Catherine Boyd, PA‐C

Study Setting

The study was conducted in a 747‐bed urban, academic medical center in the northeastern United States. The hospital's human research committee reviewed and approved the study design. The hospital has accredited residency and fellowship programs in all major specialties. Prior to July 2005, physician assistants were employed only on surgical and medical subspecialty services (ie, bone marrow transplant, interventional cardiology); none were employed on the inpatient general medicine service. There were approximately 44,000 inpatient admissions during the year of the study, with approximately 6500 of these to the general medicine service.

Description of the General Medicine Service

The General Medicine Service consisted of 8 traditional house staff teams, with 1 attending, 1 junior or senior resident, 2 interns, and 1 or 2 medical students. These teams admitted patients on a rotating basis every fourth day. On 4 of these teams, the attending was a hospitalist, with clinical responsibility for the majority of the patients admitted to the team. On the remaining 4 teams, the teaching attending was a primary care physician or medical subspecialist, responsible for the direct care of a small number of the team's patients, with the remainder cared for by private primary care physicians or subspecialists.

Description of the Physician Assistant/Hospitalist Service

The Physician Assistant/Clinician Educator (PACE) service opened in July 2005, and consisted of 15 beds localized to 2 adjacent inpatient pods, staffed by a single cadre of nurses and medically staffed by 1 hospitalist and 2 physician assistants from 7:00 _AM to 7:00 _PM on weekdays and by 1 hospitalist, 1 physician assistant, and 1 moonlighter (usually a senior medical resident or fellow) from 7:00 _AM to 7:00 _PM on weekends. A moonlighter, typically a senior resident or medical subspecialty fellow, admitted patients and covered nights on the service from 7:00 _PM to 7:00 _AM 7 days a week. The daily census goal for the service was 15 patients, limited by the number of available beds on the 2 pods, and the service accepted admissions 24 hours per day, 7 days per week, whenever beds were available. Daily morning rounds occurred at 8:00 _AM and included the hospitalist, physician assistants, nurses, a care coordinator, and a pharmacist. The PACE service did not have triage guidelines related to diagnosis, complexity, or acuity, but only accepted patients via the emergency department or via a primary care physician's office, and did not accept patients transferred from outside hospitals or from the intensive care units.

Physician Assistants

All of the physician assistants on the PACE service had prior inpatient medicine experience, ranging from 6 months to 5 years. The physician assistants worked in 3‐day to 6‐day blocks of 12‐hour shifts. Their clinical responsibilities were similar to those of interns at the study hospital, and included taking histories and performing physical examinations, writing notes and orders, reviewing and assimilating data, creating and updating patient signouts, completing discharge summaries, consulting other services as needed, and communicating with nurses and family members.

Many physician assistants also had nonclinical responsibilities, taking on physician‐mentored roles in education, quality improvement, and administration. They were involved in several initiatives: (1) developing a physician assistant curriculum in hospital medicine, (2) presenting at hospital‐wide physician assistant grand rounds, (3) surveying and tracking patient and family satisfaction on the service, (4) reviewing all 72‐hour hospital readmissions, intensive care unit transfers, and deaths on the service, and (5) maintaining the service's compliance with state regulations regarding physician assistant scope of practice and prescribing.

Hospitalists

The 3 hospitalists on the PACE service worked in 7‐day blocks of 12‐hour shifts (7:00 _AM to 7:00 _PM). They directly supervised the physician assistants and had no competing responsibilities. The hospitalists were all recent graduates of the study hospital's internal medicine residency, with no prior clinical experience beyond residency. All were planning to work on the service for 1 to 2 years before beginning a subspecialty fellowship. In addition to supervising the clinical work of the physician assistants, the hospitalists were responsible for teaching the physician assistants on rounds and in weekly didactic sessions, guided by a curriculum in hospital medicine that focused on the most common general medicine diagnoses seen on the PACE service. The medical director of the PACE service periodically reviewed each physician assistant's clinical experience, skills and knowledge base, and held semiannual feedback sessions.

