Discharge in the a.m.: A randomized controlled trial of physician rounding styles to improve hospital throughput and length of stay
Marisha Burden and Angela Keniston are cofirst authors.
Graphical Abstract
Abstract
Background
To relieve hospital capacity strain, hospitals often encourage clinicians to prioritize early morning discharges which may have unintended consequences.
Objective
We aimed to test the effects of hospitalist physicians prioritizing discharging patients first compared to usual rounding style.Design, Setting and Participants: Prospective, multi-center randomized controlled trial. Three large academic hospitals. Participants were Hospital Medicine attending-level physicians and patients the physicians cared for during the study who were at least 18 years of age, admitted to a Medicine service, and assigned by standard practice to a hospitalist team.Intervention: Physicians were randomized to: (1) prioritizing discharging patients first as care allowed or (2) usual practice. Main Outcome and Measures: Main outcome measure was discharge order time. Secondary outcomes were actual discharge time, length of stay (LOS), and order times for procedures, consults, and imaging.
Results
From February 9, 2021, to July 31, 2021, 4437 patients were discharged by 59 physicians randomized to prioritize discharging patients first or round per usual practice. In primary adjusted analyses (intention-to-treat), findings showed no significant difference for discharge order time (13:03 ± 2 h:31 min vs. 13:11 ± 2 h:33 min, p = .11) or discharge time (15:22 ± 2 h:50 min vs. 15:21 ± 2 h:50 min, p = .45), for physicians randomized to prioritize discharging patients first compared to physicians using usual rounding style, respectively, and there was no significant change in LOS or on order times of other physician orders.
Conclusions
Prioritizing discharging patients first did not result in significantly earlier discharges or reduced LOS.
INTRODUCTION
Hospital capacity strain occurs when there is a higher demand for hospital beds than the available supply. Hospital discharges frequently occur in the afternoon or evening hours,1-4 resulting in an imbalance between supply and demand for hospital beds. Delayed discharges can adversely affect patient flow throughout the hospital3, 5-7 which, in turn, can result in delays in care,8-14 increased mortality,15-17 increased length of stay (LOS),17-19 and higher costs.20
Prioritizing discharges by a specific time, such as 10 a.m. or noon, is a common tactic to improve hospital throughput.1, 7, 21 However, this practice has had mixed results, with some studies showing earlier discharge times,2, 22 while others show no impact on LOS23, 24 or even longer LOS.25 There have been no randomized studies to assess the efficacy of this practice.
Our study aimed to investigate the impact of physician rounding styles on discharge orders, discharge time, and LOS using a randomized controlled trial design and to assess whether patients were adversely impacted by a discharge-first workflow.
METHODS
Trial design overview
The study was a randomized controlled trial designed to test the effects of rounding on discharging patients first compared to usual practice. This study was approved by the Colorado Multiple Institution Review Board 19-1607 and registered through clinicaltrials.gov (NCT05370638).
Setting
The study was conducted at three hospitals including a 678-bed quaternary care center with approximately 12,000 medicine discharges per year with 75 hospitalist full-time equivalents (FTE) (52 cFTE); a 550-bed safety net hospital with approximately 10,500 medicine discharges per year with 39 hospitalists FTE (32 cFTE); and a 420-bed academic medical center with approximately 11,000 medicine discharges per year with 40 hospitalist FTE (26 cFTE). One hospital is located on the east coast and two are in the western region of the United States. They are each in separate healthcare systems.
Participants
This study included Hospital Medicine attending physicians who consented to participate in the study. Physicians were approached by members of the research team that were not in direct supervisory roles, primarily by email after it was presented in hospitalist team meetings. Physicians were recruited on a rolling basis and received a small monetary incentive to participate ($20 gift card). While the initial plan was for physicians to participate for 3 months, the full duration of the study period lasted for 6 months to account for rolling recruitment. Recruitment prioritized clinicians with 75% or higher clinical time to ensure they had sufficient numbers of discharges and patient care activities. Patients included patients the physicians cared for during the study who were at least 18 years of age, admitted to a Medicine service, and assigned by standard practice to a hospitalist team. A waiver of patient consent was approved by the institutional review board given patient data were collected as a part of standard clinical care and were extracted retrospectively following patient discharge from the hospital.
Intervention
Physicians were randomized to one of two groups: (1) prioritizing discharging patients first, as care allowed and (2) usual practice. Prioritizing discharging patients first was defined as the attending physician and accompanying team (such as advanced practice providers or learners) prioritizing seeing patients that were going to discharge that day first before seeing other patients. Prioritizing discharges first could be done by the physician or other team members as long as discharging patients were rounded on first. Participants were encouraged to enter the discharge order as soon as possible based on clinical judgment. The attending could break protocol if needed for patient care purposes. Clinical care occurred as it typically would whether a study was in place or not. Usual practice included any rounding style that the attending physician typically used.
