Changing role of EMS –analyses of non-conveyed and conveyed patients in Finland

This was a prospective study design.

A national dispatch authority operates with six regional emergency medical communication centers (EMCC) with a common and linked data management system. Incoming calls related to medical incidents are assessed in accordance with a criteria-based and nationally standardized dispatch protocol, which is formulated by national expert panel together with EMCC and the prehospital centers of the five university hospitals. A dispatcher’s training takes 18 months, and the EMCC personnel are usually not health care professionals. After triage, four categories of urgency (A, B, C and D) are used for EMS, where A refers to a life-threatening mission, B to an unknown but potentially high-risk mission, C to an urgent but not life-threatening mission and D to a non-urgent but acute situation.

In Finland, EMS is organized by 21 hospital districts and is a part of specialized healthcare. The structure of EMS is based on a four-tiered system including First Responders (FR), Basic Life Support (BLS) units, Advanced Life Support (ALS) units, on-duty Medical Supervisor units and physician manned units (HEMS and grounds units). The personnel in a BLS unit vary, and consist of firefighters, Emergency medical technicians (EMT) and/or practical or registered nurses. ALS units are the most common, and they are manned by at least one paramedic-nurse with a bachelor level degree requiring 4 years of education or a registered nurse with 1 year of additional education in prehospital emergency care. EMS units mainly operate 24/7. On-scene triaging of a patient is based on national and/or regional treatment protocols and the legal basis is described in the National Health Care Act. After triaging and treatment, units are allowed to make the non-conveyance decision based either on the standing orders or by consulting an EMS physician or doctor in primary care.

In the study area, there are central regional hospitals and municipal healthcare centers or other primary care units, where the EMS can convey patients depending on the urgency and need of the patient. In addition, with some cases it is possible to convey patients to university hospitals located in other areas. For clarity, we use the term healthcare facilities here to refer to all of the above.

The data were collected between 1st June and 30th November 2018 from different data systems of the EMS and hospital patient records in the hospital districts of Etelä-Savo, Kanta-Häme and Päijät-Häme in Finland (Fig. 1). The study area consists of both urban and rural areas, comprising 32 municipalities and a total of 482,805 inhabitants, which is 8.8% of the Finnish population. The average density of population is 26.1 inhabitants per square kilometre.

In the study area, all EMS units were able to use an electronic patient reporting system. There were two different systems in use (Merlot Medi, CGI Suomi Oy, Finland, and Codea, Codea Oy, Finland); data from these EMS databases were combined for further analyses.

The adjusted ICPC2 classification for the EMS reason for care was taken into use; hence the EMS’ electronic patient reporting systems were updated accordingly. The ICPC2 code list used was created by the Nordic Collaboration (Benchmarking) Group for EMS [23]. This list includes around one hundred ICPC2 codes, and this list was proposed to be used in Scandinavian EMS. In this study, the EMS personnel were trained in the use of the codes before the study period.

NEWS2 scores were calculated for the first values that were measured. If there were missing values, they were decoded as normal. The cut-offs for exclusion were as follows: respiration rate < 4/min or > 70/min, oxygen saturation < 50% or > 100%, systolic blood pressure < 40 mmHg or > 280 mmHg, pulse < 20/min, and temperature < 25 °C or > 45 °C. Level of consciousness was assessed by the Glasgow Coma Scale (GCS) in the electronic patient reporting systems. GCS was converted to the ACVPU scale (GCS 15 = Alert (A), GCS 14–3 = CVPU: confusion (C), verbal (V), pain (P), unresponsive (U), similarly as two other studies before [17, 24].

The NEWS2 score calculation requires information about whether supplemental oxygen has been used and if the patient has hypercapnic respiratory failure (usually due to chronic obstructive pulmonary disease, COPD) [15]. We applied a semi-supervised information extraction method to detect and extract relevant mentions from the free-text fields. First the word2vec toolkit [25] was used to train two semantic word space models in an unsupervised manner: one model was trained on a corpus of hospital clinical text (0.9 million physician and nursing notes), and the other was trained on the free text from the EMS data. Next, with a list of keywords related to oxygen administration, hypercapnic respiratory failure and COPD as the starting point (provided by domain experts), we queried the semantic models to extract words with similar meanings. The extracted keyword candidates were then analyzed by the domain experts to notice common synonyms and misspelled variants. This approach is comparable to the interactive rapid vocabulary exploration used in a recent study [26]. With the revised list, we searched and labeled all free-text fields from the EMS missions based on occurrences of these keywords. If the patient had COPD, the SpO2 was analyzed by scale 2 [15].