Study Patients

All general medicine patients admitted to the PACE service from July 1, 2005 to June 30, 2006 comprised the study population. The comparison group consisted of general medicine patients admitted to the 8 house staff general medicine teams; patients transferred from an intensive care unit (ICU) or another facility were excluded in order to match the admission criteria for the PACE service and improve comparability between the 2 study arms.

Data Collection and Study Outcomes

We obtained all patient data from the hospital's administrative databases. We identified patients assigned to the PACE service or to the comparison group based on the admitting service, team, and attending. We obtained patient demographics, insurance, admission source and discharge destination, admission and discharge times, dates, diagnoses, and diagnosis‐related groups (DRGs), as well as dates and times of transfers to other services, including to the intensive care unit. We also obtained the Medicare case‐mix index (CMI, based on DRG weight), and calculated a Charlson score based on billing diagnoses coded in the year prior to the index admission.14 Outcomes included length of stay (LOS) to the nearest hour, in‐hospital mortality, transfers to the intensive care unit, readmissions to the study hospital within 72 hours, 14 days, and 30 days, and total costs as derived from the hospital's cost accounting system (Transition Systems Inc., Boston, MA). Other outcomes included patient satisfaction as measured by responses to the Press‐Ganey survey routinely administered to a randomly selected 70% of recently discharged patients and effect on self‐reported resident work hours.

Statistical Analysis

Patient demographics, clinical characteristics, and study outcomes are presented using proportions, means with standard deviations, and medians with inter‐quartile ranges as appropriate. Unadjusted differences in outcomes between the two services were calculated using univariable regression techniques with service as the independent variable and each outcome as the dependent variable. We used logistic regression for dichotomous outcomes (readmissions, ICU transfers, and inpatient mortality), and linear regression for log‐transformed LOS and log‐transformed total costs of care. To adjust each outcome for potential confounders, we then built multivariable regression models. Each potential confounder was entered into the model one at a time as the independent variable. All variables found to be significant predictors of the outcome at the P < 0.10 level were then retained in the final model along with service as the predictor of interest. We used general estimating equations in all multivariable models to adjust for clustering of patients by attending physician. For logistic regression models, the effect size is presented as an odds ratio (OR); for log‐transformed linear regression models, the effect size is presented as the percent difference between groups. We also performed 2 subgroup analyses, limited to (1) the patients with the 10 most common discharge DRGs, and (2) patients admitted between the hours of 7:00 _AM and 7:00 _PM to remove the effects of moonlighters performing the initial admission. Except as noted above, 2‐sided P values < 0.05 were considered significant. SAS 9.1 (SAS Institute, Cary, NC) was used for all analyses.

Patient Demographics

Table 1 shows patient demographics and clinical characteristics of the PACE service and the comparison group. Patients in the comparison group were slightly older and tended to have slightly higher CMI and Charlson scores. Patients on the PACE service were more likely to be admitted at night (10:00 _PM to 7:00 _AM; 43.8% versus 30.3%; P < 0.0001). There were no significant differences in sex, race, insurance, or percentage of patients discharged to home. The 10 most common DRGs in the comparison group accounted for 37.0% of discharges, and these same DRGs accounted for 37.5% of discharges on the PACE service (Table 2).

Efficiency and Quality of Care

Table 3 compares the performance of the PACE service and the comparison group on several efficiency and quality measures. Unadjusted LOS was not significantly different, and adjusted LOS was slightly but not statistically significantly higher on the study service (adjusted LOS 5.0% higher; 95% confidence interval [CI], 0.4% to +10%). Unadjusted and adjusted total costs of care were marginally lower on the study service (adjusted total cost of care 3.9% lower; 95% CI, 7.5% to 0.3%).

