The Institutional Review Boards of the University of Alabama at Birmingham and University of Pittsburgh approved the study. In this retrospective analysis, we linked Commonwealth of Pennsylvania statewide EMS, hospital discharge and death data to determine the hospital discharge diagnoses of patients receiving successful out-of-hospital ETI.
This study included patients receiving out-of-hospital care in the Commonwealth of Pennsylvania. EMS care in Pennsylvania encompasses a range of practice settings including dense urban population centers (for example, Philadelphia and Pittsburgh), extensive suburban and remote rural areas. Care configurations include independent private and municipal agencies providing both local and regional EMS care. There are 11 independent air medical services across the Commonwealth.
Pennsylvania EMS rescuers work in both volunteer and career capacities. EMS roles in Pennsylvania include first responders, emergency medical technicians, paramedics, prehospital registered nurses and EMS physicians. Advanced life support (ALS) vehicles may have one or two ALS rescuers. Only EMS paramedics, nurses and physicians perform out-of-hospital ETI. Paramedics comprise over 90% of Pennsylvania ALS rescuers. All air medical rescuers may use neuromuscular-blockade assisted (rapid-sequence) ETI. Only select ground EMS units are permitted to use ETI facilitated by sedatives-only; ground EMS units are not permitted to perform neuromuscular-blockade assisted (rapid-sequence) ETI
Sources of Data
For this study we used three sources of data: 1) Pennsylvania Emergency Medical Services Patient Care Report (PAEMS) data, 2) Pennsylvania Health Care Cost Containment Council hospital discharge data set (PHC4), and 3) Pennsylvania Death Registry (PA Death). We used data for the three-year period January 1, 2003 through December 31, 2005.
PAEMS is a database of all Pennsylvania EMS patient care incidents. EMS agencies in Pennsylvania use electronic medical record systems that transmit patient care data to a central database. EMS services without computer access must submit patient care reports using computer scan forms. PAEMS follows the National Highway Traffic Safety Administration standards for EMS data collection and reporting. The data describe patient characteristics, nature and severity of illness, injury patterns, administered drugs, procedures and interventions, and information regarding the EMS service and out-of-hospital rescuers delivering care.17
PHC4 contains demographic, diagnostic and clinical information for all hospital discharges in the Commonwealth of Pennsylvania.18
Using standard reporting software, hospitals provide 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. The data base contains a primary and up to eight secondary discharge diagnoses, defined using the International Classification of Diseases, Clinical Modification, ninth edition (ICD-9-CM). PHC4 does not include patients dying in the Emergency Department (ED) prior to hospital admission.
PA Death contains demographic and clinical information on all deaths in the Commonwealth of Pennsylvania, including the date, time, location and attributed reason for death.19
Linkage of Data Sets
Because the three data sets (PAEMS, PHC4, PA Death) did not have unique patient identifiers (for example, name, social security number, date of birth, medical record number, etc.), we connected patient records using probabilistic linkage. Probabilistic linkage compares the values from several data fields to estimate the probability that pairs of records represent the same person or event.20–23
Many medical research studies have used probabilistic linkage.24–30
We previously described the details and results of the record linking process.10
To facilitate linkage, we limited PAEMS to successful ETI cases, and we limited PHC4 to patients 1) admitted through the Emergency Department and 2) admitted to an intensive care unit or discharged with a diagnosis of mechanical ventilation (ICD-9p 96.7–96.72), cardiopulmonary arrest (ICD-9-CM 427.4–427.5) or respiratory arrest (ICD-9-CM 799.1).
The probabilistic linking process used 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). Since an EMS patient might appear in both the PHC4 and PA Death data sets, we used a “triple match” algorithm to resolve these overlapping linkages.31
A customary practice in probabilistic linkage is to retain only record pairs with predicted match weights over an a priori
fixed threshold (-e.g., match probability >0.90).32
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 upon the probability distribution of match weights.31
Using this technique we created five probability linked data sets. We conducted separate analyses on each probability linked data set and combined the estimates using Rubin's method.33,34
We linked patient records for the period January 1, 2003 – December 31, 2005. We performed record linkage using Linksolv, version 6 (Strategic Matching Inc., Morrisonville, New York).
Selection of Patients
We included patients reported as receiving successful out-of-hospital ETI by ALS rescuers. The PAEMS data set does not contain information on unsuccessful ETIs. Rescuers define and report ETI success; there are no statewide protocols for independent confirmation by a second rescuer or physician. The PAEMS cannot identify instances where initial ETI failure was followed by successful supraglottic airway placement.
The outcomes of this study were the patient primary and secondary medical conditions. We determined and classified the outcomes differently for patients dying before hospital admission and those admitted to the hospital. ()
FIGURE Overview of medical condition classification. We classified patients dying in the field or ED as “Death Before Hospital Admission.” For patients surviving to hospital admission, we determined the primary and secondary medical conditions (more ...)
We classified patients dying in the field or Emergency Department as “Death Before Hospital Admission.” Patients dying before hospital admission do not appear in the hospital discharge data set and therefore do not have assigned ICD-9-CM discharge diagnosis codes. Due to the inconsistency of death reports and their use of the different ICD-10 classification system, we opted not to use the “reason for death” data appearing in the PA Death data set.
