Identifying Infected Emergency Department Patients Admitted to the Hospital Ward at Risk of Clinical Deterioration and Intensive Care Unit Transfer

Abstract

Objectives

An important challenge faced by emergency physicians (EPs) is determining which patients should be admitted to an intensive care unit (ICU), and which can be safely admitted to a regular ward. Understanding risk factors leading to under-triage would be useful, but these factors are not well characterized.

Methods

The authors performed a secondary analysis of two prospective, observational studies of patients admitted to the hospital with clinically suspected infection from a urban university emergency department (ED). Inclusion criteria were: adult ED patient (age 18 years or older), ward admission, and suspected infection. The primary outcome was transfer to an ICU within 48 hours of admission. Using multiple logistic regression, independent predictors of early ICU transfer were identified, and the area under the curve for the model was calculated.

Results

Of 5,365 subjects, 93 (1.7%) were transferred to an ICU within 48 hours. Independent predictors of ICU transfer included: respiratory compromise (OR 2.5, 95% CI = 1.4 to 4.3), congestive heart failure (OR 2.2, 95% CI = 1.4 to 3.6), peripheral vascular disease (OR 2.0, 95% CI = 1.1 to 3.7), systolic blood pressure < 100 mmHg (OR 1.9, 95% CI = 1.2 to 2.9), heart rate >90 beats/minute (OR 1.8, 95% CI = 1.1 to 2.8), and creatinine > 2.0 (OR 1.8, 95% CI = 1.1 to 2.8). Cellulitis was associated with a lower likelihood of ICU transfer (OR 0.33, 95% CI = 0.15 to 0.72). The area under the curve for the model was 0.73, showing moderate discriminatory ability.

Conclusions

In this preliminary study, independent predictors of ICU transfer within 48 hours of admission were identified. While somewhat intuitive, physicians should consider these factors when determining patient disposition.

INTRODUCTION

One of the many challenges faced by emergency physicians (EPs) is the appropriate disposition of patients; in particular, determining which patients should be admitted to an intensive care unit (ICU). ICUs provide essential but costly care to critically ill patients. While some patients obviously require ICU level of care (i.e., patients who are intubated or have shock requiring vasopressors), others appear stable but warrant ICU admission due to the risk of clinical deterioration. Over the past several decades, the number of ICU beds has declined, while ICU utilization has increased, making accurate risk stratification even more challenging and important.¹

Mistriaging patients has serious consequences. Prior studies demonstrate that patients who are transferred to the ICU from an inpatient ward have higher mortality rates than patients who are admitted directly to an ICU from the emergency department (ED).^2,3 This association persists in spite of comprehensive multivariable control, leading to its inclusion in the APACHE III prognostic system.⁴ Clinical deterioration on a hospital ward is a demonstrated independent predictor of mortality.⁵ Most research related to clinical deterioration on inpatient units and unanticipated ICU transfer has focused on early identification of abnormal vital signs or deployment of critical care outreach teams on the inpatient wards. While logical, these efforts have reported varying success in reducing mortality.^6–9 An alternative approach would be to move upstream in the patient’s stay, and reduce unanticipated transfers from hospital wards to ICUs by creating tools to aid EPs in making more accurate triage decisions to curtail the problem before it happens.

While the identification of patients who are likely to deteriorate early in their hospital course is important, surprisingly little ED-based data exist either describing this population or identifying risk factors for early ward-to-ICU transfer. To begin to address this need, we conducted this study in a population of patients at particular risk: ED patients being admitted to the hospital wards with suspected infection. The objectives of this study were to describe the population of patients with suspected infection who are admitted to the hospital ward from the ED and are subsequently transferred to the ICU within the first 48 hours of admission, and to identify independent predictors present in the ED that are associated with increased risk of ward-to-ICU transfer.

METHODS

Study Design

This was a secondary analysis of data from two prospective observational cohort studies. Both studies were approved by the institution’s institutional review board.

Study Setting and Population

The study was conducted during two separate time periods at an urban tertiary care university hospital with approximately 50,000 annual ED visits. The methods and selection of participants have been described in detail previously.^10,11 Briefly, the first cohort was comprised of all adult patients (aged 18 years or older) presenting to the ED with suspected infection from February 1, 2000 through February 1, 2001; suspected infection was defined by ordering of a blood culture in the ED or within 3 hours of hospital admission. The second cohort was enrolled between December 10, 2003 and September 30, 2004, and consisted of all adult ED patients (aged 18 years or older) with suspicion of infection by the clinician as documented in the medical decision-making portion of the ED chart.