We found no differences between the PACE service and comparison group in unadjusted rates of hospital readmissions within 72 hours, 14 days, and 30 days, transfer to the intensive care units, or inpatient mortality (Table 3). The associated ORs for each outcome were similar after adjusting for patient demographics and clinical characteristics including severity of illness, as well as for clustering by attending physician.

Subgroup Analyses

When the analysis was limited to the subset of patients with the 10 most common discharge DRGs, the difference in adjusted total cost of care was similar but lost statistical significance (4.0% lower on PACE service; 95% CI, 11.0% to +3.3%). In this subgroup, LOS, readmission rates, and ICU transfer rates were not different. ORs for mortality could not be calculated because there were no deaths in this subgroup on the PACE service (data not shown). When analysis was limited to daytime admissions (to remove any potential effect of admitting by a moonlighter), the difference in total cost of care was attenuated and lost statistical significance (0.2% lower on PACE service; 95%CI, 5.9% to +5.5%). No differences were seen in LOS, mortality, and ICU transfers (data not shown). However, 14‐day readmissions (but not 72‐hour or 30‐day readmissions) were lower on the PACE service (OR, 0.49; 95% CI, 0.25‐0.93).

Patient Satisfaction

Patients were similarly satisfied with their care on the PACE service and on the house staff services. In specific areas and globally, percentages of patients satisfied with their physicians and with the discharge process were not different, as measured by the Press‐Ganey survey (Press‐Ganey Associates, South Bend, IN; Figures 1 and 2). The survey distinguishes between attendings and residents, but not physician assistants; therefore, Figure 1 only includes responses to the attending questions. Given the sampling procedure of the Press‐Ganey survey, exact response rates cannot be calculated, but Press‐Ganey reports a response rate of about 40% for the English survey and about 20% for the Spanish survey.

Figure 1

Figure 2

Resident Duty Hours

Comparing the same month 1 year prior to implementation of the PACE service, mean self‐reported resident duty hours on the general medicine service were unchanged; however, self‐reported data were incomplete, and multiple changes took place in the residency program during the study period. For example, implementation of the PACE service allowed for the dissolution of one full house staff general medicine team and redistribution of these house staff to night float positions and an expanded medical intensive care unit.

Costs of Implementation

The costs associated with implementing the PACE service included physician and physician assistant salaries (2.5 full‐time physicians, 5 full‐time physician assistants, plus fringe) and night coverage by resident and fellow moonlighters (without fringe, and estimated at 50% effort given other moonlighter coverage responsibilities on subspecialty services). We estimated these costs at $257.50/patient‐day ($115/patient‐day for attending physician compensation, $110/patient‐day for physician assistant compensation, and $32.50/patient‐day for moonlighting coverage).

Study Setting

The study was conducted in a 747‐bed urban, academic medical center in the northeastern United States. The hospital's human research committee reviewed and approved the study design. The hospital has accredited residency and fellowship programs in all major specialties. Prior to July 2005, physician assistants were employed only on surgical and medical subspecialty services (ie, bone marrow transplant, interventional cardiology); none were employed on the inpatient general medicine service. There were approximately 44,000 inpatient admissions during the year of the study, with approximately 6500 of these to the general medicine service.

Description of the General Medicine Service

The General Medicine Service consisted of 8 traditional house staff teams, with 1 attending, 1 junior or senior resident, 2 interns, and 1 or 2 medical students. These teams admitted patients on a rotating basis every fourth day. On 4 of these teams, the attending was a hospitalist, with clinical responsibility for the majority of the patients admitted to the team. On the remaining 4 teams, the teaching attending was a primary care physician or medical subspecialist, responsible for the direct care of a small number of the team's patients, with the remainder cared for by private primary care physicians or subspecialists.

