Abstract
Background:
Best resuscitation practices in the post-hemostasis phase of care are poorly defined; this phase of care is characterized by a range of physiologic derangements and multiple therapeutic modalities used to address them. Using a cohort of injured patients who required an immediate intervention in the operating room or angiography suite following arrival to the emergency department, we sought to define high-intensity resuscitation (HIR) in this post-hemostasis phase of care; we hypothesized that those who would require HIR could be identified, using only data available at ICU admission.
Methods:
Clinical data was extracted for consecutive injured patients (2016–19) admitted to the ICU following an immediate procedure in the operating room or angiography suite. HIR thresholds were defined as the top decile of blood product (≥3 units) and/or crystalloid (≥4 Liters) use in the initial twelve hours of ICU care and/or vasoactive medication use between ICU hours 2–12. The primary outcome, HIR, was a composite of any of these modalities. Predictive modeling of HIR was performed using logistic regression with predictor variables selected using Least Absolute Shrinkage and Selection Operator (LASSO) estimation. Model was trained using 70% of the cohort and tested on the remaining 30%; model predictive ability was evaluated using area under receiver operator curves.
Results:
Six-hundred-and-five patients were included. Patients were 79% male, young (median age: 39 years), severely injured (median ISS: 26), and an approximately 3:2 ratio of blunt to penetrating mechanisms of injury. A total of 215 (36%) required HIR. Predictors selected by LASSO included: shock index, lactate, base deficit, hematocrit, and INR. The area under receiver operator curve for the LASSO-derived HIR prediction model was 0.82.
Conclusions:
ICU admission data can identify subsequent HIR in the post-hemostasis phase of care. Use of this model may facilitate triage, nursing ratio determination, and resource allocation.
Keywords: Resuscitation, Resource Utilization, Triage
Introduction:
Prior to definitive hemostasis in injured patients, there is consensus supporting the administration of a balanced blood product resuscitation, and minimizing intravenous crystalloid infusion.(1–6) This practice uniformity has allowed for identification of risk factors for high blood product utilization, and the development of clinical decision support tools to predict patients that will require massive transfusion.(7–9) The ability or inability to anticipate resuscitation needs has implications on triage effectiveness, resource allocation, and preventing delay in time sensitive interventions (e.g., initiation of massive transfusion protocol or procedural hemostasis in the operating room or with interventional radiology).(10) Following hemostasis, many patients continue to have significant physiologic derangements and resuscitation needs; however, optimal resuscitation practices in this critical phase of care are poorly defined and our ability to predict which patients will require continued high intensity resuscitation (HIR) is subjective.(11, 12)
Challenges with predicting resuscitation needs in the post-hemostasis phase of an injured patient’s trajectory are driven by a relative lack of evidence to define best practice and the multiple therapeutic modalities by which post-hemostasis resuscitation can be performed (e.g., blood product transfusion, intravenous crystalloid infusion, and/or vasoactive medications).(11) The practical implications of this uncertainty include: reactive clinical decision making, potential delays in treatment due to underappreciation of ongoing physiologic derangements, and underinformed bedside staffing deployment (e.g. nursing ratios). To address these issues, improved understanding of current resuscitation practice and how resuscitation needs in critical care environments are associated with clinical parameters following hemostasis are needed.
To address these knowledge gaps, we evaluated a cohort of severely injured patients admitted to a trauma specific ICU following an immediate intervention in either the operating room or angiography suite after initial assessment in the Emergency Department and prior to TSICU admission. We sought to use this cohort to define HIR in the post-hemostasis phase of care and hypothesized that using only contemporaneous vital sign, laboratory, and procedure data available at the time of TSICU admission that we could predict which subjects would go on to require HIR.
