a b s t r a c t

Hypothesis: Unplanned intensive care unit (ICU) transfer (UIT) within 48 hours of emergency department (ED) admission increases morbidity and mortality. We hypothesized that a majority of UITs do not have critical inter- ventions (CrIs) and that CrI is associated with worse outcomes.

Objective: The objective of the study is to characterize all UITs (including patients who died before ICU transfer), the proportion with CrI, and the effect of having CrI on mortality.

Design: This is a single-center, retrospective cohort study of UITs within 48 hours from 2008 to 2013 at an urban academic medical center and included patients 18 years or older without advanced directives (ADs). Critical in- tervention was defined by modified Delphi process. Data included demographics, comorbidities, reasons for UIT, length of stay, CrIs, and mortality. We calculated descriptive statistics with 95% confidence intervals (CIs).

Results: A total of 837 (0.76%) of 108 732 floor admissions from the ED had a UIT within 48 hours; 86 admitted patients died before ICU. We excluded 23 ADs, 117 postoperative transfers, 177 planned ICU transfers, and 4 with missing data. Of the 516 remaining, 65% (95% CI, 61%-69%) received a CrI. Unplanned ICU transfer reasons are as follows: 33 Medical errors, 90 disease processes not present on arrival, and 393 clinical deteriorations. Mor- tality was 10.5% (95% CI, 8%-14%), and mean length of stay was 258 hours (95% CI, 233-283) for those with CrI, whereas the mortality was 2.8% (95% CI, 1%-6%) and mean length of stay was 177 hours (95% CI, 157-197) for those without CrI.

Conclusions: Unplanned ICU transfer is rare, and only 65% had a CrI. Those with CrI had increased morbidity and mortality.

(C) 2016

Introduction

Background

A core skill of an emergency medicine physician is determining an effi- cient, accurate disposition for each patient. For cardiovascular or respirato- ry arrest, the decision for intensive care unit admission is straightforward. More complex is determining which patient is at risk for future decompensation. In 1999, the Society of Critical Care Medicine pub- lished ICU admission guidelines, but many of the criteria are vague and left to provider interpretation. The report discusses “priority 1” guidelines (ie,

? Disclaimers: There are no conflicts of interest, and there was no outside funding for this research. This abstract and research was presented at The Society of Critical Care Med- icine Annual Congress in January 21 to 25, 2016.

* Corresponding author at: 1 Boston Medical Center Place, Dowling 1 South, Boston, MA 02118.

mechanical ventilation, invasive hemodynamic monitoring, and/or vaso- active drugs) that require ICU admission [1]. The American Association for the Surgery of Trauma published guidelines recommending certain conditions (ie, multisystem trauma, traumatic brain injury with Glasgow Coma Score <=8, facial or Neck trauma with threatened airway, crush inju- ries, severe burns) that require ICU admission. The list is extensive and in- cludes Ref [2]. Other societies, such as the Infectious Disease Society of America and the American Thoracic Society, provide disease-specific guidelines (ie, community-acquired pneumonia) for ICU admission criteria, but several studies failed to validate these criteria [3-5]. It is often at the discretion of emergency providers to determine which pa- tients pose the risk for acute decompensation, requiring ICU admission.

Importance

The increasing quality-conscious atmosphere in health care necessi- tates effective use of Health care resources. Intensive care unit

http://dx.doi.org/10.1016/j.ajem.2016.05.009

0735-6757/(C) 2016

admissions are costly, provide highly Specialized care, and thus should be used efficiently. Previous studies demonstrated that delayed trans- fers to an ICU level of care are associated with increased mortality and length of stay [6-9]. As a result, early (24-72 hours) unplanned ICU transfer (UIT) from a non-ICU setting has been used as a quality mea- sure for emergency department (ED) providers.

Previous literature evaluated UIT patients early in their ED course. One prospective, observational study looked at patients with infections and found vital signs (ie, respiratory rate, systolic blood pressure, pulse), laboratory indicators (ie, high creatinine), and a history of congestive heart failure and/or peripheral vascular disease were independent pre- dictors of UIT [10]. Other retrospective studies have found other predic- tors of UIT including vital signs (tachypnea), laboratory findings (low bicarbonate), comorbidities (chronic obstructive pulmonary disease, coronary artery disease), admitting diagnoses (gastrointestinal hemor- rhage, pneumonia), and score systems (Comorbidity Points Score) [11-13]. From the literature, there is minimal consensus on which pre- dictors accurately identify patients at risk for clinical deterioration who are not critically ill on arrival to the ED.

