Association of triage hypothermia with in-hospital mortality among patients in the emergency department with suspected sepsis

Abstract

Purpose

To identify if triage hypothermia (<36.0 °C) among emergency department (ED) encounters with sepsis are independently associated with mortality.

Methods

We analyzed data from a multi-stage probability sample survey of visits to United States EDs between 2007–2015, using two inclusion approaches: an explicit definition based on diagnosis codes for sepsis and a severe sepsis definition, combining evidence of infection with organ dysfunction. We used multivariable regression to determine an association between hypothermia and in-hospital mortality.

Results

Of 1.2 billion ED encounters (95% confidence interval [CI] 1.0–1.3 billion), 3.1 million (95% CI 2.7–3.5 million) met the explicit sepsis definition; 7.4% (95% CI 7.4–7.5%) had triage hypothermia. The adjusted odds ratio (aOR) for hypothermia for in-hospital mortality was 6.82 (95% CI 3.08–15.22). The severe sepsis definition identified 3.5 million (95% 3.1–4.0 million) encounters; 30.3% (95% CI 25.0–34.6%) had triage hypothermia. The aOR for hypothermia with mortality was 4.08 (95% CI 2.09–7.95). Depending on sepsis definition, 78.1–84.4% had other systemic inflammatory response syndrome vital sign abnormalities.

Conclusion

Up to one in three patients with sepsis have triage hypothermia, which is independently associated with mortality. 10–20% of patients with hypothermic sepsis do not have other vital sign abnormalities.

Introduction

Sepsis is a major cause of morbidity and mortality across all ages, accounting for 3.2% of all hospital stays and 6.2% of hospital expenses annually, according to the Agency for Healthcare Research and Quality in 2016 [1]. The annual incidence of sepsis is estimated to be 3 cases per 1,000 population, with 215,000 deaths per year in the U.S. [2]. Early identification of patients with potential sepsis may help to reduce sepsis-associated morbidity and mortality [3] and prompt recognition and identification of patients with sepsis is an important goal in the emergency department (ED) setting [4–6]. Determining presenting patient characteristics that are associated with sepsis-related mortality would aid in the development of sepsis recognition tools, help to optimize triage workflows and could be incorporated into clinical trial enrollment criteria.

Many screening tools for children and adults utilize definitions developed around patients with the systemic inflammatory response syndrome (SIRS) [7,8]. The value of SIRS-based vital signs to screen patients with sepsis remains of importance given concerns that Sequential Organ Function Assessment based screening modalities lack sensitivity, as demonstrated in a meta-analysis including 52,849 patients [9]. Temperature criteria for both pediatric and adult SIRS include both hyperthermia (core temperature greater than 38 °C in adults or 38.5 °C if pediatric) and hypothermia (core temperature less than 36 °C) [7,8]. Among patients admitted to the intensive care unit, greater mortality has been observed among patients experiencing hypothermia compared to patients experiencing only normothermia or hyperthermia [10–13]. Importantly, patients presenting with hypothermia may not manifest other vital signs abnormalities compatible with SIRS and therefore may be missed by SIRS-based sepsis screening tools [14].

No investigation to date has evaluated whether hypothermia at the time of presentation to the ED is associated with outcomes. In this study, we aim to identify if triage-based hypothermia is an independent predictor of mortality after accounting for other vital sign abnormalities in a nationally representative database of ED encounters. Secondly, we aim to identify the prevalence of other vital sign abnormalities among hypothermic patients with sepsis.

Methods

Data source

We performed a cross-sectional analysis of the National Hospital Ambulatory Medical Care Survey (NHAMCS), a nationally representative sample survey of visits to EDs to evaluate if hypothermia on triage vital signs is an independent predictor of mortality among patients in ED with sepsis [15]. NHAMCS is conducted annually by the Centers for Disease Control and Prevention National Center for Health Statistics (NCHS) and is performed as a cross-sectional probability sample survey of visits to EDs of non-federal and short-stay hospitals in the United States. NHAMCS utilizes a probability sample which operates at four stages: 1) geographic regions as primary sampling units, 2) hospitals within primary sampling units, 3) EDs, and 4) encounters within EDs. Each deidentified record is assigned a weight that is equal to the inverse of its probability of being included in the sample. Common data elements, when present across multiple years, may be combined for the purposes of increasing sample size. Research performed using NHAMCS is approved by the NCHS Ethics Review Board. For this study, we used NHAMCS public-use data for the period 2007–2015.

