Shock Index Predicts Patient-Related Clinical Outcomes in Stroke.

Background The prognostic value of shock index ( SI ), heart rate divided by systolic blood pressure, in stroke for clinical outcomes other than mortality is not well understood. Methods and Results We examined the Get With The Guidelines-Stroke ( GWTG -Stroke) data to explore the usefulness of SI in predicting in-hospital outcomes in 425 808 acute stroke cases (mean age: 71.0±14.5 years; 48.8% male; 89.7% ischemic stroke and 10.3% intracerebral hemorrhage) admitted between October 2012 and March 2015. Compared with patients with SI of 0.5 to 0.7, patients with SI >0.7 (13.6% of the sample) had worse outcomes, with adjusted odds ratios of 2.00 (95% confidence interval [ CI ], 1.92-2.08) for in-hospital mortality, 1.46 (95% CI , 1.43-1.49) for longer length of hospital stay >4 days, 1.50 (95% CI , 1.47-1.54) for discharge destination other than home, 1.41 (95% CI , 1.38-1.45) for inability to ambulate independently at discharge, and 1.52 (95% CI , 1.47-1.57) for modified Rankin Scale score of 3 to 6 at discharge. Results were similar when analyses were confined to those with available National Institutes of Health Stroke Scale (NIHSS) or within individual stroke subtypes or when SI was additionally included in the models with or without blood pressure components. Every 0.1 increase in SI >0.5 was associated with significantly worse outcomes in linear spline models. The addition of SI to existing GWTG -Stroke mortality prediction models without NIHSS demonstrated modest improvement, but little to no improvement was noted in models with NIHSS . Conclusions SI calculated at the point of care may be a useful prognostic indicator to identify those with high risk of poor outcomes in acute stroke, especially in hospitals with limited experience with NIHSS assessment.

Clinical Perspective

What Is New?

Shock index (heart rate divided by systolic blood pressure) at presentation to the emergency department predicts patient‐related clinical outcomes in ischemic and hemorrhagic stroke.

What Are the Clinical Implications?

Shock index may be useful in providing prognostic information to patients and relatives.

Shock index may also serve as an easily available bedside assessment tool to identify high‐risk individuals and to guide clinicians in providing appropriate management for these patients.

Shock index may be particularly useful in low‐resource settings or in hospitals with limited experience with stroke‐specific assessments such as the National Institutes of Health Stroke Scale.

The shock index (SI), expressed as a ratio of heart rate to systolic blood pressure (BP), was used initially to detect hypovolemia in patients with septic or hypovolemic shock, indicated by a SI value of >0.7.1, 2 In addition, high SI has been reported to have prognostic significance in trauma3 and acute cardiovascular events such as acute myocardial infarction and stroke.4, 5, 6 Although heart rate and BP are important variables in predicting acute stroke outcome, physiologically they are linked to each other. Thus SI, which considers the ratio of these variables, could perform better in predicting outcomes. The SI may also reflect the development of Cushing reflex, which is seen with ischemic brain injury caused by any stroke subtypes. In addition to identifying patients at higher risk of mortality, SI may also predict functional outcomes depicted by discharge modified Rankin Scale and ambulatory status in acute stroke.

McCall et al6 were the first to report the relationship between this simple and readily available physiological index and acute stroke mortality. In their UK‐based multicenter study (N=2121), the authors demonstrated that whereas the higher SI is associated with increased risk of in‐hospital mortality, both low and high SIs (low SI indicating presence of Cushing reflex) were associated with early mortality within 3 and 7 days, albeit with higher odds associated with high SI compared with low SI. Although a clear plausible link exists between high or low SI and stroke mortality outcome, no study to date has examined the link between admission SI and other important stroke outcomes such as length of hospital stay and disability.

Using the Get With The Guidelines–Stroke (GWTG‐Stroke) data, we aimed to examine admission SI as a prognostic marker not just for in‐hospital mortality but also for other important stroke outcomes (ie, in‐hospital mortality, discharge destination, discharge ambulatory status, length of stay, and discharge modified Rankin score) controlling for multiple potential confounders in a large cohort of stroke patients. We were also interested in exploring whether adding SI to existing mortality prediction scores would improve the scores’ discriminative ability significantly if SI were an independent predictor of in‐hospital mortality outcome. Therefore, we also aimed to examine the added value of inclusion of SI in existing GWTG‐Stroke mortality prediction models.

Methods

The American Heart Association (AHA) and American Stroke Association GWTG‐Stroke database data collection methods have been described previously.7, 8, 9, 10 In brief, participating hospitals used the AHA's Internet‐based Patient Management Tool (Quintiles) to enter data, receive decision support, and obtain feedback via on demand reports of performance on quality measures and recorded data from consecutive admissions for acute ischemic stroke.

Trained hospital personnel abstracted data using the Internet‐based Patient Management Tool with standardized data definitions and detailed coding instructions. The Internet‐based system performs checks to ensure that the reported data are complete and internally consistent. In addition, data quality is monitored for both completeness and accuracy. Hospitals that participate must receive approval through their local institutional review boards or a waiver of individual consent under the common rule. Quintiles is the data collection coordination center for the GWTG‐Stroke program.11

The Duke Clinical Research Institute serves as the data analysis center. Hospital characteristics (ie, academic teaching status, bed size) were based on American Hospital Association data.10 Past medical history was defined on the basis of preexisting conditions, with the exclusion of conditions that were newly diagnosed during the hospital stay.