During the 6-month study intervention period, reminders about randomization assignments were sent to participating physicians before they started on service. At the end of each shift, a brief survey (described below) was sent. Two additional reminders were sent to physicians if survey data were missing. A weekly deidentified report was provided to the study team at each participating site showing the overall response rate to the daily survey.
Data collection
Surveys
After obtaining informed consent, each physician was asked to complete an enrollment survey using REDCap,26 to collect demographics and rounding style practices (Supporting Information: Figure 1). Daily survey data were collected using a secure online data collection tool that linked to each site's clinical scheduling software and asked questions regarding rounding styles, team composition, and workload (e.g., team census) (Supporting Information: Figure 2). Definitions are provided in Supporting Information: Table 1.
Patient-level data
Clinical variables were collected from each site's data warehouse of pre-existing clinical data as a part of the patient's routine medical care (Supporting Information: Table 1).
Intention-to-treat and per protocol analyses
Primary analyses utilized an intention-to-treat approach to test the research hypothesis that physicians randomized to prioritize discharging patients first would see earlier discharge times compared to those in the usual practice group. The rounding practice to which each participating physician was randomized was used as the explanatory variable. For secondary analyses, we conducted per-protocol analyses and analyses based upon the reported rounding style given we expected nonadherence with randomization (e.g., we expected up to 50% nonadherence based on both practical experience and previous studies27, 28). For the per-protocol analyses, the explanatory variable was the rounding practice style that was used by each participating physician. In order to be included in these analyses, physicians had to adhere to the rounding practice they were assigned to. Thus, the “usual practice” group included all physicians who were randomized to that group, and the “discharges first” group only included those who reported adhering to this practice on the day of patient discharge. The explanatory variable for analyses based on reported style was the rounding practice as reported by participating physicians on the day of patient discharge (Supporting Information: Table 1).
Outcome variables
The primary outcome for assessing the effects of rounding style was the time of day the discharge order was entered. Secondary outcomes were actual discharge time and LOS. To assess whether prioritizing discharging patients first caused other care delays, the average time of day over the course of each patient's hospitalization that orders for imaging, consults, or procedures were entered by participating physicians was analyzed (Supporting Information: Table 1 for definitions).
Sample size
Power calculations were based on a two-sample t test for the difference in the meantime of discharge order between study arms, with adjustment for clustering of discharges within physicians. With 30 physicians per study arm and an assumed intraclass correlation coefficient (ICC) of 0.05, the minimum detectable mean difference in discharge times at 80% power was 30 min for an ICC of 0.05 or 51 min at an ICC of 0.2.
Randomization
The randomization schedule was constructed as random permuted blocks of four physicians each with equal allocation to the two groups (prioritizing discharging patients first vs. usual practice). A single randomization schedule was produced for all three sites.
Statistical methods
Quantitative data analysis
All statistical analyses were performed using SAS Enterprise Guide 8.2 (SAS Institute Inc.). We estimated means and standard deviations (SDs) for continuous variables when approximately normally distributed (as assessed by visual inspection of histograms), medians and interquartile ranges (IQRs) when not, and frequencies for categorical variables.
We used multivariable linear mixed modeling, adjusted for patient-level, clinician-level, and team-level covariates hypothesized a priori to be associated with the outcomes. Models were adjusted for patient sex, patient age, patient insurance, patient limited English proficiency, type of patient, patient Charlson Comorbidity Index (CCI), patient discharge to postacute care, day of the week the patient was discharged, any care delivered to the patient while in an intensive care unit (ICU), years since physician training, physician percent clinical full-time equivalent (cFTE), physician gender, learners on the team, type of team, starting morning census, new patients, hospital units, first day on service, and site (see Supporting Information: Table 1 for definitions). The discharge order time and discharge time models were also adjusted for hospital LOS. The models for the secondary “reported group” analyses were also adjusted for the randomization group.
We converted time to hours since midnight on a 24-h clock for modeling. The discharging physician was included as a random effect in all models to account for the correlation between patients cared for by the same physician. Given that hospital LOS is right-skewed, this variable was log-transformed to facilitate regression analysis in both modeling as a primary outcome and as a confounder. We reported a relative difference in hospital LOS by exponentiating the coefficient, subtracting one, and expressing the result as a percentage. The patient CCI was categorized into none, mild, moderate, or severe based on the CCI score calculated from discharge diagnoses for each patient, (0, 1–2, 3–4, and >5, respectively).
To determine whether potential effect modification was supported by the data, we included several interaction terms including team starting morning census, team composition, and disposition to postacute care given we hypothesized that these might impact the findings based upon the literature to date.27, 29-31
Finally, we evaluated whether prioritizing discharging patients first was associated with the average time of day other patient care orders were entered using a noninferiority test as it was hypothesized that prioritizing discharging patients first might delay the care of other patients using mixed effects regression modeling. The one-sided test of noninferiority at the 5% level of significance involved constructing a 90% confidence interval for the mean difference between groups, as estimated by the regression coefficient for the intervention covariate. Noninferiority was established if the upper limit of the confidence interval did not exceed 30 min, our established noninferiority margin.32
Missing data
Prior to beginning analyses, we examined the data carefully to determine if any data queried from site-specific data warehouses were missing. After resolving identified issues, patients with missing data necessary for a specific analysis were excluded from that analysis and reported in all tables.
