Discharge Time of Day and 30-day Hospital Reutilization at an Academic Children’s Hospital Free

We conducted a retrospective cohort study of all admissions to a 140-bed, urban, tertiary care academic children’s hospital in the Bronx, New York, that serves a predominantly low-income racial and ethnic minority population. The hospital’s inpatient units consist of 3 general medical units for pediatric hospital medicine and subspecialty teams and a PICU. The nursery and NICU are located at a different site within the health care system and were not included in the study.

We included all children, 0 through 17 years of age, who were discharged from the hospital between July 1, 2016 and December 31, 2019. This end date was chosen to avoid confounders related to the severe acute respiratory syndrome coronavirus 2 pandemic. We excluded children who were discharged directly from the PICU, transferred to another institution, left against medical advice, or died. To identify a cohort with a typical length of stay, we excluded children whose length of stay was <12 hours or >21 days, which accounted for 3.1% and 1.2% of all hospital discharges, respectively (Fig 1). Multiple index admissions and readmissions were counted for each patient, and each readmission was counted as an index admission for subsequent readmissions. We extracted patient demographic and clinical data from the electronic health record. The hospital’s institutional review board approved this study.

The exposure of interest was the discharge time of day, defined as the timestamp recorded in the electronic health records by a unit clerk when the patient left the inpatient unit after being discharged. We determined that the final departure timestamp was a more consistent marker than the discharge order timestamp because there was significant variability in the time lag between a patient being medically ready and the discharge order being placed at the institution. Taking into account the shift change times for house staff (6 am and 6 pm) and nurses (7 am and 7 pm), and based on a consensus from pediatric hospital medicine attending physicians at the institution, morning discharge was defined a priori as a discharge between 8 am and 12:59 pm, afternoon discharge was defined as between 1 pm and 5:59 pm, and evening discharge was defined as between 6 pm and 10:59 pm. Patients who were discharged between 11 pm and 7:59 am were excluded because discharges during those hours are unlikely to be planned and are rare, accounting for <1% of all hospital discharges.

The primary outcome was all-cause hospital reutilization, a composite of emergency department (ED) visits and hospital readmissions within 30 days after the index hospital discharge. Hospital reutilization was treated as a binary variable (ie, ED visit and/or readmission versus none). Only same-hospital reutilization was captured.

Patient demographic variables included age, sex, and insurance type. Data on race, ethnicity, and preferred language were extracted but not included in the analysis because of a high proportion of missing data (eg, 15% missing ethnicity data), low data reliability (ie, a combination of data reported by self and selected by hospital personnel), and our intention to avoid reporting race and ethnicity in isolation without appropriate sociodemographic factors and social determinants.¹⁵ 

Patient clinical variables included the number of complex chronic conditions (CCCs), the presence of previous hospitalization in the past 90 days, the presence of a documented primary care physician at discharge, surgery during the index admission, PICU stay during the index admission, length of stay, hospital inpatient unit at discharge, weekend discharge, and Medicare Severity Diagnosis Related Group relative weight. CCCs were defined by using the criteria published by Feudtner et al.¹⁶  Diagnosis related group relative weights reflect the relative resource consumption of the hospitalization.¹⁷  All covariates were determined a priori on the basis of the existing literature and clinical knowledge from the authors.^18–21 

Categorial variables were summarized by using frequencies and percentages, and nonnormally distributed continuous variables were summarized by using medians and interquartile ranges. The χ², Wilcoxon rank, and Kruskal–Wallis tests were used, as appropriate, to assess the relationship between each variable of interest and the discharge time of day, as well as the relationship between each variable of interest and the outcome of interest. χ² tests were used to compare unadjusted hospital reutilization rates.

To avoid convergence issues with the binomial regression model, we used a multivariable modified least-squares regression with robust SEs (heteroskedasticity-robust standard errors, HC2) to estimate the absolute rate differences of 30-day hospital reutilization rates using the identity link,²²  adjusting for confounders as identified above. A 2-sided α value of 0.05 was used to compare the adjusted hospital reutilization rates between morning and evening, as well as between afternoon and evening.

Our initial working hypothesis aimed to test the noninferiority of evening discharge. This hypothesis was based on our initial goal to demonstrate that, although evening discharge might be operationally less preferred, it may not be unacceptably worse than morning or afternoon discharge (ie, it may be noninferior). The analysis to test the noninferiority of evening discharge compared with morning and afternoon discharge is described in detail in the Supplemental Information and Supplemental Table 5. This manuscript is focused on superiority testing conducted after evening discharge was shown to be not noninferior.

