a b s t r a c t

Objective: To derive and validate a predictive model and novel emergency medical services (EMS) screening tool for severe sepsis (SS).

Design: Retrospective cohort study.

Setting: A single EMS system and an urban, public hospital.

Patients: Sequential adult, nontrauma, nonarrest, at-risk, EMS-transported patients between January 1, 2011, and December 31, 2012 were included in the study. at-risk patients were defined as having all 3 of the following criteria present in the EMS setting: (1) heart rate greater than 90 beats/min, (2) respiratory rate greater than 20 beats/min, and (3) systolic blood pressure less than 110 mm Hg.

Interventions: None. Measurements and Main Results: Among 66,439 EMS encounters, 555 met the criteria for analysis. Fourteen percent (n = 75) of patients had SS, of which 19% (n = 14) were identified by EMS clinical judgment. In-hospital mortality for patients with SS was 31% (n = 23). Six EMS characteristics were found to be predictors of SS: older age, transport from nursing home, Emergency Medical Dispatch (EMD) 9-1-1 chief concern category of “sick person,” hot tactile temperature assessment, low systolic blood pressure, and Low Oxygen Saturation. The final predictive model showed good discrimination in derivation and validation subgroups (areas under curves, 0.843 and 0.820, respectively). Sensitivity of the final model was 91% in the derivation group and 78% in the validation group. At a predefined threshold of 2 or more points, prehospital severe sepsis (PRESS) score sensitivity was 86%.

Conclusions: The PRESS score is a novel EMS screening tool for SS that demonstrates a sensitivity of 86% and a specificity of 47%. Additional validation is needed before this tool can be recommended for widespread clinical use.

(C) 2015

Introduction

Early recognition of severe sepsis is of paramount importance in order to facilitate timely initiation of lifesaving treatment. The goal of early recognition is supported by the most recent Surviving sepsis campaign guidelines as a means of maximizing Mortality benefit, primarily from early antibiotics and intravenous fluid therapy [1-3]. Despite best care practices, however, severe sepsis mortality remains as high as 18% to 30% [3,4]. Notably, the emergency medical services (EMS) care setting is a critical health care access point for up to 40% to 50% of patients with severe sepsis [5]. However, there are currently no standardized,

? Funding support: C.P. and this work are supported by the National Institutes of Health (T32GM095442 and UL1TR000454).

* Corresponding author at: Division of Pulmonary, Allergy, and Critical Care Medicine, Emory University School of Medicine, 615 Michael St, Suite 205M, Atlanta, GA 30322. Tel.: +1 404 712 2970; fax: +1 404 712 2974.

E-mail address: [email protected] (C.C. Polito).

evidence-based Screening tools available to enable EMS providers to accurately recognize severe sepsis in the field. This recognition is a crucial first step to the provision of both supportive and definitive therapy. As the point of first medical contact, EMS recognition has the potential to positively impact patient outcomes by allowing for the development of coordinated care systems that facilitate earlier treatment in the emer- gency department (ED). Notably, this type of strategy has proven benefi- cial for other life-threatening, time-sensitive conditions including cardiac arrest, heart attack, stroke, and trauma [6-8].

Small studies suggest that EMS recognition of severe sepsis may be beneficial in reducing time to initiation of antibiotic and intravenous fluid administration [9,10]. However, these reports have used screening tools that demonstrate low sensitivity to rule out sepsis, have not been formally validated, or require point-of-care (POC) diagnostic testing such as POC venous lactate that is not readily available to most EMS providers [10-12]. In addition, the need for a practical, reliable EMS screening tool is highlighted by the finding that EMS clinical judgment is only 17% sensitive for recognizing severe sepsis [12]. This finding

http://dx.doi.org/10.1016/j.ajem.2015.04.024

0735-6757/(C) 2015

can likely be explained by a variety of factors, most important of which are the following: the absence of a validated EMS screening tool; proto- cols derived from them; the complex, dynamic, and heterogeneous na- ture of the sepsis syndrome; and the low-resource nature of ambulances.

A practical, reliable EMS screening tool would allow for earlier recognition of this life-threatening condition and fuel efforts to develop coordinated EMS-ED care delivery systems improve sepsis out- comes through expediting definitive treatment. The aim of this study was to develop a simple, reliable EMS screening tool to aid first re- sponders in detecting severe sepsis. As such, we herein report the deri- vation and validation of the prehospital severe sepsis (PRESS) score.

