Novel Score for Stratifying Risk of Critical Care Needs in Patients With Intracerebral Hemorrhage

Abstract

Objective

To develop a risk prediction score identifying patients with intracerebral hemorrhage (ICH) at low risk for critical care.

Methods

We retrospectively analyzed data of 451 patients with ICH between 2010 and 2018. The sample was randomly divided into a development and a validation cohort. Logistic regression was used to develop a risk score by weighting independent predictors of intensive care unit (ICU) needs according to strength of association. The risk score was tested in the validation cohort and externally validated in a dataset from another institution.

Results

The rate of ICU interventions was 80.3%. Systolic blood pressure (SBP), Glasgow Coma Scale (GCS) score, intraventricular hemorrhage (IVH), and ICH volume were independent predictors of critical care, resulting in the following point assignments for the Intensive Care Triaging in Spontaneous Intracerebral Hemorrhage (INTRINSIC) score: SBP 160 to 190 mm Hg (1 point), SBP >190 mm Hg (3 points); GCS 8 to 13 (1 point), GCS <8 (3 points); ICH volume 16 to 40 cm³ (1 point), ICH volume >40 cm³ (2 points); and presence of IVH (1 point), with values ranging between 0 and 9. Among patients with a score of 0 and no ICU needs during their emergency department stay, 93.6% remained without critical care needs. In an external validation cohort of patients with ICH, the INTRINSIC score achieved an area under the receiver operating characteristic curve of 0.823 (95% confidence interval 0.782–0.863). A score <2 predicted the absence of critical care needs with 48.5% sensitivity and 88.5% specificity, and a score <3 predicted the absence of critical care needs with 61.7% sensitivity and 83.0% specificity.

Conclusion

The INTRINSIC score identifies patients with ICH who are at low risk for critical care interventions.

Classification of Evidence

This study provides Class II evidence that the INTRINSIC score identifies patients with ICH at low risk for critical care interventions.

Spontaneous intracerebral hemorrhage (ICH) is the most devastating form of stroke, carrying high risk for early deterioration and short-term mortality.¹ In most institutions, patients with ICH are routinely admitted to an intensive care unit (ICU) to timely detect and adequately address potential complications such as elevation in intracranial pressure and blood pressure, airway compromise, and other complicating medical issues indiscriminate of patient demographics and clinical characteristics.^2,3 While most patients with ICH undergo interventions that require critical care and an ICU or comparable environment, often with access to neurosurgical procedures, it is currently unclear whether ICU admission and care are medically necessary for all patients with ICH. Current ICH guidelines acknowledge that admission to a stroke unit as opposed to an ICU is generally feasible for some patients with ICH.³ Admission of patients with ICH to stroke units in clinical practice is uncommon, and uniform criteria or triaging tools for accurate identification of low-risk patients who can be safely monitored in a non-ICU setting are lacking.^4,5 Similarly, there are no agreed-on triaging tools that identify patients with ICH who may forgo transfer to a tertiary care center.³

The COVID-19 pandemic has imposed unprecedented challenges on stroke care and resource use. The pandemic has already resulted in reallocation of critical care resources typically reserved for patients with stroke and ICH to accommodate the flood of critically ill patients at some hospitals in the United States and around the world⁶; this is likely to continue or to reoccur because reopening efforts are expected to result in additional waves of critically ill patients with COVID-19. As the pandemic continues, tools guiding interhospital transfers and aiding the appropriate allocation of critical care resources in patients with stroke and ICH have gained importance and urgency.^6,7 In addition to alleviating resource constraints during the current pandemic, avoiding ICU admission for patients who may not need critical care may improve health outcomes by reducing the risk of health care–associated infections and delirium more commonly encountered in a critical care environment.^8-11

Previous studies have demonstrated that routine admission to an ICU may be unnecessary for subpopulations of patients with certain neurologic conditions and neurosurgical procedures such as traumatic ICH¹² and traumatic brain injury,¹³ after brain biopsy,¹⁴ after thrombolysis for ischemic stroke,^15,16 and after carotid endarterectomy.¹⁷ There are no established parameters that allow stratification of risk of critical care needs in patients with ICH, and there is no universally agreed-on algorithm that identifies patients with ICH who may forgo ICU admission. In the present study, we aimed to develop a clinical risk score for the prediction of critical care needs in patients with primary ICH and to identify a subpopulation of patients with ICH in whom ICU care is unnecessary.