Population of interest

Our population of interest was patients with sepsis. For the purposes of this analysis, we excluded patients classified as dead on arrival and patients with missing temperature data. We used two approaches to identify patients with sepsis, using definitions previously identified in the literature. In addition, we investigated an exploratory analysis based on a definition derived from SIRS.

In our first criteria for sepsis, we used an explicit (ICD-9 based) definition of sepsis as previously developed by Filbin, et al [16]. We identified the presence of any of the following International Classification of Disease, ninth revision (ICD-9) diagnosis codes among included encounters: 038 (septicemia), 995.91 (sepsis), 995.92 (severe sepsis), or 785.52 (septic shock). In the NHAMCS dataset, up to three ED diagnosis codes fields could have entries. Encounters were included in this analysis if an explicit ICD-9 code for sepsis was present in any three of these fields. Only ICD codes placed during the ED encounter were considered.

In our second definition of sepsis, we used a definition of severe sepsis using criteria published by Wang, et al [17,18]. This definition, modified for use with NHAMCS from criteria previously established by Angus, et al [2], required evidence of organ dysfunction with infection. Organ dysfunction was defined as an ICD-9 code consistent with organ dysfunction [18], procedure of endotracheal intubation, or hypotension for age [7,19]. Evidence of infection based on ICD-9 coding required an ICD-9 code consistent with infection [18] or the presence of fever (>38.5 °C or >38.0 °C for pediatric and adult encounters, respectively) or hypothermia (<36 °C). As with the explicit definition of sepsis, only diagnosis codes placed during the ED encounter were considered.

Outcome and predictor of interest

Our outcome of interest was in-hospital mortality, which was considered to have occurred if the NHAMCS recorded death occurring while in the ED or death occurring following hospital admission. Our predictor of interest was hypothermia in triage vital signs.

Variables

We abstracted the following variables for the development of multivariable models. Age was stratified into the following groups: pediatric (0–11 years), adolescent (12–17 years), adult (18–65 years), and older adult (>65 years). Race was provided as White, Black, and other. Ethnicity was provided as Hispanic and Non-Hispanic. Expected source of payment was reclassified from the original dataset into the following subgroups: private insurance, government, self-pay/no charge, and other/unknown. Disposition was reclassified as discharged, admitted/transferred, died in ED, or other. We also abstracted the season of encounter (based on meteorological season), whether encounters presented to hospitals in metropolitan status areas, whether they arrived by emergency medical services, and year of presentation. Triage temperature was classified as euthermic, hypothermia, fever, or absent documentation, for which encounters were excluded. The NHAMCS dataset does not document route of temperature measurement. We applied criteria for SIRS vital signs using temperature, respiratory rate, and heart rate as established by the American College of Chest Physicians and Society of Critical Care Medicine Consensus Conference (ACCP/SCCM) [7] for adult encounters (≥18 years) and by the International Pediatric Sepsis Consensus Conference [8] for pediatric encounters (Supplementary Table 1). Blood pressure was defined using ACCP/SCCM criteria and American Heart Association criteria for pediatric encounters [7,19]. All vital signs, including temperature, were limited to those acquired at triage, which is performed at the beginning of an ED encounter. We did not consider subsequent vital signs as our primary aim was to assess the role of triage vital signs on outcome. Furthermore, additional vital signs are inconsistently available within the NHAMCS dataset. All imputation steps, performed for race and ethnicity, were performed by NCHS and are available within the public use dataset.

