User login
Physician Assistant‐Based General Medical Inpatient Care
In 2003 the Accreditation Council for Graduate Medical Education (ACGME) prescribed residency reform in the form of work hour restrictions without prescribing alternatives to resident based care.1 As a response, many academic medical centers have developed innovative models for providing inpatient care, some of which incorporate Physician Assistants (PAs).2 With further restrictions in resident work hours possible,3 teaching hospitals may increase use of these alternate models to provide inpatient care. Widespread implementation of such new and untested models could impact the care of the approximately 20 million hospitalizations that occur every year in US teaching hospitals.4
Few reports have compared the care delivered by these alternate models with the care provided by traditional resident‐based models of care.58 Roy et al.8 have provided the only recent comparison of a PA‐based model of care with a resident‐based model. They showed lower adjusted costs of inpatient care associated with PA based care but other outcomes were similar to resident‐based teams.
The objective of this study is to provide a valid and usable comparison of the outcomes of a hospitalist‐PA (H‐PA) model of inpatient care with the traditional resident‐based model. This will add to the quantity and quality of the limited research on PA‐based inpatient care, and informs the anticipated increase in the involvement of PAs in this arena.
Methods
Study Design and Setting
We conducted a retrospective cohort study at a 430‐bed urban academic medical center in the Midwestern United States.
Models of General Medical (GM) Inpatient Care at the Study Hospital During the Study Period
In November 2004, as a response to the ACGME‐mandated work hour regulations, we formed 2 Hospitalist‐PA teams (H‐PA) to supplement the 6 preexisting general medicine resident teams (RES).
The H‐PA and RES teams differed in staffing, admitting times and weekend/overnight cross coverage structure (Table 1). There were no predesigned differences between the teams in the ward location of their patients, availability of laboratory/radiology services, specialty consultation, social services/case management resources, nursing resources or documentation requirements for admission, daily care, and discharge.
| H‐PA Teams | RES Teams | |
|---|---|---|
| Attending physician | Always a hospitalist | Hospitalist, non‐hospitalist general internist or rarely a specialist |
| Attending physician role | Supervisory for some patients (about half) and sole care provider for others. | Supervisory for all patients |
| Team composition | One attending paired with 1 PA | Attending + senior resident + (2) interns + (2‐3) medical students |
| Rotation schedule | ||
| Attending | Every 2 weeks | Every 2 weeks |
| Physician assistant | Off on weekends | |
| House staff & medical students | Every month | |
| Weekend | No new admissions & hospitalist manages all patients | Accept new admissions |
| Admission times (weekdays) | 7 AM to 3 PM | Noon to 7 AM |
| Source of admissions | Emergency room, clinics, other hospitals | Emergency room, clinics, other hospitals |
| Number of admissions (weekdays) | 4‐6 patients per day per team | Noon to 5 PM: 2 teams admit a maximum of 9 patients total |
| 5 PM to 7 AM: 3 teams admit a maximum 5 patients each. | ||
| Overnight coverageroles and responsibilities | One in‐house faculty | 3 on call interns |
| Cross‐covering 2 H‐PA teams | Cross‐covering 2 teams each | |
| Performing triage | Admitting up to 5 patients each | |
| Admitting patients if necessary | ||
| Assisting residents if necessary | ||
| General medical consultation |
Admission Schedule for H‐PA or RES Teams
The admitting schedule was designed to decrease the workload of the house staff and to do so specifically during the periods of peak educational activity (morning report, attending‐led teaching rounds, and noon report). A faculty admitting medical officer (AMO) assigned patients strictly based on the time an admission was requested. Importantly, the request for admission preceded the time of actual admission recorded when the patient reached the ward. The time difference between request for admission and actual admission depended on the source of admission and the delay associated with assigning a patient room. The AMO assigned 8 to 12 new patients to the H‐PA teams every weekday between 7 AM and 3 PM and to the RES teams between noon and 7 AM the next day. There was a designed period of overlap from noon to 3 PM during which both H‐PA and RES teams could admit patients. This period allowed for flexibility in assigning patients to either type of team depending on their workload. The AMO did not use patient complexity or teaching value to assign patients.
Exceptions to Admission Schedule
Patients admitted overnight after the on call RES had reached their admission limits were assigned to H‐PA teams the next morning. In addition, recently discharged patients who were readmitted while the discharging hospitalist (H‐PA teams) or the discharging resident (RES teams) was still scheduled for inpatient duties, were assigned back to the discharging team irrespective of the admitting schedule.
The same medicine team cared for a patient from admission to discharge but on transfer to the intensive care unit (ICU), an intensivist led critical care team assumed care. On transfer out of the ICU these patients were assigned back to the original team irrespective of admitting schedulethe so called bounce back rule to promote inpatient continuity of care. But if the residents (RES teams) or the hospitalist (H‐PA teams) had changedthe bounce back rule was no longer in effect and these patients were assigned to a team according to the admission schedule.
Study Population and Study Period
We included all hospitalizations of adult patients to GM teams if both their date of admission and their date of discharge fell within the study period (January 1, 2005 to December 31, 2006). We excluded hospitalizations with admissions during the weekendwhen H‐PA teams did not admit patients; hospitalizations to GM services with transfer to nonGM service (excluding ICU) and hospitalizations involving comanagement with specialty servicesas the contribution of GM teams for these was variable; and hospitalizations of private patients.
Data Collection and Team Assignment
We collected patient data from our hospital's discharge abstract database. This database did not contain team information so to assign teams we matched the discharging attending and the day of discharge to the type of team that the discharging attending was leading that day.
We collected patient age, gender, race, insurance status, zip‐code, primary care provider, source of admission, ward type, time and day of admission, and time and day of discharge for use as independent variables. The time of admission captured in the database was the time of actual admission and not the time the admission was requested.
We grouped the principal diagnosis International Statistical Classification of Diseases and Related Health Problems, 9th edition (ICD‐9) codes into clinically relevant categories using the Clinical Classification Software.9 We created comorbidity measures using Healthcare Cost and Utilization Project Comorbidity Software, version 3.4.10
Outcome Measures
We used length of stay (LOS), charges, readmissions within 7, 14, and 30 days and inpatient mortality as our outcome measures. We calculated LOS by subtracting the discharge day and time from the admission day and time. The LOS included time spent in the ICU. We summed all charges accrued during the entire hospitalization including any stay in the ICU but did not include professional fees. We considered any repeat hospitalization to our hospital within 7, 14, and 30 days following a discharge to be a readmission except that we excluded readmissions for a planned procedure or for inpatient rehabilitation.
Statistical Analysis
Descriptive Analysis
We performed unadjusted descriptive statistics at the level of an individual hospitalization using medians and interquartile ranges for continuous data and frequencies and percentages for categorical data. We used chi‐square tests of association and KruskalWallis analysis of variance to compare H‐PA and RES teams.
Missing Data
Because we lacked data on whether a primary outpatient care provider was available for 284 (2.9%) of our study hospitalizations, we dropped them from our multivariable analyses. We used an arbitrary discharge time of noon for the 11 hospitalizations which did not have a discharge time recorded.
Multivariable Analysis
We used multivariable mixed models to risk adjust for a wide variety of variables. We included age, gender, race, insurance, presence of primary care physician, and total number of comorbidities as fixed effects in all models because of the high face validity of these variables. We then added admission source, ward, time, day of week, discharge day of week, and comorbidity measures one by one as fixed effects, including them only if significant at P < 0.01. For assessing LOS, charges, and readmissions, we added a variable identifying each patient as a random effect to account for multiple admissions for the same patient. We then added variables identifying attending physician, principal diagnostic group, and ZIP code of residence as random effects to account for clustering of hospitalizations within these categories, including them only if significant at P < 0.01. For the model assessing mortality we included variables for attending physician, principal diagnostic group, and ZIP code of residence as random effects if significant at P < 0.01. We log transformed LOS and charges because they were extremely skewed in nature. Readmissions were analyzed after excluding patients who died or were discharged alive within 7, 14, or 30 days of the end of the study period.
Sensitivity Analyses
To assess the influence of LOS outliers, we changed LOS to 6 hours if it was less than 6 hours, and 45 days if it was more than 45 daysa process called winsorizing. We consider winsorizing superior to dropping outliers because it acknowledges that outliers contribute information, but prevent them from being too influential. We chose the 6 hour cut off because we believed that was the minimum time required to admit and then discharge a patient. We chose the upper limit of 45 days on reviewing the frequency distribution for outliers. Similarly, we winsorized charges at the first and 99th percentile after reviewing the frequency distribution for outliers. We then log transformed the winsorized data before analysis.
Inpatient deaths reduce the LOS and charges associated with a hospitalization. Thus excess mortality may provide a false concession in terms of lower LOS or charges. To check if this occurred in our study we repeated the analyses after excluding inpatient deaths.
ICU stays are associated with higher LOS, charges, and mortality. In our model of care, some patients transferred to the ICU are not cared for by the original team on transfer out. Moreover, care in the ICU is not controlled by the team that discharges them. Since this might obscure differences in outcomes achieved by RES vs. H‐PA teams, we repeated these analyses after excluding hospitalizations with an ICU stay.
Since mortality can only occur during 1 hospitalization per patient, we repeated the mortality analysis using only each patient's first admission or last admission and using a randomly selected single admission for each patient.
Subgroup Analysis
To limit the effect of different physician characteristics on H‐PA and RES teams we separately analyzed the hospitalizations under the care of hospitalists who served on both H‐PA and RES teams.
To limit the effect of different admission schedules of H‐PA and RES teams we analyzed the hospitalizations with admission times between 11.00 AM and 4.00 PM. Such hospitalizations were likely to be assigned during the noon to 3 PM period when they could be assigned to either an H‐PA or RES team.
Interactions
Finally we explored interactions between the type of team and the fixed effect variables included in each model.
Statistical Software
We performed the statistical analysis using SAS software version 9.0 for UNIX (SAS Institute, Inc., Cary, NC) and R software (The R Project for Statistical Computing).
This study protocol was approved by the hospital's institutional review board.
Results
Study Population
Of the 52,391 hospitalizations to our hospital during the study period, 13,058 were admitted to general medicine. We excluded 3102 weekend admissions and 209 who met other exclusion criteria. We could not determine the team assignment for 66. Of the remaining 9681 hospitalizations, we assigned 2171 to H‐PA teams and 7510 to RES teams (Figure 1).
