Article Text

Download PDFPDF

Shock index as a predictor of hospital admission and inpatient mortality in a US national database of emergency departments
  1. Nour Al Jalbout1,
  2. Kamna Singh Balhara1,
  3. Bachar Hamade2,
  4. Yu-Hsiang Hsieh1,
  5. Gabor D Kelen1,
  6. Jamil D Bayram1
  1. 1 Department of Emergency Medicine, Johns Hopkins University, Baltimore, Maryland, USA
  2. 2 Department of Critical Care, University of Pittsburgh Department of Medicine, Pittsburgh, Pennsylvania, USA
  1. Correspondence to Dr Jamil D Bayram, Department of Emergency Medicine, Johns Hopkins University, Baltimore, MD 21209, USA; jbayram1{at}


Study objectives The shock index (SI), defined as the ratio of the heart rate (HR) to the systolic blood pressure (BP), is used as a prognostic tool in trauma and in specific disease states. However, there is scarcity of data about the utility of the SI in the general emergency department (ED)population. Our goal was to use a large national database of EDs in the United States (US) to determine whether the likelihood of inpatient mortality and hospital admission was associated with initial SI at presentation.

Methods Data from the National Hospital Ambulatory Medical Care Survey were retrospectively reviewed to obtain a weighted sample of all US ED visits between 2005 and 2010. All adults >18 years old who survived the ED visit were included, regardless of their chief complaint. Likelihood ratios (LR) were calculated for a range of SI values, in order to determine SI thresholds most predictive of hospital admission and inpatient mortality. +LRs >5 were considered to be clinically significant.

Results A total of 526 455 251 adult patient encounters were included in the analysis. 56.9% were women, 73.9% were white and 53.2% were between the ages of 18 and 44 years. 88 326 638 (15.7%) unique ED visits resulted in hospital admission and 1 927 235 (2.6%) visits resulted in inpatient mortality. SI>1.3 was associated with a clinically significant increase in both the likelihood of hospital admission (+LR=6.64) and inpatient mortality (+LR=5.67). SI>0.7 and >0.9, the traditional cited cut-offs, were only associated with marginal increases (+LR= 1.13; 1.54 for SI>0.7 and +LR=1.95; 2.59 for SI>0.9 for hospital admission and inpatient mortality, respectively).

Conclusions In this largest retrospective study to date on SI in the general ED population, we demonstrated that initial SI at presentation to the ED could potentially be useful in predicting the likelihood of hospital admission and inpatient mortality, which could help guide rapid and accurate acuity designation, resource allocation and disposition.

  • triage
  • death/mortality
  • hospitalisations

Statistics from

Key messages

What is already known on this subject

  • The shock index is a valuable prognostic tool in specific patient populations.

  • There is a paucity of studies on the utility of the shock index in the general ED population.

What this study adds

  • The traditional shock index values of >0.7 and >0.9 cited in the literature may not be useful in the general ED population.

  • In our largest to date retrospective study of the general ED population, a shock index of >1.3 may predict hospital admission and inpatient mortality.

  • This could help supplement the triage system by guiding accurate acuity designation and rapid patient disposition.


The shock index (SI), first described in 1967 by Allgöwer and Burri, is a clinical metric obtained by dividing the heart rate (HR) by the systolic BP (SBP), giving clinicians an early indication of the deterioration of a disease state.1–3 This can be explained by the inverse relationship of SI to the left ventricular stroke volume, with higher values predicting worse cardiac function.1 Although the term SI was originally investigated in shock states, it has been studied as a prognostic tool in other conditions of critical illness not particularly in shock. In addition to the trauma literature where an SI>0.9 has been described as an early predictor of haemorrhagic shock, mortality and need for transfusion, SI has also been studied as a predictor of haemodynamic instability, morbidity and mortality in a variety of disease states.4–12 In a study that included admitted patients with pneumonia, SI≥1.0 was associated with a higher likelihood of dying within 6 weeks from admission.8 Another study of septic patients in the ED also showed that SI≥1.0 was associated with an elevated risk for 28-day mortality, and a study of patients with active gastrointestinal bleeding had a positive association between an elevated SI and angiographic visualisation of extravasation.9 10 An additional investigation of patients admitted after ST-elevation myocardial infarction showed a 1.6-fold increase in the risk of cardiac adverse outcomes within 7 days with a SI≥0.7.11

With SI demonstrating its utility across a large variety of pathophysiological states, it may be especially useful in a diverse, undifferentiated setting such as the ED. However, literature regarding its optimal predictive cut-off and utility in the general ED patient population with a range of undifferentiated chief complaints from low acuity to critically ill is limited to very small or single-site studies including our prior single-centre study, showing an association between an SI>1.2 and a significant likelihood of hospital admission and inpatient mortality.2 13 While there is a significant number of early warning scores used in emergency medicine practice, including various five-tiered triage systems, these scores are complex with only moderate validity and reliability.14 15 In comparison, SI is a simple, one-step easy to calculate bedside metric that only relies on vital signs, available to all practice settings. We elected to study its utility as an early predictor of resource utilisation and clinical deterioration in the general ED population.

