A large portion of these hospital stays include readmissions by the same patients, and hospital readmission has been a major quality concern in diabetes care. Yet there are only a few population-based studies on readmission among diabetes patients and its risk factors. Jiang et al.³ reported more than 30% of older diabetes patients were readmitted during the same year in five states in 1999. Readmissions accounted for 55.2% of total hospital stays, and the average total hospital cost for patients with readmissions was 2.5 times higher than those with no readmissions. They also reported a high risk of readmission in ethnic minority groups and variations in readmission rates by insurance type.^4,5 Robbins and colleagues^6–8 observed racial disparities in hospital readmissions in Pennsylvania hospitals between 1994 and 2001, and noted the positive effect of public health clinics in reducing hospital readmissions among low-income diabetes patients.

A gap in the literature on hospital readmission among diabetes patients has been the lack of differentiation of scheduled and unscheduled readmissions.^3–5 Readmissions were often assumed to be unscheduled, but this may not be true, especially for those with existing complex conditions such as diabetes. The need for attention to the type of readmission has been raised,⁷ but only a small number of studies explicitly focused on unscheduled readmissions only.^6,7 No study appears to exist on the reasons for and factors associated with scheduled readmissions and whether they are similar to those associated with unscheduled readmissions.

Another gap in the literature on readmission is that studies have often focused on readmissions within one month of hospital discharge^{6, 7,9} or those of Medicare patients.^9,10 This short observation window is because hospital readmission has been widely studied in terms of its relationship to hospital performance.^7,11 This may be too narrow an approach to evaluating readmission risk from the perspective of patients, however, for whom avoiding a later readmission is as just important as avoiding an early readmission. Studies focusing on patient risk for readmission have often used a 3-month window following hospital discharge,⁵ which we adopted.

The purpose of this study was to examine and compare the demographic, socio-economic, and clinical factors associated with scheduled and unscheduled readmissions in mid-life and older diabetes patients, using a population-based dataset from California. We also examined common reasons for both scheduled and unscheduled readmissions, and estimated the cost as well as the extent of potentially preventable readmissions using the Preventive Quality Indicators (PQIs) of the U.S. Agency for Healthcare Research and Quality (AHRQ).¹²

Our sample included patients 50 years or older with diabetes as the principal or a secondary diagnosis, identified using AHRQ’s Clinical Classification System (CCS 45, diabetes mellitus without complications, and CCS 50, diabetes mellitus with complications),¹⁵ all of whom were admitted to California hospitals at least once between April and September 2006. We defined the first admission as the index hospitalization and, using the encrypted UPI, identified readmissions within three months of the index hospitalization. We counted no more than one readmission for each patient, as Jencks, Williams, and Coleman⁹ did, and focused on the characteristics of the first readmission (e.g., scheduled or unscheduled) in the analysis. Similar to Jiang et al.’s³ approach, we excluded patients with a missing UPI, with the same UPI but an inconsistent age (different by more than 1 year) or sex, or a missing admission or discharge month.

The outcome variables of this study were scheduled and unscheduled readmissions. We used the SCHED variable in the California SID,¹⁶ referring to whether an admission is scheduled at least 24 hours in advance or not. Thus, we defined unscheduled readmissions as readmissions that were not scheduled at least 24 hours in advance and scheduled readmissions as those scheduled at least 24 hours in advance. The SCHED variable is derived from the type of admission (ToA) in the California hospital inpatient data report collected by the California OSPHD, from which the California SID is developed for the HCUP. The California OSHPD¹⁵ provides detailed guidelines for each variable in the report, including the ToA, and also conducts multi-step validation checks and edits.

We examined patient socio-demographic and clinical characteristics potentially associated with readmissions variables, representing the three major categories of Andersen’s Behavioral Model of Health Services Use.¹⁷ This model posits that potential and realized health service use is determined by interactions among predisposing, enabling, and need characteristics of individuals and also the health care systems in the communities where they reside. Predisposing factors are individual characteristics existing prior to illness that determine the likelihood of seeking care and are measured by age, sex, and race. Enabling factors provide the means to utilize health services, and include both individual-level and community-level factors measured by primary payer type, resident location, and neighborhood median income. The California SID do not provide the patient’s exact income level, so we used a proxy variable, median household income for the patient’s ZIP code in 2006. Both perceived and evaluated health status determine the need for health services, measured by comorbidities, prior hospitalization history, and variables related to index hospitalizations. The severity of conditions was measured by the number of chronic conditions in the 18 body systems using the Chronic Conditions Index developed by AHRQ.¹⁸ We counted only one chronic condition per body system, as was counted by Wolff et al.,¹⁹ although some patients might have had more than one.