Qualitative analysis
Using open-ended questions from the daily surveys completed by participating physicians at the end of each shift (Supporting Information: Figure 2), we explored the facilitators and barriers to implementing a prioritizing discharging patients' first rounding style. We employed summative content analysis throughout our data collection and analyses.33, 34 A priori selected terms were identified by hand searching for occurrences of the identified terms across all comments. Frequency counts for each identified term were calculated and latent content analysis was conducted to ascertain the hypothesized underlying meanings of the words identified in the comments. From these data, we identified the facilitators and barriers to prioritizing discharging patients first.
Role of the funding source
This work was funded by the Agency for Healthcare Research and Quality (1R03HS027231-01A1). The funding source did not have any role in study design, conduct, analysis, or reporting.
RESULTS
Sixty-one physicians across three academic hospitals were randomized to one of two groups: 30 physicians were assigned to prioritize discharging patients first (intervention group) and 31 physicians were assigned to round per usual practice (control group) (Figure 1). A total of 59 physicians were included in the analysis. Baseline clinician and team demographics are shown in Table 1 and patient demographics are provided in Table 2. Supporting Information: Tables 2–4 provide adherence with randomization and clinician, team, and patient-level demographics for secondary analyses.
| Intention-to-treat randomization group | ||
|---|---|---|
| Clinician-level demographics | Round on discharging patients first (N = 30) | Round using usual style (N = 29) |
| Site, N (%) | ||
| Site 1 | 10 (33.3) | 10 (34.5) |
| Site 2 | 10 (33.3) | 9 (31.0) |
| Site 3 | 10 (33.3) | 10 (34.5) |
| Clinician years since training, mean ± SD | 7.2 ± 4.2 | 5.1 ± 3.9 |
| Missing | 0 (0) | 0 (0) |
| Clinician percent clinical full-time equivalent, mean ± SD | 0.89 ± 0.17 | 0.82 ± 0.21 |
| Missing | 0 (0) | 1 (3.3) |
| Clinician gender identity, N (%) | ||
| Woman | 15 (50) | 9 (31.0) |
| Man | 15 (50) | 20 (69.0) |
| Missing | 0 (0) | 0 (0) |
| Clinician most commonly used rounding style prior to study, N (%) | ||
| Sickest first | 14 (46.7) | 17 (58.6) |
| Decompensating patients first | 8 (26.7) | 7 (24.1) |
| Geography | 2 (6.7) | 2 (6.9) |
| New patients first | 4 (13.3) | 1 (3.5) |
| Discharges first | 1 (3.3) | 1 (3.5) |
| Other | 1 (3.3) | 1 (3.5) |
| Missing | 0 (0) | 0 (0) |
| Team-level demographicsa | Round on discharging patients first (N = 864) |
Round using usual style (N = 982) |
|---|---|---|
| Site, N (%) | ||
| Site 1 | 315 (36.5) | 335 (34.1) |
| Site 2 | 226 (26.2) | 275 (28.0) |
| Site 3 | 323 (37.4) | 372 (37.9) |
| Learners on team, N (%) | ||
| Yes | 436 (50.5) | 482 (49.1) |
| No | 371 (42.9) | 426 (43.4) |
| Missing | 57 (6.6) | 74 (7.5) |
| Type of team, N (%) | ||
| Physician only | 241 (27.9) | 298 (30.3) |
| Physician and APP, independent | 41 (4.7) | 56 (5.7) |
| Physician and APP, shared | 207 (24.0) | 218 (22.2) |
| Physician and resident | 318 (36.8) | 336 (34.2) |
| Missing | 57 (6.6) | 74 (7.5) |
| Starting morning census, mean ± SD | 11.5 ± 2.4 | 11.5 ± 2.6 |
| Physician only team | 10.2 ± 1.6 | 10.5 ± 2.1 |
| Physician and APP, independent | 16.3 ± 1.8 | 15.1 ± 2.6 |
| Physician and APP, shared | 12.7 ± 2.7 | 13.4 ± 2.6 |
| Physician and resident | 11.0 ± 1.6 | 10.6 ± 1.4 |
| Missing | 57 (6.6) | 74 (7.5) |
| New patients, N (%) | ||
| Yes | 735 (85.1) | 807 (82.2) |
| No | 72 (8.3) | 101 (10.3) |
| Missing | 57 (6.6) | 74 (7.5) |
| Number of hospital units, mean ± SD | 4.1 ± 2.0 | 4.6 ± 2.2 |
| Missing | 57 (6.6) | 74 (7.5) |
| First day on service, N (%) | ||
| Yes | 169 (19.6) | 169 (17.2) |
| No | 638 (73.8) | 739 (75.3) |
| Missing | 57 (6.6) | 74 (7.5) |
| Which type of patients did your team round on first? | ||
| Discharging patients | 419 (48.5) | 289 (29.4) |
