Racial and Ethnic Differences in 30-Day Hospital Readmissions Among US Adults With Diabetes

Key Points Question Are there differences in hospital readmission rates among racial and ethnic minorities with diabetes, and if so, what are the individual-level and hospital-level factors associated with these differences? Findings In this cohort study of 272 758 adults with diabetes, black patients had a significantly higher risk of unplanned all-cause 30-day hospital readmission than members of other racial/ethnic groups. This difference was most pronounced among lower-income patients and patients hospitalized in nonprofit, academic, or large hospitals. Meaning The observed differences in unplanned hospital readmission rates between black patients with diabetes and members of all other racial/ethnic groups reinforce the importance of identifying and addressing gaps in care that may be contributing to racial/ethnic disparities in health care quality and outcomes.


Introduction
Diabetes is among the most prevalent and costly chronic diseases in the United States, costing $327 billion annually and accounting for 1 of every 7 dollars spent on health care in the United States. 1 Diabetes disproportionately affects racial/ethnic minorities, who also have worse diabetes-related health outcomes (ie, microvascular and macrovascular outcomes) and mortality. 2,3[11] Nevertheless, none of these studies assessed the association with race separately from other important risk factors for readmission, such as age, comorbidity, economic status, reasons for hospitalization, and site of care.Hence, our understanding of individual and hospital factors contributing to these disparities remains uncertain and is an essential first step in developing policy and practice interventions aimed at improving health outcomes for all people with diabetes.
There are several, likely interrelated, reasons why members of a racial/ethnic minority may have higher rates of unplanned hospital readmission, and it has been difficult to differentiate among them. 5,6,12Specifically, readmission risk can be increased by person-level factors, such as age, sex, and comorbidity burden. 13Hospitalization-related factors, specifically the reason for index admission (ie, initial or first admission), length of stay, or history of recurrent hospitalizations, can increase readmission risk. 14Socioeconomic factors, including income, education level, insurance status, and availability of a social support network and community resources, also play a role. 14,15Finally, factors related to the quality of care received at the discharging hospital (ie, the site where the person received their care) can affect subsequent readmission risk. 16Improving care quality by reducing readmissions requires a nuanced understanding of the racial/ethnic disparities that may drive differences in readmission risk, including the interplay of these contributing variables.Using a large data set of commercially insured and Medicare Advantage beneficiaries across the United States, we examined the rates of unplanned 30-day hospital readmissions among adults with diabetes, focusing specifically on 4 ethnic/racial groups derived from US Census categories: white individuals, black individuals, Hispanic individuals, and Asian individuals.We further evaluated factors associated with readmissions across these different groups, focusing specifically on the 5 following prespecified categories: demographic characteristics, clinical factors, economic status, aspects of the index hospitalization, and characteristics of the hospital where the index hospitalization took place.Using a nationwide administrative claims database for this work allowed us to address these questions in ways not possible with publicly available hospital and US Census data, including isolating within-hospital differences and accounting for the patient-level clinical and socioeconomic factors that are the focus of this study.

Data Source
We conducted retrospective analysis of data from the OptumLabs Data Warehouse, an administrative claims database of private and Medicare Advantage enrollees across the United States, with the greatest representation in the South. 17,18

Study Participants
We included adults with an established diagnosis of diabetes who experienced an index hospitalization (ie, first-time hospital admission not preceded by another hospitalization within a 30-day period) for any cause between January

Primary Outcome
We also excluded readmissions within 1 day of index admission discharge, as these may have represented continuation of the index admission.

Statistical Analysis
We summarized patient and admission characteristics overall and by race/ethnicity for all index admissions and readmissions, testing for differences in characteristics using χ 2 tests of independence (null hypothesis, no difference among racial/ethnic groups).For each racial/ethnic group, we identified the 10 most common causes for admission and readmission, reporting the number and percentage for each.
We then assessed the association of patient risk factors with readmission risk using mixedeffects logistic regression models.To assess the relative association of clinical, admission, economic, and hospital factors to differences in risk, we estimated 5 nested models.Starting with an empty model that included only indicators for race/ethnicity, patient demographic characteristics (ie, age and sex), calendar year, and random effect for hospital (model 0), we added 4 models, sequentially: ( hospital ownership status, teaching status, urban category status, sole community hospital status, and bed number.For each model, we reported the readmission odds ratio (OR) for each racial/ethnic group.By examining how the association of race with readmission changes across models, we could evaluate the importance of each set of factors in explaining racial differences.
Finally, to estimate within-hospital disparity effects, we used the Peters-Belsen approach, which was developed for understanding wage disparities and has been used previously to assess health care disparities. 27,28Specifically, we estimated a logistic regression model corresponding to model 2 (without calendar year) using only white patients and used this model to estimate an expected probability of readmission for all patients in the cohort.For each hospital, we used these expected probabilities to calculate the observed-to-expected (OE) ratio of readmission for each racial/ethnic group at each hospital.We used model 2 rather than model 4 so that we could evaluate differences in the OE ratio across subgroups of patients according to income, hospital characteristics, and calendar year.After using the model to calculate an expected readmission risk for all patients, the observed and expected values were summed over the subgroups of interest to calculate OE ratio for that group.These were reported with 95% CIs based on the delta method for income category, year, and hospital characteristics.
We used multiple imputation with 20 imputations to account for missing income data (the only variable with missing values; missing in 18 498 of 272 758 patients [6.8%] or 31 819 of 467 324 [6.8%] of admissions) in all models.We imputed this under the assumption that data were missing at random, using patient race/ethnicity, readmission status, and clinical and demographic factors in

Study Population
We identified 272 758 adults with diabetes who experienced 467 324 index hospitalizations during the 6-year study period (