For data analysis, first, we computed unadjusted rates of scheduled and unscheduled readmissions within three months of the index hospitalization by patient demographic, socioeconomic, and clinical characteristics. Second, logistic regression models were developed to examine the predictors of scheduled and unscheduled hospital readmissions. The reference group for odds ratios was patients with no readmissions in the 0–3 months following their index hospitalizations. Third, we compared common clinical conditions between the scheduled and unscheduled readmission groups using the Clinical Classifications Software (CCS) for the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) developed by the AHRQ.¹⁵ The CCS is a uniform and standardized coding system for aggregating over 14,000 diagnosis codes into smaller, mutually exclusive, clinically meaningful categories. We ran single-level CCS algorithm against our data, which categorized all the diagnoses into about 300 clinical conditions, among which we selected and compared the top 15 conditions for the scheduled and unscheduled readmission groups.

Approximately twenty-six percent (n=32,857) of the patients were readmitted within 0–3 months of their index hospitalizations, and a majority of the readmissions (87.2%) were unscheduled (Table 1). The risks for unscheduled and scheduled readmission varied by patients’ demographic, socioeconomic, and clinical characteristics (Table 1). While adjusting for other factors, patients aged 80 or older were slightly more likely (Odds Ratio [OR]=1.07[95% CI 1.02–1.12], compared to those aged between 50 and 64) to have an unscheduled readmission. Blacks (OR=1.17[1.11–1.23]) and Hispanics (OR=1.10[1.07–1.14]) were also more likely to have an unscheduled readmission than Whites. Patients with public insurance had a higher risk for an unscheduled readmission compared to patients with private insurance (OR for Medicare patients=1.39[1.33–1.46], & OR for Medicaid patients=1.53[1.45–1.62]). Patients residing in urban areas (OR=1.16[1.09–1.22], compared to rural) and those residing in lower income neighborhoods (OR=1.11[1.07–1.16], compared to higher income neighborhoods) had higher risks for an unscheduled readmission. The risk for unscheduled readmission consistently increased as the number of chronic conditions that a patient had increased: patients with 7 or more chronic conditions were nearly three times more likely to have an unscheduled readmission than those with diabetes only (OR=2.93[2.52–3.41]). An unscheduled readmission was more likely to occur in patients who had had one or more hospitalizations in the 3 months preceding the index hospitalization (OR=2.17[2.10–2.25]). Also, the risk for an unscheduled readmission increased when the index hospitalization was an unscheduled admission (OR=1.72[1.64–1.80]), or when it ended with a transfer to another post-acute or long-term institution (OR=1.28[1.24–1.32]). As length of stay rose, the likelihood of unscheduled readmission increased.

Some factors predicting unscheduled readmissions also predicted scheduled readmissions in the same way: compared to men, women were less likely to have a scheduled readmission (OR=0.75[0.70–0.80]), but people with Medicare as the primary payer (OR=1.15[1.04–1.26]) were more likely to do so. The number of chronic conditions and length of stay also positively predicted the odds for both unscheduled and scheduled readmissions. Two other factors predicted both scheduled and unscheduled readmissions, but the directions were opposite: people aged 80 or over (OR=0.71[0.63–0.79], compared to people aged 50–64) and those with unscheduled index hospitalizations (OR=0.58[0.47–0.71], compared to scheduled index hospitalizations) were more likely to have an unscheduled readmission, but less likely to have a scheduled readmission. Unlike these factors, factors that positively predicted an unscheduled readmission but not a scheduled readmission were being an ethnic minority (compared to being White), having Medicaid as the primary payer (compared to private insurance), resident location, median income of the neighborhood, and disposition destination. Being uninsured negatively predicted a scheduled readmission only (OR=0.58[0.53–0.62], compared to people with private insurance).