| Sickest patients | 169 (19.6) | 205 (20.9) |
| Patients based on geography | 87 (10.1) | 207 (21.1) |
| New patients | 101 (11.7) | 162 (16.5) |
| Decompensating patients | 15 (1.7) | 15 (1.5) |
| Other | 16 (1.9) | 30 (3.1) |
| Missing | 57 (6.6) | 74 (7.5) |
| What was the primary reason your team rounded this way? | ||
| Usual rounding style | 116 (13.4) | 324 (33.0) |
| Randomized to round on discharging patients first | 345 (39.9) | 0 (0) |
| Sick patient(s) | 118 (13.7) | 129 (13.1) |
| Learner-driven rounding | 85 (9.8) | 109 (11.1) |
| Patient flow or capacity | 16 (1.9) | 98 (10.0) |
| Convenience (geography or other factors) | 31 (3.6) | 88 (9.0) |
| No patients ready for discharge | 49 (5.7) | 45 (4.6) |
| New admissions or transfers | 13 (1.5) | 56 (5.7) |
| First day on team | 15 (1.7) | 18 (1.8) |
| Divide and conquer | 1 (0.1) | 1 (0.1) |
| Other | 18 (2.1) | 40 (4.1) |
| Missing | 57 (6.6) | 74 (7.5) |
| Which type of patients did your team round on second?b | ||
| Discharging patients | 168 (19.4) | 177 (18.0) |
| Sickest patients | 91 (10.5) | 118 (12.0) |
| Patients based on geography | 398 (46.1) | 478 (48.7) |
| New patients | 125 (14.5) | 109 (11.1) |
| Decompensating patients | 3 (0.3) | 11 (1.1) |
| Other | 22 (2.5) | 15 (1.5) |
| Missing | 57 (6.6) | 74 (7.5) |
- Abbreviation: SD, standard deviation.
- a Team-level data are determined based on the answers provided by a participating physician to the daily evaluation questions on a day at least one patient was discharged by the physician.
- b Top three reasons for the rounding style used second in the discharges first group were convenience (N = 239, 27.7%), usual rounding style (N = 191, 22.1%), and randomized to round on discharging patients first (N = 99, 11.5%) and the usual style group were convenience (N = 316, 32.2%), usual rounding style (N = 235, 23.9%), and patient flow/capacity (N = 107, 10.9%).
| Intention-to-treat randomization group | ||
|---|---|---|
| Round on discharging patients first (N = 2118) | Round using usual style (N = 2319) | |
| Site, N (%) | ||
| Site 1 | 908 (42.9) | 952 (41.1) |
| Site 2 | 476 (22.5) | 575 (24.8) |
| Site 3 | 734 (34.7) | 792 (34.2) |
| Patient age, mean ± SD | 57.5 ± 17.1 | 58.0 ± 17.2 |
| Missing | 0 (0) | 0 (0) |
| Patient gender, N (%) | ||
| Female | 977 (46.1) | 1048 (45.2) |
| Male | 1141 (53.9) | 1271 (54.8) |
| Missing | 0 (0) | 0 (0) |
| Patient insurance | ||
| Medicare | 1009 (47.6) | 1105 (47.6) |
| Medicaid | 769 (36.3) | 845 (36.4) |
| Commercial | 159 (7.5) | 195 (8.4) |
| Indigent care | 55 (2.6) | 60 (2.6) |
| Self-pay | 70 (3.3) | 62 (2.7) |
| Other | 46 (2.2) | 36 (1.6) |
| Missing | 10 (0.5) | 16 (0.7) |
| Patient limited English proficiency, N (%) | ||
| Yes | 281 (13.3) | 287 (12.4) |
| No | 1835 (86.6) | 2029 (87.5) |
| Missing | 2 (0.09) | 3 (0.1) |
| Patient type, N (%) | ||
| Inpatient | 1680 (79.3) | 1907 (82.2) |
| Observation patient | 438 (20.7) | 412 (17.8) |
| Missing | 0 (0) | 0 (0) |
| Length of stay (days), mean ± SD | ||
| Inpatient | 6.0 ± 7.4 | 6.2 ± 8.1 |
| Observation | 1.9 ± 1.4 | 1.7 ± 1.0 |
| Missing | 0 (0) | 0 (0) |
| Length of stay (days), median (IQR) | ||
| Inpatient | 4 (2, 7) | 4 (2, 7) |
| Observation | 1 (1, 2) | 1 (1, 2) |
| Missing | 0 (0) | 0 (0) |
| Patient CCI severity levels, N (%) | ||
| None | 309 (14.6) | 330 (14.2) |
| Mild | 550 (26.0) | 630 (27.2) |
| Moderate | 550 (26.0) | 583 (25.1) |
| Severe | 707 (33.4) | 769 (33.2) |
| Missing | 2 (0.09) | 7 (0.3) |
| Patient discharged to postacute care, N (%) | ||
| Yes | 254 (12.0) | 337 (14.5) |
| No | 1864 (88.0) | 1982 (85.5) |
| Missing | 0 (0) | 0 (0) |
| Patient discharged on a weekend, N (%) | ||
| Yes | 477 (22.5) | 518 (22.3) |
| No | 1641 (77.5) | 1801 (77.7) |
| Missing | 0 (0) | 0 (0) |
| Patient received care in an ICU, N (%) | ||
| Yes | 41 (1.9) | 41 (1.8) |
| No | 2075 (98.0) | 2269 (97.8) |
| Missing | 2 (0.09) | 9 (0.4) |
- Abbreviations: CCI, Charlson Comorbidity Index; ICU, intensive care unit; IQR, interquartile range; SD, standard deviation.
