Internal and Emergency Medicine

, Volume 8, Issue 3, pp 249–254

Derivation of a nomogram to estimate probability of revisit in at-risk older adults discharged from the emergency department

Authors

    • Centre for Clinical Research in Emergency Medicine (CCREM), Western Australian Institute for Medical Research
    • School of Primary, Aboriginal and Rural Health Care, University of Western Australia
  • Sarah Fitzhardinge
    • School of Primary, Aboriginal and Rural Health Care, University of Western Australia
  • Karren Pronk
    • School of Primary, Aboriginal and Rural Health Care, University of Western Australia
  • Marani Hutton
    • Western Australian Department of HealthSouth Metropolitan Health Service
  • Yusuf Nagree
    • School of Primary, Aboriginal and Rural Health Care, University of Western Australia
  • Mark Donaldson
    • Department of Geriatric MedicineRoyal Perth Hospital
EM - ORIGINAL

DOI: 10.1007/s11739-012-0895-5

Cite this article as:
Arendts, G., Fitzhardinge, S., Pronk, K. et al. Intern Emerg Med (2013) 8: 249. doi:10.1007/s11739-012-0895-5
  • 210 Views

Abstract

Estimation of the risk of revisit to the emergency department (ED) soon after discharge in the older population may assist discharge planning and targeting of post discharge intervention in high risk patients. In this study we sought to derive a risk prediction calculator for this purpose. In a prospective observational study in two tertiary ED, we conducted a comprehensive assessment of people aged 65 and over, and followed them for a minimum of 28 days post discharge. Cox proportional hazard models relating any unplanned ED revisit in the follow up period to observed risk factors were used to compute a probability nomogram. From 1,439 patients, 189 (13.1 %) had at least one unplanned revisit within 28 days. Revisit probability was weighted towards chronic and difficult to modify risk factors such as depression, malignancy and cognitive impairment. We conclude that the risk of revisit post discharge is calculable using a probability nomogram. However, revisit is largely related to immutable factors reflecting chronic illness burden, and does not necessarily reflect poor ED care during the initial index presentation.

Keywords

Emergency departmentRevisitRisk assessmentDischarge planningAllied health personnel

Introduction

When an older person is a patient at an emergency department (ED), and is discharged, the risk of revisit to the ED should form part of that discharge decision. Up to 20 % of patients aged 65 and over who are discharged home from the ED will have an unscheduled revisit to the ED within a month [1]. Early revisit post discharge has been identified as a quality indicator of emergency care [2]; has negative consequences for patient outcomes and hospital occupancy; and is disproportionate in older people [3]. Successful ED care for older people with complex needs should encompass not only management of the acute illness or injury necessitating ED attendance, but successful transition from ED to home [4]. This presents a challenge to the traditional ED model of care, which by its nature is episodic and focussed on the acute presenting problem. An ED visit by an older person may be a sentinel marker of physical, functional and psychosocial decline, and while acute care in this population must be exemplary, high rates of early revisit post discharge could indicate suboptimal recognition of the at-risk patient. It would be especially important for ED care processes if the revisit risk was correlated with features of the acute care received on the first presentation.

With demographic change, the proportion of ED patients that are aged 65 and over is increasing [5]. A number of centres now utilise specialist multidisciplinary geriatric teams within the ED in order to optimise care of older people, though the evidence in support of such teams is variable[6]. Optimal care has many features including accurate diagnosis, recognition of concurrent geriatric syndromes, minimising iatrogenic illness and injury, avoiding unnecessary hospitalisation and organising safe discharge to reduce revisit risk [79]. To some extent, the latter two priorities are part of a spectrum of competing interests that presents a great challenge to emergency physicians. The decision to admit or discharge an older person is sometimes finely balanced, and represents a risk–benefit decision where the benefits of avoiding hospitalisation [10, 11] are pitted against the risks of discharge [12, 13].

To this end, a number of tools that enable prediction of discharge risk have been developed as a way of assisting with the discharge decision by identifying patients at increased risk of adverse event post discharge [14]. The derivation of these tools has varied from study to study, but all have been based in part or fully on self reporting questionnaires in a convenience sample of respondents. Although the tools and their use in associated studies have sometimes been used to identify ‘high risk’ patients in a dichotomised fashion, none have been used to apply a risk-probability to an individual patient. If a robust tool could be developed to calculate the risk on any one individual revisiting an ED, it could assist physicians in cases where risk–benefit discharge decisions are being considered.