Methods

Standard Protocol Approvals, Registrations, and Patient Consents

This study was approved by the Johns Hopkins University School of Medicine Institutional Review Board. Written informed consent was waived by our Institutional Review Board.

Classification of Evidence

The primary research question was to determine whether a clinical risk prediction score could identify patients with ICH at low risk for requiring critical care. This study provides Class II evidence that the Intensive Care Triaging in Spontaneous Intracerebral Hemorrhage (INTRINSIC) score identifies patients with ICH at low risk for critical care interventions.

Study Design and Setting

To develop and validate a risk prediction model for critical care needs after ICH, we retrospectively analyzed medical records of patients with ICH in the Johns Hopkins prospective stroke database. Patients presenting with primary ICH to Johns Hopkins Hospital and Johns Hopkins Bayview Medical Center between January 2010 and December 2018 were included (Johns Hopkins dataset); some of the 2014 and 2015 Bayview Medical Center records were excluded due to the inability to abstract key variables of interest during transitioning of the digital records system. Patients with in-hospital ICH or interhospital transfers were excluded. Patients who underwent care withdrawal in the emergency department (ED) were also excluded. Patients with known intracerebral metastatic disease or known arteriovenous malformation or cavernoma in the location of the hemorrhage also were excluded. For external validation of the prediction model, we used an independent dataset of patients with primary ICH admitted to the neurocritical care unit at University Hospitals Cleveland Medical Center between January 2013 and December 2015 (Cleveland dataset), as described previously.⁴ The Cleveland dataset used for external validation included interhospital transfers.

Clinical Data Collection

Data abstraction for the Johns Hopkins dataset was performed by a single reviewer for each chart, a board-certified vascular neurologist (R.F.), and a research assistant (B.J.C.). A training period for the research assistant preceded the data collection period, and initial charts were abstracted by both reviewers separately and then iteratively compared for consistency. Inconsistencies in the chart documentation were adjudicated by the board-certified vascular neurologist. A similar process was followed for the Cleveland dataset, from which data were abstracted by a neurocritical care fellow and a senior neurology resident. For both datasets, all available medical records were reviewed, including ED records, admission notes, consult notes, discharge summaries, daily progress notes, nursing notes/documentation and flowsheets, surgical/procedure notes, all vital signs and laboratory values, and medication administration records.

In the Johns Hopkins dataset, demographic data, including age, sex, and race, were collected for all patients. The presence of comorbid conditions, including hypertension, hyperlipidemia, diabetes, chronic kidney disease, atrial fibrillation, prior ICH, smoking status, and the prehospital use of antiplatelet agents, anticoagulation, and statins, was abstracted from the medical record. The following physiologic parameters at initial presentation were recorded: systolic blood pressure (SBP), diastolic blood pressure (DBP), Glasgow Coma Scale (GCS) score, international normalized ratio, serum glucose, and serum creatinine. In addition, length of ICU stay, overall length of hospital stay, and discharge disposition were extracted. Variables relevant for external score validation were obtained from the Cleveland dataset.

