Family Practice Advance Access originally published online on January 8, 2007
Family Practice 2007 24(2):158-167; doi:10.1093/fampra/cml069
Predicting the likelihood of emergency admission to hospital of older people: development and validation of the Emergency Admission Risk Likelihood Index (EARLI)
a Castlefields Health Centre, Chester Close, Runcorn WA7 2HY
b Centre for Medical Statistics and Health Evaluation, University of Liverpool, Shelley's Cottage, Brownlow Street, Liverpool L69 3GS
c Division of Primary Care, University of Liverpool, Liverpool L69 3GB
d Sefton Primary Care Trust, 5 Curzon Road, Southport PR8 6LW, UK
Correspondence to Dr David Lyon, Castlefields Health Centre, Chester Close, Runcorn WA7 2HY, UK; Email: david.lyon{at}nhs.net.
Received 2 January 2006; Revised 31 October 2006; Accepted 28 November 2006.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Objective. To develop and evaluate an evidence-based tool for predicting the likelihood of emergency admission to hospital of older people aged 75 years and over in the UK.
Methods. Prospective cohort study of older people registered with 17 general practices within Halton Primary Care Trust in the north-west of England. A questionnaire with 20 items was sent to older people aged 
75 years. Items for inclusion in the questionnaire were selected from information gleaned from published literature and a pilot study. The primary outcome measurement was an emergency admission to hospital within 12 months of completing the questionnaire. A logistic regression analysis was carried out to identify those items which predicted emergency admission to hospital. A scoring system was devised to identify those at low, moderate, high and very high risk of admission, using the items identified in the predictive modelling process.
Results. In total, 83% (3032) returned the questionnaire. A simple, six-item tool was developed and validated—the Emergency Admission Risk Likelihood Index (EARLI). The items included in the tool are as follows: do you have heart problems? [odds ratio (OR) 1.40, 95% confidence interval (CI) 1.15–1.72]; do you have leg ulcers? (OR 1.46, 95% CI 1.04–2.04); can you go out of the house without help? (OR 0.60, 95% CI 0.47–0.75); do you have problems with your memory and get confused? (OR 1.46, 95% CI 1.19–1.81); have you been admitted to hospital as an emergency in the last 12 months? (OR 2.16, CI 1.72–2.72); and would you say the general state of your health is good? (OR 0.66, 95% CI 0.53–0.82). The tool had high negative predictive value (>79%) and identified over 50% of those at high or very high risk of emergency admission. A very high score (>20) identified 6% of older people, 55% of whom had an emergency admission in the following 12 months. A low score (
10) identified 74% of the older population of whom 17% were admitted.
Conclusions. In this study, we have developed and validated a simple-to-apply tool for identifying older people in the UK who are at risk of having an emergency admission within the following 12 months. EARLI can be used as a simple triage-screening tool to help identify the most vulnerable older people, either to target interventions and support to reduce demand on hospital services or for inclusion in testing the effectiveness of different preventive interventions.
Keywords. Older people, predicting emergency admissions.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Emergency admissions to hospital have been rising.1 The majority occur in individuals who already have a prior long-term medical condition. Although people aged 65 and over make up only 16% of the population, they occupy almost two-thirds of general and acute hospital beds and account for half of the recent growth in emergency admissions.2 Several studies have shown that many unplanned admissions of older people are avoidable.3–6 However, a review of older people admitted as emergencies has shown that once an acute emergency has occurred, most (86%) do need a further admission.5 These people are typically over 65 with more than one long-term condition and social care needs. If emergency admissions are to be reduced, or at least contained, the incidence of acute episodes needs to be reduced by better management of any underlying chronic diseases, combined with support to maintain independence.
