Study design and setting

In 2009 the Adaptive process triage (ADAPT) was implemented in the ED at Odense University Hospital and is the most commonly used triage system in Denmark [7]. Prior to 2009, the severity and urgency of a patient’s condition were evaluated by an experienced nurse who measured vital values in accordance with a clinical judgment followed by blood test analysis based at a doctors prescription. Patients suffering minor complaints (e.g. sprained ankle, insect bite without systemic reactions etc.) had usually not their vital values measured or blood test analysis performed.

Participants

Eligible patients were all patients aged ≥18 years presenting to the ED with a systolic blood pressure (SBP) ≤100 mmHg registered within 3 h upon arrival during the study period. We chose to use a higher threshold (100 mmHg) of hypotension than the traditional 90 mmHg. This decision was based on increasing evidence advocating for a redefinition of arterial hypotension [8–11], which we also have underlined in a recent study within our research unit [12]. The primary date of contact defined the index date. If a patient had multiple ED visits with hypotension over the study period, only the first was included in the cohort. Patients residing outside the hospitals catchment area at the time of contact and patients without a Danish personal identification number were excluded. Patients who had visited the ED between 1 of January 1998 and 1 of January 2000 with hypotension were excluded to minimize left sided censoring. The background population, from which patients were retrieved, was all adult (≥18 years) Danish citizens living in the hospitals catchment area. Patients were followed from index date until the date of death, emigration, December 31, 2011, or completion of 90 days follow-up, whichever came first.

Data sources and processing

Population-based registers

In Denmark every Danish citizen has free individual access to tax-supported health care provided by The Danish National Health Service. At birth the Danish Civil Registration system (CRS) assigns a unique 10-digit civil personal registry number (PRN-number) to each Danish citizen and to residents upon immigration since 1968. This unique PRN-number enables accurate linkage of the Danish national registers [15]. True population-based studies are hereby possible as all patient contacts are registered and linked between all Danish registries using the patients unique PRN-number.

Outcome measures, exposure and possible confounders

We defined shock as the presence SBP ≤100 mmHg [12] and ≥1 organ failures.

The following organ failures were included: Cardiovascular, Renal, Coagulation and Hepatic. Biochemical variables (creatine, bilirubin, platelets and INR (international normalized ratio)) registered 180 days before and 1 day after the index date was used to identify renal, hepatic and coagulative failure (see Appendix for details). We used the Shock Index (SI) as a measure of cardiovascular failure. SI is calculated as the ratio of heart rate to SBP and included as a categorical variable (<0.7, 0.7–1, ≥1) [18]. We defined cardiovascular failure as SI ≥1. SBP was measured with an automated oscillometric device or manual cuff and sphygmomanometer. Heart rate was measured with ECG, palpation or pulse oximetry. The primary outcome was all-cause mortality within 7-days of the index date. Secondary outcomes were 90-day mortality as well as factors associated with death and annual IRs during the study period. The primary exposure variables were the first recorded SBP ≤100 mmHg at presentation, registered within 3 h upon arrival and the presence of ≥1 organ failures. As the laboratorial analysis of biochemical variables could exceed 3 h (due to busy hours, crowding etc.) we computed organ failures based on variables registered within 24 h after arrival to the ED.

We also included information on the additional covariates; gender, age, SBP level (90> SBP ≤100 mmHg, 80> SBP ≤90 mmHg, SBP ≤80 mmHg) and Charlson comorbidity index. The latter was used as a proxy for comorbid illness [19]. We used discharge diagnoses from the previous 10 years in order to generate the Charlson comorbidity index (CCI; 0, 1–2, >2) for each enrolled patient upon the index contact date [19].

Statistical analysis

We presented continuous and categorical data as medians (interquartile range (IQR)) and numbers (%), respectively.

Participants

Of 438,191 ED contacts a total 1646 (0.4 %) patients presented with shock and were included in the analysis. Reasons for exclusions are presented in Fig. 1 and baseline characteristics in Table 3.

