The Internal Medicine Residency Program at Saint Louis University has 83 residents, 76 of whom are categorical residents and the remainder preliminary interns. Traditionally, the program had five general floor teams, each staffed with one attending, one supervising senior resident and two interns for a 28-day rotation. Day call (where the team performed admissions and ran the code team) occurred every five days. Each resident had four days off during the block: the senior was off most pre-call days, and the two interns each had one day off between the call days. Thus, in a 28-day rotation, the entire team was together only for 16 of the days, and the senior had to serve in a ‘resitern’ capacity for eight days, in which they wrote notes for intern #1 on one day, and again for intern #2 the next day while supervising the remaining intern. There were also four days where the two interns were alone under the direct supervision of the attending.
Due to the impetus to reduce the number of inpatient teams, we implemented an alternate daily resident schedule on our general inpatient service in April of 2020. The number of inpatient attending-supervised ward teams was reduced to four, yet our program preserved the five one-resident, two-intern teams who then rotated through those four ward teams in an alternate schedule dubbed the “MarioKart”.
Each triad of resident and two interns was given a name based on a ‘MarioKart’ character to identify the triad as opposed to team name of the ward service. This also provided clarity in long-term assignment of residents throughout the year. During a given day, each resident/intern grouping is assigned to an inpatient team, with one entire resident/intern grouping having the day off. Figure 1 Part A shows how each group cycles through the care teams. For example, the four teams would be staffed by ‘Peach’’ (pink in Fig. 1) on inpatient team one, ‘Yoshi’ (yellow in Fig. 1) on inpatient team two, ‘Luigi’ (green in Fig. 1) on inpatient team three, and ‘Bowser’ (orange in Fig. 1)’ on inpatient team four, with team ‘Mario’ (red in Fig. 1)’ having two days off. After two days, team Yoshi will have two consecutive days off, having handed off team two to Mario, who will begin a run of eight days on inpatient service. When team Yoshi finishes their two days off, they take over team three for Luigi, who then receive two days off. Thus, on any given day, the MarioKart system has four resident teams working and one off for the duration of the 28-day rotation block. As a contrast, our ‘traditional’ way of scheduling is outlined in Fig. 1 Part B.
As the new schedule was adopted our distribution of patients to resident teams changed. In creating a system where the team is fully complemented every day, patients were assigned based on team census numbers. For both models, the team ‘cap’ was 16 patients, thus the maximum patient census was 80 for the traditional model and 64 for the Mariokart model.
To compare differences between the ‘traditional’ model and the MarioKart model we analyzed daily patient census numbers for the resident inpatient service and by assigned inpatient teams from July 1, 2018 to June 30, 2021. Data from the traditional model period (July 1, 2018 to March 31, 2020) were compared to data in the MarioKart period (April 1, 2020 to June 30, 2021). There are three reasons these dates and time intervals were chosen for analysis. First, collection of patient census data was part of a larger project within the Department; second, the Mariokart schedule was implemented near the beginning of the COVID-19 pandemic and the hospital census was low during the Spring of 2020; third, the three year period captured the experience of at least one full class in our program.
Missing patient census data were imputed using predictive mean matching via the mice package in R.13 Additionally, we also compared self-reported ACGME short-break violations via New-Innovations (Uniontown, OH), defined as time off of less than ten hours between work-related shifts and ACGME 80-hour violations, defined as a flag of worked hours more than 80 hours in a 7 day period. Numbers of both violations were collected for the overall program and for the general inpatient service where the MarioKart schedule was deployed during the study period. All violations were calculated as the mean number per month during the study period.
Median and interquartile ranges (IQR) were calculated for patient census data. To assess the patient care opportunities for the residents during the rotation, we created a metric called total patient days per team. This metric is calculated as the multiple of the median, lower, and upper IQR bounds of patient census by the number of days worked per rotation (24 in the traditional model, 22.5 in the MarioKart model). This metric is used to provide an estimate for comparison of the number of patient encounters a resident team would have during the duration of the 28 day rotation. A two-sided t-test was used to compare differences between the traditional and MarioKart models.
All statistical analyses were performed in R (R Studio, Boston, MA). The code for analysis is available on GitHub (https://github.com/fbuckhold3/MarioKart/). This study was formally determined by our Institutional Review Board (IRB) to not constitute Human Subjects Research and did not require IRB approval. The study followed the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) reporting guidelines.