Since the introduction of the NHS Plan7 in the UK through to Our Health, Our Care, Our Say,8 there is a push to shift resources from hospitals to the community setting in order to reduce the disruptive impact of acute unscheduled hospital admissions. Payment by Results9 throws into sharp relief the fact that emergency admissions account for a large proportion of the total health service budget. Practice-based commissioners will very quickly recognize that admission-avoidance schemes need to feature heavily in their redesign plans and for these to work, they must be targeted at the right people. There is clearly a need for a simple tool that predicts future risk of admission. It needs to be flexible enough for it to be applied quickly and easily in the clinical setting, for example when a nurse in primary care carries out a routine disease management check, and yet could be used to identify a high-risk group across a larger population. It needs to be especially appropriate for older people and include medical and social elements. The scoring method needs to be straightforward and provide an immediate result to give any admission-avoidance scheme the chance to be deployed as timely as possible. This has been successfully achieved in other disease areas.10,11
However, the evidence base for such tools in the UK is not robust and there have been calls for further research in this area.12 There have been numerous programmes of case management designed to reduce crises and acute admissions, notably in the UK, Evercare,13 the Castlefields programme14 and the London Older People's Programme15 but the selection criteria have been criticized.16 Community matrons17 and the Evercare model13 are focussed on patients with a ‘high risk’ of a subsequent acute hospital admission rather than being disease specific. In the absence of a tool, the criteria have largely been limited to patients with multiple admissions in the previous 12 months. Unfortunately, this criterion has its limitations. Roland et al.12 showed that, over time, the number of admission within the cohort of patients with a high number of admissions decline (i.e. regress to the mean) even without any active intervention. In addition, Lyratzopoulos et al.18 demonstrated that patients were most frequently readmitted in the early period after discharge, one-third within 28 days and over half within 3 months. An editorial by Morrison19 concluded that there is a clear need for better ways to identify patients at high risk of admission other than simply by their previous admission history.
The dearth of tools for use in the UK20 is in stark contrast to the American literature. American studies have proposed evidence-based methods of identifying high-risk people and have developed predictive screening tools.21–24 Although the generalizability of these tools to the UK health care setting is unknown, the findings were used to inform this study. The closest study is one based on further development of the Probability of Repeated Admission questionnaire tool (Pra Tool), testing validity in three European health care systems, including London.20 Three other studies, based in the US health care setting, although sharing a common end point of non-elective hospital admissions, were based on a selected population group.25–27 One additional European study28 produced a six-point postal questionnaire, the Sherbrooke questionnaire. However, this study's end point was a decrease in functional ability rather than an emergency admission to hospital. Two other studies on predisposing factors for emergency admissions in the elderly did not develop predictive models.5,29
A literature review of the main databases, including Medline, Cinahl and Embase, did not reveal a similar study in the UK for use in primary care. The most notable recent innovation is the Patient At Risk of Re-hospitalization (PARR) case-finding tool30 commissioned by the Department of Health. This is a hospital-based tool which is focussed only on those who have had an admission, using re-hospitalization as a ‘triggering’ event. However, this tool includes retrospective hospital-based inpatient and outpatient information as predictors and requires skill at scrutinizing different data sets. From Scotland, a study,31 ‘Scottish Patients At Risk of Readmission and Admission’, based on the PARR, uses historic data and is based on patients who have had an emergency admission in the previous 3 years. The different applications are compared and contrasted in Table 1.20,25–27,30,31
|
The aim of this study is to develop and validate a new evidence-based tool to identify older people at high risk of an acute admission to hospital in the UK that is applicable both in the clinical setting (primary, secondary and tertiary) and across larger community populations and which gives an instant risk score that can be acted upon immediately.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Setting
The study took place in Halton, a health district covering the towns of Runcorn and Widnes. It is the 21st most deprived health district with high mortality rates ranking fourth highest in the country32 (All age, All cause standardized mortality ratio of 120). The total population size for people aged 75 years of age and over was 7140. The proportion of the population from an ethnic minority is very small, 1.2%. All 17 GP practices within the primary care trust were recruited to participate in the study.
Pilot study
One practice (Castlefields) was selected to act as a pilot to obtain information and test out relevant items for inclusion in the questionnaire and to gain initial estimates for a sample size calculation. Castlefields already had a track record of proactive management of people with long-term conditions33 and read-coded all clinical encounters, including hospital episodes and some social risk factors.