Methods

Study design and patient population

A retrospective cohort study of all adult patients (age, >= 18 years) transported by Grady EMS to Grady Memorial Hospital was conducted between January 1, 2011, and December 31, 2012. All patients met a priori criteria for being at risk for having sepsis. The at-risk group was defined in order to both enrich the study population and to reflect the practical realities of how a severe sepsis screening tool might be used. This ap- proach is recommended when creating a predictive model and is similar to the approach used by EMS providers in Screening patients for stroke and heart attack, for example [13]. In these situations, screening is not performed on every EMS patient but rather is triggered by the presence of at-risk features such as unilateral weakness or chest pain, respectively. Patients were defined as being at-risk if all 3 of the following criteria were present in the EMS setting: (1) heart rate (HR) greater than 90 beats/min, (2) respiratory rate (RR) greater than 20 beats/min, and

(3) systolic blood pressure (SBP) less than 110 mm Hg. At-risk criteria were chosen based on modified systemic inflammatory response syn- drome criteria and previously published reports of the association between low EMS SBP and acute illness [1,14].

Patients were excluded if any of the following conditions were iden- tified by Emergency Medical Dispatch (EMD) call takers or by EMS initial impression on-scene: trauma injury, cardiac arrest, pregnancy, psychiat- ric emergency, or toxic ingestion. Exclusion criteria were based on

(1) existence of mature Care pathways for the condition, (2) a low like- lihood of severe sepsis being present, or (3) if the condition is not treat- ed in the main Grady ED. Patients were also excluded if the EMS patient care record could not be linked to a corresponding hospital encounter.

Study setting

Grady EMS manages the EMD of 9-1-1 medical calls for the portion of the City of Atlanta located in Fulton County, GA (88% of the city’s population). Of approximately 74,000 annual Ambulance transports by Grady EMS, approximately 30,000 are transported to Grady Memorial Hospital, a 900-bed, urban, public hospital. Emergency Medical Dispatch call takers use an integrated software system, ProQA (version 3.4.3.33; Priority Dispatch Corporation, Salt Lake City, UT), to query callers as well as categorize and prioritize caller information [15]. Emer- gency Medical Dispatch complaint categories are generated by caller answers to scripted questions supplied by the standardized EMD proto- col set. The “sick person” category is a standard classifier in the ProQA cardset and software system which is defined by Priority Dispatch as “a patient with a non-categorizable chief complaint who does not have an identifiable priority symptom” [15]. Please see the Appendix for a list of sick person nonpriority complaints.

Grady EMS ambulances are staffed with basic life support emergency medical technicians and advanced life support paramedics. The level of expertise for a given response is based on the acuity of the complaint, as provided by the caller. Information routinely captured during the on- scene evaluation and treatment phase of EMS care includes a chief con- cern-based patient history, an initial EMS impression, routine vital signs, physical examination, and a summary Clinical impression by EMS providers. The guidelines for arriving at these impressions are protocol- driven. Although temperature is not routinely measured, tactile tempera- ture assessment is performed. Emergency medical services tactile tem- perature assessment has been shown to correlate with first measured, core temperature in the ED [16].

Data abstraction

Emergency medical services and hospital electronic medical records were manually linked based on the following criteria: date and time of

Fig. 1. Patient selection⁺. Abbreviation: EMR, Emergency Medical Record. ⁺Inclusion criteria: age >= 18 y, EMS SBP b 110 mm Hg, EMS HR N 90 beats/min, EMS RR N 20 beats/min.

encounter, patient name, and date of birth. The presence of 2 of 3 criteria was required for patient inclusion. Cases that could not be linked by the defined criteria were excluded. Data abstraction was performed by trained abstractors who were overseen by a lead abstractor (C.P.). The abstractors followed procedures outlined in the study operations manual. Random audit of 5% of all abstracted charts was performed by the lead abstractor to ensure at least 95% consistency with the opera- tions manual procedures.

Table 1

Patient characteristics

Characteristic Patients with severe sepsis (n = 75)

Age (y), mean (SD)	56 (15)	49 (16)	.002
Female sex, no. (%)	35 (47)	252 (53)	.35
Race and ethnicity, no. (%)				.57
White	8 (11)	33 (7)
African American	63 (84)	417 (87)
Hispanic	1 (1)	14 (3)
Other Medical history, no. (%)	3 (4)	16 (3)
Cardiac	9 (12)	100 (21)	.07
Hypertension	31 (41)	182 (38)	.57
Diabetes	16 (21)	93 (19)	.69
Stroke	16 (21)	30 (6)	b.0001
Seizure	9 (12)	52 (11)	.76
Asthma	7 (9)	102 (21)	.02
COPD	4 (5)	40 (8)	.37
CKD	5 (7)	26 (5)	.66
Hemodialysis	4 (5)	21 (4)	.71
Cancer	8 (11)	49 (10)	.90
HIV/AIDS	11 (15)	59 (12)	.56

