Length of stay and hospital costs among patients admitted to hospital by family physicians

Study setting and population

St Paul’s Hospital is an acute care, teaching, and research hospital that serves an urban population in downtown Vancouver. This study identified all adult admissions to the family practice inpatient service between April 1, 2006, and June 30, 2008. Data were obtained from Health Record Services at Providence Health Care, which maintains records of all admissions and discharges for St Paul’s Hospital. Ethics approval was obtained from the research ethics boards of the University of British Columbia and Providence Health Care.

Patients presenting to the emergency department who require admission for medical conditions are referred to the family practice inpatient service if they have family physicians in the community with admitting privileges and do not require specialist care. Other clinically suitable patients are referred to this service, even if they do not have family physicians in the community with admitting privileges, if beds are available and the physicians attending the family practice inpatient service for 1-week rotations accept the referrals. Most of these attending physicians maintain community-based practices outside of the hospital, and none exclusively provide hospital-based care. In 2009, 22 different physicians (19 men, 3 women) covered 52 scheduled weeks of inpatient service. Family medicine residents provided an important care component to patients using this service. Two family practice medical residents attended during the day and 1 provided home-call coverage overnight and on weekends. Under the supervision of family practice physician staff, residents performed daily rounds on admitted patients and reviewed daily patient progress. On average, there were 13 to 15 patients on the family practice inpatient service at any given time.

Inpatient admissions to the family practice service were classified into 2 groups on the basis of the patients’ admitting physicians or most responsible physicians. Group A consisted of patients admitted under the care of the family physicians who were the patients’ usual health care providers in the community. Group B consisted of patients admitted under the care of family physicians who were not the patients’ usual health care providers but who were attending the inpatient service for 1-week rotations. Group B included patients who did not have primary care physicians in the community and patients whose primary care physicians did not have admitting privileges at the hospital.

Admissions were excluded if they were patient transfers from or to another acute care inpatient institution, had specialists listed as the most responsible physicians, or had missing or incomplete data for outcome variables. A sensitivity analysis was also performed to exclude admissions for patients who died while in hospital or who left against medical advice, as these variables could directly affect the outcomes of interest.

Data sources

Demographic data, most responsible diagnoses, date of discharge, expected LOS (ELOS), and resource intensity weight (RIW) were ascertained from the hospital’s electronic health records. Most responsible diagnosis was based on International Statistical Classification of Diseases and Related Health Problems, 10th revision, Canadian edition codes and represents the diagnosis that most contributed to the patients’ LOS. Using standardized methods, each admission was assigned an ELOS and RIW based on the patient’s age, most responsible diagnosis, and comorbidities.^5,6

Whether the patient had a family physician in the community was determined by self-report at admission. Patients’ residential postal codes were linked to neighbourhood-level mean household income quintiles using previously described methods.^7–10 Patients with no fixed address were assigned to the lowest income quintile. Admissions with postal codes that could not be converted (eg, low census income response rate or postal codes associated with an institution) were excluded from analyses that included neighbourhood-level incomes.

Outcome measures

The primary outcomes for this analysis were adjusted LOS and adjusted total hospital costs. Length of stay was defined as the number of acute days in hospital and represented the length of time the patient was being actively treated from date of admission to date of discharge. Adjustment of LOS was accomplished by either subtracting (difference between LOS and ELOS [LOS - ELOS]) or dividing by (acute LOS to ELOS ratio [LOS:ELOS]) ELOS.

Hospital-level costs were ascertained for each admission by tracking actual resource use. Hospital departments give standard unit cost information for resources and services. The sum of the actual resources used multiplied by their respective unit cost estimates the total cost of any admission. Adjusted hospital costs were expressed as hospital costs per RIW unit. These values are commonly used to facilitate comparisons of economic efficiency among health care entities.⁶

Data analysis

The patient characteristics of the 2 groups were compared using t tests or Mann-Whitney nonparametric tests (where appropriate) for continuous variables and χ² tests for categorical variables. Most responsible diagnoses for hospitalization were compared using Fisher exact tests to account for the small (< 5) expected cell counts. All outcome measures (LOS, LOS - ELOS, LOS:ELOS, total hospital costs, and total hospital costs per RIW unit) were compared using Mann-Whitney nonparametric tests, as these data were not normally distributed. A P value of .05 was considered statistically significant.

Multivariate linear regression models were constructed separately for each outcome of interest. The independent variable was admission group. Variables for patient age, sex, admission from a nursing home, and neighbourhood-level income were entered into the models to adjust for demographic characteristics. In the primary analysis, models were also adjusted for whether the patient died during hospitalization or left against medical advice. Having a primary care physician in the community was not included because of its obvious correlation with admission group.

Initial diagnostic tests suggested that assumptions of normality and constant variance (homoscedasticity) were being violated. For this reason, the regression models were reconstructed using a natural logarithm transformation of the outcome variables. Regression coefficients from the models including the transformed outcomes were converted into the percentage increase in the average value of the outcome per unit increase in the predictor using the following formula: 100(e^β - 1).¹¹ All analyses were performed using SPSS, version 16.0.