|Home | About | Journals | Submit | Contact Us | Français|
Risk-standardized hospital readmission rates are used as publicly reported measures reflecting quality of care. Valid risk-standardized models adjust for differences in patient-level factors across hospitals. We conducted a systematic review of peer-reviewed literature to identify models that compare hospital-level post-stroke readmission rates, evaluate patient-level risk-scores predicting readmission, or describe patient and process-of-care predictors of readmission after stroke.
Relevant English-language studies published from January 1989–July 2010 were identified using MEDLINE, PubMed, Scopus, PsycINFO, and all Ovid Evidence-Based Medicine Reviews. Authors of eligible publications reported readmission within one year after stroke hospitalization and identified one or more predictors of readmission in risk-adjusted statistical models. Publications were excluded if they lacked primary data or quantitative outcomes, reported only composite outcomes, or had fewer than 100 patients.
Of 374 identified publications, 16 met the inclusion criteria for this review. No model was specifically designed to compare risk-adjusted readmission rates at the hospital-level or calculate scores predicting a patient’s risk of readmission. The studies providing multivariable models of patient-level and/or process-of-care factors associated with readmission varied in stroke definitions, data sources, outcomes (all-cause and/or stroke-related readmission), durations of follow-up, and model covariates. Few characteristics were consistently associated with readmission.
This review identified no risk-standardized models for comparing hospital readmission performance or predicting readmission risk after stroke. Patient and system-level factors associated with readmission were inconsistent across studies. The current literature provides little guidance for the development of risk-standardized models suitable for the public reporting of hospital-level stroke readmission performance.
Readmission after hospital discharge is used as an indicator of the quality and efficiency of hospital-level care for several clinical conditions.1–3 Risk-standardized models for readmission are available for acute myocardial infarction (AMI) and heart failure (HF).4 In 2009, the Centers for Medicare & Medicaid Services (CMS) began publicly reporting hospital-level risk-standardized 30-day readmission rates for these conditions to both assist healthcare consumers in their care decisions and to drive quality improvement nationally.4–6 Risk-standardized models use hierarchical analyses to generate standardized readmission ratios (predicted over expected), and then standardize the estimate by multiplying it by the national average. High readmission rates may indicate unresolved problems at initial discharge, the quality of immediate post-hospital care, a more chronically ill population, or combinations of these factors. High readmission rates are also associated with a substantial economic burden on the healthcare system and may represent opportunities to reduce avoidable costs. Reduction of readmission rates is an important US healthcare reform goal.1, 7
Stroke is an important condition to target for efforts to reduce hospital readmission rates. It affects an estimated 795,000 people each year in the US,8 and there are more than 4.4 million stroke survivors in this country.9 Stroke is one of the 10 highest contributors to Medicare costs,10 and among the elderly, stroke and transient ischemic attacks (TIAs) are a leading cause of hospitalization.11, 12 For stroke survivors, significant disability, preventable complications, and discharge to settings with substantial requirements for ongoing care are common. Recurrent events, which occur in 185,000 stroke survivors in the US annually, are associated with higher mortality rates, greater levels of disability, and increased costs as compared with initial stroke events.13 Given the clinical and policy importance of stroke, identifying factors that contribute to readmission risk is important to assist clinicians and health care institutions in the care of stroke patients, as well as to identify opportunities to reduce avoidable hospitalizations.
Meaningful comparisons of hospital-level readmission rates to assess quality of care require a valid method that appropriately risk-adjusts for differences in patient characteristics. In order to inform the development of such models, we conducted a systematic review of peer-reviewed literature to (1) identify and evaluate existing statistical models developed to compare hospital-level post-stroke readmission rates, (2) identify and evaluate patient-level statistical models or risk scores predicting readmission, and (3) identify individual patient-level and process-of-care predictors of readmission after stroke hospitalization and evaluate the consistency of these predictors across studies.