Association of Financial and Hospital Factors With Readmission Risk Among Different Racial/Ethnic Groups
The OE readmission rate ratios for each racial/ethnic group are reported in

Discussion
A wide range of factors contribute to the heightened risk of readmissions among adults with diabetes.In this large cohort of commercially insured adults and Medicare Advantage beneficiaries with diabetes across the United States, patient-level characteristics, clinical characteristics, patient economic status, factors surrounding their index hospitalization, and the characteristics of hospitals where they received care all affected the odds of unplanned all-cause 30-day readmission.
Importantly, even after adjustment for these categories of readmission risk factors, black patients were significantly more likely to be readmitted than members of other racial/ethnic groups.This statistically significant increase in readmission risk could not be explained by other demographic factors, comorbidities, income, reason for index hospitalization, or place where they received care, although black patients did also have a higher prevalence of all these readmission risk factors than members of other racial/ethnic groups.Black patients with diabetes had a higher than expected readmission rate irrespective of where they received care, with the exception of very small hospitals b Division indicates in or adjacent to area with at least 2.5 million people; metropolitan, area with at least 50 000 and less than 2.5 million people; micropolitan, area with at least 10 000 and less than 50 000 people; and rural, area with less than 10 000 people.
) clinical factors, comprising individual comorbidities and insulin use; (2) index hospitalization factors, comprising length of stay, prior hospitalization, planned admission status, and discharge year; (3) economic factors, comprising annual household income; and (4) hospital factors, comprising the imputation model.P < .05 was considered statistically significant, and all tests were 2-tailed.All analyses were conducted using SAS statistical software version 9.3 (SAS Institute) and Stata version 13.1 (StataCorp).

JAMA Network Open | Diabetes and Endocrinology Racial
Study data were accessed using techniques and Ethnic Differences in Readmissions Among Patients With Diabetes compliant with the Health Insurance Portability and Accountability Act of 1996, and because this study involved analysis of preexisting, deidentified data, it was exempt from institutional review board approval.Patient consent was not obtained because all presented data were anonymized prior to data set creation.This retrospective cohort study was performed following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
JAMA Network Open.2019;2(10):e1913249. doi:10.1001/jamanetworkopen.2019.13249(Reprinted) October 11, 2019 2/16 Downloaded From: https://jamanetwork.com/ on 09/18/2023 1, 2009, and December 31, 2014, were discharged alive, and had at least 12 months of enrollment in private insurance or Medicare Advantage prior to index admission and at least 31 days after the index discharge date (ie, date discharged from hospital after the index admission).Hospitalizations for the principal discharge diagnoses of medical treatment of cancer, psychiatric disease, and pregnancy were excluded in accordance with hospitalwide readmission measure methods.
19,20Principal diagnoses for index hospitalizations were identified using the primary diagnosis International Classification of Diseases, Ninth Revision, Clinical 21dification (ICD-9-CM) codes from the hospitalization claims and grouped using the Agency for Healthcare Research and Quality Comorbidity and Clinical Classifications Software.21Weexcludedadmissionswith lengths of stay longer than 365 days.Diagnosis of diabetes was established by applying National Committee for Quality AssuranceHealthcare Effectiveness Data and Information Set claims-computable criteria 22 to data from the 12 months preceding the index hospitalization; patients with only gestational diabetes (ICD-9-CM codes, 648.0 and 648.8) were not included.Race/ethnicity was ascertained from OptumLabs Data Warehouse enrollment files, and individuals for whom race/ethnicity was missing or unknown (10 970 of 283 729 patients [3.9%]) were excluded.

Table 1 .
Baseline Characteristics of the Study Population at Index Hospital Admission (continued) [5.5%]) and among Asian patients, sepsis (655 [6.6%]).Congestive heart failure was a common cause of hospitalization among all groups, ranking first among black patients (6109 [6.8%]), second among white patients (16 926 [5.5%]) and Hispanic patients (1856 [4.9%]), and third among Asian patients (470 [4.7%]).Relative Contributions of the Readmission Risk Factor CategoriesResults of the nested models are illustrated in the Figure, examining the changing association of race with readmission risk as each set of factors is sequentially added to the analysis.In the first model (model 0), only patient age, sex, and race/ethnicity (the demographic characteristics category) were considered.In this model, black patients had a higher risk of readmission compared with white patients (OR, 1.15; 95% CI, 1.12-1.18;P < .001),while Asian patients had a lower risk (OR, 0.89; 95% CI, 0.83-0.96;P < .001).The lower readmission risk among Asian patients was no longer observed when disease-level factors were added (model 1) (OR, 0.97; 95% CI, 0.90-1.05;P < .001),indicating that their lower risk can be explained by lower clinical complexity.The higher risk among black patients persisted (OR, 1.10; 95% CI, 1.07-1.13;P < .001).Model 2 incorporated data on the index hospitalization, and this only slightly attenuated the increased readmission risk among black patients (OR, 1.07; 95% CI, 1.04-1.10;P < .001).Subsequent incorporation of annual household income, an indicator of financial factors, in model 3 brought the risk among black patients down to an OR of 1.06 (95% CI, 1.03-1.09;P < .001)compared with white patients.Finally, consideration of hospital characteristics of the discharging facility (model 4) did not change the increase in readmission risk among black patients (OR 1.05; 95% CI, 1.02-1.08;P < .002),and their higher risk of readmission compared with white patients persisted.

Table 3 .
OE Readmission Rate Ratio for Each Racial/Ethnic Group in Different Economic Circumstances and Hospital Settings Abbreviations: NA, not applicable; OE, observed-to-expected.a Nonteaching indicates no residents training and the hospital does not belong to or partner with a university; residency, hospital has formal resident training programs but does not belong to a university; academic, hospital belongs to a university and trains residents.