The main objective of this study was to use the National Hospital Ambulatory Medical Care Survey (NHAMCS), which is the largest national database of EDs in the United States, to determine whether the probability of hospital admission and inpatient mortality was associated with the initial SI value at presentation of the diverse patient population to the ED. We planned to derive a predictive SI threshold specific to this patient population by evaluating the performance of SI at different cut-offs, followed by subpopulation analysis based on gender, race and age.


Study design

This retrospective study examined a cohort of ED visits from 2005 to 2010 from the NHAMCS. A weighted sample including all US ED visits reported in the survey was used. The study was granted exempt status by our institution’s review board (IRB00151493).

NHAMCS is a nationally representative survey conducted by the Centers for Disease Control and Prevention and the National Center for Health Statistics.16 It uses probability sampling of ED visits in the 50 states and the District of Columbia, excluding visits to federal, military and Veterans Administration hospitals, and has an estimated 120 million encounters annually. NHAMCS personnel collect data from ED visit medical records during a randomly assigned 4-week period while being monitored by NHAMCS field representatives. NHAMCS staff members independently check 10% of the data for accuracy with an error rate ranging from 0.3% to 0.9% for various items on the survey. Between the years 2005 and 2010, the number of hospitals that agreed to participate in the survey ranged from 352 to 389. The data from NHAMCS were publicly available with free access between the years of 2005 and 2010, which constituted our study period.

Outcome measures and data analysis

Our primary outcome measures were hospital admission and inpatient mortality, as they constitute meaningful patient outcomes and can potentially serve as surrogate markers for medical acuity and increased need for resource mobilisation. All ED visits for patients over the age of 18 years were included. Encounters without sufficient data for SI calculation were excluded. Encounters that resulted in death in the ED were also excluded, as the primary outcome of interest was inpatient mortality. The initial recorded set of vital signs including HR and SBP at presentation for each visit, the visit outcome and basic patient demographics, including age, gender and race, were used for data analysis. We extracted the first recorded HR and SBPs to calculate the initial SI at presentation for each encounter, and then calculated the likelihood ratios (LR) for a broad range of thresholds of SI for both outcomes. +LRs were used since they represent statistically robust measures of diagnostic accuracy that are independent of pretest probability.17 The LR values for SI of more than 1.6 for hospital admission and more than 1.4 for inpatient mortality were statistically unstable due to increased variance in the setting of small sample size, and thus these LR values are reported but excluded from the analysis to avoid invalid estimates. Subgroup analyses stratified by race (white or non-white), gender (women/men) and age group (18–44, 45–64 or ≥65 years) were performed for the outcome of hospital admission, as prior data suggest higher likelihood of admission based on age, race and gender.18–21 The above chosen age categories were similar to those used in prior epidemiological studies, particularly national health statistics surveys.22 The relatively small sample size associated with inpatient mortality in most of the subgroups (original weighted sample size <30 in each group) precluded stratification by demographics due to the possibility of generating invalid estimates.

+LR values of >5 and >10 were considered to indicate moderate and large increases in the likelihood of the primary outcomes, respectively.17 +LR ≥5 was chosen to represent a clinically significant value for our study purposes.



A total of 567 994 402 distinct weighted adult patient ED visits between 2005 and 2010 were identified, of which 41 539 151 (7.3%) were excluded due to incomplete records and 660 207 (0.12%) died in the ED. Of the 526 455 251 weighted patient visits included for analysis, 56.9% were women, 73.9% were white and 53.2% were between the ages of 18 and 44 years (table 1).

Table 1

Demographics and disposition of 526 455 251 ED patient visits

Main results

A total of 88 326 638 (15.7%) unique ED visits resulted in inpatient admissions; 1 927 235 (2.6%) visits resulted in inpatient mortality (table 1). For hospital admissions, 11 153 882 patient encounters had SI>0.9 while 1 539 198 had an SI>1.3. For inpatient mortality, 654 531 encounters had an SI>0.9 while 185 698 had an SI>1.3. The frequencies of both outcomes at every SI cut-off are reported in tables 2 and 3.