Descriptive Analysis
We compare patients assigned to H‐PA and RES teams in Table 2. They were similar in age, gender, race, having a primary care provider or not, and insurance status. Clinically, they had similar comorbidities and a similar distribution of common principal diagnoses. Consistent with their admitting schedule, H‐PA teams admitted and discharged more patients earlier in the day and admitted more patients earlier in the work week. Patients cared for by H‐PA teams were admitted from the Emergency Room (ER) less often and were more likely to reside on wards designated as nonmedicine by nursing specialty. Hospitalizations to H‐PA teams more often included an ICU stay.
| H‐PA (n = 2171) | RES (n = 7510) | P Value | |
|---|---|---|---|
| |||
| Age | |||
| Mean | 56.80 | 57.04 | |
| Median | 56 | 56 | 0.15 |
| Interquartile range | 43‐72 | 43‐73 | |
| Age group (years), n (%) | |||
| < 20 | 10 (0.5) | 57 (0.8) | |
| 20‐29 | 186 (8.6) | 632 (8.7) | |
| 30‐39 | 221 (10.2) | 766 (10.3) | |
| 40‐49 | 387 (17.8) | 1341 (18.1) | |
| 50‐59 | 434 (20.0) | 1492 (20.2) | 0.28 |
| 60‐69 | 325 (15.0) | 974 (12.8) | |
| 70‐79 | 271 (12.5) | 1035 (13.6) | |
| 80‐89 | 262 (12.0) | 951(12.3) | |
| 90< | 75 (3.5) | 262 (3.4) | |
| Female, n (%) | 1175 (54.1) | 4138 (55.1) | 0.42 |
| Race, n (%) | |||
| White | 1282 (59.1) | 4419 (58.9) | |
| Black | 793 (36.5) | 2754 (36.7) | 0.98 |
| Other | 96 (4.4) | 337 (4.5) | |
| Primary care provider, n (%) | 0.16 | ||
| Yes | 1537 (73.2) | 5451 (74.7) | |
| Missing: 284 | 71 (3.3) | 213 (2.8) | |
| Insurance status, n (%) | |||
| Commercial/worker's comp | 440 (20.3) | 1442 (19.2) | |
| Medicare | 1017 (46.8) | 3589 (47.8) | 0.52 |
| Medicaid/others | 714 (32.9) | 2479 (33.0) | |
| Time of admission, n (%) | |||
| 0000‐0259 | 167 (7.7) | 1068 (14.2) | |
| 0300‐0559 | 244 (11.2) | 485 (6.5) | |
| 0600‐0859 | 456 (21.0) | 270 (3.6) | |
| 0900‐1159 | 782 (36.0) | 1146 (15.3) | <0.001 |
| 1200‐1459 | 299 (13.8) | 1750 (23.3) | |
| 1500‐1759 | 155 (7.1) | 1676 (22.3) | |
| 1800‐2359 | 68 (3.1) | 1115 (14.9) | |
| Time of discharge, n (%) | |||
| 2100‐0859 | 36 (1.7) | 174 (2.3) | |
| 0900‐1159 | 275 (12.7) | 495 (6.6) | |
| 1200‐1459 | 858 (39.6) | 2608 (34.8) | <0.001 |
| 1500‐1759 | 749 (34.6) | 3122 (41.6) | |
| 1800‐2059 | 249 (11.5) | 1104 (14.7) | |
| Missing | 4 | 7 | |
| Day of week of admission, n (%) | |||
| Monday | 462 (21.3) | 1549 (20.6) | |
| Tuesday | 499 (23.0) | 1470 (19.6) | |
| Wednesday | 430 (19.8) | 1479 (19.7) | 0.001 |
| Thursday | 400 (18.4) | 1482 (19.7) | |
| Friday | 380 (17.5) | 1530 (20.4) | |
| Day of week of discharge, n (%) | |||
| Monday | 207 (9.5) | 829 (11.0) | |
| Tuesday | 268 (12.3) | 973 (13.0) | |
| Wednesday | 334 (15.4) | 1142 (15.2) | |
| Thursday | 362 (16.7) | 1297 (17.3) | 0.16 |
| Friday | 485 (22.3) | 1523 (20.3) | |
| Saturday | 330 (15.2) | 1165 (15.5) | |
| Sunday | 185 (8.5) | 581 (7.7) | |
| Admit to non‐medicine wards, n (%) | 1332 (61.4) | 2624 (34.9) | <0.001 |
| Transfer to ICU (at least once), n (%) | 299 (13.8) | 504 (6.7) | <0.001 |
| Admit from ER No (%) | 1663 (76.6) | 6063 (80.7) | <0.001 |
| 10 most frequent diagnosis (%) | Pneumonia (4.9) | Pneumonia (5.5) | |
| Congestive heart failure; nonhypertensive (4.2) | Congestive heart failure; nonhypertensive (3.9) | ||
| Sickle cell anemia (3.9) | Nonspecific chest pain (3.7) | ||
| Chronic obstructive pulmonary disease and Bronchiectasis (3.3) | Urinary tract infections(3.6) | ||
| Diabetes mellitus with complications (3.2) | Skin and subcutaneous tissue infections (3.3) | ||
| Urinary tract infections (3.2) | Sickle cell anemia (3.3) | ||
| Asthma (3.0) | Pancreatic disorders (not diabetes) (2.8) | ||
| Nonspecific chest pain (3.0) | Asthma (2.8) | ||
| Pancreatic disorders (not diabetes) (2.9) | Chronic obstructive pulmonary disease and Bronchiectasis (2.6) | ||
| Septicemia (2.2) | Diabetes mellitus with complications (2.6) | ||
| Average number of comorbidities mean (95% CI) | 0.39 (0.37‐0.42) | 0.38 (0.36‐0.39) | 0.23 |
In unadjusted comparisons of outcomes (Table 3), hospitalizations on H‐PA teams had higher lengths of stay and charges than hospitalizations on RES teams, possibly higher inpatient mortality rates but similar unadjusted readmission rates at 7, 14, and 30 days
| H‐PA (n = 2171) | RES (n = 7150) | % Difference* (CI) | P Value | |
|---|---|---|---|---|
| ||||
| LOS | Median (IQR) | Median (IQR) | ||
| Days | 3.17 (2.03‐5.30) | 2.99 (1.80‐5.08) | +8.9% (4.71‐13.29%) | <0.001 |
| Charges | ||||
| US Dollars | 9390 (6196‐16,239) | 9044 (6106‐14,805) | +5.56% (1.96‐9.28%) | 0.002 |
| Readmissions | n (%) | n (%) | Odds Ratio (CI) | |
| Within 7 days | 147 (6.96) | 571 (7.78) | 0.88 (0.73‐1.06) | 0.19 |
| Within14 days | 236 (11.34) | 924 (12.76) | 0.87 (0.75‐1.01) | 0.07 |
| Within 30 days | 383 (18.91) | 1436 (20.31) | 0.91 (0.80‐1.03) | 0.14 |
| Inpatient deaths | 39 (1.8) | 95 (1.3) | 1.36 (0.90‐2.00) | 0.06 |
Multivariable Analysis
LOS
Hospitalizations to H‐PA teams were associated with a 6.73% longer LOS (P = 0.005) (Table 4). This difference persisted when we used the winsorized data (6.45% increase, P = 0.006), excluded inpatient deaths (6.81% increase, P = 0.005), or excluded hospitalizations that involved an ICU stay (6.40%increase, P = 0.011) (Table 5).
| Overall | Subgroup: Restricted to Physicians Attending on Both H‐PA and RES Teams* | Subgroup: Restricted to Hospitalizations Between 11.00 AM and 4.00 PM | ||||
|---|---|---|---|---|---|---|
| % Difference (CI) | P Value | % Difference (CI) | P Value | % Difference (CI) | P Value | |
| ||||||
| LOS | 6.73% (1.99% to 11.70%) | 0.005 | 5.44% (0.65% to 11.91%) | 0.08 | 2.97% (4.47% to 10.98%) | 0.44 |
| Charges | 2.75% (1.30% to 6.97%) | 0.19 | 1.55% (3.76% to 7.16%) | 0.57 | 6.45% (0.62% to 14.03%) | 0.07 |
| Risk of Readmission | Adjusted OR (95%CI) | P Value | Adjusted OR (95% CI) | P Value | Adjusted OR (95% CI) | P Value |
| Within 7 days | 0.88 (0.64‐1.20) | 0.42 | 0.74 (0.40‐1.35) | 0.32 | 0.90 (0.40‐2.00) | 0.78 |
| Within14 days | 0.90 (0.69‐1.19) | 0.46 | 0.71 (0.51‐0.99) | 0.05 | 0.87 (0.36‐2.13) | 0.77 |
| Within 30 days | 0.89 (0.75‐1.06) | 0.20 | 0.75 (0.51‐1.08) | 0.12 | 0.92 (0.55‐1.54) | 0.75 |
| Inpatient mortality | 1.27 (0.82‐1.97) | 0.28 | 1.46 (0.67‐3.17) | 0.33 | 1.14 (0.47‐2.74) | 0.77 |
| Analysis With Winsorized Data | Analysis After Excluding Inpatient Deaths | Analysis After Excluding Patients With ICU Stays | ||||
|---|---|---|---|---|---|---|
| % Difference (CI) | P Value | % Difference (CI) | P Value | % Difference (CI) | P Value | |
| ||||||
| LOS | 6.45% (4.04 to 8.91%) | 0.006 | 6.81% (2.03 to 11.80%) | 0.005 | 6.40% (1.46 to 11.58%) | 0.011 |
| Charges | 2.67 (1.27 to 6.76%) | 0.187 | 2.89% (1.16 to 7.11%) | 0.164 | 0.74% (3.11 to 4.76%) | 0.710 |
Charges
Hospitalizations to H‐PA and RES teams were associated with similar charges (Table 4). The results were similar when we used winsorized data, excluded inpatient deaths or excluded hospitalizations involving an ICU stay (Table 5).
Readmissions
The risk of readmission at 7, 14, and 30 days was similar between hospitalizations to H‐PA and RES teams (Table 4).
Mortality
The risk of inpatient death was similar between all hospitalizations to H‐PA and RES teams or only hospitalizations without an ICU stay (Table 4). The results also remained the same in analyses restricted to first admissions, last admissions, or 1 randomly selected admission per patient.
Sub‐Group Analysis
On restricting the multivariable analyses to the subset of hospitalists who staffed both types of teams (Table 4), the increase in LOS associated with H‐PA care was no longer significant (5.44% higher, P = 0.081). The charges, risk of readmission at 7 and 30 days, and risk of inpatient mortality remained similar. The risk of readmission at 14 days was slightly lower following hospitalizations to H‐PA teams (odds ratio 0.71, 95% confidence interval [CI] 0.51‐0.99).
The increase in LOS associated with H‐PA care was further attenuated in analyses of the subset of admissions between 11.00 AM and 4.00 PM (2.97% higher, P = 0.444). The difference in charges approached significance (6.45% higher, P = 0.07), but risk of readmission at 7, 14, and 30 days and risk of inpatient mortality were no different (Table 4).
Interactions
On adding interaction terms between the team assignment and the fixed effect variables in each model we detected that the effect of H‐PA care on LOS (P < 0.001) and charges (P < 0.001) varied by time of admission (Figure 2a and b). Hospitalizations to H‐PA teams from 6.00 PM to 6.00 AM had greater relative increases in LOS as compared to hospitalizations to RES teams during those times. Similarly, hospitalizations during the period 3.00 PM to 3.00 AM had relatively higher charges associated with H‐PA care compared to RES care.