Variable	Total (%)
N (%)	1,646 (100 %)
Age in years, Median (IQR)	71 (56–80)
Sex (%)
Male	885 (53.8 %)
Female	761 (46.2 %)
Age in age groups, yr (%)
18–39	146 (8.9 %)
40–64	496 (30.1 %)
65–84	745 (45.3 %)
85+	259 (15.7 %)
Charlson Comorbidity Index (%)
0	495 (30.1 %)
1	627 (38.1 %)
>2	524 (31.8 %)
Vital variables
Systolic blood pressure, Median (IQR)	88 (80–94)
Diastolic blood pressure, Median (IQR)	53 (44–62)
Heart rate, Median (IQR)	100 (85–114)
Shock Index (SI), n (%)
SI, Median (IQR)	1.2 (1.0–1.3)
SI ≤0.7	94 (5.9)
0.7> SI ≤1	267 (16.6 %)
SI >1.0	1,245 (77.5 %)
Number of organ failures, n (%)
1	1,155 (70.2 %)
2	389 (23.6 %)
3+	114 (6.2 %)
Site of organ failure (%)
Cardiovascular	1,245 (76.6 %)
Renal	382 (23.2 %)
Coagulation	545 (33.1 %)
Hepatic	78 (4.7 %)

^aValues expressed as total number (fraction) and medians [25 percentile-75 percentile] as appropriate

The median SBP on presentation was 88 mmHg (IQR, 80–94 mmHg) with a median SI of 1.2 (IQR, 1.0–1.3). The most frequent organ failure was cardiovascular present in 76.6 % (1245) of the patients). One organ failure was present in 70.2 % (1155), 23.6 % (389) had 2 and 6.2 % (114) had ≥3 failures (Table 3). The proportion of admittance to non-surgical specialties was 49.8 % (820), whereas 36.8 % (605) patients were evaluated exclusively in the ED (Table 1). In the period 2007–2011, 784 patients were assessed in the ED of which 19.7 % (150) had a prehospital contact to a physician (MECU), and 21.6 % (169) were admitted to the ICU (Table 2).

Incidence of shock

The yearly crude IR are shown in Fig. 2 together with the standardized IR. The mean annual IR of shock was 63.2 cases per 100,000 pyar (95 % CI: 60.2–66.3). The IR increased from 53.8 to 80.6 cases per 100,000 pyar, during the period 2000–2011, with an average adjusted annual increase of 2.6 % (95 % CI: 1.1–4.2). The average annual increase using standardized estimates was 2.7 % (95 % CI: 1.0–4.9). The estimated incidence rates stratified by sex and age group with incidence rate ratios are shown in Fig. 3. Men aged 85+ had a fifty-time higher IR than men aged 18–39 years.

Mortality among patients with shock

Among patients presenting with shock 380/1646 died within 7 days ((23.1 % (95 % CI: 21.1–25.1)) and a total 670/1646 died within 90 days (40.7 % (95 % CI: 38.3–43.1),) (Table 4). Trend analysis of the annual 7-, and 90-day mortality proportions did not show any significant change during the entire observation period (7-day mortality: P _trend  = 0.513 and 90-day mortality: P _trend  = 0.674). Kaplan-Meier curves are shown in Fig. 4 with the overall estimated probability of 90-day survival stratified into age (Fig. 4a), Charlson comorbidity index (Fig. 4b), organ failures (Fig. 4c) and systolic blood pressure (Fig. 4d).

		0–7 days					8–90 days
	N, total (%)	N, died (%)	Crude HR (95 % CI)	p Value	Adjusted HR (95 % CI) a	p Value	N, died (%)	Crude HR (95 % CI)	p Value	Adjusted HR (95 % CI) a	p Value
Gender
Female (reference)	761 (46.2)	167 (21.9)	1		1		133 (22.4)	1		1
Male	885 (53.8)	213 (24.1)	1.12 (0.91–1.37)	0.289	1.10 (0.90–1.35)	0.354	157 (23.4)	1.05 (0.83–1.32)	0.686	1.03 (0.82–1.30)	0.813
Age (continous)			1.03 (1.03–1.04)	<0.001	1.03 (1.03–1.04)	<0.001		1.04 (1.04–1.05)	<0.001	1.04 (1.03–1.05)	<0.001
Comorbidity level
0 (reference)	495 (30.1)	95 (19.2)	1		1		53 (13.3)	1		1
1 to 2	627 (38.1)	134 (21.4)	1.13 (0.87–1.47)	0.372	0.85 (0.65–1.11)	0.229	106 (21.5)	1.71 (1.23–2.37)	0.001	1.23 (0.88–1.71)	0.229
>2	524 (31.8)	151 (28.8)	1.56 (1.21–2.02)	0.001	1.10 (0.85–1.43)	0.470	131 (22.9)	3.06 (2.22–4.20)	<0.001	2.03 (1.47–2.80)	<0.001
Number of organ failures
1 (reference)	1,155 (70.2)	206 (17.8)	1		1		189 (19.9)	1		1
2	389 (23.6)	126 (32.4)	1.96 (1.57–2.44)	<0.001	1.90 (1.53–2.38)	<0.001	85 (32.3)	1.78 (1.37–2.30)	<0.001	1.76 (1.36–2.28)	<0.001
3+	102 (6.2)	48 (47.1)	3.08 (2.25–4.22)	<0.001	3.13 (2.28–4.30)	<0.001	16 (29.6)	1.67 (1.00–2.79)	0.048	1.75 (1.05–2.92)	0.033

