The study took place between 2002 and 2005 during consecutive workshops at the simulation centre of the University of Basel, Basel, Switzerland. The workshops were marketed as "unique learning experience of relevant medical emergencies in a patient simulator" and physicians could take part on a voluntary basis. No formal previous training was required to participate and, during the workshop, no training or teaching was provided prior to the simulation. Thus, the participants' performance reflected their current knowledge and skills. Participants were general practitioners involved in emergency duties or hospital physicians of different specialities (internal medicine, cardiology, intensive care) and status (staff physicians, residents, juniors). The study was approved by the local ethical committee and written informed consent was obtained from all participants.
A high-fidelity patient simulator (Human Patient Simulator, METI®, Sarasota, FL, USA) was used. Features of this simulator include palpable pulses, spontaneous breathing with visible thoracic excursion, eyes with spontaneous lid movements, and a speaker in the mannequin's head that broadcasts the voice of an operator to give the illusion that the "patient" can talk. However, the simulator is unable to detect and/or record the depth of chest compressions and the adequacy of mask ventilation. A cannula was placed in a peripheral vein to allow for intravenous administration of drugs. A commercially available manual defibrillator was placed next to the bed. All participants received a 15 min structured instruction on the technicalities of the simulator.
This is a prospective randomized study. Each resuscitation team consisted of a nurse and either three general practitioners or three hospital physicians. The nurse belonged to the simulator team and was instructed to display a helpful attitude, but to be active on commands only.
Using sealed envelopes a stratified randomization according to the participants' profession was employed to assign an equal number of teams composed of either general practitioners or hospital physicians to two different versions of a scenario of a simulated witnessed cardiac arrest: version "ad-hoc" mimics reality in that only one physician, randomly selected from his/her team, was present at the start of the scenario and the remaining two physicians were summoned to help upon the onset of the cardiac arrest; in version "preformed" all three physicians were present right from the start of the scenario. Pilot experiments revealed that a time period of approximately 5 min during which preformed teams together receive information about the patient's history and subsequently assess together the patient is sufficient to structure the team, and that longer time periods feasible within the settings of simulation offer no significant advantage.
Prior to the simulation, teams were instructed that they were the responsible physicians for the "patient" and that a nurse, fully familiar with all technicalities of the simulator and the equipment, would help them upon request. Teams of general practitioners were informed that the scenario would take place in a group practice where all three of them would work. Teams of hospital physicians were informed that the scenario would take place in the ambulatory part of a hospital where all three of them would work. In "ad-hoc" teams, two randomly selected members were then led to a room adjacent to the simulator and the remaining physician was instructed that help from his/her colleagues would be immediately available on request. Thereafter, the case history was given to the one remaining physician of the "ad-hoc" teams or to all three physicians of the "preformed" teams.
The "patient" was a 66 year old man who felt dizzy after an uneventful bicycle stress test. Upon entering the simulator room, the physician(s) encountered a talkative "patient" connected to a monitor showing sinus rhythm. The "patient" did not feel dizzy anymore but volunteered a detailed account of that episode. In addition, the "patient" complained of stiff muscles in both thighs. Two minutes after the physician(s) had entered the simulator, a cardiac arrest occurred due to ventricular tachycardia displayed on the monitor. With the onset of the cardiac arrest, the "patient" closed his eyes, ceased to speak and to breathe, and pulses were no longer palpable. As our aim was to study the effects of team-building during the early phase of a cardiac arrest, we ensured that all ad-hoc teams were complete ≤ 20 sec after the start of the cardiac arrest: in case the first physician of the "ad-hoc" group did not call for his colleagues within 15 sec they were immediately sent to the simulator. Regardless of any measures taken the patient stayed in cardiac arrest for 3 min. Thereafter, sinus rhythm could be achieved by defibrillation. To avoid a potentially traumatic experience the death of the "patient" was prevented by the nurse who, after six minutes, suggested appropriate measures.
Upon completion of the scenario participants were given a questionnaire and asked to rate on a 11-point Likert scale  the realism of the scenario, the realism of their own behaviour, and the realism of the behaviour of their colleagues (0 = "not at all realistic", 5 = "somewhat realistic", 10 = "very realistic"); the quality of their team's performance (0 = "very low performance", 5 = "average performance", 10 = "very high performance"); the stress felt during simulation, and the stress felt during a real cardiac arrest (0 = "no stress at all felt", 5 = "some stress felt", 10 = "very high stress felt"). A video-assisted debriefing concluded the simulation.
Using frame-in-frame technology, the teams' performance and the monitor displaying the "patient's" vital signs were simultaneously recorded. Data were coded based on the video-tapes recorded during simulation by two independent observers. Inter-observer agreement was assumed if the difference of timing of events was less than 5 sec. In this case, the shorter of two different timings was used for further analysis. Disagreements of more than five seconds in the timing of events were solved by jointly reviewing the videotapes.
Hands-on time was defined as cardiac massage or defibrillation. Each defibrillation was rated as 10 sec of hands-on time. Interruptions of cardiac massage to allow for ventilation were rated as continuous cardiac massage if the interruption was ≤ 10 sec. The first appropriate intervention was defined as first execution of either precordial thump, ventilation, cardiac massage, or defibrillation. Chest compression rates were calculated during the third minute after the onset of the cardiac arrest using a previously published formula : compression rate = (compressions per 60-second segment) × 60/(60 – total pause time in the 60-second segment), where pause time indicates periods of time in which ≥ 2 seconds pass without chest compressions.
All utterances during the first 3 min after the cardiac arrest were noted and classified according to a predefined checklist partly based on the adapted Leadership Behaviour Description Questionnaire : Decision what should be done was defined as any utterance, regardless whether correct or followed, on measures to be performed (e.g. we should defibrillate); Decision on how things should be done was defined as any utterance, regardless whether correct or followed, on how to perform a measure (e.g. the next countershock should be performed with 360 Joule); Direction/Command was defined as any utterance, regardless whether correct or followed, prompting a colleague to do something or do it differently (e.g. you should perform the massage quicker); Task assignment was defined as any utterance, regardless whether correct or followed, that assigned a team member to a particular task. Reflection was defined as any utterance, regardless whether correct or followed, with the potential of prompting a colleague or the team to assess the situation (e.g. what should we do next?). Other utterance was defined as any utterances that did not fit in one of the above categories.
The primary outcome was the hands-on time during the first three minutes of the cardiac arrest. Secondary outcomes included the timing of measures of resuscitation and leadership utterances. A difference of ≥ 10% (i.e. a difference ≥ 18 sec in the first 180 sec of the arrest) in the primary outcome hands-on time was considered to be of clinical significance. Interruptions of cardiac massage of this magnitude are associated with poorer survival rate and worse neurological outcomes [18, 19]. A power analysis revealed that 45 teams had to be studied in each group to detect this difference with significance levels of 0.05 and 90% power. Anticipating a 10% rate of technical difficulties or major protocol deviations we planed to include 50 teams of general physicians and 50 teams of hospital physicians in the study. Data were analysed using SPSS (version 15.0), a commercially available statistical software. Cohen's Kappa for inter-rater reliability, general linear modelling, stepwise multiple linear regression, and Student's t-test were used as appropriate. A p < 0.05 was considered to represent statistical significance.