ABSTRACT
Background: Data on the perfomance of physician assistants in a medical intensive care unit are scarce.
Objective: To compare clinical outcomes between patients admitted to a resident-run MICU and a PA-run MICU.
Methods: Retrospective analysis of prospectively collected MICU data was performed for 5,346 patients admitted to an MICU from January 2004 through January 2007; 3,971 patients were admitted to a resident-run MICU (resident group) and 1,375 to a PA-run MICU (PA group).
Results: The groups were relatively similar, though the resident group had a higher rate of renal insufficiency (25% vs. 22%, P = .05) and the PA group had a higher rate of cerebrovascular accidents (5.6% vs. 4%, P = .02).
Hospital length of stay (LOS) was similar, with a median of 9 days in the PA group and 8 days in the resident group (P = .59). MICU LOS was slightly longer for the PA group: a median of 2.58 days (1.55-4.86) vs. 2.33 days (1.39-4.16) in the resident group. After matching by propensity score, we could not confirm this increased LOS. There was no difference in hospital mortality or in ICU mortality between the two groups either in uncontrolled or controlled analyses. Survival analyses showed no difference in 28-day survival between the two groups.
Conclusion: A PA-run MICU has similar outcomes when compared to a resident-run MICU.
In the 1960s, when it became apparent that more primary care providers were needed to meet the growing needs of the American health care system, the physician assistant (PA) and nurse practitioner (NP) professions were introduced. This movement was supported by a report from the Office of Technology Assessment in 1986, which concluded that in primary and ambulatory care, PAs and NPs are capable of providing care whose quality is equivalent to that of physicians.
1 PAs and NPs now practice in different outpatient and inpatient settings, including critical care units.
The need for these nonphysician clinicians was augmented by the implementation of the Accreditation Council for Graduate Medical Education (ACGME) duty-hour restriction for resident physicians.2 This was followed by a report from the Institute of Medicine in 2008 recommending further reduction of resident physician workloads, naps during extended shifts, and a 16-hour limit for shifts without naps.3
With the growing shortage of intensive care providers,4 many teaching hospitals hired PAs and NPs to provide care in critical care units. Several studies have shown that the tasks and activities performed by acute care physician assistants and nurse practitioners are similar to those performed by resident physicians, with similar patient outcomes in emergency departments,5,6 pediatric intensive care units,7 and trauma centers.8-10 However, data on the performance of PAs in medical critical care units are scarce.11-13
At Henry Ford Hospital (Detroit, Michigan), a tertiary care medical center, there are two medical intensive care unit (MICU) models: a resident-run MICU and a PA-run MICU, both having the same level of supervision by critical care attending physicians and critical care fellows. The aim of this study was to compare patient outcomes for the PA-run MICU team and the resident-run MICU team.
MATERIALS AND METHODS
The medical intensive care units Henry Ford Hospital has two MICU models: a 16-bed MICU run by a team of four PAs, a critical care fellow, and an attending critical care physician; and a 32-bed MICU run by two teams, each consisting of 4 to 6 second-year internal medicine residents, a critical care fellow, and an attending critical care physician. Nighttime coverage is provided by the residents on the resident-run MICU under supervision by an in-house critical care fellow, and the PA-run MICU is covered by another in-house critical care fellow. The patient population is derived from direct admissions from the emergency department, transfers from other inpatient floors in the hospital, transfers from satellite emergency departments, and transfers from other hospitals. The duties of the PAs are summarized in Table 1 and are similar to those of residents.
Data collection Data were prospectively collected on all patients admitted to all MICUs from January 2004 through January 2007 as part of our involvement with Project IMPACT Essentials (Project Impact Critical Care Data System; Society of Critical Care Medicine; Des Plaines, IL). The study design was approved by the Institutional Review Board of the Henry Ford Health Sciences Center. Demographic variables collected on all patients included age, gender, admitting diagnoses, patient origin prior to ICU admission, and insurance status. Clinical data collected on admission included all data required to calculate the mortality prediction model at time 0 (MPM-0). These variables include whether the patient had any of the following events around the time of admission: tachycardia (heart rate >150 beats per minute), arrhythmia, hypotension (systolic BP <90 mm Hg), cerebrovascular accident (CVA), coma, intracerebral mass, GI bleeding, and need for mechanical ventilation. The variables also included whether patients received cardiopulmonary resuscitation (CPR) in the 24 hours prior to admission and whether they had liver cirrhosis, elevated serum creatinine (≤2 mg/dL) levels, or metastatic cancer. Other data collected on admission were the name of the staff physician and whether the patient was being readmitted to the ICU.
The outcome measures were ICU length of stay (LOS), hospital LOS, ICU mortality, and hospital mortality. ICU and hospital discharge destinations were also collected. The cohort was divided into two groups; patients admitted to the PA-run MICU (PA group) and patients admitted to the resident-run MICU (resident group). Patients were excluded from the study if an MPM-0 score could not be calculated based on the admission diagnosis.
Data analysis Statistical analysis was performed to compare the PA group and the resident group. Baseline characteristics were compared using the unpaired t test for continuous variables and the chi-square test for dichotomous variables. Data that were not normally distributed, such as ICU and hospital LOS, were analyzed using nonparametric tests, such as the Wilcoxon rank sum test. Nonparametric results were reported as median and interquartile ranges. All multivariate analyses were done using a backward stepwise selection process. In this process, group was always maintained in the analyses; but all other variables were removed, starting with the variable associated with the highest P value until all remaining variables had a P ≤.05. Also, for multivariate analyses of LOS, the data were transformed using the natural log so as to better satisfy assumptions of normality. Analyses evaluating length of stay were done after excluding patients who died in the ICU or in the hospital. Mortality outcomes were also evaluated using multivariate logistic regression. To evaluate the impact of team allocation on survival, survival curves were generated and compared using the log-rank test. Finally, the Cox proportional hazards model was used to adjust for the observed differences in baseline characteristics between the two groups in evaluating survival. An a level <.05 was considered to indicate statistical significance. Although there were multiple comparisons between groups, we did not correct for these multiple comparisons so as to have the greatest opportunity to highlight potential differences between these two groups.