Evaluation of pre-hospital administration of adrenaline (epinephrine) by emergency medical services for patients with out of hospital cardiac arrest in Japan: controlled propensity matched retrospective cohort study

Study settings

In Japan, the fire departments of municipal governments are responsible for emergency medical services, dispatching ambulances from fire stations. Each ambulance team consists of three emergency medical services personnel, usually including at least one emergency lifesaving technician trained to provide advanced life support (advanced airway management, placement of intravenous line, and defibrillation). Ambulances with doctors are also available but are not routinely used.14 Emergency lifesaving technicians certified after training in hospital are allowed to administer adrenaline (epinephrine) and perform endotracheal intubation with remote instruction from a doctor. Administration of adrenaline to patients with out of hospital cardiac arrest by emergency lifesaving technicians started in April 2006 with a small number of certified technicians, which gradually increased. Thus, during the study period Japanese emergency medical services consisted of a mixture of ambulance teams with and without the ability to administer pre-hospital adrenaline (a relatively small fraction of patients with out of hospital cardiac arrest received pre-hospital adrenaline), which allowed us to examine the effectiveness of pre-hospital adrenaline administered by emergency medical services. The emergency medical services resuscitation procedures follow Japanese guidelines for cardiopulmonary resuscitation, based on the American Heart Association guidelines current at that time. If patients with out of hospital cardiac arrest do not respond to initial cardiopulmonary resuscitation and defibrillation, a dose of 1 mg intravenous adrenaline can be given under instruction from a doctor (other routes or other drugs are not permitted). Repeated doses can be given. Emergency medical services have no discretion to stop resuscitation at the scene; they are supposed to resuscitate and transport patients with out of hospital cardiac arrest to hospital if they are not definitely dead (in cases of decapitation and torso transection, for example). Details have been described elsewhere.10 11 14 15

Study design

We used a controlled propensity matched retrospective cohort study to compare outcome in patients who received (adrenaline group) or did not receive (control group) pre-hospital adrenaline by emergency medical services. We used national registry data obtained from the All-Japan Utstein Registry of patients with out of hospital cardiac arrest. We restricted our comparison to pre-hospital use of adrenaline because the registry data did not include information on use in hospital. We could assume that almost all of the patients reaching the hospital without return of spontaneous circulation received adrenaline at the hospital.11 16 We adopted time dependent propensity score matching to assure comparability between adrenaline and control groups by balancing severity of the cardiac arrest and characteristics of the patients (see table 1⇓) at each time interval (every minute from 6 to 30 minutes) from initiation of cardiopulmonary resuscitation by emergency medical services to administration of adrenaline.17 All the analyses including propensity score calculation and outcome comparison were performed separately among those with an initial rhythm of ventricular fibrillation or pulseless ventricular tachycardia (VF/VT) and among those with non-VF/VT, because the initial rhythms definitively determine the cardiopulmonary resuscitation procedures and prognoses.

Study participants

We included patients aged 15-94 who had an out of hospital cardiac arrest witnessed by a bystander and were registered between January 2007 and December 2010 (figure⇓). In some cases with no witness, cardiopulmonary resuscitation might have been not indicated but still carried out so we excluded them because emergency medical services have no discretion to terminate resuscitation. We also excluded patients who arrested after the arrival of emergency medical services (as we focused on cardiac arrest in situations without medical personnel), those who were given adrenaline after return of spontaneous circulation (re-arrest cases), those in whom arrest was attributable to external causes (such as trauma, drowning, poisoning, and asphyxia), and those with missing, contradictory, or outlying data (such as negative or long (>2 hours) response interval). We also excluded those who were transported by ambulance without an emergency lifesaving technician or by ambulance with doctors.