The predictor variable was SI, which is calculated as heart rate divided by systolic BP. These variables were abstracted by trained hospital personnel from the first values recorded in the emergency department record. For descriptive purposes, 3 categories of SI were defined: SI <0.50, SI 0.50 to 0.70 as the normal range, and SI >0.70. The outcomes of interest are patient‐relevant outcomes of in‐hospital mortality; poor discharge outcome, defined as discharge other than home; poor functional status at discharge, defined as inability to ambulate independently; disability at the time of discharged, defined as modified Rankin Scale (mRS) 3 to 6; and acute hospital length of stay (LOS). For this study purpose, only intracerebral hemorrhages were included as hemorrhagic stroke, and other hemorrhages presenting with stroke symptoms such as subarachnoid hemorrhages or subdural hematoma were excluded.

Patient Inclusion and Exclusion

A total of 694 623 ischemic stroke patients were admitted between October 2012 to April 1, 2015, at 1732 sites after excluding hospitals with >25% missing medical history. Further exclusions included patients with missing admission heart rate or missing admission systolic BP (206 sites, n=232 475); patients who transferred out, who left against medical advice, or whose discharge disposition was missing, not determined, or unable to be determined (8 sites, n=13 618); and patients who were transferred in (4 sites, n=66 658). Consequently, a total of 381 872 ischemic stroke patients from 1514 sites were included in the current study (Figure S1A). Similar inclusion and exclusion criteria were applied for hemorrhagic stroke (intracerebral hemorrhage) over the same study period (Figure S1B). Comparison of patients who were included and excluded showed that a majority of the characteristics are not significantly different between the 2 groups (data not shown).

Statistical Analysis

Patient characteristics are summarized and compared by SI groups. To examine the association of SI and study outcomes, adjusted multivariable logistic regression models with generalized estimating equations to account for in‐hospital clustering were created for each binary outcome. Multiple imputation with 25‐fold imputations were used to impute missing data for covariates (see Table S1 for missing data) to account for possible confounders. If a patient had missing medical history, it was assumed that the medical condition did not occur. National Institutes of Health Stroke Scale (NIHSS) score and hospital characteristics were not imputed.

Regression models controlled for various important confounders that are known or likely to influence the relationship between predictor and outcome. We adjusted demographic characteristics such as age, sex, race, insurance, and comorbidities (prosthetic heart valve, previous stroke, coronary artery disease or prior myocardial infarction, diabetes mellitus, peripheral vascular disease, hypertension, heart failure). Other covariates included risk factors such as body mass index (kg/m2), smoking status, and ambulatory status before the current event, and whether the patient arrived at the hospital during off‐hours (regular hours defined as 7 am to 6 pm Monday–Friday). We also adjusted for other relevant biochemical and physiological markers at the time of admission: glucose (mg/dL), creatinine (mg/dL), diastolic BP (mm Hg), international normalized ratio, total cholesterol, HDL (high‐density lipoprotein), LDL (low‐density lipoprotein), and triglycerides. Other clinical factors included presence or absence of thrombotic complications of deep venous thrombosis and pulmonary embolism. We also controlled for the following hospital characteristics: region, hospital type, number of beds, rural versus urban location, annual volume of respective ischemic/hemorrhagic stroke admissions, annual volume of tissue plasminogen activator administration (ischemic stroke patients only), and primary stroke center. For those with NIHSS data (n=344 100), we also adjusted for NIHSS, and analyses were conducted as subanalyses.

We repeated the analysis by including SI along with systolic BP instead of diastolic BP. Given high collinearity, heart rate was not included in the model. We also examined the associations by including SI alone without BP components or heart rate. We further assessed the association between both systolic BP and heart rate on all outcomes using multivariable logistic regression models.

To better understand the observed associations between SI and outcomes examined within each SI category, we also fitted both linear and nonlinear models. The latter was done by way of linear splines. First, restricted cubic spline models were used (data not shown) as inference to help us to see the shape of the curves and choose the knots (or cut points) for linear splines.12 SI was modeled as cubic splines with knots at 5th, 50th and 95th percentiles of SI. Based on the results from cubic spline, SI was modeled by using linear splines with knots at 0.5 and 1.5.

Finally, we examined the usefulness of adding the SI to existing GWTG‐Stroke in‐hospital mortality risk scores9, 13 by calculating C‐index with or without SI in the original GWTG‐Stroke models with or without NIHSS (categorical as well as every 0.1 increase in SI). Estimates of reclassification were also assessed using net reclassification index analysis. Calibration plots were also constructed using the continuous SI models.

Results

The study cohort consisted of 425 808 patients with a mean±SD age of 71.04±14.53 years (48.8% male), of whom 89.7% had ischemic stroke and 10.3% had intracerebral hemorrhage. The mean±SD ages were 70.4±14.6 and 69.5±14.8 years for patients ischemic and hemorrhagic stroke, respectively. The majority of patients were white; black patients represented ≈18% and 20% of the sample with ischemic and hemorrhagic stroke subtypes, respectively (versus 69% and 63% white, respectively).

Table 1 shows the total sample characteristics for selected variables (overall and by SI categories). Statistically significant differences were noted among the 3 SI groups. Those with higher SIs (>0.70) were more likely to be younger, and fewer were male. They were more likely to arrive during off‐hours and tended to have more severe stroke, depicted by NIHSS (median NIHSS values for SI of <0.5, 0.5–0.7, and >0.7 were 3, 4, and 6, respectively), and were less likely to be independently ambulatory. Notable differences in cardiovascular comorbidities included significantly higher prevalence of peripheral vascular disease and heart failure with significantly lower prevalence of hypertension compared with the other 2 categories. They had significantly higher heart rate and lower systolic BP on admission and were associated with significantly higher rates of poor outcomes for all outcomes examined.