Primary Outcome

The primary outcome was the need for a critical care intervention at any time point during the hospitalization. We operationalized the outcome definition by abstracting the presence of a critical care intervention documented from the medical chart, as described previously.^16,18,19 Specifically, critical care interventions included uncontrolled hypertension requiring titration of IV antihypertensives (either as continuous infusions or as IV bolus administrations exceeding the frequency permitted in a non-ICU setting), vasopressor use for symptomatic hypotension, symptomatic bradycardia or tachycardia requiring titration of rate-augmenting or rate-controlling IV medications or procedures (such as external cardiac pacing or cardioversion), need for invasive hemodynamic monitoring, uncontrolled hyperglycemia requiring continuous IV insulin infusions, respiratory compromise requiring mechanical ventilation, IV fluid management exceeding the capability of the ward, management of elevated intracranial pressure and cerebral edema, and neurosurgical interventions such as placement of an external ventricular drain (EVD) or decompressive craniectomy.¹⁶ Our definition of an ICU intervention also included any event or complication that required monitoring in an ICU setting even if no immediate critical care intervention was performed such as a progressive decrease in mental status with impaired airway protection, increasing oxygen requirement, detection of potentially life-threatening arrhythmia, or high-risk anticoagulation within a few days after ICH (e.g., because of the presence of a mechanical heart valve). To meet the outcome for these specified indications for monitoring only, the medical records had to explicitly indicate that the patient was in the ICU because of the specific need for monitoring.

Neuroimaging Analysis

ICH location on admission CT scan was categorized as deep, lobar, cerebellar, or brainstem. ICH volume was calculated by the ABC/2 method in both datasets.²⁰ In the Johns Hopkins dataset, all images were reviewed by 1 of 2 investigators (R.F. or B.J.C.). Similarly, in the Cleveland dataset, all images were reviewed by either a neurocritical care fellow or a senior neurology resident. Images for 10% of patients in the Johns Hopkins data were reviewed by both reviewers, and an intraclass correlation coefficient for a 2-way mixed-effects model was determined to assess interrater agreement of ICH volume (intraclass correlation 0.97, 95% confidence interval [CI] 0.95–0.99), as described previously²¹; a formal interrater analysis was not conducted in the Cleveland dataset. The presence of intraventricular hemorrhage (IVH) was recorded. Hematoma expansion was defined as a proportional increase of >33% or an absolute increase >6 cm³ (if baseline ICH volume ≤15 cm³) from the initial ICH volume.²²

Statistical Analysis

Statistical analysis was performed with Stata version 15 (StataCorp, College Station, TX). Continuous variables were analyzed with the Wilcoxon rank-sum test, and categorical variables were analyzed with the Pearson χ² test.

The prediction model was developed by using a random sample of 50% of the Johns Hopkins dataset (development cohort) and subsequently tested on the remaining 50% (validation cohort). In addition, the score was externally validated in the Cleveland dataset.

For score development, simple logistic regression in the development cohort of the Hopkin dataset included demographic, clinical, physiologic, and imaging variables available on initial presentation. Medical comorbid conditions were not included in model building because the full extent of comorbid conditions is often not known to the clinician until after triaging decisions have been made. A multivariable logistic regression model of predictors of critical care needs was developed with the use of demographic variables, ICH score components, and predictors significantly associated with the need for ICU care in the simple logistic regression analyses at a threshold of p < 0.1. Independent predictors with p < 0.1 in multivariable analysis were included as score variables in the final score after variables with strong collinearity were omitted. Continuous variables associated with the need for ICU care were transformed into categorical variables for score model generation, and the Akaike information criterion was used for selection of the most parsimonious model. For final risk score generation, we assigned points to each score variable category proportional to its regression coefficients rounded to the nearest integer.

Score model discrimination was assessed by using the area under the receiver operating characteristic curve (AUC). Calibration was assessed with the Hosmer-Lemeshow test to determine goodness of fit. Sensitivity, specificity, positive predictive value, and negative predictive value for various score cut points are presented.

Data Availability

Anonymized data can be made available to qualified investigators on reasonable request to the corresponding author.