Goals

The purpose of this study was to examine all UITs to the ICU within 48 hours of admission from the ED to a non-ICU setting as an indicator of quality of care in the ED. Although some studies have focused on specific diseases [10] and others have focused on generalized early UIT [11-13], our goal was to capture all rapid clinical deteriorations to better define the epidemiology. We included patients who were transferred from any non-ICU setting to any ICU (medical, cardiac, surgical, neurocritical) and patients who died within 48 hours of ED admission before ICU transfer. We sought to characterize the reasons for UITs (Table 1) and describe the outcomes. In addition, we examined which patients re- ceived critical interventions (CrIs), that is, cardiopulmonary resuscita- tion (CPR), unplanned operative intervention, and/or invasive hemodynamic monitoring.

We hypothesized that a large proportion of those with UIT do not undergo CrIs and are transferred for closer monitoring or nursing care. In addition, we suspected that ICU transfer patients requiring a CrI would have worse outcomes than transfer patients who did not receive a CrI.

Methods

Study design and setting

We performed a 5-year (June 2008 to June 2013) retrospective, cohort study using standardized chart review methods. The study

site is a single, 500-bed tertiary care urban hospital including more than 40 adult ICU beds and an ED that sees more than 130 000 visits each year. Approximately 80% of ICU admissions are referred from the ED.

A 5-year time frame was chosen to capture an adequate sample size for the study by estimating approximately 20 000 non-ICU admissions annually at the study site from prior years and a UIT rate ranging anywhere from 1% to 9% based on recent literature [14-16]. Using the lowest possible UIT rates to abstract data, we estimated approximately 200 cases of UIT per year. The 5-year time frame would therefore pro- duce an underestimated minimum sample size of approximately 1000 patients, which would be sufficient to provide reasonable confidence in- tervals around the major estimates of interest including annual propor- tion of non-ICU admissions with UIT, proportion with CrI, and reasons for UIT.

This study was approved by the Boston University Medical Campus Institutional Review Board with a waiver of informed consent.

Selection of participants

The study sample included patients 18 years or older admitted to a non-ICU setting from the ED. Additional criteria included patients who were transferred to an ICU (medical, surgical, cardiac, or neurologic), within 48 hours and those who died within 48 hours of admission to a non-ICU setting. We excluded those with active comfort-measures- only, do-not-resuscitate (DNR), and do-not-intubate (DNI) orders on admission; postoperative complications; and planned ICU transfers. An electronic query of the hospital’s clinical data warehouse generated a list of patients who met inclusion criteria.

We used 5 independent categories to define reasons for UIT (see Table 1). The 3 categories included in the investigation of CrIs, and out- come data were medical error leading to transfer (including diagnostic errors or other therapeutic errors), condition not present on arrival leading to deterioration and ICU transfer, and condition present on ar- rival leading to deterioration and ICU transfer. The 2 categories excluded from this analysis were patients admitted to the floor with a planned ICU transfer or a postoperative transfer to the ICU.

Methods and measurements

We observed procedures that are recommended to enhance the va- lidity of chart review studies [17-20]. Three chart reviewers completed a pilot review of 30 charts to develop a standardized methodology and create consensus for future abstraction. Criteria for case inclusion and exclusion were reviewed; and important variables, defined. The princi- pal investigator trained chart abstractors in data abstraction (Appendix I). Reviewers used standardized abstraction forms, which contained no

Table 1

Categories and definitions of UIT within 48 hours

Reason for transfer Definition Examples

Medical error Any diagnostic or treatment error that leads to clinical decompensation

Present on arrival Clinical deterioration is related to an illness or injury that was present on arrival to the ED

Not present on arrival Unpredictable patient course that cannot be attributed to

deterioration of the presenting illness.

86-year-old woman with suprapubic abdominal pain admitted for a urinary tract infection; found to have pneumoperitoneum and a sigmoid perforation.