Patients without severe P

sepsis (n = 480)

Primary outcome measure

The primary outcome measure was an inpatient diagnosis of severe sepsis, including septic shock, within the first 48 hours of hospital arrival. The time cutoff was selected in order to exclude cases of hospital-acquired severe sepsis. Medical record review was performed for each subject, and severe sepsis was defined as present if “severe sepsis” or “septic shock” was listed as a diagnosis in the clinical documentation of the inpatient care team [1,17-19]. Emergency medical services and hospital demo- graphics, biologic and physiologic data, admission diagnoses, and hospital outcomes were collected for each patient.

Statistical analyses

Data were collected and entered into REDCap, an online, Health In- surance Portability and Accountability Act-compliant database. For de- scriptive analysis, median values with interquartile ranges are reported. Student t test and ?² (or Fisher exact) tests were used as appropriate to report differences in means and proportions, respectively. Hosmer- Lemeshow test was used to determine the goodness of fit of the model. To derive and validate the predictive model, the cohort was divided into derivation (80%) and validation (20%) subgroups using a random number generator [20]. To build the predictive model, univariable logistic regression analysis was performed on EMS variables consistent with potential predictors of severe sepsis. Variables were chosen for univariable analysis based on biologic plausibility, or if there was a significant difference in the distribution of patients with and without severe sepsis. Infectious signs and symptoms were grouped into a composite category consisting of reported fever, cough, or infection due to small sample size of individual symptoms and instability in the model when symptoms were run individually. Shock, respiratory failure, and respiratory arrest were grouped into a composite risk factor for the same reason. Seizure was not modeled as a risk factor due to

model instability.

Variables associated with a P value less than .10 were retained in a

Abbreviations: CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; HIV/AIDS, human immunodeficiency virus/Acquired immunodeficiency syndrome.

Study approval

The study protocol was reviewed and approved by the Emory Emergency Medicine Departmental Review Committee, the Emory Insti- tutional Review Board, and the Grady Research Oversight Committee.

Results

Descriptive analysis

Among 66,439 EMS transports to Grady Memorial Hospital between January 1, 2011, and December 31, 2012, 555 met the entry criteria, of

which 13.5% (n = 75) had severe sepsis (Fig. 1). Fourteen (19%) of 75 patients with severe sepsis were accurately identified by EMS providers. Baseline characteristics of patients with and without severe sepsis were compared (Table 1). Patients with severe sepsis were older (56 vs 50 years, P = .002), more likely to have a history of stroke (21% vs 6%, P b .0001), and less likely to have a history of asthma (9% vs 21%, P = .02).

Table 2

EMS characteristics

multivariable model, and variables associated with a P value less than

.05 were retained in the final predictive model. Stepwise selection proce- dures were used to further evaluate the final predictive model. We used cross-validation techniques to assess the appropriateness of our model.

Characteristic Patients with severe sepsis (n = 75)

EMD chief concern category, no. (%)

Patients without severe P

sepsis (n = 480)

These techniques changed the number of predictors from 6 to 5, but did not result in significant change in the point estimates. The final model was tested in both the derivation and validation subgroups to de-

Chest pain 4 (5) 58 (12) .08

cardiac symptoms 1 (1) 7 (2) 1.00

Difficulty in breathing 17 (23) 161 (34) .06

Diabetes-related complaint 2 (3) 23 (5) .56

termine performance characteristics including sensitivity, specificity,	Stroke	1 (1)	9 (2)	1.00
and predictive values, both positive and negative. A risk classification	Unconscious	9 (12)	38 (8)	.24
table generated from the model was reviewed in order to select a highly	Seizure	4 (5)	39 (8)	.40

sensitive cut point for risk classification [21]. This strategy was chosen in order to minimize the number of false-negative sepsis screens. However, this strategy also unavoidably increases the number of false-positive screens, a common characteristic of Screening tests.

Using a previously described method, based on point estimate- weighted values for each predictor, the predictive model was converted into a PRESS clinical risk prediction score [22]. A highly sensitive point threshold was chosen to classify patients as low or increased risk for having severe sepsis [22]. All statistical analyses were performed using SAS (version 9.3; SAS Institute Inc, Cary, NC).

Sick person 30 (40) 79 (16) b.0001

Abdominal pain 3 (4) 15 (3) .72

Hemorrhage 2 (3) 21 (4) .76

Other 2 (3) 29 (6) .24

Transport from location, no. (%)

Residence 43 (57) 330 (69) .05

Nursing home 22 (29) 29 (6) b.0001

Other 7 (9) 99 (21) .02

Not documented 1 (1) 12 (3) 1.00

All dispatch categories were defined and determined by use of Priority Dispatch Corporation software.