Description of the Physician Assistant/Hospitalist Service

The Physician Assistant/Clinician Educator (PACE) service opened in July 2005, and consisted of 15 beds localized to 2 adjacent inpatient pods, staffed by a single cadre of nurses and medically staffed by 1 hospitalist and 2 physician assistants from 7:00 _AM to 7:00 _PM on weekdays and by 1 hospitalist, 1 physician assistant, and 1 moonlighter (usually a senior medical resident or fellow) from 7:00 _AM to 7:00 _PM on weekends. A moonlighter, typically a senior resident or medical subspecialty fellow, admitted patients and covered nights on the service from 7:00 _PM to 7:00 _AM 7 days a week. The daily census goal for the service was 15 patients, limited by the number of available beds on the 2 pods, and the service accepted admissions 24 hours per day, 7 days per week, whenever beds were available. Daily morning rounds occurred at 8:00 _AM and included the hospitalist, physician assistants, nurses, a care coordinator, and a pharmacist. The PACE service did not have triage guidelines related to diagnosis, complexity, or acuity, but only accepted patients via the emergency department or via a primary care physician's office, and did not accept patients transferred from outside hospitals or from the intensive care units.

Physician Assistants

All of the physician assistants on the PACE service had prior inpatient medicine experience, ranging from 6 months to 5 years. The physician assistants worked in 3‐day to 6‐day blocks of 12‐hour shifts. Their clinical responsibilities were similar to those of interns at the study hospital, and included taking histories and performing physical examinations, writing notes and orders, reviewing and assimilating data, creating and updating patient signouts, completing discharge summaries, consulting other services as needed, and communicating with nurses and family members.

Many physician assistants also had nonclinical responsibilities, taking on physician‐mentored roles in education, quality improvement, and administration. They were involved in several initiatives: (1) developing a physician assistant curriculum in hospital medicine, (2) presenting at hospital‐wide physician assistant grand rounds, (3) surveying and tracking patient and family satisfaction on the service, (4) reviewing all 72‐hour hospital readmissions, intensive care unit transfers, and deaths on the service, and (5) maintaining the service's compliance with state regulations regarding physician assistant scope of practice and prescribing.

Hospitalists

The 3 hospitalists on the PACE service worked in 7‐day blocks of 12‐hour shifts (7:00 _AM to 7:00 _PM). They directly supervised the physician assistants and had no competing responsibilities. The hospitalists were all recent graduates of the study hospital's internal medicine residency, with no prior clinical experience beyond residency. All were planning to work on the service for 1 to 2 years before beginning a subspecialty fellowship. In addition to supervising the clinical work of the physician assistants, the hospitalists were responsible for teaching the physician assistants on rounds and in weekly didactic sessions, guided by a curriculum in hospital medicine that focused on the most common general medicine diagnoses seen on the PACE service. The medical director of the PACE service periodically reviewed each physician assistant's clinical experience, skills and knowledge base, and held semiannual feedback sessions.

Study Patients

All general medicine patients admitted to the PACE service from July 1, 2005 to June 30, 2006 comprised the study population. The comparison group consisted of general medicine patients admitted to the 8 house staff general medicine teams; patients transferred from an intensive care unit (ICU) or another facility were excluded in order to match the admission criteria for the PACE service and improve comparability between the 2 study arms.

Data Collection and Study Outcomes

We obtained all patient data from the hospital's administrative databases. We identified patients assigned to the PACE service or to the comparison group based on the admitting service, team, and attending. We obtained patient demographics, insurance, admission source and discharge destination, admission and discharge times, dates, diagnoses, and diagnosis‐related groups (DRGs), as well as dates and times of transfers to other services, including to the intensive care unit. We also obtained the Medicare case‐mix index (CMI, based on DRG weight), and calculated a Charlson score based on billing diagnoses coded in the year prior to the index admission.14 Outcomes included length of stay (LOS) to the nearest hour, in‐hospital mortality, transfers to the intensive care unit, readmissions to the study hospital within 72 hours, 14 days, and 30 days, and total costs as derived from the hospital's cost accounting system (Transition Systems Inc., Boston, MA). Other outcomes included patient satisfaction as measured by responses to the Press‐Ganey survey routinely administered to a randomly selected 70% of recently discharged patients and effect on self‐reported resident work hours.