A 34-year-old man, T2 paraplegic admitted to the floor with a urinary tract infection and transferred to the ICU for septic shock requiring vasopressors and a central line.

A 62-year-old man, daily alcohol drinker, suffered a fall with multiple Facial fractures requiring operative repair; transferred to the ICU for alcohol withdrawal.

Postsurgical complication Patient requiring unexpected ICU level of care postoperatively. A 42-year-old man with history of multiple abdominal surgeries who

went to the OR for an incarcerated ventral hernia. Intraoperatively, necrotic bowel found and patient transferred postoperative to the ICU for fluid resuscitation and monitoring.

Planned ICU transfer on admission Patient who is admitted to a non-ICU level of care (operating

room or floor or interventional radiology) and then the ICU as part of regular, planned care.

A 71-year-old man with metastatic Lung cancer to his spine that underwent an emergent Decompressive laminectomy; transferred postoperative for hourly neurologic checks.

Interrater agreement“>identifiable information and a unique study ID number for each form. A separate untitled log was kept on a password-protected computer that linked study ID numbers to medical record numbers.

We used a modified Delphi to develop a list of CrIs [21]. Five experts (2 intensive care physicians and 3 emergency medicine physicians) de- veloped an explicit list of CrIs. We used the initial statement from the Society of Critical Care Medicine and American Association for the Sur- gery of Trauma and through an iterative process developed a prespecified list of interventions that would be considered CrIs (Table 2).

Analysis

We examined interrater agreement by using 2 other blinded inde- pendent reviewers who reabstracted a 10% random sample of charts for key independent variables. We used ? to assess interrater reliability for these key predictors. We held scheduled meetings with chart ab- stractors and study coordinators to resolve disputes and review coding rules throughout the 3-month data abstraction period. We used SAS 9.3 to perform all analyses.

Results

Characteristics of study subjects

From June 1, 2008, to June 1, 2013, there were an estimated 520 202 visits to the ED, and 108 732 were admitted to a non-ICU setting. A total of 923 patients were identified as being transferred to an ICU setting from the ED or expired within 48 hours (0.9%). Of these 923 patients, we found 837 (91%) were transferred to an ICU, and 86 (9.3%) died (Figure). Among those who were transferred, we excluded 4 due to missing information, 23 DNR and DNI, 117 had a postoperative compli- cation, and 177 patients had a planned ICU transfer. A total of 516 eligi- ble patients had a UIT within 48 hours. Those who received a CrI were similar in demographic characteristics to those who did not receive a CrI (Table 3). Among patients who died, we excluded 6 due to missing

Table 2 Frequency and percentage of critical interventions within those transferred to an ICU and/ or those who died within 48 hours

Critical interventions	n	%
Unplanned emergent intubation	92	27.5%
Unplanned emergent positive airway pressure ventilation	43	12.9%
Cardiopulmonary Resuscitation	10	3.0%
Defibrillation or electrical cardioversion	3	0.9%
Transvenous or transcutaneous pacing	1	0.3%
Continuous vasoactive drugs	46	13.8%
N 2 units blood transfusion within 12 hours of transfer	54	16.2%
N 4 liters intravenous fluid bolus within 12 hours of transfer	31	9.3%
Continuous benzodiazepine, barbiturate, or other Sedative agent	31	9.3%
Continuous nitroglycerin or other antihypertensive agent	24	7.2%
Continuous dextrose infusion	7	2.1%
Continuous naloxone infusion	1	0.3%
Continuous (or frequency N every 1 hour) albuterol nebulization	10	3.0%
Emergent dialysis (defined by dialysis required for pH b 7.1, respiratory	3	0.9%
compromise [Respirations N 35], toxic drug level, potassium N 7, calcium N 15, neurologic compromise related to uremia) Unplanned emergent endoscopy	22	0.9%
Unplanned emergent Interventional Radiology embolization	5	1.5%
Unplanned emergent catheterization	3	0.9%
Invasive neurologic monitoring	1	0.3%
Invasive hemodynamic monitoring	78	23.4%
Mannitol or hypertonic saline infusions	2	0.6%
Unplanned thrombolytic infusion (for pulmonary embolism or stroke)	8	2.4%
Unplanned emergent pericardiocentesis	0	0.0%
Unplanned emergent cardiac assist devices	0	0.0%
Unplanned emergent operative intervention	47	14.1%
High oxygen requirement (requiring more than nasal cannula oxygenation)	77	23.1%
Other	29	8.7%

information, 46 DNR and DNI, 2 with postoperative complications, and 1 had a planned ICU transfer. Thus, 31 dead patients were included.

