Trends in Length of Stay and Readmissions in Children’s Hospitals

This retrospective cohort study included hospitalization encounters in the Pediatric Health Information System (PHIS) (Children’s Hospital Association, Lenexa, KS) database. Both “inpatient” and “observation” admissions were included together because of the variability and clinical similarities between many patients with these administrative billing statuses in children’s hospitals.^28,29  PHIS is an administrative database containing demographic, diagnostic, and procedural data from 51 tertiary care children’s hospitals, located in 25 states and representing 17 of the 20 major metropolitan areas in the United States.³⁰ 

Discharges from 42 participating hospitals between January 1, 2013, and December 31, 2017, were included. Each hospitalization was treated independently because only a small proportion of patients (<10%) were hospitalized multiple times within a given year and correlation of LOS across hospitalizations was low (Pearson’s correlation = 0.09), making model convergence challenging. Medical complexity was identified by using the complex chronic conditions (CCCs) algorithm developed by Feudtner et al.^31,32  CCCs are defined as medical conditions expected to last at least 12 months, involve several organ systems or 1 system severely enough to require specialty pediatric care, and have a high probability of hospitalization. The severity of illness for each encounter was measured by using the Hospital Resource Intensity Score for Kids (HRISK),³³  which calculates pediatric-specific relative cost weights on the basis of the encounter’s assigned group and severity of illness level from the All Patient Refined Diagnosis Related Groups (APR-DRG) (3M HIS, Minneapolis, MN) software. For descriptive purposes, APR-DRGs were further categorized into clinical service lines (medical or surgical) on the basis of the principal diagnosis or procedure of the APR-DRG. Service line definitions were developed by PHIS participating hospitals and the Children’s Hospital Association.³⁰  Within these general service lines, condition groups were defined on the basis of diagnosis or procedure-related codes for a particular procedure or organ system. For example, a patient with pneumonia with parapneumonic effusion assigned to APR-DRG 137 would fall under the medical respiratory condition group. A patient with pneumonia with parapneumonic effusion who underwent video-assisted thorascopic surgery would be assigned APR-DRG 121 and fall under the surgical respiratory condition group.

We excluded hospitalizations for obstetric care, normal newborn care, and those in which the patient died, and therefore could not be readmitted (n = 20 903), or left against medical advice. We also excluded condition groups with <1000 annual encounters in the entire cohort (eg, Hematology: Surgical, Substance Abuse: Medical, Mental Health: Surgical). Hospitals that did not participate in PHIS throughout the duration of the study period and hospitals that did not report LOS in hours were also excluded (n = 9 of 51 PHIS hospitals).

The main outcomes were hospital LOS (measured in hours) and 14-day all-cause readmission (measured from the index hospitalization date of discharge). A 14-day all-cause readmission interval was selected instead of the conventional 30-day window, because earlier readmissions may better reflect hospital care provided during the index admission while minimizing capture of unrelated readmissions.^34,35 

We summarized categorical variables with frequencies and percentages and compared discharge demographics and clinical characteristics across years from 2013 to 2017. Because of the nonnormal distribution of LOS, we log-transformed LOS before analysis. We then back-transformed LOS estimates for presentation. In unadjusted analyses, we calculated the mean annual LOS stratified by medical and surgical discharges and condition groups. We used generalized linear models to test for trends across study years in LOS and readmissions, as well as for trends in patients’ age and severity of illness (ie case-mix index calculated as the HRISK²⁷ ). We used logistic regression to assess for trends in binary demographic and clinical characteristics including CCCs, HRISK, and ICU admission. Factors found to be significantly changing over time at P < .10 were incorporated as fixed effects into adjusted models by using generalized linear mixed effects models. To account for hospital clustering, we included hospital as a random effect. Using these models, we produced estimates of annual mean LOS and readmission rate. A Bonferroni-corrected significance threshold of P < .001 was used to account for multiple comparisons. Next, we estimated theoretical cumulative hours reduced by summing the decrease in annual hours in 2014 to 2017 relative to 2013 (calculated by differencing each year’s mean adjusted LOS from the mean adjusted LOS in 2013 and multiplying by the number of discharges in that year). Finally, we conducted a subgroup analysis of LOS and readmissions for patients with CCCs versus patients without CCCs. Statistical analyses were performed by using SAS v.9.4 (SAS Institute, Inc, Cary, NC). This study was considered nonhuman subjects research by our university’s institutional review board.

There were 3 355 815 discharges included in the study, representing 2 156 474 individuals across 42 hospitals (Table 1). Seventy-five percent of discharges were within medical service lines and 25% were within surgical service lines. Overall, 48% of the discharges were for patients 0 to 4 years old. CCCs were associated with 39% of all hospitalizations. During the study period, there were modest, but statistically significant, temporal increases in mean age (6.4 years [SE 0.1] in 2013 versus 6.7 years [SE 0.1] in 2017, P < .001) and decreases in the proportion of children identified as non-Hispanic white and those with Medicaid (Table 2). Statistically significant temporal increases were also noted in severity (HRISK mean score of 2.3 [SE 0.1] in 2013 versus 2.5 [SE 0.1] in 2017, P < .001), the proportion of hospitalized children with a CCC (40.1% in 2013 versus 41.5% in 2017, P < .001), and the proportion of children admitted to an ICU (10.8% in 2013 versus 12.3% in 2017, P < .001) (Table 2).

