Understanding which people with dementia are at risk of inappropriate care and avoidable transitions to hospital near the end-of-life: a retrospective cohort study

Abstract

Background

transitions between care settings near the end-of-life for people with dementia can be distressing, lead to physical and cognitive deterioration, and may be avoidable.

Objective

to investigate determinants of end-of-life hospital transitions, and association with healthcare use, among people with dementia.

Design

retrospective cohort study.

Setting

electronic records from a mental health provider in London, linked to national mortality and hospital data.

Subjects

people with dementia who died in 2007–2016.

Methods

end-of-life hospital transitions were defined as: multiple admissions in the last 90 days (early), or any admission in the last three days of life (late). Determinants were assessed using logistic regression.

Results

of 8,880 people, 1,421 (16.0%) had at least one end-of-life transition: 505 (5.7%) had early, 788 (8.9%) late, and 128 (1.5%) both types. Early transitions were associated with male gender (OR 1.33, 95% CI 1.11–1.59), age (>90 vs <75 years OR 0.69, 95% CI 0.49–0.97), physical illness (OR 1.52, 95% CI 1.20–1.94), depressed mood (OR 1.49, 95% CI 1.17–1.90), and deprivation (most vs least affluent quintile OR 0.58, 95% CI 0.37–0.90). Care home residence was associated with fewer early (OR 0.63, 95% CI 0.53 to 0.76) and late (OR 0.80, 95% CI 0.65 to 0.97) transitions. Early transitions were associated with more hospital admissions throughout the last year of life compared to those with late and no transitions (mean 4.56, 1.89, 1.60; P < 0.001).

Conclusions

in contrast to late transitions, early transitions are associated with higher healthcare use and characteristics that are predictable, indicating potential for prevention.

Transitions, Electronic medical records, Dementia, End-of-life, Emergency hospital admissions

Key points

• End-of-life transitions among patients with dementia are frequent occurring in one in six people before death.

• Early and late transition are associated with distinct socio-demographic and clinical characteristics

• Early transitions are associated with high levels of hospital use throughout the last year of life.

• Living in a care home is associated with fewer late and early transitions

• Differences suggest that late transitions are less predictable than early transitions, and are being driven by external factors.

Introduction

Dementia brings serious challenges for provision of care near the end-of-life [1, 2]. Care often becomes fragmented, and people with dementia can experience distressing moves from one care setting to another (‘transitions’) [3]. Such transitions can precipitate both physical and cognitive deterioration [4], and are frequently avoidable [5–7]. While deaths in hospital for people with dementia in England have fallen in recent years, they remain high [8], and Emergency Department visits in the last year of life are increasing [9].

Hospital transitions near the end-of-life have previously been defined as ‘burdensome’ if they occur either in the last three days of life (late transitions), or where there are multiple transitions in the last 90 days of life (early transitions) [3]. Burdensome transitions are associated with other indicators of poor end-of-life care, such as use of feeding tubes, presence of stage IV decubitus ulcers, and late enrollment to hospice (specialist palliative care) services [3]. These transitions have been little studied outside the US, where samples have been limited to Medicare beneficiaries [10–12]. In addition, little is known about the sociodemographic and clinical characteristics associated with late and early end-of-life transitions, and whether these differ. Our aim was to investigate the frequency and correlates of end-of-life hospital transitions among people with previously diagnosed dementia in an English sample to understand better the determinants of these transitions, the association with hospital use in the last year of life, and who is most at risk.

Methods

Setting and data source

This retrospective observational study used data from the South London and Maudsley National Health Service (NHS) Foundation Trust Biomedical Research Centre (SLaM BRC) Case Register and its Clinical Record Interactive Search (CRIS) application. SLaM provides comprehensive mental health services to a geographic catchment area of over 1.2 million residents in four south London boroughs (Croydon, Lambeth, Lewisham, Southwark). The CRIS application comprises a research repository of anonymised, structured and open-text data derived from electronic health records within SLaM [13, 14].