Methods:
Study Subjects & Data Management
We combined the institutional trauma registry of a high volume, level-one trauma center with Electronic Health Record (EHR) data for injured patients admitted to the Trauma Surgery Intensive Care Unit (TSICU) between January 1, 2016, and December 31, 2019. Patients with burn injuries were excluded. We included patients who were over 18 years old and who were admitted to the TSICU after an immediate procedure either in the operating room or angiography suite following initial assessment in the emergency department. Immediate procedure was defined as a subject’s initial disposition from the emergency department being a procedural unit (operating room or angiography). The decision to limit our cohort to those who received an operation or interventional radiology procedure prior to TSICU admission was made to allow for the assumption that these patients are hemostatic, and therefore use their subsequent TSICU care to make inferences about the post-hemostasis phase of care. We excluded patients transferred from another institution, admitted to medical or a neurologic specific critical care unit, following drowning, or those who were missing TSICU crystalloid data. The use of patient data from the trauma registry and linked data from the EHR was performed after review and approval of our Institutional Review Board (STUDY00009850) and follows Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for observational studies.(13)
Data on subject demographics, injury characteristics, clinical outcomes, TSICU admission dates and times, hemodynamics, laboratory values, procedure data, and blood product, crystalloid, and vasoactive medication utilization data were collected. The methodology by which date and time stamp data for these parameters were linked and organized were similar to those published previously.(11) All crystalloid fluids (boluses, maintenance fluids, and rider fluids) were captured. For the purposes of our analyses, TSICU admission time after transfer from a procedural unit (operating room or angiography suite) served as the start of the post-hemostasis resuscitation phase of care. Blood product, crystalloid, and vasopressor utilization were recorded over the initial 12-hours of TSICU admission. This 12 -hour window was selected based on prior work in this cohort examining crystalloid administration as well as for its representing a common length of nursing and/or intensivist shifts.(14, 15) Continuous data are presented as medians with interquartile ranges and proportions as percentages. We performed comparisons between patients requiring HIR and those that did not using chi-square and Wilcoxon rank sum tests.
Resuscitation Metrics & Associated Clinical Outcomes
Subjects requiring HIR were defined as requiring any of the following in the initial 12 hours following TSICU admission 1) ≥ 4 liters of crystalloid (approximate top decile) or 2) ≥ 3 units of blood products (whole blood, packed red blood cells, plasma, or platelets; approximate top decile), or 3) vasoactive medication (phenylephrine, norepinephrine, vasopressin, epinephrine, or dobutamine) between hours 2–12 of TSICU admission. We excluded vasopressor use during the first 2 hours following TSICU admission as this often reflects the use of these drugs to counter the effects of anesthetic agents, rather than a significant resuscitation requirement.(16) For the purposes of prediction – the primary outcome of interest was HIR, a composite of one or more of these three modalities. Secondary outcomes for comparisons between HIR to non-HIR patients included inpatient mortality, time from ICU admission to death, ICU length of stay, duration of mechanical ventilation, and hospital length of stay.
Significant Resuscitation Prediction Modeling & Performance Evaluation
Candidate predictor variables were limited to data that is temporally relevant to TSICU admission physiology (vital signs, laboratory values, and data from the immediately preceding procedure data) and readily available to providers at TSICU admission. Initial TSICU admission vital signs (heart rate and systolic blood pressure) and laboratory values (complete blood count, basic metabolic panel, coagulation studies and blood gas) were extracted for all subjects. Patients for whom vital signs were not recorded within one hour following admission and/or laboratory values were not available within a window one hour before and two hours after TSICU admission were excluded from analysis. For subjects with multiple laboratory values in this window, the mean of available values was used.