We identified relevant peer-reviewed publications by searching the following databases: (1) Ovid MEDLINE (January 1989–July 15, 2010); (2) PubMed (January 1989–July 15, 2010); (3) Scopus, an Elsevier abstract and citation database (January 1989–July, 15 2010); (4) Ovid PsycINFO (January 1989–July 15, 2010); and (5) all Ovid Evidence-Based Medicine Reviews, including ACP Journal Club (January 1991–July 15, 2010) and the Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Database of Abstracts of Reviews of Effectiveness, Cochrane Methodology Register, Health Technology Assessment, and National Health Service Economic Evaluation Database (3rd Quarter 2010). The initial search used the Medical Subject Heading (MeSH) term stroke (exploded). A search was then performed using the MeSH term risk (exploded) and the terms model*, predict*, use*, util*, and risk* (using * for truncation; terms combined using an OR statement). This was followed by a search using the terms patient readmission, readmission, and rehosp* (terms combined using an OR statement). Finally, the stroke, risk/model/prediction, and readmission terms were combined, limiting the results to English-language publications and human cohorts. This search was performed independently and replicated by two authors (EL-L, SS); the search identified 374 publications.
Inclusion and exclusion criteria were defined a priori and then applied to the identified publications. Eligible studies were those that included stroke patients who were hospitalized and/or enrolled in stroke registries and focused on one or more predictors of readmission in risk-adjusted statistical models. Studies using data collected from a randomized clinical trial to examine the risk-adjusted effect of participants’ characteristics on readmission (independent of the effect of the intervention) were also included. Studies without primary data (e.g., reviews, letters, editorials, and methods papers), abstracts, studies reporting results from a case series or case report, dissertations, pediatric studies, studies published prior to 1989, and studies including fewer than 100 patients were excluded. Studies that lacked quantitative outcomes, did not report readmission outcomes within 1 year, only reported readmission as part of a composite outcome, were limited to TIA and/or hemorrhagic stroke patients (unless these patients were part of a mixed cohort that included ischemic stroke patients), or focused on patient disease subgroups (e.g., diabetes) were also excluded.
Five authors (SB, KB, EL-L, JL, SS) independently reviewed the titles and abstracts of the 374 identified studies, excluding 329 based on inclusion and exclusion criteria (Figure 1). Two authors (ELL, JL) verified these exclusions. The full-text publications of the remaining 45 potentially eligible studies were reviewed in detail. Of these, 29 were excluded based on the predefined criteria.
Data were abstracted from the remaining 16 studies using a standardized form that included study objective, design, time period, sample size and characteristics, and location; definition of population/cohort, stroke, and outcome measure; data sources used to ascertain patient characteristics and follow-up readmissions; and the presence and statistical significance of candidate variables evaluated as either primary predictors or covariates in multivariable models for all-cause readmission, stroke-specific readmission, or stroke-related readmission (composite readmission outcome that included stroke). Three authors (SJ, EL-L, EW) abstracted the data and 2 authors (SB, JL) reviewed the results. Disagreements in assessment and data extraction were resolved by consensus.
Among the 16 studies meeting review criteria,14–29 none provided models developed for the purpose of comparing hospital-level readmission rates (Aim 1) and none reported patient-level statistical models or risk scores predicting readmission (Aim 2). All 16 studies identified patient-level and/or process-of-care predictors of post-stroke readmission (Aim 3). Characteristics of these studies are presented in Table 1. The majority of studies were conducted in populations within the United States (10 of the 16 studies),14–16, 19, 21–24, 26, 27 with the remaining cohorts from Canada,20 Australia,18, 25 Singapore,28 and Taiwan.17, 29 Thirteen studies used administrative data,14–16, 18–20, 22–24, 26–29 five used medical record abstraction,14, 15, 21, 22, 25 and three used data from patient interviews.15, 17, 21
Study samples varied from 228 to 366,551 patients, and methods for patient selection varied across studies, with some restricted to those with ischemic stroke14, 15, 23, 24, 26–28 and others including a broader cohort of stroke patients.16–22, 25, 29 Most identified the index stroke based on ICD-9 or diagnostic related group (DRG) codes;16–24, 26–29 however, the specific codes varied across studies, as did the inclusion of codes from either primary or secondary diagnoses. Several studies further limited their populations to patients who were veterans,16, 19 65 years or older,16, 21, 23, 26, 27 free of stroke for at least one year prior to the index event,20, 22, 28 had at least one limitation in either activities of daily living (ADL) or instrumental ADLs,17 had at least three health care encounters during the follow-up period,19 or were cared for by physicians classified as a general internist, family physician, or hospitalist.24
The selection of outcomes differed across studies. Fourteen studies reported all-cause readmission,14–17, 19–21, 23–29 with five of these also reporting stroke-related outcomes; 19, 23, 26–28 two studies reported stroke-related readmission as the sole outcome.18, 22 Duration of follow-up was either within 30 days (Table 2)16, 17, 23–27 or 1 year (Table 3),14, 15, 18–22, 28, 29 with one study reporting models for both time periods.26 Readmission rates at both time points were high and varied across studies: 30-day all-cause readmission ranged from 6.5% to 24.3%, 1-year all-cause readmission from 30.0% to 62.2%, 30-day stroke-related readmission from 7.4% to 9.4%, and 1-year stroke-related readmission from 10.5% to 31.1%.