Statistical Analysis

Patient demographics, clinical characteristics, and study outcomes are presented using proportions, means with standard deviations, and medians with inter‐quartile ranges as appropriate. Unadjusted differences in outcomes between the two services were calculated using univariable regression techniques with service as the independent variable and each outcome as the dependent variable. We used logistic regression for dichotomous outcomes (readmissions, ICU transfers, and inpatient mortality), and linear regression for log‐transformed LOS and log‐transformed total costs of care. To adjust each outcome for potential confounders, we then built multivariable regression models. Each potential confounder was entered into the model one at a time as the independent variable. All variables found to be significant predictors of the outcome at the P < 0.10 level were then retained in the final model along with service as the predictor of interest. We used general estimating equations in all multivariable models to adjust for clustering of patients by attending physician. For logistic regression models, the effect size is presented as an odds ratio (OR); for log‐transformed linear regression models, the effect size is presented as the percent difference between groups. We also performed 2 subgroup analyses, limited to (1) the patients with the 10 most common discharge DRGs, and (2) patients admitted between the hours of 7:00 _AM and 7:00 _PM to remove the effects of moonlighters performing the initial admission. Except as noted above, 2‐sided P values < 0.05 were considered significant. SAS 9.1 (SAS Institute, Cary, NC) was used for all analyses.

Patient Demographics

Table 1 shows patient demographics and clinical characteristics of the PACE service and the comparison group. Patients in the comparison group were slightly older and tended to have slightly higher CMI and Charlson scores. Patients on the PACE service were more likely to be admitted at night (10:00 _PM to 7:00 _AM; 43.8% versus 30.3%; P < 0.0001). There were no significant differences in sex, race, insurance, or percentage of patients discharged to home. The 10 most common DRGs in the comparison group accounted for 37.0% of discharges, and these same DRGs accounted for 37.5% of discharges on the PACE service (Table 2).

Efficiency and Quality of Care

Table 3 compares the performance of the PACE service and the comparison group on several efficiency and quality measures. Unadjusted LOS was not significantly different, and adjusted LOS was slightly but not statistically significantly higher on the study service (adjusted LOS 5.0% higher; 95% confidence interval [CI], 0.4% to +10%). Unadjusted and adjusted total costs of care were marginally lower on the study service (adjusted total cost of care 3.9% lower; 95% CI, 7.5% to 0.3%).

We found no differences between the PACE service and comparison group in unadjusted rates of hospital readmissions within 72 hours, 14 days, and 30 days, transfer to the intensive care units, or inpatient mortality (Table 3). The associated ORs for each outcome were similar after adjusting for patient demographics and clinical characteristics including severity of illness, as well as for clustering by attending physician.

Subgroup Analyses

When the analysis was limited to the subset of patients with the 10 most common discharge DRGs, the difference in adjusted total cost of care was similar but lost statistical significance (4.0% lower on PACE service; 95% CI, 11.0% to +3.3%). In this subgroup, LOS, readmission rates, and ICU transfer rates were not different. ORs for mortality could not be calculated because there were no deaths in this subgroup on the PACE service (data not shown). When analysis was limited to daytime admissions (to remove any potential effect of admitting by a moonlighter), the difference in total cost of care was attenuated and lost statistical significance (0.2% lower on PACE service; 95%CI, 5.9% to +5.5%). No differences were seen in LOS, mortality, and ICU transfers (data not shown). However, 14‐day readmissions (but not 72‐hour or 30‐day readmissions) were lower on the PACE service (OR, 0.49; 95% CI, 0.25‐0.93).