Least absolute shrinkage and selection operator (LASSO) estimation was performed to determine which predictor variables would be included in the model using the ‘caret’ package in R Studio (Version 1.4.1717; Boston, MA).(17–19) Briefly, LASSO estimation selects predictor variables for inclusion in the model by limiting the absolute value of the sum of estimated logistic regression predictor variable coefficients to a value, lambda (λ), reducing the value of some predictor variable coefficients to zero resulting in their being removed from the model.(19) The optimal value of λ was selected using k-fold cross validation (k=5).(17–21) Candidate predictor variables included TSICU admission: shock index, lactate, base deficit, hematocrit, platelet count, creatinine, and INR as well as pre-TSICU admission procedure duration all of which were included in the LASSO estimation analysis. While other potentially relevant candidate predictors exist, these were selected because of their being real-time indicators of physiology which are both commonly available at the time of TSICU admission and being clearly discernable using retrospective methodology.
Logistic regression modeling using the LASSO-selected variables was used to predict HIR. For the logistic regression prediction model, an area under the receiver operating curve (AUROC) was created to quantify the sensitivity and specificity for predicting the composite HIR outcome. For the purposes of model training and testing, our cohort was divided randomly with 70% of the cohort used to train the predictive model and 30% to test it.(22) All statistics and graphic generation were performed using R Studio.
Results:
Study Subjects
Six-hundred-and-five (605) subjects were included in our analysis (Inclusion/Exclusion Diagram - Figure 1; Cohort Demographics and Variable Overview - Table 1). Overall, subjects were predominantly male (79%), young (median age 39 years old; [IQR: 28, 52]), severely injured (ISS: 26 [IQR: 17–38]) and contained an approximately 3:2 mix of blunt and penetrating mechanisms. One-hundred-and-thirty-three (22%) had severe head injuries (AIS Head ≥3). Inpatient mortality of the cohort was 11%.
Figure 1.
Cohort Inclusion & Exclusion Criteria
Table 1.
Cohort Demographics and Variable Overview
| Characteristic | N = 6051 |
|---|---|
| Age | 39 (28, 52) |
| Sex (Male) | 480 (79%) |
| Race | |
| Asian | 50 (8.3%) |
| Black | 110 (18%) |
| Native American | 16 (2.6%) |
| Unknown/Not Recorded | 22 (3.6%) |
| Pacific Islander | 10 (1.7%) |
| White | 397 (66%) |
| Disposition from Emergency Department | |
| Angiography Suite | 42 (6.9%) |
| Operating Room | 563 (93%) |
| Time to Procedure After Emergency Department Arrival (Hours) | |
| Angiography Suite | 2.63 (1.59, 4.2) |
| Operating Room | 1.17 (0.65, 2.78) |
| Trauma Mechanism | |
| Blunt | 351 (58%) |
| Penetrating | 250 (41%) |
| Other | 4 (0.7%) |
| Injury Severity Score | 26 (17, 38) |
| Head Abbreviated Injury Score | |
| 0 | 426 (70%) |
| 1 | 3 (0.5%) |
| 2 | 43 (7.1%) |
| 3 | 36 (6.0%) |
| 4 | 23 (3.8%) |
| 5 | 73 (12%) |
| 6 | 1 (0.2%) |
| ICU Admission Lactate (mmol/L) | 3.00 (1.96, 4.40) |
| ICU Admission Base Deficit (mmol/L) | −0.9 (−3.7, 0.8) |
| ICU Admission Platelet Count (thousand) | 154 (119, 197) |
| ICU Admission Hematocrit (%) | 32.7 (29.0, 36.0) |
| ICU Admission Creatinine (mg/dL) | 0.79 (0.66, 0.99) |
| ICU Admission INR | 1.30 (1.20, 1.40) |
| ICU Admission Shock Index2 | 0.77 (0.63, 0.94) |
| Procedure Duration (Hours) | 1.98 (1.32, 2.80) |
Median (IQR); n (%)
Time between ICU Admission Heart Rate and Systolic Blood Pressure documentation: Median 0 minutes, Mean 1.2 minutes, Range 0–39 minutes
Resuscitation Metrics & Associated Clinical Outcomes
The distributions of crystalloid, blood product, and vasoactive medication use in the study cohort are shown in Figure 2. In the initial 12 hours following TSICU admission, 88 (15%) received ≥ 4 L of intravenous crystalloid, 67 (11%) received ≥ 3 units of blood product, and 143 patients (24%) required vasoactive medications between TSICU hours 2–12. Two-hundred-and-fifteen (36%) required at least one of these three (primary outcome). A depiction of overlapping high-intensity resuscitation requirements among these patients can be found in Figure 3. Clinical parameters and outcome associations with high-intensity resuscitation are depicted in Table 2. Among subjects who died during their inpatient care, median time from hospital admission to death was 140 hours (IQR: 83–407) in the group that did not require HIR compared to 58 hours (IQR: 19–129) in the HIR subjects (p=0.002).