Table 2

Likelihood of admission by shock index

Table 3

Likelihood of inpatient mortality by shock index

In our study population of general ED patients, initial SI>1.3 was associated with a moderate increase in the likelihoods of both hospital admission and inpatient mortality, with +LRs of 6.64 (95% CI 6.62 to 6.65) and 5.67 (95% CI 5.64 to 5.69), respectively (tables 2 and 3). The sensitivity for both outcomes at SI>1.3 was very low while the specificity ranged from 98.3% to 99.7% for inpatient mortality and hospital admission, respectively (tables 2 and 3). At SI>1.3, 15.64% of patients were misclassified to have an outcome of hospital admission, and 4.03% were misclassified to have inpatient mortality (tables 2 and 3). SI>0.7 and SI>0.9, which represent frequently cited SI thresholds,1–3 were not significant predictors of the likelihood of hospital admission and inpatient mortality (tables 2 and 3).

For hospital admission, +LR for SI>0.7 was the lowest at 1.13 (95% CI 1.13 to 1.14). At SI>0.9, +LR was 1.95 (95% CI 1.95 to 1.95) and at SI below 0.7 +LR ranged from 1.21 to 2.3. Moderate increases in likelihood of admission (+LR>5) were seen at SI values of 1.2 and greater. For inpatient mortality, +LR for SI>0.7 was 1.54 (95% CI 1.54 to 1.55) and +LR for SI>0.9 was 2.59 (95% CI 2.59 to 2.60).

+LR for hospital admission at SI>1.3, when stratified by age, gender and race, was higher for non-white patients (9.09, 95% CI 9.05 to 9.14) versus white patients (6.03, 95% CI 6.00 to 6.04) and higher for men (8.37, 95% CI 8.34 to 8.40) versus women (5.29, 95% CI 5.26 to 5.30) (tables 4 and 5). The likelihood of admission in all three age groups increased with increasing SI thresholds, with highest +LR for admission for those aged 45–64 years (8.07, 95% CI 8.02 to 8.10) compared with other age groups (table 6). Subgroup analysis on inpatient mortality was not performed due to statistical limitations, as the sample size is too small to make a valid estimate of LRs for subgroup data analysis. However, there was an inclination towards higher LR with higher SI values.

Table 4

Likelihood of admission by shock index, stratified by race

Table 5

Likelihood of admission by shock index, stratified by gender

Table 6

Likelihood of admission by shock index, stratified by age


This large multicentre study shows that an initial SI>1.3 at presentation to the ED is associated with significant increase in the likelihoods of both hospital admission and inpatient mortality. Our current retrospective study included 526 455 251 weighted ED patient visits with a wide spectrum of disease states across the US from NHAMCS data sets, of which 88 326 638 were admitted and 1 927 235 died as inpatients.

Although the percentage of patients presenting to the ED with SI>1.3 is relatively small compared with the entire cohort included (0.52% and 0.27% for patients at risk for admission and inpatient mortality, respectively), the absolute number of these patients remains significantly high at 2.8 million and 1.4 million, respectively. Importantly, SI>1.3 is where both probabilities, collectively, of hospital admission and inpatient mortality display a significant increase with high +LR, providing clinicians with a potential threshold for earlier interventions. These findings are in line with the results of our prior single-site study of 58 336 ED patient encounters, where Balhara et al demonstrated a significant association with SI>1.2 and no association with a SI>0.9.13 The multicentre cohort included in our current study helps in supporting the conclusions of Balhara et al on a larger more robust scale.

SI has been previously described as a predictive metric of mortality and morbidity in specific patient populations, at cut-offs (0.7, 0.9) lower than our findings: Montoya et al evaluated more than 1000 patients in a national trauma registry, showing higher mortality in 24 hours for patients with SI>0.97; a study conducted by Reinstadler et al demonstrated that patients with recent myocardial infarction had higher myocardial damage with elevated SI above 0.62 and independently associated with major adverse cardiac events at 12 months12; Wira et al concluded that 38.6% of septic patients presenting to the ED with sustained SI elevation required vasopressors within 72 hours of ED admission, compared with 11.6% of those without a sustained SI elevation (OR 4.42; p<0.0001)23; Rady et al conducted a prospective study of 275 general adult ED patients, and concluded that SI>0.9 could identify acute critical illness and was associated with higher total admissions and intensive care unit admissions.2 These lower thresholds were unable to distinguish those at risk for admission or mortality. This may be due to the different types of cohorts studied as our cohort included all comers to the ED with wide spectrum of chief complaints and not patients with specific disease states.

