The study protocol and consent forms were approved by Promedica Toledo Hospital and the University of Toledo institutional review boards (IRBs). We conducted a classical prospective cohort study with pre-specified intervals for data collection and follow-up using standardized protocols. Patients were recruited and consented between 2/25/2019 and 4/30/2019. After recruitment concluded, an IRB-approved amendment to protocol expanded the study sample to include patients who were missed and not consented during the recruitment window. The amendment complied with “common rule” regulations.22
This study was conducted in Promedica Toledo hospital, an 800-bed tertiary center located in Toledo, Ohio. Promedica Health System includes 13 hospitals and serves Northwest Ohio and Southeast Michigan.
Internal medicine services are divided into teaching services and non-teaching services. At the time of admission, patients are screened and assigned to either service type by a board-certified physician. Admission assignments are rotated between teams and depend on the team’s daily census. Teaching services have a cap of 20 patients per team, unlike hospitalists who lack a cap. The admitting teaching service had a cap of 12 new admissions every 24 h, unlike non-teaching services, where no gap existed. Teaching services are staffed by 19 physicians. Non-teaching services are covered by 22 hospital-employed physicians or 24 locum tenens physicians. Locum hospitalists and employed hospitalists could cover any of the non-teaching teams non-preferentially. There was no mid-level support provided in non-teaching services. Further details about intervention are described in a supplementary file.
The cohort study used dynamic recruitment, and the study population included all consecutive adult patients 18 years and older admitted for any medical reason to IM services at Promedica Toledo Hospital during the recruitment period. Patients were excluded if they refused to join the study, retracted their consent post recruitment, or were transferred to another non-IM service during hospitalization. Patients transferred to the intensive care unit after admission to the IM department were not excluded. Patients who were missed at the recruitment period were included, and their data were collected retrospectively.
Screening and Consenting Protocol
Every 12 h, all patients admitted to IM services were screened for eligibility and assigned to physicians to obtain consents, baseline characteristics, and contact information of patients and family members for outcomes ascertainment post-discharge. A team of 10 physicians contributed to consenting over 65 consecutive days, and at least three physicians obtained consents daily.
Data extracted by physicians included demographics, food security (yes/no), residence before admission, insurance type, principal admission diagnosis, and comorbidities. The primary exposures in this study were the types of admitting physicians: teaching IM services, employed hospitalist services, or locum hospitalist services. Further details about the methodology used to collect variables and exposure are described in a supplementary file.
The primary outcome of the study was the LOS. The secondary outcomes of the study were HC, inpatient mortality, 30-day readmission, and 30-day mortality. Hospitalization outcomes included LOS, HC, and inpatient mortality. Length of stay was measured in integer days. A stay length of 1 day was defined as being admitted to IM services for any duration of a calendar day between 12:00 a.m. and 11:59 p.m. Hospital costs included the sum of variable direct and fixed direct costs. Senior analysts in Promedica’s financial department provided hospital cost outcomes for analysis. Further details regarding HC are described in supplementary. Inpatient mortality (yes/no) was recorded as a “yes” if the patient died during their index hospitalization.
Post-hospitalization outcomes included the 30-day all-cause mortality rate and the 30-day all-cause readmission rate. All-cause readmission included any admission for any medical reason to any hospital, excluding elective non-urgent admissions (i.e., elective surgery) and admissions to psychiatry hospitals. Both observation and inpatient status were counted as admissions.
Both active and passive ascertainment captured 30-day mortality and readmission outcomes. A group of 4 physicians contacted all enrolled patients on or shortly after 31-days post-discharge, and inquired about readmission within the 30 days post-discharge. Patients who did not answer were contacted through other methods, including email, text messages, voice messages, or via proxy through family members or other contacts based on patient preference. In these situations where the patient was not reached directly by phone, at least two additional separate attempts to contact the patient, family members, or friends for outcome ascertainment per patient preference.
Passive ascertainment was determined by checking Promedica electronic health records and the University of Toledo health medical records for readmission or mortality. Health records were checked at least 3 months post-discharge to maximize data accuracy. For patients with whom an investigator failed to contact post-discharge, surrogates were used to assess vitality. Surrogates included clinic visits, emergency visits, or laboratory checks at any period after discharge.
Patients were determined to have survived 30 days or more post-discharge if they were contacted successfully, or a surrogate was used as evidence of vitality. Readmission events and dates were determined if the patient reported readmission or readmission was evident in checking electronic medical records. Readmission and mortality for patients who were missed during consenting were determined passively.
Physicians collected data from collection sheets used at the screening process, Promedica and University of Toledo electronic health records, communication with patient/family members, and Epic’s Care Everywhere health network. All outcomes were collected or calculated electronically to maximize accuracy. After compiling all data, it was checked extensively for accuracy.
Promedica Health System utilizes Epic software, which supports a health network connecting all hospitals that utilize Epic; this feature is called Care Everywhere®. Data was also mounted from Care Everywhere to cover a large geographical area, including most referral hospitals in Northwest Ohio and Southeast Michigan. Referral health systems that use Care Everywhere include Mercy Health, Cleveland Clinic, Ohio State University Medical Center, Henry Ford Health System, University of Michigan, and Beaumont Health System.23 Lucas County, including Toledo city, is served by three separate hospital systems: Promedica Health Care, Mercy Health, and University of Toledo Medical Center.24