Table 1

Characteristics and Outcomes for the Study Population and by SI Categories for Selected Variables

Variable	Overall (N=425 808)	SI <0.50 (n=203 807)	SI 0.50–0.70 (n=163 980)	SI >0.70 (n=58 021)	P Value
Characteristics
Age, y	71.04±14.53	72.39±13.57	70.01±15.01	69.21±15.88	<0.0001
Sex (male)	207 763 (48.79)	101 863 (49.98)	79 539 (48.51)	26 361 (45.43)	<0.0001
Stroke type					<0.0001
Ischemic	381 872 (89.68)	180 229 (88.43)	149 002 (90.87)	52 641 (90.73)
Hemorrhagic	43 936 (10.32)	23 578 (11.57)	14 978 (9.13)	5380 (9.27)
Race/ethnicity					<0.0001
Other	13 629 (3.20)	6424 (3.15)	5321 (3.25)	1884 (3.25)
Asian	13 134 (3.09)	6705 (3.29)	4848 (2.96)	1581 (2.73)
Hispanic (any race)	30 802 (7.24)	14 502 (7.12)	12 187 (7.44)	4113 (7.09)
Black	79 285 (18.63)	38 601 (18.95)	30 139 (18.39)	10 545 (18.18)
White	288 753 (67.85)	137 491 (67.49)	111 397 (67.97)	39 865 (68.75)
Missing	205 (0.05)	84 (0.04)	88 (0.05)	33 (0.06)
Arrived during off‐hours (yes)a	188 736 (44.32)	89 970 (44.14)	72 638 (44.30)	26 128 (45.03)	0.0007
Initial NIHSS[Link] (missing, 19.19%), median (IQR)	4 (1–10)	3 (1–9)	4 (1–10)	6 (2–14)	<0.0001
Ambulatory status independent	127 089 (31.60)	63 606 (33.06)	49 701 (32.10)	13 782 (25.08)	<0.0001
Medical history (yes)
Previous stroke/TIA	134 334 (31.56)	63 784 (31.30)	52 209 (31.85)	18 342 (31.62)	0.0019
CAD/prior MI	104 509 (24.55)	51 914 (25.48)	38 659 (23.58)	13 936 (24.03)	<0.0001
Diabetes mellitus	144 248 (33.88)	69 071 (33.90)	55 983 (34.15)	19 194 (33.09)	<0.0001
PVD	20 170 (4.74)	9479 (4.65)	7662 (4.67)	3029 (5.22)	<0.0001
Hypertension	330 174 (77.56)	167 235 (82.07)	122 128 (74.50)	40 811 (70.36)	<0.0001
Heart failure	39 167 (9.20)	16 217 (7.96)	15 526 (9.47)	7424 (12.80)	<0.0001
Smoker (yes)	73 445 (17.25)	32 259 (15.83)	30 084 (18.35)	11 102 (19.14)	<0.0001
BMI, kg/m2	28.07±6.78	28.16±6.56	28.14±6.89	27.50±7.18	<0.0001
Heart rate, beats/min	82±18	71±12	86±14	105±20	<0.0001
Systolic BP, mm Hg	160±31	177±27	149±23	126±23	<0.0001
Glucose, mg/dL	145±72	141±67	146±75	153±82	<0.0001
Cholesterol, mg/dL	170±48	173±48	167±48	158±48	<0.0001
Hospital characteristics
Primary stroke center (yes)	186 487 (43.80)	89 369 (43.85)	71 698 (43.72)	25 420 (43.81)	0.74
Region					<0.0001
West	77 212 (18.15)	36 559 (17.95)	29 849 (18.22)	10 804 (18.64)
South	150 665 (35.41)	73 109 (35.90)	57 350 (35.00)	20 206 (34.86)
Midwest	77 841 (18.30)	37 440 (18.39)	29 773 (18.17)	10 628 (18.33)
Northeast	119 722 (28.14)	56 514 (27.75)	46 879 (28.61)	16 329 (28.17)
Missing	368 (0.09)	185 (0.09)	129 (0.08)	54 (0.09)
Outcomes
In‐hospital mortality	27 903 (6.55)	11 895 (5.84)	9601 (5.85)	6407 (11.04)	<0.0001
Died as inpatient
LOS >4 d (missing, n=12 884)	164 760 (39.90)	75 788 (37.96)	62 682 (39.42)	26 290 (48.43)	<0.0001
Independent ambulatory status
No (missing, n=42 454)	188 221 (49.10)	88 897 (48.07)	71 606 (48.16)	27 718 (55.70)	<0.0001
mRS >2	146 562 (61.34)	68 182 (60.03)	55 344 (60.05)	23 036 (69.42)	<0.0001

Values presented are mean±SD for continuous data and n (%) for categorical data, except as noted. BMI indicates body mass index; BP, blood pressure; CAD, coronary artery disease; IQR, interquartile range; LOS, length of stay; MI, myocardial infarction; mRS, modified Rankin Scale; NIHSS, National Institute of Health Stroke Scale; PVD, peripheral vascular disease; SI, stroke index; TIA, transient ischemic attack.

Daytime regular hours were defined as 7 am to 6 pm Monday to Friday; all other times (including all day Saturday and Sunday) were considered off‐hours.

