Background
The scientific evidence for an effect of prehospital emergency medical services (EMS) on patient survival is limited and mainly based on case series and cohort studies [
1,
2]. As stated by Callaham [
2]: "There is more solid scientific evidence about topics such as herbal medicine, acupuncture, hives, and constipation than there is about the entire practice of EMS."
Two previous reviews on the effect of advanced life support (ALS) vs. basic life support (BLS) found contradicting results on patient outcome [
3,
4]. A recent study found impaired survival with ALS compared to BLS in a subgroup of trauma patients with Glasgow Coma Score (GCS) < 9 [
5]. However, these studies only included patients treated by paramedical personnel. Other studies have shown that physician treatment may increase survival and add life years for some groups of patients, especially patients with trauma, cardiac arrest and respiratory failure [
6,
7].
The closest any trial on the effect of prehospital treatment by physicians has come to a randomized, controlled design was published in 1987 [
8]. The study examined blunt trauma patients receiving treatment by a paramedic/nurse or physician/nurse crew. Although dispatch of each crew was considered random, depending on rotation of calls or distance to scene, the design might carry inherent bias because differences in time to arrival to scene may influence the results. The study showed a decrease in mortality in the group treated by physicians. Since then, ethical considerations about informed consent have obstructed randomization – in Europe by laws and a directive [
9], making controlled studies the highest possible level of evidence in this area.
The effect of physician based prehospital treatment using updated transport logistics has only been reviewed in combination with helicopter emergency medical services (HEMS) [
10]. Therefore, the aim of this study was by means of a systematic review of controlled studies to examine whether physicians as opposed to other paramedical personnel increase patient survival in prehospital treatment and if so, to identify the patient groups that gain most benefit.
Methods
Original, controlled studies comparing prehospital physician treatment with treatment by paramedical personnel were included. We required that treatment by physicians was an additional therapeutic intervention. We included studies with trauma-patients or patients with acutely developed known or unclear medical, surgical or psychiatric conditions or worsening of such. Studies using the outcome measures: survival, mortality or derivates were considered valid regardless of follow up time. Studies published in any language were included.
Search strategy
Studies were identified by a search in the databases PubMed, EMBASE and Cochrane. The search strands are specified in additional file
1. In all databases, the search was limited to studies published from January 1, 1990 to November 24, 2008. In addition, we conducted a comprehensive search by the feature "related articles" in PubMed and through cross-references from already included original articles and from other reviews.
We systematically excluded studies not meeting the inclusion criteria in a hierarchical manner. First the title of a study, as it appeared from the search pages in the respective databases, was read and searched for the first of four predefined exclusion criteria: 1) Not prehospital treatment: studies that did not involve treatment of acutely ill patients out of hospital or any other medical institution, studies of organizational or safety-related issues in the prehospital setting and studies of inter hospital transfer of patients, 2) Not physicians: studies evaluating the effect of other health personnel, procedures or equipment, but not the effect of physicians, 3) Not controlled: all studies that did not explicitly compare a group of patients treated by physicians with a group of patients treated by paramedical personnel, including studies that compared physicians with paramedical personnel indirectly utilizing historical control studies and 4) Not survival: studies that did not use survival, mortality or derivates of these as outcome measures. If the study could not be excluded based on criteria 1, we searched for criteria 2 and so forth. If a study could not be excluded based on title, the abstract was searched. If exclusion could not be done based on the abstract, the entire article was searched. For each study we registered whether it was excluded by title, abstract or article and by criteria 1, 2, 3 or 4. Two reviewers (Bøtker MT and Bakke SA) conducted the searches in duplicate, and discrepancies regarding exclusion were solved by consensus with a third reviewer (Christensen EF). In cases of further doubt, whether a study was eligible for inclusion, an email was sent to the corresponding author of the article for clarification. The two reviewers independently extracted study details from the included articles. We predefined data to be extracted: study design, type and number of patients in each group, prehospital setting, raw survival data, adjustment and results. Discrepancies regarding data extraction were solved by consensus with the third reviewer.
Study assessment
The studies were categorized according to study type: 1) controlled cohort studies, i.e. studies comparing outcome with different treatment over a period of time 2) system comparison studies, i.e. studies comparing outcome in two geographically separate areas and 3) before-and after studies, i.e. studies comparing outcome before and after a change in personnel. We ranked the studies in tables according to the number of patients in the intervention group: 1) less than 100 patients 2) 100–1000 patients and 3) more than 1000 patients. Afterwards we sorted them according to result: 1) studies demonstrating increased survival with physician treatment 2) studies not demonstrating a significant difference and 3) studies demonstrating decreased survival with physician treatment. A result was considered statistically significant if 95% confidence intervals were exceeded or p values were < 0.05.
For studies necessitating subgroup analysis, we attempted to assess, whether a subgroup analysis was intended a priori or whether it was a post hoc decision. Studies with subgroup analysis were initially categorized according to overall findings and ranked according to total number of patients. Afterwards, the results were categorized according to subgroup result and ranked according to the number of patients in the sub-intervention group. For some trauma studies using the TRISS methodology [
11] based on Trauma Score [
12] and Injury Severity Score (ISS) [
13], interpretation was necessary to clarify whether the difference was significant or not. The Z statistic (± 1.96 required for significance) was utilized to describe the deviation in mortality (or survival) in a study group from the Major Trauma Outcome Study (MTOS) population [
14] or the W statistic to describe how many more/less survivors than expected from the MTOS benchmark there were. In one study, CANALS – adjusting for variables like Revised Trauma Score, Injury Severity Score, age, sex, type of accident and type and number of treatment, was used [
15].