Discussion
We found that hospitalizations to our H‐PA teams had longer LOS but similar charges, readmission rates, and mortality as compared to traditional resident‐based teams. These findings were robust to multiple sensitivity and subgroup analyses but when we examined times when both types of teams could receive admissions, the difference in LOS was markedly attenuated and nonsignificant.
We note that most prior reports comparing PA‐based models of inpatient care predate the ACGME work hour regulations. In a randomized control trial (1987‐1988) Simmer et al.5 showed lower lengths of stay and charges but possibly higher risk of readmission for PA based teams as compared to resident based teams. Van Rhee et al.7 conducted a nonrandomized retrospective cohort study (1994‐1995) using administrative data which showed lower resource utilization for PA‐based inpatient care. Our results from 2005 to 2006 reflect the important changes in the organization and delivery of inpatient care since these previous investigations.
Roy et al.8 report the only previously published comparison of PA and resident based GM inpatient care after the ACGME mandated work hour regulations. They found PA‐based care was associated with lower costs, whereas we found similar charges for admissions to RES and H‐PA teams. They also found that LOS was similar for PA and resident‐based care, while we found a higher LOS for admissions to our H‐PA team. We note that although the design of Roy's study was similar to our own, patients cared for by PA‐based teams were geographically localized in their model. This may contribute to the differences in results noted between our studies.
Despite no designed differences in patients assigned to either type of team other than time of admission we noted some differences between the H‐PA and RES teams in the descriptive analysis. These differences, such as a higher proportion of hospitalizations to H‐PA teams being admitted from the ER, residing on nonmedicine wards or having an ICU stay are likely a result of our system of assigning admissions to H‐PA teams early during the workday. For example patients on H‐PA teams were more often located on nonmedicine wards as a result of later discharges and bed availability on medicine wards. The difference that deserves special comment is the much higher proportion (13.8% vs. 6.7%) of hospitalizations with an ICU stay on the H‐PA teams. Hospitalizations directly to the ICU were excluded from our study which means that the hospitalizations with an ICU stay in our study were initially admitted to either H‐PA or RES teams and then transferred to the ICU. Transfers out of the ICU usually occur early in the workday when H‐PA teams accepted patients per our admission schedule. These patients may have been preferentially assigned to H‐PA teams, if on returning from the ICU the original team's resident had changed (and the bounce back rule was not in effect). Importantly, the conclusions of our research are not altered on controlling for this difference in the teams by excluding hospitalizations with an ICU stay.
Hospitalizations to H‐PA teams were associated with higher resource utilization if they occurred later in the day or overnight (Figure 2a and b). During these times a transition of care occurred shortly after admission. For a late day admission the H‐PA teams would transfer care for overnight cross cover soon after the admission and for patients admitted overnight as overflow they would assume care of a patient from the nighttime covering physician performing the admission. On the other hand, on RES teams, interns admitting patients overnight continued to care for their patients for part of the following day (30‐hour call). Similar findings of higher resource utilization associated with transfer of care after admission in the daytime11 and nighttime12 have been previously reported. An alternative hypothesis for our findings is that the hospital maybe busier and thus less efficient during times when H‐PA teams had to admit later in the day or accept patients admitted overnight as overflow. Future research to determine the cause of this significant interaction between team assignment and time of admission on resource utilization is important as the large increases in LOS (up to 30%) and charges (up to 50%) noted, could have a potentially large impact if a higher proportion of hospitalizations were affected by this phenomenon.
Our H‐PA teams were assigned equally complex patients as our RES teams, in contrast to previous reports.8, 13 This was accomplished while improving the resident's educational experience and we have previously reported increases in our resident's board pass rates and in‐service training exam scores with that introduction of our H‐PA teams.14 We thus believe that selection of less complex patients to H‐PA teams such as ours is unnecessary and may give them a second tier status in academic settings.
Our report has limitations. It is a retrospective, nonrandomized investigation using a single institution's administrative database and has the limitations of not being able to account for unmeasured confounders, severity of illness, errors in the database, selection bias and has limited generalizability. We measured charges not actual costs,15 but we feel charges are a true reflection of relative resource use when compared between similar patients within a single institution. We also did not account for the readmissions that occur to other hospitals16 and our results do not reflect resource utilization for the healthcare system in total. For example, we could not tell if higher LOS on H‐PA teams resulted in lower readmissions for their patients in all hospitals in the region, which may reveal an overall resource savings. Additionally, we measured in‐hospital mortality and could not capture deaths related to hospital care that may occur shortly after discharge.
ACGME has proposed revised standards that may further restrict resident duty hours when they take effect in July 2011.3 This may lead to further decreases in resident‐based inpatient care. Teaching hospitals will need to continue to develop alternate models for inpatient care that do not depend on house staff. Our findings provide important evidence to inform the development of such models. Our study shows that one such model: PAs paired with hospitalists, accepting admissions early in the workday, with hospitalist coverage over the weekend and nights can care for GM inpatients as complex as those cared for by resident‐based teams without increasing readmission rates, inpatient mortality, or charges but at the cost of slightly higher LOS.
- ACGME‐Common Program Requirements for Resident Duty Hours. Available at: http://www.acgme.org/acWebsite/dutyHours/dh_ComProgrRequirmentsDutyHours0707.pdf. Accessed July 2010.
- ,,,,.Non‐housestaff medicine services in academic centers: models and challenges.J Hosp Med.2008;3(3):247–255.
- ACGME. Duty Hours: Proposed Standards for Review and comment. Available at: http://acgme‐2010standards.org/pdf/Proposed_Standards. pdf. Accessed July 22,2010.
- Agency for Health Care Policy and Research. HCUPnet: A tool for identifying, tracking, and analyzing national hospital statistics. Available at: http://hcup.ahrq.gov/HCUPnet.asp. Accessed July2010.
- ,,,,.A randomized, controlled trial of an attending staff service in general internal medicine.Med Care.1991;29(7 suppl):JS31–JS40.
- ,.Replacing an academic internal medicine residency program with a physician assistant‐‐hospitalist model: a Comparative Analysis Study.Am J Med Qual.2009;24(2):132–139.
- ,,.Resource use by physician assistant services versus teaching services.JAAPA.2002;15(1):33–42.
- ,,, et al.Implementation of a physician assistant/hospitalist service in an academic medical center: impact on efficiency and patient outcomes.J Hosp Med.2008;3(5):361–368.
- AHRQ. Clinical Classifications Software (CCS) for ICD‐9‐CM. Available at: http://www.hcup‐us.ahrq.gov/toolssoftware/ccs/ccs.jsp#overview. Accessed July2010.
- AHRQ. HCUP: Comorbidity Software, Version 3.4.;Available at: http://www.hcup‐us.ahrq.gov/toolssoftware/comorbidity/comorbidity.jsp. Accessed July2010.
- ,,, et al.Effect of short call admission on length of stay and quality of care for acute decompensated heart failure.Circulation.2008;117(20):2637–2644.
- ,,,.Post‐call transfer of resident responsibility: its effect on patient care.J Gen Intern Med.1990;5(6):501–505.
- ,,,,,.The effect of nonteaching services on the distribution of inpatient cases for internal medicine residents.Acad Med.2009:84(2):220–225.
- ,,, et al.Allowing for better resident education and improving patient care: hospitalist‐physician assistant teams fill in the gaps.J Hosp Med.2007;2[S2]:1–39.
- .The distinction between cost and charges.Ann Intern Med.1982;96(1):102–109.
- ,,.Rehospitalizations among patients in the Medicare Fee‐for‐Service Program.N Engl J Med.2009;360(14):1418–1428.
In 2003 the Accreditation Council for Graduate Medical Education (ACGME) prescribed residency reform in the form of work hour restrictions without prescribing alternatives to resident based care.1 As a response, many academic medical centers have developed innovative models for providing inpatient care, some of which incorporate Physician Assistants (PAs).2 With further restrictions in resident work hours possible,3 teaching hospitals may increase use of these alternate models to provide inpatient care. Widespread implementation of such new and untested models could impact the care of the approximately 20 million hospitalizations that occur every year in US teaching hospitals.4
Few reports have compared the care delivered by these alternate models with the care provided by traditional resident‐based models of care.58 Roy et al.8 have provided the only recent comparison of a PA‐based model of care with a resident‐based model. They showed lower adjusted costs of inpatient care associated with PA based care but other outcomes were similar to resident‐based teams.
The objective of this study is to provide a valid and usable comparison of the outcomes of a hospitalist‐PA (H‐PA) model of inpatient care with the traditional resident‐based model. This will add to the quantity and quality of the limited research on PA‐based inpatient care, and informs the anticipated increase in the involvement of PAs in this arena.
Methods
Study Design and Setting
We conducted a retrospective cohort study at a 430‐bed urban academic medical center in the Midwestern United States.
Models of General Medical (GM) Inpatient Care at the Study Hospital During the Study Period
In November 2004, as a response to the ACGME‐mandated work hour regulations, we formed 2 Hospitalist‐PA teams (H‐PA) to supplement the 6 preexisting general medicine resident teams (RES).
The H‐PA and RES teams differed in staffing, admitting times and weekend/overnight cross coverage structure (Table 1). There were no predesigned differences between the teams in the ward location of their patients, availability of laboratory/radiology services, specialty consultation, social services/case management resources, nursing resources or documentation requirements for admission, daily care, and discharge.
| H‐PA Teams | RES Teams | |
|---|---|---|
| Attending physician | Always a hospitalist | Hospitalist, non‐hospitalist general internist or rarely a specialist |
| Attending physician role | Supervisory for some patients (about half) and sole care provider for others. | Supervisory for all patients |
| Team composition | One attending paired with 1 PA | Attending + senior resident + (2) interns + (2‐3) medical students |
| Rotation schedule | ||
| Attending | Every 2 weeks | Every 2 weeks |
| Physician assistant | Off on weekends | |
| House staff & medical students | Every month | |
| Weekend | No new admissions & hospitalist manages all patients | Accept new admissions |
| Admission times (weekdays) | 7 AM to 3 PM | Noon to 7 AM |
| Source of admissions | Emergency room, clinics, other hospitals | Emergency room, clinics, other hospitals |
| Number of admissions (weekdays) | 4‐6 patients per day per team | Noon to 5 PM: 2 teams admit a maximum of 9 patients total |
| 5 PM to 7 AM: 3 teams admit a maximum 5 patients each. | ||
| Overnight coverageroles and responsibilities | One in‐house faculty | 3 on call interns |
| Cross‐covering 2 H‐PA teams | Cross‐covering 2 teams each | |
| Performing triage | Admitting up to 5 patients each | |
| Admitting patients if necessary | ||
| Assisting residents if necessary | ||
| General medical consultation |
Admission Schedule for H‐PA or RES Teams
The admitting schedule was designed to decrease the workload of the house staff and to do so specifically during the periods of peak educational activity (morning report, attending‐led teaching rounds, and noon report). A faculty admitting medical officer (AMO) assigned patients strictly based on the time an admission was requested. Importantly, the request for admission preceded the time of actual admission recorded when the patient reached the ward. The time difference between request for admission and actual admission depended on the source of admission and the delay associated with assigning a patient room. The AMO assigned 8 to 12 new patients to the H‐PA teams every weekday between 7 AM and 3 PM and to the RES teams between noon and 7 AM the next day. There was a designed period of overlap from noon to 3 PM during which both H‐PA and RES teams could admit patients. This period allowed for flexibility in assigning patients to either type of team depending on their workload. The AMO did not use patient complexity or teaching value to assign patients.
