Retrospective study of more than 5 million emergency hospital admissions: epidemiology and outcomes for people with dementia

doi:10.21203/rs.3.rs-1593977/v1

Download PDF

Research Article

Retrospective study of more than 5 million emergency hospital admissions: epidemiology and outcomes for people with dementia

https://doi.org/10.21203/rs.3.rs-1593977/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

People with dementia (PwD) as emergency acute hospital admissions may be at higher risk of inappropriate care and poorer outcomes including longer hospitalisations and higher risk of emergency re-admission or death. Since 2009, numerous national and local initiatives in England have sought to improve hospital care for PwD. We compared outcomes of emergency admissions for patients aged 65+ with and without dementia across three points in time.

Methods

We analysed emergency admissions (EAs) from the Hospital Episodes Statistics datasets for England 2010/11, 2012/13 and 2016/17. Dementia upon admission was based on a diagnosis in the patient’s hospital records within the last five years. Outcomes were length of hospital stays (LoS), long stays (>=15 days), emergency re-admissions (ERAs) and deaths within 30 days of discharge. Covariates included patient demographics, pre-existing health and reasons for admission. Hospital indices of dementia care relating to staff training, hospital governance and patient-focus were derived from National Audit of Dementia survey returns. Hierarchical multivariate regression analysis was used - separately for males and females – to evaluate patient group differences on each outcome adjusted for covariates and to examine associations with the dementia care indices.

Results

We included 178 acute hospitals with a total of 2,569,007 EAs for male patients and 3,011,099 for female patients, of which 356,992 (13.9%) were male PwD and 561,349 (18.6%) female PwD. Covariate-adjusted differences in LoS changed little over time and in 2016/17 were 17% (95%CI 15%-18%) and 12% (10%-14%) longer for male and female PwD respectively. Adjusted excess risk of an ERA for PwD reduced over time to 17% (15%-18%) for males and 17% (16%-19%) for females. Adjusted mortality for PwD was around 30% higher throughout. No associations were found between outcomes and hospital indices of dementia care.

Conclusions

Over the six-year period, covariate-adjusted hospital LoS, ERA and mortality rates for PwD showed little beneficial change relative to similar patients without dementia, despite many national and local efforts to improve hospital care for this patient group. Group differences after adjustment were mostly small and may reflect factors beyond hospital control. Greater emphasis is needed on avoiding unnecessary hospitalisations for PwD.

Dementia

Hospital outcomes

Hospital episode statistics (HES)

Emergency admissions

Length of stay

mortality

dementia training

neighbourhoods and dementia

The number of people with dementia in the UK is forecast to almost double between 2021 and 2051, to around 1.7 million (p23).[1] People with dementia are more likely to experience a hospitalisation[2, 3] and estimates suggest that at any one time up to 25%, or possibly more, of beds in UK hospitals are occupied by someone with dementia.[4, 5] Numerous studies have reported hospital in-patients with dementia to be at higher risk of inappropriate care and poorer outcomes, including poorer inter-personal care,[6] greater risk of adverse events and complications (e.g. a fall, dehydration, bed sores, infections),[4, 7, 8] longer hospitalisations and higher risk of death or of emergency re-admission.[7, 9, 10]

Against this background the first national strategy for dementia in England was introduced in 2009, with improved quality of care in general hospitals a key objective.[11] This was followed by a series of related initiatives, including the first national audit of dementia (NAD) care in general hospitals in 2010, the introduction in 2013 by Health Education England of progressive increases in provision of foundation-level dementia training for all staff looking after patients with dementia,[12, 13] publication of the Five Year Forward View vision for the NHS[14] and Dementia Roadmap web-resource[15] in 2014, and publication of the Dementia Core Skills Education and Training Framework in 2015 (updated 2018)[16] detailing the essential skills and knowledge necessary across health and social care. Numerous other local and national initiatives to improve care, such as the Butterfly Scheme[17] and John’s Campaign[18] have also become widespread. Yet despite this considerable activity, evaluation has been lacking on whether disparities in hospital outcomes for people with and without dementia have reduced as a result.

An understanding of the true extent of the differences in outcomes for people with dementia (PwD) compared to people without dementia (PwoD) and the factors responsible, is important for the design of effective innovations and interventions. Accurate comparisons are made complex because PwD differ from other in-patients in many ways: they are generally older[19] and may have a greater illness burden upon admission, [20] they may also have different patterns of co-morbidities[20] and be admitted for different conditions.[20–22] Very few studies have controlled for such differences to any appreciable degree. Hence it is not clear to what extent the care received in hospital per se contributes to differences in observed outcomes, leading to calls for greater attention to be paid to this issue.[10]

Aims

We used the NHS-Digital Hospital Episode Statistics (HES) dataset for England.[23] We analysed emergency admissions for physical health conditions, for which PwD are typically looked after on general wards by non-specialist staff and which are the great majority of all admissions. Our aims were:

To describe the epidemiology of emergency hospital admissions for PwD and how this compares to PwoD, across the period 2010 to 2016.

To examine the trends in hospital outcomes for PwD and PwoD across the period and the extent to which differences in outcomes can be accounted for by differences in patient characteristics.

To assess the relationships between NAD-reported hospital initiatives around care for PwD and patient outcomes over this period.

Source

We extracted individual patient data from the admitted patient care (APC) database within HES for the financial years 2010/11, 2012/13 and 2016/17 to coincide with NAD survey years. The APC dataset contains detailed data on all admissions to National Health Service (NHS) hospitals in England.[23] Professional coders enter the data retrospectively using information collected from the medical records. Up to 20 diagnosis fields are used to capture the primary diagnosis for each admission plus other relevant diagnoses, using ICD-10 codes. We obtained diagnosis data for every FY from 2005/6 to 2016/17. Following the method of previous studies,[24] patients were classified as having comorbid dementia if their record included an ICD10 dementia code as secondary diagnosis for that admission, or as a primary or secondary diagnosis for any previous admission at any hospital over the prior 5 FYs. For this purpose we adapted the set of ICD-10 codes used by the NAD survey in 2016/17 (See Box A1).[25] Individuals identified as having dementia at any point in a FY were assumed to have that status for any other admission during that FY.