Results

Patient Characteristics and Critical Care Interventions

A total of 451 patients with primary ICH in the Johns Hopkins dataset were included. The median age was 62 years (interquartile range 54–77 years); 54.1% were male; and 51.0% were Black (table 1). Three hundred sixty-two patients (80.3%) underwent critical care interventions, while 19.7% of patients remained free of critical care needs for the entirety of their hospitalization. Table 1 shows the baseline characteristics of patients with and without critical care needs.

Table 1.

Baseline Characteristics of Patients With ICH Stratified by Need for ICU Care

A total of 323 (71.6%) patients developed critical care needs during their ED stay, while another 39 (8.6%) patients newly developed critical care needs after their ED stay (figure 1). The most common ICU interventions were IV medication infusions for uncontrolled hypertension (67.0%; table 2), mechanical ventilation (47.5%), hyperosmolar therapy for management of cerebral edema (47.5%), and EVD placement (22.8%). Table 2 shows a complete list of critical care interventions. Most patients underwent more than a single critical care intervention (53.4%), while 23.3% of patients had need for blood pressure control as their sole critical care need.

Figure 1. Timing of Critical Care Needs After ICH.

Flowchart illustrates the need for critical care interventions as a function of time. ED = emergency department; ICH = intracerebral hemorrhage; ICU = intensive care unit.

Table 2.

Types of Intervention Among Patients With ICH With Critical Care Needs

Development of a Critical Care Risk Prediction Model

Simple logistic regression in the development cohort of the Johns Hopkins dataset identified the following demographic, clinical, physiologic, and imaging characteristics associated with critical care needs: age (odds ratio [OR] 0.74 per 10-year increase; p = 0.007), Black race (OR 2.4; p = 0.013), GCS score (OR 1.54 per 1-point decrease in GCS score; p < 0.001), SBP (OR 1.31 per 10-point increase; p < 0.001), DBP (OR 1.48 per 10-point increase; p < 0.001), serum potassium (OR 0.52 per 1-mmol/L increase; p = 0.007), ICH volume (OR 1.80 per 10 cm³ increase; p < 0.001), IVH (OR 6.87; p < 0.001), and hydrocephalus (OR 3.71; p = 0.009). Multivariable logistic modeling, including demographic variables, components of the ICH score, and predictors significantly associated with need for ICU care in univariate analysis, identified GCS score, SBP, ICH volume, and IVH as independent predictors of critical care needs in the development cohort with p < 0.1 (table 3). DBP and hydrocephalus were not included in the model because of collinearity with SBP and IVH, respectively. For score development, the best model included GCS score with cut points of 8 and 14, SBP with cut points of 160 and 190 mm Hg, ICH volume with cut points of 15 and 40 cm³, and presence of IVH. Points for the individual components of the 4-item INTRINSIC score were assigned on the basis of the regression coefficients of the final score model in the development cohort (table 3). The risk score ranges from 0 to 9, with higher scores indicating higher risk for needing critical care (figure 2A).

Table 3.

Multivariable Logistic Regression Analyses: Predictors of Critical Care Interventions in the Development Cohort

Figure 2. INTRINSIC Score Components and Performance in the External Validation Cohort.

(A) Determination of the Intensive Care Triaging in Spontaneous Intracerebral Hemorrhage (INTRINSIC) score for predicting critical care needs in intracerebral hemorrhage (ICH). (B) Receiver operating characteristic curve for the score model predicting presence of critical care needs in the Cleveland external validation cohort. Area under the curve (AUC) of 0. 823 shows that the model has good discrimination ability. GCS = Glasgow Coma Scale; IVH = intraventricular hemorrhage; SBP = systolic blood pressure.