Main results

Within the 516 patients with a UIT, 334 patients received at least 1 CrI found in Table 2. The most common CrIs were an unplanned emer- gent intubation (92), invasive hemodynamic monitoring (78), and high O₂ requirement (77). Among patients who were transferred and received a CrI, the total number of CrIs ranged from 1 to 7, and 92.5% re- ceived 3 CrIs or fewer. In dead patients, only 3 of 31 did not receive a CrI. The most common CrIs were CPR and unplanned emergent intubation, with 75% of the expired patients receiving one or the other or both.

For patients with UIT, the median (interquartile range) hospital length of stay for those with a CrI was 184.2 (186) hours, com- pared to 143.9 (125.3) hours in those who did not receive a CrI (Table 4). The mean LOS for those receiving a CrI was on average 80.7 hours (95% confidence interval [CI], 48.7-112.7) longer than those re- ceiving no CrI. Percent mortality was higher in those receiving a CrI (10.5%; 95% CI, 7.6%-14.3%) compared to receiving no CrI (2.8%; 95%

CI, 1%-6.5%).

Interrater agreement

The chance-corrected AGreement for the inclusion variable (? = 0.81) and the admission criteria variable (? = 0.84) was almost perfect for the 93 charts abstracted to determine interrater agreement. The chance-corrected agreement for the transfer category variable (? = 0.67) indicated substantial chance-corrected agreement.

Limitations

There are limitations inherent in retrospective chart reviews that we attempted to minimize with methods recommended to improve the va- lidity of data gathering.

We used an explicit coding scheme for reasons for UIT and tested its reliability. However, the reasons for UIT can be subject to hindsight bias. For example, whether a patient with a history of heavy alcohol use who had no clinical signs of withdrawal in the ED and later developed severe alcohol withdrawal requiring ICU care was coded as present on arrival (chronic alcoholism at risk for Acute withdrawal) or unanticipated as the cause for acute deterioration was a coding judgment that led to var- iation in categorization of the reason for UIT.

Our list of CrIs, although based on available guidelines and devel- oped by an explicit process, could allow potential misclassification if it was not complete. However, our outcome of CrI was well defined and had high reliability in coding. In addition, the end point of having a CrI could lead to misclassification due to the appropriateness of the inter- vention. For example, a patient with a gastrointestinal hemorrhage who had an unplanned but negative endoscopy would still be consid- ered a UIT with CrI. Although we considered having a CrI a more appro- priate outcome rather than just transfer to an ICU, having a CrI that is not therapeutic may also have limitations to identify patients who had adverse outcomes related to delay to ICU transfer.

This study used mortality as an outcome measure but included pa- tients who were made comfort measures only, DNR, and/or DNI during their ICU stay. This problem is intrinsic to all similar retrospective liter- ature to date. These patients should be excluded or considered a prespecified subgroup in studies of UIT and quality measurement of ED Disposition decisions.

This is a single-center study, and our results may not be generaliz- able to other centers with different specifications for ICU level of care. For example, the study hospital necessitates ICU admission for all con- tinuous Insulin infusions or noninvasive ventilation. At other centers, the proportion of patients with UIT and the reasons for UIT may vary.

Figure. Flow diagram showing derivation of study sample.