Table 3

EMS vital signs
Characteristic	Patients with severe sepsis (n = 75)	Patients without severe sepsis (n = 480)	P
Tactile temperature, no. (%)
Hot	27 (36)	56 (12)	b.0001
Normal	38 (51)	358 (76)	b.0001
Cool	9 (12)	55 (12)	.92
Cold	1 (1)	5 (1)	.59
SBP (mm Hg), median (IQR)	90 (83-98)	100 (90-106)	b.0001
HR (beats/min), median (IQR)	123 (112-140)	114 (104-130)	.01
RR (beats/min), median (IQR)	26 (22-30)	24 (22-28)	.07
O2 saturation (%), median (IQR)	92 (87-96)	96 (92-99)	b.0001
Glucose (mg/dL), median (IQR)	134 (94-165)	123 (102-168	.70
GCS, median (IQR)	14 (9-15)	15 (14-15)	b.0001

Emergency medical services characteristics of patients are listed in Table 2. Patients with severe sepsis were more likely to have been categorized by medical dispatch as a sick person (40% vs 16%, P b

.0001) and to have been transported from a nursing home (29% vs 6%, P b .0001). Patients with severe sepsis were also more likely to have had a hot tactile temperature (36% vs 21%, P b .0001), lower SBP (90 mm Hg [interquartile range, or IQR, 83-98 mm Hg] vs 100 mm Hg [IQR, 90-106 mm Hg], P b .0001)], higher HR (123 [IQR, 112-140]

beats/min vs 114 [IQR, 104-130] beats/min, P = .01], lower oxygen saturation [92% [IQR, 87%-96%] vs 96% [IQR, 92%-99%], P b .0001)],

and lower Glasgow Coma Scale (GCS) (14 [IQR, 9-15] vs 15 [IQR, 14-15], P b .0001; Table 3).

The following initial EMS impression categories were more frequently documented in patients with severe sepsis: respiratory failure or arrest (4% vs 0.4%, P = .02), shock (4% vs 0.6%, P = .04),

Development and validation of the predictive model

Using univariable logistic regression analysis, the following variables were found to be significant predictors of severe sepsis in the derivation subgroup: older age modeled in tertiles, absent medical history of asth- ma, medical history of stroke, transport from nursing home, EMD chief concern category of sick person, initial EMS impression of a composite of shock, respiratory failure or arrest, initial EMS impression of acutely altered mental status or unconscious state, hot tactile temperature as- sessment, low SBP, elevated HR, elevated RR, low oxygen saturation, and low GCS (Table 5).

Table 5

Univariable logistic regression

Univariable analysis (n = 441)^a

acutely altered mental status or unconscious status (28% vs 11%,

P b .0001), and a composite category of fever, infection, or cough (15% vs 8%, P = .04; Table 4). The following initial EMS impression categories were more frequently documented in patients without severe sepsis: chest pain (1% vs 11%, P = .01), asthma (0% vs 7%,

P = .01), and seizure (0% vs 8%, P = .01).

In-hospital mortality for patients with severe sepsis was 31% (n = 23) as compared with 5% (n = 25) for those without severe sepsis (P b .0001).

Predictor variable? Odds ratio (95% CL) P

Demographics Age (y), tertiles

b40	Reference	–
50-59	7.65 (2.30-25.45)	b.001
>=60	6.73 (1.92-23.52)	b.01
Sex (M/F)	1.36 (0.78-2.39)	.28
Race (AA/W) Medical history	0.81 (0.30-2.21)	.51
Asthma (Y/N)	0.33 (0.13-0.86)	.02
Stroke (Y/N)	5.10 (2.39-10.91)	b.0001
Cancer (Y/N)	1.31 (0.56-3.11)	.53
HIV (Y/N)	0.96 (0.41-2.23)	.92
Diabetes (Y/N) EMS characteristics	1.44 (0.75-2.77)	.28

Table 4

Initial EMS impression

Impression Patients with severe sepsis (n = 75)

Patients without severe P

sepsis (n = 480)