Figure 2.
Distributions of different resuscitative modalities in study cohort. A) Intravenous crystalloid distribution. Dashed red line at 4 liters represents threshold for high intensity resuscitation based on crystalloid use B) Blood product distribution. Dashed red line at 3 units represents threshold for high intensity resuscitation based on blood product use. Five subjects with > 20 units of total blood product use are excluded from this histogram C) Vasoactive medication distribution.
Figure 3.
Venn diagram depicting two-hundred and fifteen subjects that required at least one modality of high intensity resuscitation (Blood product ≥3 units or crystalloid ≥ 4 liters in the first 12 hours after TSICU admission or persistent vasopressor use between TSICU hours 2–12). Diagram created with ‘eulerr’ package in R.
Table 2.
Clinical Parameters & Outcome Associations with High Intensity Resuscitation
| Characteristic | Did Not Require HIR N = 3901 | Required HIR N = 2151 | p-value2 |
|---|---|---|---|
| Injury Severity Score | 22 (14, 34) | 34 (22, 50) | <0.001 |
| Hospital Length of Stay (Days) | 10 (6, 23) | 16 (7, 33) | 0.001 |
| ICU Length of Stay (Days) | 4 (3, 6) | 8 (4, 17) | <0.001 |
| Ventilator Days | 2 (1, 3) | 5 (2, 12) | <0.001 |
| In-Hospital Mortality | 19 (4.9%) | 49 (23%) | <0.001 |
Median (IQR); n (%)
Wilcoxon rank sum test; Pearson’s Chi-squared test
High-Intensity Resuscitation Prediction Modeling & Performance Evaluation
LASSO estimation selected five predictor variables from eight candidate predictors to be included in our logistic regression prediction models. These included TSICU admission: 1) shock index, 2) lactate, 3) INR, 4) base deficit, and 5) hematocrit. TSICU admission creatinine and platelet count along with pre-TSICU procedure length were not included in the model. The relative predictive importance of each variable included in LASSO analyses and the proportion of missing data for each candidate predictor variable can be found in Supplemental Figures 1 and 2, respectively. The logistic regression model utilizing these LASSO selected predictor variables (Supplemental Table 1) achieved an AUROC of 0.82 for the composite outcome (Figure 4). The sensitivity of the prediction model was 0.90, specificity 0.44, positive predictive value 0.71 and negative predictive value 0.75.
Figure 4.
Area under receiver operator curve graph for LASSO derived model’s ability to identify high intensity resuscitation patients. This model achieved an AUROC of 0.82.