Study Protocol

Patients were stratified into three groups: those who were admitted to the general hospital ward and did not require ICU transfer over the first 48 hours (non-event), those who were admitted to a ward and were transferred to the ICU within 48 hours (event), and those who were admitted directly to the ICU from the ED (excluded from primary analysis). Patients transferred to the ICU for reasons other than deterioration in clinical condition (e.g. elective conscious sedation for a procedure) were also excluded from analysis.

Patients with suspected infection were prospectively identified on a daily basis, and trained research assistants abstracted data from the ED medical records using a previously reported methodology.^10,11 The data recorded included demographic information, medical history, pertinent physical exam findings, and laboratory results. We retrospectively determined which patients met our primary outcome (transfer from an inpatient ward to the ICU within 48 hours), and reviewed the medical chart to determine the reason for transfer. A secondary endpoint was in-hospital mortality.

Data Analysis

The population was characterized using descriptive statistics. Next, univariate analysis was performed to identify potential predictors of our primary outcome. We used a relaxed p-value of < 0.2 to identify candidate predictors for the multivariate model. We subsequently used multivariable logistic regression to select independent predictors of early ICU transfer. We used a forward-selection technique, allowing a variable to stay in the model when the p-value was less than 0.05. We built a multivariate model and report odds ratios (ORs), corresponding 95% confidence intervals (CI), and p-values for each variable. To describe overall discrimination of the model, we report the area under the receiver-operating-characteristics curve. To ensure model stability, we a priori determined the model should have at most one predictor per ten outcomes, as commonly recommended.¹²

RESULTS

Characteristics of Study Subjects

There were 5,749 patients with suspected infection who met study criteria and were admitted to the hospital. Of these, 335 (5.8%) were admitted directly to the ICU, and 49 (0.8%) were transferred to an ICU for reasons other than clinical deterioration; these were excluded from further analysis. This left a final cohort of 5,365 who were initially admitted to a non-ICU setting (Figure 1). Patients who were transferred to the ICU within 48 hrs were older than those who were not, had a higher burden of certain co-morbidities, and were more likely to have low blood pressure, tachypnea, and respiratory compromise (Table 1).

Figure 1.

Outline of patient enrollment and composition of patient groups.

Table 1.

Demographics and Distribution of Covariates of the Study Population

	Admitted directly to ICU (n=335)	ICU transfer within 48 hours (n=93)	No ICU transfer in 48 hours (n=5,272)	p-value*
Demographics
Age in years (mean ±SD)	63 (±19)	66 (±18)	61 (±20)	0.008
Age ≥ 65 years old (%)	52	61	46	0.003
Sex (% male)	48	54	47	0.175
Nursing home resident (%)	17	11	10	0.863
Comorbid Illnesses
Cancer (%)	15	24	17	0.076
Cerebrovascular disease (%)	10	13	8.3	0.108
CHF (%)	18	30	13	<0.001
COPD (%)	15	22	12	0.005
Dementia (%)	8	8.6	7.5	0.682
DM (%)	30	26	25	0.773
HIV (%)	2.7	5.4	6.4	0.681
Immunosuppression (%)	23	34	30	0.313
History of myocardial infarction (%)	9.3	11	8.6	0.466
Peripheral vascular disease (%)	7.8	16	7.9	0.004
Renal failure (%)	18	9.7	2.3	<0.001
Vital signs
Initial sBP, mmHg (mean ± SD)	121 (±34)	126 (±28)	132 (±29)	0.029
Initial dBP, mmHg (mean ±SD)	64 (±19)	65 (±19)	67 (±20)	0.291
Minimum sBP, mmHg (mean ±SD)	100 (±31)	111 (±23)	122 (±26)	<0.001
Minimum dBP, mmHg (mean ±SD)	51 (±19)	57 (±19)	63 (±20)	0.015
Initial HR (mean ±SD)	100 (±24)	95 (±23)	93 (±21)	0.348
Maximum HR (mean ±SD)	107 (±26)	102 (±24)	96 (±21)	0.007
Minimum sBP <100 mmHg (%)	53	32	17	<0.001
Maximum HR >90 (%)	70	70	57	0.014
Respiratory compromise (%)	21	18	5.6	<0.001
Laboratory results
WBC >15,000 (%)	43	37	24	0.004
Bands >5% (%)	3	9.7	6.8	0.271
Creatinine >2.0 (%)	30	29	16	0.001
Suspected Source of Infection
Intra-abdominal (%)	9.3	12	11	0.884
Lower respiratory (%)	34	37	23	0.002
Neutropenic fever (%)	0	0	1.3	0.267
Skin/soft tissue (%)	6.6	7.5	22	0.001
UTI/pyelonephritis (%)	13	16	10	0.066