Tables S1 through S3 show the characteristics of all variables included in the models and the outcomes of the overall sample and the ischemic and hemorrhagic stroke groups in detail. Figures S2A through S2C demonstrate the sample's SI distribution for the total sample and for the ischemic and hemorrhagic stroke groups. With large numbers, the overall P values for trends are significant, although the material differences between groups are relatively minor. Those with higher SI in both stroke types were more likely to be younger, female, prior anticoagulant users with higher international normalized ratios, and current smokers and to have higher prevalence of heart failure, lower cholesterol, and higher glucose levels on admission. Not surprisingly, history of hypertension and higher systolic BP and lower levels of heart rate were less likely to be associated with higher SI. Higher NIHSS was associated with higher SI in ischemic stroke, but the trend appeared to be U shaped in hemorrhagic stroke; high NIHSS was associated with both low and high SI.

Higher SI is also associated with poor outcomes of mortality, ambulatory status at discharge, and disability outcome (mRS >2) when crude comparisons were made. In‐hospital mortality for those with SI >0.7 was almost double that of the other SI categories (11.0% versus ≈5.9%, respectively). Also in this category, 55.7% patients were not able to ambulate independently compared with ≈48% in other 2 categories. The crude rate of mRS >2 outcome also increased ≈10% in those with SI >0.7 compared with the other categories. LOS (>4 days) appears to be differently associated with SI depending on stroke type; a higher proportion of ischemic stroke patients with longer LOS had higher SI (35.9% with SI <0.5, 38.1% with SI 0.5–0.7, and 48.3% with SI >0.7), whereas the reverse was true for hemorrhagic stroke (53.9%, 52.2%, and 49.9%, respectively).

As expected, the individual components of SI, heart rate and systolic BP, are also independent predictors of outcomes. Both lower heart rate and higher systolic BP are also independently associated with outcomes. Repeating the analyses including SI or replacing diastolic BP with systolic BP did not alter the results (data not shown).

Table 2 shows the odds ratios and corresponding 95% confidence intervals for the relationship between low (<0.5) and high (>0.7) SI compared with normal SI (0.5–0.7) for each selected outcome in the whole sample and separately for ischemic and hemorrhagic stroke subtypes after controlling for potential confounders listed in the study methods. The results consistently showed the higher SI group (SI >0.7) associated with worse outcomes including in‐hospital mortality, longer LOS, likelihood of being discharged to a destination other than home, inability to ambulate independently at the time of hospital discharge, and association with higher likelihood of disability defined as mRS >2 (ie, discharge mRS 3–6).

Table 2

ORs and Corresponding 95% CIs for Those With Admission SI <0.5 and >0.7 Compared With SI 0.5–0.7 for Study Outcomes for All Participants and by Ischemic and Hemorrhagic Stroke in GWTG‐Stroke October 2012 to March 2015

Outcomes	SI <0.5 vs 0.5–0.7 OR (95% CI)	P Value	SI >0.7 vs 0.5–0.7 OR (95% CI)	P Value
In‐hospital mortality
All	0.86 (0.83–0.89)	<0.0001	2.00 (1.92–2.08)	<0.0001
Ischemic stroke	0.79 (0.76–0.82)	<0.0001	1.97 (1.87–2.06)	<0.0001
Hemorrhagic stroke	1.01 (0.96–1.08)	0.64	1.85 (1.69–2.02)	<0.0001
LOS (>4 vs 0–4 d)
All	0.87 (0.86–0.89)	<0.0001	1.46 (1.43–1.49)	<0.0001
Ischemic stroke	0.86 (0.84–0.87)	<0.0001	1.53 (1.49–1.56)	<0.0001
Hemorrhagic stroke	0.97 (0.92–1.02)	0.21	0.95 (0.89–1.02)	0.14
Discharge destination other than home
All	0.89 (0.87–0.90)	<0.001	1.50 (1.47–1.54)	<0.001
Ischemic stroke	0.88 (0.86–0.90)	<0.001	1.49 (1.45–1.53)	<0.001
Hemorrhagic stroke	0.97 (0.91–1.03)	0.32	1.53 (1.38–1.68)	<0.001
Independent ambulatory status at discharge (no vs yes)
All	0.89 (0.87–0.91)	<0.0001	1.41 (1.38–1.45)	<0.0001
Ischemic stroke	0.89 (0.87–0.91)	<0.0001	1.40 (1.36–1.44)	<0.0001
Hemorrhagic stroke	0.89 (0.83–0.96)	0.0011	1.46 (1.32–1.62)	<0.0001
Disability (mRS 3–6 vs 0–2)
All	0.88 (0.86–0.90)	<0.0001	1.52 (1.47–1.57)	<0.0001
Ischemic stroke	0.88 (0.86–0.90)	<0.0001	1.49 (1.45–1.54)	<0.0001
Hemorrhagic stroke	0.99 (0.90–1.08)	0.81	1.83 (1.59–2.09)	<0.0001

Models adjusted for patient and hospital characteristics including systolic and diastolic blood pressure (see Methods for details). CI indicates confidence interval; GWTG‐Stroke, Get With The Guidelines–Stroke; LOS, length of stay; mRS, modified Rankin Scale; OR, odds ratio; SI, stroke index.

Table 3 shows the results for those with NIHSS data available (n=344 100) with additional adjustment for NIHSS score. Results were broadly similar to the whole sample except for the nonsignificant association of the hemorrhagic stroke subtype with longer LOS, although a point estimate showed a similar direction (odds ratio: 1.07). Based on cubic spline results, linear spline models were constructed for these outcomes using knots at 0.5 and 0.7 to better understand the dose‐response relationship within each SI category. Odds ratios and their corresponding 95% confidence intervals were calculated by incremental 0.1‐U increase in SI, and results are presented in Table 4. Consistent with earlier results, an incremental increase of 0.1 U of SI in a linear fashion was associated with worse outcomes for SI >0.5. With a lower number of events, the results appeared to be less consistent for hemorrhagic stroke. The cubic splines, which were constructed for inference purposes of finding relevant knots for linear spline models, are shown as Figures S4A–S4C to S5A–S5C.

