Identification and validation of objective triggers for initiation of resuscitation management of acutely ill non-trauma patients: the INITIATE IRON MAN study

Background While there are clear national resuscitation room admission guidelines for major trauma patients, there are no comparable alarm criteria for critically ill nontrauma (CINT) patients in the emergency department (ED). The aim of this study was to define and validate specific trigger factor cut-offs for identification of CINT patients in need of a structured resuscitation management protocol. Methods All CINT patients at a German university hospital ED for whom structured resuscitation management would have been deemed desirable were prospectively enrolled over a 6-week period (derivation cohort, n = 108). The performance of different thresholds and/or combinations of trigger factors immediately available during triage were compared with the National Early Warning Score (NEWS) and Quick Sequential Organ Failure Assessment (qSOFA) score. Identified combinations were then tested in a retrospective sample of consecutive nontrauma patients presenting at the ED during a 4-week period (n = 996), and two large external datasets of CINT patients treated in two German university hospital EDs (validation cohorts 1 [n = 357] and 2 [n = 187]). Results The any-of-the-following trigger factor iteration with the best performance in the derivation cohort included: systolic blood pressure < 90 mmHg, oxygen saturation < 90%, and Glasgow Coma Scale score < 15 points. This set of triggers identified > 80% of patients in the derivation cohort and performed better than NEWS and qSOFA scores in the internal validation cohort (sensitivity = 98.5%, specificity = 98.6%). When applied to the external validation cohorts, need for advanced resuscitation measures and hospital mortality (6.7 vs. 28.6%, p < 0.0001 and 2.7 vs. 20.0%, p < 0.012) were significantly lower in trigger factor-negative patients. Conclusion Our simple, any-of-the-following decision rule can serve as an objective trigger for initiating resuscitation room management of CINT patients in the ED. Supplementary Information The online version contains supplementary material available at 10.1186/s13049-021-00973-4.

Background Trauma patients are regularly admitted directly to the resuscitation room of an emergency department (ED) at dedicated trauma centers. The relevant alert criteria (e.g., trauma mechanism and pattern, Glasgow Coma Scale [GCS] score < 9 points, systolic blood pressure [SBP] < 90 mmHg, need for intubation) are clearly specified by national guidelines [1]. However, there are no comparable triggers for initiating resuscitation room management of critically ill nontrauma (CINT) patients, posing a challenge for their identification by emergency medical services (EMS) or triage nurses. Early studies on nontrauma resuscitation room management have found that CINT patients account for about 1.5-2.0% of all patient contacts in German EDs [2,3] and that the nontrauma vs. trauma ratio of resuscitation room patients may be as high as 4:1 in Germany and Belgium [4]. Although the 30-day mortality of CINT patients is around twice that of trauma patients [2], a recent survey showed that only ~ 50% of the participating German EDs use alert criteria to initiate interdisciplinary resuscitation room management of CINT patients, which are mostly limited to classic acute medical conditions triaged according to the Airway, Breathing, Circulation, Disability, and Exposure (ABCDE) scheme [5]. After establishing a structured resuscitation room protocol based on the ABCDE algorithm (Interdisciplinary Resuscitation Room Management of Acutely Ill Nontraumatic Patients [IRON MAN] protocol; Additional file 1: Fig. S1), we recognized that, aside from prehospital intubation [6], there are no objective triggers available in the literature. The aim of this study was therefore to specify and validate objective trigger factor cut-offs for initiating structured resuscitation management in non-intubated CINT patients. A total of 4 different datasets were used: a prospective derivation cohort, a retrospective validation cohort, and two external validation cohorts (Fig. 1).

Study design-overview
The different cohorts in this study are shown in Fig. 1. As a first step, several combinations of quantitative variable thresholds for initiating resuscitation room care were evaluated and optimized in a small prospectively assembled cohort of CINT patients (derivation cohort). The most promising iterations of these trigger factor combinations were compared with established emergency medicine scores. The sensitivity and specificity of the iterations were tested in an independent dataset of consecutive ED patients from the same institution, including non-CINT patients (internal validation cohort). The performance of the most promising trigger factor combination was then evaluated in two large external Keywords: Resuscitation room, Conservative shock room, Nontrauma patients, Emergency department, Trigger factor  Demographic variables, routine biochemistry tests, and physiologic parameters were obtained for each subject during acute care (Table 1). It was irrelevant whether the patients had actually been treated in the resuscitation room or elsewhere in the ED. Emergency patients who bypassed the ER based on predefined treatment pathways, guideline recommendations, or local agreements (e.g., ST-elevation myocardial infarction or prehospital cardiac arrest) were excluded. Prehospital intubated patients were also excluded, as such emergency patients require structured resuscitation management and care anyway. Sepsis was defined according to the second American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) consensus criteria (Sepsis-2) [7]. The requirement for informed consent for Emergency medicine score

Internal validation cohort
Electronic chart records from all nontrauma patients admitted to the ED of the University Hospital Muenster in November 2020 (n = 1069) were reviewed by four senior ED physicians with emergency and critical care expertise. Actual room allocation (resuscitation room or elsewhere within the ED) was retrieved from the hospital information system (Orbis; Dedalus Healthcare, Bonn, Germany) by data query. Resuscitation room stays for planned interventions (e.g., cardioversion) were excluded.
After blinding for actual allocation, the reviewers determined for each case whether there had been an indication for a structured resuscitation management (e.g., IRON MAN). Each case was reviewed by at least two referees. If there was no consensus, a third expert (PK) made the final decision. Cases with purely administrative recordings (n = 30; ED bypassing patients) or incomplete/missing data (n = 46; mainly self-presenters) were excluded. The final dataset comprised 993 patients.