A secondary analysis was performed to narrow the outcome of interest from 30-day all-cause reutilization to 30-day unplanned hospital reutilization. All ED visits were assumed to be unplanned. Determining unplanned readmissions using the timing of hospital registration was previously validated,²³  but this data field was not consistently available in our cohort. Any readmission that was planned at least 24 hours previously was considered planned. Hospital readmissions were assumed to be unplanned if children were readmitted through the ED, readmitted directly to the PICU, or transferred from another hospital. Readmissions were assumed to be planned if children were readmitted from home through the operating room or had chemotherapy as their principal diagnosis. The principal diagnoses of the readmissions were reviewed to identify potential outliers. These readmission assumptions were further validated through an in-depth medical record review conducted by a single reviewer (JL) on a nonrandomly selected subset of 10.3% of these readmissions, preferentially selecting the readmissions with ambiguous principal diagnoses. Within this subset, 94.0% of records were categorized correctly on the basis of the assumptions, and those who were incorrectly categorized were reclassified. For children who were readmitted through a different mechanism (ie, 18.9% of all readmissions, eg, readmitted through a clinic), a medical record review was performed to determine the planned or unplanned nature of the readmissions. In total, a manual medical record review was conducted on 27.3% of all readmissions (1189 out of 4356) to determine the nature of the readmission.

To address the nonindependence of outcomes due to multiple admissions, a second sensitivity analysis was performed in which children with multiple admissions during the study period had 1 admission randomly selected for inclusion in the analysis. A multivariable modified least-squares regression model was again generated to estimate adjusted hospital reutilization rate differences, as described above.

We also used a covariate balancing propensity score in an inverse probability treatment weighting (IPTW) model for a third sensitivity analysis. All covariates were determined a priori, as described above. The propensity scores were created independently of the outcome. The propensity score balance was examined using the standardized mean difference of the covariates between the 2 groups, both before and after the IPTW (see Supplemental Figs 2 and 3).

All statistical analyses were performed by using Stata version 17.0 (StataCorp, College Station, TX), and 2-sided P values < .05 were considered statistically significant.

A total of 24 994 hospital discharges from 17 314 unique patients met the inclusion criteria. Of these discharges, 80.1% were from patients residing in the same 42-square-mile borough as the hospital, located within 10 miles.²⁴  In total, 6103 (24.4%) discharges occurred in the morning, 13 786 (55.2%) discharges occurred in the afternoon, and 5105 (20.4%) discharges occurred in the evening.

Patients discharged in the evening were older compared with those discharged in the morning or afternoon (Table 1). The proportions of patients with 1 or more CCC or hospitalization in the past 90 days were higher in patients discharged in the evening than those in patients discharged in the morning, whereas the proportions were similar in patients discharged in the evening compared with those in patients discharged in the afternoon. The length of stay and the proportion of patients who were in the PICU during hospitalization were not different across the groups based on discharge time (P = .31 and P = .43, respectively).

The unadjusted 30-day all-cause hospital reutilization rate in the cohort was 17.4% (4356/24 994), which was a composite of the unadjusted 30-day ED visit rate of 8.3% (2073/24 994) and unadjusted 30-day readmission rate of 10.1% (2525/24 994). The unadjusted 30-day hospital reutilization rate was 14.1% in patients discharged in the morning, which is significantly lower than the 18.2% reutilization rate in the afternoon or the 19.3% reutilization rate in the evening (P < .001). In patients with 30-day hospital reutilization, there was a higher proportion of patients with public insurance (83.2% vs 76.2%), 1 or more CCC (43.9% vs 21.6%), and a history of hospitalization in the 90 days before the index admission (36.5% vs 12.9%) compared with those without 30-day hospital reutilization (P < .001 for all, Table 2). The proportion of weekend discharge was not different between the 2 groups (P = .10).

The adjusted 30-day hospital reutilization rate was lowest in the morning (6.1%, 95% confidence interval [CI] 4.1%–8.2%), followed by the afternoon (9.0%; 95% CI 7.0%–11.0%) and evening (10.1%, 95% CI 8.0%–12.3%, Table 3). Morning discharge had a significantly lower adjusted 30-day all-cause hospital reutilization rate compared with evening discharge (P < .001), whereas afternoon discharge did not have a significantly different 30-day hospital reutilization rate compared with evening discharge (P = .06).

In general, the results were consistent across multiple sensitivity analyses (Table 4). The unadjusted 30-day unplanned readmission rate was 7.4%, and the unadjusted 30-day unplanned hospital reutilization rate was 14.8%. Morning discharge had a significantly lower adjusted 30-day unplanned hospital reutilization rate than evening discharge (P < .001), whereas afternoon discharge did not have a significantly different rate compared with evening discharge (P = .06). Similarly, when narrowing the cohort to 1 randomly selected admission per patient (cohort described in Supplemental Tables 6 and 7), morning discharge had a significantly lower adjusted 30-day unplanned hospital reutilization rate than evening discharge (P < .001), whereas afternoon discharge did not have a significantly different rate than evening discharge (P = .05). The results were also consistent when the propensity score model was used (Table 4).

In this single-center retrospective cohort study, the adjusted all-cause 30-day hospital reutilization rate was higher in evening discharge than morning discharge, whereas the rates were not significantly different between evening and afternoon discharges. These results were generally consistent across multiple sensitivity analyses.

One of the strengths of our study is the use of multiple sensitivity analyses planned a priori to enhance the reliability of the findings from the primary analysis. In particular, our analyses included an inverse probability treatment weighting model, which balanced the differences in demographic and clinical characteristics among patients between discharge time groups and reduced bias in estimating hospital reutilization rates.

