The institutional review boards of the University of Pittsburgh, University of Utah, and University of Alabama at Birmingham approved the study.
In this retrospective analysis, we linked statewide EMS, hospital discharge, and death data from the Commonwealth of Pennsylvania to determine the relationship between rescuer procedural experience and patient survival after out-of-hospital tracheal intubation.
We studied patients treated by out-of-hospital EMS rescuers in Pennsylvania. Pennsylvania EMS care is diverse and encompasses a range of care configurations and practice settings. Independent private and municipal agencies provide both local and regional EMS care. Pennsylvania EMS practice settings include dense urban population centers (for example, Philadelphia and Pittsburgh), as well as extensive suburban and remote rural areas. Eleven independent air medical services provide care across the commonwealth.
Pennsylvania EMS rescuers work in both volunteer and career capacities and include first responders, emergency medical technicians, paramedics, out-of-hospital registered nurses, and EMS physicians. Advanced life support vehicles may have one or two advanced life support rescuers. Only EMS paramedics, nurses, and physicians are allowed to perform out-of-hospital tracheal intubation. Composing more than 90% of Pennsylvania advanced life support rescuers, paramedics perform more than 94% of out-of-hospital tracheal intubation. Although all air medical rescuers may use neuromuscular-blockade-assisted (rapid sequence) tracheal intubation, select ground EMS units are allowed to use tracheal intubation facilitated by sedatives only.
We used 3 sources of data: Pennsylvania Emergency Medical Services Patient Care Report data, Pennsylvania Health Care Cost Containment Council hospital discharge data, and the Pennsylvania Death Registry.
The Pennsylvania Emergency Medical Services Patient Care Report is a database of all Pennsylvania EMS patient care incidents. In Pennsylvania, all EMS agencies must use electronic medical record systems that transmit patient care data to a central database. EMS services without computer access must submit patient care reports by using computer scan forms. Following the National Highway Traffic Safety Administration standards for EMS data collection and reporting, the Pennsylvania Emergency Medical Services Patient Care Report contains data about patient characteristics, nature and severity of illness, injury patterns, administered drugs, procedures and interventions, and information about the EMS service and out-of-hospital rescuers delivering care.18
We used Pennsylvania Emergency Medical Services Patient Care Report data for the 6-year period January 1, 2000, to December 31, 2005.
The Pennsylvania Health Care Cost Containment Council contains demographic, diagnostic, and clinical information on all hospital discharges in the commonwealth.19
Hospitals use standard software to report basic demographic (patient age, sex), clinical (the date, time, and location of hospital admission, the discharge status and hospital length of stay) and diagnostic information (primary and up to nine secondary International Classification of Diseases, Ninth Revision
] discharge diagnoses). The Pennsylvania Health Care Cost Containment Council includes only patients surviving to hospital admission; the data set does not include patients dying in the emergency department (ED) before hospital admission. We used data for the 3-year period January 1, 2003, through December 31, 2005.
The Pennsylvania Death Registry contains demographic and clinical information on all deaths in the commonwealth.20
We used death data for the 3-year period January 1, 2003, through December 31, 2005.
Selection of Participants
We studied patients receiving successful out-of-hospital tracheal intubation by advanced life support rescuers, including EMS paramedics, nurses, and physicians. Rescuers self-reported tracheal intubation success in the Pennsylvania Emergency Medical Services Patient Care Report; there are no statewide protocols for independent confirmation by a second rescuer or physician. The Pennsylvania Emergency Medical Services Patient Care Report does not include information on unsuccessful tracheal intubations or post–tracheal intubation tube placement events.
We determined the outcomes of patients receiving tracheal intubation during 2003 to 2005. To determine the cumulative experience of rescuers performing these tracheal intubations, we used the longer overlapping period 2000 to 2005.