| Intention-to-treat randomization group | ||
|---|---|---|
| Round on discharging patients first (N = 2118) | Round using usual style (N = 2319) | |
| Discharge order time, mean (SD) | 13:03 ± 2 h:31 mina | 13:11 ± 2 h:33 mina |
| Mean difference (min) (95% confidence Interval) | ||
| Unadjusted results | −8.0 (−32.6, 16.7), p = .5271 | |
| Adjusted resultsb | −13.6 (−30.3. 3.0), p = .1093 | |
| Discharge time, mean (SD) | 15:22 ± 2 h:50 min | 15:21 ± 2 h:50 min |
| Mean difference (min) (95% confidence interval) | ||
| Unadjusted results | 0.40 (−24.4, 25.2), p = .9729 | |
| Adjusted resultsb | −5.3 (−19.2. 8.6), p = .4534 | |
| Length of stay (h), median (IQR) | 75 (45, 141) | 78 (46, 144) |
| Mean percentage decrease (95% confidence interval) | ||
| Unadjusted results | −2.4% (−10.0, 5.9%), p = .5619 | |
| Adjusted resultsb | 2.0% (−2.8, 7.1%), p = .4186 | |
| Procedure order time, mean (SD)c | (N = 2201)10:49 ± 3 h:57 min | (N = 2354) 11:11 ± 3 h:53 min |
| Mean difference (min) (95% confidence interval) | ||
| Unadjusted results | −18.5 (−33.8, −3.2) p = .0466 | |
| Adjusted resultsb | −20.5 (−42.2, 1.2) p = .0644 | |
- Abbreviations: CCI, Charlson Comorbidity Index; FTE, full-time equivalent; ICU, intensive care unit; IQR, interquartile range; SD, standard deviation.
- a Observations missing discharge order time, 29 (1.4%) and 44 (1.9%), respectively.
- b Models adjusted for patient sex, patient age, patient insurance, patient limited English proficiency, type of patient, patient CCI, patient discharge to postacute care, day of the week the patient was discharged, any care delivered to the patient while in an ICU, years since physician training, physician percent clinical FTE, physician gender identity, learners on team, type of team, starting morning census, new patients, hospital units, first day on service, and site (Supporting Information: Table 1 for definitions).
- c All patients for whom a note was written by a participating physician on a day they were scheduled to work (N is thus higher than 4437 because of this as patients that were not discharged in the study are still included in this analysis).
| Concept | Definition | Frequency mentioned, N (%) |
Exemplar quotes |
|---|---|---|---|
| Advantage(s) of discharges first | Positive outcome or advantage noted when rounding on patients anticipated to discharge first | 1 (0.2%) | We had only one guaranteed discharge today and saw her first; this was fortunate because even though we thought she was all ready for discharge and just waiting on infusion company to come at 10 a.m., we actually identified an issue that we had previously missed she needed a double lumen, not single lumen PICC and were able to address it early and avoid a delay. (Evaluation ID: 15506237) |
| Anticipated discharges | Patients who were expected to discharge at the start of rounding or patients identified as early discharges | 112 (17.3%) | We were able to prioritize discharges at same time as new patients today based on geography and acuity of conditions. (Evaluation ID: 15715302) Ended up with 6 discharges today…4 were anticipated and we saw these first, though 2 out of 4 had barriers to morning discharge. (Evaluation ID: 15521310) |
| Care coordination factors | Coordination of care including social work, care management, home O2, medications, or other aspects of patient care in preparation for discharge | 46 (7.1%) | Discharge delays included: (1) Home O2 evaluation and arrangement of transport for COVID + patient, (2) complex coordination requiring OT evaluation, SW coordination, (3) addiction medicine evaluation of patient with recommendations and coordinated follow-up, (4) diabetes education of patient, (5) coordination of home O2 in patient with no funding available/self-pay. (Evaluation ID: 15926166) |
| Consult factors | Making decisions about the order in which to round on patients based on patient needing additional evaluation or consultation or waiting for additional evaluation or consultation before the patient can be discharged | 48 (7.4%) | We had four possible discharges today but all had barriers that would prevent early discharge and already had orders ready (i.e., waiting on something other than our team, like a consulting team, procedure, or insurance authorization), so we rounded geographically; two of those patients ended up discharging today and were not delayed by our rounds. (Evaluation ID: 15506232) |
| Disadvantage(s) of discharges first | Negative outcome or disadvantage noted when rounding on patients anticipated to discharge first | 27 (4.2%) | We had one predicted (discharge) and one who was not previously predicted and saw and discharged both around noon due to a very busy morning with sick patients. I am very glad we did not round on them first because even short delays in care for the sicker patients could have been seriously harmful. (Evaluation ID: 15521136) …seeing discharges first does not translate into early discharges (waiting for procedures, PT/OT, oxygen, tests, seeing how lunch goes, etc.). Saw discharges on the home unit first, then the rest of the patients on the unit, then rounded geographically and eventually got around to discharges on other units. (Evaluation ID: 15521438) Irritated that I rounded on the discharging patients first. Two known discharges had known issues to address during the day prior to discharge (with new consults, waiting for tests, etc.) and I knew they would not be able to leave until later in the afternoon. By rounding on them first before I had all the necessary information I was misguided when putting in discharge orders and counseling the patients. I had to see both a second time. (Evaluation ID: 16243585) |
