Profiling hospitals based on emergency readmission: A multilevel transition modelling approach

https://doi.org/10.1016/j.cmpb.2011.03.003Get rights and content

Abstract

Emergency readmission is seen as an important part of the United Kingdom government policy to improve the quality of care that patients receive. In this context, patients and the public have the right to know how well different health organizations are performing. Most methods for profiling estimate the expected numbers of adverse outcomes (e.g. readmission, mortality) for each organization. A number of statistical concerns have been raised, such as the differences in hospital sizes and the unavailability of relevant data for risk adjustment. Having recognized these statistical concerns, a new framework known as the multilevel transition model is developed. Hospital specific propensities of the first, second and further readmissions are considered to be measures of performance, where these measures are used to define a new performance index. During the period 1997 and 2004, the national (English) hospital episodes statistics dataset comprise more than 5 million patient readmissions. Implementing a multilevel model using the complete population dataset could possibly take weeks to estimate the parameters. To resolve the problem, we extract 1000 random samples from the original data, where each random sample is likely to lead to differing hospital performance measures. For computational efficiency a Grid implementation of the model is developed. Analysing the output from the full 1000 sample, we noticed that 4 out of the 5 worst performing hospitals treating cancer patients were in London. These hospitals are known to be the leading NHS Trusts in England, providing diverse range of services to complex patients, and therefore it is inevitable to expect higher numbers of emergency readmissions.

Introduction

Profiling hospitals on the basis of outcomes (e.g. readmission or mortality) is rapidly becoming a widely used method in healthcare policy and research [1], [2] with a focus on comparing the patients’ outcomes of hospitals. It can be used to generate feedback to healthcare purchasers (or providers) for possible intervention policies to reduce levels of an outcome. Profiling could help patients in selecting a hospital, for instance, patient choice has been at the heart of the United Kingdom (UK) government public service reform agenda to empower patients and reduce inequalities in access to healthcare [3].

Under new government plans the National Health Service (NHS) hospitals will face financial penalties if patients are readmitted as an emergency within 30 days of being discharged [4]. Hospitals in England will be paid for initial treatment but not paid again if a patient is brought back in with a related problem. In this context, it is important to know how well different NHS hospitals are doing, taking into account the process of risk adjustment for possible differences in case mix, e.g. patient severity and socioeconomic differences.

There are a number of negative implications for ranking hospitals through various sources of the media, and this apparently sheds light to those hospitals with the β€˜best’ and β€˜worst’ ranks. The effects of such scrutiny are undoubtedly complex and require careful analysis of results. Therefore, the process of developing a method for generating the results needs more immediate attention [5].

In England emergency readmission is used as an indicator in the performance rating framework, where lower rates of emergency readmission was accepted surrogate for better healthcare [6]. The method profiled hospitals based on the estimates of the standardized percentage of emergency admissions within 28 days of a discharge from hospital (patients aged 16 and over). An emergency admission within 28 days of discharge from hospital (respectively greater than 28 days) is classified as readmitted (respectively non-readmitted), hence regarded as a quality issue, else it is just an unplanned admission. The process takes into account differences between types of patients by their age and gender. This process is known as the risk adjustment step to account for possible differences in patient case mix.

For each hospital, the observed number of readmissions is determined. Given the case mix, the expected number of readmissions for hospital k is estimated (Ε·k). The ratio of the observed to expected readmissions is defined to be the hospitals’ standardized readmission rate (SRk), which is multiplied by the national average rate of readmission (Ο•), providing an indicator value (as a %) known as the β€˜risk-adjusted readmission rate’ (ΞΌkΒ =Β 100Β Γ—Β SRkΒ Γ—Β Ο•), which forms the basis for comparisons between hospitals.

There are a number of issues related to this method. The expected number of readmissions (Ε·k) could be misleading, where the number of emergency admissions to hospital k is multiplied by the national average rate of readmission. In some cases, such as Barts and The London National Health Service (NHS) Trust, who provides diverse range of treatments and services to complex patients, it is inevitable to expect higher numbers of emergency admissions. On the other hand, Trusts treating less severe patients could experience lower numbers of emergency admissions. Therefore, the estimation of Ε·k for Barts and The London NHS Trust would be higher than Trusts treating less severe patients, hence inaccurate estimation of Ε·k. In the UK, the term Trust stands for a single hospital or a set of hospitals in a small region. To be consistent we use the term hospitals throughout this article.