We linked the 3 data sets to connect out-of-hospital tracheal intubation and patient outcomes. Because the data sets did not have unique patient identifiers (for example, name, social security number, date of birth, and medical record number), we connected patient records with probabilistic linkage. Probabilistic linkage compares the values from several data fields (for example, date, time, age, sex, and geographic region) to estimate the probability that pairs of records match. 21–22
Many medical research studies have used probabilistic linkage.25–31
A more comprehensive description of the record linkage process is given in Appendix E1, Table E1
(available online at http://www.annemergmed.com
). To optimize record linkage, we narrowed the Pennsylvania Emergency Medical Services Patient Care Report to tracheal intubation cases. We limited the Pennsylvania Health Care Cost Containment Council to patients (1) admitted through the ED and (2) admitted to an ICU or discharged with a diagnosis of mechanical ventilation (ICD-9p
96.7 to 96.72), cardiopulmonary arrest (ICD-9
427.4 to 427.5), or respiratory arrest (ICD-9
We probabilistically linked the 3 data sets by using combinations of the following variables: date and time of encounter, patient age, patient sex, patient race, receiving hospital facility, EMS agency location, and patient geographic location (minor civil division). Because an EMS patient might appear in both the Pennsylvania Health Care Cost Containment Council and Pennsylvania Death Registry data sets, we used a “triple match” algorithm to resolve these overlapping linkages.32
A customary practice in probabilistic linkage is to retain only record pairs with predicted match weights over an a priori fixed threshold (eg, match probability >0.90).33
However, this approach often results in low match rates and may inadvertently exclude true matches just below the defined threshold. To avoid this outcome, we used a multiple imputation procedure, creating a series of linked data sets based on the probability distribution of match weights.32
Using this technique, we created 5 probability-linked data sets. We conducted separate analyses on each probability-linked data set and combined the estimates using Rubin’s method.34,35
We linked patient records for the period January 1, 2003, to December 31, 2005. We performed record linkage with Linksolv, version 6 (Strategic Matching Inc., Morrisonville, NY).
Patient survival to hospital discharge was the primary outcome, determined from Pennsylvania Death Registry and Pennsylvania Health Care Cost Containment Council records. If the patient appeared in the Pennsylvania Health Care Cost Containment Council data set, we used the reported discharge status (alive/dead). If a patient did not appear in the Pennsylvania Health Care Cost Containment Council data set but had a death record on the date of encounter, we classified the patient as dead. If the patient appeared in both the Pennsylvania Health Care Cost Containment Council and Pennsylvania Death Registry, we used the outcome reported in Pennsylvania Health Care Cost Containment Council. We identified outcomes of patients intubated during 2003 to 2005 only.
Because of their differing prognoses and airway management approaches, we separately analyzed the cardiac arrest, medical nonarrest, and trauma (major injury) nonarrest subsets. We defined cardiac arrests as patients receiving cardiopulmonary resuscitation (CPR) chest compressions, receiving automated external defibrillator use, or who exhibited an ECG rhythm of ventricular fibrillation, ventricular tachycardia, pulseless electrical activity, or asystole. We classified all other patients as nonarrests.
The Pennsylvania Emergency Medical Services Patient Care Report did not contain standard measures of trauma acuity such as the Abbreviated Injury Score.36
We therefore defined trauma (major injury) nonarrests as patient incidents involving assault, shooting, stabbing, fall, or fire, or bicycle, motorcycle, pedestrian, recreational, or other vehicular crash. We also included cases with major injury situational modifiers such as flail chest, burns greater than 10%, face or airway burns, vehicular extrication greater than 20 minutes, fall greater than 20 feet, extremity paralysis, vehicular speed greater than 40 miles per hour, vehicular speed change greater than 20 miles per hour, vehicular deformity greater than 20 inches, passenger compartment intrusion greater than 12 inches, vehicular rollover, passenger ejection, death in same vehicle, pedestrian/ vehicle crash greater than 5 miles per hour, pedestrian thrown/ run over, and motorcycle crash greater than 20 miles/hour. The Pennsylvania Emergency Medical Services Patient Care Report a priori defined these categories. We defined all other nonarrest patients as medical nonarrests.
For cardiac arrests, we used the covariates patient age, patient sex, major injury/trauma bystander-witnessed cardiac arrest, bystander CPR, EMS automated external defibrillator use, EMS response time (dispatch to arrival on scene), rescuer cumulative patient contacts, EMS agency population setting, and year of encounter. We included trauma/major injury as a covariate in the cardiac arrest model to account for traumatic cardiac arrest cases in the data set. We adjusted for bystander-witnessed arrest, bystander CPR, EMS automated external defibrillator use, and EMS response time because of their identified relationships with out-of-hospital cardiac arrest outcome.37
For the medical and trauma nonarrests, we used the covariates patient age, patient sex, pulse, systolic blood pressure, Glasgow Coma Scale score, rescuer cumulative patient contacts, EMS agency population setting, and year of encounter. Because of the absence of trauma severity covariates, we did not incorporate Injury Severity Scores in the trauma nonarrest model. The Pennsylvania Emergency Medical Services Patient Care Report did not have information on the use of rapid-sequence or sedation-facilitated intubation or the administration of neuromuscular-blocking or sedative agents. Therefore, we did not adjust for the use of rapid sequence or sedation-facilitated intubation.