Analytic methods used to examine predictors of readmission included logistic regression,14, 15, 17–20, 29 proportional hazards regression,16, 23, 26–28 generalized estimating equations,24 truncated negative binomial regression,22 log-linear analysis,25 and instrumental variables estimation.21 Only four of the 13 studies using data from multiple sites reported adjusting analyses for site or patient clustering within site.23, 24, 26, 27 Accounting for death within the study period varied by analytic method. Most studies utilizing proportional hazards models reported censoring for events,16, 23, 26, 27 whereas studies utilizing other types of analytic models either excluded patients who died prior to the interview/analysis date (in either primary or secondary analyses),17, 19, 29 adjusted for death,19 or did not specify in the methods.18, 21, 24, 25 Most studies reported excluding patients who died during the index hospitalization,14–16, 18–21, 23, 25–29 with only one including in-hospital death as a covariate for risk-adjustment 22 and one not reporting the information.24 None presented measures of model performance or power calculations to determine whether the study had a sample size adequate to detect the associations of interest.
There was little consistency in the variables presented in analytic models across studies (Tables 2 and and3).3). Of the 15 studies that clearly stated the covariates used in their models, commonly included demographic variables were age,14, 15, 17–20, 22–29 sex,14, 15, 17–20, 22–29 and race.14–17, 19, 22–24, 26–28 Nearly all studies included a stroke severity scale14–16 or individual variables related to stroke severity.14–17, 19, 22, 23, 25–27, 29 Of these severity indicators, length of index hospitalization14–17, 19, 22, 26, 29 and discharge location/need for nursing care14–17, 22, 26, 27 were the most common. Most studies included one or more cardiovascular-related premorbid or comorbid conditions either as individual variables14–20, 22–24, 26–28 or as part of a comorbidity index,19, 23, 29 with many specifically including diabetes.15, 16, 18, 20, 22–24, 26–28 The definition of these and other model variables and the reporting of the magnitude and direction of the association were inconsistent, with eight studies reporting results for only the primary variable(s) of interest.20, 21, 23–28 Variables associated with higher readmission rates in at least two studies included advanced age,18, 19, 29 longer hospital stay,16, 22, 29 poorer post-stroke physical functioning,15, 17 and an increased number of prior hospitalizations.16, 19 Other variables associated with readmission were insurance type,19, 26 stroke type,19, 22 incident stroke,17, 19 discharge destination,17, 21, 26 diabetes,18, 22, 28 physician specialty,25, 27, 29 and hospital characteristics/certification status.19, 22, 23, 29 Table 4 provides detailed information on variables significantly associated with readmission identified in Tables 2 and and33.
This systematic review did not identify any studies reporting on statistical models for comparing or predicting post-stroke readmission rates among hospitals or scores predicting the risk of readmission. Only 16 studies were identified that presented multivariable models of individual patient-level and/or process-of-care factors associated with readmission. Among these, there was considerable variability in case definitions, analytic approaches (including censoring for patient deaths), outcome definitions, follow-up periods, and model covariates.
Identifying patient-level characteristics is important to assist clinicians caring for stroke patients as well as for developing models that appropriately risk-adjust for case-mix differences to compare outcomes across hospitals. There was considerable heterogeneity among studies in the selection of patient-level covariates considered to be associated with hospital readmission. The small subset of studies reporting significance levels for patient-level covariates makes drawing conclusions about important predictors of readmission for stroke survivors difficult. In addition, these studies provide limited detail to ascertain whether clinical conditions were present at admission or may have occurred as complications during the hospitalization. Differentiating pre-existing conditions from complications is essential in order to avoid risk adjusting for conditions that may be a result of hospital care. A number of the studies further restricted their cohorts by age, physical limitations, or prior events. Such narrowing of inclusion criteria limit the generalizability of results to larger, more representative stroke populations.