CHF = congestive heart failure; COPD = chronic obstructive pulmonary disease; MI = myocardial infarction; DM = diabetes mellitus; HIV = human immunodeficiency syndrome; sBP = systolic blood pressure; dBP = diastolic blood pressure; HR = heart rate (beats/minute); WBC = white blood cell count; UTI = urinary tract infection; bpm = beats per minute. Respiratory compromise was defined as respiratory rate >20 breaths per minute or oxygen saturation <90% by pulse oximetry. Immunosuppression was defined as history of HIV or malignancy, transplant recipient, current treatment with systemic steroids, or current neutropenic fever.

^*P-values are reported for the univariate comparisons of patients transferred to the ICU within 48 hours versus patients not transferred to the ICU within 48 hours

Main Results

Of the 5,365 patients in the study analysis, 93 (1.7%) met our primary outcome of ward-to-ICU transfer within 48 hours. The median time to ICU transfer in this group was 20 hours (25%–75% IQR = 12–30 hours). The most common reason for transfer was worsening respiratory status (Table 2). Among the ward-to-ICU transfers, 39 of 93 patients (42%) were intubated at some point during their ICU course. Using the covariates associated with early ICU transfer identified in our univariate analysis, we performed a multivariate analysis and identified seven independent predictors for ICU transfer. Respiratory compromise (defined as respiratory rate >20 breaths/ minute or oxygen saturation <90% by pulse oximetry), congestive heart failure (CHF), peripheral vascular disease, moderate hypotension (defined as minimum systolic blood pressure [sBP] < 100 while in the ED), tachycardia, and elevated creatinine were associated with increased risk of ICU transfer (Table 3). Patients with cellulitis were less likely than other patients to require subsequent ICU transfer. The area under the receiver operator characteristics curve (AUC) for our model was 0.73 (95% CI = 0.68 to 0.78; likelihood ratio chi-square 66.6, p < 0.0001), demonstrating moderate discriminatory ability.

Table 2.

Reasons patients were transferred to the ICU after admission to an inpatient ward. Shock was defined as hypotension or development of acidosis.

Reason for ICU Transfer	Number (%)
Worsening respiratory status	45 (48)
Shock	33 (35)
Altered mental status	9 (9.6)
Cardiopulmonary arrest	4 (4.3)
Other*	13 (14)

^*Other: transfer for closer monitoring of underlying medical problem, transfer for closer monitoring post-procedure, cardiac concerns (atrial fibrillation with rapid ventricular rate; development of heart block; development of pericardial effusion; myocardial ischemia), hyperglycemia requiring continuous insulin infusion, concern for potential aortic erosion, and post-seizure.

Total number exceeds 100% as 11 patients had two reasons for transfer.

Table 3.

Independent Predictors of Subsequent ICU Transfer Among Patients Admitted to a Hospital Ward from the ED

Variable	Adjusted OR	95% CI	p-value
Respiratory compromise in the ED	2.5	1.4–4.3	0.002
Congestive heart failure	2.2	1.4–3.6	0.001
Peripheral vascular disease	2.0	1.1–3.7	0.02
sBP <100 in the ED	1.9	1.2–2.9	0.007
HR >90 in the ED	1.8	1.1–2.8	0.02
Creatinine >2.0	1.8	1.1–2.8	0.02
Cellulitis	0.33	0.15–0.72	0.01

OR = odds ratio; sBP = systolic blood pressure; HR = heart rate; ICU = intensive care unit

The mortality rate for patients who were transferred to the ICU was 24% (22 out of 93), compared to a mortality rate of 19% (62 out of 335) among those admitted directly to the ICU from the ED (p = 0.3). Subjects who did not require early ICU transfer had a lower mortality rate (188 out of 5,272 patients, or 4%) than those admitted directly to the ICU and those transferred early in their hospital course (p < 0.001 for both comparisons).