Table 3

ORs and Corresponding 95% CIs for Those With Admission SI <0.5 and >0.7 Compared With SI 0.5–0.7 for Study Outcomes for All Participants (N=344 100) and by Ischemic Stroke (n=315 612) and Hemorrhagic Stroke (n=28 488) in GWTG‐Stroke October 2012 to March 2015 for Those With NIHSS Data

Models	SI <0.5 OR (95% CI)	P Value	SI >0.7 OR (95% CI)	P Value
In‐hospital mortality
All	0.91 (0.88–0.94)	<0.0001	1.53 (1.46–1.61)	<0.0001
Ischemic stroke	0.89 (0.85–0.92)	<0.0001	1.54 (1.46–1.62)	<0.0001
Hemorrhagic stroke	0.98 (0.90–1.07)	0.70	1.39 (1.23–1.58)	<0.0001
Length of stay (>4 vs 0–4 d)
All	0.90 (0.88–0.92)	<0.0001	1.39 (1.36–1.43)	<0.0001
Ischemic stroke	0.89 (0.88–0.91)	<0.0001	1.42 (1.38–1.46)	<0.0001
Hemorrhagic stroke	0.99 (0.93–1.05)	0.64	1.07 (0.98–1.17)	0.16
Discharge destination other than home
All	0.94 (0.92–0.95)	<0.001	1.28 (1.24–1.31)	<0.001
Ischemic stroke	0.93 (0.92–0.95)	<0.001	1.28 (1.24–1.32)	<0.001
Hemorrhagic stroke	0.95 (0.88–1.03)	0.18	1.23 (1.07–1.42)	0.004
Independent ambulatory status at discharge (no vs yes)
All	0.94 (0.92–0.97)	<0.0001	1.25 (1.21–1.29)	<0.0001
Ischemic stroke	0.95 (0.93–0.97)	<0.0001	1.25 (1.21–1.28)	<0.0001
Hemorrhagic stroke	0.89 (0.81–0.97)	0.0081	1.29 (1.12–1.49)	<0.0001
Disability (mRS 3–6 vs 0–2)
All	0.94 (0.91–0.97)	<0.0001	1.28 (1.23–1.33)	<0.0001
Ischemic stroke	0.93 (0.91–0.96)	<0.0001	1.27 (1.21–1.33)	<0.0001
Hemorrhagic stroke	1.01 (0.91–1.13)	0.83	1.39 (1.15–1.69)	0.0007

Models were adjusted for patient and hospital characteristics, including systolic blood pressure, diastolic blood pressure, and NIHSS. See Methods for details. CI indicates confidence interval; GWTG‐Stroke, Get With The Guidelines–Stroke; mRS, modified Rankin Scale; NIHSS, National Institute of Health Stroke Scale; OR, odds ratio; SI, stroke index.

Table 4

ORs and Corresponding 95% CIs for Every Incremental Increase in SI 0.1 Within Each SI Category (Coded as Linear Spline With Knots at 0.5 and 0.7) for Study Outcomes for All Patients and by Ischemic and Hemorrhagic Stroke Patients Separately in GWTG‐Stroke (October 2012–March 2015)

Outcomes	SI 0.0–0.5 OR (95% CI)	SI 0.5–0.7 OR (95% CI)	SI >0.7 OR (95% CI)
In‐hospital mortality
All	0.91 (0.89–0.94)	1.35 (1.26–1.45)	1.22 (1.10–1.35)
Ischemic stroke	1.00 (0.97–1.04)	1.37 (1.26–1.50)	1.20 (1.07–1.35)
Hemorrhagic stroke	0.83 (0.79–0.87)	1.22 (1.08–1.38)	1.26 (1.04–1.52)
LOS (>4 vs 0–4 d)
All	1.02 (1.00–1.03)	1.20 (1.16–1.25)	1.10 (1.04–1.17)
Ischemic stroke	1.01 (1.00–1.03)	1.22 (1.18–1.27)	1.13 (1.06–1.20)
Hemorrhagic stroke	1.07 (1.03–1.11)	1.00 (0.91–1.10)	0.96 (0.83–1.12)
Discharge destination other than home
All	1.01 (1.00–1.03)	1.18 (1.13–1.22)	1.16 (1.09–1.23)
Ischemic stroke	1.02 (1.01–1.04)	1.18 (1.13–1.23)	1.16 (1.09–1.23)
Hemorrhagic stroke	0.96 (0.91–1.01)	1.14 (1.00–1.30)	1.19 (0.97–1.46)
Independent ambulatory status at discharge (no vs yes)
All	1.02 (1.01–1.04)	1.17 (1.12–1.21)	1.12 (1.05–1.19)
Ischemic stroke	1.03 (1.01–1.04)	1.16 (1.12–1.21)	1.12 (1.05–1.19)
Hemorrhagic stroke	1.00 (0.95–1.06)	1.18 (1.02–1.35)	1.11 (0.89–1.39)
Disability (mRS 3–6 vs 0–2)
All	1.00 (0.98–1.02)	1.20 (1.14–1.26)	1.17 (1.08–1.26)
Ischemic stroke	1.01 (0.99–1.04)	1.19 (1.13–1.26)	1.16 (1.07–1.26)
Hemorrhagic stroke	0.94 (0.87–1.01)	1.15 (0.95–1.40)	1.29 (0.93–1.79)

Models were adjusted for baseline patient and hospital characteristics, including systolic and diastolic blood pressure. CI indicates confidence interval; GWTG‐Stroke, Get With The Guidelines–Stroke; LOS, length of stay; mRS, modified Rankin Scale; OR, odds ratio; SI, stroke index.