Discussion

Our study contributes important insight into understanding the use of UIT as a measure of quality care and an adverse event screening tool. By capturing all patients who decompensated within 48 hours of admis- sion but did not survive to ICU admission and reviewing all ICU trans- fers, we were able to examine the rate of such events in all patients who were at risk for this event. Other studies have either focused on 1 type of ICU (medical ICU), 1 disease state (suspected infection or sepsis), or only included patients admitted to an ICU (and not dead patients) [10-13]. Our analysis provides a robust estimate of the frequency of this event in the population at risk. Second, by identifying patients who required a CrI, we provide a more precise estimate of those who

suffered morbidity potentially related to ED treatment and triage deci- sions. One recent Belgian study examined the prevalence and prevent- ability of adverse events that led to UIT, but no attempt was made to examine what, if any, CrIs were required subsequent to UIT [22]. Our study determined that approximately one-third of patients with UIT did not have CrIs. As an observational study, our goal was not to deter- mine the “appropriateness” of these non-CrI transfers. However, includ- ing all patients with UIT as an indicator of poor Quality of ED care, where all UITs trigger a mandatory review, would appear overly sensitive.

Furthermore, this study provides perspective of UIT as a crude event rate similar to 30-day readmissions to a hospital after discharge. This all- cause definition is used in calculating the national average readmission rate and a hospital’s specific readmission rate. However, such crude

Table 3

Demographics/characteristics of those transferred to ICU and those who died within 48 hours

Received CrI (n = 334) No CrI (n = 182) Died^a (n = 31) Age, mean (SD) 56.5 (18.5) 54.8 (17.5) 68.9 (13.7)

Sex (%)

Male	57.2%	58.2%	51.6%
Female	42.8%	41.8%	48.4%
Race (%) Asian	1.8%	2.2%	3.2%
Black	47.0%	51.1%	64.5%
Hispanic	11.7%	12.1%	12.9%
White	37.7%	34.1%	16.1%
Other/unknown	1.8%	0.6%	3.2%

^a Patients who died before ICU transfer.

measures may perform poorly as indicators of quality or identification of preventable adverse outcomes. A recent analysis of 30-day readmis- sion rates reported that there was a very weak correlation between hos- pitals and the rate of preventable readmissions, all readmissions, and potentially preventable readmissions [23].

The decision to transfer a patient to an ICU who does not require a CrI could be affected by both provider and institution factors such as nurse to physician ratio, hospital protocols, and capability of interven- tions in a non-ICU setting. Using a crude measure of UIT may be a mea- sure of hospital staffing ratios more than a measure of quality.

Unplanned ICU transfer remains a recommended and well-accepted

screening tool used to identify medical or treatment errors in the ED. The indicator was developed by the Australian and New Zealand College of Anesthetists and the Australian Council on Healthcare Standards as an indicator for patient safety and is a trigger recommended by the In- stitute for Healthcare Improvement Global Trigger Tools for Measuring Adverse Events (IHI GTT G8) to identify adverse events [24,25].

Unplanned ICU transfer can be a useful screening tool to identify po- tential problems. We found that 6.3% of cases were related to Medical errors, and 76% had decompensation of a condition present on arrival. However, when crude event rates such as UITs are used for such high- stakes activities as pay for performance or public reporting, the limita- tions of this measure must be recognized. Our study suggests that any intervention designed to reduce the rate of UITs or prospectively indi- cate a patient at risk for UIT will require a very large multicenter pro- spective clinical trial to demonstrate its validity and cost-effectiveness given our observation that only 0.9% of all ED admissions had a UIT. Rec- ommendations that specific patients such as those with respiratory con- ditions or sepsis receive “better triage” or “closer monitoring” do not provide useful guidance [13].

There is substantial variation in the prevalence of UITs, and this esti- mate is affected by the amount of time used for identification (within 24-72 hours of admission from the ED). Posa et al [16] reported that 1% to 9% of all ICU admissions were unplanned, and Delgado et al [13] found that 2.4% of UITs from 13 US community EDs were within 24 hours of ED admission [13,16]. The sensitivity of using UIT as a trigger for preventable adverse event detection has varied from 18.6% to 25.9% [22,26]. A systematic review concluded that the percentage of surgical and medical adverse events requiring ICU admission ranged

Table 4

hospital outcomes comparing those with and without CrI

CrI No CrI Difference in means

Inpatient LOS

Mean (SD)	257.7 (231.3)	177.1 (137.7)	80.7
95% CI around mean	232.8-282.6	156.7-197.4	48.6-112.7
Median (IQR) Mortality	184.2 (186)	143.9 (125.3)
%	10.5%	2.8%
95% CI	(7.6%-14.3%)	(1.0%-6.5%)

Abbreviation: IQR, interquartile range.

from 1.1% to 37.2%. Furthermore, the preventability of the AEs varied from 17% to 76.5% [27].