EMS personnel measured the influence of alcohol by a breathalyzer test or clinically. These cases we coded as yes or no into the analyses. The urban-rural classification was analyzed according to the Finnish Environment Institute (SYKE) classification (Fig. 2) [27]. For further statistical analysis, a Spatial Network Analysis was executed for every EMS patient case observation [28, 29]. With the Spatial Network Analysis, the fastest route from every EMS patient mission location to the nearest (non-conveyed patients) or realized (conveyed patients) healthcare facility was calculated. The opening hours of the healthcare facilities were considered in the analysis.

In order to achieve this, three data sets were used when performing the analysis: 1) EMS missions with coordinates, time stamp and other additional information, 2) healthcare facilities with coordinates and opening hours and 3) Digiroad (CC BY 4.0), the national road and street database [30]. All datasets were quality-assessed by their locations and their attribute information and modified, if necessary.

The fastest route from the EMS mission location to the nearest healthcare facility was computed by using the Closest Facility method [31]. The Finnish national road and street database Digiroad (CC BY 4.0) was used as the Network Dataset of the analysis and speed limits of the road network were defined as the cost attribute [30]. Hierarchy and restriction attributes were also defined so that the routing prefers larger main roads, and avoids pedestrian lanes and small forest roads completely.

Categorical variables were characterized using frequencies and percentages and continuous variables were characterized by using medians and IQR (interquartile range), because variables were not normally distributed. Differences between non-conveyed and conveyance groups were tested using Chi-Square test (categorical variables) or with Mann-Whitney U-test (continuous variables). Univariate associations between non-conveyed and conveyed patients and study variables were studied using logistic regression analysis. Multiple logistic regression analysis included variables that were clinically and statistically significant in univariate analysis, and the expected rural-urban area, because it partly measures the same thing as distance to the nearest healthcare facilities. NEWS2 points were analyzed separately, since it is suitable only for patients over 16 years of age. Results were presented with odds ratios together with 95% confidence intervals and p-values. Statistical analyses were carried out using SAS for Windows version 9.4 (SAS Institute Inc., Cary, NC, USA), and p values < 0.05 were considered to be statistically significant. The age groups in this study were based on the Finnish national classification provided by Statistics Finland. Distance to health care facilities was classified for the analysis purpose.

This study has some limitations. First, although this study had a prospective design, the registries used were not primarily designed for scientific research, which is why we had to exclude a number of patients. For instance, 3718 patients were missing or had an incorrect or unclear social security number. Second, the use of adjusted ICPC2 classification for emergency care was only recently adopted; consequently, its usage was not established despite the training of the EMS personnel. In one study area, due to a human error, ICPC2 codes Z25 and Z29 (social problems) were missing, although those codes were quite rare. We only looked at the main reason for care, even if another ICPC2 code, also chosen by EMS, might have provided more information about the patient. Third, when we analyzed the NEWS2 points, we interpreted the missing values as normal. The final NEWS2 scores obtained may thus be lower. We analyzed the use of oxygen and whether the patient had a COPD with the text mining method (machine learning). Even if we noted spelling mistakes etc., it is possible that we did miss some information. For example, trade names of drugs change constantly. Furthermore, the level of consciousness was assessed by GCS, which we converted to the AVCPU scale. In the ACVPU scale, C means new confusion [15], which is difficult to detect retrospectively, but it does not change this analysis compared to the conversion of GSC to AVPU. Fourth, the EMS units’ non-conveyance decision-making was influenced by several other factors, which were not analyzed in this study. Requirements of the EMS, organizational support, guidelines and human factors differ between study areas. Fifth, it should be noted that in this article we did not assess the outcomes, follow-up care or patient safety of non-conveyed and conveyed patients.

However, despite of these limitations, we feel that the aims of this study are justified. In this prospective study design we collected data from all EMS missions from three areas over six-month period. The dataset is large and a wide range of variables were tested. The characteristics of EMS patients were described, differences between non-conveyed and conveyed patients were shown, and multivariate logistic regression model was found.