The rates were adjusted for age and Chronic Condition Indicators. Error bars indicate 95% confidence intervals.
Solid line through the middle of each box indicates the mean rate; dotted lines, ±1 SD of the mean; box boundaries, ±2 SD of the mean; error bars, minimum and maximum rates.
Berry JG, Toomey SL, Zaslavsky AM, et al. Pediatric readmission prevalence and variability across hospitals. JAMA. doi:10.1001/jama.2012.188351.
eFigure 1. Study Cohort
eTable 1. Hospital and Patient Characteristics of the Study Cohort Compared with a Nationally Representative Sample of Hospitalized Children
Jay G. Berry, Sara L. Toomey, Alan M. Zaslavsky, Ashish K. Jha, Mari M. Nakamura, David J. Klein, Jeremy Y. Feng, Shanna Shulman, Vincent W. Chiang, William Kaplan, Matt Hall, Mark A. Schuster. Pediatric Readmission Prevalence and Variability Across Hospitals. JAMA. 2013;309(4):372–380. doi:10.1001/jama.2012.188351
Author Affiliations: Divisions of General Pediatrics (Drs Berry, Toomey, Nakamura, Shulman, Chiang, and Schuster and Messrs Klein, Feng, and Kaplan), Infectious Diseases (Dr Nakamura), and Emergency Medicine (Dr Chiang), Boston Children's Hospital, Boston, Massachusetts; Departments of Health Care Policy (Dr Zaslavsky) and Pediatrics (Drs Berry, Toomey, Nakamura, Chiang, and Schuster), Harvard Medical School, Boston; Department of Health Policy and Management, Harvard School of Public Health, Boston (Dr Jha); and Children's Hospital Association, Overland Park, Kansas (Dr Hall).
Importance Readmission rates are used as an indicator of the quality of care that patients receive during a hospital admission and after discharge.
Objective To determine the prevalence of pediatric readmissions and the magnitude of variation in pediatric readmission rates across hospitals.
Design, Setting, and Patients We analyzed 568 845 admissions at 72 children's hospitals between July 1, 2009, and June 30, 2010, in the National Association of Children's Hospitals and Related Institutions Case Mix Comparative data set. We estimated hierarchical regression models for 30-day readmission rates by hospital, accounting for age and Chronic Condition Indicators. Hospitals with adjusted readmission rates that were 1 SD above and below the mean were defined as having “high” and “low” rates, respectively.
Main Outcome Measures Thirty-day unplanned readmissions following admission for any diagnosis and for the 10 admission diagnoses with the highest readmission prevalence. Planned readmissions were identified with procedure codes from the International Classification of Diseases, Ninth Revision, Clinical Modification.
Results The 30-day unadjusted readmission rate for all hospitalized children was 6.5% (n = 36 734). Adjusted rates were 28.6% greater in hospitals with high vs low readmission rates (7.2% [95% CI, 7.1%-7.2%] vs 5.6% [95% CI, 5.6%-5.6%]). For the 10 admission diagnoses with the highest readmission prevalence, the adjusted rates were 17.0% to 66.0% greater in hospitals with high vs low readmission rates. For example, sickle cell rates were 20.1% (95% CI, 20.0%-20.3%) vs 12.7% (95% CI, 12.6%-12.8%) in high vs low hospitals, respectively.
Conclusions and Relevance Among patients admitted to acute care pediatric hospitals, the rate of unplanned readmissions at 30 days was 6.5%. There was significant variability in readmission rates across conditions and hospitals. These data may be useful for hospitals' quality improvement efforts.