Table 5 shows the comparison of C‐indexes calculated with or without inclusion of SI for the original GWTG‐Stroke models with or without NIHSS for mortality outcome. There were modest improvements in C‐statistics by adding SI (categorically or every 0.1‐U incremental increase in SI) to the original model without stroke severity,9 but little to no improvement when SI was added to the model that also included NIHSS.13 However, in the subsample with available NIHSS data, exclusion of NIHSS from the model performed less well. Table 5 also shows the results of the net reclassification index with values >0 for the model with SI, confirming the reclassification improvement of the new model (with SI) compared with the model without SI.

Table 5

Mortality Outcome Prediction by Including SI in the Original GWTG‐Stroke and GWTG‐Stroke NIHSS Indexes in This Cohort

Model	C‐Index Comparison		Model With SI, NRI (95% CI)
	With SI	Without SI
GWTG‐Stroke without NIHSSa
In‐hospital mortality (overall population)	0.816	0.809	0.250 (0.238–0.262)a
In‐hospital mortality (ischemic stroke subgroup)	0.767	0.757	0.192 (0.163–0.222)a
In‐hospital mortality (hemorrhagic stroke)	0.744	0.741	0.232 (0.218–0.246)a
Available NIHSS Data GWTG‐Stroke With NIHSS Modelb
In‐hospital mortality (overall population)	0.875	0.873	0.802
In‐hospital mortality (ischemic stroke subgroup)	0.855	0.852	0.767
In‐hospital mortality (hemorrhagic stroke)	0.852	0.852	0.734
SI as 0.1 Increment
In‐hospital mortality (overall population)	0.818	0.809	…
In‐hospital mortality (ischemic stroke subgroup)	0.771	0.757	…
In‐hospital mortality (hemorrhagic stroke)	0.748	0.741	…

CI indicates confidence interval; GWTG‐Stroke, Get With The Guidelines–Stroke; mRS, modified Rankin Scale; NIHSS, National Institute of Health Stroke Scale; NRI, net reclassification improvement; SI, shock index.

NRI >0 means reclassification improvement of the new model with SI compared with the model without SI.

NRI is for model with SI without NIHSS.

Figure shows the calibration plots for overall population (Figure 1A), and then for ischemic stroke and intracerebral hemorrhage cohorts separately (Figure 1B and 1C, respectively) using the continuous SI models (ie, the last 3 models in Table 5. This demonstrates the good fit of model prediction from the observed data.

Figure 1

Calibration plots for (A) the overall population, (B) the ischemic stroke cohort, and (C) the intracerebral hemorrhage cohort.

Discussion

To our knowledge, this study is the first to examine the association between the SI and relevant clinical outcomes of acute hospital LOS, discharge destination, ambulatory status at the time of discharge, and disability outcome. We are the first to examine this association for specific stroke subtypes, ischemic and hemorrhagic stroke, using the largest sample to date, with >420 000 patients with unselected consecutively admitted strokes to stroke services across the United States.

We found consistent predictive value of admission SI by clear demonstration of significantly worse outcomes with higher levels of SI, including mortality. This association appears to be linear; every incremental increase in SI of 0.1 above an SI value of 0.5 appears to be an important prognostic factor after taking into account individual prognostic factors and clinical and hospital characteristics. McCall and colleagues previously examined the link between SI and acute stroke mortality in a much smaller sample (n=2121). They found similar results demonstrating that high SI value was associated with increased inpatient mortality (odds ratio: 1.85; 95% confidence interval, 1.17–2.92).6 Of note, their population was predominantly white (>95%), and they were not able to control for some key variables of initial stroke severity, such as NIHSS or poststroke complications. Our results are broadly similar, with an overall estimated odds ratio of 2.0 for mortality. We were able to examine the stroke type–specific in‐hospital mortality outcome, and the results appeared to be consistent for both stroke subtypes.

Acute stroke causes an increase in sympathetic outflow to the heart in an attempt to increase arterial BP and total peripheral resistance to maintain cerebral blood flow, accompanied by bradycardia known as Cushing reflex. Although Cushing reflex is noticeable only with more severe ischemic injury (direct effect of ischemic injury or secondary to hemorrhagic stroke with resultant edema), the subtle physiological changes may have already been occurring in any ischemic stroke depending on the severity of ischemic injury. Kalmer et al14 showed that the Cushing reflex indicated by higher systolic BP and bradycardia predicted poor intracranial perfusion and high intracranial pressure even with minimally invasive procedures. Furthermore, mortality after a subarachnoid hemorrhage was predicted by high intracranial pressure, which demonstrated poor intracranial perfusion.15

A lower SI may also have prognostic significance in stroke, serving as an early marker of the Cushing reflex. Conversely, large strokes with dysphagia or posterior circulation strokes presenting with vomiting present a high risk of aspiration pneumonia and may be associated with higher SI, as they are more likely to cause higher heart rate and lower systolic BP. Consequently, a potential U‐ or J‐shaped relationship was expected between SI and selected outcomes; however, we did not find the U‐shaped relationship that was observed in the study by McCall et al.6 Of note, McCall and colleagues did not adjust for poststroke complications, which could influence the outcomes examined.