External validation cohorts OBSERvE study
The Observation of Critically Ill Patients in the Resuscitation Room of the Emergency Department (OBSERvE) study [2] was a prospective observational study in 532 CINT patients aged ≥ 18 years treated in the resuscitation room at the ED of the University Hospital of Leipzig (Germany) between September 2014 to August 2015.

OBSERvE-Duesseldorf study
The OBSERvE-Duesseldorf (OBSERvE-DUS) study is an ongoing retrospective cohort study of all CINT patients aged ≥ 18 years treated in the resuscitation room of the ED at the University Hospital Duesseldorf. Data on allocation to the resuscitation room from March 2018 to October 2018 were retrieved from the Patient Data Management System (COPRA; COPRA System GmbH, Berlin, Germany) by data query.
In both OBSERvE-cohorts, referral to the resuscitation room was at the discretion of a senior ED physician with expertise in emergency and critical care. The qualitative criteria used for admission to the resuscitation room (no cut-off values were set or specified) are listed in Additional file 3: Table S1 [2]. Numerical trigger factors or thresholds were not specified. After excluding intubated patients (see explanation above) and cases with missing data, the final datasets consisted of 357 patients from the OBSERvE-study and 187 patients from the OBSERvE-DUS study, respectively ( Table 1).

Derivation of objective trigger factor iterations
Unlike other risk or mortality scores, objective trigger factors for initiating resuscitation room management must be readily available upon arrival at the ED and determined before room assignment. We also wanted to capture the enumerative/qualitative ABCDE logic with only a few, objective surrogate parameters. Briefly, we used preliminary decision tree models and receiver-operator characteristic (ROC) curves (SPSS v26, IBM Corporation, Armonk, NY, USA) to identify possible cut-off values for the following (on admission readily available) vital signs: respiratory rate, SBP, and peripheral oxygen saturation (SpO 2 ) (A/B); heart rate (C); GCS score (D); and body temperature (E). Subsequently, an interdisciplinary team of intensivists and emergency physicians combined the identified objective cut-offs into several, slightly different trigger factor iterations (hereafter called  IRON MAN iterations). These combinations were then tested and further optimized in the derivation cohort (see results for more detail). Frequencies of the vital sign cutoffs that were used are shown in Additional file 2: Fig.  S2.

Comparison with established severity scores
The IRON MAN iterations were compared with the Quick Sequential Organ Failure Assessment (qSOFA) score [8] and the National Early Warning Score (NEWS) endorsed by the National Health System of the United Kingdom [9]. The NEWS system (0-20 points) recommends "emergency assessment by a clinical team/critical care outreach team with critical-care competencies" at ≥ 7 points; accordingly, a score ≥ 7 was considered as positive ( Table 2).

Outcome measures
In the derivation cohort, we analyzed the performance of the scores and different IRON MAN iterations. Outcomes analyzed were: • the detection of cases of (defined as the percentage of cases in the total cohort identified as "positive"), and • the detection of deaths (defined as the percentage of non-survivors (during hospital stay) identified as "positive" by any score or iteration, respectively).
In the internal validation cohort, the primary outcome was sensitivity, specificity, and positive and negative predictive values (PPV and NPV), respectively, of the different iterations and comparators. As a secondary outcome, we analyzed whether patients with an indication for a structured resuscitation management had actually been treated in the resuscitation room or elsewhere in the ED. In the external validation cohorts, we compared resource use and 30-day mortality between CINT patients that were "positive" or "negative", based on the best performing IRON MAN iteration from the internal validation cohort, respectively.

Statistical analysis
Data are presented as absolute numbers, percentages, or medians with corresponding 25th and 75th percentiles (interquartile range) as appropriate. Differences between groups were analyzed with a two-sided Mann-Whitney U test in the case of continuous variables, and the chisquared test was used to compare categorical variables. Receiver operator characteristic (ROC) curve analysis was carried out and 2 × 2 contingency tables were used to calculate the area under the ROC curve (AUC), sensitivity, specificity, PPV and NPV for different trigger factor combinations. The distribution of the time-toevent variables were estimated using the Kaplan-Meier method with log-rank testing. All tests were two-sided and statistical significance was set at p < 0.05. SPSS v26 (IBM Corporation, Armonk, NY, USA) and Prism v8.4.3 (GraphPad Inc, San Diego, CA, USA) were used for statistical analyses and to plot the data.