Inclusion and exclusion criteria

Hospitals

Using the APC database we identified general acute hospitals and excluded other providers such as day, specialist, palliative and rehabilitation units. We also excluded a few atypically small providers and acute hospitals that did not contribute data to all three FYs 2010/11, 2012/13 and 2016/17.

Patients

For each hospital in each FY, we extracted data for people aged 65 or above with at least one EA for a physical health condition. A hospital “spell” was defined as a continuous stay in hospital, made up of one or more finished consultant episodes (FCE) of care. Spells had to start and finish within the same FY (due to data limits) but patients could have multiple spells in a FY. We included spells coded as EAs and excluded elective admissions, day cases and regular day and night attenders. We also excluded spells where dementia itself or another mental health condition was the main speciality or primary diagnosis, or for palliative care.

Outcomes

Outcomes for patients were computed for each complete spell. Length of hospital stay (LoS) was calculated as the number of days between admission and subsequent discharge including all consultant episodes over that period. Long stays were spells of 15 days or more. An Emergency re-admission (ERA) was an admission for an emergency within 30 days of a previous discharge. Death was a death whilst in hospital or within 30 days of discharge. ERA or Death are competing outcomes, so were also analysed in combination. Full details of each outcome are provided in Box 1.

Box 1 Details of the patient outcomes and explanatory factors used in the study

Patient outcomes

Length of hospital stay (LoS). Defined as the number of consecutive days between admission and subsequent discharge, including all consultant episodes during that period.

Long stay. A spell of 15 days or more. Applies to approximately 20% of all spells.

Emergency re-admission (ERA). An admission for an emergency within 30 days of a previous discharge.

Death. Defined as death while in hospital or within 30 days of discharge. Based on a flag in the dataset identifying each patient who met this criterion, derived from Office for National Statistics mortality data.

ERA or death. These can be considered competing outcomes. We therefore also analysed them as a combined outcome.

Explanatory factors

Demographic characteristics: age on admission (in 5-year age-bands); ethnicity (in 10 categories); residential area Index of Multiple Deprivation quintile (IMD 2010 for FY 2010/11; IMD 2015 for FYs 2012/13 and 2016/17); residential area low income quintile (% of the Lower Super Output Area (LSOA) population in receipt of pension credit or income support, expressed as a national quintile).

Spell characteristics: month of admission (in seasonal quarters); admitted on a weekend (yes or no); source of admission (usual residence, other hospital, residential care/care home, other/unknown).

Pre-existing health: The 30 Elixhauser comorbidities, based on the International Classification of Diseases version 10 (ICD-10) codes associated with each secondary diagnosis.^{20, 21} A comorbidity was coded as present if an associated ICD-10 code was recorded for the current admission or any other admission over the present or previous financial year. We excluded codes for dementia in the construction of the comorbidity “other (non-dementia) neurodegenerative disorders”; numbers with AIDS/HIV were very small and were collapsed with lymphoma). Number of emergency admissions in the previous financial year (coded from 0 to 6, and 7 plus).

Admitting condition: the primary reason for admission and type of treatment given in the spell, recorded in the HES as the core Heath Resource Group (HRG) subchapter for the first episode in a spell.²² Subchapters applying to fewer than around 200 PwD were merged with closely related subchapters, or where a related subchapter did not exist combined into a catch-all “other” admissions category.

Number of consultant episodes: the number of different specialist consultants providing an episode of care to the patient during the spell, coded as 1, 2, 3, or 4 plus.

Discharge destination: The place to where the patient was discharged, coded as usual residence, other hospital, residential care/care home, died in hospital, other/unknown. Discharge destination was excluded from models where the outcome was death or death/ERA combined.

Explanatory factors

Reasons for an EA varied by gender and a few medical procedures were gender-specific. We therefore conducted separate analyses for males and females. Using the APC dataset we constructed a wide range of patient demographic, hospital spell and health-related factors. These were used to characterise and compare the patient cohorts and also as covariates in regression analyses. We used a fine-grained categorisation to capture as much association with patient group as possible. Factors were grouped into subsets as summarised below. Full details of each explanatory factor are provided in Box 1.

Demographic characteristics: age on admission (in 5-year age-bands); ethnicity; residential area Index of Multiple Deprivation quintile; residential area low income quintile.

Spell characteristics: month of admission (in seasonal quarters); weekend admission; source of admission (usual residence, other hospital, residential care/care home, other/unknown).

Pre-existing health: comorbidities (30 Elixhauser comorbidities[26, 27]); number of EAs in the previous financial year.

Admitting condition: the primary reason for admission and type of treatment given (based on the first episode core Heath Resource Group (HRG) subchapter).[28]

Number of consultant episodes: number of specialist consultants providing care over the spell.

Discharge destination: The place discharged to (usual residence, other hospital, residential care/care home, died in hospital, other/unknown).

Measures of dementia care

Measures of hospital dementia care were derived from the hospital returns to the NAD Occupational Checklist (OC) surveys in FYs 2010, 2012 and 2016.[29] Data from the surveys was used to construct hospital-level scores for each FY on three indices of dementia care: Training, Governance and Patient-focus. Scores were weighted sums of responses to the OC items, rescaled to range from 0 (no items met) to 4 (all items met). Scores are not directly comparable across years due to changes in the OC items. Non-response from some hospitals resulted in usable samples of 170 in 2010, 175 in 2012 and 172 in 2016. Full details of our methodology along with the OC items and assigned weights are provided in the Additional File supplement B and Tables B2 to B4.