For hospital admission, +LR stratified by race and gender for SI>1.3 was highest among non-white and male patients (9.09 and 8.37, respectively). While it is difficult to pinpoint specific causes for these differences without further substratifying for possible confounders, mainly age and comorbidities within gender and race, prior data support higher admission rates for non-white patients for certain disease states.18 19 There are also limited data to support higher admission rates for men in stroke, myocardial infarction and heart failure.18–20 However, admission rates for women are generally higher in national surveys.21 The higher LRs for admissions noted for men in our study are likely multifactorial in origin and could be related to the distribution of ages and comorbidities within each gender.

Within each age group, +LR for admission increased as SI increased. However, +LR for admission at SI>1.3 was lower (4.6) in the elderly age group (≥65 years) compared with +LRs of 6.79 and 8.07 in 18–44 year-olds and 45–64 year-olds, respectively. A possible explanation for this discrepancy could be that the threshold to admit elderly patients may be lower than that for other age groups, regardless of SI, due to multiple comorbidities, possible polypharmacy and limitations in activities of daily living. The use of antihypertensive medications including beta blockers in this population may also contribute to lower +LR in this group. Kristensen et al recently studied the association between SI and 30-day mortality and demonstrated that old age, hypertension and beta blocker or calcium channel blocker use weakened the association between SI (≥1) and 30-day mortality.24 Our results vary potentially due to the difference in our primary outcome, the multicentre larger patient population studied and the lack of control for medication administration.

Although there is a wide range of triage tools used in the ED, including the Emergency Severity Index (ESI) and the Manchester Triage System (MTS), SI can potentially play a role in improving the performance of these tools. Both ESI and MTS have been widely adopted due to numerous strengths including linkage to anticipated ED resource utilisation. While both have demonstrated good predictive capacities for admission, their ability to predict mortality is best at cases classified into higher urgency categories.14 15 25 Importantly, they are both subjective and prone to practice variation and individual judgement with inter-rater reliability reported to range from k=0.46 to 0.91 for ESI and k=0.55–0.78 for MTS.14 15 25 SI is an objective easily calculated metric that can be an adjunct to these ED-specific scoring systems and can potentially increase their predictive accuracy. Further studies are needed to compare the SI with other scoring systems and to study its additive impact on the performance of these scores in predicting admission and mortality.


Our study has certain limitations. This was a retrospective study using a national data cohort, which is representative rather than comprehensive, with 7.3% of the visits being excluded due to incomplete records. Although the analysis included a large number of eligible patients, the actual number of patients with an SI>1.3 for both outcomes, potentially affected by the use of SI as a triage tool, is much lower. Importantly, the primary diagnosis was not uniform across patient encounters and could be an important confounder. A group of patients might have had disease states that were not a result of low stroke volume, thus potentially skewing our results and explaining the unique SI cut-off in our study. We also could not account for the secular trends in the care of shock or sepsis in the ED, which may heavily impact the outcome of patients with increased SI. Moreover, the available data in the survey do not contain additional patient characteristics, such as comprehensive medication profiles, medical comorbidities, ED interventions and medications, all which are possible confounders contributing to the outcome.16 Notably, we were unable to stratify our analysis to exclude patients on beta blockers and other antihypertensives, whose SI at presentation may have differed from what is physiologically anticipated, mainly among elderly patients. Level of care for admitted patients was not available and analysis could not be stratified by hospital type, which can provide different levels of care to patients and thus affecting overall outcome. Owing to the relatively smaller sample size of patients with inpatient mortality, subgroup analysis for this outcome was not included due to the possibility of generating statistically unstable estimates, limiting further identification of the discriminatory ability of SI among different groups. Finally, inpatient mortality may also have been subject to confounding, as death may not have been a direct consequence of the presenting condition at the time of the ED visit, but a result of interventions or new developments in the course of the disease during the hospital stay. However, our results can inform future prospective studies with hypotheses for validation.


SI is a metric that is easily available to all ED providers. Our study of 526 455 251 ED visits in a large nationally representative cohort suggests that a SI>1.3 at presentation may be helpful as one of the predictors for admission and inpatient mortality for all patients early in their ED course, potentially assisting in prompt resource allocation and disposition. While this retrospective study has its own limitations, SI>1.3 was consistently associated with an increased likelihood of admission and inpatient mortality across different age groups, gender and race compared with other SI thresholds including the often cited cut-offs (0.7 and 0.9). Further research including prospective studies is needed to validate our SI cut-off, compare its accuracy with other widely used early warning scores and assess its value as a clinical decision tool supplementing the existing triage systems.


All coauthors have had the opportunity to review the final manuscript and have provided their permission to submit. All who contributed to the manuscript are listed as authors. The manuscript is an honest, accurate and transparent account of the study being reported and no important aspects of the study have been omitted.



  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Disclaimer We received no support from any organization for the submitted work. There are no relationships or activities that could appear to have influenced the submitted work.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Patient consent for publication Next of kin consent obtained.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.