There are a number of ways to identify patients who may be at high risk of future emergency admission. They include the following.
- Clinical knowledge, which is the default position in the NHS. There is little research evidence in this area. Although clinicians may be able to identify those currently at high risk, they are less able to identify those who may be at risk in the future (The King’s Fund 2005).
- Threshold modelling, which is rules based, and identifies those at high risk who meet a set of criteria. Case finding has usually been based on threshold modelling such as identifying patients with repeated emergency admissions as a marker of high risk of future admissions. But the utility of this approach has been questioned as, over four to five years, admission rates and bed use among high-risk patients (those over 65 with at least two emergency admissions in one year) fall to the mean rate for older people (38 per cent of admissions in index year, 10 per cent the following year, and 3 per cent at five years)(Roland et al 2005).
Alternative threshold modelling techniques such as identifying patients at high risk through a questionnaire administered by a GP practice have also been tried. The Emergency Admission Risk Likelihood Index (EARLI) is an example of this (Lyon et al 2007). It comprises a six-item questionnaire used to identify patients over 75 who are at high risk of admission. The tool correctly identified more than 50 per cent of those at high or very high risk of emergency admission, and more than 79 per cent of those who were not at risk. However, this method does not take account of changes in health status, unless repeated regularly. - Predictive modelling, in which data are entered into a statistical model in order to calculate the risk of future admission. Predictive modelling is thought to be the best available technique (The King’s Fund 2005).
Several predictive models calculate the risk of future emergency admission for patients with one or more previous admissions; using information about the patient’s age, gender and socio-demographic characteristics. These include the Patients at Risk of Re-Hospitalisation (PARR) and Scottish Patients at Risk of Readmission and Admission (SPARRA) models (see Appendix 1) (Billings et al 2006; NHS Scotland Information Services Division 2006).
Other models, including The King’s Fund’s Combined Predictive Model, the Predicting Emergency Admissions Over the Next Year (PEONY) model, and the Reduce Emergency Admissions Risk model (Prism), use further data from primary care records such as prescribing or diagnosis and medical test results (The King’s Fund 2006; Donnan et al 2008; Welsh Assembly Government Department for Health and Social Services 2007).
Different models have focused on different population groups – for example, those with a prior history of emergency hospital admission (PARR) and those aged over 65 (SPARRA) – whereas the Combined, PEONY and Prism models include all patients registered with a GP or PCT. Testing the various models results in varying degrees of accuracy in predicting future admission (see Appendix 1). Those models that include data from primary care records perform around 10 per cent better than those that rely on secondary care data alone.
In order to improve the performance of predictive models, detailed data on individual patients need to be available.
For further information, please visit:
http://www.kingsfund.org.uk/sites/files/kf/Avoiding-Hospital-Admissions-Sarah-Purdy-December2010.pdf