Discussion
Few studies met our inclusion criteria. In the two groups most studied – trauma patients and patients with out of hospital cardiac arrest – an improved survival with physician treatment was found. Indications that survival might be increased for other patient groups were found, but an increase in survival could not be demonstrated for broader, unselected patient groups attended by EMS.
We found no randomized controlled studies. Because this was suspected, we had chosen to include controlled studies a priori. This was done in order to pursue the best available evidence, in line with the intention of evidence based medicine [
43]. However, we only found 26 controlled studies for inclusion and many of the studies included very few patients. The studies were inhomogeneous and as a consequence we did not perform a true meta-analysis.
The risk of publication bias is a well-recognized limitation of systematic reviews [
44]. We sought to minimize this by including studies in other languages than English in order to avoid bias introduced by the tendency to publish very unique results in an English journal and otherwise in a journal of native language. The large number of studies not fulfilling our inclusion criteria demonstrates the degree of difficulty in constructing a concise search in this area. This is mainly caused by a huge variability in the terms used for physicians, emergency services and transport platforms in the prehospital setting. It increases the risk that includable studies were missed by the original search. In addition, there was a considerable difference between the number of studies located in PubMed, Cochrane and EMBASE. This is probably caused by our rigorous search method that was initially intended for PubMed. We intended to counteract these limitations by an initial systematic search in all three databases and subsequently by systematically using the "related articles" function in the PubMed database and searching cross-references in articles from already included studies.
One of the confounding factors that may influence the results was transportation, i.e. physicians were transported by helicopter, the other crew by ground ambulance [
15,
21,
22,
24,
26,
27,
31‐
34,
37,
40]. However, the confounding influence may only be minor because the studies demonstrated an increase as well as a decrease and no change in survival suggesting absence of any systematic bias. The observation is in accordance with the fact that the effect on survival with helicopter transport has not yet been clarified [
7,
45]. Lossius et al [
6] estimated that the major part of life years gained could be explained by treatment and only ten percent by fast helicopter transport. Consequently, we made no attempt to adjust for transportation logistics. Another potential confounder is the merging of comparisons between not only differences between prehospital physician vs. non-physician treatment but also between other differences in system logistics [
18,
28,
30,
35,
36] and comparison of survival before and after a larger reorganization [
38].
In 14 studies, an increased survival with physician treatment was demonstrated. One of these studies on trauma patients was large-scaled and well designed system comparison study [
35]. In particular three other studies were appropriately designed [
20,
25,
31]. One of these was a system comparison study [
25], another included few patients in the intervention group [
31], reducing the strength of this study, but one was large-scaled and found an increased survival in a priori defined subgroups [
20]. In three studies finding increased survival with physician treatment, the strength was limited by a low number of patients in the intervention group [
23,
39,
40] and two studies had obvious confounding factors because only physicians carried defibrillators [
16] and because the addition of physician treatment was part of other organizational improvements [
38]. Three studies revealed a reduced survival with physician treatment. In one of these, a higher ISS and a lower GCS in the intervention group than in the control group remained unadjusted for [
26]. In another, the patients were divided into subgroups depending on admittance to ICU or not and it was not specified why this division was chosen [
29]. Hence, the result should be interpreted with caution. One of the studies finding a decrease in survival with physician treatment was well-designed [
30] and possible explanations for the impaired survival with ALS found in some studies are prolonged on-scene time in patients requiring in-hospital definitive treatment (e.g. patients with penetrating trauma) and suboptimal endotracheal intubation [
4]. Nine studies did not demonstrate any significant difference in survival. In particular three studies were well designed, but all included a limited number of patients in the intervention group or relevant subgroup [
18,
21,
22] increasing susceptibility to type 2 errors. One of these examined patients in cardiac arrest after blunt trauma [
22]. These patients have a very serious prognosis and an expected survival of nearly zero. It should be noted that survivors (n = 2) were only found in the group treated by physicians. In one small study, exclusion of patients dying in the ambulance induced a risk of selection bias [
27]. In one study survival rates were very high in both intervention and control group (97.2% and 97.5%) reflecting that for patients with an a priori low risk of dying, physicians do not increase survival [
19].
Thus, the quality and strength of the evidence supporting an increase in survival with physician treatment was variable and could be influenced by publication bias. However, many of the studies not finding a significant difference in survival were susceptible to type 2 errors and the evidence supporting a decrease in survival was very sparse and mostly of questionable quality. Hence, the results are encouraging – in particular for the most studied group – trauma patients. Since trauma patients and patients with out of hospital cardiac arrest were the only specified groups studied, many conditions seen in the prehospital setting remain unexamined. Future research should address this aspect.
It was beyond the scope of our review to consider that the only two studies analyzing admission rates, demonstrated that physicians were able to complete treatment of more patients on site and thus avoid unnecessary hospital admission [
19,
20]. Avoided admissions may be of importance not only for the individual patient, but also have potential socioeconomic implications. This is an obvious topic for future research.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
MTB conceived and designed the study, carried out one duplicate of the search and study details extraction, and drafted the manuscript. SAB carried out the other duplicate of the search and study details extraction, and helped drafting the manuscript. EFC participated in conceiving the study, participated in its design and revised the manuscript. All authors have read and agreed to the content.