Population

The records of all patients older than 60 years with a diagnosis of dementia were retrieved from the SLaM BRC Case Register. Diagnosis of dementia was determined from structured fields based on ICD-10 codes F00x–03x and supplemented by an open-text data search developed using General Architecture for Text Engineering software [15]. A link to Office for National Statistics (ONS) mortality records was used to restrict the cohort to individuals who had died. CRIS is linked to Hospital Episodes Statistics (HES) covering all NHS trusts in England, providing individual-level information on hospital admissions for the last 12 months of life.

Outcome variables

The primary outcome was the presence or not of a hospital transition near the end-of-life. Two types of transition were defined, based on the work of Gozalo et al.: any hospital transition in the last three days of life (late), and multiple hospital transitions in the last 90 days of life (early). Multiple transitions are defined as either more than two hospitalisations for any reason or more than one hospitalisation for pneumonia, urinary tract infection, dehydration, or sepsis (Appendix 1) [3, 12]. Individuals with both early and late transitions were included in the early transition group for analysis.

Covariates

Data on age, gender, ethnicity, subtype of dementia, and date when the diagnosis of dementia was first recorded were extracted from CRIS. Date of death was obtained from ONS records. Socioeconomic position was measured using the 2015 Index of Multiple Deprivation (IMD), a measure of relative deprivation for 32,844 neighbourhoods in England termed Lower Super Output Areas (LSOAs). IMD was derived from the LSOA associated with the patient’s most recent address, and converted into quintiles (1 - most deprived, 5 - least deprived quintile). Dementia severity was estimated from the most recently recorded Mini-Mental State Examination (MMSE) score, either as structured data, or from an open-text data search. A percentage MMSE score was calculated, and analysed as a categorical variable: severe cognitive impairment (less than 33%), moderate (33–49% and 50–66%) and mild (more than 66%).

The most recent Health of the Nation Outcome Scale (HoNOS) was also obtained from CRIS. HoNOS contains 12 sub-scales covering behaviour, impairment, symptoms, and social functioning/context, with each scored on a five-point scale ranging from 0 (no problem) to 4 (severe/very severe problem) [16]. HoNOS scores were dichotomised (scores 0 and 1— no or minor problems, scores 2 to 4—mild to severe problems) to facilitate interpretation. Care home residence was derived from a database of known care home addresses linked to CRIS. We used HES ICD-10 diagnosis codes for the last hospital admission before death to calculate the Charlson Comorbidity Index of multimorbidity. Health conditions for building the Charlson Comorbidity Index were identified using translation codes from Sundarajan et al. [17], and scores were calculated using the amended HES-based weights from Bottle & Aylin [18].

Health care use in the last year of life

We used consecutive ‘finished consultant episodes’ (FCEs) from HES to identify the start and end of all non-elective hospital admissions during the last year of life, and used this to calculate the monthly cumulative number of admissions during the last year of life.

Statistical analysis

Descriptive statistics were used to describe the study population in terms of sociodemographic and clinical characteristics. A Chi-squared test for trend was used to examine the proportion of end-of-life hospital transitions observed by year of death. We calculated the mean cumulative number of hospital admissions in the last 12 months of life for those who had early, late and no end-of-life transitions, and used Kruskal–Wallis tests followed by Dunn’s post hoc test with Bonferroni correction to understand differences between groups.

Unadjusted and multivariable logistic regression modelling was used to investigate the sociodemographic and clinical factors associated with the two types of end-of-life hospital transition, analysed separately. Generalised estimating equations (GEE) were used to account for geographical clustering at the LSOA level. Explanatory variables were selected based on a priori hypothesis and significance in unadjusted analysis (P-value < 0.1). As the Charlson Comorbidity Index was available only for individuals who had at least one hospital admission, we did not include it in the multivariable GEE models. Differences between the Charlson index score between individuals who had early, late and no end-of-life transitions were examined using a Kruskal–Wallis test.

The analysis was performed using STATA version 12 (StataCorp LP, College Station, TX, USA).

Ethics statement

The source database is approved for secondary analysis by the Oxfordshire Research Ethics Committee C (reference 08/H0606/71 + 5).

Patient involvement

This project was planned with input from the Patient and Public involvement (PPI) groups at the Cicely Saunders Institute and the SLaM BRC. An ‘expert by experience’ has been part of this project since the study design.