In Table 3, we present mean adjusted LOS for discharges over the study period. Comparing 2013 with 2017, the overall mean LOS declined by 1.8 hours (3%), from 61.1 hours in 2013 to 59.3 hours in 2017 (P < .001). Reductions in adjusted LOS were noted in both medical (3.6% decline; P < .001) and surgical (2% decline; P < .001) service lines (Fig 1A). When ranked by the theoretical reduction in cumulative hours, medical respiratory, medical digestive, medical neuroscience, medical infectious disease, and medical neonatal care were the condition groups with the greatest aggregate declines over the study period. When stratified by percent change, the condition groups with the greatest proportional decreases in LOS were surgical infectious disease (−14.8%), surgical trauma care (−14.6%), surgical digestive services (−14.3%), surgical neuroscience (−10.4%), and medical endocrinology (−10.1%, P < .001 for all; Supplemental Table 5). Three condition groups experienced statistically significant increases in their LOS over the study period: medical mental health services (+8.4%), medical oncology services (+0.8%), and surgical cardiac care (+2.1%; Supplemental Table 5).

In our CCC subgroup analysis (Supplemental Table 7), average LOS was longer for patients with CCCs compared with those without CCCs (85.9 hours versus 44.1 hours in 2017); however, overall LOS for the CCC group was stable over the study period (87.3 hours in 2013 versus 85.9 hours in 2017, P = .97). In contrast, the overall adjusted LOS for the non-CCC group declined by 4.6% over the study period (46.3 hours in 2013 versus 44.1 hours in 2017, P < .001). Reductions were seen in mean adjusted LOS for patients without CCCs in both medical (4.4% decline, P < .001) and surgical (5.4% decline; P < .001) service lines.

Overall, adjusted 14-day readmission rates did not change over time (7.0% in 2013 vesus 6.9% in 2017; P = .03). When stratified by service line, medical service line–adjusted readmissions rates declined by 4.6% (7.6% in 2013 versus 7.3% in 2017, P < .001), whereas surgical service line readmission rates increased by 7.2% (5.2% in 2013 versus 5.6% in 2017, P < .001) (Table 4, Fig 1B). Readmission rates remained stable for most medical subgroups studied. Four medical condition groups demonstrated decreased readmission rates (endocrinology, −17%; cardiac care, −9.8%; infectious diseases, −7.9%; and digestive diseases, −6.8%; Supplemental Table 6). Only the medical respiratory condition group demonstrated an increased readmission rate over the study period (+8.6%; Table 4). Within the surgical service line, several surgical condition groups showed a positive trend in readmissions; however, no individual condition group trend met statistical significance (Supplemental Table 6).

In our subgroup analysis (Supplemental Table 8), the 14-day adjusted readmission rate was higher in 2017 for those discharges associated with a CCC (13.8%) than those without CCCs (3.2%). Within the CCC population, the overall adjusted readmission rate declined over the study period (14.1% in 2013 versus 13.8% in 2017, P < .001), whereas the readmission rate for those without CCCs remained stable (3.2% in 2013 versus 3.2% in 2017, P = .022). Further stratifying the CCC population, the medical service line–adjusted readmissions declined by 6.8% (16.1% in 2013 versus 15.0% in 2017; P < .001), whereas surgical service line readmission rate increased (9.1%. in 2013 versus 9.8% in 2017, P < .001). The only individual surgical condition group whose readmissions trend met statistical significance was the non-CCC surgical otolaryngology condition group, increasing from 2.8% in 2013 to 3.2% in 2017 (P < .001).

Efforts to reduce LOS and readmissions provide key strategies to improve inpatient care quality, increase hospital capacity, and reduce health care costs.^36,37  Optimization of these efforts may be challenged by a growing and increasingly complex patient population.^18,24  In our study, we found that from 2013 to 2017, when adjusting for temporal differences in complexity and demographic factors, aggregate LOS across a national sample of US children’s hospitals decreased by 3% and overall readmissions remained stable. These findings suggest that children’s hospitals are managing pediatric inpatients more efficiently despite increasing patient complexity.

The modest annual decrease in mean LOS observed in the current study may have limited clinical consequence for an individual hospitalization; however, even small changes in LOS can meaningfully impact hospital systems. Excess LOS may reduce functional bed capacity and turnover, which can in turn delay elective admissions and increase emergency department (ED) boarding. Previous adult and pediatric literature has indicated that ED boarding is associated with increased safety events, increased patient morbidity and mortality,³⁸  higher hospital costs, and overall longer hospital LOS.^39–41  Cancellation of elective procedures because of high hospital occupancy can also reduce hospital revenue and net income.⁴²  As health care institutions face pressure to improve efficiency while simultaneously containing costs, even small reductions in LOS can translate into important differences for capacity and throughput within both an individual hospital and hospital system.