Exceptions to Admission Schedule
Patients admitted overnight after the on call RES had reached their admission limits were assigned to H‐PA teams the next morning. In addition, recently discharged patients who were readmitted while the discharging hospitalist (H‐PA teams) or the discharging resident (RES teams) was still scheduled for inpatient duties, were assigned back to the discharging team irrespective of the admitting schedule.
The same medicine team cared for a patient from admission to discharge but on transfer to the intensive care unit (ICU), an intensivist led critical care team assumed care. On transfer out of the ICU these patients were assigned back to the original team irrespective of admitting schedulethe so called bounce back rule to promote inpatient continuity of care. But if the residents (RES teams) or the hospitalist (H‐PA teams) had changedthe bounce back rule was no longer in effect and these patients were assigned to a team according to the admission schedule.
Study Population and Study Period
We included all hospitalizations of adult patients to GM teams if both their date of admission and their date of discharge fell within the study period (January 1, 2005 to December 31, 2006). We excluded hospitalizations with admissions during the weekendwhen H‐PA teams did not admit patients; hospitalizations to GM services with transfer to nonGM service (excluding ICU) and hospitalizations involving comanagement with specialty servicesas the contribution of GM teams for these was variable; and hospitalizations of private patients.
Data Collection and Team Assignment
We collected patient data from our hospital's discharge abstract database. This database did not contain team information so to assign teams we matched the discharging attending and the day of discharge to the type of team that the discharging attending was leading that day.
We collected patient age, gender, race, insurance status, zip‐code, primary care provider, source of admission, ward type, time and day of admission, and time and day of discharge for use as independent variables. The time of admission captured in the database was the time of actual admission and not the time the admission was requested.
We grouped the principal diagnosis International Statistical Classification of Diseases and Related Health Problems, 9th edition (ICD‐9) codes into clinically relevant categories using the Clinical Classification Software.9 We created comorbidity measures using Healthcare Cost and Utilization Project Comorbidity Software, version 3.4.10
Outcome Measures
We used length of stay (LOS), charges, readmissions within 7, 14, and 30 days and inpatient mortality as our outcome measures. We calculated LOS by subtracting the discharge day and time from the admission day and time. The LOS included time spent in the ICU. We summed all charges accrued during the entire hospitalization including any stay in the ICU but did not include professional fees. We considered any repeat hospitalization to our hospital within 7, 14, and 30 days following a discharge to be a readmission except that we excluded readmissions for a planned procedure or for inpatient rehabilitation.
Statistical Analysis
Descriptive Analysis
We performed unadjusted descriptive statistics at the level of an individual hospitalization using medians and interquartile ranges for continuous data and frequencies and percentages for categorical data. We used chi‐square tests of association and KruskalWallis analysis of variance to compare H‐PA and RES teams.
Missing Data
Because we lacked data on whether a primary outpatient care provider was available for 284 (2.9%) of our study hospitalizations, we dropped them from our multivariable analyses. We used an arbitrary discharge time of noon for the 11 hospitalizations which did not have a discharge time recorded.
Multivariable Analysis
We used multivariable mixed models to risk adjust for a wide variety of variables. We included age, gender, race, insurance, presence of primary care physician, and total number of comorbidities as fixed effects in all models because of the high face validity of these variables. We then added admission source, ward, time, day of week, discharge day of week, and comorbidity measures one by one as fixed effects, including them only if significant at P < 0.01. For assessing LOS, charges, and readmissions, we added a variable identifying each patient as a random effect to account for multiple admissions for the same patient. We then added variables identifying attending physician, principal diagnostic group, and ZIP code of residence as random effects to account for clustering of hospitalizations within these categories, including them only if significant at P < 0.01. For the model assessing mortality we included variables for attending physician, principal diagnostic group, and ZIP code of residence as random effects if significant at P < 0.01. We log transformed LOS and charges because they were extremely skewed in nature. Readmissions were analyzed after excluding patients who died or were discharged alive within 7, 14, or 30 days of the end of the study period.
Sensitivity Analyses
To assess the influence of LOS outliers, we changed LOS to 6 hours if it was less than 6 hours, and 45 days if it was more than 45 daysa process called winsorizing. We consider winsorizing superior to dropping outliers because it acknowledges that outliers contribute information, but prevent them from being too influential. We chose the 6 hour cut off because we believed that was the minimum time required to admit and then discharge a patient. We chose the upper limit of 45 days on reviewing the frequency distribution for outliers. Similarly, we winsorized charges at the first and 99th percentile after reviewing the frequency distribution for outliers. We then log transformed the winsorized data before analysis.
Inpatient deaths reduce the LOS and charges associated with a hospitalization. Thus excess mortality may provide a false concession in terms of lower LOS or charges. To check if this occurred in our study we repeated the analyses after excluding inpatient deaths.
ICU stays are associated with higher LOS, charges, and mortality. In our model of care, some patients transferred to the ICU are not cared for by the original team on transfer out. Moreover, care in the ICU is not controlled by the team that discharges them. Since this might obscure differences in outcomes achieved by RES vs. H‐PA teams, we repeated these analyses after excluding hospitalizations with an ICU stay.
Since mortality can only occur during 1 hospitalization per patient, we repeated the mortality analysis using only each patient's first admission or last admission and using a randomly selected single admission for each patient.
Subgroup Analysis
To limit the effect of different physician characteristics on H‐PA and RES teams we separately analyzed the hospitalizations under the care of hospitalists who served on both H‐PA and RES teams.
To limit the effect of different admission schedules of H‐PA and RES teams we analyzed the hospitalizations with admission times between 11.00 AM and 4.00 PM. Such hospitalizations were likely to be assigned during the noon to 3 PM period when they could be assigned to either an H‐PA or RES team.
Interactions
Finally we explored interactions between the type of team and the fixed effect variables included in each model.
Statistical Software
We performed the statistical analysis using SAS software version 9.0 for UNIX (SAS Institute, Inc., Cary, NC) and R software (The R Project for Statistical Computing).
This study protocol was approved by the hospital's institutional review board.
Results
Study Population
Of the 52,391 hospitalizations to our hospital during the study period, 13,058 were admitted to general medicine. We excluded 3102 weekend admissions and 209 who met other exclusion criteria. We could not determine the team assignment for 66. Of the remaining 9681 hospitalizations, we assigned 2171 to H‐PA teams and 7510 to RES teams (Figure 1).
Descriptive Analysis
We compare patients assigned to H‐PA and RES teams in Table 2. They were similar in age, gender, race, having a primary care provider or not, and insurance status. Clinically, they had similar comorbidities and a similar distribution of common principal diagnoses. Consistent with their admitting schedule, H‐PA teams admitted and discharged more patients earlier in the day and admitted more patients earlier in the work week. Patients cared for by H‐PA teams were admitted from the Emergency Room (ER) less often and were more likely to reside on wards designated as nonmedicine by nursing specialty. Hospitalizations to H‐PA teams more often included an ICU stay.
| H‐PA (n = 2171) | RES (n = 7510) | P Value | |
|---|---|---|---|
| |||
| Age | |||
| Mean | 56.80 | 57.04 | |
| Median | 56 | 56 | 0.15 |
| Interquartile range | 43‐72 | 43‐73 | |
| Age group (years), n (%) | |||
| < 20 | 10 (0.5) | 57 (0.8) | |
| 20‐29 | 186 (8.6) | 632 (8.7) | |
| 30‐39 | 221 (10.2) | 766 (10.3) | |
| 40‐49 | 387 (17.8) | 1341 (18.1) | |
| 50‐59 | 434 (20.0) | 1492 (20.2) | 0.28 |
| 60‐69 | 325 (15.0) | 974 (12.8) | |
| 70‐79 | 271 (12.5) | 1035 (13.6) | |
| 80‐89 | 262 (12.0) | 951(12.3) | |
| 90< | 75 (3.5) | 262 (3.4) | |
| Female, n (%) | 1175 (54.1) | 4138 (55.1) | 0.42 |
| Race, n (%) | |||
| White | 1282 (59.1) | 4419 (58.9) | |
| Black | 793 (36.5) | 2754 (36.7) | 0.98 |
| Other | 96 (4.4) | 337 (4.5) | |
| Primary care provider, n (%) | 0.16 | ||
| Yes | 1537 (73.2) | 5451 (74.7) | |
| Missing: 284 | 71 (3.3) | 213 (2.8) | |
| Insurance status, n (%) | |||
| Commercial/worker's comp | 440 (20.3) | 1442 (19.2) | |
| Medicare | 1017 (46.8) | 3589 (47.8) | 0.52 |
| Medicaid/others | 714 (32.9) | 2479 (33.0) | |
| Time of admission, n (%) | |||
| 0000‐0259 | 167 (7.7) | 1068 (14.2) | |
| 0300‐0559 | 244 (11.2) | 485 (6.5) | |
| 0600‐0859 | 456 (21.0) | 270 (3.6) | |
| 0900‐1159 | 782 (36.0) | 1146 (15.3) | <0.001 |
| 1200‐1459 | 299 (13.8) | 1750 (23.3) | |
| 1500‐1759 | 155 (7.1) | 1676 (22.3) | |
| 1800‐2359 | 68 (3.1) | 1115 (14.9) | |
| Time of discharge, n (%) | |||
| 2100‐0859 | 36 (1.7) | 174 (2.3) | |
| 0900‐1159 | 275 (12.7) | 495 (6.6) | |
| 1200‐1459 | 858 (39.6) | 2608 (34.8) | <0.001 |
| 1500‐1759 | 749 (34.6) | 3122 (41.6) | |
| 1800‐2059 | 249 (11.5) | 1104 (14.7) | |
| Missing | 4 | 7 | |
| Day of week of admission, n (%) | |||
| Monday | 462 (21.3) | 1549 (20.6) | |
| Tuesday | 499 (23.0) | 1470 (19.6) | |
| Wednesday | 430 (19.8) | 1479 (19.7) | 0.001 |
| Thursday | 400 (18.4) | 1482 (19.7) | |
| Friday | 380 (17.5) | 1530 (20.4) | |
| Day of week of discharge, n (%) | |||
| Monday | 207 (9.5) | 829 (11.0) | |
| Tuesday | 268 (12.3) | 973 (13.0) | |
| Wednesday | 334 (15.4) | 1142 (15.2) | |
| Thursday | 362 (16.7) | 1297 (17.3) | 0.16 |
| Friday | 485 (22.3) | 1523 (20.3) | |
| Saturday | 330 (15.2) | 1165 (15.5) | |
| Sunday | 185 (8.5) | 581 (7.7) | |
| Admit to non‐medicine wards, n (%) | 1332 (61.4) | 2624 (34.9) | <0.001 |
| Transfer to ICU (at least once), n (%) | 299 (13.8) | 504 (6.7) | <0.001 |
| Admit from ER No (%) | 1663 (76.6) | 6063 (80.7) | <0.001 |
| 10 most frequent diagnosis (%) | Pneumonia (4.9) | Pneumonia (5.5) | |
| Congestive heart failure; nonhypertensive (4.2) | Congestive heart failure; nonhypertensive (3.9) | ||
| Sickle cell anemia (3.9) | Nonspecific chest pain (3.7) | ||
| Chronic obstructive pulmonary disease and Bronchiectasis (3.3) | Urinary tract infections(3.6) | ||
| Diabetes mellitus with complications (3.2) | Skin and subcutaneous tissue infections (3.3) | ||
| Urinary tract infections (3.2) | Sickle cell anemia (3.3) | ||
| Asthma (3.0) | Pancreatic disorders (not diabetes) (2.8) | ||
| Nonspecific chest pain (3.0) | Asthma (2.8) | ||
| Pancreatic disorders (not diabetes) (2.9) | Chronic obstructive pulmonary disease and Bronchiectasis (2.6) | ||
| Septicemia (2.2) | Diabetes mellitus with complications (2.6) | ||
| Average number of comorbidities mean (95% CI) | 0.39 (0.37‐0.42) | 0.38 (0.36‐0.39) | 0.23 |
In unadjusted comparisons of outcomes (Table 3), hospitalizations on H‐PA teams had higher lengths of stay and charges than hospitalizations on RES teams, possibly higher inpatient mortality rates but similar unadjusted readmission rates at 7, 14, and 30 days
| H‐PA (n = 2171) | RES (n = 7150) | % Difference* (CI) | P Value | |
|---|---|---|---|---|
| ||||
| LOS | Median (IQR) | Median (IQR) | ||
| Days | 3.17 (2.03‐5.30) | 2.99 (1.80‐5.08) | +8.9% (4.71‐13.29%) | <0.001 |
| Charges | ||||
| US Dollars | 9390 (6196‐16,239) | 9044 (6106‐14,805) | +5.56% (1.96‐9.28%) | 0.002 |
| Readmissions | n (%) | n (%) | Odds Ratio (CI) | |
| Within 7 days | 147 (6.96) | 571 (7.78) | 0.88 (0.73‐1.06) | 0.19 |
| Within14 days | 236 (11.34) | 924 (12.76) | 0.87 (0.75‐1.01) | 0.07 |
| Within 30 days | 383 (18.91) | 1436 (20.31) | 0.91 (0.80‐1.03) | 0.14 |
| Inpatient deaths | 39 (1.8) | 95 (1.3) | 1.36 (0.90‐2.00) | 0.06 |
Multivariable Analysis
LOS
Hospitalizations to H‐PA teams were associated with a 6.73% longer LOS (P = 0.005) (Table 4). This difference persisted when we used the winsorized data (6.45% increase, P = 0.006), excluded inpatient deaths (6.81% increase, P = 0.005), or excluded hospitalizations that involved an ICU stay (6.40%increase, P = 0.011) (Table 5).