Statistical models

Descriptive statistics were used to characterise and contrast the patient samples.

Patients and spells were clustered within hospitals. Multi-level models (MLMs) such as random effect, fixed effect and mixed-effect models attempt to model this structure and are often recommended for this type of data. This study involved a large number of separate analyses and MLMs frequently failed to converge to a solution due to very large samples and model complexity. An alternative approach consists of analysing the data as a single-level model and utilising cluster-robust estimates of variance to account for inter-correlations between observations within clusters.[30-32] This approach makes fewer assumptions than MLMs and is appropriate when the focus is not on effects at the cluster level[31]. This method consistently converged and was used for our primary analyses. As a sensitivity, we also report results from MLMs where convergence was achieved.

Length of stay demonstrated a highly skewed distribution so was treated within our models as a count variable with a negative binomial distribution to account for overdispersion. Long spells, ERAs, deaths, and combined ERA/death were analysed as binary variables using robust Poisson regression to obtain direct estimates of relative risks.[33, 34]

Analyses focused on the extent to which explanatory factors could account for differences in outcomes between patients with and without dementia. We ran hierarchical regression models, with each model building on the previous by adding a further subset of covariates. The initial model (M1) included dementia status alone; M2 added the subset of demographic factors; M3 hospital spell characteristics; M4 measures of pre-existing health; M5 admitting condition (HRG subchapter); M6 number of consultant episodes; and M7 discharge destination. Due to the very high volume of data and model complexity, we analysed each FY as a separate cross-sectional dataset.

Number of consultant episodes (M6) and discharge destination (M7) can reflect the health care needs of a patient, but might also reflect events in hospital: for example, a hospital-acquired infection may result in an additional consultant episode. In the analysis we make a distinction between these “post-admission” factors and the other “pre-admission” factors.

For each outcome we analysed relationships to the hospital-level staff training, governance and patient focus index scores. These analyses added in main effects for all three indices plus interaction terms with patient group, where the latter estimated the risk of the outcome for PwD relative to PwoD per 1-point change in the index. These analyses adjusted for pre-admission factors only, to avoid over-control should dementia care impact on the post-admission factors. Analyses were conducted between outcomes and care scores within each FY, then to test for a lagged effect of care scores, outcomes in 2016/17 were regressed on care scores in 2012/13.

Sensitivity analyses

Full details of the sensitivity analyses including rationales and statistical models are provided in Additional File 1. Briefly, these included: exclusion of spells of <2 days; exclusion of spells with psychiatric comorbidities; analysis of a 1:1 matched sample of spells; use of MLMs (on a 50% random sample to increase convergence); re-analysis of LoS using a Fine and Grey survival model accounting for death in hospital as a competing risk; [22]; re-analysis of binary outcomes using logistic regression.[35]

The sample consisted of 178 hospitals. Table 1 provides the numbers of eligible patients and EAs for each FY, broken down by dementia status and sex. The total sample consisted of 2,569,007 spells across 1,685,723 male patients, of which 356,992 spells (13.9%) involved PwD, and 3,011,099 spells across 2,028,743 female patients of which 561,349 spells (18.6%) involved PwD. Less than 6% of spells in any FY extended into the following FY and were thereby excluded (similar across patient groups).

Table 1 Summary of analysis samples^a
	Males						Females
	2010/11		2012/13		2016/17		2010/11		2012/13		2016/17
	PwD	PwoD	PwD	PwoD	PwD	PwoD	PwD	PwoD	PwD	PwoD	PwD	PwoD
Patients
Number of patients	49,473	452,865	62,248	480,923	84,856	555,358	92,656	532,866	108,740	549,229	131,589	613,663
% within the FY	9.8	90.2	11.5	88.5	13.3	86.7	14.8	85.2	16.5	83.5	17.7	82.3
Emergency admissions
Number of spells	87,924	659,615	111,846	706,090	157,222	846,310	153,131	756,483	182,956	786,434	225,262	906,833
% within the FY	11.8	88.2	13.7	86.3	15.7	84.3	16.8	83.2	18.9	81.1	19.9	80.1
^aTotals across 178 hospitals common to all three years PwD patients with dementia; PwoD patients without dementia

Sample descriptive statistics

Numbers of patients and spells increased year-on-year for PwD and PwoD of both sexes, though proportionately more for PwD (Table 1).

Descriptives of the hospital spells were similar across the three FYs, so here we report on FY 2016/17 (Table 2); for 2010/11 and 2012/13 see Tables A1 and A2. PwD of both sexes were much more likely to be older. The patient groups had similar large majorities with White ethnicity. They were also similar in residential IMD though PwD were slightly more likely to live in areas in the two lowest income quintiles. Calendar months of admission were similar but PwD were a little more likely to be admitted on a weekend. The great majority of admissions were from a patient’s “usual residence” with only small numbers being admitted from care homes. Most discharges were to a patient’s usual residence, though PwD were more likely to be coded as discharged to a care home. Around 4.5% of both groups were discharged to another hospital.

Both male and female PwD were more likely to have experienced one or more EAs in the previous FY. Mean numbers of comorbidities were similar for both groups, though patients with dementia were more likely to have fluid and electrolyte disorders, additional neurodegenerative disorders and depression, but less likely to have cancer or be obese.

Regarding reasons for admission, most HRG subchapters were represented by less than 2% of patients in either group. Some conditions were more frequently encountered in PwD and others less frequently. This included higher rates of orthopaedic trauma and renal procedures/disorders, but lower rates of cardiac disorders and digestive system disorders. PwD were also more likely to have more than one consultant episode during their spell in hospital.