Clinicians, hospitals, health systems, patients, and purchasers of health care are using readmission rates as an indicator of the quality of care that patients receive during a hospital admission and after discharge.1- 3 About 20% of hospitalized elderly Medicare beneficiaries are readmitted within 30 days, and the readmission rates vary greatly across hospitals.4,5 This variation is believed to indicate that a substantial proportion of readmissions may be preventable.6,7 The Affordable Care Act mandates that the Centers for Medicare & Medicaid Services reduce Medicare payments to hospitals with excessively high readmission rates.8
Although readmissions for adults have been the subject of substantial research,1- 8 readmissions for children have received less attention.9- 11 However, there has been a recent increase in interest in pediatric readmissions. For example, the Pediatric Quality Measures Program, established by the Children's Health Insurance Program Reauthorization Act, has identified pediatric readmissions as one of the first measures it will develop.12,13 In addition, the federal Partnership for Patients initiative has challenged hospitals to reduce pediatric readmissions by 20%.14,15
To understand potential opportunities to improve pediatric practice and reduce readmissions, information is needed on which diseases have the highest number of readmissions and whether there are differences in readmission rates across hospitals. Therefore, we analyzed data from 72 children's hospitals to examine the percentage of hospitalized children who have unplanned readmissions, which admission diagnoses have the most readmissions, and whether readmission rates vary across hospitals.
We conducted a retrospective analysis of the National Association of Children's Hospitals and Related Institutions (NACHRI) Case Mix Comparative data set of patients 18 years and younger who were discharged between July 1, 2009, and June 30, 2010, from 72 acute care children's hospitals in 34 states. The hospitals voluntarily submitted their data to NACHRI so that comparative analyses could be performed to identify best practices and clinical areas needing care improvement.16 The NACHRI Case Mix is the largest data set of children's hospitals that links patients across hospitalizations, enabling readmissions analyses. Boston Children's Hospital's institutional review board approved the study with an informed consent exemption.
We excluded index admissions for labor and delivery, newborns with a routine birth, and chemotherapy and for patients who left against medical advice, were transferred to another acute care hospital, or died (eFigure).17 Pregnant adolescents admitted for pregnancy-related (eg, pre-eclampsia) and other health problems (eg, asthma) and births for newborns with a nonroutine condition (eg, congenital cytomegalovirus) were included. We used the Agency for Healthcare Research and Quality (AHRQ) Kids' Inpatient Database (KID) 2009 to compare characteristics of our NACHRI cohort with a nationally representative sample of hospitalized children (eTable).18
The first unplanned admission within 30 days of an index admission was defined as a readmission. Additional admissions within 30 days were not counted as readmissions or index admissions. An additional admission after 30 days was counted as a new index admission. We measured readmissions following all-condition admissions and following the 10 condition-specific admissions with the highest readmission prevalence (ie, admissions with the greatest number of readmissions). We used All-Patient Refined Diagnosis-Related Groups (APR-DRG) version 25 (3M Health Information Systems) to classify the primary diagnosis for each admission. APR-DRGs include 316 mutually exclusive groupings of clinically related diagnosis and procedural codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM).19,20 We refer to each APR-DRG as a diagnosis throughout the article. We ranked each diagnosis by the number of associated readmissions to determine which diagnoses had the most readmissions.
We measured unplanned readmissions for any reason (ie, all-cause) because patients might be readmitted for related conditions even if the index admission and readmission diagnoses differ. We also examined readmissions with the APR-DRG and with the APR-DRG Major Diagnostic Category that match, respectively, the diagnosis and the organ system/related etiology of the index admission. To exclude planned readmissions, pediatric specialists reviewed every ICD-9-CM procedure code (n = 4066) and identified procedures for their specialty that are usually planned—defined as ones that are scheduled in advance in more than 80% of cases (eg, spinal fusion, Nissen fundoplication). They identified 2418 such procedures; when one of them was coded as the primary procedure, the readmission was classified as a planned readmission.
We adjusted for age and chronic conditions because their association with readmission risk is believed to be related to intrinsic patient factors rather than quality of care.21 Variation in readmission risk by age and chronic conditions was significant (P < .001 for each; Table 1 and Table 2). Moreover, there was statistically significant heterogeneity in age and chronic conditions across hospitals (P < .001). Therefore, these adjustments accounted for hospitals with a greater proportion of patients with a high readmission risk due to the types of patients they served and lessened the likelihood of hospitals being inappropriately labeled as an outlier.