A retrospective analysis of 12 months' unplanned hospital admissions of the Castlefields 75 year olds and over was carried out. The practice database was interrogated for likely predisposing risk factors. Nine predictors appeared to be independently associated with unplanned admissions in older people aged over 75 years: male gender, history of falls in the previous 12 months, ischaemic heart disease, respiratory disease, atrial fibrillation, cancer, having leg ulceration, living alone without help and having difficulty with mobility.
Development of the questionnaire
The predictors found in the pilot study were combined with criteria from the American literature21–24 to produce a simple self-completed questionnaire. The questionnaire (Fig. 1) consisted of 20 ‘yes/no’ questions with no option for ‘don't know’. The questionnaire complied with seven of the 13 factors described by Edwards et al.34 as being conducive to improving return rates. The questionnaires were posted to all those aged 75 years and over. They were accompanied by a personalized letter from each person's GP and a fact sheet as to the purpose of the research. A prepaid envelope was provided along with contact details of the research team. There were no financial incentives.
|
Selection of sample
The Health Agency's database was used to identify potential participants. Of the 7140 people aged 75 years and over, 6951 people registered with 16 practices were available to participate. The pilot practice (Castlefields) was excluded as it had been used to generate the items on the questionnaire. Sample size for a logistic regression was calculated using the formula given in Hseih et al.35 Assuming simple random sampling, a 5% significance level and 90% power, comparisons were made of the sample size necessary to detect a minimum relative odds of 2 for emergency admission in the exposed compared to the unexposed groups, using four main predictor (exposure) variables, namely, ischaemic heart disease, respiratory disease, history of fall and leg ulcers identified in the pilot study, allowing for adjustment for other predictor variables in the model using a variance inflation factor. To account for the added complexity of sampling people from within GP practices (clusters), the sample size was then inflated by a design effect of up to three, based on several potential scenarios and then again to allow for non-response of up to 25%. The final sample size was 4000. A stratified random sample of 150 males and 150 females were sampled from each practice; all males and females were included if fewer than 150 males and/or females were available. The rationale for the design of the study including the sample size calculation is described in more detail in Lancaster et al.36
Administration of questionnaire
Two research assistants double checked the list of people to be included in the sample with appropriate practices. This was to avoid the unfortunate circumstance of sending a questionnaire to someone who had died recently and to ensure that age and postcodes fitted the selection criteria. The GPs were given the option to exclude anyone from the study whom they felt was unsuitable, such as someone in the terminal phase of cancer or suffering from a severe mental illness.
As this was a time-consuming process and required visiting all the practices, the questionnaires were sent out in batches practice by practice over an 8-month period. In addition, two reminder letters were sent out at 3-week intervals. When requested, the research assistants visited nursing homes to help complete the questionnaire. All the returned questionnaires were scanned via an optical reader into an Access database.
Monitoring for admissions or deaths within the time period
All unplanned hospital medical admissions were observed for 12 months following the date of completion of the questionnaire for each participant using Hospital Episodes Statistics (HES) data covering the monitoring period 22 March 2002 to 26 November 2003. Records extracted from HES were linked to the questionnaire records using patient NHS numbers. Where an NHS number was missing, matches were made using surname, forename and date of birth. Records in the study sample were also linked to the Public Health Mortality File, which holds details of all deaths.
Statistical analysis
Univariable associations between the emergency admission outcome and the 20 binary (yes/no) questionnaire items were assessed using logistic regression models. The associations were modelled while controlling for stratification by using fixed GP practice and gender effects. A P-value 0.20 was used to assess statistical significance of the effects for inclusion in a multivariable logistic regression. Multivariable logistic regression models were then fitted using a forward stepwise selection procedure, controlling for fixed practice and gender effects. Model calibration was assessed using the Hosmer–Lemeshow goodness-of-fit statistic.37 Residual and diagnostic plots were examined to check the validity of the fitted model38 and a sensitivity analysis was carried out to investigate the robustness of the model by omitting influential points identified in the plots. The predicted probabilities of emergency admission for each person were calculated from the final model and used to assess the discriminative ability of the model37 by calculating the area under the receiver operating characteristic curve (summarized as AUC). A sensitivity analysis was also carried out to assess the effect of missing data on the final model.