Table 2 Summary statistics for the FY2016/17 hospital spell sample
	Males				Females
	PwD		PwoD		PwD		PwoD
	n	%	n	%	n	%	n	%
Total number of spells	157,222	15.7	846,310	84.3	225,262	19.9	906,833	80.1
Patient demographics
Age on admission
65-69	7,678	4.9	169,864	20.1	5,639	2.5	146,453	16.1
70-74	14,383	9.1	173,295	20.5	12,251	5.4	155,706	17.2
75-79	26,287	16.7	169,595	20	26,633	11.8	167,804	18.5
80-84	39,959	25.4	157,220	18.6	48,444	21.5	173,062	19.1
85-89	41,290	26.3	112,019	13.2	65,877	29.2	149,195	16.5
90plus	27,625	17.6	64,317	7.6	66,418	29.5	114,613	12.6
Mean (SD)	82.8	7.1	77.4	7.9	85.3	7.0	79.2	8.4
Ethnicity
White	138,301	88.0	742,366	87.7	201,304	89.4	800,216	88.2
Black African	556	0.4	2,394	0.3	463	0.2	2,502	0.3
Black Caribbean	2,223	1.4	6,504	0.8	2,215	1.0	6,665	0.7
Indian	1,969	1.3	13,003	1.5	2,038	0.9	13,240	1.5
Pakistani	1,242	0.8	7,415	0.9	1,059	0.5	7,493	0.8
Bangladeshi	455	0.3	2,099	0.2	343	0.2	1,859	0.2
Black other/mixed	546	0.3	2,162	0.3	532	0.2	2,180	0.2
Asian other/mixed	1,001	0.6	5,676	0.7	855	0.4	5,079	0.6
Other ethnic group/mixed	1,896	1.2	9,233	1.1	2,445	1.1	9,616	1.1
Unknown	9,033	5.7	55,458	6.6	14,008	6.2	57,983	6.4
Residential area IMD quintile (2015)
1 (most deprived)	33,314	21.2	157,829	18.6	47,757	21.2	175,311	19.3
2	33,361	21.2	162,172	19.2	47,121	20.9	177,471	19.6
3	32,490	20.7	175,850	20.8	46,630	20.7	186,761	20.6
4	29,997	19.1	174,308	20.6	43,773	19.4	184,915	20.4
5 (least deprived)	26,802	17	164,738	19.5	38,338	17.0	171,569	18.9
Missing	1,258	0.8	11,413	1.3	1,643	0.7	10,806	1.2
Residential area low income quintile
1 (Highest income)	23,782	15.1	160,884	19	32,081	14.2	159,919	17.6
2	26,750	17	160,757	19	37,522	16.7	165,548	18.3
3	29,676	18.9	160,080	18.9	44,088	19.6	173,059	19.1
4	33,906	21.6	162,843	19.2	49,929	22.2	182,889	20.2
5 (Lowest income)	37,662	24	170,290	20.1	54,198	24.1	192,804	21.3
Missing	5,446	3.5	31,456	3.7	7,444	3.3	32,614	3.6
Spell characteristics
Month of admission
March-May	37,213	23.7	203,777	24.1	52,927	23.5	216,490	23.9
June-August	40,444	25.7	212,012	25.1	57,540	25.5	226,831	25.0
September-November	39,609	25.2	210,016	24.8	56,596	25.1	225,064	24.8
December-February	39,956	25.4	220,505	26.1	58,199	25.8	238,448	26.3
Weekend admission (Sat or Sun)
Yes	42,850	27.3	198,640	23.5	63,061	28.0	217,259	24.0
No	114,372	72.7	647,670	76.5	162,201	72.0	689,574	76.0
Source of admission
Usual residence
Other hospital	144,838	92.1	803,496	94.9	207,217	92.0	863,261	95.2
Residential care/care home	7,508	4.8	37,159	4.4	9,726	4.3	36,694	4.0
Other/unknown	4,334	2.8	2,723	0.3	7,597	3.4	3,892	0.4
Pre-existing health
Elixhauser comorbidities
Congestive heart failure	39,417	25.1	202,291	23.9	51,346	22.8	196,205	21.6
Cardiac arrhythmias	76,040	48.4	353,101	41.7	91,329	40.5	323,307	35.7
Valvular disease	21,232	13.5	128,501	15.2	28,920	12.8	132,025	14.6
Pulmonary circulation disorders	6,328	4.0	49,814	5.9	10,010	4.4	59,111	6.5
Peripheral vascular disorders	20,842	13.3	120,149	14.2	15,813	7.0	71,935	7.9
Hypertension	100,879	64.2	548,018	64.8	148,327	65.8	595,216	65.6
Paralysis	9,420	6.0	32,274	3.8	9,750	4.3	28,989	3.2
Other (non-dementia) neurodegenerative disorders	42,281	26.9	84,884	10.0	41,195	18.3	78,402	8.6
Chronic pulmonary disease	45,256	28.8	275,439	32.5	57,954	25.7	296,598	32.7
Diabetes, uncomplicated	47,901	30.5	241,537	28.5	53,166	23.6	206,816	22.8
Diabetes, complicated	8,481	5.4	42,046	5.0	6,783	3.0	28,480	3.1
Hypothyroidism	12,104	7.7	50,595	6.0	40,355	17.9	151,341	16.7
Renal failure	47,385	30.1	199,247	23.5	61,871	27.5	191,114	21.1
Liver disease	5,837	3.7	43,384	5.1	5,538	2.5	36,790	4.1
Peptic ulcer disease excluding bleeding	3,993	2.5	23,668	2.8	3,155	1.4	18,250	2.0
Lymphoma + AIDS/HIV	1,681	1.1	20,301	2.4	1,603	0.7	15,512	1.7
Metastatic cancer	4,942	3.1	69,534	8.2	3,557	1.6	54,727	6.0
Solid tumour without metastasis	22,882	14.6	164,637	19.5	12,564	5.6	98,556	10.9
Rheumatoid arthritis/collagen vascular diseases	7,094	4.5	45,765	5.4	17,016	7.6	91,989	10.1
Coagulopathy	3,642	2.3	22,810	2.7	3,529	1.6	17,849	2.0
