Methods
Setting
In 1997 a large regional health maintenance organization in Colorado mandated that all its inpatients be admitted by a pulmonology or critical care hospitalist team to the exclusion of their primary care physicians. Rose Medical Center, a 420-bed private community hospital in Denver, Colorado, serves as a family practice residency training site in which residents care for patients under the guidance of resident faculty and community primary care physicians. We recognized the health maintenance orgaization’s program as a natural experiment and an opportunity to address some of the design limitations of prior studies by comparing the care delivered simultaneously by these 3 inpatient models.
Subjects and study design
We conducted a retrospective cohort study of all patients admitted between April 1997 and March 1998 with a primary diagnosis of pneumonia as identified by codes from the International Classification of Diseases, Ninth Revision. We studied pneumonia care because of the high incidence of pneumonia in our institution and the existence of a valid, population-based measure of disease severity, the Pneumonia Severity Index (PSI; see Statistical Methods). In addition, focusing on 1 diagnosis allowed for a direct and detailed analysis of the process of care. To eliminate potential biases produced by different outpatient physician specialties, we excluded patients who did not have a family physician as a primary care provider. Patients also were excluded if they were younger than 18 years, had human immunodeficiency virus or acquired immunodeficiency syndrome, had exclusively nosocomial pneumonia, or had the diagnosis of pneumonia subsequently ruled out. Ninety-seven patients were included. Data were collected by standardized chart abstraction and entered into a Microsoft Access database. The chart abstraction was performed by research assistants who had no knowledge of whether the attending physician was a primary care physician or a hospitalist.
Patients were grouped based on the status of their inpatient attending physicians of record. The critical care hospitalists represented a group of subspecialty critical care pulmonologists contracted by a large health maintenance organization to care for all its inpatients. The family physician hospitalists represented rotating family medicine faculty who worked exclusively in the hospital for 8 weeks each year. Some patients under the care of family physician hospitalists originated from the residency’s continuity clinic, and others came from the practices of approximately 30 community family physicians for whom the family physician hospitalist acts as a hospitalist service. Community family medicine primary care physicians cared for their own patients and continued their outpatient practices. Family medicine house staff was involved in the care of inpatients from all groups.
Patients’ insurance plans rather than physician referral or self-referral determined whether they were assigned to the critical care hospitalist, family physician hospitalist, or primary care physician group. Accordingly, there were instances when a primary care physician had some patients admitted to the critical care hospitalist because the patient was a member of this particular health maintenance organization and also because the physician cared directly for his other patients in the hospital.
Outcome measures
We looked at 5 primary patient- and policy-oriented outcomes that have been validated as indicators of quality of inpatient care: in-hospital mortality, length of stay, hospital charges, 7-day readmissions, and 7- and 30-day returns to the emergency department.18,19 We included Colorado data from the Healthcare Cost and Utilization Project, when available, to serve as a reference standard.20 In addition, we chose 7 validated secondary “process of care”outcomes21 to further describe the practice behaviors of the 3 groups: documentation of lifestyle modification counseling (tobacco cessation, exercise, etc), documentation of end-of-life counseling, compliance with contemporary guidelines from the American Thoracic Society for treating community-acquired pneumonia,21 length of stay in intensive care, and the use of chest radiographs and blood and sputum cultures.
Statistical methods
All statistical analyses were performed with SAS version 6.12. Patient demographic and clinical characteristics were analyzed with chi-square test and analysis of variance, when appropriate. We controlled for disease severity with the PSI, a well-validated, population-based severity of illness score for inpatients with pneumonia.22,23
Multiple and logistic regressions were used to control for disease severity and potential confounders. Our models included the PSI class and those characteristics that were statistically significantly different (ie, diagnosis of hypertension) or showed a trend toward difference and were felt to be potentially clinically significant (diagnosis of diabetes, effusion on chest x-ray, mental status at admission), in addition to sex and age. Logarithmic transformations of non-normal data were conducted, when appropriate. We eliminated as extreme outliers 3 cases (2 under the care of critical care hospitalists and 1 under the care of family physician hospitalists) whose outcomes were 2 standard deviations or more beyond the mean. For example, 1 patient had a rare clotting disorder, stayed in intensive care for 30 days, required an orphan drug at $6000 per dose, and had charges well in excess of $1 million. This study was approved by the Rose Medical Center and Health One Institutional Review Board.