DISCUSSION

In this study of ED patients with suspected infection, we identified several predictors of clinical deterioration requiring early transfer to an ICU. The strongest predictor was respiratory compromise, as evidenced by tachypnea or hypoxia. In our study, 37% of patients who were transferred to the ICU within 48 hours of admission were diagnosed with pneumonia, and 42% of patients who were transferred required intubation. Patients who were transferred from the ward to the ICU also had more comorbid illnesses. In our multivariate analysis, CHF, peripheral vascular disease, and renal insufficiency were independent predictors of ICU transfer. Cellulitis as the source of infection was associated with a better outcome in our study, which correlates with the tendency of skin and soft tissue infections to cause less severe illness,¹³ and perhaps to be recognizable by patients earlier in the course of disease.

In a recent study of patients with community-acquired pneumonia, Renaud et al. also found that tachycardia, tachypnea, and hypoxia were associated with early ICU transfer, and found that patients who had at least one comorbid illness (i.e., CHF, chronic pulmonary disease, and renal disease) were more likely to be undergo early ICU transfer.¹⁴ Hypoxemia and tachycardia were predictive of ICU transfer in a study of febrile ED patients by Knott et al., although the ORs were only 1.10 and 1.03, respectively.¹⁵ CHF and chronic airway disease were associated with early ICU transfer in a study by Tam et al.¹⁶ Additional predictors of ICU transfer identified by Renaud et al. in their study of community acquired pneumonia were male sex, multilobar infiltrates or pleural effusion, pH < 7.35, elevated blood urea nitrogen and hyponatremia.¹⁴ Other elements of the nomogram created by Tam et al.¹⁶ to predict early ICU transfer were male sex, night or weekend admission, and presence of femur fracture, acute pancreatitis, liver disease, and pneumonia. In their study of febrile patients presenting to an ED, Knott et al.¹⁵ found that jaundice was most highly correlated with early ICU transfer (OR 5.12).

The findings that hypoxia, tachypnea, and underlying chronic pulmonary disease are associated with early ICU transfer underscore the importance of careful evaluation of respiratory status as part of ICU risk stratification. The relationship between comorbid illness and early ICU transfer suggests that these patients have less physiologic reserve, and correlates with our knowledge that preexisting medical disease is also associated with increased mortality in patients with severe sepsis.^17,18

We found that even transient hypotension and tachycardia were predictive of early ICU transfer, and tachycardia was also predictive of early ICU transfer in two of the previous studies of unanticipated ICU transfer.^14,15 In studies evaluating detection of and response to clinical deterioration on the inpatient ward, patients typically have normal vital signs on admission to the inpatient ward from the ED¹⁹; therefore, they appear hemodynamically stable to the ED provider at the time of ward admission. Marchick et al.²⁰ demonstrated that in septic ED patients, transient hypotension is an independent risk factor for in-hospital mortality. Concordant with prior studies, this suggests that even if vital signs normalize after appropriate resuscitation in the ED, these subjects are at higher risk of unanticipated ICU transfer and possibly death.

While age was correlated with ICU transfer in our univariate analysis, it did not reach statistical significance in multivariate analysis (adjusted OR 1.5; 95% CI = 0.97 to 2.4; p = 0.067). This may be due to co-linearity with CHF and peripheral vascular disease, or may be related to our choice to dichotomize at 65 years of age. Renaud et al. found that age less 80 years of age was associated with higher likelihood of ICU transfer,¹⁴ whereas Tam et al. found that older age was predictive of early ICU transfer.¹⁶ In studies of febrile adults, increasing age has been shown to be associated with increasing risk of serious illness,^21,22 and in adults, incidence of severe sepsis increases sharply with increasing age.¹⁷