One of the major concerns in the literature is the process of risk adjustment to account for possible differences in patient case mix [7], [8]. Patient characteristics may be important predictors of emergency readmission; however, few variables have been consistently identified [9], [10], [11]. Age and gender were found to be the two predictors that are mostly insignificant to account for variations among patients [12], [13], [14]. Other crucial failures include the inadequacy to account for patient severity and socioeconomic status. Studies by Amarasingham et al. [15] and Lyratzopoulos et al. [16] found that more deprived patients has a higher risk of readmission. From a policy perspective, a validated risk-standardized model to profile hospitals readmission rates is currently unavailable in the literature [10].

The performance rating framework defines readmission for adults as an emergency or unplanned admission to the same hospital within 28 days following discharge. In the literature, the time window for defining readmission varied according to the purpose of the study, generally from 30 to 90 days [9], [10]. Demir [18] reviewed over one hundred articles related to patient readmissions, and found that there is no consensus in defining readmission, and developed a number of methodologies that objectively defined readmission, e.g. 41, 9, 37 and 8 days for chronic obstructive pulmonary disease, stroke, congestive heart failure and hip & thigh patients, respectively, whereas performance ratings framework used 28 days across all clinical conditions. The choice of time windows will inevitably have an effect on the outcome of profiling hospitals.

The majority of the profiling efforts are based on statistical techniques, where the interest lies in the estimation of the expected numbers of an outcome. However, a number of statistical concerns have been raised (see Section hyperlinksec:statisticalIssues2 for further details), such as the differences in hospital sizes, unavailability of relevant data for risk adjustment, and the precision of the hospital-specific estimates, which may all vary greatly. Having recognized these statistical concerns, and some of the limitations based on the current method (described above), we developed a new framework for profiling hospitals based on patient readmissions.

Patient’s previous history of readmissions is known to be the most significant variable in determining the risk of readmission [15], [2], [16]. Based on earlier studies, a transition model was implemented to incorporate individual patient’s history of readmissions to determine the risk of future readmissions [43]. Here, the fixed effect of first, second, third (and so on) readmissions was estimated. Motivated from this work, we assumed that every hospital has its own propensity for first readmission, second readmission, and further readmissions. Hence, the transition model is extended to the multilevel analogue to estimate hospital specific propensities of being readmitted as a high risk group patient. High risk group of readmission is defined to be a readmission within a time window following discharge (greater than the time window is low risk group of readmission), hence a binary outcome. The hospital-specific estimates are interpreted as indicators of hospital performance. For instance, some hospitals may have higher propensity for first readmissions than others. Using these hospital specific transition effects a new performance index is introduced.

In the next section a literature review is compiled together to illustrate the statistical and clinical aspects of hospital profiling, examining past, present and future methods. We discuss the limitations of these methods and provide the grounds for developing a multilevel transition model. Section 3 describes the data; Section 4.1 briefly introduces the transition model; Section 4.2 formulates the multilevel transition model and derives the likelihood function; Section 4.3 introduces a new performance index from the estimated hospital specific transition effects; Section 5 illustrates the application of MTM using the English national dataset; discussion and conclusion are in Section 6.

Section snippets

Statistical issues and methods

The chronology of statistical issues and methods can be stratified into three developments. Early attempts in the 1970s involved the measurement of quality of care via excess variation [21], [22], [23], where these variations included the comparison of resources and expenditures among hospitals. During the mid 1980s the analysis fitted logit models [24], [25], [26], [27] to outcomes using administrative characteristics as predictors. The ranking of hospitals are usually based on the ratio of

Experimental studies

Before we commence any profiling efforts, a number of steps had to be taken with the data, such as clustering the data and objectively defining readmission for the clustered patient sub-groups.