Results

A total of 9,842 decedents who had a clinical diagnosis of dementia were identified. After removing duplicates, elective admissions, and people younger than 60 years, a total of 8,800 individuals and 16,447 hospital admissions remained for the time period 2007–2016 (Appendix 2). Most were women, white British, and the mean age at death was 85.5 years. The most common diagnosis was Alzheimer’s disease followed by unspecified and vascular dementia (Table 1). The mean time interval from diagnosis to death was 2.9 years. An MMSE score was available for 79.6% of the cohort, 1,230 (17.4%) of whom had severe dementia, 2,514 (35.6%) had mild dementia; 35.2% of individuals were care home residents. The mean Charlson Comorbidity Index Score (available for 6,991 individuals) was 3.5 (SD 4.2). The most common HoNOS problems recorded were cognitive problems, physical illness or disability problems, and problems with activities of daily living (Table 1).

Table 1.

Population characteristics

		Total		No transition		Late transition (Transitions last 3 days)		Early transition (Multiple transitions last 90 days)
		No.	%a	No.	%a	No.	%a	No.	%a
Sex (total n = 8,880)	Female	5,394	60.7	4,608	61.8	453	57.5	333	52.6
Ethnicity (total n = 8,490)	British	6,272	73.9	5,306	74.5	534	70.5	432	70.7
	Other white	913	10.8	748	10.5	98	12.9	67	11.0
	African Caribbean	863	10.2	710	10.0	84	11.1	69	11.3
	Other	442	5.2	357	5.0	42	5.5	43	7.0
Categories of Age at death (total n = 8,880)	<75	786	8.9	641	8.6	67	8.5	78	12.3
	75 to 79	1,087	12.2	887	11.9	113	14.3	87	13.7
	80 to 84	1,943	21.9	1,613	21.6	174	22.1	156	24.6
	85 to 89	2,503	28.2	2,116	28.4	227	28.8	160	25.3
	≥90	2,561	28.8	2,202	29.5	207	26.3	152	24.0
Year of death (total n = 8,880)	2007–2008	728	8.2	595	8.0	75	9.5	58	9.2
	2008–2009	778	8.8	649	8.7	77	9.8	52	8.2
	2009–2010	848	9.6	692	9.3	92	11.7	64	10.1
	2010–2011	943	10.6	805	10.8	78	9.9	60	9.5
	2011–2012	992	11.2	834	11.2	95	12.1	63	10.0
	2012–2013	1,117	12.6	936	12.6	97	12.3	84	13.3
	2013–2014	1,060	11.9	909	12.2	81	10.3	70	11.1
	2014–2015	1,226	13.8	1,024	13.7	97	12.3	105	16.6
	2015–2016	1,188	13.4	1,015	13.6	96	12.2	77	12.2
IMD quintiles closer to death (total n = 8,829)	1	2,612	29.6	2,160	29.1	225	28.7	227	36.2
	2	3,131	35.5	2,636	35.5	278	35.4	217	34.6
	3	1,567	17.8	1,324	17.9	136	17.3	107	17.0
	4	871	9.9	745	10.1	81	10.3	45	7.2
	5	648	7.3	551	7.4	65	8.3	32	5.1
Primary Dementia Diagnosis (total n = 8,880)	Alzheimer	3,636	41.0	3,073	41.2	318	40.4	245	38.7
	Unspecified	2,282	25.7	1,911	25.6	202	25.6	169	26.7
	Vascular	2,104	23.7	1,752	23.5	188	23.9	164	25.9
	Other	858	9.7	723	9.7	80	10.2	55	8.7
Years from diagnosis mean(SD) (total n = 8,789)		2.9	(2.4)	3.0	(2.4)	2.9	(2.4)	2.3	(2.1)
Patient lives in a care home* (total n = 8,880)		3,126	35.2	2,706	36.3	242	30.7	178	28.1
MMSE (total n = 7,070)	<33%	1,230	17.4	1,042	17.6	97	15.3	91	17.8
	33–49%	1,420	20.1	1,181	19.9	138	21.8	101	19.8
	50–66%	1,906	27.0	1,614	27.2	167	26.3	125	24.5
	≥67%	2,514	35.6	2,088	35.2	232	36.6	194	38.0
Charlson Comorbidity Index mean(SD) (total n = 6,991)		3.5	(4.2)	3.4	(4.1)	3.7	(4.2)	4.1	(4.7)
Aggressive, disruptive or agitated behaviour (total n = 7,830)		1,834	23.4	1,567	23.8	149	22.0	118	21.1
Problems with occupation and activities (total n = 7,681)		2,881	37.5	2,420	37.4	248	37.1	213	38.9
Problems with living conditions (total n = 7,748)		963	12.4	810	12.4	74	11.1	79	14.3
Problems with activities of daily living (total n = 7,817)		6,231	79.7	5,245	79.7	531	78.6	455	81.8
Problems with relationships (total n = 7,798)		1,434	18.4	1,217	18.5	119	17.7	98	17.6
Other mental and behavioural problems (total n = 7,750)		2,047	26.4	1,723	26.4	157	23.3	167	30.2
Problems associated with hallucinations (total n = 7,795)		903	11.6	771	11.8	68	10.0	64	11.5
Physical illness or disability problems (total n = 7,834)		5,602	71.5	4,694	71.2	457	67.4	451	80.4
Cognitive problems (total n = 7,829)		7,126	91.0	6,015	91.3	607	89.5	504	90.0
Problem drinking or drug-taking (total n = 7,830)		155	2.0	122	1.9	20	3.0	13	2.3
Non-accidental self-injury (total n = 7,832)		97	1.2	81	1.2	–	–	–	–
Problems with depressed mood (total n = 7,804)		886	11.4	725	11.0	69	10.2	92	16.6