| Discharge to postacute care | Discharge to a setting other than home for additional treatment or care including outside hospital transfers, long-term acute care, skilled nursing facility, or subacute rehab | 24 (3.7%) | Discharge patients were for facility, which usually happens late. (Evaluation ID: 15654959) |
| Family-related factors | Making decisions about the order in which to round on patients or determining whether a patient can be discharged based on interactions with the patient's family or caregiver, including goals of care conversation, conversations about patient placement, or delays in leaving the hospital due to transportation | 21 (3.2%) | Only planned/anticipated discharge regarding assessment of PO tolerance. Discharge work completed and then patient/family in disagreement about postdischarge plan. Family had to sort out plan among themselves prior to discharge (patient needed family support for safe discharge). (Evaluation ID: 16586863) |
| First day | Making decisions about the order in which to round on patients based on this being the first day a member of the primary team (physician, advanced practice provider, or learner) was on service with this team | 13 (2.0%) | It is difficult to round on discharging patients first when it is (the) first day on service. (Evaluation ID: 16516168) |
| Geographic | Making decisions about the order in which to round on patients based on geographic location of patients | 71 (11.0%) | Geography matters. Got to see all discharges first without having to skip around very much. (Evaluation ID: 15521110) Eight patients on eight different units, discharges on nonadjoining units, tons of extra travel time plus additional travel time to multidisciplinary rounds equals slower rounding, later orders, delayed care. (Evaluation ID: 15521200) |
| New patient(s) | Making decisions about the order in which to round on patients based on whether any patients were newly admitted to the team | 44 (6.8%) | We had a high census with a lot of new patients and could not waste time moving back and forth between units…(Evaluation ID: 15521268) |
| No discharges | No patients were identified as ready to discharge | 48 (7.4%) | None of the patients this morning were clearly ready for discharge, they were pending placement so there was no rush to see (them). (Evaluation ID: 15530879) |
| Nursing factors | Making decisions about the order in which to round on patients or determining whether a patient can be discharged based on nursing availability or outstanding nursing-related tasks | 4 (0.6%) | Able to discharge one patient in am, second patient in afternoon waiting on diabetes education. (Evaluation ID: 15690134) |
| Patient acuity | Making decisions about the order in which to round on patients based on whether any patients assigned to the team were in need of immediate medical attention | 131 (20.2%) | No patients were sick and new patients were stable, so decided to round on discharging patients first (Evaluation ID: 15521121) I had some patients who would be quick discharges since they were admitted for observation overnight but there were other sicker patients with symptoms. (Evaluation ID: 15521204) |
| Patient condition factor(s) | Making decisions about the order in which to round on patients or determining whether a patient can be discharged based on a patient's medical readiness for discharge | 49 (7.6%) | Several discharges, but all complex, awaiting recommendations, HD, and procedures. (Evaluation ID: 15530955) |
| Patient-related factor(s) | Making decisions about the order in which to round on patients or determining whether a patient can be discharged based on interactions with the patient, including situations where patient does not agree with care plan or placement or having to manage request of patient | 37 (5.7%) | Significant morning time spent on one AMA discharge who needed surgery and was high risk for medical worsening without further Rx prior to discharge; one planned discharge was delayed secondary to coordination of postdischarge home monitoring, which patient later decided no longer wanted to utilize. (Evaluation ID: 15531070) |
| Procedure factor(s) | Making decisions about the order in which to round on patients or determining whether a patient can be discharged based on the status of a procedure | 40 (6.2%) | Discharging patients all needed to wait for something before they could go: labs, PT/OT evaluation, procedure. (Evaluation ID: 15521351) |
| Team composition factors | Making decisions about the order in which to round on patients based on the composition of the team or any reference to the composition of the team | 112 (17.3%) | Giving the resident autonomy by waiting to see their patients until after we had card flipped on them. (Evaluation ID: 15521524) Being with a learner with 16 (or) 17 patients really slows things down. We round on our patients and then immediately meet up with the advanced practice provider fellow so do not have time to do our discharges or the learner's discharges. (Evaluation ID: 16487673) |