Nursing home transport (Y/N) 8.87 (4.30-18.29) b.0001

Respiratory failure or arrest

3 (4) 2 (0.4) .02

EMD chief concern

DIB (Y/N) 0.70 (0.37-1.31) .26

Shock 3 (4) 3 (0.6) .04

Chest pain 1 (1) 53 (11) .01

Diabetes (Y/N) 0.90 (0.20-4.02) .88

Sick person (Y/N) 3.32 (1.84-6.01) b.0001

Difficulty in breathing

14 (19) 87 (18) .91

Altered or LOC (Y/N) 0.96 (0.32-2.85) .94

Initial EMS impression

Asthma 0 (0) 34 (7) .01 Shock, RF, or arrest (Y/N) 7.17 (1.74-29.53) .006

Pulmonary edema

0 (0) 8 (2) .60

DIB (Y/N) 1.37 (0.70-2.69) .35

Diabetes (Y/N) 0.70 (0.16-3.08) .64

Abdominal pain 7 (9) 28 (6) .30 Altered or LOC (Y/N) 2.91 (1.49-5.68) .002

Nausea, vomiting, diarrhea

2 (3) 19 (4) .75

Fever, cough, infection (Y/N) 2.30 (0.87-4.62) .11 EMS vital signs

Stroke 0 (0) 1 (1) 1.00

Altered or LOC 21 (28) 52 (11) b.0001

Seizure 0 (0) 36 (8) .01

Dehydration 3 (4) 20 (4) 1.00

Dizzy or weak 9 (12) 28 (6) .05

Diabetes 3 (4) 31 (6) .60

Hemorrhage 1 (1) 16 (3) .71

Fever	3 (4)	6 (1)	.11
Infection	8 (11)	29 (6)	.14
Cough	0 (0)	1 (1)	1.00
Fever, infection,	11 (15)	36 (8)	.04
cough Other	1 (1)	42 (9)	0.03
Not documented	1 (1)	7 (1)	1.00

Abbreviation: LOC, loss of consciousness.

Hot tactile temperature (Y/N) 3.81 (2.02-7.18) b.0001 SBP, per 1-mm-Hg increase 0.95 (0.93-0.97) b.0001

HR, per 1-beat/min increase 1.01 (1.00-1.03) .02

RR, per 1-beat/min increase 1.04 (1.00-1.08) .046

Oxygen saturation, per 1% increase 0.94 (0.91-0.97) b.0001

Blood glucose, per 1-mg/dL increase 1.00 (1.00-1.00) .12

GCS, per 1-point increase (3-15) 0.86 (0.80-0.92) b.0001

Abbreviations: AA/W, African American/white; CL, confidence limit; DIB, difficulty in breathing; HIV, human immunodeficiency virus; M/F, male/female; RF, respiratory failure; LOC, loss of consciousness; Y/N, yes/no.

* All variables modeled as binary categorical predictors (1, present; 0, absent), unless

otherwise stated. Sex modeled as male vs female (reference); race modeled as African American vs White (reference). Age, SBP, HR, RR, oxygen saturation, GCS, and blood glu- cose modeled as continuous variables.

^a Analysis performed in the derivation subgroup.

Table 6

Multivariable Logistic Regression

	Multivariable analysis (n = 441)a
	Predictor variable?	Odds ratio	95% CL	P
	Demographics
	Age (y), tertiles
	b40	Reference	–	–
	50-59	3.83	1.05-14.07	.04
	>= 60	1.63	0.39-6.75	.50
	Medical history
	Asthma (Y/N)	0.45	0.14-4.41	.17
	Stroke (Y/N)	1.88	0.65-5.43	.24
	EMS characteristics
	Nursing home transport (Y/N)	4.47	1.77-11.25	b.01
	EMD dispatch complaint
	Sick person (Y/N)	2.46	1.12-5.40	.03
	Initial EMS impression
	Shock, RF, or arrest (Y/N)	0.61	0.05-6.76	.68
	Altered or LOC (Y/N)	1.49	0.55-4.03	.43
	EMS vital signs
Hot tactile temperature (Y/N)		2.52	1.10-5.74	.03
SBP, per 1-mm-Hg increase		0.96	0.94-0.99	b.01
HR, per 1-beat/min increase		1.00	0.98-1.02	.92
RR, per 1-beat/min increase		0.98	0.92-1.05	.61
Oxygen saturation, per 1% increase		0.94	0.90-0.99	.01
GCS, per 1-point increase (3-15)		0.98	0.87-1.11	0.77

Abbreviations: CL, confidence limit; LOC, loss of consciousness; RF, respiratory failure; Y/N, yes/no.

* All variables modeled as binary categorical predictors unless otherwise stated. Sex modeled as male vs female (reference); race modeled as African American vs white (reference). Age, SBP, HR, RR, oxygen saturation, GCS, and blood glucose modeled as continuous variables.

^a Analysis performed in the derivation subgroup.