Model Validation and Performance

In the validation cohort of the Johns Hopkins dataset, the final score model achieved an AUC of 0.880 (95% CI 0.833–0.928), and model calibration was good (Hosmer-Lemeshow p = 0.38). Each 1-point increase in the INTRINSIC score was associated with a 3-fold increase in odds for a critical care intervention (OR 3.00, 95% CI 2.13–4.22). The odds of ICU needs in patients with a score ≥1 were >37 times higher than in patients with a score of 0 (OR 37.36, 95% CI 10.26–136.04). Table 4 shows score performance measures for predicting the absence of critical care needs in the Johns Hopkins internal validation cohort at various cut points. Among patients who remained without any ICU needs, 38.6% had a score of 0, and a score of 0 predicted the absence of critical care needs with 98.3% specificity; 85.0% of patients with a score of 0 remained free of critical care needs (table 4). A score of <2 vs ≥2 predicted the absence of critical care needs with 56.8% sensitivity and 88.4% specificity, and a score of <3 vs ≥3 predicted the absence of critical care needs with 88.0% sensitivity and 76.8% specificity (table 4).

Table 4.

Sensitivity, Specificity, Positive Predictive Value, and Negative Predictive Value for Absence of Critical Care Needs at Various INTRINSIC Score Cut Points in the Johns Hopkins Internal and Cleveland External Validation Cohorts

For external validation, we tested the performance of the INTRINSIC score in the Cleveland dataset, an independent dataset from a different institution. In this external dataset of 385 patients with ICH with complete information on score variables, INTRINSIC predicted the need for critical care with an AUC of 0.823 (95% CI 0.782–0.863) (figure 2B). Each 1-point increase in the score was associated with an almost 2-fold increase in odds for a critical care intervention (OR 1.96, 95% CI 1.70–2.26). The odds of ICU needs in patients with a score ≥1 were >6 times higher than in patients with a score of 0 (OR 6.22, 95% CI 2.49–15.55). A score of <2 vs ≥2 predicted the absence of critical care needs with 48.5% sensitivity and 88.5% specificity, and a score of <3 vs ≥3 predicted the absence of critical care needs with 61.7% sensitivity and 83.0% specificity (table 4). Among the 31 patients with a score of 0, 1 patient (3.2%) required critical care interventions that may be considered particularly time critical such as intubation or placement of an EVD. Among the 106 patients with a score <2, 6 patients (5.7%) required intubation or EVD placement, while 122 of 279 (43.7%) patients with a score ≥2 required either of these interventions.

Score Performance When Blood Pressure Control Is Not Considered Critical Care

Because some stroke units and intermediate care units are capable of administering continuous infusions of IV antihypertensives, we performed a sensitivity analysis in the Johns Hopkins dataset applying the INTRINSIC score to predict ICU needs and the absence thereof after redefining the outcome by classifying patients as not needing ICU care if their only critical care intervention was blood pressure control. Even under this paradigm, each 1-point increase in the INTRINSIC score was associated with a 1.92-fold increase in the odds of ICU needs (95% CI 1.59–2.33), and the score achieved an AUC of 0.794 (95% CI 0.738–0.850) in the validation cohort. Among the 131 (29.0%) patients with a score of 0 or who scored points only for blood pressure control, 86.3% either remained without any critical care intervention or had need for blood pressure control as their only critical care intervention.

A Critical Time Window for Declaring Critical Care Needs

To define the timing of onset for critical care needs, we recorded whether ICU needs were present during the ED stay, developed thereafter, or both. Almost 90% of patients requiring critical care developed their critical care needs while still in the ED (323 of 362, 89.2%; figure 1). Conversely, among all patients who were free of critical care needs by the end of their ED stay, 69.5% remained free of critical care needs for the remainder of their hospitalization (figure 1 and table 2). Among the 103 (22.8%) patients with blood pressure control as the only critical care need, 96 (93.2%) declared those needs during the ED stay.