Prior studies have attempted to better identify patients at risk for clinical decompensation and possible UIT [10-13]. Emergency medicine physicians must maintain a High clinical suspicion for potentially ill pa- tients. However, they cannot be 100% sensitive in the triage decision for all patients who could potentially decompensate within 24 to 72 hours of hospital admission without admitting patients unnecessarily to an ICU. Such resource overutilization could worsen ED boarding, delay ad- missions of other more critical patients to the ICU, and increase the costs of medical care.

In summary, our study indicates that UIT is rare when viewed from the perspective of all patients who are admitted to a non-ICU level of care. Quality reviews should focus on patients who underwent a CrI as these patients are at increased risk for morbidity and mortality. Finally, our study indicates the need for further research to develop and pro- spectively test interventions to reduce UITs.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.ajem.2016.05.009.

References

1. Guidelines for intensive care unit admission, discharge, and triage Task Force of the American College of Critical Care Medicine, Society of Critical Care MedicineCrit Care Med 1999;27:633-8.
2. Jenkins T. American Association for the Surgery of Trauma. ICU Admission Criteria. https://admissiontoicucriteriahildreth-deidentified-june-2014-pdf.pdf. [accessed September 08, 2015].
3. Brown SM, Jones BE, Jephson AR, Dean NC. Infectious Disease Society of America/ American Thoracic Society 2007. Validation of the Infectious Disease Society of America/American Thoracic Society 2007 guidelines for severe community- acquired pneumonia. Crit Care Med 2009;37:3010-6. http://dx.doi.org/10.1097/ CCM.0b013e3181b030d9.
4. Phua J, See KC, Chan YH, Widjaja LS, Aung NW, Ngerng WJ, et al. Validation and clin- ical implications of the IDSA/ATS minor criteria for severe community-acquired pneumonia. Thorax 2009;64:598-603. http://dx.doi.org/10.1136/thx.2009.113795.
5. Renaud B, Santin A, Coma E, Camus N, Van Pelt D, Hayon J, et al. Association between timing of intensive care unit admission and outcomes for emergency department patients with community-acquired pneumonia. Crit Care Med 2009;37:2867-74. http://dx.doi.org/10.1097/CCM.0b013e3181b02dbb.
6. Goldhill DR, White SA, Sumner A. Physiological values and procedures in the 24 h before ICU admission from the ward. Anaesthesia 1999;54:529-34.
7. Liu V, Kipnis P, Rizk NW, Escobar GJ. Adverse outcomes associated with delayed in- tensive care unit transfers in an integrated healthcare system. J Hosp Med 2012;7: 224-30. http://dx.doi.org/10.1002/jhm.964.
8. Mardini L, Lipes J, Jayaraman D. Adverse outcomes associated with delayed intensive care consultation in medical and surgical inpatients. J Crit Care 2012;27:688-93. http://dx.doi.org/10.1016/j.jcrc.2012.04.011.
9. Lundberg JS, Perl TM, Wiblin T, Costigan MD, Dawson J, Nettleman MD, et al. Septic shock: an analysis of outcomes for patients with onset on hospital wards versus in- tensive care units. Crit Care Med 1998;26:1020-4.
10. Kennedy M, Joyce N, Howell MD, Lawrence Mottley J, Shapiro NI. Identifying infect- ed emergency department patients admitted to the hospital ward at risk of clinical deterioration and intensive care unit transfer. Acad Emerg Med 2010;17:1080-5. http://dx.doi.org/10.1111/j.1553-2712.2010.00872.x.
11. Caterino JM, Jalbuena T, Bogucki B. Predictors of acute decompensation after admis- sion in ED patients with sepsis. Am J Emerg Med 2010;28:631-6. http://dx.doi.org/ 10.1016/j.ajem.2009.04.020.
12. Farley H, Zubrow MT, Gies J, Kolm P, Mascioli S, Mahoney DD, et al. Emergency de- partment tachypnea predicts transfer to a higher level of care in the first 24 hours after ED admission. Acad Emerg Med 2010;17:718-22. http://dx.doi.org/10.1111/j. 1553-2712.2010.00796.x.
13. Delgado MK, Liu V, Pines JM, Kipnis P, Gardner MN, Escobar GJ. Risk factors for un- planned transfer to intensive care within 24 hours of admission from the emergency department in an integrated healthcare system. J Hosp Med 2013;8:13-9. http://dx. doi.org/10.1002/jhm.1979.
14. Goldhill DR, McNarry AF, Hadjianastassiou VG, Tekkis PP. The longer patients are in hospital before intensive care admission the higher their mortality. Intensive Care Med 2004;30:1908-13. http://dx.doi.org/10.1007/s00134-004-2386-2.
15. Molina JAD, Seow E, Heng BH, Chong WF, Ho B. Outcomes of direct and indirect medical intensive care unit admissions from the emergency department of an acute care hospital: a retrospective cohort study. BMJ Open 2014;4:e005553. http://dx.doi.org/10.1136/bmjopen-2014-005553.
16. Posa PJ, Yonkee DE, Fields WL. Development and implications of an interdisciplinary quality assurance monitor on unplanned transfers into the intensive care units. J Nurs Care Qual 1992;6:51-5.
17. Gilbert EH, Lowenstein SR, Koziol-McLain J, Barta DC, Steiner J. Chart reviews in emergency medicine research: where are the methods? Ann Emerg Med 1996; 27:305-8.
18. Lowenstein SR. Medical record reviews in emergency medicine: the blessing and the curse. Ann Emerg Med 2005;45:452-5. http://dx.doi.org/10.1016/j.annemergmed. 2005.01.032.
19. Worster A, Bledsoe RD, Cleve P, Fernandes CM, Upadhye S, Eva K. Reassessing the methods of medical record review studies in emergency medicine research. Ann Emerg Med 2005;45:448-51. http://dx.doi.org/10.1016/j.annemergmed.2004.11. 021.
20. Kaji AH, Schriger D, Green S. Looking through the retrospectoscope: reducing bias in emergency medicine chart review studies. Ann Emerg Med 2014;64:292-8. http:// dx.doi.org/10.1016/j.annemergmed.2014.03.025.
21. Lindsay P, Schull M, Bronskill S, Anderson G. The development of indicators to mea- sure the quality of clinical care in emergency departments following a modified- Delphi approach. Acad Emerg Med 2002;9:1131-9.
22. Marquet K, Claes N, De Troy E, Kox G, Droogmans M, Schrooten W, et al. One fourth of unplanned transfers to a higher level of care are associated with a highly