Obesity	5,631	3.6	58,297	6.9	7,432	3.3	63,969	7.1
Weight loss	8,307	5.3	45,116	5.3	9,734	4.3	43,413	4.8
Fluid and electrolyte disorders	59,493	37.8	205,671	24.3	85,691	38.0	236,530	26.1
Blood loss anaemia	742	0.5	3,939	0.5	765	0.3	3,431	0.4
Deficiency anaemia	18,675	11.9	86,761	10.3	28,776	12.8	102,924	11.3
Alcohol abuse	10,056	6.4	56,409	6.7	4,922	2.2	22,305	2.5
Drug abuse	308	0.2	2,289	0.3	416	0.2	1,353	0.1
Psychoses	3,679	2.3	7,557	0.9	5,665	2.5	9,456	1.0
Depression	22,030	14.0	63,272	7.5	40,353	17.9	103,427	11.4
Number of Elixhauser comorbidities
Mean (SD)	4.2	2.4	3.8	2.5	3.8	2.3	3.6	2.4
No. of emergency admissions in previous FY
None	71,714	45.6	555,413	65.6	110,567	49.1	597,795	65.9
1	36,825	23.4	150,124	17.7	54,316	24.1	165,213	18.2
2	20,423	13.0	65,443	7.7	27,982	12.4	69,956	7.7
3	11,255	7.2	31,656	3.7	14,139	6.3	33,082	3.6
4	6,486	4.1	16,433	1.9	7,795	3.5	16,927	1.9
5	3,628	2.3	9,855	1.2	4,143	1.8	9,425	1.0
6	2,265	1.4	5,432	0.6	2,272	1.0	5,065	0.6
7+	4,626	2.9	11,954	1.4	4,048	1.8	9,370	1.0
Admitting condition type and severity
Spell Core HRG subchapter
AA+AB Nervous System Procedures and Disorders/Pain management	11,082	7.0	54,298	6.4	15,682	7.0	58,462	6.4
BZ Eyes and Periorbita Procedures and Disorders	484	0.3	3,557	0.4	809	0.4	4,023	0.4
CZ Mouth Head Neck and Ears Procedures and Disorders	4,593	2.9	21,590	2.6	7,956	3.5	24,062	2.7
DZ Respiratory System Procedures and Disorders	32,857	20.9	150,957	17.8	40,678	18.1	157,863	17.4
EA Cardiac Procedures	1,282	0.8	28,628	3.4	925	0.4	17,212	1.9
EB Cardiac Disorders	14,213	9.0	116,120	13.7	18,981	8.4	118,334	13.0
FZ Digestive System Procedures and Disorders	10,913	6.9	86,077	10.2	16,420	7.3	101,275	11.2
GA+GB Hepatobiliary and Pancreatic System Open, Laparoscopic, Endoscopic Procedures	501	0.3	6,933	0.8	568	0.3	6,800	0.7
GC Hepatobiliary and Pancreatic System Disorders	1,372	0.9	16,810	2.0	1,893	0.8	16,572	1.8
HA Orthopaedic Trauma Procedures	7,374	4.7	26,107	3.1	21,098	9.4	59,261	6.5
HB+HR Orthopaedic Non-Trauma + Reconstruction Procedures	1,055	0.7	7,901	0.9	1,732	0.8	10,079	1.1
HC Spinal Procedures and Disorders	1,097	0.7	9,177	1.1	2,158	1.0	13,557	1.5
HD Musculoskeletal and Rheumatological Disorders	2,891	1.8	21,271	2.5	6,608	2.9	31,961	3.5
JA+JB+JC Breast/Burn/Skin Procedures and Disorders	651	0.4	6,111	0.7	1,174	0.5	7,097	0.8
JD Skin Disorders	2,591	1.6	18,477	2.2	4,583	2.0	21,230	2.3
KA+KC Endocrine System/Metabolic Disorders	1,790	1.1	9,919	1.2	3,563	1.6	13,899	1.5
KB Diabetic Medicine	1,599	1.0	5,808	0.7	2,060	0.9	4,883	0.5
LA Renal Procedures and Disorders	17,510	11.1	47,142	5.6	24,121	10.7	50,194	5.5
LB Urological and Male Reproductive System Procedures and Disorders	7,855	5.0	45,462	5.4	1,622	0.7	8,675	1.0
MA+MB Female Reproductive System Procedures/Disorders	NA	NA	NA	NA	463	0.2	3,697	0.4
QZ Vascular Procedures and Disorders	1,353	0.9	15,647	1.8	2,067	0.9	12,789	1.4
SA Haematological Procedures and Disorders	1,332	0.8	15,469	1.8	2,067	0.9	15,149	1.7
VA Multiple Trauma	1,759	1.1	6,738	0.8	3,820	1.7	10,837	1.2
WA_A Immunology, Infectious Diseases, Poisoning	4,472	2.8	19,873	2.3	5,584	2.5	19,807	2.2
WA_B Complications of procedures	331	0.2	5,953	0.7	288	0.1	5,003	0.6
WA_D Unexplained symptoms or abnormal findings	3,602	2.3	8,531	1.0	5,204	2.3	10,299	1.1
RC+WA_C+WA_E+ZZ Miscellaneous	8,562	5.4	22,231	2.6	13,398	5.9	25,631	2.8
missing or N/A “Not applicable”	14,101	9.0	69,523	8.2	19,740	8.8	78,182	8.6
Post-admission factors
Number of consultant episodes in spell
1	70,029	44.5	463,175	54.7	104,635	46.5	496,509	54.8
2	55,680	35.4	248,449	29.4	78,466	34.8	267,575	29.5
3	20,691	13.2	85,012	10	28,501	12.7	92,634	10.2
4 or more	10,822	6.9	49,674	5.9	13,660	6.1	50,115	5.5
Discharge destination
Usual residence	119,024	75.7	732,906	86.6	171,005	75.9	778,821	85.9
Other hospital	7,363	4.7	37,582	4.4	9,760	4.3	39,990	4.4
Residential care/care home	13,022	8.3	15,239	1.8	21,260	9.4	23,608	2.6
Died	14,364	9.1	50,493	6	17,456	7.7	49,037	5.4
Other/unknown/NA	3,449	2.2	10,090	1.2	5,781	2.6	15,377	1.7