The prognostic value of high SI has been previously reported in conditions such as sepsis and severe volume depletion.1, 16 Previous studies have used SI to assess outcomes in patients with community‐acquired pneumonia and acute circulatory failure and to predict those who would require immediate intensive therapy.1, 16, 17, 18 Moreover, a recent study that used prospectively collected audit data showed that an SI >1.0 was predictive of inpatient mortality in individuals aged >90 years.19 It should be noted that SI can vary during the process of care, as BP management in stroke may influence its values, and thus it is expected that SI would fluctuate during the course of admission. This may have attenuated the observed relationships.

It was unknown whether incorporating SI into prognostic scoring systems to improve existing scoring systems such as GWTG‐Stroke,9, 13 iScore,20 SOAR (Stroke Subtype, Oxford Community Stroke Project Classification, Age, Prestroke Modified Rankin) score,21 and mSOAR (modified SOAR)22 would improve the clinical prediction rules in stroke. We extended our study aims to examine the usefulness of adding SI to existing GWTG‐Stroke scores.9, 13 The addition of SI in the absence of NIHSS data appeared to improve the prediction of the GWTG‐Stroke prognosis scoring system. Indeed, both heart rate and systolic BP can be accurately measured and recorded in any hospital setting without special training. Furthermore, McCall et al already demonstrated that using SI has greater predictive ability than systolic BP with regard to short‐term mortality.6 In clinical practice, information on vital signs can be readily abstracted from routine examination; therefore, SI can be easily calculated, and using SI could improve prediction of outcomes for individual patients as well as risk adjustment. Consequently, SI is particularly useful in jurisdictions where standardized stroke scales that require training (eg, NIHSS) are difficult to implement in the emergency setting.

SI is a very simple index derived from 2 readily available vital signs—heart rate and systolic BP—and its value at the point of care can provide valuable prognostic information. Indeed, Lee et al demonstrated in the GUSTO‐I (Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries‐I) that both heart rate and BP are among 5 predictors of 30‐day mortality after acute myocardial infarction.23 Contemporary findings in the field of cardiology showed similar results; in a cohort of 7187 patients with ST‐segment elevation–myocardial infarction, admission SI ≥0.7 was also associated with 1.6‐ and 1.5‐fold increased risk of 7‐ and 30‐day major adverse cardiovascular events, respectively.5

The key strengths of our work include the large sample size and prospective data collection. We were able to conduct robust statistical analysis with control for potential confounders including poststroke complications. We were able to examine the outcomes by specific stroke subtypes and robustly control for initial stroke severity in those with NIHSS data available. Furthermore, our study has good ethnic mix and thus is potentially applicable to other populations including black, Asian, and Hispanic ethnic groups. The net reclassification index confirms the reclassification improvement of SI, and calibration plots demonstrate the model's fitness. This shows the usefulness of SI in a clinical setting because SI is an easily obtainable physiological parameter. Moreover, as a physiological index, the relationship between SI and outcome is unlikely to be different between different races or sexes; therefore, the results are generalizable.

Our study has some limitations. Hospital‐based study outcomes are confined to in‐hospital outcomes (ie, up to the time of hospital discharge); however, the outcomes assessed are important for stroke patients and their significant others. Moreover, prediction of LOS and discharge destination forms an important aspect of service provision and assessment of social care needs. We acknowledge large amounts of missing data for some variables; however, we have used multiple imputation to account for this. As an observational study, causality between SI and poor stroke outcomes cannot be drawn. Nevertheless, the observed associations have plausible explanations, observed prospectively and consistent with the wider literature of cardiovascular outcome epidemiology. Because data were from a hospital‐based registry, some stroke patients might not have been included, such as those who died before admission. Patients and hospitals may not be entirely representative of the US population, but the sample population is comparable to all US patients hospitalized with stroke.24

We were not able to assess the effect of unmeasured or residual confounders; however, we were able to control for a comprehensive list of potential confounding variables. A substantial proportion of patients had missing NIHSS data; however, that was because of nonrandom missingness, and repeating the analyses in those who had NIHSS data showed similar results. About of a third of patients were excluded for various reasons including missing data, and this could potentially limit generalizability; however, we were able to impute for missing variables, and the percentage of missing values is generally low or very low. As a physiological parameter, the relationship between SI and outcome is unlikely to be influenced by population characteristics; therefore, the findings are generalizable and applicable to any stroke patient in predicting likely outcome at the point of care.

In summary, our study shows that SI is a significant predictor of important patient‐related acute stroke outcomes including mortality, acute hospital LOS, discharge destination, ambulatory status at the time of discharge, and poststroke disability. Our robust statistical analysis, taking into account multiple confounding variables, showed a relationship between SI and clinical outcomes; however, further studies are required to demonstrate prognostic utility before it could be widely adopted as a clinical tool. This information may be useful in clinical practice for managing stroke patients, to identify those with high risk of poor outcomes from the point of contact, particularly if NIHSS is not available, and to better inform patients and their significant others about the prognosis of these important outcomes.

Author Contributions

Myint and Smith conceived the idea and developed the analysis plan with critical input from the coauthors. Sheng analyzed the data. Myint and Smith drafted the article with input from all coauthors. All authors contributed to interpretation of the results and made important intellectual contributions to the article.

Sources of Funding

The Get With The Guidelines–Stroke (GWTG‐Stroke) program is currently supported in part by a charitable contribution from Bristol‐Myers Squibb/Sanofi Pharmaceutical Partnership and the American Heart Association Pharmaceutical Roundtable. GWTG‐Stroke has been funded in the past through support from Boehringer‐Ingelheim and Merck. These funding agencies did not participate in the design or analysis, article preparation, or approval of this study.

