A Comparison of the Clinical Characteristics of Short-, Mid-, and Long-Term Mortality in Patients Attended by the Emergency Medical Services: An Observational Study

Aim: The development of predictive models for patients treated by emergency medical services (EMS) is on the rise in the emergency field. However, how these models evolve over time has not been studied. The objective of the present work is to compare the characteristics of patients who present mortality in the short, medium and long term, and to derive and validate a predictive model for each mortality time. Methods: A prospective multicenter study was conducted, which included adult patients with unselected acute illness who were treated by EMS. The primary outcome was noncumulative mortality from all causes by time windows including 30-day mortality, 31- to 180-day mortality, and 181- to 365-day mortality. Prehospital predictors included demographic variables, standard vital signs, prehospital laboratory tests, and comorbidities. Results: A total of 4830 patients were enrolled. The noncumulative mortalities at 30, 180, and 365 days were 10.8%, 6.6%, and 3.5%, respectively. The best predictive value was shown for 30-day mortality (AUC = 0.930; 95% CI: 0.919–0.940), followed by 180-day (AUC = 0.852; 95% CI: 0.832–0.871) and 365-day (AUC = 0.806; 95% CI: 0.778–0.833) mortality. Discussion: Rapid characterization of patients at risk of short-, medium-, or long-term mortality could help EMS to improve the treatment of patients suffering from acute illnesses.


Introduction
Emergency medical services (EMS) constitute the gateway to health care systems for patients with acute conditions.EMS are confronted every day with acute illnesses demanding precise responses in short intervals of time.On arrival at the scene, the EMS must rapidly assess the clinical characteristics of the patient in order to determine the severity of their condition, perform appropriate and timely therapeutic support, and, if necessary, transfer the patient to a referral hospital center [1,2].
To standardize the clinical presentations in prehospital critical care, novel scoring systems have been proposed to evaluate and predict the risk of early clinical worsening [3].Currently, precision medicine has made great strides in improving prehospital care and emergency departments (EDs), providing bedside scores composed of various clinical, physiological, comorbidity-related, and/or analytical variables as invaluable support in the decision-making process [4].
However, acute illnesses can endanger survival not only at the first stages, but sometimes even days, months, or years later, at which clinical worsening could supervene.Patients admitted to intensive care units (ICUs) exhibit higher mortality rates than the general population, even several years after hospital discharge [5,6].Several investigations show that this excess mortality is present in patients treated by both EMS and ED after the onset of severe acute disease [7,8].On the other hand, aging appears to be a key contributor to this mortality; however, evidence suggests that other factors, such as the degree of senescence or the previous functional reserve, play a critical role in excess mortality [5].Indeed, younger patients admitted to the ICU due to respiratory disease and with a significant comorbidity burden have significantly elevated long-term mortality rates compared to elderly patients without preexisting conditions and infectious diseases [9].
Prehospital clinical characteristics may be useful for categorizing short-term versus long-term mortality.Current risk prognostic models assume a linear relationship between risk factors and clinical outcomes; in contrast, day-to-day patients can present heightened complexity involving a multiplicity of interrelated conditions [10].In this sense, machine learning or algorithms developed based on artificial intelligence allow the analysis of massive amounts of data and the subsequent development of predictive models.These real-time tools, delivered electronically, will help us to gain a better understanding of the complexity of prehospital critical care landscapes, support clinical decision-making, increase the accuracy and timeliness of diagnosis, and provide prognostic predictions [11,12].
The goal of the present study is to compare the characteristics (all the prehospital predictors, including demographic variables; standard vital signs; prehospital laboratory tests; and comorbidities) of patients presenting short-, mid-, or long-term mortality, and to derive and validate three risk models to determine the aforementioned mortality outcomes.