Selection of cases for study of effect of pre-hospital administration of adrenaline (epinephrine) in patients with out of hospital cardiac arrest. EMS=emergency medical service; ROSC=return of spontaneous circulation; ELST=emergency lifesaving technician

• Download figure
• Open in new tab
• Download powerpoint

Data collection and variables

Since 2005, the All-Japan Utstein Registry, a national registry managed by the Fire and Disaster Management Agency of Japan, has prospectively registered all patients with out of hospital cardiac arrest transported to hospital by emergency medical services in Japan.15 Emergency medical services staff collected data using an Utstein-style form,18 including age, sex, initial rhythm (VF/VT, pulseless electrical activity, asystole, and other), presumed cause of arrest, witnessed or not, description of bystander (family member, colleague, and other), bystander cardiopulmonary resuscitation including defibrillation with an automated external defibrillator, instruction in cardiopulmonary resuscitation to bystander by staff at emergency medical services control centre, activities of emergency medical services (defibrillation, airway management, intravenous line placement, administration of adrenaline), time course of activities (timing of the first dose of adrenaline was recorded; timings of additional doses, airway management, and placement of intravenous line were not recorded), and patient outcomes (spontaneous return of circulation before admission to hospital and long term survival). Failed attempts at airway management or intravenous line placement were not recorded. Information on treatment and return of spontaneous circulation after arrival at hospital was also not recorded. Intravenous line placement was not included in any analyses because it is a part of adrenaline administration procedures. Adrenaline doses were recorded but not included in the analyses because total doses were unknown without information on doses in hospital. If patients reached hospital without return of spontaneous circulation, control patients might have received adrenaline and patients in the adrenaline group might have received additional doses.11 16

The timings of activities of emergency medical services were measured with a clock synchronised within the emergency medical services system, recorded on the activity log forms, and then transcribed to the Utstein-style form. One of the three crew members took charge of data recording; they were trained to record and report the activities with the timing using the designated forms. The physician treating the patient at the hospital determined the cause of arrest clinically. Emergency medical services staff in charge of the data collection contacted the physician to seek information on survival and neurological function at one month or at discharge, whichever was earlier. If the patient survived, the physician assessed neurological function using the Glasgow-Pittsburgh cerebral performance category score: 1 (good), 2 (moderately disabled), 3 (severely disabled), 4 (vegetative state), or 5 (dead).18

The fire departments sent the data to the Fire and Disaster Management Agency, where they were compiled.14 The agency checked the data for missing or contradictory values and requested the fire departments to correct any errors detected to assure the data quality. Of the witnessed patients (n=144 074), 3636 (2.5%) had missing timing data; of those otherwise eligible for the analyses (n=100 677), 4598 (4.6%) had contradictory, outlying, or missing data in the variables that were used for propensity score calculation (figure⇑).

In this study, patients administered adrenaline were defined as those with data on timing of administration; patients without such data were categorised as not receiving adrenaline. We assumed that this categorisation would underestimate rather than overestimate the effects of pre-hospital use of adrenaline and would lead to conservative results. Some patients who actually received adrenaline might have lacked data on timing.

Time dependent propensity score matching

Time dependent propensity score matching is a sequential risk set matching that uses time dependent propensity scores. We created pairs of adrenaline and control patients (1:1 matching) within a risk set at each time interval from the start of cardiopulmonary resuscitation by emergency medical services in a chronological order.17 19 The risk set at time t consisted of those who are at risk of treatment at time t, excluding those who have attained return of spontaneous circulation (no indication of treatment by emergency medical services) or reached hospital (no possibility of treatment by emergency medical services) by time t. The sequential matching should not depend on future information to avoid selection biases. Therefore, patient who later received adrenaline as well as those who never received it were included in the risk set and could be selected as controls (a control group consisting of patients who never received adrenaline alone would accumulate those with favourable outcomes because patients who were slow to attain return of spontaneous circulation might have received adrenaline whereas those who attained return of spontaneous circulation relatively early would not).17 19

The time dependent propensity scores were predicted values, denoted by S, in the following Cox proportional hazard model:

h(t)=h₀(t)exp(b₁x₁+b₂x₂+b₃x₃+...+b_px_p),

S=b₁x₁+b₂x₂+b₃x₃+...+b_px_p

where h(t) and h₀(t) are the hazard function and corresponding baseline hazard function, respectively, which are both constants for given t; t was the time interval from the start of cardiopulmonary resuscitation by emergency medical services to administration of adrenaline by emergency medical services; and b₁, b₂, b₃, ... b_p and x₁, x₂, x₃, ... x_p were estimated coefficients and covariates, respectively.