MMSE: mini-mental state examination; SD: standard deviation; IMD: index of multiple deprivation (1 = most deprived, 5 = most affluent).

^aColumn % *Patients living in a care home by the time of diagnosis. – numbers not reported because < 10 individuals per cell.

1,421 (16.0%) people had at least one end-of-life hospital transition; 788 (8.9%) had a late transition, 505 (5.7%) had early transitions, and 128 (1.5%) had both types of end-of-life transition (Appendix 3) The proportion of people undergoing a late transition decreased from 10.3% in 2007–08 to 8.1% in 2015–16 (Chi-squared (1) = 8.31, P-value = 0.004). Early transitions did not change significantly over the nine years (chi-squared (1) = 0.05, P-value = 0.821).

People with early transitions had more hospital admissions during their last year of life (mean 4.56 SD 2.48) than those with late transitions (mean 1.89 SD 1.50) and those without an end-of-life transition (mean 1.60 SD 1.66) (Kurskal–Wallis X2(2) = 1,167.60, P-value < 0.001) (Figure 1). This difference in the cumulative number of hospital admissions was present from 12 months before death for individuals with early transitions (Appendix 4). In contrast, people with late transitions had similar numbers of hospital admissions compared to those with no end-of-life transition until the last month before death.

Figure 1.

Mean of the cumulative number of hospital admissions in the last year of life for individuals with dementia by type of end-of-life transition.

In unadjusted analysis, late transitions were positively associated with being male, ‘other white’ ethnicity, and having drinking or drug-taking problems. Late transitions were negatively associated with living in a care home and physical illness or disability problems. Early transitions were positively associated with being male, ‘other’ ethnicity, other mental and behavioural problems, physical illness or disability and problems with depressed mood. Older age was negatively associated with early transitions, as was living in the least deprived areas. Individuals who lived in care homes were less likely to have had an early transition (Appendix 5). Mean Charlson comorbidity score was higher in those experiencing early transitions compared to late and those with no transitions (Kruskal–Wallis X2(2) = 13.8, P-value = 0.001).

In the multivariable analysis, late transitions remained negatively associated with living in a care home and positively associated with drinking and drug taking. Early transitions remained positively associated with male gender, having physical illness or disability problems and problems with depressed mood, and negatively associated with older age, living in a less deprived area, and living in a care home (Table 2).