Discussion:
This work is the describes the heterogeneous and multi-modality resuscitative needs of injured patients following hemostasis, defines high-intensity resuscitation in this phase of care, and predicts the need for high-intensity resuscitation using data available at the time of ICU admission.(12) Our analysis demonstrated a wide range in resuscitation requirements and overlapping use of three main therapeutic modalities – blood products, crystalloid, and vasopressors. As expected, those identified as requiring high-intensity resuscitation had significantly worse outcomes with respect to inpatient mortality, critical care length of stay, and total length of stay. More importantly, we developed and tested a novel prediction model using hemodynamic and laboratory data available at the time of ICU admission that accurately identifies patients who will subsequently require HIR. This model has comparable sensitivity and specificity to models used to predict massive transfusion in the pre-hemostasis phase of care.(7–9)
The ability to accurately identify which patients will require HIR could have significant utility in enhancing physician/nursing triage capabilities, resource allocation, and early prognostication. In the wake of the SARS-CoV-2 pandemic, institutional resources for intensive care unit staffing availability and patient to nursing ratios have been challenging.(23, 24) If a prediction model for HIR could be successfully incorporated into critical care practice, patient placement and nursing assignment ratios could be proactively adjusted. Additionally, patients requiring high-intensity resuscitation were more than four times more likely to die than those who did not require high-intensity resuscitation and died much more rapidly (Median time to death: 58 hours vs. 140 hours). While early establishment of code status and goals of care are important for all patients, patients identified early as being likely to require HIR would provide an early prompt to clinicians to have these conversations with patients and families early and engage palliative care services when appropriate. Finally, models like the one presented here are amenable to being integrated into electronic medical records where they can passively collect the data required to predict HIR needs and generate alerts for critical nurses and intensivists if a given patient is at high risk of requiring HIR thereby decreasing the cognitive load on these providers when triaging their multiple patient care responsibilities.
These results should be interpreted in the context of several important limitations. The retrospective nature of our study prevented the capture of potentially important variables (e.g., damage control procedures or transitions to comfort-directed care) and outcomes (e.g., return to IR/OR for hemostasis). It also forced the arbitrary creation of a period representative of ICU admission physiology for predictor variables that contains a small overlap with the resuscitation window rather than the ideal where all predictor data would have been captured at ICU admission time zero with resuscitative outcomes being measured after. While prospective validation of this model is needed, the novelty of this predictive model using only commonly available variables present on critical care admission needs to be considered. Second, shock index had the highest level of relative importance of any variable included in LASSO-facilitated creation of our prediction model (Supplemental Figure 1). While there is a body of literature describing the association of shock index with mortality and resuscitation requirements in injured patients, the use of shock index in the post-hemostasis/ICU phase of care has not been validated.(25–28) Third, the definition of “high-intensity” post-hemostasis resuscitation though novel, is arbitrary based on our high-volume, institutional practice. The thresholds for blood product and crystalloid use were selected as they approximate the top decile of each therapeutic modality in our study cohort, and while we expect that our practice is comparable to like centers, prospective multi-center validation of such a prediction model will likely require adjustment. Finally, missing hemodynamic and lab data was an issue in our cohort (0–22% of included predictor variables were missing); while this is attributable to our pre-defined acceptable time windows for predictor variables to be considered representative of ICU admission physiology rather than true missingness, it is nonetheless possible that this missing data introduces bias (e.g., patients with more severe physiologic derangements may have their initial vital signs recorded outside of the one hour post admission window because nurses responsible for charting vital signs are occupied providing care). Standardized timing of vital sign and laboratory studies coordinated by research personnel will be performed in the prospective validation study.
In conclusion, this work represents an advance in our understanding of the post-hemostasis phase of care. Simple laboratory and hemodynamic data available at the time of ICU admission are predictive of which patients will require HIR which has implications on patient outcomes, our prognostic capabilities, and resource utilization in critical care environments.
Supplementary Material
Supplemental Digital Content 1: STROBE Statement—checklist of items that should be included in reports of observational studies
Supplemental Digital Content 2: Author Conflict of Interest Forms
Supplemental Table 1. High-Intensity Resuscitation Logistic Regression Prediction Model
Supplemental Table 2. Unadjusted Associations Between Candidate Predictors and High-Intensity Resuscitation
Supplemental Table 3. Missing Predictor Data
Supplemental Figure 1. LASSO Variable Importance Plot
Financial support:
NIH T32GM121290
Footnotes
Disclosures: The above group of authors has no conflicts of interest to disclose. JTACS Disclosure form has been supplied and is provided as supplemental digital content.
