients who died were not resuscitated, in accordance with "do not resuscitate" orders from the patient or family. Three of these patients were receiving only comfort care at the time of death. One patient, who had no "do not resuscitate" orders, apparently died as a direct result of GIH. Another patient, who refused blood transfusions, died of complications related to GIH.
Stratification by ICU Admission Criteria
Thirty-four patients satisfied ICU admission criteria after the
initial 2 h treatment in the ED; all but 2 of these met criteria
within 48 h of disposition from the ED. All 34 patients also would
have satisfied ICU admission criteria, even if 4 h had been
allowed for initial stabilization in the ED. Table 1 lists the reasons patients met ICU admission criteria. Of these 34
patients, 29 were actually admitted to the ICU. Thirty-six additional
patients (18%) who were admitted to the ICU never fulfilled the ICU
admission criteria at any time after the initial 2 h in the ED.
Among patients who never met ICU admission criteria, there were no
significant differences in age, APACHE II, BP, number of comorbid
illnesses, or admission PT between those who went to the ICU and those
who did not.
|
|
15) were
independently associated with subsequently meeting ICU admission
criteria (Table 3
). The sensitivity for predicting meeting ICU admission criteria after
having one or more of these four present in the first 2 h of
treatment in the ED was 88%, and specificity was 74% with an area
under the ROC curve of 0.86. For those who satisfied two or more
criteria, the specificity increased to 95%, while the sensitivity
decreased to 47%. If patients met one or more of the
BLEED1
criteria, the sensitivity of meeting ICU admission
criteria was 94%, with a specificity of 37% (ROC area
= 0.83).
|
Analysis of Outcomes
The low mortality (4%) noted in this cohort precluded meaningful
analysis of variables that predict mortality. Hospital lengths of stay
were significantly greater in patients who met ICU admission criteria
(9.5 ± 1.3 days vs 5.5 ± 0.3 days; p = 0.004). Among those who
never fulfilled ICU admission criteria, hospital admission was longer
for those admitted to the ICU than for those not admitted to ICU
(6.6 ± 0.6 days vs 5.2 ± 0.3 days; p = 0.04).
Discussion
This study demonstrates that roughly half of patients admitted to
our medical ICU failed to meet even our local physiologic admission
criteria at any time subsequent to their initial ED care. Four
variables, easily obtained during the first 2 h of
hospitalization, were strongly associated with subsequently meeting
physiologic criteria for admission to our ICU. These include the
admission APACHE II score ( 15), the PT INR (> 1.2), new
neurologic dysfunction, and systolic hypotension (< 90 mm Hg). The
presence of one or more of these factors had a sensitivity of 88% and
specificity of 74% in identifying patients who would satisfy criteria
for admission to our ICU during their hospital course.
To our knowledge, this is the first study that identifies variables
predictive of the development of unstable physiologic parameters that
could warrant care in an intensive or special care unit. Kollef and
colleagues1
found that BLEED criteria predict increased
in-hospital morbidity and mortality. However, some definitions
(eg, what is meant by active conditions, and at what level
does mental status change?) were not well described in the article, and
the relationship of outcomes with intensive care were not studied.
Although the BLEED risk stratification system was not developed to
predict the need for ICU care, satisfaction of one or more of the BLEED
criteria carried a sensitivity of 94% and specificity of 37% for
meeting our local ICU admission criteria (ROC area = 0.83). In
contrast, the presence of one or more of our four criteria, systolic BP
< 90 mm Hg, PT INR > 1.2, new neurologic dysfunction or APACHE II
score 15, predicted satisfaction of our ICU admission criteria with
a sensitivity of 88% and specificity of 74% (ROC area =
0.86). Reduction of triage decisions to these four readily quantifiable
variables may provide a simple tool for identifying patients who are
most likely to benefit from more specialized care and monitoring. Any
predictive tool used to allocate ICU resources should seek to maximize
both sensitivity (to minimize the likelihood of not admitting a patient
who truly requires ICU admission) and specificity (to minimize
unnecessary ICU admissions). For some, the 88% sensitivity of our
model may provide a major practical limitation if the goal is to avoid
inadvertent admission of critically ill patients to the hospital floor.
In fact, the current practice of the physicians of our hospital, each
with his/her own clinical skills, yielded similar sensitivity of 85%
and specificity of 78%. The criteria suggested by our study, while not
a "gold standard," offer objective physiologic parameters which,
when combined with clinical judgment, could improve triage decisions.