Unanticipated ICU transfer of admitted patients has previously been associated with worse outcomes. In our study, this occurrence was not as striking, and we did not find a difference in the point estimate mortality rate of patients who were transferred vs those who were admitted directly to an ICU (24% vs. 19%). This pattern is consistent with prior studies, although we report lower mortality rates for ICU-transferred patients. Goldhill and Sumner reported that transferred patients had a mortality rate of 53% compared to 30% for patients admitted directly to the ICU.² In a study of septic patients, Knott et al. found that patients who developed shock while in the ICU had a 39% mortality rate, compared to a 70% mortality rate for subjects who developed shock on an inpatient ward.¹⁵ This pattern of higher mortality among ward patients who are transferred to the ICU is interesting considering that those admitted directly to the ICU from the ED are likely sicker during their ED course. It has been theorized that there is a “golden hour” in septic shock between the onset of global tissue hypoxia and multiorgan failure, during which treatments for septic shock may have maximal benefit^23,24; therefore, this mortality difference may reflect delays in initiation of critical medical interventions. In a study of subjects who developed septic shock after hospital admission, there was almost a fourteen-fold difference in the time from onset of shock to the initiation of inotropic support in subjects who developed septic shock in the ICU versus those who developed shock on the ward (median delay of 22.5 minutes vs. 310 minutes, respectively; p = 0.037).³ Young et al. found that for patients who clinically deteriorate on a hospital ward, a longer delay from the onset of hemodynamic instability to ICU admission was an independent risk factor for in-hospital mortality,²⁵ further strengthening the theory that the higher mortality rates associated with ICU transfer may reflect delays in initiation of critical medical therapies.

LIMITATIONS

One limitation is the low number of events (93) that was found across our study population. Additionally, for our primary outcome measure, we chose a cutoff of 48 hours as our indicator of early transfer to the ICU. Some might argue that 24 hours is a more practical cutoff; others might argue that 72 hours is more appropriate. Another limitation is that our data were obtained through chart reviews; the accuracy of the data is dependent on the quality of the medical records and data abstraction.

We also limited the scope of our study to patients with suspected infection as opposed to a broader population of patients being admitted to the hospital for undifferentiated conditions. As different hospitals may be able to manage different levels of medical acuity on the inpatient wards, the results of this study may not be generalizable to other hospitals, particularly non-tertiary care hospitals. We dichotomized the continuous variables of age, blood pressure, and heart rate because dichotomous variables are easier to apply in clinical prediction rules than continuous variables, and for consistency with prior work.¹⁰ Thresholds selected for dichotomization were determined a priori. As a result, this model may not use these variables in an optimal fashion, and our model may underestimate their effects.

We used selection techniques to develop our initial model, but did not have an independent study sample on which to validate the rule; therefore, this initial set of covariates may be considered exploratory and warrants validation. Between the first and second cohorts, our ED started routinely obtaining lactate levels in patients with suspected infection, which creates a potential for selection bias within our population. One might expect that this policy change would decrease the rate of early ICU transfer as normotensive patients with a high lactate would be admitted directly to the ICU; instead we found no differences in early ICU transfers between the cohorts (first cohort: 48 out of 3,240 patients, or 1.5%; second cohort: 45 out of 2,125 patients, or 2.1%; p = 0.09). Finally, our model demonstrated only moderate predictive value for identifying patients at risk of early ICU transfer, with an AUC of 0.73.

CONCLUSIONS

We found that respiratory compromise, congestive heart failure, peripheral vascular disease, transient hypotension, transient tachycardia, and elevated creatinine were independently associated with early transfer to the intensive care unit. Cellulitis was a marker of patients at lower risk. These results have substantial clinical face validity, and consideration of these factors may assist the emergency physician when determining whether a patient with suspected infection should be admitted to an ICU or inpatient ward. Ward-to-ICU transfer in our patient population was a relatively rare event (less than 2%). Most research on the topic of unanticipated ICU transfer has focused on methods for early detection of and response to clinical deterioration on the inpatient ward, with conflicting results as to whether these strategies reduce mortality rates.^6–9 This underscores the importance of proper risk stratification by emergency physicians. Given the expense of ICU care, accounting for 1% of the gross domestic product,^26,27 and our current economic climate, it is possible we will see a further reduction in ICU bed availability. Under these conditions, accurate ICU risk stratification will become more challenging and important. Improving the accuracy of these decisions may enable better allocation of ICU resources to patients who require this level of care, and either reduce the incidence of early clinical deterioration or ensure that these patients are in a setting where deterioration may be more rapidly and appropriately addressed. Our study was a preliminary investigation into this topic, and further studies are warranted. Potential avenues for future studies include looking at an unselected patient population, using biomarkers for risk-stratification, and assessing for trends in vital signs over the ED course.