| Overall | Subgroup: Restricted to Physicians Attending on Both H‐PA and RES Teams* | Subgroup: Restricted to Hospitalizations Between 11.00 AM and 4.00 PM | ||||
|---|---|---|---|---|---|---|
| % Difference (CI) | P Value | % Difference (CI) | P Value | % Difference (CI) | P Value | |
| ||||||
| LOS | 6.73% (1.99% to 11.70%) | 0.005 | 5.44% (0.65% to 11.91%) | 0.08 | 2.97% (4.47% to 10.98%) | 0.44 |
| Charges | 2.75% (1.30% to 6.97%) | 0.19 | 1.55% (3.76% to 7.16%) | 0.57 | 6.45% (0.62% to 14.03%) | 0.07 |
| Risk of Readmission | Adjusted OR (95%CI) | P Value | Adjusted OR (95% CI) | P Value | Adjusted OR (95% CI) | P Value |
| Within 7 days | 0.88 (0.64‐1.20) | 0.42 | 0.74 (0.40‐1.35) | 0.32 | 0.90 (0.40‐2.00) | 0.78 |
| Within14 days | 0.90 (0.69‐1.19) | 0.46 | 0.71 (0.51‐0.99) | 0.05 | 0.87 (0.36‐2.13) | 0.77 |
| Within 30 days | 0.89 (0.75‐1.06) | 0.20 | 0.75 (0.51‐1.08) | 0.12 | 0.92 (0.55‐1.54) | 0.75 |
| Inpatient mortality | 1.27 (0.82‐1.97) | 0.28 | 1.46 (0.67‐3.17) | 0.33 | 1.14 (0.47‐2.74) | 0.77 |
| Analysis With Winsorized Data | Analysis After Excluding Inpatient Deaths | Analysis After Excluding Patients With ICU Stays | ||||
|---|---|---|---|---|---|---|
| % Difference (CI) | P Value | % Difference (CI) | P Value | % Difference (CI) | P Value | |
| ||||||
| LOS | 6.45% (4.04 to 8.91%) | 0.006 | 6.81% (2.03 to 11.80%) | 0.005 | 6.40% (1.46 to 11.58%) | 0.011 |
| Charges | 2.67 (1.27 to 6.76%) | 0.187 | 2.89% (1.16 to 7.11%) | 0.164 | 0.74% (3.11 to 4.76%) | 0.710 |
Charges
Hospitalizations to H‐PA and RES teams were associated with similar charges (Table 4). The results were similar when we used winsorized data, excluded inpatient deaths or excluded hospitalizations involving an ICU stay (Table 5).
Readmissions
The risk of readmission at 7, 14, and 30 days was similar between hospitalizations to H‐PA and RES teams (Table 4).
Mortality
The risk of inpatient death was similar between all hospitalizations to H‐PA and RES teams or only hospitalizations without an ICU stay (Table 4). The results also remained the same in analyses restricted to first admissions, last admissions, or 1 randomly selected admission per patient.
Sub‐Group Analysis
On restricting the multivariable analyses to the subset of hospitalists who staffed both types of teams (Table 4), the increase in LOS associated with H‐PA care was no longer significant (5.44% higher, P = 0.081). The charges, risk of readmission at 7 and 30 days, and risk of inpatient mortality remained similar. The risk of readmission at 14 days was slightly lower following hospitalizations to H‐PA teams (odds ratio 0.71, 95% confidence interval [CI] 0.51‐0.99).
The increase in LOS associated with H‐PA care was further attenuated in analyses of the subset of admissions between 11.00 AM and 4.00 PM (2.97% higher, P = 0.444). The difference in charges approached significance (6.45% higher, P = 0.07), but risk of readmission at 7, 14, and 30 days and risk of inpatient mortality were no different (Table 4).
Interactions
On adding interaction terms between the team assignment and the fixed effect variables in each model we detected that the effect of H‐PA care on LOS (P < 0.001) and charges (P < 0.001) varied by time of admission (Figure 2a and b). Hospitalizations to H‐PA teams from 6.00 PM to 6.00 AM had greater relative increases in LOS as compared to hospitalizations to RES teams during those times. Similarly, hospitalizations during the period 3.00 PM to 3.00 AM had relatively higher charges associated with H‐PA care compared to RES care.
Discussion
We found that hospitalizations to our H‐PA teams had longer LOS but similar charges, readmission rates, and mortality as compared to traditional resident‐based teams. These findings were robust to multiple sensitivity and subgroup analyses but when we examined times when both types of teams could receive admissions, the difference in LOS was markedly attenuated and nonsignificant.
We note that most prior reports comparing PA‐based models of inpatient care predate the ACGME work hour regulations. In a randomized control trial (1987‐1988) Simmer et al.5 showed lower lengths of stay and charges but possibly higher risk of readmission for PA based teams as compared to resident based teams. Van Rhee et al.7 conducted a nonrandomized retrospective cohort study (1994‐1995) using administrative data which showed lower resource utilization for PA‐based inpatient care. Our results from 2005 to 2006 reflect the important changes in the organization and delivery of inpatient care since these previous investigations.
Roy et al.8 report the only previously published comparison of PA and resident based GM inpatient care after the ACGME mandated work hour regulations. They found PA‐based care was associated with lower costs, whereas we found similar charges for admissions to RES and H‐PA teams. They also found that LOS was similar for PA and resident‐based care, while we found a higher LOS for admissions to our H‐PA team. We note that although the design of Roy's study was similar to our own, patients cared for by PA‐based teams were geographically localized in their model. This may contribute to the differences in results noted between our studies.
Despite no designed differences in patients assigned to either type of team other than time of admission we noted some differences between the H‐PA and RES teams in the descriptive analysis. These differences, such as a higher proportion of hospitalizations to H‐PA teams being admitted from the ER, residing on nonmedicine wards or having an ICU stay are likely a result of our system of assigning admissions to H‐PA teams early during the workday. For example patients on H‐PA teams were more often located on nonmedicine wards as a result of later discharges and bed availability on medicine wards. The difference that deserves special comment is the much higher proportion (13.8% vs. 6.7%) of hospitalizations with an ICU stay on the H‐PA teams. Hospitalizations directly to the ICU were excluded from our study which means that the hospitalizations with an ICU stay in our study were initially admitted to either H‐PA or RES teams and then transferred to the ICU. Transfers out of the ICU usually occur early in the workday when H‐PA teams accepted patients per our admission schedule. These patients may have been preferentially assigned to H‐PA teams, if on returning from the ICU the original team's resident had changed (and the bounce back rule was not in effect). Importantly, the conclusions of our research are not altered on controlling for this difference in the teams by excluding hospitalizations with an ICU stay.
Hospitalizations to H‐PA teams were associated with higher resource utilization if they occurred later in the day or overnight (Figure 2a and b). During these times a transition of care occurred shortly after admission. For a late day admission the H‐PA teams would transfer care for overnight cross cover soon after the admission and for patients admitted overnight as overflow they would assume care of a patient from the nighttime covering physician performing the admission. On the other hand, on RES teams, interns admitting patients overnight continued to care for their patients for part of the following day (30‐hour call). Similar findings of higher resource utilization associated with transfer of care after admission in the daytime11 and nighttime12 have been previously reported. An alternative hypothesis for our findings is that the hospital maybe busier and thus less efficient during times when H‐PA teams had to admit later in the day or accept patients admitted overnight as overflow. Future research to determine the cause of this significant interaction between team assignment and time of admission on resource utilization is important as the large increases in LOS (up to 30%) and charges (up to 50%) noted, could have a potentially large impact if a higher proportion of hospitalizations were affected by this phenomenon.
Our H‐PA teams were assigned equally complex patients as our RES teams, in contrast to previous reports.8, 13 This was accomplished while improving the resident's educational experience and we have previously reported increases in our resident's board pass rates and in‐service training exam scores with that introduction of our H‐PA teams.14 We thus believe that selection of less complex patients to H‐PA teams such as ours is unnecessary and may give them a second tier status in academic settings.