For patients surviving to hospital admission, we determined the primary and secondary medical conditions from ICD-9-CM hospital discharge diagnosis codes appearing in the PHC4 data set. At the completion of a patient's hospital course, hospital personnel typically assign diagnosis codes (International Classification of Diseases, ninth edition (ICD-9)) reflecting the medical conditions associated with the patient's hospitalization, as well as the patient's comorbidities. Hospitals typically use this information to support billing efforts. PHC4 contains the primary and up to eight secondary ICD-9-CM diagnosis codes for each hospitalization.
To examine the concurrent medical conditions of the study population, we also calculated Charlson comorbidity index scores for each hospitalized patient.35
The Charlson index is a widely used system for characterizing patient comorbidities, drawing upon information regarding 17 chronic medical conditions (myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatologic disease, peptic ulcer disease, liver disease, diabetes, hemiplegia or paraplegia, renal disease, cancer, and acquired immunodeficiency disease syndrome (AIDS)). Each condition receives a different weight (1, 2 or 6 points). The total Charlson score consists of the sum of the individual weighted scores and ranges from 0–37, reflecting increasing risks of one-year mortality. For example, a patient with a history of myocardial infarction has Charlson Score of 1, corresponding to 10% probability of death within one year.36
For example, a patient with a history of myocardial infarction and leukemia has Charlson Score of 2, corresponding to 19% probability of death within one year. In scientific analyses, authors typically categorize the Charlson score as 0, 1, 2 or ≥3. Deyo, et al. adapted the Charlson index for use with ICD-9-CM administrative data sets.37
Primary Data Analysis
We grouped the primary and secondary discharge diagnoses of admitted ETI patients according to major ICD-9-CM disease categories. () We opted to group the cases using the ICD-9-CM system due to the wide recognition of this method.
TABLE Discharge diagnoses of patients receiving out-of-hospital endotracheal intubation, Commonwealth of Pennsylvania 2003–2005. Percentages reflect estimates for each imputed set, combined by Rubin's method. Each hospital record contained one primary (more ...)
In addition to the major ICD-9-CM categories, we also highlighted selected disease subgroups with potential relevance to airway management. We defined a subgroup for sepsis, septicemia, bacteremia and septic shock (003.1, 022.3, 031.2, 038–038.9, 040.82, 422.92, 449, 659.3, 670.20–670.24, 673.3, 771.81, 771.83, 785.52, 790.7, 995.90–995.94). Under endocrine disorders we identified the subgroup with diabetes, hypoglycemia and hyperglycemia conditions (ICD-9-CM 250–250.9, 251.0–251.2). Under neurological disorders we identified epilepsy, seizures and convulsions (ICD-9-CM 345–345.91, 649.40–649.44, 779.0, 780.3–780.39) as well as anoxic brain injury and encephalopathy (348.1–348.39, 349.82).
Circulatory system disease subgroups included myocardial infarction and ischemic heart disease (ICD-9-CM 410–414.9), pulmonary embolism (415.1–415.19), dysrhythmias (425–427.9), congestive heart failure and cardiomyopathies (425, 425.1–425.2, 425.4–425.9, 428–428.9), and hemorrhagic (430–432.9) and thrombotic stroke (433.437.9). Respiratory disease subgroups included pneumonia and influenza (ICD-9-CM 480–488.9), chronic obstructive pulmonary disease and allied conditions (490–496.9), and other respiratory diseases (510–519.9). Among genitourinary disorders we identified acute and chronic renal failure (584–586).
Injury subgroups included skull fractures and traumatic brain injury (ICD-9-CM 800–804.9, 850–854.9, 959.0–959.09), spinal fractures and spinal cord injuries (805–806.9, 952–952.9), limb fractures, dislocations, amputations, crush and vascular injuries (807–839.9, 885–887.7, 895–897.7, 900–904.9, 925–929), chest, abdomen and pelvis internal injuries (860–869.9) and burns (940–949.9). We identified all poisonings, drug and alcohol disorders (291–292.9, 303.0–305.9, 960–989.9).
Under symptoms, signs and ill-defined conditions, we identified the subgroup syncope, coma and altered mental status (780.0–780.2, 780.97), chest pain (786.5–786.59), and shock, hypovolemia, dehydration, hypotension and anaphylaxis (276.5–276.52, 458–458.9, 785.5–785.51, 785.59, 958.4, 995.0, 999.4).
We analyzed the data using descriptive statistics. For each of the five imputed match sets, we examined the primary discharge diagnosis, calculating the binomial proportion of cases in each major ICD-9-CM disease category and subgroup. We determined the proportion of each disease category and subgroup relative to all successful ETI as well as to admitted patients only. We analyzed the secondary diagnoses in a similar fashion, examining all eight secondary discharge diagnoses of admitted patients and calculating the proportion of each secondary disease category relative to admitted patients only.
Utilizing the primary and all eight secondary discharge diagnoses, we calculated the Charlson comorbidity index using the Stata “Charlson” module. Combining the primary and secondary diagnoses and the Charlson score allowed estimations from each of the five imputed match sets using Rubin's method, implemented using the Stata “mi” command.33
Because Rubin's method cannot combine medians, we examined the Charlson median and interquartile range for each imputed set separately. We analyzed the data using Stata 11.2 (Stata, Inc., College Station, Texas).