Study Design
A prospective, multicenter, ambulance-based study was conducted on adult patients (>18 years) with unselected acute disease who were evaluated and managed by EMS and transferred to the ED.
Data were extracted from two consecutive studies conducted under identical design standards: "Prehospital identification of prognostic biomarkers in time-dependent diseases -HITS study-" (ISRCTN48326533) and "Identification of biomarkers of clinical-risk deterioration in prehospital care-preBIO study-" (ISRCTN49321933), which followed the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) statement (Supplementary Materials, p3) and were approved by the institutional review board of the Public Health Service (reference: PI-049-19/PI-GR-19-1258).

Study Setting
From 8 October 2019 to 31 January 2022, forty-eight basic life support (BLS), five advanced life support (ALS), and four hospitals (one minor general district hospital and three university tertiary hospitals) in the Spanish provinces of Salamanca, Segovia, and Valladolid (Spain) participated in the study.The Public Health System (SACYL) managed and coordinated all medical services.
In an inbound call center (1-1-2 phone number), a teleoperator geo-locates and profiles the emergency.Next, the call is transferred to a medical dispatcher (a physician or registered nurse), who, via a guided interview, dispatches the most appropriate emergency ambulance team to the incident, i.e., an ALS-composed of two emergency medical technicians (EMTs), an emergency registered nurse (ERN), and a physician-or a BLS, formed by two EMTs.All EMS providers adhered to international guidelines for BLS and ALS.

Population
Consecutive adult patients (>18 years) with acute disease who were transported to the ED were enrolled uninterruptedly.For inclusion in the study, every patient had to be examined by the ALS physician, and based on the objective and structured clinical evaluation and the complementary tests available bedside, the physician decided the need for transfer to the ED and the ambulance type (ALS or BLS).
Cases involving minors; unobserved and unrecovered cardiac arrest; ongoing psychiatric disorders; documented end-stage disease; pregnant women (evident or probable); refusal of transfer; failed prehospital blood test; lack of informed consent; and inability to complete follow-up (365 days) were excluded.

Outcome
The primary outcome was noncumulative mortality (all-cause and in-and out-hospital) in the following time windows: short-term mortality (30 days), mid-term mortality (31 to 180 days), and long-term mortality (181 to 365 days).All non-survivor patients were reassessed by the PI.
Secondary outcomes included prehospital and hospital ALS (noninvasive and invasive mechanical ventilation and/or vasoactive agents) and ICU admission.

Data Collection
ALS providers collected the epidemiological variables (sex at birth, age, attention area, nursing home residence, and type of ambulance for transfer) during the first contact with the patient.Next, the ERN measured the complete standardized set of vital signs (respiratory rate, oxygen saturation, blood pressure, heart rate, temperature, and Glasgow coma scale) and proceeded to perform the prehospital blood test (venous blood gases, ions, hemoglobin, renal panel, lactate, and glucose), with simple fresh venous blood drawn at the same time as the venous line was cannulated, as part of a workflow.ALS physicians gathered data concerning electrocardiographic rhythm and ST segment disturbance, as well as prehospital ALS or special follow-up (advanced airway management and/or use of vasoactive agents) and suspected prehospital diagnoses.A LifePAK ® 15 monitordefibrillator (Physio-Control, Inc., Redmond, WA, USA) was used to determine blood pressure, oxygen saturation, heart rate, temperature, and electrocardiographic rhythm.Biomarkers were assessed with Siemens 10,736,515 EPOC BGEM BUN Test Cards using the epoc ® POC instrument (Siemens Healthcare GmbH, Erlangen, Germany) following the manufacturer's instructions.
Finally, a research associate from every hospital, via electronic medical record check after a one-year follow-up of prehospital support, collected the seventeen comorbidity categories to compute the age-adjusted Charlson comorbidity index (aCCI), hospital admissions, ICU admission, advanced airway management and/or use of vasoactive agents, and noncumulative mortality (all-cause and in-and out-hospital).