The model was validated in two ways. Firstly, bootstrap validation was used to assess internal bias in model discrimination.39 The bootstrap procedure took 500 random samples with replacement from the original sample, of equal size (n = 3032). The mean value of the AUCs found from fitting 500 models using these samples was subtracted from the AUC of the original sample model to determine the minimum internal bias. Secondly, the model was externally validated using a split-sample approach by first developing a model using data from one half of the district (Widnes) and then applying it to data from the other half (Runcorn).
The final stage of the analysis was to use the model to develop a risk score. The parameter estimates, rounded to whole numbers, from the model were used to allocate a risk score to each person on a simple linear scale based on their questionnaire responses. Ranges of scores that practically divided people into very high, high-, moderate- and low-risk groups for emergency admission were devised and compared, in terms of model performance, to the cut-off that optimized model performance. This was done by determining estimates of both sensitivity (percentage of total emergency admission cases correctly identified by the model at the specified cut-off) and specificity (percentage of total non-admissions correctly identified by the model at this cut-off), together with the corresponding positive predictive value (percentage of emergency admission cases identified by the model that were true admissions) and negative predictive value (percentage of non-admission cases identified by the model that were true non-admissions).
A risk algorithm to use for calculation of scores and associated predicted probabilities of emergency admission for adoption in general practice was also devised.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Of the 6951 eligible participants, using the sampling system described earlier, 3999 people were selected. A total of 350 individuals were removed from this sample because of the following reasons: 187 had died, 77 had an ineligible age or postcode, 71 had left the area and 15 were removed on the advice of GPs.
A total of 3649 questionnaires were sent out over an 8-month period. In all, 302 (8.3%) people declined to participate and 305 (8.4%) did not respond, even after a second reminder. Of all, 3032 people (83%) completed and returned the questionnaire.
A total of 696 individual patients (23%) in the study sample had at least one emergency admission to hospital during the 12 months following completion of the questionnaire. Some patients were admitted more than once during the relevant period—giving a total number of 1050 emergency admissions for study patients. Within the monitoring period, 246 patients in the study sample died, of these 169 had had at least one emergency admission.
The results of the univariable analysis are given in Table 2 and show that 16 of the 20 items were significantly associated with having an emergency admission (P < 0.20) and were selected for the multivariable analysis. Of these 16 predictors, six were included in the final multivariable model (Table 3) and there were no significant interactions between variables. The Hosmer–Lemeshow goodness-of-fit statistic showed that the model fit was acceptable (chi-square = 9.47 on 8 d.f.; P = 0.304). The AUC was 0.695 [95% confidence interval (CI) 0.671–0.719], indicating that the model gave acceptable discrimination.37 In the sensitivity analysis to assess the robustness of the model, an influential group of 61 subjects who had answered favourably on the six items in the fitted model were omitted, but did not substantially affect the parameter estimates.
|
|
In the final model, out of 3032 returned questionnaires, 395 (13%) had to be excluded because of missing data on one or more of the questionnaire items. However, a missing data sensitivity analysis using simple imputation methods indicated that the missing data did not affect the overall results or discriminative ability of the model and so the original model was retained.
The bootstrap validation procedure took 500 random samples of full size (n = 3032), with replacement, from the original sample. The mean value of the AUC from fitting the final model to these 500 samples was 0.690, 0.005 less than the AUC estimated from the original sample, indicating minimum internal bias.40 When bootstrapping was performed with stepwise modelling, the mean AUC among all possible models was 0.67, 0.025 less than the original model. To externally validate the model, it was developed using data from one half of the health district (Widnes) only. The stepwise selection procedures obtained the same predictors on the Widnes data as for the model fitted to the full data set. When this model was tested on the data from the other half of the district (Runcorn), the external estimate of the AUC was 0.669 (95% CI 0.630–0.709).