Our report has limitations. It is a retrospective, nonrandomized investigation using a single institution's administrative database and has the limitations of not being able to account for unmeasured confounders, severity of illness, errors in the database, selection bias and has limited generalizability. We measured charges not actual costs,15 but we feel charges are a true reflection of relative resource use when compared between similar patients within a single institution. We also did not account for the readmissions that occur to other hospitals16 and our results do not reflect resource utilization for the healthcare system in total. For example, we could not tell if higher LOS on H‐PA teams resulted in lower readmissions for their patients in all hospitals in the region, which may reveal an overall resource savings. Additionally, we measured in‐hospital mortality and could not capture deaths related to hospital care that may occur shortly after discharge.
ACGME has proposed revised standards that may further restrict resident duty hours when they take effect in July 2011.3 This may lead to further decreases in resident‐based inpatient care. Teaching hospitals will need to continue to develop alternate models for inpatient care that do not depend on house staff. Our findings provide important evidence to inform the development of such models. Our study shows that one such model: PAs paired with hospitalists, accepting admissions early in the workday, with hospitalist coverage over the weekend and nights can care for GM inpatients as complex as those cared for by resident‐based teams without increasing readmission rates, inpatient mortality, or charges but at the cost of slightly higher LOS.
In 2003 the Accreditation Council for Graduate Medical Education (ACGME) prescribed residency reform in the form of work hour restrictions without prescribing alternatives to resident based care.1 As a response, many academic medical centers have developed innovative models for providing inpatient care, some of which incorporate Physician Assistants (PAs).2 With further restrictions in resident work hours possible,3 teaching hospitals may increase use of these alternate models to provide inpatient care. Widespread implementation of such new and untested models could impact the care of the approximately 20 million hospitalizations that occur every year in US teaching hospitals.4
Few reports have compared the care delivered by these alternate models with the care provided by traditional resident‐based models of care.58 Roy et al.8 have provided the only recent comparison of a PA‐based model of care with a resident‐based model. They showed lower adjusted costs of inpatient care associated with PA based care but other outcomes were similar to resident‐based teams.
The objective of this study is to provide a valid and usable comparison of the outcomes of a hospitalist‐PA (H‐PA) model of inpatient care with the traditional resident‐based model. This will add to the quantity and quality of the limited research on PA‐based inpatient care, and informs the anticipated increase in the involvement of PAs in this arena.
Methods
Study Design and Setting
We conducted a retrospective cohort study at a 430‐bed urban academic medical center in the Midwestern United States.
Models of General Medical (GM) Inpatient Care at the Study Hospital During the Study Period
In November 2004, as a response to the ACGME‐mandated work hour regulations, we formed 2 Hospitalist‐PA teams (H‐PA) to supplement the 6 preexisting general medicine resident teams (RES).
The H‐PA and RES teams differed in staffing, admitting times and weekend/overnight cross coverage structure (Table 1). There were no predesigned differences between the teams in the ward location of their patients, availability of laboratory/radiology services, specialty consultation, social services/case management resources, nursing resources or documentation requirements for admission, daily care, and discharge.
| H‐PA Teams | RES Teams | |
|---|---|---|
| Attending physician | Always a hospitalist | Hospitalist, non‐hospitalist general internist or rarely a specialist |
| Attending physician role | Supervisory for some patients (about half) and sole care provider for others. | Supervisory for all patients |
| Team composition | One attending paired with 1 PA | Attending + senior resident + (2) interns + (2‐3) medical students |
| Rotation schedule | ||
| Attending | Every 2 weeks | Every 2 weeks |
| Physician assistant | Off on weekends | |
| House staff & medical students | Every month | |
| Weekend | No new admissions & hospitalist manages all patients | Accept new admissions |
| Admission times (weekdays) | 7 AM to 3 PM | Noon to 7 AM |
| Source of admissions | Emergency room, clinics, other hospitals | Emergency room, clinics, other hospitals |
| Number of admissions (weekdays) | 4‐6 patients per day per team | Noon to 5 PM: 2 teams admit a maximum of 9 patients total |
| 5 PM to 7 AM: 3 teams admit a maximum 5 patients each. | ||
| Overnight coverageroles and responsibilities | One in‐house faculty | 3 on call interns |
| Cross‐covering 2 H‐PA teams | Cross‐covering 2 teams each | |
| Performing triage | Admitting up to 5 patients each | |
| Admitting patients if necessary | ||
| Assisting residents if necessary | ||
| General medical consultation |
Admission Schedule for H‐PA or RES Teams
The admitting schedule was designed to decrease the workload of the house staff and to do so specifically during the periods of peak educational activity (morning report, attending‐led teaching rounds, and noon report). A faculty admitting medical officer (AMO) assigned patients strictly based on the time an admission was requested. Importantly, the request for admission preceded the time of actual admission recorded when the patient reached the ward. The time difference between request for admission and actual admission depended on the source of admission and the delay associated with assigning a patient room. The AMO assigned 8 to 12 new patients to the H‐PA teams every weekday between 7 AM and 3 PM and to the RES teams between noon and 7 AM the next day. There was a designed period of overlap from noon to 3 PM during which both H‐PA and RES teams could admit patients. This period allowed for flexibility in assigning patients to either type of team depending on their workload. The AMO did not use patient complexity or teaching value to assign patients.
Exceptions to Admission Schedule
Patients admitted overnight after the on call RES had reached their admission limits were assigned to H‐PA teams the next morning. In addition, recently discharged patients who were readmitted while the discharging hospitalist (H‐PA teams) or the discharging resident (RES teams) was still scheduled for inpatient duties, were assigned back to the discharging team irrespective of the admitting schedule.
The same medicine team cared for a patient from admission to discharge but on transfer to the intensive care unit (ICU), an intensivist led critical care team assumed care. On transfer out of the ICU these patients were assigned back to the original team irrespective of admitting schedulethe so called bounce back rule to promote inpatient continuity of care. But if the residents (RES teams) or the hospitalist (H‐PA teams) had changedthe bounce back rule was no longer in effect and these patients were assigned to a team according to the admission schedule.
Study Population and Study Period
We included all hospitalizations of adult patients to GM teams if both their date of admission and their date of discharge fell within the study period (January 1, 2005 to December 31, 2006). We excluded hospitalizations with admissions during the weekendwhen H‐PA teams did not admit patients; hospitalizations to GM services with transfer to nonGM service (excluding ICU) and hospitalizations involving comanagement with specialty servicesas the contribution of GM teams for these was variable; and hospitalizations of private patients.
Data Collection and Team Assignment
We collected patient data from our hospital's discharge abstract database. This database did not contain team information so to assign teams we matched the discharging attending and the day of discharge to the type of team that the discharging attending was leading that day.
We collected patient age, gender, race, insurance status, zip‐code, primary care provider, source of admission, ward type, time and day of admission, and time and day of discharge for use as independent variables. The time of admission captured in the database was the time of actual admission and not the time the admission was requested.
We grouped the principal diagnosis International Statistical Classification of Diseases and Related Health Problems, 9th edition (ICD‐9) codes into clinically relevant categories using the Clinical Classification Software.9 We created comorbidity measures using Healthcare Cost and Utilization Project Comorbidity Software, version 3.4.10
Outcome Measures
We used length of stay (LOS), charges, readmissions within 7, 14, and 30 days and inpatient mortality as our outcome measures. We calculated LOS by subtracting the discharge day and time from the admission day and time. The LOS included time spent in the ICU. We summed all charges accrued during the entire hospitalization including any stay in the ICU but did not include professional fees. We considered any repeat hospitalization to our hospital within 7, 14, and 30 days following a discharge to be a readmission except that we excluded readmissions for a planned procedure or for inpatient rehabilitation.
Statistical Analysis
Descriptive Analysis
We performed unadjusted descriptive statistics at the level of an individual hospitalization using medians and interquartile ranges for continuous data and frequencies and percentages for categorical data. We used chi‐square tests of association and KruskalWallis analysis of variance to compare H‐PA and RES teams.
Missing Data
Because we lacked data on whether a primary outpatient care provider was available for 284 (2.9%) of our study hospitalizations, we dropped them from our multivariable analyses. We used an arbitrary discharge time of noon for the 11 hospitalizations which did not have a discharge time recorded.
Multivariable Analysis
We used multivariable mixed models to risk adjust for a wide variety of variables. We included age, gender, race, insurance, presence of primary care physician, and total number of comorbidities as fixed effects in all models because of the high face validity of these variables. We then added admission source, ward, time, day of week, discharge day of week, and comorbidity measures one by one as fixed effects, including them only if significant at P < 0.01. For assessing LOS, charges, and readmissions, we added a variable identifying each patient as a random effect to account for multiple admissions for the same patient. We then added variables identifying attending physician, principal diagnostic group, and ZIP code of residence as random effects to account for clustering of hospitalizations within these categories, including them only if significant at P < 0.01. For the model assessing mortality we included variables for attending physician, principal diagnostic group, and ZIP code of residence as random effects if significant at P < 0.01. We log transformed LOS and charges because they were extremely skewed in nature. Readmissions were analyzed after excluding patients who died or were discharged alive within 7, 14, or 30 days of the end of the study period.
Sensitivity Analyses
To assess the influence of LOS outliers, we changed LOS to 6 hours if it was less than 6 hours, and 45 days if it was more than 45 daysa process called winsorizing. We consider winsorizing superior to dropping outliers because it acknowledges that outliers contribute information, but prevent them from being too influential. We chose the 6 hour cut off because we believed that was the minimum time required to admit and then discharge a patient. We chose the upper limit of 45 days on reviewing the frequency distribution for outliers. Similarly, we winsorized charges at the first and 99th percentile after reviewing the frequency distribution for outliers. We then log transformed the winsorized data before analysis.
Inpatient deaths reduce the LOS and charges associated with a hospitalization. Thus excess mortality may provide a false concession in terms of lower LOS or charges. To check if this occurred in our study we repeated the analyses after excluding inpatient deaths.
ICU stays are associated with higher LOS, charges, and mortality. In our model of care, some patients transferred to the ICU are not cared for by the original team on transfer out. Moreover, care in the ICU is not controlled by the team that discharges them. Since this might obscure differences in outcomes achieved by RES vs. H‐PA teams, we repeated these analyses after excluding hospitalizations with an ICU stay.
Since mortality can only occur during 1 hospitalization per patient, we repeated the mortality analysis using only each patient's first admission or last admission and using a randomly selected single admission for each patient.
Subgroup Analysis
To limit the effect of different physician characteristics on H‐PA and RES teams we separately analyzed the hospitalizations under the care of hospitalists who served on both H‐PA and RES teams.
To limit the effect of different admission schedules of H‐PA and RES teams we analyzed the hospitalizations with admission times between 11.00 AM and 4.00 PM. Such hospitalizations were likely to be assigned during the noon to 3 PM period when they could be assigned to either an H‐PA or RES team.
Interactions
Finally we explored interactions between the type of team and the fixed effect variables included in each model.
Statistical Software
We performed the statistical analysis using SAS software version 9.0 for UNIX (SAS Institute, Inc., Cary, NC) and R software (The R Project for Statistical Computing).
This study protocol was approved by the hospital's institutional review board.