In multivariable logistic regression analysis, the following predictors remained significant: older age modeled in tertiles, transport from nursing home, EMD chief concern category of sick person, hot tactile temperature, low SBP, and low oxygen saturation (Table 6). These predictors were retained for the final predictive model (Table 7).

In the final model, Hosmer-Lemeshow goodness-of-Fit test demon- strated good model fit (?² statistic = 6.34, P = .61). Performance characteristics of the model were determined in both the derivation and validation subgroups (area under curve [AUC] derivation, 0.843; AUC validation, 0.820; Fig. 2). Using a highly sensitive cut point of predicted probability greater than 3%, the sensitivity and specificity were measured in both the derivation and validation groups and are reported in Table 8.

Development of the PRESS score

The predictive model was used to generate the PRESS score and an estimate of points-based risk using the same 6 risk factors used in the model: an EMD chief concern category of sick person, EMS transport from a nursing home, Older patient age, hot tactile temperature assess- ment, lower SBP, and lower oxygen saturation. The PRESS score demon- strated a sensitivity of 86% and a specificity of 47% (Table 8). A

Table 7

Final predictive model

Final predictive model (n = 441)

Predictor variable Odds ratio 95% CL P

Age (y), tertiles

b40 Reference – –

50-59 4.28 1.20-15.38 .03

Fig. 2. Receiver operating characteristic curves for derivation and validation subgroups.*.

*AUC derivation, 0.843; AUC validation, 0.820.

prescreening flow sheet and final PRESS score sheet can be seen in Fig. 3 and Table 9, respectively.

Discussion

The PRESS screening tool is simple, practical, and reliable and demonstrates a sensitivity of 86% and a specificity of 47%. One of the advantages of the PRESS score is that it comprises various types of rou- tinely and practically collected EMS data including the following 6 risk factors: an EMD chief concern category of sick person, EMS transport from a nursing home, older patient age, hot tactile temperature assess- ment, low SBP, and low oxygen saturation.

The potential impact of EMS recognition of severe sepsis is consider- able. Just as EMS identification of STEMI allows for coordinated care that is streamlined to achieve the goal of Door-to-balloon times of less than 90 minutes, EMS recognition of severe sepsis could potentially allow for shortened door-to-antibiotic times that maximize benefit to pa- tients. The effectiveness of targeting other time-sensitive treatments in- cluding intravenous fluids and early goal-directed therapy is unknown [4,23]. Diagnostic challenges currently limit Prehospital identification of severe sepsis, arguably resulting in delay of initiation of lifesaving treatment. In fact, a recent epidemiologic study showed that although the average prehospital care interval was greater than 45 minutes for EMS patients with severe sepsis, only 37% received prehospital intrave- nous access [5].

Small studies of EMS identification of severe sepsis have been shown to improve patient outcomes. In a study by Studnek et al [9], patients with severe sepsis who were identified by EMS had a shorter time to first antibiotics in the ED (70 vs 122 minutes, P = .003) and a shorter time from ED triage to early goal-directed therapy initiation as com- pared with patients who were not identified by EMS (69 vs 131

Table 8

Performance characteristics of the predictive model and PRESS score

>= 60 2.19 0.56-8.66 .26

Nursing home transport (Y/N) 4.73 2.01-11.13 b.001 EMD complaint: sick person (Y/N) 3.04 1.45-6.37 b.01

Characteristic Model derivation (n = 441)

Model validation (n = 114)

PRESS score

Hot tactile temperature (Y/N) 2.90 1.35-6.23 b.01 SBP, per 1-mm-Hg increase 0.96 0.93-0.99 b.01 O₂ saturation, per 1% increase 0.95 0.91-0.99 b.01

Abbreviations: CL, confidence limit; Y/N, yes/no.

Sensitivity 91% 78% 86%

Specificity 34% 26% 47%

Positive predictive value 17% 16% 19%

Negative predictive value 96% 86% 96%

Fig. 3. Prescreening flow sheet.

minutes, P = .001). In another study by Guerra et al [10], EMS identifi- cation of severe sepsis using a tool that included POC venous lactate was associated with an in-hospital mortality rate of 13.6% as compared with 50% in patients who were not identified or treated by EMS.