We categorized age as less than 1 year, 1 to 4 years, 5 to 12 years, and 13 to 18 years. To identify chronic conditions, we used AHRQ's Chronic Condition Indicator (CCI) classification system,22- 24 which dichotomizes approximately 14 000 ICD-9-CM diagnosis codes into chronic or nonchronic conditions and aggregates chronic conditions into 1 of 18 mutually exclusive clinical groups (Table 2). We included all groups in our analyses; however, we combined 2 groups, “complications of pregnancy, childbirth, and the puerperium” and “certain conditions originating in the perinatal period,” because of their small size in our cohort. We also adjusted for the number of CCI groups for each index admission as an indicator of medical complexity.24,25
We assessed differences in readmission rates by patients' insurance type (public, private, no insurance, other), race/ethnicity (black, Latino, white, and other), and length of stay (LOS) and by children's hospital characteristics, including freestanding vs nonfreestanding, geographic region, and number of annual index admissions. Because adjustment for these characteristics raises the possibility of adjusting for factors that inappropriately influence quality,2,26 we did not include them as case-mix adjusters. Instead, we assessed the degree to which the variation in adjusted readmission rates across hospitals was explained by these characteristics.
We used hierarchical logistic regression models to assess differences in readmission rates by patient and hospital characteristics. We began with hospital random-effect models, each with 1 fixed effect per characteristic of interest. A multiple degrees-of-freedom block test was used to determine statistical significance for each characteristic. Characteristics with an effect of P < .20 were then entered, simultaneously, into a multivariable model. Preplanned pairwise comparisons of subgroups (eg, Latino vs white) were performed in the multivariable model.
To assess variation in readmission rates across hospitals, we began with a hospital random-effect model without any fixed effects. We subsequently derived an adjusted model by adding fixed effects for each case-mix adjuster (age, each CCI group, and CCI count). To assess model overfitting, we built models with 1, 2 (all combinations), and 3 case-mix adjusters and compared the Akaike information criterion (AIC) among the models.27 As additional case-mix adjusters were added, the AIC decreased, suggesting that the model with all 3 case-mix adjusters was not overfit. In the adjusted model, all 3 case-mix adjusters remained significantly associated with readmission risk (P < .001), suggesting that each adjuster was explaining a significant amount of variance beyond the other adjusters.
We estimated the variation of adjusted readmission rates across all hospitals and used the statistical significance of the variance component for the hospital random effect to partition the variation into systematic variation vs variation due to chance. The 2-sided significance threshold was P < .05. We then added fixed effects for additional patient and hospital characteristics (eg, patient race/ethnicity and type of children's hospital) into the adjusted model to assess whether the variability in adjusted readmission rates across hospitals was reduced when accounting for these characteristics.
There were 450 index admissions (<0.1%) excluded from analysis because of missing data for variables that were necessary to define the study cohort (eFigure). For assessment of the effects of additional patient and hospital characteristics, 17 464 (3.1%) and 25 643 (4.5%) index admissions that were missing race/ethnicity and insurance status, respectively, were excluded in bivariate analysis; 42 117 index admissions (7.4%) were excluded in multivariable analysis because of missing race/ethnicity, insurance status, or both. Data were analyzed using SAS version 9.3 (SAS Institute).
We examined 568 845 index admissions for readmission within 30 days (eFigure). The 30-day readmission rate was 6.5% (n = 36 734); among readmitted children, 39.0% (n = 14 325) were readmitted in the first 7 days and 61.6% (n = 22 628) in the first 14 days. Median age at admission was 3 years (interquartile range, 0-10), and 55.2% (n = 299 812) had public insurance (Table 1). Thirty-five percent (n = 197 972) had 1 CCI, and 29.9% had 2 or more CCIs. The most common CCIs among children in the cohort were congenital anomalies (n = 111 588; 19.6%), such as congenital heart disease; respiratory diseases (n = 101 775; 17.9%), such as asthma; and neurologic diseases (n = 84 450; 14.8%), such as cerebral palsy (Table 2). Our cohort had a distribution of age, race/ethnicity, and insurance type similar to the nationally representative sample of hospitalized children in the KID. However, more children in our cohort had 2 or more CCIs and the hospitals had a larger number of annual admissions (eTable).