The model included all the variables obtainable in the field as predictors, except for the description of the bystander (family member or not), which violated the proportionality assumption, and the time from the onset of cardiopulmonary resuscitation by emergency medical services to defibrillation, which caused multicollinearity. The predictors, categorised in table 1⇓, were year, sex, age, bystander cardiopulmonary resuscitation, bystander defibrillation, cardiopulmonary resuscitation instruction given, initial rhythm, time interval from call to emergency medical services to patient contact, number of attempts at defibrillation by emergency medical services, and advanced airway management. The model also included prefecture categories according to the proportions of patients with out-of-hospital cardiac arrest witnessed by a bystander who received adrenaline by emergency medical services (four categories: <5%, 5-<15%, 15-<30%, and ≥30%) to adjust for regionally different environments influencing pre-hospital use of adrenaline.

We created pairs having the closest propensity scores within a calliper (in each pair, the difference in the propensity scores should be within this predetermined width), a quarter of the standard deviation of the propensity scores, without replacement, using a user developed psmatch2 program in Stata,20 as described by Guo and Fraser.21 In the matching process, the timing of administration of adrenaline from the start of cardiopulmonary resuscitation by emergency medical services for the matching was restricted from t=6 to t=30 (minutes); administration at t<6 was considered to be technically difficult and administration at t>30 was rare. The matched pairs were removed from the subsequent procedures. The matching created 1990 pairs for VT/VF and 9058 pairs for non-VF/VT patients.

We showed the imbalance between groups of patients using differences of proportions as all variables were categorical (appendix 1). We used imbalance proportion defined as the sum of absolute differences of proportions between groups among all levels (categories) in each variable and using standardised difference of proportions defined as

√(P₁−P₂)′∑⁻¹(P₁−P₂)

where a (k−1) dimensional proportion vector P₁−P₂ in a variable with k levels is considered to approximately follow (k−1) dimensional normal distribution

N_k−1(0,∑)

Here, P₁ and P₂ are (k−1) dimensional proportion vectors of a variable with k levels in VF/VT and non-VF/VT groups, respectively. ∑ is a (k−1) dimensional diagonal matrix with the (i,i) element being variance of the i-th element (i=1,2,...,k−1) of the proportion vector P₁−P₂.

Endpoints

The endpoints were overall survival and neurologically intact survival with the Glasgow-Pittsburgh cerebral performance category score 1-2 at one month or at discharge, whichever was earlier. Most of the previous studies used survival at one month or at discharge with and without neurological outcomes, which allowed us to compare our findings with the previous ones.

Statistical analysis

After describing the participants’ characteristics (before and after matching), we applied conditional logistic regression models to the matched pairs, with and without adjustment for the variables that were not included in the development of the propensity score, to determine the relation between administration of adrenaline and the endpoints. As the outcomes were measured only once at a fixed time point (at one month or discharge), we did not use a Cox proportional hazard model for the outcome comparisons. The covariates for adjustment in the logistic regression models were bystander description, time from the onset of cardiopulmonary resuscitation by emergency medical services to hospital arrival, time from the onset of cardiopulmonary resuscitation by emergency medical services to the first defibrillation, and presumed cause of arrest. As the sample sizes were determined by the number of patients registered in the database, we calculated the power of the Wald tests of the parameters being equal to 0 (null hypothesis) under a significance level of 5% (two sided).22 All analyses were performed with Stata version 12.1 (StataCorp, College Station, TX, USA).