Risk score
A simplified scoring scale was constructed from the parameter estimates of the final model. The score is calculated starting with a score = 10. The score is then adjusted accordingly if the answer is yes to any of the following questions:
- Do you have heart problems? +3
- Do you have leg ulcers? +4
- Can you go out of the house without help? –5
- Do you have problems with your memory and get confused? +4
- Have you been admitted to hospital as an emergency in the last 12 months? +8
- Would you say the general state of your health is good? –4
- Do you have leg ulcers? +4
If the answer is no or don't know, no change is made to the score.
The lowest possible score is 1 and the highest 29. The starting score of 10 is to avoid the possibility of a minus score.
Using this scale, a binary cut off score of greater than 6 (corresponding to a predicted probability of 0.21) optimized both sensitivity (63.4%, 95% CI 59.5–67.1%) and specificity (63.8%, 95% CI 61.7–65.9%) and gave a negative predictive value of 85% with 222 (15%) identified as false negatives and positive predictive value of 35% with 738 (65%) identified as false positives. A scoring system of 1–10, 11–15, 16–20 and 21–29 was considered to usefully categorize subjects into low-, moderate-, high- and very high risk groups, as illustrated in Table 4. The sensitivity, specificity and positive and negative predictive values obtained by taking each of these cut-offs in turn to identify patients at risk (i.e. low versus moderate/high/very high; low/moderate versus high/very high; low/moderate/high versus very high) are also shown in the table. In each case, the negative predictive values were high (79%) and indicate that the score does well at identifying non-admission cases. The positive predictive values indicate that for those identified as being at high or very high risk just over 50% of the patients identified as being at risk of an emergency admission will in fact be one. An Excel macro is available from the authors for making the score and predicted probability calculations for a given set of questionnaire responses.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
We have taken a systematic approach to identifying important risk factors that have been combined into a simple tool, the Emergency Admission Risk Likelihood Index (EARLI, Fig. 2). The risk factors considered here were chosen from the literature and previous pilot work and were therefore evidence based. The tool has very high specificity and negative predictive values, suggesting that it is very effective at identifying non-admission cases, for whom further monitoring or follow-up would not be needed. However, with low sensitivity, it is less effective at identifying true emergency admissions, and there is therefore less certainty that an identified case will indeed become an emergency admission. In practice, based on the positive predictive values (>50%), this means that twice as many cases would need to be followed up as potential emergency admissions than would actually occur. We consider that this is practicable and would be worth implementing in high-risk populations. For this reason, the decision to choose four rather than three distinct risk groups was to enable practices to decide the prioritization of resources. In this way, the tool can select a small proportion (6%) of the total population at very high risk of emergency admission. Scarce resources, such as the community matrons17 or the Advanced Primary Nurses of Evercare,13 could then first of all be deployed to these people with greatest risk. If resources permitted, the high-risk group (8%) could also then be targeted. The 158 individuals (6%) identified at very high risk in our study accounted for 18% of the total subsequent admissions. If admissions can be reduced by a third in this group alone, then the Public Service Agreement (PSA) target41 for reductions in acute admissions would be achieved.