The aim of this study was to derive a risk probability calculator that could potentially be used to predict return visit risk in at risk older people discharged from the ED.

Methods

A prospective study was conducted at two large referral hospital EDs conducted over 12 months in 2009–2010. Institutional Ethics Committee approval at each site was obtained. We enrolled consecutive patients aged 65 and over who underwent a comprehensive assessment by a care coordination team (CCT) in the ED prior to discharge. The composition and role of the CCT have been described elsewhere [15, 16]. In brief, the teams consist of experienced allied health personnel who work in conjunction with ED medical and nursing staff with a view to facilitating safe discharge of suitable older people from the ED. The CCT were present in the ED 7 days a week, but not outside the hours 2100-0700. Patients could be referred to the CCT for comprehensive assessment by any ED clinical staff member. This could be on the basis of clinical judgement, or the use of a risk screening sticker that prompted referral if the answer to any one of four screening questions was yes (Fig. 1).
https://static-content.springer.com/image/art%3A10.1007%2Fs11739-012-0895-5/MediaObjects/11739_2012_895_Fig1_HTML.gif
Fig. 1

Risk screen

Consenting enrolled patients were comprehensively assessed by at least one CCT staff member with data contemporaneously collected from the patient (and family/carer where relevant) to capture the circumstances of their acute presentation; their medical comorbid status; their pre-existing and current functional needs; and what if any interventions were instituted by the CCT in order to facilitate patient discharge (Table 1). The median (IQR) time for assessment and management was 60 (40–100) min.
Table 1

Variables collected from study patients during assessment

Demographic and acute presentation

Presence of co-morbidities

Allied health assessment

Post discharge referral

Gender

Airways disease

Living arrangements

Medical services e.g. memory clinic

Age

Cardiac failure

Hours/week of formal community help

Allied health services e.g. OT home visit

Ethnicity

Atrial fibrillation

Hours/week of informal community help

Community services e.g. meals on wheels

Primary language

Ischaemic heart disease

Mobility assessment

Government/social security services

Triage code in ED

Back pain

Visual assessment

 

Mode of transport to ED

Depression

Hearing assessment

 

ED length of stay

Anxiety

Urinary continence

 

ICD10 discharge diagnosis

Dementia

Faecal continence

 

ED attended

Diabetes mellitus

Cognitive assessment

 

Number of ED visits in past year

Peripheral atherosclerosis

Number of falls in past 6 months

 
 

Malignancy

Weight loss in past 6 months

 
 

Alcoholism

Number of medications

 

The primary outcome measure was any unplanned ED revisit in the 28 days post discharge. All patients were followed up for a minimum of 28 days through established data processes [17], and, where necessary, telephone contact. A Cox proportional hazards model relating ED revisit to each of the predictive variables measured (Table 1) was constructed. Variables that were neither statistically significant predictors of revisit nor confounders of other variables were eliminated to reach a final best fit parsimonious model. Testing for interaction terms was conducted. We derived a probability nomogram by weighting individual risk factors on the basis of their hazard ratios estimated by the model, using methods described by Harrell et al. [18]. A p value of 0.05 was used as the threshold for statistical significance. Data analysis was performed using Stata (StataCorp, TX, USA) and R (R Development, Vienna, Austria) software.

Results

We enrolled 1,439 consecutive patients aged 65 and over in the study period who underwent complete comprehensive assessment and were discharged home after CCT review. The circumstances of ED attendance for this study population is summarised in Table 2.
Table 2

Summary of study population on arrival (n = 1,439 patients)

Parameter

(n, % unless indicated)

Age (mean, SD)

78, 8

Female

806, 56

Mode of arrival to ED

 Ambulance

815, 57

 Non-ambulance transport

624, 43

Australasian Triage Scale

 2 (urgent)

268, 19

 3 (semi-urgent)

570, 40

 4/5 (not urgent)

601, 42

Most common individual presenting problems

 Fall with no major injury

181, 13

 Atraumatic joint/back pain

112, 8

 Chest pain

107, 7

At 28 days, 189 (13.1 %) of patients had made at least one unplanned ED revisit with another 20 (1.4 %) patients dying in the follow-up period. 88 (46.6 %) of revisits were clearly unrelated to the index presenting problem at enrolment, in the remaining cases the revisit was for the same problem or a problem that may have been related to the index attendance. Revisit rates at 3 and 90 days were 4.4 and 23.2 %, respectively.