Reductions in overall adjusted LOS varied across patient populations and conditions. The largest reductions in cumulative LOS were seen within the medical respiratory, medical digestive, and medical neuroscience condition groups. When stratified by complexity, the non-CCC population experienced a greater decline in adjusted LOS than the CCC population. The reasons for these declines are likely multifactorial and may reflect both inpatient and outpatient factors including greater implementation of clinical practice guidelines,⁴³  targeted discharge initiatives,⁴⁴  and improved outpatient management of some common acute inpatient pediatric conditions (eg, asthma, urinary tract infections, gastroenteritis).^15–17,45  We also identified rising LOS over time in a few specific condition groups, such as medical mental health. This particular example corresponds with previous findings revealing substantial increases in mental health encounters at children’s hospitals in the last decade⁴⁶  and may partially explain the observed increases in LOS for this condition group, and potentially others.

Shorter LOS has been associated with higher readmission rates in certain neonatal⁴⁷  and adult conditions.⁴⁸  Hospitalizations among children with CCCs have also been associated with longer LOS and more frequent readmissions compared with children without CCCs.^{18,21,26,49–51}  Few researchers, however, have evaluated LOS and readmission rates after discharge for a broad range of pediatric medical and surgical conditions. In our study, we found that overall readmission rates have remained stable, and the adjusted readmission rate for the CCC population declined over the study period. It has been suggested that robust care coordination and outpatient chronic condition management reduce both hospitalizations and ED visits for children with complex medical needs.^52–55  Future exploration of specific condition groups and individual diagnoses with reduced readmission rates could provide an opportunity to understand how lower readmission rates can be achieved despite increasing medical complexity. Conversely, identifying diagnosis groups with higher rates of readmission, such as medical respiratory and select surgical service lines, provides an important step toward developing targeted quality initiatives to improve these metrics at the hospital and national level.

Similar to authors of previous studies, we demonstrated that overall, mean LOS is short and readmission rates are low for pediatric hospitalizations.^56,57  Because of these low rates, setting reliable benchmarks to measure hospital performance has been challenging. Furthermore, we illustrate that neither LOS nor readmissions should be targeted in isolation. For example, the medical respiratory and surgical otolaryngology condition groups demonstrated reduced LOS but were also associated with a significant increase in readmissions. Identifying preventable readmissions is challenging in administrative data; however, this finding suggests that there may be limits to how far LOS can be reduced without impacting readmissions and other postdischarge health care use. Additional investigations to define expected condition-specific LOS and readmission rates, as well as understand the reasons for differences in readmission rates (including potential unintended consequences of efforts focused on reducing LOS), will be important to best inform quality initiatives.

In this study, we extend previous findings that the proportion of inpatient hospital use attributable to children with CCCs is rising.^18,21,49  Advances in medical and surgical technology have led to improved survival of preterm infants and children with congenital conditions, resulting in a higher prevalence of children with CCCs nationally.^58–60  Additionally, care for some common acute inpatient pediatric conditions may be shifting to the outpatient setting,^15–17  altering the proportion of patients with CCCs in the inpatient setting. Furthermore, regionalization and referral trends result in a greater percentage of children with specialty needs receiving care at higher-volume referral centers.^13,14,61  As major drivers of pediatric hospital use, children with medically complex conditions require high-value systems and comprehensive care coordination. As health care moves from a fee-for-service to a value-based, population-centered model, it will be increasingly important for children’s hospitals to not only identify and address outlier inpatient conditions associated with prolonged LOS and excess readmissions, but also continue investigations to improve inpatient to outpatient care transitions, enhance care coordination across the continuum, and identify and intervene on the social determinants of hospital use.

Our study has several limitations. Use of administrative data limits evaluation of complete patient-level clinical data and confirmation of diagnostic coding accuracy. Second, grouping by service line is well-described; however, combining similar hospitalizations into condition groups may obscure important differences between individual diagnoses and treatment teams represented within a single condition group. Third, because we used an all-cause readmission measure, we may have captured planned readmissions as well as readmissions unrelated to the index hospitalization, although we attempted to minimize this by using a 14-day readmission window instead of the conventional 30-day window. Additionally, readmissions to a different hospital were not captured; thus, our findings underestimate true readmission rates.⁶²  Furthermore, although researchers in studies focused on adult populations have used readmission rates as a surrogate for adverse outcomes, including suboptimal discharge planning, poor surgical care quality, and premature discharge,^63,64  a cause and effect relationship has not been established for pediatric patients. Therefore, readmission rates may not be a sensitive or comprehensive measure of quality in pediatric inpatient care. Finally, our assessments are restricted to a subset of tertiary care children’s hospitals, potentially limiting the generalizability of our findings for other community or non-PHIS hospital settings.

With our findings, we suggest that children’s hospitals are providing increasingly efficient care by decreasing LOS without a rise in adjusted readmission rates despite an increasingly complex inpatient population. Such information may have important implications for hospitals and policy makers as they seek to provide high-value care in the most efficient hospital environments given finite health care resources.