Social Media Summary, Tags, and Author Handles:
This work describes the multimodality resuscitations used in the post hemostasis phase of care, defines “high-intensity resuscitation”, and utilizes predictive modeling to identify patients who will require high intensity resuscitation using data available at ICU admission.
@WeykampMike, @traumabryce, @gok_hmc, @brakenridge_md, @harborviewmc, @HMCTraumaT32
Level of Evidence: Level IV
References:
- 1.Holcomb JB, Tilley BC, Baraniuk S, Fox EE, Wade CE, Podbielski JM, et al. Transfusion of plasma, platelets, and red blood cells in a 1:1:1 vs a 1:1:2 ratio and mortality in patients with severe trauma: the PROPPR randomized clinical trial. Jama. 2015;313(5):471–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Cannon JW, Khan MA, Raja AS, Cohen MJ, Como JJ, Cotton BA, et al. Damage control resuscitation in patients with severe traumatic hemorrhage: A practice management guideline from the Eastern Association for the Surgery of Trauma. J Trauma Acute Care Surg. 2017;82(3):605–17. [DOI] [PubMed] [Google Scholar]
- 3.Kasotakis G, Sideris A, Yang Y, de Moya M, Alam H, Investigators IaHRtI. Aggressive early crystalloid resuscitation adversely affects outcomes in adult blunt trauma patients: an analysis of the Glue Grant database. J Trauma Acute Care Surg. 2013;74(5):1215–21; discussion 21–2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Robinson BRH, Cohen MJ, Holcomb JB, Pritts TA, Gomaa D, Fox EE, et al. Risk Factors for the Development of Acute Respiratory Distress Syndrome Following Hemorrhage. Shock. 2018;50(3):258–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Holcomb JB, del Junco DJ, Fox EE, Wade CE, Cohen MJ, Schreiber MA, et al. The prospective, observational, multicenter, major trauma transfusion (PROMMTT) study: comparative effectiveness of a time-varying treatment with competing risks. JAMA Surg. 2013;148(2):127–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Weykamp MB, Stern K, Brakenridge SC, Robinson BR, Wade CE, Fox EE, et al. PREHOSPITAL CRYSTALLOID RESUSCITATION: PRACTICE VARIATION AND ASSOCIATIONS WITH CLINICAL OUTCOMES. Shock. 2023;59(1):28–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Nunez TC, Voskresensky IV, Dossett LA, Shinall R, Dutton WD, Cotton BA. Early prediction of massive transfusion in trauma: simple as ABC (assessment of blood consumption)? J Trauma. 2009;66(2):346–52. [DOI] [PubMed] [Google Scholar]
- 8.McLaughlin DF, Niles SE, Salinas J, Perkins JG, Cox ED, Wade CE, et al. A predictive model for massive transfusion in combat casualty patients. J Trauma. 2008;64(2 Suppl):S57–63; discussion S. [DOI] [PubMed] [Google Scholar]
- 9.Yücel N, Lefering R, Maegele M, Vorweg M, Tjardes T, Ruchholtz S, et al. Trauma Associated Severe Hemorrhage (TASH)-Score: probability of mass transfusion as surrogate for life threatening hemorrhage after multiple trauma. J Trauma. 2006;60(6):1228–36; discussion 36–7. [DOI] [PubMed] [Google Scholar]
- 10.Meyer DE, Vincent LE, Fox EE, OʼKeeffe T, Inaba K, Bulger E, et al. Every minute counts: Time to delivery of initial massive transfusion cooler and its impact on mortality. J Trauma Acute Care Surg. 2017;83(1):19–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Beni CE, Arbabi S, Robinson BRH, O’Keefe GE. Acute intensive care unit resuscitation of severely injured trauma patients: Do we need a new strategy? J Trauma Acute Care Surg. 2021;91(6):1010–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Braasch MC, Turco LM, Cole EM, Brohi K, Winfield RD. The evolution of initial-hemostatic resuscitation and the void of posthemostatic resuscitation. J Trauma Acute Care Surg. 2020;89(3):597–601. [DOI] [PubMed] [Google Scholar]