Statistical analysis

We reported estimates using survey-weighting procedures accounting for the NHAMCS sampling design. In keeping with guidelines established by NCHS, estimates with fewer than 30 records or which have a relative standard error greater than 30% were considered unstable. Demographics were provided as estimates or proportions, using 95% confidence intervals (CI). We used multivariable logistic regression to identify if hypothermia was an independent predictor of mortality in patients after adjusting for other vital signs collected at triage in each of the cohorts identified using different definitions of sepsis. Univariable logistic regression was used to identify if each variable was possibly associated with mortality and variables with p<0.10 were included in the multivariable logistic regression models. For the final multivariable model, p<0.05 defined statistical significance. We evaluated rates of other vital sign abnormalities for patients with hypothermia when using each definition of sepsis. Analyses were conducted using the survey package (version 3.36) [20] in R, version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria, https://www.R-project.org/).

Sensitivity analysis

We repeated our analysis using the explicit definition for sepsis, but additionally excluding patients with death in the ED and patients with ED ICD-9 diagnosis codes of 427.5 (cardiac arrest), 427.41 (ventricular fibrillation), or 427.42 (ventricular flutter) [21]. We repeated the multivariable analysis for an outcome of death following hospital admission.

Exploratory analysis

We performed an exploratory analysis using a combination of SIRS with evidence of infection (SIRS estimate) to create an alternative definition of sepsis. As the white blood cell count (WBC) is not available in NHAMCS, this was performed using methods derived from Horeczko et al [22,23]. Encounters having two vital sign criteria were considered to have SIRS and encounters having no vital sign criteria were considered not to have SIRS. For encounters meeting one vital sign criteria, we randomly assigned 50% of such encounters as meeting the WBC count criteria for SIRS. This was referred to as the “medium” SIRS estimate. Of patients with SIRS, those having confirmed or suspected infection on the basis of diagnosis coding as established by Angus et al [2], were included as having sepsis. We then assumed that all patients having one SIRS criteria would fulfill the WBC criterion (the “maximum” SIRS estimate). We then assessed our outcomes in a model in which no patients meeting one SIRS criteria would fulfill the WBC criterion (the “minimum” SIRS estimate). For each analysis, patients with SIRS and an ICD-9 code corresponding for an infection were classified as sepsis. We performed logistic regression to assess if our predictor of interest was associated with mortality, after adjustment with other vital signs [2].

Results

Study inclusion

During the included timeframe, an estimated 1.2 billion encounters (95% C 1.0–1.3 billion) occurred. Of these, <0.1% of encounters were classified as dead on arrival and 5.5% lacked triage temperature data and were excluded from further analysis. The included cohort was 94.5% of the total estimated patient encounters in NHAMCS for these years and represented 1.1 billion total encounters (95% CI 1.0–1.2 billion). Demographics of the population overall and by sepsis definition are provided in Supplementary Table 2.

Explicit definition of sepsis

Using the explicit definition of sepsis, we identified that 0.3% of patients, or 3.1 million (95% CI 2.7–3.5 million) patients met the criteria for sepsis. This figure represented data from 724 raw counts within the NHAMCS dataset. Among patients with sepsis, most were euthermic, had normal respiratory rates and blood pressures, but there was a high prevalence of tachycardia. 7.4% (95% CI 7.4–7.5%) of patients were hypothermic when using this definition of sepsis and 13.2% (95% CI 13.1–13.2%) of encounters ended with in-hospital mortality.

Severe sepsis definition

Using the definition for severe sepsis, 0.3% of patients, or 3.5 million (95% 3.1–4.0 million) patients, met the criteria for sepsis from 787 raw encounters, of which 127 overlapped with the explicit sepsis definition. A higher proportion of patients in this group had temperature instability and hypotension, though most had normal heart and respiratory rates. 30.3% (95% CI 25.0–34.6) had hypothermia in this group, and in-hospital mortality was 10.8% (95% CI 8.1–13.5%).

Association of triage hypothermia with mortality

Using the explicit definition of sepsis, univariable analysis identified an association between mortality and variables of temperature, respiratory rate, and blood pressure. In multivariable analysis, hypothermia was independently associated with mortality after adjusting for covariates (Table 1). Using the severe sepsis definition, we identified a univariable association between mortality and all vital signs. After accounting for covariates, hypothermia was independently associated with mortality (Table 2).