In bivariate analyses, readmission rates varied significantly (P < .001) for each case-mix adjuster (ie, age, CCI group, and CCI count). Readmission rates were higher for children aged 13 to 18 years (7.6%) than for children aged 5 to 12 years (6.1%), 1 to 4 years (6.2%), and less than 1 year (6.2%) (P < .001) (Table 1). Among the CCI groups, the highest and lowest readmission rates were observed for children with neoplasms (21.1%) and chronic respiratory diseases (6.1%), respectively (Table 2). Readmission rates increased as CCI count increased: 5.4% for 1 CCI, 9.4% for 2 CCIs, 12.4% for 3 CCIs, and 16.8% for 4 or more CCIs (P < .001) (Table 1). Patients' age, CCI group, and CCI count remained significantly associated with the likelihood of readmission in multivariable analysis (P < .001 for all).
Readmission rates varied by patients' insurance type, race/ethnicity, and LOS; by hospitals' number of annual admissions; and by hospital type (Table 1). For example, readmission rates were 6.9% for patients with public insurance, 5.9% for private insurance, 4.5% for no insurance, and 6.2% for other insurance (P < .001). The rates were 6.9% for black individuals, 6.5% for Latino individuals, 6.5% for white individuals, and 6.2% for patients with other race/ethnicity (P < .001). Readmission rates were higher for patients with a longer LOS; rates were 4.6% for patients with a LOS 1 to 2 days, 6.2% for LOS 3 to 4 days, 7.8% for LOS 5 to 6 days, and 11.2% for LOS 7 days or longer (P < .001). Insurance type, race/ethnicity, and LOS remained significantly associated with the likelihood of readmission in multivariable analysis.
Unadjusted readmission rates varied significantly across hospitals (P < .001), and this variation persisted after adjusting for age, CCI group, and CCI count (P < .001). Significant variation was observed when measuring all-cause readmissions and when measuring readmissions for the same diagnosis as the index admissions (P < .001 for each). To illustrate the magnitude of variation in all-cause readmissions across hospitals, we described the rates for hospitals 1 SD above vs below the mean: adjusted rates were 28.6% greater in hospitals above vs below (7.2% [95% CI, 7.1%-7.2%] vs 5.6% [95% CI, 5.6%-5.6%], respectively). The corresponding rate was 65.4% greater for hospitals 2 SD above vs below the mean (8.1% [95% CI, 8.1%-8.1%] vs 4.9% [95% CI, 4.9%-4.9%], respectively) (Figure 1). Moreover, a multivariable model that additionally adjusted for patients' insurance type, race/ethnicity, and LOS did not significantly change the amount of variance in readmission rates across hospitals (P = .25).
Figure 2 lists the 10 admission diagnoses with the highest readmission prevalence. Collectively, these admission diagnoses account for 27.7% (n = 10 162) of all readmissions. The highest rates for condition-specific unadjusted 30-day readmissions were for admissions for anemia or neutropenia (22.5%; n = 1092), ventricular shunt procedures (18.1%, n = 675), and sickle cell anemia crisis (16.9%, n = 1019). The remaining 7 admission diagnoses had readmission rates between 2.6% and 6.9% and included appendectomy, gastroenteritis, seizure, and 4 respiratory conditions: asthma, bronchiolitis, pneumonia, and upper respiratory tract infection (eg, croup).