We hypothesize several reasons why hospital reutilization rates were higher in patients discharged in the evening. First, there may be confounding by indication (eg, medical complexity, reason for admission, and complicated hospital course) that was inadequately accounted for, despite adjusting for available demographic and clinical variables. The description of our cohort divided by the discharge time of day indicates that a larger proportion of patients discharged in the evening have 1 or more complex chronic conditions and previous hospitalizations than those discharged in the morning. We suspect that children discharged in the evening are more likely to have complicated discharge needs and, therefore, more opportunities for suboptimal execution of discharge plans than children discharged in the morning with more straightforward discharge needs. By adjusting for patient-level factors (ie, complexity of patients and their hospital course), we attempted to establish adjusted hospital reutilization rates that capture the differences in the outcome stemming from potential system-level differences in discharge processes at different times of the day. A proportion of the adjusted rate difference in the outcomes may still stem from patient-level characteristics that were not captured, despite our use of primary analysis and additional sensitivity analyses in an attempt to account for these differences. Second, apart from confounding by indication, there may be true system-level barriers to high-quality discharges in the evening within our system. These barriers may include incomplete handoff between providers, a higher patient-to-provider ratio, inadequate preparation when caregivers unexpectedly request an evening discharge, limited access to discharge medications, and difficulty arranging outpatient follow-up appointments. We expect some of these barriers to be generalizable beyond our center. Further identification of the facilitators and barriers to achieving high-quality discharges at all hours and a delineation of their relationships will be important for designing effective interventions.

The unadjusted 30-day emergency department visit rate in our cohort was 8.3%, which is higher than the previously reported rate of 6.2% in a study using a large administrative database.²⁵  Meanwhile, the unadjusted 30-day unplanned readmission rate in our cohort was 7.4%, which is comparable to previously reported rates of 6.5%,²⁰  7.2%,²⁵  and 8.2%.¹⁸  These findings may reflect the characteristics of the local pediatric population and their tendency to overly rely on ED visits instead of primary care physician visits.

From an operational perspective, weekend and evening discharge share similar characteristics with regard to staffing and access to resources. The authors of previous studies have examined the association between weekend discharge and readmission.^18,26  A single-center study did not reveal an association,²⁶  whereas a multicenter study using a large administrative database revealed an overall higher rate of 30-day readmission with weekend discharge compared with weekday discharge.¹⁸  We initially expected that facilitators and barriers to safe evening discharge would be similar to those for safe weekend discharge. Interestingly, in our cohort, despite having a higher 30-day hospital reutilization rate in evening discharge, there was no difference in hospital reutilization rate between weekend and weekday discharge. This finding suggests that the relationship between the system-level factors and the quality of evening discharge may be different from the relationship between the system-level factors and the quality of weekend discharge. A similar observation was noted in a meta-analysis of studies in an adult ICU in which evening or nighttime discharge was associated with an increased risk of hospital mortality compared with daytime discharge, whereas weekend discharge was not associated.²⁷ 

As an observational study, there are limitations to highlight. Although we adjusted for >10 demographic and clinical variables, there may be other confounders, including other clinical or social determinant of health variables that were unavailable to be collected. These factors could influence the association between the discharge time of day and hospital reutilization. For instance, the distance from home to the hospital is an unmeasured confounder in this study that could influence the family’s ability to leave the hospital at certain times of the day or impact their decision to revisit the hospital after discharge. Second, inaccurate electronic health record data may have resulted in the misclassification of the exposure and covariates. Third, our study captured hospital reutilization only within our health care system. Although most patients in our study resided within 10 miles of the hospital and we expect most patients to return to the same hospital, it is unknown whether patient’s characteristics, such as medical complexity, influence their decision to seek care elsewhere. However, we expect that the misclassification of hospital reutilization from seeking care at nonaffiliated facilities would be independent of the discharge time of day, and therefore, it is not expected to substantially bias the reutilization outcome. Fourth, this is a single-center study, and generalization is limited by specific institutional factors related to patient population, staffing, and organizational structure. It is unclear how these factors may bias the results. Although generalizability is limited, our cohort is large, spans over 3 years, and includes patients with all diagnoses. Lastly, hospital reutilization is only 1 aspect of health outcomes after hospital discharge. Other outcomes, such as patient satisfaction, return to baseline health status, and comprehension and adherence to discharge instructions would have enriched the study but were not feasible to obtain in a retrospective study.

We found that children discharged in the evening had a higher 30-day hospital reutilization rate than those discharged in the morning. The findings of our study highlight a need to critically examine the discharge process at all hours and identify modifiable factors to achieve consistently high-quality discharge from the hospital. Further studies may include multicenter studies to enhance the generalizability of the findings, as well as a mixed-methods approach to understanding facilitators and barriers to timely, efficient, and safe discharges from the perspectives of both health care providers and caregivers.

We wish to acknowledge the Montefiore Einstein Center for Health Data Innovations including Kasra Jabbary Moghaddam, MS, MA, and Erin Henninger, MPH, for their contribution in data collection.