Table 1.

Univariable and multivariable analysis including patients with an explicit definition of sepsis.

Variable	Unadjusted odds ratio (95% confidence interval)	P	Adjusted odds ratio (95% confidence interval)	P
Temperature
Euthermia	Ref	--	Ref	--
Fever	0.66 (0.31–1.41)	0.281	0.62 (0.27–1.41)	0.255
Hypothermia	6.03 (2.78–13.10)	<0.001	6.82 (3.06–15.22)	<0.001
Heart rate
Normal	Ref	--
Tachycardia	0.78 (0.44–1.40)	0.411
Absent documentation	0.38 (0.07–1.96)	0.248
Respiratory rate
Normal	Ref	--	Ref	--
Tachypnea	2.29 (1.29–4.08)	0.005	2.73 (1.45–5.15)	0.002
Absent documentation	1.84 (0.36–9.47)	0.466	3.19 (0.55–18.41)	0.197
Blood pressure
Normotensive	Ref	--	Ref	--
Hypotensive	2.65 (1.31–5.33)	0.007	2.68 (1.31–5.49)	0.008
Absent documentation	0.60 (0.15–2.43)	0.472	0.32 (0.08–1.29)	0.110

Table 2.

Univariable and multivariable analysis including patients using a definition of severe sepsis.

Variable	Unadjusted odds ratio (95% confidence interval)	P	Adjusted odds ratio (95% confidence interval)	P
Temperature
Euthermia	Ref	--	Ref	--
Fever	1.77 (0.77–4.03)	0.178	1.27 (0.56–2.89)	0.564
Hypothermia	3.28 (1.73–6.24)	<0.001	4.08 (2.09–7.95)	<0.001
Heart rate
Normal	Ref	--	Ref	--
Tachycardia	2.05 (1.12–3.75)	0.020	2.53 (1.26–5.09)	0.009
Absent documentation	2.01 (0.57–7.06)	0.275	1.65 (0.38–7.20)	0.505
Respiratory rate
Normal	Ref	--	Ref	--
Tachypnea	2.67 (1.46–4.89)	0.002	2.65 (1.42–4.93)	0.002
Absent documentation	3.91 (1.4–10.93)	0.010	2.71 (1.05–6.96)	0.040
Blood pressure
Normotensive	Ref	--	Ref	--
Hypotensive	0.61 (0.35–1.07)	0.083	0.72 (0.41–1.26)	0.251
Absent documentation	2.32 (0.54–10.04)	0.260	1.8 (0.55–5.93)	0.333

Abnormal vital signs among patients with sepsis

Using either definition of sepsis, the majority of patients had normal heart and respiratory rates (Table 3). With an explicit definition of sepsis, 78.1% (95% CI 65.2–91.0%) had at least one other documented abnormal SIRS vital sign. Using a severe sepsis definition, 84.4% (95% CI 78.6–90.1%) of hypothermic patients had at least one other documented abnormal SIRS vital sign. Among patients meeting an explicit definition of sepsis, 60.8% (95% CI 44.6–77.0%) presented during the fall or winter months, as did 56.4% (95% CI 47.8–64.9%) of the population identified using the severe sepsis definition.

Table 3.

Percentages (with 95% confidence intervals) of triage vital sign findings among encounters meeting sepsis criteria and with hypothermia

Variable	With sepsis, explicit definition	With severe sepsis
	Percent (95% confidence interval)	Percent (95% confidence interval)
Heart rate
Normal	50.7 (33.7–67.7)*	69.3 (62.4–76.3)
Tachycardia	45.7 (28.7–62.7)*	28.2 (21.2–35.1)
Absent documentation	3.6 (0.0–8.6)*	2.5 (0.8–4.2)*
Respiratory rate
Normal	57.0 (40.2–73.8)	71.1 (64.2–78.1)
Tachypnea	43.0 (26.2–59.8*	24.2 (17.6–30.8)
Absent documentation	0.0 (0.0–0.0)*	4.7 (1.6–7.7)*
Blood pressure
Normotensive	66.0 (49.9–82.1)	33.9 (26.4–41.5)
Hypotensive	22.2 (8.7–35.7)*	62.5 (54.7–70.4)
Absent documentation	11.7 (0.0–24.2)*	3.5 (0.5–6.6)*

^*Derived from a low number of raw counts, which limit reliability of estimates as per National Center for Health Statistics guidelines.