Descriptive statistics on outcomes

In all FYs mean LoS was longer for PwD of both sexes (Table 3; Figure 1a). Mean LoS declined slightly over time for both groups and in 2016/17 was 9.7 versus 6.2 days for males, and 9.6 versus 6.7 days for females. Rates of spells >=15 days had a very similar pattern (Figure 1b). In all FYs PwD were more likely to experience an ERA and males more-so than females. Rates of ERAs were similar in 2010 and 2012, but increased a little in 2016/17, though slightly less for PwD (Figure 1c). By contrast, rates of deaths in hospital or within 30 days of discharge demonstrated a small reduction year-on-year for both patient groups and both sexes (Figure 1d), though remained substantially higher for PwD. The increase in ERAs and decrease in deaths essentially cancelled each other out in the combined outcome, leaving it stable over time, although higher throughout for PwD (Figure A1a).

Table 3 Summary of samples and hospital outcomes
	Males						Females
	2010/11		2012/13		2016/17		2010/11		2012/13		2016/17
	PwD	PwoD	PwD	PwoD	PwD	PwoD	PwD	PwoD	PwD	PwoD	PwD	PwoD
Length of stay (days)
Number of spells^a	87,924	659,615	111,846	706,090	157,222	846,310	153,131	756,483	182,956	786,434	225,262	906,833
Mean	10.6	7.0	10.0	6.7	9.7	6.2	10.6	7.8	10.0	7.3	9.6	6.7
SD	15.7	11.3	14.3	10.5	14.9	10.5	15.4	12.0	13.9	11.1	14.5	10.9
Median	5	3	5	3	4	3	5	4	5	3	4	3
IQR	1, 13	1, 8	1, 13	1, 8	1, 12	1, 7	1, 14	1, 10	1, 13	1, 9	1, 12	1, 8
Maximum	211	314	231	265	304	292	264	336	227	348	272	282
Length of stay>=15 days
Number of spells^a	87,924	659,615	111,846	706,090	157,222	846,310	153,131	756,483	182,956	786,434	225,262	906,833
Number with the outcome	20,215	88,022	24,537	88,614	31,747	96,935	36,309	119,201	41,223	115,340	46,691	118,250
%	23.0	13.3	21.9	12.5	20.2	11.5	23.7	15.8	22.5	14.7	20.7	13.0
Emergency re-admission within 30 days
Number of spells^ab	71,885	558,879	92,756	603,625	132,585	729,461	127,605	643,531	154,340	674,366	192,624	785,113
Number with the outcome	18,493	100,191	23,246	105,542	35,272	148,432	28,142	104,322	33,764	107,822	43,948	143,682
%	25.7	17.9	25.1	17.5	26.6	20.3	22.1	16.2	21.9	16.0	22.8	18.3
Death within 30 days
Number of spells^ab	85,944	644,771	111,131	701,059	156,713	841,535	149,619	733,083	181,882	779,772	224,575	900,149
Number with the outcome	14,010	62,314	17,267	64,662	21,382	66,526	21,788	65,550	24,731	66,457	27,774	63,933
%	16.3	9.7	15.5	9.2	13.6	7.9	14.6	8.9	13.6	8.5	12.4	7.1
Emergency re-admission or death within 30 days
Number of spells^ab	80,941	597,325	104,765	650,638	146,996	776,835	140,456	677,786	170,700	722,053	210,144	829,239
Number with the outcome	32,471	162,414	40,501	170,151	56,635	214,833	49,905	169,793	58,477	174,215	71,702	207,551
%	40.1	27.2	38.7	26.2	38.5	27.7	35.5	25.1	34.3	24.1	34.1	25.0
^aExcludes spells without a discharge date ^bNumber of spells varied by outcome depending upon extent of follow-up post-discharge and availability of a mortality flag

Hierarchical regression analysis

Findings of the hierarchical regression analyses controlling for demographic, health and other differences between the patient groups are summarised in Table 4. Results - in the form of relative risks (RRs) for males and females by financial year - are given for selected models: (a) with no control for covariates (model M1); (b) with control for all pre-admission factors (M5); (c) with further control for the post-admission factors (M6 or M7). Table A3 provides results for all seven steps in the hierarchical analysis. Figures 2 and A1 graph the results for all seven steps, expressed in the form of the “excess” risk for patients with dementia compared to those without.

Length of stay

In unadjusted analysis, male PwD experienced spells 52% to 55% longer on average than male PwoD in all FYs. Mean spells for female PwD were 36% to 37% longer in 2010 and 2012, but 43% longer in 2016 (Table 4; Figures 2a and 2b). However, these differences diminished with increasing adjustment for covariates. After inclusion of all pre-admission covariates, the excess mean length of spell was 29% to 35% for males and 19% to 25% for females in all FYs. This reduced further with inclusion of the post-admission covariates, to around 15% for males and 10% for females in all FYs.