Unadjusted readmission rates varied significantly across hospitals for 8 of 10 diagnoses (P < .001 for each) with the highest number of readmissions (upper respiratory infection [P = .07] and ventricular shunt procedures [P = .48] were the exceptions). For these 8 diagnoses, the adjusted readmission rates were 17.0% to 66.0% greater for hospitals with readmission rates that were 1 SD above vs below the mean (Figure 2). The corresponding adjusted rates were 36.8% to 174.8% greater for hospitals 2 SD above vs below the mean. The differences in adjusted rates across hospitals varied by diagnosis. For example, 3.6% (95% CI, 3.5%-3.6%) vs 1.8% (95% CI, 1.8%-1.8%) were the adjusted readmission rates for asthma between hospitals 2 SD above vs below the mean, respectively. The corresponding rates for sickle cell were 24.9% (95% CI, 24.8%-25.1%) vs 9.9% (95% CI, 9.8%-10.0%).
For all-condition admissions, 48.3% of readmissions were for a diagnosis involving the same organ system or a related etiology as the index admission. For each condition-specific admission, 27.3% to 86.2% of readmissions were for a diagnosis involving the same organ system or a related etiology as the index admission (Table 3).
For 9 of 10 index admission diagnoses, the most common readmission diagnosis was the same as the index diagnosis. Sickle cell had the highest percentage of readmissions (79.4%, n = 809) that were for the same diagnosis as the index admission. Appendectomy was an exception to this pattern; its most common readmission diagnosis was postoperative infection (29.7%, n = 213).
In a national sample of children's hospitals, 6.5% of hospitalized children experienced a readmission within 30 days that was likely to have been unplanned. The 10 condition-specific diagnoses with the highest number (ie, prevalence) of readmissions had readmission rates that ranged from 3% to 23%. There were statistically significant variations in readmission rates across hospitals, both for all-condition and most condition-specific admissions. These hospital variations persisted when case mix–adjusted for patients' age and type and number of CCIs.
Further investigation is necessary to understand the reasons for variation in readmission rates across children's hospitals. As found in Medicare readmission studies, the variation may indicate differences in care during the index hospitalization (eg, quality of the discharge instructions28); differences in postdischarge care (eg, access to primary care for a follow-up visit5); or differences in community factors (eg, availability of paid leave for parents to care for recuperating children29). The variation could also be due to area differences in the tendency to hospitalize children or in the availability of hospital beds.30,31 In addition, although we accounted for age and chronic conditions, the variation may have been influenced by patient-specific factors, such as an unpredictable disease course (eg, disease progression near the end of life).32
The threshold at which readmission rate variation across hospitals becomes clinically meaningful has not been determined.33 In this study, the difference in readmission rates across hospitals was relatively small for some of the conditions. Greater readmission rate variation across hospitals was sometimes observed for diseases with higher readmission rates (eg, sickle cell crisis). Higher readmission rates may have occurred because these diseases are, in general, more complex or more difficult to treat.11 Nonetheless, some hospitals and their local health systems had relatively low readmission rates for these diseases, suggesting that the natural course of the diseases might not be the only cause of higher readmission rates. Additional studies are needed to understand how lower readmission rates can be achieved.
The opportunity to reduce pediatric readmissions depends on whether they are preventable. The administrative data used in the present study do not enable a conclusive determination of readmission preventability. However, we sought to exclude readmissions for planned procedures from our analyses, and the assessment of variation in readmission rates involved a comparison of observed vs expected rates for each hospital. Moreover, prior studies report that a considerable proportion of readmissions for many of the diagnoses in our study is preventable. For example, 4 of the 10 most prevalent readmission conditions were for the ambulatory care sensitive conditions (ACSCs) of asthma, gastroenteritis, pneumonia, and seizure.34- 36 It is believed that hospitalizations for ACSCs can be prevented with high-quality outpatient care.34- 36 Some readmissions may have been prevented with improved inpatient care. For instance, in a single-center study, nearly one-fourth of readmissions following admission for seizure were due to medication adverse events found to have occurred during the index admission.37 Nearly all of the readmissions in studies of appendectomy and ventricular shunt operations were due to postoperative complications.38- 40 In studies of individual hospitals, readmission rates for asthma, bronchiolitis, and sickle cell disease have been reduced with improved discharge planning and follow-up care.41- 43
Eliminating disparities in pediatric care may also help reduce pediatric readmissions. As in prior studies,9,11,44- 47 readmission rates were higher for black and Latino children and for children with public insurance. Children with these attributes are less likely to have a usual source of outpatient care48 or a medical home that could help maintain their health after hospital discharge.49 Adults with public insurance are less likely to visit their doctor within 30 days of a hospitalization,50 and they are more likely to use the emergency department for nonemergent care.51 Further investigation is needed to determine whether these patterns of health services utilization are associated with increased readmission risk in children.