Sensitivity analysis

A sensitivity analysis using an explicit definition of sepsis but additionally excluding patients with death or cardiac arrest in the ED demonstrated similar results to the primary analysis (Supplementary Table 3).

Exploratory analysis

Of 254,043 raw encounters in the dataset, 134,329 (52.9%) had no SIRS vital signs, 97,513 (38.4%) had one SIRS vital sign, 19,572 (7.7%) had two SIRS vital signs, and 2,629 (1.0%) had three SIRS vital signs. After random assignment of WBC count meeting the SIRS criteria in 50% of encounters with one SIRS vital sign and accounting for sampling weights, we identified SIRS in 28.0% of included encounters. 17.0% of encounters met ICD-9 criteria for infection. A diagnosis of sepsis was identified in 5.8% (95% CI 5.8–5.8%), 2.5% (95% CI 2.4–2.6%), 9.4% (95% CI 9.1–9.6%) meeting the medium, minimum, and high incidence rates estimates for sepsis, respectively. Proportions of in-hospital mortality varied between 0.8–2.0%, with the highest in the minimum-incidence SIRS group (Supplementary Table 4). For each approach, hypothermia was independently associated with an outcome of in hospital mortality (Supplementary Table 5).

Discussion

In this large, nationally representative sample of children and adults presenting to the emergency department meeting criteria for sepsis, hypothermia was independently associated with increased adjusted odds of mortality. We demonstrated an association between triage hypothermia and mortality among encounters in the ED with suspected sepsis regardless of methodology used to identify potential cases: an explicit sepsis definition, a severe sepsis definition, and, in exploratory analyses, sepsis estimates identified using SIRS criteria. Using an explicit definition of sepsis, the adjusted effect size of the association between hypothermia and mortality was greater than the observed effect of presenting tachypnea and hypotension with mortality in a multivariable model. These findings have potentially important implications for risk stratification of patients with suspected infection and possible sepsis in the ED setting. Hypothermia may also be an important characteristic when determining inclusion criteria for sepsis-related clinical investigations in the ED.

A growing number of studies have identified hypothermia as associated with mortality among adults with sepsis. In an earlier meta-analysis of forty-two studies of patients with sepsis, temperature was negatively correlated with mortality, with the highest proportion of mortality observed among patients with documented hypothermia [24]. To the best of our knowledge, our study is the largest to examine hypothermia relative to other presenting vital sign aberrations in the context of sepsis at time of ED triage. A recent analysis of adult patients with severe sepsis defined according to Sepsis-2 criteria admitted to fifty-nine ICUs in Japan identified an odds ratio of 1.76 (95% CI 1.13–2.73) for mortality among patients with hypothermia as compared to those with fever [25]. Notably, hypothermia was associated with lower utilization of sepsis bundles, suggesting this symptom may be less recognized as an indicator of sepsis. In our study the odds of mortality were four to six-fold greater in multivariable analysis for hypothermia as compared to hypotension. As a distinguishing feature of shock, hypotension may be more immediately recognizable as an ominous finding to clinicians and also be more readily addressable with both fluids and vasoactive medications compared to hypothermia, which may incite less alarm and could disproportionately occur in a frailer population. Notably, Kushimoto et al. noted that patients with hypothermia and sepsis were older and had a lower BMI compared to patients with sepsis and fever [25].