Unadjusted rates of long spells (>=15 days) for PwD were more than 70% higher for males and 50%-60% higher for females in all FYs (Table 4; Figures 2c and 2d). Differences reduced substantially upon adjustment for patient demographics, then more slowly with the addition of further covariates. With all pre-admission covariates included, males were 33% to 37% more likely to have a long stay, and females 20% to 25% more likely. Inclusion of post-admission covariates resulted in further reductions to around 20% for males and around 15% for females in all FYs.

Table 4 Summary of group comparisons from hierarchical regression analysis
	Males			Females
	2010/11	2012/13	2016/17	2010/11	2012/13	2016/17
	RR (95% CI)	RR (95% CI)	RR (95% CI)	RR (95% CI)	RR (95% CI)	RR (95% CI)
Length of stay (count in days)
Uncontrolled (M1)	1.52 (1.49-1.55)	1.50 (1.47-1.53)	1.55 (1.52-1.58)	1.37 (1.34-1.40)	1.36 (1.33-1.39)	1.43 (1.40-1.46)
Adjusted for pre-admission factors (M5)	1.31 (1.29-1.33)	1.29 (1.27-1.32)	1.35 (1.33-1.38)	1.20 (1.18-1.22)	1.19 (1.17-1.21)	1.26 (1.23-1.28)
Adjusted for pre- and post-admission factors (M7)	1.15 (1.12-1.17)	1.13 (1.12-1.15)	1.17 (1.15-1.18)	1.09 (1.07-1.11)	1.09 (1.07-1.10)	1.12 (1.10-1.14)
Length of stay>=15 days
Uncontrolled (M1)	1.72 (1.68-1.76)	1.75 (1.70-1.79)	1.76 (1.72-1.80)	1.50 (1.47-1.54)	1.54 (1.50-1.57)	1.59 (1.55-1.63)
Adjusted for pre-admission factors (M5)	1.33 (1.30-1.35)	1.35 (1.32-1.38)	1.38 (1.36-1.41)	1.20 (1.18-1.22)	1.22 (1.20-1.25)	1.26 (1.23-1.28)
Adjusted for pre- and post-admission factors (M7)	1.19 (1.17-1.22)	1.20 (1.18-1.23)	1.22 (1.19-1.24)	1.13 (1.11-1.15)	1.15 (1.13-1.17)	1.16 (1.14-1.18)
Emergency re-admission within 30 days
Uncontrolled (M1)	1.44 (1.40-1.47)	1.43 (1.41-1.46)	1.31 (1.28-1.33)	1.36 (1.34-1.39)	1.37 (1.34-1.40)	1.25 (1.23-1.27)
Adjusted for pre-admission factors (M5)	1.27 (1.25-1.30)	1.24 (1.22-1.26)	1.16 (1.14-1.17)	1.26 (1.24-1.28)	1.24 (1.22-1.26)	1.16 (1.14-1.17)
Adjusted for pre- and post-admission factors (M7)	1.29 (1.26-1.31)	1.25 (1.23-1.27)	1.17 (1.16-1.19)	1.28 (1.26-1.30)	1.25 (1.23-1.27)	1.17 (1.16-1.19)
Death within 30 days
Uncontrolled (M1)	1.69 (1.65-1.72)	1.68 (1.65-1.72)	1.73 (1.69-1.76)	1.63 (1.60-1.66)	1.60 (1.56-1.63)	1.74 (1.71-1.77)
Adjusted for pre-admission factors (M5)	1.34 (1.32-1.37)	1.34 (1.31-1.37)	1.33 (1.31-1.36)	1.33 (1.30-1.36)	1.30 (1.27-1.32)	1.37 (1.34-1.39)
Adjusted for pre- and post-admission factors (M6)	1.33 (1.31-1.36)	1.32 (1.30-1.35)	1.30 (1.28-1.33)	1.32 (1.30-1.35)	1.29 (1.27-1.32)	1.35 (1.33-1.38)
Emergency re-admission or death within 30 days
Uncontrolled (M1)	1.48 (1.46-1.50)	1.48 (1.46-1.50)	1.39 (1.38-1.41)	1.42 (1.40-1.43)	1.42 (1.40-1.44)	1.36 (1.35-1.38)
Adjusted for pre-admission factors (M5)	1.28 (1.26-1.30)	1.26 (1.25-1.28)	1.21 (1.20-1.22)	1.27 (1.25-1.28)	1.25 (1.24-1.27)	1.22 (1.21-1.23)
Adjusted for pre- and post-admission factors (M6)	1.28 (1.26-1.29)	1.26 (1.24-1.27)	1.20 (1.19-1.22)	1.27 (1.25-1.28)	1.25 (1.24-1.26)	1.22 (1.21-1.23)
RR = Relative Risk (strictly, the Incident Rate Ratio)

Emergency re-admissions

Unadjusted rates of ERAs within 30 days were 43% to 44% higher for male PwD and 36% to 37% higher for female PwD in 2010/11 and 2012/13 respectively, but in 2016/17 just 31% and 25% higher. In all three FYs the group difference in ERA rates decreased as more pre-admission covariates were added, but changed only a little with inclusion of post-admission factors, while remaining lower for 2016/17 throughout (Figures 2e and 2f). In the fully adjusted models for 2016/17 the ERA rate was just 17% higher for both males and females with dementia. Although this represents a reduced group difference compared to earlier FYs, this was driven by an increase in ERAs for PwoD, rather than a reduction for PwD (Table 3).

Deaths within 30 days

Unadjusted rates of deaths were 68% to 73% higher for male PwD in all three FYs. This reduced to 33% to 34% after adjustment for patient demographics but changed only slightly as additional covariates were added (Figure 2g). The pattern for female PwD was similar except that excess deaths were somewhat higher in 2016/17, with a fully adjusted excess rate in 2016/17 of 35% (Table 4; Figure 2h).