Statistical Analysis
Descriptive results and the associations between the outcomes and the analyzed variables were assessed by Student's t-test, the Mann-Whitney U test, or the chi-squared test, when appropriate.Absolute values and percentages were used for categorical variables, and median interquartile ranges (IQR) were used for continuous variables because they did not follow a normal distribution.To determine the variables associated with the outcome, the following process was performed for each noncumulative mortality.(i) A univariate comparison was used to select variables with p < 0.001 criterion.(ii) The selected variables were included in a multivariate logistic regression with forward and backward stepwise variable selection.Note that continuous variables included in the model were not categorized.The results from the logistic regression were evaluated using model metrics (Akaike's Information Criteria (AIC) and Bayesian information criteria (BIC)) and the area under the curve of the receiver operating characteristic curve (AUC).Moreover, the models were internally validated by bootstrapping (1000 iterations), and the Nagelkerke R2 index and the Somers' Dxy index were reported.Data were collected and registered in a database generated with IBM SPSS Statistics for Apple version 20.0 software (IBM Corp, Armonk, NY, USA).The caseload entry system was tested in order to delete unclear or ambiguous items and to verify the adequacy of the data-gathering system.Missing values were completely random; therefore, a listwise deletion method was used since it does not induce biased means, variances, or regression weight modification (note that three patients were removed from the final predictive model procedure due to missing values).The sample size needed for the present study was n = 185, based on the following considerations: a statistical power (1 -β) of 80%, a significance level (α) of p = 0.05, a proportion of the sample in the case group (q1) = 0.1, and an estimated odds ratio of 2.
All calculations and analyses were performed using our own codes, as well as R packages and base functions in R, version 4.2.2 (http://www.R-project.org, the R Foundation for Statistical Computing, Vienna, Austria accessed on 18 June 2023).

Results
A total of 4830 patients were included in the final analysis cohort (see Supplementary Materials, Figure S1), with noncumulative mortalities at 30, 180, and 365 days of 10.8% (523 cases), 6.6% (321 cases), and 3.5% (170 cases), respectively.The median age was 64 years in survivors and 79 years in non-survivors.In a total of 1615 cases (42.3%), the survivors were female; this was also the case for 200 (38.2%),127 (39.6%), and 60 (35.3%) of the noncumulative mortalities at 30, 180, and 365 days, respectively.Two-thirds of the patients underwent ALS, climbing to 79.7% in the 30-day nonsurvivors.Cases with a major comorbidity burden and nursing home origin had significantly elevated mortality rates in all analyzed periods (see Table 1).Table 2 reports the numerical distribution of mortality according to the age-adjusted Charlson comorbidity index.On-scene vital signs, baseline cardiac rhythms, and prehospital blood tests are listed in Table 1.Short-term mortality cases presented a superior incidence of ALS interventions onscene, with 12.4% of noninvasive mechanical ventilation, 30.6% of invasive mechanical ventilation, and 16.3% of vasoactive agents, with a marked incidence of acute life-threatening illness, especially sepsis (19.5%), stroke (18.5%), and cardiac arrest (10.3%).As the time window lengthens, the causes of mortality change, highlighting exacerbations of preexisting comorbidities, e.g., heart failure, chronic obstructive pulmonary disease/dyspnea, or syncope.Hospital inpatient admittance rates, ICU admissions, and hospital ALS interventions were most intense in the short-term mortality cluster, showing a linear decrease in the rest of the mortality groups and the lowest incidence in survivors (Table 3).Table 4 summarizes the predictive models of each mortality.Some of the final selected variables were repeated for all three models (age, partial pressure of carbon dioxide, hemoglobin, and aCCI), but some were exclusive to one outcome-for 30-day mortality, heart rate, ocular Glasgow coma scale, calcium, chlorine, and creatinine were exclusive; for 180-day mortality, nursing home origin, oxygen saturation, tachyarrhythmia, and pH were exclusive; and for 365-day mortality, no variables were exclusive.It is important to highlight that the oxygen saturation/fraction of inspired oxygen ratio was associated with both 30-day and 365-day mortality; for the last case, except for hemoglobin, no analytical parameters were selected.The metrics used to evaluate the models showed that they improved as we considered fewer variables (as better models occurred when increasing mortality time), with results of 1779 and 1901, 1785 and 1893, and 1208 and 1246, respectively, for the AIC and BIC of 30-, 180-, and 365-day mortality.Supplementary Tables S1, S2, and S3 show the univariate analyses of 30-, 180-, and 365-day mortality, respectively.