The univariate association between the primary outcome of interest (any unplanned revisit within 28 days) and potential risk factors is shown in Table 3. Because of the large number of index visit diagnoses, the analysis for each diagnosis is not shown, but there was no significant difference between revisit and no revisit groups with respect to the initial discharge diagnosis. Using Cox proportional hazards models, we created multiple models until we derived a parsimonious model of best fit that included both significant predictive variables and significant confounders (Table 4).
Table 3

Univariate analysis

Variable

Any revisit (n = 189)

No revisit (n = 1,250)

Significance

Age in years (median)

80

78

<0.01

Male

50 %

42 %

0.04

Australian born

50 %

47 %

NS

Non-urgent triage code for index visit

40 %

42 %

NS

Ambulance transport for index visit

58 %

56 %

NS

Index ED length of stay in hours (median)

5.4

5.1

NS

Nil ED attendances in past year

57 %

74 %

<0.01

Airways disease

6 %

4 %

NS

Cardiac failure

8 %

5 %

0.04

Atrial fibrillation

16 %

7 %

<0.01

Ischaemic heart disease

20 %

9 %

<0.01

Back pain

6 %

3 %

0.03

Depression

4 %

1 %

<0.01

Anxiety

3 %

1 %

0.01

Diabetes mellitus

14 %

10 %

NS

Peripheral atherosclerosis

2 %

1 %

NS

Malignancy

13 %

5 %

<0.01

Alcoholism

2 %

1 %

NS

Living alone independently

37 %

30 %

0.03

No formal community help

54 %

61 %

NS

No informal community help

78 %

84 %

NS

Mobility safe unaided

57 %

62 %

NS

Nil visual aids

67 %

70 %

NS

Total urinary continence

68 %

71 %

NS

Total faecal continence

88 %

88 %

NS

Moderate to severe cognitive deficit

23 %

16 %

0.02

Nil falls in past 6 months

57 %

59 %

NS

>5 kg weight loss in past 6 months

17 %

15 %

NS

6 or more medications

49 %

36 %

<0.01

Discharge referral to medical services

15 %

14 %

NS

Discharge referral to allied health services

15 %

13 %

NS

Discharge referral to community services

10 %

7 %

NS

Discharge referral to government services

4 %

3 %

NS

Table 4

Multivariate analysis

Variable

HR (95 % CI)

Prior registrationsa

1.3 (1.1–1.4)

Ageb

1.02 (1.00–1.03)

Gender

 Male

1.4 (1.0–2.0)

 Female

1.0*

Medications

 6+

1.4 (1.0–2.1)

 0–5

1.0*

SIS

 0–2

1.2 (0.7–2.1)

 3–4

1.5 (0.9–2.6)

 5–6

1.0*

Malignancy

2.6 (1.4–4.8)

CCT intervention

1.3 (0.8–1.7)

Depression

3.2 (1.0–8.4)

* Referent value

aPer registration over zero

bPer year over 65

The predictive nomogram derived from this final model is shown in Fig. 2. To use the nomogram, each risk factor is individually assessed, and if present, assigned the appropriate number of corresponding points (top line). Some risk factors are binary, others scaled in either a linear (e.g. number of prior ED registrations in the previous 12 months) or nonlinear (e.g. age, SIS) fashion. The points allocated on the basis of each risk factor are then summed to obtain total points for the patient, which is used to calculate the probability of having no unplanned revisit in the 28 days after discharge (bottom line). In practice, the figure nomogram can be replaced by a simple electronic spreadsheet (see online supplementary data) that will automatically calculate the probability of revisit based on risk data entered into the spreadsheet, avoiding the need for use of the paper based nomogram presented in Fig. 2.
https://static-content.springer.com/image/art%3A10.1007%2Fs11739-012-0895-5/MediaObjects/11739_2012_895_Fig2_HTML.gif
Fig. 2

Derived nomogram

Discussion

In this study of older adults that require comprehensive allied health assessment prior to ED discharge, we have found that the risk of ED revisit within 28 days of discharge is largely related to chronic immutable factors rather than characteristics associated with the index presentation. The implications of this for clinical ED practice are important. ED revisit by an older person may be considered a failure of discharge decision making or planning, but the factors that influence revisit are not strongly related to the acute presenting problem. ED care has a focus on acute illness and injury, but even exemplary initial acute care is unlikely to have a meaningful impact on the risk of revisit in some populations. A different approach is required.