Patient Satisfaction

Patients were similarly satisfied with their care on the PACE service and on the house staff services. In specific areas and globally, percentages of patients satisfied with their physicians and with the discharge process were not different, as measured by the Press‐Ganey survey (Press‐Ganey Associates, South Bend, IN; Figures 1 and 2). The survey distinguishes between attendings and residents, but not physician assistants; therefore, Figure 1 only includes responses to the attending questions. Given the sampling procedure of the Press‐Ganey survey, exact response rates cannot be calculated, but Press‐Ganey reports a response rate of about 40% for the English survey and about 20% for the Spanish survey.

Figure 1

Figure 2

Resident Duty Hours

Comparing the same month 1 year prior to implementation of the PACE service, mean self‐reported resident duty hours on the general medicine service were unchanged; however, self‐reported data were incomplete, and multiple changes took place in the residency program during the study period. For example, implementation of the PACE service allowed for the dissolution of one full house staff general medicine team and redistribution of these house staff to night float positions and an expanded medical intensive care unit.

Costs of Implementation

The costs associated with implementing the PACE service included physician and physician assistant salaries (2.5 full‐time physicians, 5 full‐time physician assistants, plus fringe) and night coverage by resident and fellow moonlighters (without fringe, and estimated at 50% effort given other moonlighter coverage responsibilities on subspecialty services). We estimated these costs at $257.50/patient‐day ($115/patient‐day for attending physician compensation, $110/patient‐day for physician assistant compensation, and $32.50/patient‐day for moonlighting coverage).

Study Setting

The study was conducted in a 747‐bed urban, academic medical center in the northeastern United States. The hospital's human research committee reviewed and approved the study design. The hospital has accredited residency and fellowship programs in all major specialties. Prior to July 2005, physician assistants were employed only on surgical and medical subspecialty services (ie, bone marrow transplant, interventional cardiology); none were employed on the inpatient general medicine service. There were approximately 44,000 inpatient admissions during the year of the study, with approximately 6500 of these to the general medicine service.

Description of the General Medicine Service

The General Medicine Service consisted of 8 traditional house staff teams, with 1 attending, 1 junior or senior resident, 2 interns, and 1 or 2 medical students. These teams admitted patients on a rotating basis every fourth day. On 4 of these teams, the attending was a hospitalist, with clinical responsibility for the majority of the patients admitted to the team. On the remaining 4 teams, the teaching attending was a primary care physician or medical subspecialist, responsible for the direct care of a small number of the team's patients, with the remainder cared for by private primary care physicians or subspecialists.

Description of the Physician Assistant/Hospitalist Service

The Physician Assistant/Clinician Educator (PACE) service opened in July 2005, and consisted of 15 beds localized to 2 adjacent inpatient pods, staffed by a single cadre of nurses and medically staffed by 1 hospitalist and 2 physician assistants from 7:00 _AM to 7:00 _PM on weekdays and by 1 hospitalist, 1 physician assistant, and 1 moonlighter (usually a senior medical resident or fellow) from 7:00 _AM to 7:00 _PM on weekends. A moonlighter, typically a senior resident or medical subspecialty fellow, admitted patients and covered nights on the service from 7:00 _PM to 7:00 _AM 7 days a week. The daily census goal for the service was 15 patients, limited by the number of available beds on the 2 pods, and the service accepted admissions 24 hours per day, 7 days per week, whenever beds were available. Daily morning rounds occurred at 8:00 _AM and included the hospitalist, physician assistants, nurses, a care coordinator, and a pharmacist. The PACE service did not have triage guidelines related to diagnosis, complexity, or acuity, but only accepted patients via the emergency department or via a primary care physician's office, and did not accept patients transferred from outside hospitals or from the intensive care units.

Physician Assistants

All of the physician assistants on the PACE service had prior inpatient medicine experience, ranging from 6 months to 5 years. The physician assistants worked in 3‐day to 6‐day blocks of 12‐hour shifts. Their clinical responsibilities were similar to those of interns at the study hospital, and included taking histories and performing physical examinations, writing notes and orders, reviewing and assimilating data, creating and updating patient signouts, completing discharge summaries, consulting other services as needed, and communicating with nurses and family members.