    preventable adverse event: a patient record review in six Belgian hospitals. Crit Care Med 2015;43:1053-61. http://dx.doi.org/10.1097/CCM.0000000000000932.

    Boccuti C, Casillas G. Aiming for fewer hospital U-turns: the Medicare hospital read- mission reduction program. 2015 n.d. http://kff.org/medicare/issue-brief/aiming- for-fewer-hospital-u-turns-the-medicare-hospital-readmission-reduction-pro- gram/. [accessed December 16, 2015].

23. Accreditation and the NSQHS Standards. Safety and quality. n.d. http://www. safetyandquality.gov.au/our-work/accreditation-and-the-nsqhs-standards/. [accessed December 16, 2015].
24. Institute for Healthcare Improvement: IHI global trigger tool for measuring adverse events.

    n.d. http://www.ihi.org/resources/pages/tools/ihiglobaltriggertoolformeasuringaes.aspx. [accessed December 16, 2015].

    Soop M, Fryksmark U, Koster M, Haglund B. The incidence of adverse events in Swedish hospitals: a retrospective medical record review study. Int J Qual Health Care 2009;21:285-91. http://dx.doi.org/10.1093/intqhc/mzp025.

25. Vlayen A, Verelst S, Bekkering GE, Schrooten W, Hellings J, Claes N. Incidence and pre- ventability of adverse events requiring intensive care admission: a systematic review. J Eval Clin Pract 2012;18:485-97. http://dx.doi.org/10.1111/j.1365-2753.2010.01612.x.