This study has several limitations. The data set did not include information on readmissions to a different hospital, which likely led to an undercounting of readmissions. One adult study reports that nearly 20% of heart failure readmissions are to a different hospital.52 Patients at children's hospitals might not experience as many readmissions to a different hospital as adult patients, who often use multiple hospitals for their inpatient care.53 Children, especially those with higher medical complexity, may be more likely to have their inpatient care needs met within a single institution.
Furthermore, although our cohort included nearly one-fifth of all hospitalized children in the United States, it did not include hospitalizations at nonchildren's hospitals, which in general might have a smaller volume of patients and provide inpatient care for children with less severe illnesses. We relied on administrative data to define the study cohort, identify race/ethnicity, identify hospitalizations and readmissions for specific diseases, and exclude planned readmissions; limitations in the ability of administrative data to provide complete clinical information as well as errors and variation in coding practices across institutions could affect our results.54 The data set did not contain information on outpatient health services, which may be associated with the likelihood of readmission.5
Despite these limitations, we found substantial readmission rate variation across children's hospitals that remained after controlling for patient age and chronic conditions. If hospitals with the highest readmission rates in this study were able to achieve the rates of the best performing hospitals, then the overall count of readmissions would be much smaller. It is possible that the distribution of pediatric readmission rates in this study could help hospitals interpret their own performance, identify target conditions for quality improvement, and determine whether an examination of the causes of their readmissions would be useful.
Our study provides a broad look at variation in pediatric readmission rates across a large number of children's hospitals in the United States. Among patients admitted to acute care pediatric hospitals, the rate of unplanned readmissions at 30 days was 6.5%, and there was significant variability in readmission rates across conditions and hospitals.
Corresponding Author: Jay G. Berry, MD, MPH, Division of General Pediatrics, Boston Children's Hospital, Harvard Medical School, 300 Longwood Ave, Boston, MA 02115 (email@example.com).
Author Contributions: Dr Berry had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Berry, Toomey, Zaslavsky, Jha, Nakamura, Feng, Shulman, Chiang, Hall, Schuster.
Acquisition of data: Berry, Nakamura, Shulman, Hall, Schuster.
Analysis and interpretation of data: Berry, Toomey, Zaslavsky, Jha, Nakamura, Klein, Feng, Shulman, Chiang, Kaplan, Hall, Schuster.
Drafting of the manuscript: Berry, Schuster.
Critical revision of the manuscript for important intellectual content: Berry, Toomey, Zaslavsky, Jha, Nakamura, Klein, Feng, Shulman, Chiang, Kaplan, Hall, Schuster.
Statistical analysis: Berry, Zaslavsky, Klein, Feng, Hall.
Obtained funding: Berry, Toomey, Jha, Schuster.
Administrative, technical, or material support: Feng, Shulman, Kaplan.
Study supervision: Schuster.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: Funding was provided by the Agency for Healthcare Research and Quality (U18 HS020513, Principal Investigator: Dr Schuster). Dr Berry was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (K23 HD058092).
Role of the Sponsor: The funders had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.
This article was corrected for errors on February 6, 2013.