Disclosures

Dr Bhatt discloses the following relationships—Advisory Board: Cardax, Elsevier Practice Update Cardiology, Medscape Cardiology, Regado Biosciences; Board of Directors: Boston VA Research Institute, Society of Cardiovascular Patient Care, TobeSoft; Chair: American Heart Association Quality Oversight Committee; Data Monitoring Committees: Baim Institute for Clinical Research (formerly Harvard Clinical Research Institute, for the PORTICO trial, funded by St. Jude Medical, now Abbott), Cleveland Clinic, Duke Clinical Research Institute, Mayo Clinic, Mount Sinai School of Medicine, Population Health Research Institute; Honoraria: American College of Cardiology (Senior Associate Editor, Clinical Trials and News, http://ACC.org; Vice‐Chair, ACC Accreditation Committee), Baim Institute for Clinical Research (formerly Harvard Clinical Research Institute; RE‐DUAL PCI clinical trial steering committee funded by Boehringer Ingelheim), Belvoir Publications (Editor in Chief, Harvard Heart Letter), Duke Clinical Research Institute (clinical trial steering committees), HMP Global (Editor in Chief, Journal of Invasive Cardiology), Journal of the American College of Cardiology (Guest Editor; Associate Editor), Population Health Research Institute (for the COMPASS operations committee, publications committee, steering committee, and USA national coleader, funded by Bayer), Slack Publications (Chief Medical Editor, Cardiology Today's Intervention), Society of Cardiovascular Patient Care (Secretary/Treasurer), WebMD (CME steering committees); Other: Clinical Cardiology (Deputy Editor), NCDR‐ACTION Registry Steering Committee (Chair), VA CART Research and Publications Committee (Chair); Research Funding: Abbott, Amarin, Amgen, AstraZeneca, Bayer, Boehringer Ingelheim, Bristol‐Myers Squibb, Chiesi, Eisai, Ethicon, Forest Laboratories, Idorsia, Ironwood, Ischemix, Lilly, Medtronic, PhaseBio, Pfizer, Regeneron, Roche, Sanofi Aventis, Synaptic, The Medicines Company; Royalties: Elsevier (Editor, Cardiovascular Intervention: A Companion to Braunwald's Heart Disease); Site Co‐Investigator: Biotronik, Boston Scientific, St. Jude Medical (now Abbott), Svelte; Trustee: American College of Cardiology; Unfunded Research: FlowCo, Merck, PLx Pharma, Takeda. Dr Fonarow is a member of the GWTG Executive Committee; has served as a consultant to Janssen (modest), receives research support from PCORI (significant), and is an employee of UCLA, which holds a patent on retriever devices for stroke. Dr Myint has received support from Viforpharma to attend an advisory meeting. Dr Reeves has received salary support from the Michigan Stroke GWTG Registry Dr Saver is an employee of the University of California. Dr Saver has served as an unpaid site investigator in multicenter trials run by Medtronic and Stryker for which the UC Regents received payments on the basis of clinical trial contracts for the number of subjects enrolled. Dr Saver receives funding for services as a scientific consultant regarding trial design and conduct to Medtronic/Covidien, Stryker, Neuravi, BrainsGate, Pfizer, Squibb, Boehringer Ingelheim (prevention only), ZZ Biotech, and St. Jude Medical. Dr Saver serves as an unpaid consultant to Genentech advising on the design and conduct of the PRISMS trial; neither the University of California nor Dr Saver received any payments for this voluntary service. The University of California has patent rights in retrieval devices for stroke. Dr Schwamm reports being the principal investigator of an investigator‐initiated study of extended‐window intravenous thrombolysis funded by the National Institutes of Neurological Disorders and Stroke (http://clinicaltrials.gov/show/NCT01282242) for which Genentech provided alteplase free of charge to Massachusetts General Hospital as well as supplemental per‐patient payments to participating sites; serving as chair of the AHA/ASA GWTG stroke clinical work group and hospital accreditation Science Committee and Quality Oversight Committees, co‐chair of Mission‐Lifeline: Stroke; serving as a stroke systems consultant to the Massachusetts Department of Public Health; and serving as a scientific consultant to LifeImage regarding user interface design and usability, and regarding trial design and conduct to Lundbeck (international steering committee, DIAS3, 4 trial), Penumbra (data and safety monitoring committee, Separator 3D and MIND trials) and Medtronic (Victory AF and Stroke AF trials). Dr Smith is a member of the American Heart Association Get With The Guidelines Steering Committee; has received research support from the Canadian Institutes for Health Research, Brain Canada, Alberta Innnovates—Health Solutions, and the Canadian Partnership Against Cancer. The remaining authors have no disclosures to report.

Table S1. Characteristics and Outcomes by Shock Index

Table S2. Characteristics and Outcomes by Shock Index, Acute Ischemic Stroke Cohort

Table S3. Characteristics and Outcomes by Shock Index, Intracerebral Hemorrhage Cohort

Figure S1A. Study cohort development and exclusions: acute ischemic stroke.

Figure S1B. Study cohort development and exclusions: intracerebral hemorrhage.

Figure S2A. Shock index distribution of the whole sample.

Figure S2B. Shock index distribution of acute ischemic stroke patients.

Figure S2C. Shock index distribution of intracerebral hemorrhage patients.

Figure S3. Outcomes vs shock index (SI; SI was model as cubic splines) truncated at SI value of 1.5 to remove extreme outliers.

Figure S4. Outcomes vs shock index (SI; SI was model as cubic splines).

Figure S5. Outcomes vs shock index (SI; SI was model as cubic splines).

Figure S6. Outcomes vs shock index (SI; SI was model as cubic splines).

Click here for additional data file.