Acquired clinical experience outside of tracheal intubation procedures may affect patient outcomes. Therefore, for each tracheal intubation we also determined the paramedic’s cumulative number of patient contacts between January 1, 2000, and the date of the tracheal intubation.
We classified EMS agency population setting as urban, nonurban, or air medical. Although the Pennsylvania Emergency Medical Services Patient Care Report contains the minor civil division of a patient encounter, the distribution of urban and rural settings may vary within these regions. Also, the Pennsylvania Emergency Medical Services Patient Care Report does not use standard federal urban/rural coding systems.38
Because our intention was to broadly characterize EMS provider practice setting (not the precise geographic location of the patient), we classified urban EMS agencies as services located in the greater Allentown, Erie, Harrisburg, Lancaster, Philadelphia, Pittsburgh, Reading, Wilkes-Barre, and York areas. We used zip codes of the EMS agencies to confirm their locations. We classified other EMS agencies as nonurban services. Because air medical helicopters may cross geographic boundaries and have distinctly different practice settings, we classified these agencies in a separate air medical category.
We divided patient age into the intervals less than or equal to 6, 7 to 17, and greater than 17 years old. We divided pulse into the intervals less than or equal to 40, 40 to 80, and greater than 80 beats/min and systolic blood pressure into the intervals less than or equal to 60, 61 to 100, 101 to 140, and greater than 140 mm Hg. We divided Glasgow Coma Scale score into the intervals 3 to 8, 9 to 12, and 13 to 15. We divided rescuer cumulative patient contacts to the intervals less than 1,000, 1,001 to 2,000, 2,001 to 4,000, and greater than 4,000. We divided EMS response time to the intervals 0 to 3, 4 to 6, 7 to 10, and greater than 10 minutes. We selected these ordinal categories because they optimized multivariable model fit.
Although the outcomes analysis encompassed January 1, 2003, to December 31, 2005, we sought to account for each rescuer’s accumulated proficiency before this period. Therefore, for each tracheal intubation we defined cumulative tracheal intubation experience as the accumulated number of tracheal intubations since January 1, 2000. Accurate procedural data before January 1, 2000, were not available. Because the Pennsylvania Emergency Medical Services Patient Care Report does not record unsuccessful tracheal intubations, these figures included only successful tracheal intubations.
Primary Data Analysis
To evaluate the association between patient survival and rescuer tracheal intubation experience, we fit multivariable models with generalized estimating equations.39
The general form of the models was
We defined patient survival as the primary dependent outcome. We defined rescuer cumulative tracheal intubation experience as the key independent variable. If the data set attributed a tracheal intubation to more than one rescuer, we used the experience level of the rescuer with the higher cumulative tracheal intubation experience. If the 2 rescuers had the same cumulative tracheal intubation experience, we selected the individual with the higher number of cumulative patient contacts.
We separately analyzed cardiac arrest, medical nonarrest, and trauma nonarrest patients. For the cardiac arrest patients, we adjusted for patient age, patient sex, major injury/trauma bystander-witnessed cardiac arrest, bystander CPR, EMS automated external defibrillator use, EMS response time (dispatch to arrival on scene), rescuer cumulative patient contacts, EMS agency population setting, and year of encounter. For the nonarrest patients, we adjusted for patient age, patient sex, pulse, systolic blood pressure, Glasgow Coma Scale score, rescuer cumulative patient contacts, EMS agency population setting, and year of encounter.
Because each rescuer may have performed several tracheal intubations, we used generalized estimating equations to account for clustering, applying independent covariance structure. We repeated the multivariable analysis on each of the 5 probability-linked sets, combining the results using Rubin’s method. We implemented Rubin’s method through the SAS procedure MIANALYZE.34,35
We analyzed the data with Stata 10.0 (StataCorp, College Station, TX) and SAS v.9.2 (SAS Institute, Inc., Cary, NC).
In the primary analysis, if the data set attributed multiple rescuers to a tracheal intubation, we used the tracheal intubation experience of the most experienced rescuer. We repeated the analysis with the tracheal intubation experience of the least experienced rescuer in these scenarios. We repeated the analysis separately for each EMS agency population setting (urban, nonurban, air medical).