- 13.von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol. 2008;61(4):344–9. [DOI] [PubMed] [Google Scholar]
- 14.Stimpfel AW, Aiken LH. Hospital staff nurses’ shift length associated with safety and quality of care. J Nurs Care Qual. 2013;28(2):122–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Webster J, McLeod K, O’Sullivan J, Bird L. Eight-hour versus 12-h shifts in an ICU: Comparison of nursing responses and patient outcomes. Australian Critical Care. 2019;32(5):391–6. [DOI] [PubMed] [Google Scholar]
- 16.Wu M, Ghassemi M, Feng M, Celi LA, Szolovits P, Doshi-Velez F. Understanding vasopressor intervention and weaning: risk prediction in a public heterogeneous clinical time series database. J Am Med Inform Assoc. 2017;24(3):488–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Ranstam J, Cook JA. LASSO regression. British Journal of Surgery. 2018;105(10):1348-. [Google Scholar]
- 18.Max K. Building Predictive Models in R Using the caret Package. Journal of Statistical Software. 2008;28. [Google Scholar]
- 19.Yokoyama T, Nakahara S, Kondo H, Miyake Y, Sakamoto T. Novel score for predicting early emergency endovascular therapy in trauma care using logistic LASSO regression. Acute Med Surg. 2022;9(1):e774. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Caughlin D. R for HR: An Introduction to Human Resource Analytics Using R 2023. Available from: https://rforhr.com/. Accessed 08-01-2023.
- 21.S J. Machine Learning 2023. Available from: https://bookdown.org/ssjackson300/Machine-Learning-Lecture-Notes/. Accessed 08-01-2023.
- 22.Gholamy A, Kreinovich V, Kosheleva O. Why 70/30 or 80/20 relation between training and testing sets: A pedagogical explanation. 2018. [Google Scholar]
- 23.Andel SA, Tedone AM, Shen W, Arvan ML. Safety implications of different forms of understaffing among nurses during the COVID-19 pandemic. J Adv Nurs. 2022;78(1):121–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Lasater KB, Aiken LH, Sloane DM, French R, Martin B, Reneau K, et al. Chronic hospital nurse understaffing meets COVID-19: an observational study. BMJ Qual Saf. 2021;30(8):639–47. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Mutschler M, Nienaber U, Münzberg M, Wölfl C, Schoechl H, Paffrath T, et al. The Shock Index revisited - a fast guide to transfusion requirement? A retrospective analysis on 21,853 patients derived from the TraumaRegister DGU. Crit Care. 2013;17(4):R172. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Biney I, Shepherd A, Thomas J, Mehari A. Shock Index and Outcomes in Patients Admitted to the ICU With Sepsis. CHEST. 2015;148(4):337A. [Google Scholar]
- 27.Cannon CM, Braxton CC, Kling-Smith M, Mahnken JD, Carlton E, Moncure M. Utility of the shock index in predicting mortality in traumatically injured patients. J Trauma. 2009;67(6):1426–30. [DOI] [PubMed] [Google Scholar]
- 28.Vandromme MJ, Griffin RL, Kerby JD, McGwin G, Jr., Rue LW, 3rd, Weinberg JA. Identifying risk for massive transfusion in the relatively normotensive patient: utility of the prehospital shock index. J Trauma. 2011;70(2):384–8; discussion 8–90. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental Digital Content 1: STROBE Statement—checklist of items that should be included in reports of observational studies
Supplemental Digital Content 2: Author Conflict of Interest Forms
Supplemental Table 1. High-Intensity Resuscitation Logistic Regression Prediction Model
Supplemental Table 2. Unadjusted Associations Between Candidate Predictors and High-Intensity Resuscitation
Supplemental Table 3. Missing Predictor Data
Supplemental Figure 1. LASSO Variable Importance Plot