| Test result factors | Making decisions about the order in which to round on patients or determining whether a patient can be discharged based on the status of a medical test | 32 (4.9%) | Saw all discharges early but most leaving late in the day due to pending tests, SW visits, evaluation to see if they are continuing to improve, etc. (Evaluation ID: 15521205) |
| Unanticipated discharge | Patients who were not expected to discharge at the start of rounding but ended up leaving, including patients who leave against medical advice | 40 (6.2%) | We had zero predicted discharges but ended up discharging 5 later in the day, all in the afternoon, either because placement became available or patient's course changed (in one case choosing home hospice). (Evaluation ID: 16586855) |
| Workload factor(s) | Making decisions about the order in which to round on patients based on staffing, patient volume, the day surveyed being a busy day, or admission or discharge volume | 20 (3.1%) | Very chaotic morning with new patients and many possible discharges with a lot of contingencies. (Evaluation ID: 15521351) No discharges completed on rounds as resident only day and insufficient staffing available to complete (discharge) work while continuing to round and see/staff four new patients. (Evaluation ID: 16243603) It feels really inefficient to round on discharging patients in far satellite units during high-capacity times when our service is full… (Evaluation ID: 15521440) |
- Abbreviations: AMA, against medical advice; HD, hemodialysis; O2, oxygen; OT, occupational therapy; PICC, peripherally inserted central catheter; PO, per os or “by mouth”; PT, physical therapy; Rx, prescription; SW, social work.
From February 9, 2021, to July 31, 2021, participating physicians were scheduled for 2318 shifts, for which 2156 (93.0%) had an associated completed physician daily evaluation. During this timeframe, 4437 patients were discharged by participating physicians, for which 4189 (94.4%) had an associated completed physician daily evaluation. There was an almost 20% higher percentage of patients discharged by physicians randomized to prioritize discharging patients first compared to those prioritizing discharges first as their usual practice style (control group) (Table 1, Supporting Information: Table 2).
In primary adjusted analyses comparing physicians by randomized assignment (intention-to-treat), findings showed no significant difference for discharge order time (13:03 ± 2 h:31 min vs. 13:11 ± 2 h:33 min, p = .11) or actual discharge time (15:22 ± 2 h:50 min vs. 15:21 ± 2 h:50 min, p = .45) for physicians randomized to prioritize discharging patients first compared to physicians randomized to usual rounding style, respectively. Similarly, there was no significant change in LOS (75 h [IQR: 45, 141] vs. 78 h [46, 144], p = .42), respectively (Table 3). Unadjusted and adjusted analyses showed similar results.
In secondary per protocol adjusted analyses comparing physicians that adhered to their randomization assignment compared to physicians who rounded using the usual style, mean discharge order time was 12:34 ± 2 h:31 min versus 13:11 ± 2 h:33 min, respectively, p = .001 with similar trends for actual discharge time; however, there was not a significant difference in LOS with LOS of 78 h (IQR 46, 145) versus 78 h (IQR 46, 144), respectively, p = .15. In secondary reported group adjusted analyses comparing physicians that reported prioritizing discharging patients first to those that used some other style, mean discharge order time was 12:37 ± 2 h:33 min versus 13:28 ± 2 h:26 min, respectively, p < .001 with similar trends for actual discharge time. There was no significant difference in LOS between the two groups (79 h [IQR 46, 145] vs. 75 h [IQR 44, 140], respectively, p = 0.30) (Supporting Information: Table 5). Model statistics for all models are reported in Supporting Information: Tables 6–18.
In addition, in effect modification analyses, we did not find that the time of day the discharge order was entered into the electronic health record (EHR), time of day the patient left the hospital, or hospital LOS varied according to whether the patient was discharged to postacute care, the number of patients assigned to a team at 7 a.m. each morning, or by team composition for patients discharged by a physician who was randomized to prioritize discharging patients first compared to physicians in the control group (Supporting Information: Tables 19–21).
In adjusted models for an average time of day orders for procedures, consults, and imaging entered into the EHR, we found that physicians randomized to prioritizing discharging patients first did not see later order times compared to usual care 10:49 ± 3 h:57 m versus 11:11 ± 3 h:53 m, respectively (α = .05, noninferiority margin = 30 min) (Table 3).
Across 2156 surveys completed by physicians during the study period, 501 surveys contained unstructured comment data, for a total of 648 comments submitted. Content analysis of unstructured survey data indicated that there are several factors that influence physician rounding styles including patient acuity, team composition, readiness for discharge, geography, and balancing new patients. Participants reported that rounding on discharging patients first could impede workflow causing additional work and travel and that prioritizing discharges over sick patients could cause patient harm (Table 4).