Results
Study Population
Of the 52,391 hospitalizations to our hospital during the study period, 13,058 were admitted to general medicine. We excluded 3102 weekend admissions and 209 who met other exclusion criteria. We could not determine the team assignment for 66. Of the remaining 9681 hospitalizations, we assigned 2171 to H‐PA teams and 7510 to RES teams (Figure 1).
Descriptive Analysis
We compare patients assigned to H‐PA and RES teams in Table 2. They were similar in age, gender, race, having a primary care provider or not, and insurance status. Clinically, they had similar comorbidities and a similar distribution of common principal diagnoses. Consistent with their admitting schedule, H‐PA teams admitted and discharged more patients earlier in the day and admitted more patients earlier in the work week. Patients cared for by H‐PA teams were admitted from the Emergency Room (ER) less often and were more likely to reside on wards designated as nonmedicine by nursing specialty. Hospitalizations to H‐PA teams more often included an ICU stay.
| H‐PA (n = 2171) | RES (n = 7510) | P Value | |
|---|---|---|---|
| |||
| Age | |||
| Mean | 56.80 | 57.04 | |
| Median | 56 | 56 | 0.15 |
| Interquartile range | 43‐72 | 43‐73 | |
| Age group (years), n (%) | |||
| < 20 | 10 (0.5) | 57 (0.8) | |
| 20‐29 | 186 (8.6) | 632 (8.7) | |
| 30‐39 | 221 (10.2) | 766 (10.3) | |
| 40‐49 | 387 (17.8) | 1341 (18.1) | |
| 50‐59 | 434 (20.0) | 1492 (20.2) | 0.28 |
| 60‐69 | 325 (15.0) | 974 (12.8) | |
| 70‐79 | 271 (12.5) | 1035 (13.6) | |
| 80‐89 | 262 (12.0) | 951(12.3) | |
| 90< | 75 (3.5) | 262 (3.4) | |
| Female, n (%) | 1175 (54.1) | 4138 (55.1) | 0.42 |
| Race, n (%) | |||
| White | 1282 (59.1) | 4419 (58.9) | |
| Black | 793 (36.5) | 2754 (36.7) | 0.98 |
| Other | 96 (4.4) | 337 (4.5) | |
| Primary care provider, n (%) | 0.16 | ||
| Yes | 1537 (73.2) | 5451 (74.7) | |
| Missing: 284 | 71 (3.3) | 213 (2.8) | |
| Insurance status, n (%) | |||
| Commercial/worker's comp | 440 (20.3) | 1442 (19.2) | |
| Medicare | 1017 (46.8) | 3589 (47.8) | 0.52 |
| Medicaid/others | 714 (32.9) | 2479 (33.0) | |
| Time of admission, n (%) | |||
| 0000‐0259 | 167 (7.7) | 1068 (14.2) | |
| 0300‐0559 | 244 (11.2) | 485 (6.5) | |
| 0600‐0859 | 456 (21.0) | 270 (3.6) | |
| 0900‐1159 | 782 (36.0) | 1146 (15.3) | <0.001 |
| 1200‐1459 | 299 (13.8) | 1750 (23.3) | |
| 1500‐1759 | 155 (7.1) | 1676 (22.3) | |
| 1800‐2359 | 68 (3.1) | 1115 (14.9) | |
| Time of discharge, n (%) | |||
| 2100‐0859 | 36 (1.7) | 174 (2.3) | |
| 0900‐1159 | 275 (12.7) | 495 (6.6) | |
| 1200‐1459 | 858 (39.6) | 2608 (34.8) | <0.001 |
| 1500‐1759 | 749 (34.6) | 3122 (41.6) | |
| 1800‐2059 | 249 (11.5) | 1104 (14.7) | |
| Missing | 4 | 7 | |
| Day of week of admission, n (%) | |||
| Monday | 462 (21.3) | 1549 (20.6) | |
| Tuesday | 499 (23.0) | 1470 (19.6) | |
| Wednesday | 430 (19.8) | 1479 (19.7) | 0.001 |
| Thursday | 400 (18.4) | 1482 (19.7) | |
| Friday | 380 (17.5) | 1530 (20.4) | |
| Day of week of discharge, n (%) | |||
| Monday | 207 (9.5) | 829 (11.0) | |
| Tuesday | 268 (12.3) | 973 (13.0) | |
| Wednesday | 334 (15.4) | 1142 (15.2) | |
| Thursday | 362 (16.7) | 1297 (17.3) | 0.16 |
| Friday | 485 (22.3) | 1523 (20.3) | |
| Saturday | 330 (15.2) | 1165 (15.5) | |
| Sunday | 185 (8.5) | 581 (7.7) | |
| Admit to non‐medicine wards, n (%) | 1332 (61.4) | 2624 (34.9) | <0.001 |
| Transfer to ICU (at least once), n (%) | 299 (13.8) | 504 (6.7) | <0.001 |
| Admit from ER No (%) | 1663 (76.6) | 6063 (80.7) | <0.001 |
| 10 most frequent diagnosis (%) | Pneumonia (4.9) | Pneumonia (5.5) | |
| Congestive heart failure; nonhypertensive (4.2) | Congestive heart failure; nonhypertensive (3.9) | ||
| Sickle cell anemia (3.9) | Nonspecific chest pain (3.7) | ||
| Chronic obstructive pulmonary disease and Bronchiectasis (3.3) | Urinary tract infections(3.6) | ||
| Diabetes mellitus with complications (3.2) | Skin and subcutaneous tissue infections (3.3) | ||
| Urinary tract infections (3.2) | Sickle cell anemia (3.3) | ||
| Asthma (3.0) | Pancreatic disorders (not diabetes) (2.8) | ||
| Nonspecific chest pain (3.0) | Asthma (2.8) | ||
| Pancreatic disorders (not diabetes) (2.9) | Chronic obstructive pulmonary disease and Bronchiectasis (2.6) | ||
| Septicemia (2.2) | Diabetes mellitus with complications (2.6) | ||
| Average number of comorbidities mean (95% CI) | 0.39 (0.37‐0.42) | 0.38 (0.36‐0.39) | 0.23 |
In unadjusted comparisons of outcomes (Table 3), hospitalizations on H‐PA teams had higher lengths of stay and charges than hospitalizations on RES teams, possibly higher inpatient mortality rates but similar unadjusted readmission rates at 7, 14, and 30 days
| H‐PA (n = 2171) | RES (n = 7150) | % Difference* (CI) | P Value | |
|---|---|---|---|---|
| ||||
| LOS | Median (IQR) | Median (IQR) | ||
| Days | 3.17 (2.03‐5.30) | 2.99 (1.80‐5.08) | +8.9% (4.71‐13.29%) | <0.001 |
| Charges | ||||
| US Dollars | 9390 (6196‐16,239) | 9044 (6106‐14,805) | +5.56% (1.96‐9.28%) | 0.002 |
| Readmissions | n (%) | n (%) | Odds Ratio (CI) | |
| Within 7 days | 147 (6.96) | 571 (7.78) | 0.88 (0.73‐1.06) | 0.19 |
| Within14 days | 236 (11.34) | 924 (12.76) | 0.87 (0.75‐1.01) | 0.07 |
| Within 30 days | 383 (18.91) | 1436 (20.31) | 0.91 (0.80‐1.03) | 0.14 |
| Inpatient deaths | 39 (1.8) | 95 (1.3) | 1.36 (0.90‐2.00) | 0.06 |
Multivariable Analysis
LOS
Hospitalizations to H‐PA teams were associated with a 6.73% longer LOS (P = 0.005) (Table 4). This difference persisted when we used the winsorized data (6.45% increase, P = 0.006), excluded inpatient deaths (6.81% increase, P = 0.005), or excluded hospitalizations that involved an ICU stay (6.40%increase, P = 0.011) (Table 5).
| Overall | Subgroup: Restricted to Physicians Attending on Both H‐PA and RES Teams* | Subgroup: Restricted to Hospitalizations Between 11.00 AM and 4.00 PM | ||||
|---|---|---|---|---|---|---|
| % Difference (CI) | P Value | % Difference (CI) | P Value | % Difference (CI) | P Value | |
| ||||||
| LOS | 6.73% (1.99% to 11.70%) | 0.005 | 5.44% (0.65% to 11.91%) | 0.08 | 2.97% (4.47% to 10.98%) | 0.44 |
| Charges | 2.75% (1.30% to 6.97%) | 0.19 | 1.55% (3.76% to 7.16%) | 0.57 | 6.45% (0.62% to 14.03%) | 0.07 |
| Risk of Readmission | Adjusted OR (95%CI) | P Value | Adjusted OR (95% CI) | P Value | Adjusted OR (95% CI) | P Value |
| Within 7 days | 0.88 (0.64‐1.20) | 0.42 | 0.74 (0.40‐1.35) | 0.32 | 0.90 (0.40‐2.00) | 0.78 |
| Within14 days | 0.90 (0.69‐1.19) | 0.46 | 0.71 (0.51‐0.99) | 0.05 | 0.87 (0.36‐2.13) | 0.77 |
| Within 30 days | 0.89 (0.75‐1.06) | 0.20 | 0.75 (0.51‐1.08) | 0.12 | 0.92 (0.55‐1.54) | 0.75 |
| Inpatient mortality | 1.27 (0.82‐1.97) | 0.28 | 1.46 (0.67‐3.17) | 0.33 | 1.14 (0.47‐2.74) | 0.77 |
| Analysis With Winsorized Data | Analysis After Excluding Inpatient Deaths | Analysis After Excluding Patients With ICU Stays | ||||
|---|---|---|---|---|---|---|
| % Difference (CI) | P Value | % Difference (CI) | P Value | % Difference (CI) | P Value | |
| ||||||
| LOS | 6.45% (4.04 to 8.91%) | 0.006 | 6.81% (2.03 to 11.80%) | 0.005 | 6.40% (1.46 to 11.58%) | 0.011 |
| Charges | 2.67 (1.27 to 6.76%) | 0.187 | 2.89% (1.16 to 7.11%) | 0.164 | 0.74% (3.11 to 4.76%) | 0.710 |
Charges
Hospitalizations to H‐PA and RES teams were associated with similar charges (Table 4). The results were similar when we used winsorized data, excluded inpatient deaths or excluded hospitalizations involving an ICU stay (Table 5).
Readmissions
The risk of readmission at 7, 14, and 30 days was similar between hospitalizations to H‐PA and RES teams (Table 4).
Mortality
The risk of inpatient death was similar between all hospitalizations to H‐PA and RES teams or only hospitalizations without an ICU stay (Table 4). The results also remained the same in analyses restricted to first admissions, last admissions, or 1 randomly selected admission per patient.
Sub‐Group Analysis
On restricting the multivariable analyses to the subset of hospitalists who staffed both types of teams (Table 4), the increase in LOS associated with H‐PA care was no longer significant (5.44% higher, P = 0.081). The charges, risk of readmission at 7 and 30 days, and risk of inpatient mortality remained similar. The risk of readmission at 14 days was slightly lower following hospitalizations to H‐PA teams (odds ratio 0.71, 95% confidence interval [CI] 0.51‐0.99).