To our knowledge, 3 other EMS screening tools have been developed for severe sepsis [10-12]. They include (1) the Guerra protocol that uses POC lactate, (2) the Robson Screening Tool, and (3) the BAS 90-30-90 [10-12]. These screening tools are arguably suboptimal for a variety of reasons. In a small, pilot study, the Guerra protocol demonstrated a low sensitivity of 48% and is also limited by the fact that POC lactate is not currently available in most EMS systems, including ours. The Robson screening tool was first described as a perspective piece in 2009 by Robson et al [1,11] and is an adaptation of the Surviving Sepsis Cam- paign diagnostic criteria. It uses modified systemic inflammatory re- sponse syndrome criteria, the presence of a suspected infection, and

Table 9

Prehospital severe sepsis score^a

Risk factor Points

EMD chief concern: sick person 3

Nursing home transport 4

Age (y)

18-39 0

40-59 4

>= 60 2

4. Hot tactile temperature	3
5. SBP (mm Hg) 100-109	0
90-99	1
80-89	2
70-79	3
60-69	4
b60	5

6. Oxygen saturation (%)

>= 90 0

80-89 1

70-79 3

60-69 4

b60 5

Total points (0-24)

>=2 points = increased risk for severe sepsis.

^a For use in at-risk patients only. See prescreening flow sheet.

measures of end-organ dysfunction including SBP, oxygen saturation, anuria, lactic acidosis, and prolonged bleeding from injury or gums. In a validation study, the Robson screening tool demonstrated a sensitivity of 93% but requires incorporation of data that may not be routinely available in most EMS settings. Finally, the BAS 90-30-90 is a tool recommended for use in Swedish EMS guidelines that uses 3 clinical indicators: SBP b 90 mm Hg, RR N 30 beats/min, and oxygen saturation b 90% [24,25]. The BAS 90-30-90 tool has demonstrated a sensitivity of 81%, lower than that of the PRESS score.

Our study has several important limitations including its retrospec- tive design. Although arguably the most suitable type of design that practically lends itself to building a large predictive model, it also intro- duces the potential for misclassification of disease. Misclassification may be present in our study because the primary outcome measure, di- agnosis of severe sepsis, was determined using inpatient clinician diagnosis, rather than an independent review by an expert panel. Although this definition has been used in previous studies, it is still possible that this method could result in missed cases of sepsis which would lead to a lower sensitivity of the screening tool [17-19]. The extent to which this potential limitation compromises the validity of our study is unknown.

Although validated internally, it is also noteworthy that our study was conducted at a single center, which limits the external validity of our findings. The PRESS score will need to be validated in other popula- tions before widespread application by 9-1-1 EMS services can be recom- mended. Finally, the PRESS score was developed from a pragmatic standpoint, in that, its use is not meant for use on all patients in the EMS setting. Although this should not be considered a limitation of the study, it is an important point of clarification to ensure appropriate use of the tool in the future. Specifically, all patients in our study had abnor- mal EMS vital signs (SBP b 110 mm Hg, HR N 90 beats/min, and RR N 20 beats/min). Extrapolating use of the tool to all EMS patients would likely yield lower sensitivity and specificity than has been reported herein.

The PRESS score is a PRESS screening tool that has been both derived and validated using routinely collected EMS clinical data. Our hope is that future studies will test the potential benefit of pairing the PRESS score with early, EMS-appropriate interventions and hospital prearrival alert systems that facilitate rapid ED triage and resource allocation for these critically ill patients.

Conclusion

The PRESS screening tool was derived and validated in this study using routinely collected EMS data with a sensitivity of 86% and a specificity of 47%. Additional validation studies are needed before this tool can be recommended for use in the public sphere by EMS services. Supplementary data to this article can be found online at http://dx.

doi.org/10.1016/j.ajem.2015.04.024.