^aCox proportional hazard model adjusted for sex, age as a continuous variable, Charlson comorbidity level (0, 1–2, >2) and number of organ failures. Patients who died during the first 7 days after admission were excluded from the analyses of 8- to 90-day mortality

Prognostic factors of death among patients with shock

In the multivariate analysis patients with organ failures of 2 (HR = 1.90 (95 % CI, 1.53–2.38)) and ≥3 (HR = 3.13 (95 % CI, 2.28–4.30)) had a higher rate as compared to the reference within 0–7 days. Concordantly, patients with 2 failures (HR = 1.76 (95 % CI, 1.36–2.28)) and ≥3 (HR = 1.75 (95 % CI, 1.05–2.92)) failures had a higher rate as compared to the reference within 8–90 days. Age depicted an increased risk of death within 7 days, whereas comorbidity was not a significant predictor. Within 8–90 days, predictors; age, and Charlson comorbidity index >2 were associated with increased risk of death (Table 4).

Study strengths and limitations

In this study, we analyzed a large cohort of acutely ill, undifferentiated patients arriving to the ED. We had no loss to follow-up do to the unique personal registration numbers in Denmark. The Danish public healthcare system, with a complete, independently and prospectively recorded medical history, made it possible to identify all included patients in the population-based registries. We were hereby able to compute robust estimates on incidence, all-course mortality and predictive factors for death.

The blood pressure measurements were registered prospectively and as a routine documentation and triage in the ED population. In order to avoid possible overestimation of the IR, we excluded patient with residency outside the catchment area and a previously reported admission with SBP ≤100 mmHg in the years 1998–99. To minimize bias from repeated measurements we used the first contact with shock, within the study period.

There are limitations and possible bias that must be kept in mind when interpreting our findings. This was a single-center, retrospective study from a University Hospital ED serving a well-defined catchment area and is the primary and only hospital in this area of Denmark. The results may, however, not necessarily be generalized to other hospitals. Although our ED is the only on serving this part of Denmark, we are not able to adjust for patients living in our catchment area, who have had contact to other hospitals. However, in order minimize this proportion (n = 516, Fig. 1) we excluded patients living in municipalities outside of our ED catchment.

An important limitation is the proportion of patients who were not included as a SBP was not measured upon arrival (n = 273,774). These patients suffered minor complaints and the triaging nurses did not measure SBP based on a clinical judgment. These circumstances also apply for the proportion of patients, who did not have blood test performed upon arrival (n = 698). However, the retrospective data at hand are a reflection of the everyday procedures in our ED and not necessarily collected for research purposes.

We defined hypotension as SBP ≤100 mmHg, based on increasing evidence supporting a higher threshold, as opposed to the traditional 90 mmHg [11, 12]. We used the first recorded SBP value registered and did not have the possibility to examine individual dynamic trends by serial measurements. Although a more detailed definition of hypotension taking into account a patient’s baseline blood pressure as well as repeated measurements in the ED would be ideal, it was not feasible in this study. However, a single measurement approach is a common clinical applied triage method in emergency medicine settings.

Another important limitation is the number of organ failures defining our cohort. Metabolic failure was not included, as arterial punctures were not systematically collected. Moreover, respiratory frequencies and Glasgow Coma Scale were not consistently registered, whereby organ failures related to the respiratory system, and failure of the central nervous system were not included. We used a Shock Index ≥1 to define cardiovascular failure, as this index has been shown to prognosticate outcome across several etiologies of shock and critical illnesses [18, 29–36]. Ideally, cardiac output measurements would have been desirable but not feasible based on the present design. As not all variables for assessing organ failure were available, the incidence rate and the mortality outcomes should be interpreted bearing this in mind.

Furthermore, we acknowledge the presence of a physician in the prehospital setting (MECU) (from 2006 and onwards) could induce referral bias, as certain “high-risk” patients are prone to be transported directly from the pre-hospital setting to the operational theater or ICU, and thereby by-pass the ED. Moreover, in the period 2000–2008 (prior to the implementation of the ADAPT algorithm) blood pressure and blood test were taken only if the acute care ED personal deemed it appropriate whereby our outcomes could be susceptible to selection bias.

Lastly, a significant proportion of patients were evaluated in the ED and either discharged, died or admitted to the ICU (Table 2). The later could be susceptible to information bias, as the registration of ICU admission directly from the ED was not consistently documented during the period of observation. Although limited, we had missing values on covariates; ICD-codes (2 patients) and HR (40 patients).