Intended for healthcare professionals

Analysis

Reducing emergency admissions through community based interventions

BMJ 2016; 352 doi: https://doi.org/10.1136/bmj.h6817 (Published 28 January 2016) Cite this as: BMJ 2016;352:h6817
  1. Emma Wallace, general practice lecturer1,
  2. Susan M Smith, professor of general practice1,
  3. Tom Fahey, professor of general practice1,
  4. Martin Roland, professor of health services research2
  1. 1HRB Centre for Primary Care Research, Department of General Practice, Royal College of Surgeons in Ireland, 123 St Stephens Green, Dublin 2, Ireland
  2. 2Cambridge Centre for Health Services Research, University of Cambridge, UK
  1. Correspondence to: E Wallace emmawallace{at}rcsi.ie
  • Accepted 30 November 2015

Evidence for current interventions is limited. Emma Wallace and colleagues discuss the alternatives

Reducing emergency admissions to hospital, both as a measure of care quality and to contain spiralling healthcare expenditure, is gathering interest internationally. Emergency admissions in the United Kingdom rose by 47% from 1998 to 2013, from 3.6 million to 5.3 million, with only a 10% increase in population over this period.1 These admissions are expensive; in 2012 they cost the NHS £12.5bn (€16.8bn; $18.3bn).1

Emergency admission is used as a performance measure for healthcare systems. One of the quality measures for accountable care organisations under the US Affordable Care Act2 is to reduce emergency admissions for three chronic medical conditions: chronic obstructive pulmonary disease (COPD), congestive heart failure, and asthma.3 UK policy makers took a step further and introduced a financial incentive for general practitioners to identify the 2% of their practice population at highest risk of emergency admission and to manage them proactively (case management).

We discuss the uncertainties around identification, prevention, and management of patients at high risk of emergency admission and suggest alternative approaches.

Limited potential for reducing admissions

Risk prediction models use clinical, demographic, and healthcare use data to identify people at risk of emergency admission.4 5 6 A systematic review identified 27 models to predict future emergency admission in community dwelling adults.7 The six best performing models showed good discrimination for the outcome of future emergency admission (c statistics 0.79-0.83).7

We used two models—the Scottish patients at risk of readmissions and admission (SPARRA) model8 and the UK Nuffield trust model9—to estimate their likely effects on emergency admissions. SPARRA (version 3) was developed in a cohort of over 3.5 million Scottish people,8 and the Nuffield trust model was developed in a cohort of over 1.8 million English people.9 Both can be used by GPs to identify the 2% of patients at highest risk of emergency admission.

Applying the SPARRA model to its study population with a risk band of ≥50% equates to the top 1.6% of the population—53 394 people, responsible for 31 775 emergency admissions (10.7% of total emergency admissions) (table 1). Using the risk band of ≥40% would represent 3.1% of the population and 17.7% of emergency admissions. So an intervention that was 100% effective at preventing emergency admissions applied to the top 1.6% or 3.1% of the population could reduce emergency admissions by a maximum of 10.7% or 17.7%.

Table 1

Predicted and observed emergency admissions to acute hospitals* in Scotland using SPARRA (1 Apr 2012 to 31 Mar 2013)

View this table:

Applying the Nuffield trust model with a risk band of ≥0.3 represents the top 1.7% of the study population and equates to 31 005 people, responsible for 13.4% of admissions. A lower cut-off of ≥0.2 would identify 63 588 people (3.5% of the study population; 22.2% of admissions) (table 2). Thus, using these models to identify the 1.6% or 1.7% most at risk would target at most either 10.7% or 13.4% of all emergency admissions.

Table 2

Nuffield Trust risk prediction model (five sites)

View this table:

Now consider how the SPARRA risk model may be applied in general practice. In the validation cohort (table 1), 9.2% of patients were admitted as an emergency in the following year. Thus, a hypothetical Scottish general practice with a population of 10 000 patients could expect 920 patients to have at least one emergency admission a year. The 1.6% of these 10 000 patients identified for case management equates to 160 people. Based on the proportion of patients admitted in the SPARRA ≥50% risk band, we can expect 95 of these high risk patients to be admitted at least once in the following year. Thus, 825 patients (90%) having at least one emergency admission would not be identified using this approach, highlighting the limitations of applying risk prediction models for the purposes of reducing overall emergency admissions.

Which emergency admissions are preventable?