The increase in LOS associated with H‐PA care was further attenuated in analyses of the subset of admissions between 11.00 AM and 4.00 PM (2.97% higher, P = 0.444). The difference in charges approached significance (6.45% higher, P = 0.07), but risk of readmission at 7, 14, and 30 days and risk of inpatient mortality were no different (Table 4).
Interactions
On adding interaction terms between the team assignment and the fixed effect variables in each model we detected that the effect of H‐PA care on LOS (P < 0.001) and charges (P < 0.001) varied by time of admission (Figure 2a and b). Hospitalizations to H‐PA teams from 6.00 PM to 6.00 AM had greater relative increases in LOS as compared to hospitalizations to RES teams during those times. Similarly, hospitalizations during the period 3.00 PM to 3.00 AM had relatively higher charges associated with H‐PA care compared to RES care.
Discussion
We found that hospitalizations to our H‐PA teams had longer LOS but similar charges, readmission rates, and mortality as compared to traditional resident‐based teams. These findings were robust to multiple sensitivity and subgroup analyses but when we examined times when both types of teams could receive admissions, the difference in LOS was markedly attenuated and nonsignificant.
We note that most prior reports comparing PA‐based models of inpatient care predate the ACGME work hour regulations. In a randomized control trial (1987‐1988) Simmer et al.5 showed lower lengths of stay and charges but possibly higher risk of readmission for PA based teams as compared to resident based teams. Van Rhee et al.7 conducted a nonrandomized retrospective cohort study (1994‐1995) using administrative data which showed lower resource utilization for PA‐based inpatient care. Our results from 2005 to 2006 reflect the important changes in the organization and delivery of inpatient care since these previous investigations.
Roy et al.8 report the only previously published comparison of PA and resident based GM inpatient care after the ACGME mandated work hour regulations. They found PA‐based care was associated with lower costs, whereas we found similar charges for admissions to RES and H‐PA teams. They also found that LOS was similar for PA and resident‐based care, while we found a higher LOS for admissions to our H‐PA team. We note that although the design of Roy's study was similar to our own, patients cared for by PA‐based teams were geographically localized in their model. This may contribute to the differences in results noted between our studies.
Despite no designed differences in patients assigned to either type of team other than time of admission we noted some differences between the H‐PA and RES teams in the descriptive analysis. These differences, such as a higher proportion of hospitalizations to H‐PA teams being admitted from the ER, residing on nonmedicine wards or having an ICU stay are likely a result of our system of assigning admissions to H‐PA teams early during the workday. For example patients on H‐PA teams were more often located on nonmedicine wards as a result of later discharges and bed availability on medicine wards. The difference that deserves special comment is the much higher proportion (13.8% vs. 6.7%) of hospitalizations with an ICU stay on the H‐PA teams. Hospitalizations directly to the ICU were excluded from our study which means that the hospitalizations with an ICU stay in our study were initially admitted to either H‐PA or RES teams and then transferred to the ICU. Transfers out of the ICU usually occur early in the workday when H‐PA teams accepted patients per our admission schedule. These patients may have been preferentially assigned to H‐PA teams, if on returning from the ICU the original team's resident had changed (and the bounce back rule was not in effect). Importantly, the conclusions of our research are not altered on controlling for this difference in the teams by excluding hospitalizations with an ICU stay.
Hospitalizations to H‐PA teams were associated with higher resource utilization if they occurred later in the day or overnight (Figure 2a and b). During these times a transition of care occurred shortly after admission. For a late day admission the H‐PA teams would transfer care for overnight cross cover soon after the admission and for patients admitted overnight as overflow they would assume care of a patient from the nighttime covering physician performing the admission. On the other hand, on RES teams, interns admitting patients overnight continued to care for their patients for part of the following day (30‐hour call). Similar findings of higher resource utilization associated with transfer of care after admission in the daytime11 and nighttime12 have been previously reported. An alternative hypothesis for our findings is that the hospital maybe busier and thus less efficient during times when H‐PA teams had to admit later in the day or accept patients admitted overnight as overflow. Future research to determine the cause of this significant interaction between team assignment and time of admission on resource utilization is important as the large increases in LOS (up to 30%) and charges (up to 50%) noted, could have a potentially large impact if a higher proportion of hospitalizations were affected by this phenomenon.
Our H‐PA teams were assigned equally complex patients as our RES teams, in contrast to previous reports.8, 13 This was accomplished while improving the resident's educational experience and we have previously reported increases in our resident's board pass rates and in‐service training exam scores with that introduction of our H‐PA teams.14 We thus believe that selection of less complex patients to H‐PA teams such as ours is unnecessary and may give them a second tier status in academic settings.
Our report has limitations. It is a retrospective, nonrandomized investigation using a single institution's administrative database and has the limitations of not being able to account for unmeasured confounders, severity of illness, errors in the database, selection bias and has limited generalizability. We measured charges not actual costs,15 but we feel charges are a true reflection of relative resource use when compared between similar patients within a single institution. We also did not account for the readmissions that occur to other hospitals16 and our results do not reflect resource utilization for the healthcare system in total. For example, we could not tell if higher LOS on H‐PA teams resulted in lower readmissions for their patients in all hospitals in the region, which may reveal an overall resource savings. Additionally, we measured in‐hospital mortality and could not capture deaths related to hospital care that may occur shortly after discharge.
ACGME has proposed revised standards that may further restrict resident duty hours when they take effect in July 2011.3 This may lead to further decreases in resident‐based inpatient care. Teaching hospitals will need to continue to develop alternate models for inpatient care that do not depend on house staff. Our findings provide important evidence to inform the development of such models. Our study shows that one such model: PAs paired with hospitalists, accepting admissions early in the workday, with hospitalist coverage over the weekend and nights can care for GM inpatients as complex as those cared for by resident‐based teams without increasing readmission rates, inpatient mortality, or charges but at the cost of slightly higher LOS.
- ACGME‐Common Program Requirements for Resident Duty Hours. Available at: http://www.acgme.org/acWebsite/dutyHours/dh_ComProgrRequirmentsDutyHours0707.pdf. Accessed July 2010.
- ,,,,.Non‐housestaff medicine services in academic centers: models and challenges.J Hosp Med.2008;3(3):247–255.
- ACGME. Duty Hours: Proposed Standards for Review and comment. Available at: http://acgme‐2010standards.org/pdf/Proposed_Standards. pdf. Accessed July 22,2010.
- Agency for Health Care Policy and Research. HCUPnet: A tool for identifying, tracking, and analyzing national hospital statistics. Available at: http://hcup.ahrq.gov/HCUPnet.asp. Accessed July2010.
- ,,,,.A randomized, controlled trial of an attending staff service in general internal medicine.Med Care.1991;29(7 suppl):JS31–JS40.
- ,.Replacing an academic internal medicine residency program with a physician assistant‐‐hospitalist model: a Comparative Analysis Study.Am J Med Qual.2009;24(2):132–139.
- ,,.Resource use by physician assistant services versus teaching services.JAAPA.2002;15(1):33–42.
- ,,, et al.Implementation of a physician assistant/hospitalist service in an academic medical center: impact on efficiency and patient outcomes.J Hosp Med.2008;3(5):361–368.
- AHRQ. Clinical Classifications Software (CCS) for ICD‐9‐CM. Available at: http://www.hcup‐us.ahrq.gov/toolssoftware/ccs/ccs.jsp#overview. Accessed July2010.
- AHRQ. HCUP: Comorbidity Software, Version 3.4.;Available at: http://www.hcup‐us.ahrq.gov/toolssoftware/comorbidity/comorbidity.jsp. Accessed July2010.
- ,,, et al.Effect of short call admission on length of stay and quality of care for acute decompensated heart failure.Circulation.2008;117(20):2637–2644.
- ,,,.Post‐call transfer of resident responsibility: its effect on patient care.J Gen Intern Med.1990;5(6):501–505.
- ,,,,,.The effect of nonteaching services on the distribution of inpatient cases for internal medicine residents.Acad Med.2009:84(2):220–225.
- ,,, et al.Allowing for better resident education and improving patient care: hospitalist‐physician assistant teams fill in the gaps.J Hosp Med.2007;2[S2]:1–39.
- .The distinction between cost and charges.Ann Intern Med.1982;96(1):102–109.
- ,,.Rehospitalizations among patients in the Medicare Fee‐for‐Service Program.N Engl J Med.2009;360(14):1418–1428.
- ACGME‐Common Program Requirements for Resident Duty Hours. Available at: http://www.acgme.org/acWebsite/dutyHours/dh_ComProgrRequirmentsDutyHours0707.pdf. Accessed July 2010.
- ,,,,.Non‐housestaff medicine services in academic centers: models and challenges.J Hosp Med.2008;3(3):247–255.
- ACGME. Duty Hours: Proposed Standards for Review and comment. Available at: http://acgme‐2010standards.org/pdf/Proposed_Standards. pdf. Accessed July 22,2010.
- Agency for Health Care Policy and Research. HCUPnet: A tool for identifying, tracking, and analyzing national hospital statistics. Available at: http://hcup.ahrq.gov/HCUPnet.asp. Accessed July2010.
- ,,,,.A randomized, controlled trial of an attending staff service in general internal medicine.Med Care.1991;29(7 suppl):JS31–JS40.
- ,.Replacing an academic internal medicine residency program with a physician assistant‐‐hospitalist model: a Comparative Analysis Study.Am J Med Qual.2009;24(2):132–139.
- ,,.Resource use by physician assistant services versus teaching services.JAAPA.2002;15(1):33–42.
- ,,, et al.Implementation of a physician assistant/hospitalist service in an academic medical center: impact on efficiency and patient outcomes.J Hosp Med.2008;3(5):361–368.
- AHRQ. Clinical Classifications Software (CCS) for ICD‐9‐CM. Available at: http://www.hcup‐us.ahrq.gov/toolssoftware/ccs/ccs.jsp#overview. Accessed July2010.
- AHRQ. HCUP: Comorbidity Software, Version 3.4.;Available at: http://www.hcup‐us.ahrq.gov/toolssoftware/comorbidity/comorbidity.jsp. Accessed July2010.
- ,,, et al.Effect of short call admission on length of stay and quality of care for acute decompensated heart failure.Circulation.2008;117(20):2637–2644.
- ,,,.Post‐call transfer of resident responsibility: its effect on patient care.J Gen Intern Med.1990;5(6):501–505.
- ,,,,,.The effect of nonteaching services on the distribution of inpatient cases for internal medicine residents.Acad Med.2009:84(2):220–225.
- ,,, et al.Allowing for better resident education and improving patient care: hospitalist‐physician assistant teams fill in the gaps.J Hosp Med.2007;2[S2]:1–39.
- .The distinction between cost and charges.Ann Intern Med.1982;96(1):102–109.
- ,,.Rehospitalizations among patients in the Medicare Fee‐for‐Service Program.N Engl J Med.2009;360(14):1418–1428.
Copyright © 2011 Society of Hospital Medicine