|
Of the six predictive factors retained in the final model, two related to medical outcomes and one each to general health, social problems, mental health and previous emergency hospitalization. Heart disease is well recognized as a frequent cause for emergency admissions.2,3,21–24 In EARLI ‘Heart Disease’ becomes ‘Heart Problems’ which potentially includes coronary heart disease, heart failure, valve disease and dysrhythmias. The selection of ‘leg ulcers’ into the model is interesting. One possible reason is the potential concurrent deterioration of or increased severity of underlying diseases, such as diabetes. Although diabetes itself was not found to be independently related to an emergency admission, it may be that those with leg ulcers are more vulnerable. The standard treatment for venous leg ulcers is the application of compression bandages.42 However, a large variability in quality of care has been observed.43 "Can you get out of the house without help?" relates to an individual's mobility and independence. It makes sense that someone answering yes is less likely to experience a subsequent emergency admission. Many people with dementia are at an increased risk of being admitted as an emergency and some because there are no suitable alternatives.44 A previous emergency admission was the most important factor as has been reported in many previous publications.21–24 The general health question has also previously been cited as an indicator for future emergency admission.45,46
The major advantages and implications of EARLI are significant. It does not require a retrospective, time-consuming trawl through hospital and primary care databases, which are often incomplete and out of date. It can be applied to large numbers of people easily, providing a simple way to identify a small proportion of high-risk patients for case management, and gives a real-time prediction of the risk of admission in the next 12 months. The tool complements disease-specific approaches as it is not resource intensive and provides a quick and easy way to identify and recruit high-risk people into clinical trials for testing the effectiveness of primary care intervention strategies for reducing emergency admissions. It can be posted to patients for self-completion, or where achieving a high response rate is difficult, particularly in areas with diverse populations, it can be completed while patients are sitting in waiting rooms, queuing for their flu vaccination or during a consultation with any health professional. At six questions long and with a simple scoring system, it would take up only a few moments, a not too daunting prospect for even the most hard-pressed GP or other health worker.
Possible limitations of EARLI are that it does not use precise medical terms and the response therefore depends on the perception of the individual answering the questions. To avoid confusion questions used plain language with simple binary yes/no responses. The tool was designed using a survey of people aged 75 and over, and it needs testing to ensure that it is applicable to a younger age range, in particular those 65–74 years old. The tool was developed in Widnes and Runcorn, which are almost uniformly deprived and with a small ethnic minority. Further work needs to be done therefore to validate the tool in a variety of settings to check its generalizablility. Non-response bias may also be an issue since non-responders might be among the very high risk groups. Anecdotal evidence for non-response in our study gained from telephone follow-up focussed on those with dementia or in nursing homes. We attempted to minimize this bias by contacting all GPs before letters were sent out to check eligibility and by including letters signed by the patients' own GP to give creditability to the study as well as adding meaning and purpose for the participants.
Conclusion
In this study, we have developed and validated a tool for identifying older people in the UK who are at risk of having an emergency admission within the following 12 months. The tool is evidence based, is easy to complete and can be applied cheaply in a variety of clinical and community settings for identifying a cohort of elderly people at high risk of an emergency admission whether they have had a previous emergency admission or not. In particular, it facilitates selection of people for targeted interventions including personal care plans, recently introduced by the Government, which in turn will help achieve the PSA targets and reduce the demand on hospital care.
What is already known about this topic
No evidence-based tool currently exists for identifying older people at risk of having an emergency admission to hospital in the UK. People are currently identified by criteria relating to multiple admissions in the previous 12 months.
What this study adds
An evidence-based, simple-to-use screening tool consisting of six items has been developed and validated to use in identifying older people at increased risk of emergency admission to hospital.
The tool can be used in research to identify individuals for appropriate preventative intervention strategies and in general practice to target resources more effectively.
|
Declaration |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Funding: This study was funded by the NHS Executive North West Research and Development Fund (grant number RDO/28/2/19).
Ethical approval: None.
Conflicts of interest: None.
|
Acknowledgments |
|---|
We would like to acknowledge the participation in the study of Karen Williamson, Rachel Shellien, Tracey Flute, Julia Miller, the GPs and practice managers in all 17 practices in Runcorn and Widnes and thank them for their support.
|
Notes |
|---|
Lyon D, Lancaster GA, Taylor S, Dowrick C and Chellaswamy H. Predicting the likelihood of emergency admission to hospital of older people: development and validation of the Emergency Admission Risk Likelihood Index (EARLI). Family Practice 2007; 24: 158–167.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
1 Tackling NHS Emergency Admissions: Policy Into Practice (1997) London: NHS Confederation and the Royal College of Physicians.