References

1. Dellinger RP, Levy MM, Rhodes A, Annane D, Gerlach H, Opal SM, et al. Surviving sepsis campaign: international guidelines for management of severe sepsis and sep- tic shock: 2012. Crit Care Med 2013;41(2):580-637.
2. Martin GS, Mannino DM, Eaton S, Moss M. The epidemiology of sepsis in the United States from 1979 through 2000. N Engl J Med 2003;348(16):1546-54.
3. Gaieski DF, Edwards JM, Kallan MJ, Carr BG. Benchmarking the incidence and mor- tality of severe sepsis in the United States. Crit Care Med 2013;41(5):1167-74.
4. Pro CI, Yealy DM, Kellum JA, Huang DT, Barnato AE, Weissfeld LA, et al. A randomized trial of protocol-based care for early septic shock. N Engl J Med 2014;370(18): 1683-93.
5. Seymour CW, Rea TD, Kahn JM, Walkey AJ, Yealy DM, Angus DC. Severe sepsis in Pre-hospital emergency care: analysis of incidence, care, and outcome. Am J Respir Crit Care Med 2012;186(12):1264-71.
6. European Heart Rhythm A, Heart Rhythm S, Zipes DP, Camm AJ, Borggrefe M, Buxton AE, et al. ACC/AHA/ESC 2006 guidelines for management of patients with ventricular arrhythmias and the prevention of sudden cardiac death: a report of the American College of Cardiology/American Heart Association Task Force and the European Society of Cardiology Committee for Practice Guidelines (Writing Committee to Develop Guidelines for Management of Patients With Ventricular Arrhythmias and the Prevention of Sudden Cardiac Death). J Am Coll Cardiol 2006;48(5):e247-346.
7. Jauch EC, Saver JL, Adams Jr HP, Bruno A, Connors JJ, Demaerschalk BM, et al. Guide- lines for the early management of patients with acute ischemic stroke: a guideline for healthcare professionals from the American Heart Association/American Stroke As- sociation. Stroke 2013;44(3):870-947.
8. American College of Emergency P, Society for Cardiovascular A, Interventions, O’Gara PT, Kushner FG, Ascheim DD, et al. 2013 ACCF/AHA guideline for the man- agement of ST-elevation myocardial infarction: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guide- lines. J Am Coll Cardiol 2013;61(4):e78-140.
9. Studnek JR, Artho MR, Garner Jr CL, Jones AE. The impact of emergency medical ser- vices on the ED care of severe sepsis. Am J Emerg Med 2012;30(1):51-6.
10. Guerra WF, Mayfield TR, Meyers MS, Clouatre AE, Riccio JC. Early detection and treatment of patients with severe sepsis by Prehospital personnel. J Emerg Med 2013;44(6):1116-25.
11. Robson W, Nutbeam T, Daniels R. Sepsis: a need for Prehospital intervention? Emerg Med J 2009;26(7):535-8.
12. Wallgren UM, Castren M, Svensson AE, Kurland L. Identification of adult septic pa- tients in the prehospital setting: a comparison of two screening tools and clinical judgment. Eur J Emerg Med 2014;21(4):260-5.
13. Trillo-Alvarez C, Cartin-Ceba R, Kor DJ, Kojicic M, Kashyap R, Thakur S, et al. Acute lung injury prediction score: derivation and validation in a population-based sam- ple. Eur Respir J 2011;37(3):604-9.
14. Seymour CW, Cooke CR, Heckbert SR, Copass MK, Yealy DM, Spertus JA, et al. Prehospital systolic blood pressure thresholds: a community-based outcomes study. Acad Emerg Med 2013;20(6):597-604.
15. International Academies of Emergency Dispatch (IAED). Medical Priority Dispatch Systems (MPDS) v12.2: Protocol 26 (Sick Person), ProQA Software v3.4.3.33. Salt Lake City, Utah, USA: Priority Dispatch Corp; 2012.
16. Anderson BP, Polito CC, Martin GS, Isakov A, Yancey A, Wilson D, et al. Hot or not: does EMS tactile temperature correlate with core temperature and risk of sepsis? Crit Care Med 2013;41:A261.
17. Wilson DK, Polito CC, Haber MJ, Yancey II A, Martin GS, Isakov A, et al. Patient factors associated with identification of sepsis in the ED. Am J Emerg Med 2014;32(10): 1280-1.
18. Vishnupriya K, Falade O, Workneh A, Chandolu S, Landis R, Elizabeth K, et al. Does sepsis treatment differ between primary and overflow intensive care units? J Hosp Med 2012;7(8):600-5.
19. Sevransky JE, Martin GS, Shanholtz C, Mendez-Tellez PA, Pronovost P, Brower R, et al. Mortality in sepsis versus non-sepsis induced acute lung injury. Crit Care 2009;13(5):1-6 [R150].
20. Picard RE, Berk KN. Data splitting. Am Stat 1990;44(2):140-7.
21. Wilson JM, Jungner G. Principles and practice of screening for disease. World Health Organization Public Health Papers, 34; 1968 1-163.
22. Sullivan LM, Massaro JM, D’Agostino Sr RB. Presentation of multivariate data for clinical use: the Framingham Study risk score functions. Stat Med 2004;23(10):1631-60.
23. Investigators A, Group ACT, Peake SL, Delaney A, Bailey M, Bellomo R, et al. Goal- directed resuscitation for patients with early septic shock. N Engl J Med 2014; 371(16):1496-506.
24. Blomberg TJ. treatment guidelines SLAS 2011; 2011[Available from: www.flisa.nu].
25. Council S.C.. Medical guidelines for the ambulance services in Stockholm; 2012[Available from: http://www.webbhotell.sll.se/Global/prehospitala/ dokument/Medicinska%20riktlinjer/Nyheter%20%20Ambulanssjukv%A5rdens% 20behandlingsriktlinjer%202012.pdf].