When the INTRINSIC score was applied to patients in the Johns Hopkins validation cohort without critical care needs by the end of their ED stay, a score of 0 vs ≥1 was associated with >14-fold increased odds of remaining without critical care (OR 14.48, 95% CI 1.79–117.32) and had 95.8% specificity in identifying patients who remained without ICU needs for the remainder of their hospitalization (i.e., few patients were falsely classified as not needing critical care). Among patients with a score of 0 and no ICU needs during their ED stay, 94.4% remained without critical care needs (figure 3). Similarly, among patients with a score of <2 and no ICU needs during the ED stay, 83.3% remained without critical care needs. Among patients with a score <2 who had no critical care needs by the end of their ED stay, 2 patients required intubation and/or EVD placement (figure 3). Both patients required intubation for airway protection due to worsening mental status in the setting of cerebral edema; however, the decline in mental status occurred subacutely over the course of several hours, in keeping with the expected time trajectory of cerebral edema progression.

Figure 3. Timing and Nature of Critical Care Needs in Patients With ICH in the Johns Hopkins Validation Cohort, Stratified by INTRINSIC Score Values.

BP = blood pressure; ED = emergency department; EVD = external ventricular drain; HT = hyperosmolar therapy; ICH = intracranial hemorrhage; ICU = intensive care unit; INTRINSIC = Intensive Care Triaging in Spontaneous Intracerebral Hemorrhage.

Discussion

In this study, we developed and validated a simple risk prediction score for critical care needs in patients with primary ICH. The INTRINSIC score consists of SBP, GCS score, IVH, and ICH volume on initial presentation, reflecting the driving clinical predictors of need for critical care interventions in patients with ICH. The components of the score are easy to obtain and readily available at the time of presentation. SBP and GCS score were the strongest predictors of critical care need, consistent with the notion that the most common reasons for critical care included the need for blood pressure control and initiation of mechanical ventilation for airway protection due to decreased level of consciousness. GCS score has been a critical component of previous ICH models of mortality and overall length of ICU stay.^23-25 While SBP and GCS score can be immediately obtained at the bedside or from the chart, the presence of IVH and ICH volume (by the ABC/2 method) can be easily determined from a single head CT without the need for specific radiologic expertise. The robustness of the INTRINSIC score is highlighted by its excellent performance in internal and external validation in an independent dataset.

In addition, INTRINSIC performed well in a paradigm in which blood pressure control is not considered a critical care intervention, and the predictive accuracy of the score is increased when information about the ED course is incorporated. Prior studies have reported the feasibility of admitting subpopulations of patients with ICH to a step-down rather than an ICU environment according to defined criteria.^4,5 However, admission criteria vary between institutions and are based largely on expert opinion. In addition, a predefined empirical set of step-down admission criteria typically requires that all criteria are fulfilled. A scoring system has the advantage of accounting for tradeoffs between various variables; for example, a score of 1 or 2 can be achieved by various combinations of values of the respective score variables, increasing the flexibility of its use. Because each score value is accompanied by a specific sensitivity-specificity profile, the clinician is equipped with a more nuanced risk-benefit analysis of a potential triaging decision.

In clinical practice, the INTRINSIC score may be most useful when predicting the absence of critical care needs to identify which patients may be triaged to a non-ICU setting. Each score cut point has a different sensitivity-specificity tradeoff. To prioritize patient safety, it is desirable to avoid falsely classifying patients as not needing critical care when in fact they do. In addition, high specificity may be particularly desirable when contemplating potential patient transfer (or no transfer) to tertiary centers. We therefore propose a cut point that predicts the absence of critical care needs with high specificity (low false positives) such as a score of <2, which predicted the absence of critical care with 88.5% specificity in the external validation cohort. With increasing resource constraints such as at the height of the COVID-19 pandemic when open ICU beds are a rarity, a higher score cut point such as <3 could be considered. While this cut point would allow triaging a higher number of patients to a non-ICU setting, it comes at the expense of a slightly lower specificity of 83.0%. Remaining free of critical care needs during the ED stay increases the predictive accuracy of the score further, and the rates of falsely classifying a patient with ICH as not needing critical care with a score of 0 or 1 are comparable to the lower end of the spectrum of reported unplanned ICU readmission (bounce back) rates.^26-29