Both hypothermia and fever have been described as potentially adaptive mechanisms in response to acute infection. Fever may favorably stimulate the immune system, helping to fight systemic infection, albeit with the expense of additional energy expenditure. Alternatively, hypothermia has been postulated as an energy conserving strategy [26,27]. Pre-clinical studies have demonstrated an advantageous role for hypothermia as compared to fever in rats with severe endotoxemia [28–30]. Hypothermia has been associated with immune dysfunction in patients with sepsis, though some studies have called this mechanism into question. Wiewel et al. observed comparable levels of interleukin-6 (IL-6), IL-8 and IL-10 in septic adults with and without hypothermia, as well as comparable WBC responsiveness to lipopolysaccharide, as measured by TNF-alpha and IL-1 beta [31]. Whether the association between hypothermia and mortality observed in the present and previous studies is mediated by delayed recognition and treatment, unique inflammatory profiles, host characteristics, or a combination of factors is unclear.

Hypothermic patients may be more likely to be missed by existing screening tools that rely on vital sign abnormalities, even those that include hypothermia as part of the screen. This is suggested by our analysis demonstrating that a large proportion of patients with sepsis and hypothermia (up to one in five) do not have other vital sign abnormalities. Sepsis screening tools that rely on the presence of two or greater SIRS vital signs may fail to adequately identify hypothermic patients, despite the potentially greater risk for worse outcomes among these patients [32–35]. Our data support the need for further investigation to develop sepsis screening tools that better incorporate the relative importance of hypothermia as an indicator of patient state, with the aim of earlier recognition of potentially sicker patients.

A limitation of the present study lies in the use only of triage vitals without the availability of additional vital signs in the NHAMCS dataset. It is possible that some hypothermic patients may have been cold because of exposure, as suggested by a higher proportion (approximately 60% using either methodology) presenting during the fall and winter seasons. This might particularly be the case in younger patients who may be more vulnerable to exposure. However, the effect of hypothermia was not modified by the addition of season in multivariable models containing vital signs and season (results not shown). Furthermore, in exploratory analyses excluding patients with cardiac arrest in the ED, our results were similar, suggesting that moribund patients in the ED who were hypothermic did not solely account for the observed effect.

The findings from this study are subject to important limitations. The NHAMCS dataset may have errors with respect to data abstraction and coding [36,37]. Some investigators have noted, for example, care factors in the NHAMCS dataset which are incongruent with patient outcomes [38]. While our results demonstrated consistency in the relationship between predictors of interest and outcome, despite using both a narrow and broad definition of sepsis, the estimated numbers of patients identified with sepsis varied according to the definition. The accurate identification of patients with sepsis using administrative data remains a complex issue, with a variety of methods having been proposed [2,16–18,22,23]. We did not have laboratory values which would have allowed the development of criteria using more recent sepsis definitions [39]. The consistency of our findings regardless of method nonetheless demonstrate an important association of hypothermia with deleterious outcomes. The effect of the observed association between hypothermia and mortality may be mediated by the patient having already underwent a resuscitation, potentially in the prehospital setting and prior to vital signs acquisition. Our sensitivity analysis excluding patients with cardiac arrest demonstrated similar results, suggesting that this was not a major confounder. Finally, we were unable to provide reliable statistics of some parameters of interest because of low numbers in subgroups. Despite these limitations, the results from our study identify an important association between triage hypothermia and mortality and outcome among patients with sepsis, demonstrating an area in need of further investigation.

Conclusion

In this analysis from a national survey dataset, we identified that triage hypothermia was present in 6–8% of patients in the emergency department with sepsis, regardless of criteria for sepsis used. Hypothermia was an independent predictor of mortality in patients with sepsis in multivariable analysis. Further work is needed to prospectively assess the role of hypothermia the rapid screening of patients with sepsis in the emergency department.

Highlights.

• Using an explicit definition for sepsis, 6% of patients present with hypothermia.

• Hypothermia is an independent risk factor for in-hospital mortality.

• 10–20% of patients with hypothermic sepsis lack other triage vital sign abnormalities.

Abbreviations

aOR

adjusted odds ratio

CI

confidence interval

ED

emergency department

ICD

international classification of disease

NHAMCS

National Hospital Ambulatory Medical Care Survey

NCHS

National Center for Healthcare Statistics

SIRS

systemic inflammatory response syndrome

WBC

white blood cell