Furthermore, the high specificity (95%) associated with satisfaction
of two or more criteria strongly supports triaging such patients to an
intensive (or, possibly, intermediate) care unit. Indeed, predictive
tools (eg, the rapid shallow breathing index as a predictor
for weaning outcomes), with similar predictive characteristics,
7
8
have contributed to improved outcomes when applied as
part of a clinical algorithm.9
Development of such tools
to correctly identify patients with GIH who will require ICU care also
has implications for length of stay. In this study, patients who were
admitted to ICU but failed to meet admission criteria had a hospital
stay that was an average of 1.4 days longer than that of similarly ill
patients who were not admitted to the ICU.
The very low mortality rate (4%) noted in this relatively small sample precludes any meaningful analysis of predictors for dying of GIH or analysis of whether risk factors for requiring ICU care also predict death. In this regard, the findings of Kollef et al1 have been discussed above. Rockall et al2 found that advanced age, hypotension, other comorbidities, active bleeding, and rebleeding independently predicted mortality in a cohort of 5,810 patients with upper GIH. Clason et al3 reported that age of 60 years, clinical shock on admission, and rebleeding were independent predictors of mortality in 326 cases of upper GIH. Katschinski et al4 found that mortality was associated with advanced age, blood in the stomach on endoscopy, and rebleeding in 2,217 patients with upper GIH. Shock on admission and active bleeding during endoscopy were associated with rebleeding. Rebleeding, like hypotension on admission, is a common variable defining risk in these studies, and it is reiterated as a risk in our study. Unfortunately, very strong predictors of rebleeding have not been well defined, and clinicians cannot predict with great certainty which patients will subsequently rebleed. The above studies help to identify the cohort of patients with GIH who are at risk of morbidity and/or mortality with GIH. However, these studies do not address appropriate hospital disposition of this at-risk population. Our study is unique in that its goal was to define variables, present on admission, that predict the subsequent need for intensive care.
The absence of universally accepted ICU admission criteria limits the ability to generalize our findings to institutions with similar admission criteria. Some clinicians may disagree with application of only physiologic criteria to determine disposition, and others may disagree with the specific thresholds employed in our hospital. Nonetheless, Table 1 demonstrates that the patients sorted into the ICU group for this study are likely to meet ICU admission criteria at most hospitals. The utility of such stringent criteria makes it more likely that we excluded some patients who would be included at other hospitals. We also recognize that in some hospitals, the number of ICU beds is so limited that GIH patients are admitted only if they are receiving mechanical ventilation; admission is determined by logistic imperative rather than by what physicians would prefer if resources were available. In such situations, our findings are not likely to be very helpful.
Our study does not address whether patients satisfying our ICU admission criteria would necessarily benefit from intensive care. We are unaware of any data published to date that demonstrate improved outcomes for any subgroup of patients with GIH cared for in an ICU. We previously reported that in-hospital mortality was reduced by 33% in patients with GIH admitted to our ICU after initiation of a formal critical care program.10 These retrospective data suggest that organized ICU care may benefit sicker patients with GIH. Would a well-organized intermediate care unit suffice for all or a subset of these patients? In the absence of such studies, physicians and hospitals are left to devise criteria for admission to ICU with the presumption that patients with the most severe physiologic derangements are most likely to benefit from the additional vigilance, monitoring, and resources afforded in these settings. We have been careful to avoid suggesting that this study has identified risks for requiring ICU care, since both the criteria used to define the at-risk population and the benefits of this care can be disputed. Nevertheless, if one accepts that ICU care is beneficial in selected cases and that the selection process employed in our study is reasonable, then our data may aid in identifying those patients who are most likely to benefit from ICU admission.
In conclusion, this study demonstrates that many more patients were
admitted to the ICU at our hospital for GIH than met reasonable
physiologic criteria, and that admission to ICU was associated with
prolonged lengths of hospital stay. Our data suggest that hypotension,
neurologic impairment, coagulapathy, and APACHE II score 15 on
admission are the best predictors of subsequent physiologic
instability. The presence of one or more of these in the first several
hours of admission could be useful in defining a cohort of at-risk
patients, both in the clinical arena and for prospective studies
designed to determine the efficacy of intensive care in very ill
patients with GIH.
Footnotes
Abbreviations: APACHE = acute physiology and chronic health evaluation; BLEED = active Bleeding, hypotension (Low BP), coagulopathy (Elevated PT), Erratic mental status changes, and the presence of one or more unstable comorbid Diseases; ED = emergency department; INR = international normalized ratio; GIH = GI hemorrhage; PT = prothrombin time; ROC = receiver operating characteristic
Received for publication October 26, 1999. Accepted for publication February 1, 2000.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