The risk nomogram we have developed allows an estimation of the probability of revisit after discharge. This has some potential benefit over existing risk prediction tools. Knowing the estimated probability, and the factors that are contributing to that probability, could allow a stratified approach to post discharge care. At one end of the spectrum will be patients at low risk (<10 %) of revisit who could be discharged with standard procedures. At the other end, patients whose estimated probability is well in excess of the threshold seen in population based studies (30 % or more) could potentially benefit from closer discharge planning or post discharge intervention by a multidisciplinary service experienced in the care of older people with chronic illness, though the success rate of such programs on ED revisit is mixed [1921].

The overall rate of 28 day revisit in our study population of 13 % is similar to that found in similar patient cohorts in other jurisdictions [1, 3, 12, 13]. There is no agreed figure for what constitutes an acceptable standard of revisit rate, but rates that are manifestly higher than the 10–25 % range found in the literature should raise alarm. This is where a risk probability calculator offers refinement to clinical practice, so that when a patient is identified who has a risk of revisit that is high, it might support targeted intervention of that patient and the identifiable risks post discharge.

Factors that particularly contribute to an increased probability of revisit in our study show similarities with those risk factors found by other researchers. The TRST tool includes five fields, three of which (polypharmacy, cognitive impairment and a history of recent hospital use) are similar to domains in our nomogram [22]. Similarly the ISAR tool contains six fields, with the same three similar domains to those found in our study [23]. The outcome measure in ISAR is a composite one of death, institutionalisation or functional decline rather than rehospitalisation.

Our study identifies the highest risk of revisit to be associated with patients with depression, which is not a feature of other risk tools. It is increasingly recognised that depression is common in older ED attendees[24], and that depression is significantly associated with health resource usage[25].Depression may be associated with social isolation, poor motivation and poor health literacy that may all contribute to contact with ED services. We find moderate cognitive impairment contributes more to risk than either normal cognition or profound impairment, it is possible that the latter is already recognised as having a need for services and support that slightly ameliorates the risk.

We emphasize again that the revisits of some older patients post discharge is inevitable, and does not represent a failure of ED care. As we have shown in this study, the patients at highest risk for revisit are those with factors that may be resistant to intervention, especially any intervention that can be meaningfully undertaken during the typically brief period of ED care.

There are limitations in the applicability of our nomogram. Firstly the nomogram was derived from a population that had already undergone a brief risk screen, and were referred to CCT for comprehensive assessment, in other words it is not necessarily applicable to all older patients being discharged from ED though it is likely that patients negative on a risk screen would also have a low predicted probability of revisit. Secondly, this paper reports the derivation of the nomogram, and it needs validation in a separate population. Thirdly and most importantly, knowing the estimated probability of revisit will not assist clinical practice if we are unable to successfully intervene in those patients at highest risk. Further studies to test whether this is possible are planned http://www.ANZCTR.org.au/ACTRN12612000798864.aspx.

In conclusion, from this large study of prospectively enrolled patients undergoing comprehensive allied health assessment prior to discharge we have developed a nomogram to estimate the probability of early revisit to the ED post discharge in patients aged 65 and over. This has the potential to refine clinical practice and enable targeted intervention of high risk patients in the post discharge period, though both of these assertions need to be tested in future trials.

Acknowledgments

CCT members at each site provided invaluable assistance with the study. Mr Michael Phillips and Ms Sally Burrows assisted with statistical analysis. The research was funded by a grant from the State Health Research Advisory Council of Western Australia.

Conflict of interest

None.

Supplementary material

11739_2012_895_MOESM1_ESM.xlsx (13 kb)
Supplementary material 1 (XLSX 13 kb)

Copyright information

© SIMI 2013