Table 2.

Multivariable logistic regression of characteristics of individuals with dementia by type of end-of-life transition

	Late transitions (Transitions last 3 days) multivariate		Early transitions (Multiple transitions last 90 days) multivariate
	OR	95% CI	OR	95% CI
Sex
Female	Ref		Ref
Male	1.15	(0.97 to 1.37)	1.33	(1.11 to 1.59)
Ethnicity
British	Ref		Ref
Other white	1.10	(0.84 to 1.44)	1.02	(0.78 to 1.35)
African Caribbean	1.15	(0.89 to 1.50)	1.00	(0.74 to 1.35)
Other	1.20	(0.84 to 1.72)	1.29	(0.89 to 2.86)
Categories of Age at death
<75	Ref		Ref
75 to 79	1.40	(0.97 to 2.01)	0.81	(0.55 to 1.19)
80 to 84	1.21	(0.87 to 1.67)	0.87	(0.63 to 1.20)
85 to 89	1.15	(0.83 to 1.61)	0.72	(0.52 to 1.00)
≥90	1.05	(0.74 to 1.49)	0.69	(0.49 to 0.97)
Year of death
2007–2008	Ref
2008–2009	0.93	(0.61 to 1.42)
2009–2010	1.19	(0.79 to 1.78)
2010–2011	0.85	(0.56 to 1.29)
2011–2012	0.92	(0.61 to 1.39)
2012–2013	0.88	(0.60 to 1.30)
2013–2014	0.73	(0.48 to 1.11)
2014–2015	0.78	(0.53 to 1.15)
2015–2016	0.79	(0.53 to 1.16)
IMD quintiles
(Most deprived) 1			Ref
2			0.77	(0.62 to 0.94)
3			0.71	(0.56 to 0.91)
4			0.54	(0.39 to 0.76)
(Most affluent) 5			0.58	(0.37 to 0.90)
Care home residence
No	Ref		Ref
Yes	0.80	(0.65 to 0.97)	0.63	(0.53 to 0.76)
Other mental and behavioural problems
No			Ref
Yes	0.91	(0.75 to 1.09)	1.04	(0.84 to 1.29)
Physical illness or disability problems
No	Ref		Ref
Yes	0.85	(0.71 to 1.02)	1.52	(1.20 to 1.94)
Problem drinking or drug-taking
No	Ref
Yes	1.61	(1.00 to 2.59)
Problems with depressed mood
No			Ref
Yes			1.49	(1.17 to 1.90)

OR: odds ratio; IMD: index of multiple deprivation.

Discussion

In a large sample of people with a clinical diagnosis of dementia and who subsequently died, end-of-life hospital transitions were frequent, occurring in one in six people before death. While late transitions were more common than early transitions, they have decreased over time. In contrast, early transitions show no reduction and are associated with high levels of hospital use throughout the last year of life. Both types of end-of-life transition were less frequent among individuals living in a care home. Early transitions were more frequent among men, younger age groups, more deprived groups, and people experiencing problems with depressed mood or physical or disability problems. Late transitions were more frequent among people with drinking or drug taking problems.

Overall, 16.0% of the cohort in our study experienced an end-of-life hospital transition, which is lower than cited rates in the US [3, 12], but higher than elsewhere in Europe [19]. These differences between studies are likely to be due to a combination of factors including differences in the population, the time period, and the health care setting. In a study of nursing home residents with advanced cognitive impairment in the US, Gozalo et al. found 19.0% had at least one end-of-life transition, and they were more common among people with non-white ethnicity, men, and people without advance directives and do-not-hospitalise orders [3]. Teno et al. investigated the change over time in late and early end-of-life transitions among a cohort of fee-for-service Medical Beneficiaries, and found that transitions in the last 3 days of life increased between 2000 and 2009 whereas multiple hospitalisations in the last 90 days of life decreased in the same time period [12]. Aaltonen et al. found that 9.5% of people with dementia in residential care in Finland experienced at least one end-of-life transition [19]. This study found that both early and late end-of-life transitions were associated with younger age, male gender, and living in sheltered housing (compared to nursing homes). However, no information was available on illness related factors such as the severity of cognitive or physical impairment.