The most common 15 diagnoses for both unscheduled and scheduled readmissions, covering more than 50% of all of those readmissions, are listed in Table 2. The most frequent condition among diabetes patients with an unscheduled readmission was congestive heart failure (8.8%), followed by diabetes mellitus with complications (7.2%), septicemia (5.8%), and pneumonia (3.9%). Among patients with scheduled readmissions, diabetes mellitus with complications (7.1%) was the third most common diagnosis, following coronary atherosclerosis (11.4%) and complications of device, implant, or graft (7.3%). Only 6 out of 15 conditions—congestive heart failure, diabetes mellitus with complications, complications of device, complications of surgical procedures or medical care, coronary atherosclerosis, and cardiac dysrhythmias—were common to both unscheduled and scheduled readmissions. In addition, 3 out of those 6 conditions—congestive heart failure, complication of device, and coronary atherosclerosis—were in very different ranks among unscheduled and scheduled readmissions.

We found about one fifth of the unscheduled readmissions were potentially preventable, for which nearly 27,500 inpatient days and 72.7 million dollars were spent. Those readmissions might have been avoided if good inpatient and transition care and outpatient follow-up had been provided.^5,6 Considering the large variations among states in readmission rates, lengths of stay, and costs for inpatient care, it would be difficult to determine national estimates based on our data. Nevertheless, the economic burden of caring for people with diabetes, along with patient-level safety concerns related to unscheduled readmissions, reinforces the need for continuing efforts to innovate transition and chronic care models and policies to support such models.

Our study is the first population-based study reporting that the predictors of scheduled and unscheduled hospital readmissions among diabetes patients are not similar. Several patient socio-demographic and economic factors—such as ethnicity, living in an urban area, and living in a lower median-income neighborhood—predicted unscheduled readmissions, but none of them predicted scheduled readmissions. Being uninsured, having an unscheduled index hospitalization, and being among the older elderly (aged 80+) decreased the odds of scheduled readmissions. Kossovky et al.’s²³ study appears to be the only published paper on scheduled as well as unscheduled readmissions. Examining 31-day readmissions using patients discharged from medical units in a single hospital in Geneva over a year, between 1995 and 1996, they reported findings consistent with ours: people aged 75 or older and those admitted through the emergency room for an index hospitalization in their study were less likely to have scheduled readmissions.

This is important because scheduled readmissions may be an indicator of better access to advanced procedures or treatments (e.g., elective chemotherapy, radiotherapy, or inpatient diagnostic procedures) and to better care coordination (i.e., the patient has a relationship with a provider who arranges a scheduled admission, avoiding an emergency admission). If so, the lower scheduled readmissions among the older elderly (80+), people with no insurance, and those with unscheduled index hospitalizations--controlling for all other factors, including comorbidities-- suggests adverse disparity in care for those groups. Therefore more empirical studies on the extent and characteristics of scheduled readmissions and potential disparities in receiving scheduled readmissions are needed.

Limitations of this study should be mentioned. We analyzed publicly available hospital discharge datasets’ so, as in any secondary data analysis study, inaccurate or missing codes may exist, limiting the reliability of the reports of scheduled and unscheduled readmissions (the ToA variable) and also comorbidities. We counted all-cause readmissions and could not determine whether or not and to what extent a readmission was related to an index admission, which would be an important topic for future research. The validity of PQIs as the readmission quality indicators should be evaluated further. Because the dataset does not allow the linking of patient data across years, we examined one year of data. We had data on discharge and admission months, but data on exact dates were not available. No detailed data were available on the type and severity of diabetes, nor on the availability of primary care physicians and discharge planning services. Lastly, we analyzed California data only.

Despite these limitations, this study addressed several gaps in the literature on hospital readmissions of patients with diabetes. Using a recent, large, population-based dataset including mid-life and older diabetes patients with all types of insurance, we reported several socio-economic and clinical factors associated with both scheduled and unscheduled readmissions, and showed the predictors of the two types of readmissions were not parallel. The lower scheduled readmissions in the older and uninsured groups imply a potential lower access to advanced, coordinated follow-up services for these groups. The consistent, prevalent unscheduled readmissions among mid-life and older diabetes patients provide a rationale for continuing efforts to improve the organization and process of acute and chronic care and promote quality and timely primary care, thereby ultimately saving unnecessary social costs and ensuring patient safety.