Most risk prediction models are designed to predict all emergency admissions regardless of cause, but many, such as those for acute appendicitis, are unavoidable. It would be preferable to identify the subset of emergency admissions that might be prevented with intensified primary care management, referred to as ambulatory care sensitive (ACS) admissions. These account for about 20% of all emergency admissions, with over half occurring in people aged ≥65.10 11 Definitions of an ACS condition vary internationally, but the list of conditions used by the state health department of Victoria, Australia, is commonly used in the UK (box).12 13 Risk prediction models have been developed that specifically identify patients at high risk of ACS admissions (rather than all cause admissions).14 15

Ambulatory care sensitive conditions*

Acute conditions

  • Cellulitis

  • Dehydration

  • Dental conditions

  • Ear, nose, and throat infections

  • Gangrene

  • Gastroenteritis

  • Nutritional deficiencies

  • Pelvic inflammatory disease

  • Perforated/bleeding ulcer

  • Pyelonephritis

Chronic conditions

  • Angina

  • Asthma

  • Chronic obstructive pulmonary disease

  • Congestive heart failure

  • Convulsions and epilepsy

  • Diabetes complications

  • Hypertension

  • Iron deficiency anaemia

Vaccine preventable conditions

  • Influenza

  • Pneumonia

  • *As defined by the state health department of Victoria, Australia

The effect of reducing ACS admissions can be quantified; we would expect about 20% of the SPARRA population at highest risk of future emergency admission, responsible for 10.7% of all emergency admissions, to be amenable to prevention—that is, 2.1% of all emergency admissions (this figure may vary depending on the patient demographics of the practice population). A typical general practice having identified and proactively managed their 160 highest risk patients can therefore hope to prevent approximately 19 of the expected 95 emergency hospital admissions a year (figure 1). The limited literature in this area indicates that up to 18% of all ACS admissions could be prevented, based on all local authorities performing at the level of those in the best performing fifth.11 Thus, although proactive management may result in other benefits, it would only prevent four emergency admissions.

Figure1

Figure 1 Application of a risk prediction model to a hypothetical general practice population of 10 000 patients to identify the top 1.6% most at risk of at least one emergency admission over the next year

Case management may not reduce admissions

As part of UK policy to reduce emergency admissions GPs are incentivised to case manage patients identified as high risk.16 They do this by creating a written personalised care plan that includes healthcare provider information, medical history, and key action points.17 Practices must also provide timely telephone access and should contact all patients discharged from hospital within three days to facilitate care coordination.

Evidence shows that case management improves patient satisfaction with care, promoting high levels of professional satisfaction and reducing caregiver strain,18 19 20 21 22 but its impact on reducing future emergency admissions has not been demonstrated in systematic reviews of randomised controlled trials (RCTs).23 24 25 26 A systematic review and meta-analysis of case management for adults with long term conditions in primary care (n=36 studies) found that case management had a small significant effect on patient satisfaction, in both the short term (standardised mean difference (SMD) 0.26 (0.16 to 0.36)) and long term (SMD 0.35 (0.04 to 0.66)) but did not reduce use of primary or secondary care or costs of care.26 Subgroup analysis showed that case management delivered by a multidisciplinary team, including social workers, had a small non-significant effect on reducing use of secondary care in the short term, which may merit further investigation.26 Current evidence does not support case management as an effective intervention for reducing emergency admissions, despite the effort it requires from the primary care team.

Time for a shift in focus?

Virtual wards, which use the same staffing and processes of a hospital ward, but patients are cared for at home, were hoped to reduce emergency admissions.27 But evaluations of this model in the UK and United States have found that it did not achieve the anticipated reductions in emergency admissions, even for ACS conditions.27 28 Several other approaches merit consideration.

Firstly, targeting specific conditions might reduce emergency admissions. In the UK five conditions account for more than 50% of all ACS admissions: urinary tract infection and pyelonephritis (16%); COPD (12%); pneumonia (10%); ear, nose, and throat infections (9%); and convulsions or epilepsy (7%).10 Some evidence shows that community based interventions may work in reducing these types of admissions. A Cochrane review of seven primary care RCTs found that integrated disease management for COPD—including patient education, self management, structured follow-up, and exercise—successfully reduced the number of patients with one or more respiratory admissions over 12 months (20 of 100 patients in the intervention group (95% confidence interval 15 to 27), compared with 27 of 100 in the control group; odds ratio 0.68 (0.47 to 0.99, P=0.04).29 So using risk prediction models to identify patients at high risk of admission for these conditions, and then targeting intervention efforts at these patients, could have an effect on overall admission rates.