DISCUSSION
The key findings of this study are that in primary analyses, physicians prioritizing discharging patients first (1) did not result in earlier discharges or significantly reduced LOS, (2) did not cause delays in placing other patient care orders, and (3) prioritizing rounding on discharging patients first without incorporating other factors such as their location and illness severity led to perceptions of inefficiency.
Our study shows that in primary analyses prioritizing discharging patients first did not result in earlier discharges or significantly reduced lengths of stay. It is common practice for organizations to try to improve throughput by asking clinicians to round on discharging patients first or to discharge a set number of patients by a specific time (e.g., “discharge before 10 a.m.”). Some studies have suggested that clinicians may hold onto patients longer when incentivized to prioritize earlier discharges35 and perhaps that could also explain why we did not see a reduction in LOS though incentives in this study were very small. We also noted a several-hour delay in both study groups from discharge order time to actual discharge time suggesting a potential area of interest to focus on. Of note, despite prioritizing discharging patients first, there were no significant delays in other care orders indicating that clinicians likely adapt their workflows to the needs of their patients.
In secondary analyses (per protocol and per reported group), findings indicate that discharges occurred earlier when physicians prioritized discharging patients first, however, LOS was not impacted and, in fact, with a trend toward longer LOS in those that prioritized discharges first in the reported group. Hospital LOS has been suggested as a system measure for hospital-wide flow36 and when applying business concepts of throughput to a hospital setting, throughput is defined as when a product is processed to a finished good (i.e., when the patient leaves the hospital).37 These concepts paired with the study's results indicate that prioritizing discharges first does not necessarily improve throughput given LOS was not impacted in either primary or secondary analyses. Our findings suggest that it may be more effective to support clinicians to use their clinical judgment to determine priorities, rather than focusing solely on discharging patients or trying to hit a specific time target.
While prioritizing early discharges has been a widely utilized tactic to improve throughput,1, 7, 21 previous work has shown that few clinicians actually round in this manner.27 A small proportion of physicians in our study (3.3%) reported rounding in this fashion prior to participation in this research; however, we saw much higher numbers in the control group for the study suggesting that the study may have impacted the practices of both groups or that clinicians may prioritize discharges and workflows without recognizing their practice is as such. Previous work has shown that implementing this rounding style may be facilitated or inhibited by various operational factors (e.g., geography of patients), patient factors (e.g., acuity), and clinician factors (e.g., multiple competing priorities),27, 28 and this was also seen in our qualitative work. Additionally, prioritizing discharging patients first was perceived to result in inefficiencies such as rework and travel time to units across the hospital.
This work highlights some areas for future study. In secondary analyses for effect modification, we did not find significant interaction effects for randomization groups, per protocol, or reported rounding style and whether a patient was discharged to postacute care, starting morning census, or team composition. This work does not answer the question of what operational factors may lead to earlier discharges or improved efficiency. Previous work has suggested that workload negatively impacts throughput21, 29 and leads to longer LOS and increased costs.29, 38 Additionally, this work did not assess the impact of early discharge on emergency department (ED) boarding or on peak occupancy levels. Given LOS was not reduced, we doubt that the intervention would have impacted ED boarding or peak occupancy levels. A study by Kirubarahan et al. did not find an association between early discharges and ED LOS,24 but others have.5, 39
This study has several strengths. This study represents the first randomized controlled trial on the practice of prioritizing discharging patients first. We had high survey response rates and low dropout rates. Our study also has some limitations. The intervention was based on changing behavior, which can be challenging to sustain and adopt. Participants were offered nominal incentives and the incentives were not based on adherence to randomization assignment. We did achieve an almost 20% change in physician behavior, which suggests that the intervention did increase adoption, however, did not achieve its purpose (e.g., significant reductions in LOS). Prioritizing discharging patients first may have not resulted in an earlier discharge order. Participants were encouraged to enter the discharge order as soon as possible based on clinical judgment. Our results suggest that the intervention did lead to earlier orders in secondary analyses, but no subsequent improvement in LOS. Next, the concept of prioritizing discharging patients first is not simple, as multiple team members may be doing different things at different times, and there is a risk of misclassification. We saw a much higher utilization of prioritizing discharging patients first than expected. This could be due to the fact that most clinicians have a rounding style that adapts to the context of a given day which was seen in both the quantitative and qualitative results. The mean morning census seen in this study may be a limitation to generalizability at some institutions. Finally, our study did have contamination of study arms, which is a known risk of this study design, given some clinicians do prioritize discharging patients first as their usual practice style. Our secondary analyses help clarify the effect of the intervention when contamination of study arms occurs and these secondary analyses suggest that prioritizing discharges first did not improve throughput.
CONCLUSION
In a real-world setting, prioritizing discharging patients first did not result in earlier discharge order times or reduce hospital LOS suggesting that this intervention did not impact throughput.
ACKNOWLEDGMENTS
The authors report funding from the Agency for Healthcare Research and Quality (1R03HS027231-01A1) for this work. Dr. Burden and Ms. Keniston report funding from the Total Worker Health Pilot Grant.
CONFLICT OF INTEREST STATEMENT
The authors declare no conflict of interest.