The proportion of patients requiring ICU care in our cohort was ≈80% and is substantially higher than in the general ischemic stroke population and stroke patients undergoing IV thrombolysis.¹⁸ However, accurate identification of the remaining 20% of patients with ICH who never require critical care resources is significant considering the annual incidence of ICH in the United States of 40,000 to 60,000 per year,³⁰ with substantial potential for preserving ICU resources, personnel, and costs. Most hospitals routinely transfer patients with acute ICH to tertiary centers with neuro-ICUs, and stroke/ICH is one of the most common reasons for interhospital transfers from rural hospitals to academic medical centers.³¹ Criteria on whether and when to transfer a patient to a higher level of care are often subjective and ill-defined and are often premised on the estimated risk of requiring a neurosurgical intervention; however, it remains unclear which factors and clinical features necessitate intensive care and transfer to a tertiary center. While critical for most patients, transfer or escalation of care for some patients may be unnecessary, delay treatment, and result in additional costs. The INTRINSIC score was not developed to predict the need for interhospital transfer, and the cohort in which the score was developed excluded interhospital transfers; however, the score performed well in external validation in a cohort that included interhospital transfers, and a score of 0 or 1 may be useful in identifying patients who may not require patient transfer.

Our study has several limitations. By virtue of limiting the number of score variables, any clinical risk score entails simplification at the expense of accuracy of outcome prediction. Decision-making about the appropriate monitoring environment is complex and multifactorial and may depend on other variables and factors not assessed in the present study, including medical comorbid conditions, additional imaging data (such as presence of a spot sign), or additional information on medication history, including the use of anticoagulants. As with the use of a risk prediction score in medicine in general, the information derived from our score should be integrated with any other clinically relevant information and is intended to serve as a tool to aid the clinical decision-making process. This is a retrospective analysis of patients from 2 single stroke centers, and although we validated our score in an external independent dataset, generalizability to other ICH populations must be cautioned. Last, our results are not generalizable to patients with traumatic ICH or ICH in the setting of vascular malformations. Despite these limitations and the need for prospective validation, we believe that our risk prediction score is a valuable tool that will aid clinicians with the triaging of patients with ICH by reliably predicting the risk of critical care interventions. Similarly, our score identifies patients with ICH at low risk for critical care interventions, and patients with a score <2 may be considered for management in a stroke unit without critical care capabilities.

Glossary

AUC

area under the receiver operating characteristics curve

CI

confidence interval

COVID-19

coronavirus disease 2019

DBP

diastolic blood pressure

ED

emergency department

EVD

external ventricular drain

GCS

Glasgow Coma Scale

ICH

intracerebral hemorrhage

ICU

intensive care unit

INTRINSIC

Intensive Care Triaging in Spontaneous Intracerebral Hemorrhage

IVH

intraventricular hemorrhage

OR

odds ratio

SBP

systolic blood pressure

Appendix. Authors

Study Funding

R.F. is supported by a career development grant from the National Institute of Neurologic Disorders and Stroke (K23NS101124) and The Morningstar Foundation. A.A. is supported by the National Center for Advancing Translational Sciences of the NIH under the Miami CTSI KL2 Career Development Award (UL1TR002736). R.F.G. is supported by a grant from the National Institute on Aging (K24AG052573).

Disclosure

R. Faigle, B.J. Chen, R. Krieger, E.B. Marsh, A, Alkhachroum, and W. Xiong report no disclosures relevant to this manuscript. V.C. Urrutia is supported by grants from Genentech (ML42239) and is site principal investigator for Thrombolysis in Imaging-Eligible, Late-Window Patients to Assess the Efficacy and Safety of Tenecteplase (TIMELESS; a multicenter randomized clinical trial funded by Genentech) outside the submitted work. R.F. Gottesman reports no disclosures relevant to this manuscript. Partial content of this manuscript was presented as an abstract at the 145th Annual Meeting of the American Neurological Association, 2020. Go to Neurology.org/N for full disclosures.