Introduction of a primary care pay for performance scheme in England was associated with a fall in emergency admissions for incentivised ACS conditions (such as epilepsy and congestive heart failure) compared with those that were not incentivised (such as cellulitis and urinary tract infections or pyelonephritis).30 Other studies have demonstrated similar reductions in emergency admissions for incentivised conditions in primary care.31 32

Secondly, targeting end of life care may have an effect.33 A large retrospective analysis of 29 538 UK adults who received home based end of life nursing care found that they were significantly more likely to die at home than matched controls (unadjusted odds ratio 6.16 (5.94 to 6.38, P<0.001)).33 Emergency admissions were also significantly lower among those who received the intervention (0.14 versus 0.44 emergency admissions per person between January 2009 and November 2011, P<0.001).33

Thirdly, variation in medical practice should be considered. GP referral rates are influenced by individual tolerance of clinical uncertainty and access to services (especially out of hours).34 However, the extent to which referral patterns affect rates of emergency admission remains largely unknown. Variation is also a problem in secondary care, with significant differences in emergency admission rates for ACS conditions between countries.35 36 37 In an analysis of UK emergency admission rates (2008-11) a threefold variation was found in the rates of admissions for 14 ACS conditions across 129 hospitals, largely explained by local socioeconomic differences.35 36 Concentrating efforts on reducing admissions in more deprived areas, known to have higher levels of multimorbidty, may help.38

Lastly, focusing efforts on emergency departments and inpatients may reduce overall use of hospital beds. To date, RCTs based in emergency departments have largely targeted older patients through risk screening for focused geriatric assessment with post discharge follow-up, with mixed results.39 40 41 However, a systematic review and meta-analysis of 42 RCTs found that interventions to reduce the risk of 30 day readmission in medical and surgical inpatients were successful overall (pooled relative risk 0.82 (0.73 to 0.91)).42 Successful interventions usually had five or more components targeting patient factors (including multimorbidity, functional capacity, socioeconomic factors, and self care) as well as caregiver capabilities. Typically these complex interventions were coordinated after inpatients were discharged by at least two healthcare providers who made regular contact with the patient, including home visits.42 A second systematic review (n=26 RCTs) of inpatients with chronic conditions reported that interventions initiated during hospital admission and continued after discharge (through home visits or telephone follow-up) for a minimum of one month were effective in reducing readmissions at 180 days (pooled odds ratio 0.77 (0.62 to 0.96)) and 365 days (0.58 (0.46 to 0.75)).43 An evidence synthesis of interventions to reduce the length of inpatient hospital stay reported that, although evidence was varied and frequently lacked a robust study design, a range of interventions showed promise.44 Examples of such multidisciplinary interventions include improved or early supported discharge programmes and use of care pathways.44

Moving forward

An inherent feature of risk stratification models is that they can, at best, identify only a minority of patients who will experience an emergency admission.45 Current evidence does not support community initiated case management as an effective method of reducing emergency admissions. Other approaches to consider include identifying and directing community based interventions at specific ACS admissions, inpatient interventions targeting the transition from secondary to primary care to reduce readmissions, and interventions that focus on reducing the length of hospital inpatient stay.

Key messages

  • Risk prediction tools to identify people at high risk of future emergency admission have limitations

  • Current evidence does not support community initiated case management as an effective way of reducing emergency admissions

  • Efforts should focus on the prediction of admissions for conditions that are more amenable to prevention in the community, particularly those accounting for most emergency admissions

  • Interventions that involve coordinated care of inpatients as they are discharged to primary care and those targeting the length of inpatient stay seem promising

Notes

Cite this as: BMJ 2016;352:h6817

Footnotes

  • Acknowledgements: We thank Rachel Porteous and Theo Georghiou for providing additional data for the SPARRA and Nuffield risk prediction models, respectively.

  • Contributors and sources: This article is based on the combined experience of EW, a GP who is currently conducting her PhD on risk prediction of emergency admissions in older community dwelling people, SS, an academic GP whose research interests focus on multimorbidty and associated health outcomes including healthcare utilisation, TF, an academic GP with significant research experience in risk prediction models for use in primary care, and MR, a health service researcher who has carried out research on hospital referral by GPs for over 20 years. MR is guarantor.

  • Funding source: Emma Wallace is funded by the Health Research Board (HRB) of Ireland under the Research Training Fellowship for Healthcare Professionals award, grant no. HPF/2012/20. This research was conducted as part of the HRB Scholar’s programme in Health Services Research under grant no. PhD/2007/16 at the HRB Centre for Primary Care Research, grant HRC/2007/1.

  • Competing interests: We have read and understood BMJ policy on declaration of interests and have no relevant interests to declare.

  • Provenance and peer review: Not commissioned; externally peer reviewed.

References

View Abstract