Methodology

Multilevel transition model (MTM) is an extension of the transition model. The next section briefly describes the transition model, where MTM is formulated in Section 4.2.

Results

The study comprised 167 NHS Acute and Foundation Trusts in England. In Section 3 the HES dataset was clustered and three sub-groups were obtained. The dominant disease in cluster 2 was cancer patients, diseases of the circulatory system and diseases of the respiratory system were the main condition for cluster 3 and cluster 4 patients, respectively. Cluster 2, cluster 3 and cluster 4 comprise 337,953, 3,270,325 and 1,133,398 patient readmissions, respectively. Implementing MTM in R (LME4

Discussion and conclusion

The White Paper set by the new UK government [4] clearly highlights the importance of rating clinical departments according to the quality of care they receive, and thus a commitment to make the β€˜NHS more accountable to patients’. Related to this, improving healthcare outcomes is at the heart of this White Paper, focusing on outcomes and the quality standards that deliver them. The objectives are to reduce mortality, readmissions and morbidity. To achieve these goals, in 2010/2011 the

Conflict of interest

None declared.

References (47)

  • A.H. Leyland et al.

    League tables and acute myocardial infarction

    Lancet

    (1998)
  • V. Sundararajan et al.

    New ICD-10 version of the Charlson comorbidity index predicted in-hospital mortality

    Journal of Clinical Epidemiology

    (2004)
  • R. Walter

    The revolving door of hospital re-admissions

    Caring

    (1998)
  • J.C. Luthi et al.

    Is readmission to hospital an indicator of poor process of care for patients with heart failure?

    Quality and Safety of Health Care

    (2004)
  • DoH, Department of Health. http://www.dh.gov.uk/en/Publichealth/Healthinequalities/DH_149 (accessed...
  • DoH, Department of Health, White Paper: Equity and excellence – Liberating the NHS....
  • A. Luthi

    Performance reports on quality – prototypes, problems and prospects

    New England Journal of Medicine

    (1995)
  • Healthcare Commission, Healthcare Commission, 2008 performance ratings....
  • L.I. Iezzoni

    Risk Adjustment for Measuring Health Care Outcomes

    (1994)
  • B. Landon et al.

    Judging hospitals by severity adjusted mortality rates: the case of CABG surgery

    Inquiry

    (1996)
  • J.H. Lichtman et al.

    Predictors of hospital readmission after stroke: a systematic review

    Stroke

    (2010)
  • M.M. Desai et al.

    Statistical models and patient predictors of readmission for acute myocardial infarction: a systematic review

    Circ Cardiovasc Qual Outcomes

    (2009)
  • J.S. Ross et al.

    Statistical models and patient predictors of readmission for heart failure: a systematic review

    Archives in Internal Medicine

    (2008)
  • M. Ashton et al.

    The association between the quality of inpatient care and early readmission

    Annals of Internal Medicine

    (1995)
  • C.F. Ko et al.

    A survey of hospital readmission in elderly patients

    Hong Kong Medical Journal

    (1996)
  • B. Pearson et al.

    Unplanned readmission to hospital: a comparison of the views of general practitioners and hospital staff

    British Geriatrics Society

    (2002)
  • R. Amarasingham et al.

    An automated model to identify heart failure patients at risk for 30-day readmission or death using electronic medical record data

    Medical Care

    (2010)
  • G. Lyratzopoulos et al.

    Factors influencing emergency medical readmission risk in a UK district general hospital: a prospective study

    BioMed Central Emergency Medicine

    (2005)
  • E. Demir, Modelling readmission: defining and developing a framework for profiling hospitals, Unpublished PhD thesis,...
  • H. Goldstein et al.

    League tables and their limitations: statistical issues in comparisons of institutional performance

    Journal of the Royal Statistical Society Series A

    (1996)
  • P. Diggle et al.

    Analysis of Longitudinal Data

    (2002)
  • J. Wennberg et al.

    Small area variations in health care delivery

    Science

    (1973)
  • K. McPherson et al.

    Small area variations in the use of common surgical procedures: an international comparison of New England, England and Norway

    New England Journal of Medicine

    (1982)
  • Cited by (5)

    View full text