Derivation and performance of different IRON MAN iterations
The 108 patients in the derivation cohort had a mean age of 68 years and 47% were female. According to the inclusion criteria, none of the patients was intubated by EMS. The most common reason for admission was infection/ sepsis, followed by acute cardiovascular, neurologic, or respiratory problems. In total, 44.4% of patients had to be transferred to the ICU, and the in-hospital mortality rate was 15.7% (Table 1).
When looking at individual cases, a key finding was that some patients with "mild" concurrent A/B/C problems could only be identified as positive by cumulative iterations. However, patients with e.g., a fulminant isolated C problem were more often detected using a conservative any-of-the-following iteration. Another finding was that if the threshold for the GCS score was set to ≤ 13 points, several patients with sole (but critical) D problems would not have been detected at all. To address this problem, we also tried using the cumulative qSOFA score with the original cutoffs (i.e., GCS < 15) in combination very low threshold values for SpO 2 and SBP ( Table 2).
Several approaches were adopted to adequately assess oxygen supplementation; we obtained the best results when we used SpO 2 under room air (i.e., the value before oxygen administration by EMS) without taking into account the oxygen supplementation that was actually provided. Individual aspects of patient history, comorbidities, or type of admission (i.e., with or without accompaniment by an out-of-hospital emergency physician) had no significant effect on the results (data not shown). The three most promising IRON MAN trigger factor iterations are shown in Table 2.

Internal validation
We next tested the internal validity of the three IRON MAN iterations in a consecutive sample of nontrauma patients who presented at our ED in November 2020. According to the blinded chart review, 68 of 993 patients (6.85%) should have been treated by a structured resuscitation management protocol (~ 2.1 per day). The AUC, sensitivity, specificity, PPV, and NPV of the different iterations and comparators are shown in Table 3. Iteration 3 showed the best results, with a sensitivity of 98.5% (i.e., missing just 1 CINT patient) and specificity of 98.6% (i.e., 13 false alarms). Interestingly, the majority of positive Most of the 80 patients detected by iteration 3 were assigned to the neurology (52.5%) and internal medicine (42.5%) departments, and the most common Manchester Triage System categories were orange (65.0%) and yellow (27.5%), followed by red and green (both 3.75%). The rate of ICU admissions (32.5 vs. 4.3%; p < 0.0001) and deaths (22.5 vs. 0.6%; p < 0.0001) was significantly higher among Table 2 Components and scores of established scoring systems and most promising trigger factor iterations AVPU, Alert, Verbal, Pain, Unresponsive; GCS, Glasgow Coma Scale; HR, heart rate; n.a., not available; NEWS, National Early Warning Score; qSOFA, Quick Sequential Organ Failure Assessment Score; RR, respiratory rate; SBP, systolic blood pressure; SpO 2 , peripheral oxygen saturation *The NEWS score (0-20 points) recommends "emergency assessment by a clinical team/critical care outreachteam with critical-care competencies" at ≥7 points. Accordingly, a NEWS ≥7 cut-off was considered as positive n.a n.a n.a patients identified by iteration 3 compared to those that were not detected by iteration 3. After unblinding the actual treatment room assignment, we found that 29/68 patients (42.6%) had been originally allocated to normal ED rooms (obvious undertriage). Only 27/925 non-CINT patients (2.9%) treated in a dedicated resuscitation room would not have needed this resource (possible over-triage) based on the experts' ex post opinion. These data are consistent with our clinical impression that CINT patients are often not recognized as such by healthcare professionals when admitted to the ED.

Discussion
In the present INITIATE IRON MAN study, we aimed to identify objective trigger factors for the identification of non-intubated ED patients in need of a structured resuscitation protocol for the first time. Using several complementary patient cohorts, we demonstrated that a relatively simple and pragmatic "any-of-the-following" rule can identify a large proportion of such CINT patients in the ED with high sensitivity and specificity. Specifically, this rule includes as variables out-of-hospital vasopressor use, SBP < 90 mmHg, SpO 2 < 90%, and GCS score < 15 points. In keeping with the MARVELinspired title of the study, the mnemonic for the initiation of the IRON MAN protocol in our ED is therefore "ViSiON rule" (Out-of hospital Vasopressor administration, Systolic Blood Pressure, Oxygen Saturation, Nonnormal Consciousness). This rule does not intend to replace diagnosis-based alarm criteria but should be used as an adjunctive vital parameter-based basic default for emergency medical services (EMS) and triage nurses. The added value of this rule is that the initiation of any structured resuscitation protocol is based on specific and entirely objective thresholds-and not on subjective judgment or even gut feeling. However, additional studies are needed to determine whether the ViSiON rule improves patient safety, clinical outcome, and resource utilization in the ED.
In the absence of a true gold standard for the identification of CINT patients in the ED, we compared the performance of ViSiON rule with that of two established severity scoring systems. However, the NEWS was developed to identify patients at risk of early deterioration or