Many physician assistants also had nonclinical responsibilities, taking on physician‐mentored roles in education, quality improvement, and administration. They were involved in several initiatives: (1) developing a physician assistant curriculum in hospital medicine, (2) presenting at hospital‐wide physician assistant grand rounds, (3) surveying and tracking patient and family satisfaction on the service, (4) reviewing all 72‐hour hospital readmissions, intensive care unit transfers, and deaths on the service, and (5) maintaining the service's compliance with state regulations regarding physician assistant scope of practice and prescribing.

Hospitalists

The 3 hospitalists on the PACE service worked in 7‐day blocks of 12‐hour shifts (7:00 _AM to 7:00 _PM). They directly supervised the physician assistants and had no competing responsibilities. The hospitalists were all recent graduates of the study hospital's internal medicine residency, with no prior clinical experience beyond residency. All were planning to work on the service for 1 to 2 years before beginning a subspecialty fellowship. In addition to supervising the clinical work of the physician assistants, the hospitalists were responsible for teaching the physician assistants on rounds and in weekly didactic sessions, guided by a curriculum in hospital medicine that focused on the most common general medicine diagnoses seen on the PACE service. The medical director of the PACE service periodically reviewed each physician assistant's clinical experience, skills and knowledge base, and held semiannual feedback sessions.

Study Patients

All general medicine patients admitted to the PACE service from July 1, 2005 to June 30, 2006 comprised the study population. The comparison group consisted of general medicine patients admitted to the 8 house staff general medicine teams; patients transferred from an intensive care unit (ICU) or another facility were excluded in order to match the admission criteria for the PACE service and improve comparability between the 2 study arms.

Data Collection and Study Outcomes

We obtained all patient data from the hospital's administrative databases. We identified patients assigned to the PACE service or to the comparison group based on the admitting service, team, and attending. We obtained patient demographics, insurance, admission source and discharge destination, admission and discharge times, dates, diagnoses, and diagnosis‐related groups (DRGs), as well as dates and times of transfers to other services, including to the intensive care unit. We also obtained the Medicare case‐mix index (CMI, based on DRG weight), and calculated a Charlson score based on billing diagnoses coded in the year prior to the index admission.14 Outcomes included length of stay (LOS) to the nearest hour, in‐hospital mortality, transfers to the intensive care unit, readmissions to the study hospital within 72 hours, 14 days, and 30 days, and total costs as derived from the hospital's cost accounting system (Transition Systems Inc., Boston, MA). Other outcomes included patient satisfaction as measured by responses to the Press‐Ganey survey routinely administered to a randomly selected 70% of recently discharged patients and effect on self‐reported resident work hours.

Statistical Analysis

Patient demographics, clinical characteristics, and study outcomes are presented using proportions, means with standard deviations, and medians with inter‐quartile ranges as appropriate. Unadjusted differences in outcomes between the two services were calculated using univariable regression techniques with service as the independent variable and each outcome as the dependent variable. We used logistic regression for dichotomous outcomes (readmissions, ICU transfers, and inpatient mortality), and linear regression for log‐transformed LOS and log‐transformed total costs of care. To adjust each outcome for potential confounders, we then built multivariable regression models. Each potential confounder was entered into the model one at a time as the independent variable. All variables found to be significant predictors of the outcome at the P < 0.10 level were then retained in the final model along with service as the predictor of interest. We used general estimating equations in all multivariable models to adjust for clustering of patients by attending physician. For logistic regression models, the effect size is presented as an odds ratio (OR); for log‐transformed linear regression models, the effect size is presented as the percent difference between groups. We also performed 2 subgroup analyses, limited to (1) the patients with the 10 most common discharge DRGs, and (2) patients admitted between the hours of 7:00 _AM and 7:00 _PM to remove the effects of moonlighters performing the initial admission. Except as noted above, 2‐sided P values < 0.05 were considered significant. SAS 9.1 (SAS Institute, Cary, NC) was used for all analyses.