We found that late end-of-life transitions decreased over the time period. This coincides with a fall in hospital deaths among people with dementia [8], and is likely to be a consequence of end-of-life care policies and strategies in England, which have been dominated by a focus on the place of death as an indicator of end-of-life care quality [20]. In contrast, we found that early end-of-life transitions remained stable over the time period, and were associated with almost three-fold more hospital admissions throughout the last year of life. This highlights the importance of developing policies focused on innovations in care and support that target the early transitions group. While around 40% of people with dementia die in hospital [8], only 16% of our cohort experienced end-of-life hospital transitions. This is likely to reflect the fact that many people who die in hospital are admitted before their last three days of life.

Similarly to Aaltonen et al. [19], in our cohort there was little overlap between late and early transitions, with 1.5% of the cohort experiencing both types of hospital transitions. We were able to examine a range of sociodemographic and clinical characteristics, and we found the two types of transition were associated with distinct characteristics. Early transitions, but not late transitions, were associated with male gender, younger age, relative deprivation, physical problems, problems with depression, and more comorbidities. These differences suggest that early transitions are more predictable than late transitions, and may occur as a result of failure of community support systems, indicating potential scope for development of preventative interventions. Studies in other patient groups (for example older people, people with cancer) have identified similar characteristics as being associated with more aggressive end-of-life care [21, 22].

Living in a care home was associated with fewer late and early transitions. These findings are consistent with a study in Canada which found individuals not living in a long-term facility experienced a marked increase in hospital admissions at the end-of-life compared to those who did [23], and a previous study from our group which showed that care home residents are less likely to have Emergency Department attendances during the last year of life [9]. Together these findings may reflect a role of care homes in hospital admission avoidance. Further research is needed to understand the underlying mechanisms.

We found no association between aggressive, disruptive or agitated behaviour and hospital transitions. Previous studies have found these behaviours are common among people with dementia in the hospital setting [24], and can precipitate transitions between care settings [25]. This difference may reflect the inclusion in our study of people with mild and moderate, as well as severe, dementia.

All studies using routine data are limited by the data available. In this study we did not have information about social or community care that might influence hospital utilisation among these individuals. We also did not have information about the appropriateness of hospital transition for individual patients: some transitions near the end-of-life are necessary. Because information about comorbidities was derived from hospital data, we were unable to measure comorbidities for individuals with no hospital admissions in the last year of life (21.3% of the sample). However, linking administrative to clinical data provided a rich data resource, allowing examination of variables such as physical symptoms and cognitive impairment. Access to hospital admission data covering the last 12 months of life enabled the relevance of end-of-life transitions within the wider health system to be explored. Our study included, but was not limited to, care home residents, allowing examination of the association of place of residence with end-of-life hospital transitions. Identifying people with dementia from a clinical resource reduced the risk of bias from incomplete death certification of dementia [13]. However, our study population was limited to people with a diagnosis of dementia recorded in the clinical records within a specialist psychiatric hospital, which is likely to limit the generalisability.

Policies to improve end-of-life care over the past decade have focussed on reducing hospital deaths. These policies have coincided with a fall in late, but not early, end-of-life transitions among people with dementia. Innovations in care and support that reduce early end-of-life hospital transitions are needed. Given the high hospital use associated with these transitions, such innovations have potential to benefit both individuals and the health care system. The distinct sociodemographic and clinical characteristics associated with early end-of-life transitions should be used to guide efforts to develop interventions that reduce reliance on hospital care for people with dementia in their last months of life.

Supplementary data mentioned in the text are available to subscribers in Age and Ageing online.

Declaration of Source of Conflicts of Interest: None.

Declaration of Source of funding: K.E.S. is funded by a National Institute for Health Research (NIHR) Clinician Scientist Fellowship (CS-2015-15-005). The data resource and R.S. are funded by the NIHR Biomedical Research Centre and Dementia Biomedical Research Unit at South London and Maudsley NHS Foundation Trust and King’s College London. This article represents independent research funded by the NIHR. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health and Social Care.