Discussion
To our knowledge, the present prospective, multicenter, ambulance-based study, conducted in adults with unselected acute disease who were evaluated and managed by EMS and transferred to the ED, is unique because it analyzed on-scene point of care testing (POCT) and demographic, epidemiological, physiological, electrocardiographic, and comorbid characteristics to detect short-, mid-, and long-term mortality.
Our short-term mortality (30 days) was higher (10.8%) than that in a large study carried out in the Danish National Health System involving 219,323 patients transported by ambulance, with a cumulative 30-day mortality of 7.2% [13].Nonetheless, it was significantly lower than that obtained in another study in Finland on patients transported by Helicopter Emergency Medical Service, which showed a 30-day mortality prevalence of 27% [7].Such differences could result from the assumption that the Danish study analyzed all the patients transferred, and the Finnish study included patients referred with high priority by helicopter, whereas our study involved selected patients previously evaluated by physician or nurse dispatchers (1-1-2 emergency coordination center) and tagged as high-priority patients by the on-scene ALS physician.Accordingly, the pretest probability of our study appears to range in the middle of the two populations under analysis, suggesting a relationship with the mortality observed in each study.

Discussion
To our knowledge, the present prospective, multicenter, ambulance-based study, conducted in adults with unselected acute disease who were evaluated and managed by EMS and transferred to the ED, is unique because it analyzed on-scene point of care testing (POCT) and demographic, epidemiological, physiological, electrocardiographic, and comorbid characteristics to detect short-, mid-, and long-term mortality.
Our short-term mortality (30 days) was higher (10.8%) than that in a large study carried out in the Danish National Health System involving 219,323 patients transported by ambulance, with a cumulative 30-day mortality of 7.2% [13].Nonetheless, it was significantly lower than that obtained in another study in Finland on patients transported by Helicopter Emergency Medical Service, which showed a 30-day mortality prevalence of 27% [7].Such differences could result from the assumption that the Danish study analyzed all the patients transferred, and the Finnish study included patients referred with high priority by helicopter, whereas our study involved selected patients previously evaluated by physician or nurse dispatchers (1-1-2 emergency coordination center) and tagged as high-priority patients by the on-scene ALS physician.Accordingly, the pretest probability of our study appears to range in the middle of the two populations under analysis, suggesting a relationship with the mortality observed in each study.
The cluster of mid-term mortality (31 to 180 days) and long-term mortality (181 to 365 days) has been under-studied in the scientific literature but, in terms of mortality rate, was very comparable to the cumulative 1-year mortality of surviving patients admitted to an ICU [14], or to the 31-365-day mortality cluster in the aforementioned Finnish study [7].The observed excess mortality in patients following acute life-threatening illness and after ICU admission could result from post-intensive care syndrome (PICS), characterized by a combination of physical, cognitive, and mental symptoms involving a post-discharge deterioration in quality of life and correlated with poorer long-term outcomes [15].A comparable pathophysiological mechanism may be observed in patients managed by EMS in critical condition who require prehospital critical care [9,16,17].
The short-term mortality model showed excellent prognostic performance (AUC = 0.93), outperforming other scoring systems used to detect early mortality and comprising variables such as age, comorbidities, vital signs, and lactate [13,18].Lactate has been extensively reported as a quick biomarker of metabolic stress and tissue hypoxia, and has been demonstrated to be a strong predictor of short-and long-term mortality in several clinical circumstances, including prehospital care [19].However, other biomarkers included in the model remain less well known [calcium, chloride, hemoglobin, creatinine, and blood urea nitrogen].Hypocalcemia, decreased hemoglobin, and/or abnormal renal function are associated with worse outcomes in acute cardiac and trauma diseases [20,21].Mid-term and long-term mortality models presented poorer performances as compared to short-term mortality models.