Patient Demographics

Table 1 shows patient demographics and clinical characteristics of the PACE service and the comparison group. Patients in the comparison group were slightly older and tended to have slightly higher CMI and Charlson scores. Patients on the PACE service were more likely to be admitted at night (10:00 _PM to 7:00 _AM; 43.8% versus 30.3%; P < 0.0001). There were no significant differences in sex, race, insurance, or percentage of patients discharged to home. The 10 most common DRGs in the comparison group accounted for 37.0% of discharges, and these same DRGs accounted for 37.5% of discharges on the PACE service (Table 2).

Efficiency and Quality of Care

Table 3 compares the performance of the PACE service and the comparison group on several efficiency and quality measures. Unadjusted LOS was not significantly different, and adjusted LOS was slightly but not statistically significantly higher on the study service (adjusted LOS 5.0% higher; 95% confidence interval [CI], 0.4% to +10%). Unadjusted and adjusted total costs of care were marginally lower on the study service (adjusted total cost of care 3.9% lower; 95% CI, 7.5% to 0.3%).

We found no differences between the PACE service and comparison group in unadjusted rates of hospital readmissions within 72 hours, 14 days, and 30 days, transfer to the intensive care units, or inpatient mortality (Table 3). The associated ORs for each outcome were similar after adjusting for patient demographics and clinical characteristics including severity of illness, as well as for clustering by attending physician.

Subgroup Analyses

When the analysis was limited to the subset of patients with the 10 most common discharge DRGs, the difference in adjusted total cost of care was similar but lost statistical significance (4.0% lower on PACE service; 95% CI, 11.0% to +3.3%). In this subgroup, LOS, readmission rates, and ICU transfer rates were not different. ORs for mortality could not be calculated because there were no deaths in this subgroup on the PACE service (data not shown). When analysis was limited to daytime admissions (to remove any potential effect of admitting by a moonlighter), the difference in total cost of care was attenuated and lost statistical significance (0.2% lower on PACE service; 95%CI, 5.9% to +5.5%). No differences were seen in LOS, mortality, and ICU transfers (data not shown). However, 14‐day readmissions (but not 72‐hour or 30‐day readmissions) were lower on the PACE service (OR, 0.49; 95% CI, 0.25‐0.93).

Patient Satisfaction

Patients were similarly satisfied with their care on the PACE service and on the house staff services. In specific areas and globally, percentages of patients satisfied with their physicians and with the discharge process were not different, as measured by the Press‐Ganey survey (Press‐Ganey Associates, South Bend, IN; Figures 1 and 2). The survey distinguishes between attendings and residents, but not physician assistants; therefore, Figure 1 only includes responses to the attending questions. Given the sampling procedure of the Press‐Ganey survey, exact response rates cannot be calculated, but Press‐Ganey reports a response rate of about 40% for the English survey and about 20% for the Spanish survey.

Figure 1

Figure 2

Resident Duty Hours

Comparing the same month 1 year prior to implementation of the PACE service, mean self‐reported resident duty hours on the general medicine service were unchanged; however, self‐reported data were incomplete, and multiple changes took place in the residency program during the study period. For example, implementation of the PACE service allowed for the dissolution of one full house staff general medicine team and redistribution of these house staff to night float positions and an expanded medical intensive care unit.

Costs of Implementation

The costs associated with implementing the PACE service included physician and physician assistant salaries (2.5 full‐time physicians, 5 full‐time physician assistants, plus fringe) and night coverage by resident and fellow moonlighters (without fringe, and estimated at 50% effort given other moonlighter coverage responsibilities on subspecialty services). We estimated these costs at $257.50/patient‐day ($115/patient‐day for attending physician compensation, $110/patient‐day for physician assistant compensation, and $32.50/patient‐day for moonlighting coverage).