2 Department of Health. National Beds Inquiry (2000) London: The Stationery Office, Department of Health.
3 Bogg J, Lee A, Mair F, et al. Study of Emergency Medical Admissions at Aintree: Summary of Results (1994) Liverpool: University of Liverpool Health and Community Care Research Unit.
4 McDonagh MS, Smith DH, Goddard M. Measuring appropriate use of acute beds: a systematic review of methods and results. Health Policy (2000) 53:157–184.[CrossRef][Web of Science][Medline]
5 Coast J, Ingles A, Frankel S. Alternatives to hospital care: what are they and who should decide. Br Med J (1997) 312:162–166.[Web of Science]
6 Sanderson C, Dixon J. Conditions for which onset or hospital admission is potentially preventable by timely and effective ambulatory care. J Health Serv Res Policy (2000) 5:222–230.[Medline]
7 Department of Health. The NHS Plan: A Plan for Investment, a Plan for Reform (2003) London: The Stationery Office.
8 Department of Health. Our Health, Our Care, Our Say: A New Direction for Community Services (2006) London: The Stationery Office.
9 Department of Health. The NHS in England: the Operating Framework for 2006/2007 (2006) London: The Stationery Office.
10 Blue L, Lang E, McMurray J, et al. Randomised controlled trial of specialist nurse intervention in heart failure. Br Med J (2001) 323:715–718.
11 Ram FSF, Wedizicha JA, Wright J, Greenstone M. Hospital at home for patients with acute exacerbations of chronic obstructive pulmonary disease: systematic review of evidence. Br Med J (2004) 329:315.
12 Roland M, Dusheiko M, Gravelle H, Parker S. Follow up of people aged 65 and over with a history of emergency admissions: analysis of routine admission data. Br Med J (2005) 330:289–292.
13 National Primary Care Research and Development Centre. Evercare Evaluation Interim Report: Implications for Supporting People With Long-Term Conditions (2005) London.
14 Lyon D, Miller J, Pine K. The Castlefields Integrated Care Model: the evidence summarised. J Integr Care (2006) 14:7–12.
15 Jones V, Lowe E, Gilmour K, Walton D, Andrews J. London Older People's Service Development Programme—Final Report (2003) London: Department of Health, Social Services Inspectorate.
16 King's Fund. Predictive Risk Project: Literature Review (2005) London.
17 Department of Health. Supporting People With Long Term Conditions—Liberating the Talents of Nurses Who Care for People With Long Term Conditions (2005) London: Department of Health.
18 Lyratzopoulos G, Havely D, Gemmell I, Good GA. Factors influencing emergency readmission risk in a UK district general hospital: a prospective study. BMC Emerg Med (2005) 5:1.[CrossRef][Medline]
19 Morrison J. Identifying people at high risk of emergency admission [editorial]. Br Med J (2005) 330:266.
20 Wagner JT, Bachmann LM, Boult C, et al. Predicting the risk of hospital admission in older persons—validation of a brief self-administered questionnaire in three European countries. J Am Geriatr Soc (2006) 54:1271–1276.[CrossRef][Web of Science][Medline]
21 Boult C, Dowd B, McCaffrey D, et al. Screening elders for risk of hospital admission. J Am Geriat Soc (1993) 41:811–817.[Web of Science][Medline]
22 Pacala JT, Boult C, Boult L. Predictive validity of a questionnaire that identifies older persons at risk for hospital admission. J Am Geriat Soc (1995) 43(4):347–377.
23 Freedman HD, Beck A, Robertson B, Calonge N. Using a mailed survey to predict hospital admission, within 4.5 months of completing the questionnaire, among patients older than 80. J Am Geriat Soc (1996) 44:689–692.[Web of Science][Medline]
24 Marcantonio ER, McKean S, Goldfinger M, et al. Factors associated with unplanned hospital readmission among patients 65 years of age and older in a Medicare Managed Care Plan. Am J Med (1999) 107:1317.