Minimizing unnecessary readmissions is an integral component of quality improvement efforts, as readmissions have adverse consequences for both the patient and the health care system. One study reported that approximately 90% of readmissions within 30 days were likely to be unplanned and that these accounted for $17.4 billion in Medicare expenditures.30 The average hospital stay for rehospitalized patients was 13.2% longer than the stay for patients in the same DRG who had not been hospitalized within the prior 6 months.30 Readmissions may reflect a number of factors affecting the transition from inpatient to outpatient care including potentially avoidable complications that interfere with recovery, poor discharge planning or execution of care, a new health problem, deterioration of a chronically ill patient, noncompliance or misunderstanding of discharge instructions, adverse drug reactions, family dynamics or living arrangements, or socioeconomic factors preventing access to appropriate care.1 Short-term outcomes may reflect poor transitions of care at discharge; for example, patients may not be able to care for themselves at home, know whom to call with questions or if symptoms worsen, or understand their immediate health care needs.1 Caregivers may also be inadequately prepared to care for the patient, and community clinicians may not be sufficiently organized or have the necessary resources to deal with a patient’s needs. Bed availability may also impact readmission rates as some suggest increased availability may create a demand for readmission.31 Long-term outcomes may be affected by inpatient care as well as outpatient clinical management and secondary prevention efforts. Additional work is needed to better understand and address the complex potential causes for early post-stroke hospital readmission.
An interdisciplinary expert writing group assembled by the American Heart Association identified seven attributes of risk-adjustment models for use in public reporting of healthcare providers’ outcomes.32 These include a clear and reproducible definition of the patient sample, clinical coherence of variables selected for models, sufficiently high-quality and timely data, designation of a referent time to differentiate complications from covariates, selection of appropriate outcomes using a standardized period of outcome assessment, and analyses that account for multi-level organization data. Our comprehensive review did not identify any studies that met these criteria. In the absence of any studies that provide validated, risk-standardized models for comparing hospital stroke readmission rates, our systematic review represents one of the best approaches to identify possible components that should be included in the development of such a model. These include age, physical functioning, prior hospitalizations, and comorbid conditions (e.g., diabetes, prior stroke), but further research is warranted given the heterogeneity of case definitions and associations with readmission outcomes across studies.
This study has several limitations. We conducted a comprehensive review of the peer-reviewed literature, but did not include studies from the “gray literature” such as agency reports, doctoral dissertations, or conference proceedings. Additionally, only English language studies were included. Due to the relatively small number of identified peer-reviewed papers, a well designed study to determine patient-level predictors of hospital readmission in a representative stroke population could contribute substantively to our understanding of factors associated with readmission, as well as inform the development of statistical models that can be used to compare hospital-level performance. Stroke readmissions, occurring in nearly one quarter of stroke patients annually, create significant burden on the healthcare system and are an important target for quality improvement efforts. The lack of a validated risk-standardized statistical model that accounts for differences in patient characteristics represents an important research gap that needs to be addressed to help direct quality improvement efforts, including the development of measures to compare the quality and efficiency of hospital-level care for stroke patients.
The authors thank Dr. Harlan Krumholz for his advice in framing the literature review and suggestions on the manuscript.
Funding sources: This project was supported by grant numbers R01 NS043322-01 and R01 NS043322 (ARRA) from the National Institute for Neurological Disorders and Stroke, and by contract number HHSM-500-2008-0025I-MIDS Task Order T0001 with the Centers for Medicare & Medicaid Services. The analyses upon which this publication is based were performed under Contract Number HHSM-500-2008-0025I Task Order T0001, entitled “Measure & Instrument Development and Support (MIDS)-Development and Re-evaluation of the CMS Hospital Outcomes and Efficiency Measures,” funded by the Centers for Medicare & Medicaid Services, an agency of the U.S. Department of Health and Human Services. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government. The authors assume full responsibility for the accuracy and completeness of the ideas presented.
Disclosures: The authors report no conflicts of interest.