Models of prehospital and in-hospital mortality have explored age as an intrinsic hazard for adverse outcomes.Age seems to play an overlapping role in all patients in our three mortality groups.In agreement with previous studies, however, age is not a determining condition for mortality, whereas the burden of comorbidities prior to the event-a variable that has also been identified as a factor associated with morbidity in the short, mid, and long term-may be more critical [22,23].In addition, other variables, such as elevated partial pressure of carbon dioxide and decreased prehospital hemoglobin, remained independent variables in all the analyzed time courses.Decreased hemoglobin has been shown to be a biomarker associated with enhanced frailty and pluripathology, increasing the risk of all-cause mortality in elderly patients, as well as a poor prognostic factor in acute pathologies [24].Similarly, hypercapnia has been associated with tissue hypoperfusion in patients with respiratory diseases, both acute and chronic, showing increased mortality compared to patients with normocapnia [25].On the other hand, the oxygen saturation/fraction of inspired oxygen ratio has already been validated as a predictive variable for early clinical deterioration in prehospital care [26], and the present study shows that this variable has a predictive capacity in the mid-and long-term, whereas oxygen saturation in isolation is only observed as a predictive variable in the short-and mid-term [27].The models used in this work could a priori seem complex due to the number of variables included; however, the actual informatization of medicine could ease the inclusion of the models in daily practice.The integration of models into the monitor/device that captures all the information is now a reality that illustrates how our proposed models could help clinicians.
The study was not free of limitations.First, a convenience sample was made.To minimize bias, recruitment was performed in an uninterrupted way, involving several ambulance stations and hospitals (one minor general district hospital and three university tertiary hospitals) in urban and countryside areas, and involving cases with unselected acute disease.Despite this, the final result included a significantly aging population, a mirror representation of the population pyramid in Spain's surrounding area [28].Second, the data extractors were not blinded.To avoid possible cross-contamination, the EMS providers were unaware of the hospital follow-up data, and likewise, the hospital investigators were blinded to the variables collected during prehospital care.Only the data manager and PI had full access to the data.Third, the study was undertaken during the COVID-19 pandemic.SARS-CoV-2 has produced an exponential increase in pneumonia with multisystem involvement, ICU admissions, and excess mortality [29].In addition, particularly during the first waves, the incidence of emergency calls to EMS dropped dramatically, so hidden mortality from acute life-threatening illness is plausible.Further studies are needed to quantify and identify the true pandemic dimension [30].Finally, the models used are compounded by a large number of variables, a fact that is a handicap for clinical use in prehospital care.Additionally, to determine bedside analytical variables, the use of POCTs-which are devices of proven usefulness and reliability, but with a very uneven implementation-is mandatory, a circumstance that is capable of limiting the models' utilization.To encourage the introduction of scoring systems or risk models (from simple clinical scores to complex models based on artificial intelligence), the future involves the incorporation in ambulances of portable computers that allow on-scene access to electronic medical records, with the option to use a variety of scores or models.

Table 1 .
Cont.The Mann-Whitney U test or chi-squared test was used, as appropriate.c Tachycardia rhythm includes sinus tachycardia, atrial fibrillation, atrial flutter, supraventricular tachycardia, and ventricular tachycardia.d Bradycardia rhythm includes sinus bradycardia, first-degree atrioventricular (AV) block, Mobitz type I 2nd-degree AV block, Mobitz type II 2nd-degree AV block, and third-degree AV block.

Table 2 .
Numerical distribution of mortality according to age-adjusted Charlson comorbidity index.

Table 3 .
Principal outcomes and other determinants, mortality cluster vs. survivors.

Table 4 .
Odds ratios of multivariate logistic regression.