25 Damush TM, Smith DM, Perkins AJ, et al. Risk factors for non-elective hospitalization in frail and older adult, inner-city outpatients. Gerontologist (2004) 44(1):68–75.
26 Reuben DB, Keeler E, Seeman TE, et al. Development of a method to identify seniors at high risk for high hospital utilization. Med Care (2002) 40:782–793.[CrossRef][Web of Science][Medline]
27 Shelton P, Sager MA, Schraeder C. The Community Assessment Risk Screen (CARS): identifying elderly persons at risk for hospitalization or emergency department visit. Am J Manag Care (2000) 6:925–933.[Web of Science][Medline]
28 Hebert R, Bravo G, Komer-Bitensky N, Voyer L. Predictive validity of a postal questionnaire for screening community-dwelling elderly individuals at risk of functional decline. Age Ageing (1996) 25(78):159–167.
29 Taylor RC, Ford GG. Functional geriatric screening: a critical review of current developments. J R Coll Gen Pract Occas Pap (1987) 30:49–51.
30 Billings J, Mijanovich T, Dixon J, et al. Case Finding Algorithm for Patients at Risk of Re-hospitalisation (2005) York: King's Fund, Health Dialog Analytic Solutions, and New York University.
31 Delivering for Health Information Programme. SPARRA: Scottish Patients at Risk of Readmission and Admission (2006) NHS Scotland.
32 Department Office of Deputy Prime Minister. The English Indices of Deprivation (2004) London: The Stationery Office.
33 Department of Health. Supporting People with Long Term Conditions (2005) London: The Stationery Office.
34 Edwards P, Roberts I, Clarke M, et al. Increasing response rates to postal questionnaires: systematic review. Br Med J (2002) 324:1183–1185.
35 Hseih FY, Bloch DA, Larsen MD. A simple method of sample size calculation for linear and logistic regression. Stat Med (1998) 17:1623–1634.[CrossRef][Web of Science][Medline]
36 Lancaster GA, Chellaswamy H, Taylor S, Lyon D, Dowrick C. Design of a clustered observational study to predict emergency admissions in the elderly: statistical reasoning in clinical practice. J Eval Clin Pract (2007) (in press).
37 Homer DW and Lemeshow S. Applied Logistic Regression (2000) 2nd edn. New York: Wiley.
38 Collett D. Modelling Binary Data (2003) 2nd edn. Boca Raton, FL: Chapman & Hall/CRC.
39 Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med (1996) 15:361–387.[CrossRef][Web of Science][Medline]
40 Steyerberg EW, Harrell FE, Borsboom GJJM, et al. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol (2001) 54:774–781.[CrossRef][Web of Science][Medline]
41 Department of Health. National Standards, Local Action: Health and Social Care Standards and Planning Framework 2005/6-2007/08 (2004) London: The Stationery Office.
42 Compression therapy for leg ulcer. Effective Health Care Bulletin (1997) University of York, NHS Centre for Reviews and Dissemination.
43 Logan R, Thomas S, Harding E, et al. A comparison of sub-bandage pressures produced by experienced and inexperienced bandagers. J Wound Care (1992) 1:23–26.[Medline]
44 Stevens A, Raftery J, Mant J, Simpson S. Health Care Needs Assessment (2004) Volume 2. Oxford: Radcliffe Publishing. 239–305.
45 Bowling A. Just one question: if one question works, why ask several? J Epidemol Community Health (2005) 59:342–354.[CrossRef]
46 Lee Y. The predictive value of self-assessed general, physical and mental health on functional decline and mortality in older adults. J Epidemol Community Health (2000) 54:123–129.[CrossRef]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  P. T. Donnan, D. W. T. Dorward, B. Mutch, and A. D. Morris Development and Validation of a Model for Predicting Emergency Admissions Over the Next Year (PEONY): A UK Historical Cohort Study Arch Intern Med, July 14, 2008; 168(13): 1416 - 1422. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||