Emergency readmission or death

When analysed as a combined outcome, excess rates of ERAs+deaths for PwD were a little lower in 2016/17 compared to previous FYs for both sexes. Rates showed a sharp decline for both sexes in all three FYs after adjustment for patient demographics, followed by a more gradual decline with increasing covariate control (Figures A1b and A1c).

Relationships between patient outcomes and hospital dementia care scores

Hospital scores on the training, governance and patient-focus dementia care indices demonstrated high variability, indicating an ability to differentiate between hospitals. Mean scores on all three indices increased over time, implying improved provision for PwD (Table A5), while correlations between indices were small to moderate (Table A6). Little was found in the way of statistically significant relationships between patient outcomes and care scores, either within each FY or in the lagged analysis (Table 5; Table A7). Although a scattering of significant main effects and interactions were observed this was to be expected given the large number of comparisons made, and RRs were small – nearly all in the range 0.98 to 1.02 – with little consistency in direction of effect or across genders or FYs.

Table 5 Relationships between patient outcomes 2016/17 and hospital dementia care scores adjusted for pre-admission covariates
		Males			Females
		Training	Governance	Patient-focus	Training		Governance	Patient-focus
		RR^a (95% CI)	RR^a (95% CI)	RR^a (95% CI)	RR^a (95% CI)		RR^a (95% CI)	RR^a (95% CI)
Length of stay (days)
2016/17	Main effect^b	*1.03 (**1.00-1.05)	0.99 (0.97-1.01)	1.01 (0.97-1.06)	*1.03 (**1.00-1.05)		0.99 (0.96-1.01)	1.02 (0.98-1.07)
	Interaction^c	1.01 (0.99-1.04)	1.01 (0.99-1.04)	0.99 (0.95-1.04)	1.02 (0.99-1.04)		1.02 (1.00-1.05)	0.97 (0.92-1.01)
2016/17 with 2012/13^d	Main effect	1.02 (0.99-1.04)	1.00 (0.98-1.02)	0.99 (0.94-1.03)	1.02 (0.99-1.05)		1.00 (0.98-1.02)	0.99 (0.94-1.04)
	Interaction	0.97 (0.93-1.00)	1.01 (0.98-1.03)	1.02 (0.96-1.07)	0.97 (0.93-1.01)		1.01 (0.98-1.03)	1.01 (0.96-1.06)
Length of stay>=15 days
2016/17	Main effect	1.04 (**1.01-1.07)	0.98 (0.961.01)	1.03 (0.97-1.09)	*1.03 (**1.00-1.07)	0.98 (0.95-1.01)		1.04 (0.99-1.10)
	Interaction	1.01 (0.98-1.04)	1.02 (0.99-1.04)	0.98 (0.93-1.04)	1.02 (0.99-1.04)	*1.02 (**1.00-1.05)		0.94* (0.89-1.00)
2016/17 with 2012/13	Main effect	1.01 (0.97-1.04)	0.99 (0.97-1.02)	1.01 (0.96-1.07)	1.01 (0.97-1.05)	0.99 (0.97-1.02)		1.02 (0.96-1.08)
	Interaction	0.98 (0.95-1.02)	1.00 (0.98-1.03)	1.02 (0.98-1.07)	0.98 (0.95-1.02)	1.00 (0.98-1.03)		1.02 (0.97-1.06)
Emergency re-admission within 30 days
2016/17	Main effect	0.99 (0.97-1.01)	1.00 (0.98-1.01)	1.03 (0.99-1.08)	0.99 (0.97-1.01)	1.00 (0.98-1.01)		*1.04 (**1.00-1.08)
	Interaction	0.99 (0.98-1.01)	0.99 (0.97-1.01)	0.99 (0.96-1.03)	1.00 (0.99-1.02)	0.99 (0.98-1.01)		0.96 (**0.94-0.99)
2016/17 with 2012/13	Main effect	0.99 (0.96-1.01)	1.00 (0.99-1.02)	1.00 (0.96-1.03)	0.98 (0.95-1.00)	1.01 (0.99-1.02)		1.00 (0.97-1.04)
	Interaction	1.00 (0.98-1.03))	0.99 (0.98-1.00)	0.99 (0.95-1.03)	1.01 (0.99-1.03)	0.99 (0.98-1.00)		*0.97 (**0.94-1.00)
Death within 30 days
2016/17	Main effect	0.99 (0.97-1.02)	1.00 (0.97-1.02)	1.01 (0.96-1.05)	0.99 (0.97-1.02)	0.99 (0.96-1.01)		1.00 (0.95-1.06)
	Interaction	1.01 (0.98-1.04)	1.00 (0.98-1.02)	1.02 (0.98-1.05)	1.00 (0.98-1.03)	*1.02 (**1.00-1.04)		1.03 (0.991.06)
2016/17 with 2012/13	Main effect	1.00 (0.97-1.02)	1.00 (0.98-1.02)	1.02 (0.98-1.07)	1.00 (0.97-1.03)	0.99 (0.97-1.01)		1.03 (0.98-1.09)
	Interaction	0.98 (0.95-1.00)	1.00 (0.98-1.02)	1.03 (0.98-1.08)	0.98 (0.96-1.00)	1.02 (1.00-1.03)		1.02 (0.98-1.06)
Emergency re-admission or death within 30 days
2016/17	Main effect	0.99 (0.97-1.00)	1.00 (0.98-1.01)	1.03 (1.00-1.05)	0.99 (0.98-1.01)	0.99 (0.98-1.01)		*1.03 (**1.00-1.06)
	Interaction	1.00 (0.99-1.01)	0.99 (0.98-1.00)	1.00 (0.98-1.02)	1.00 (0.99-1.01)	1.01 (1.00-1.01)		0.98 (0.961.00)
2016/17 with 2012/13	Main effect	0.99 (0.97-1.01)	1.00 (0.99-1.01)	1.00 (0.98-1.03)	0.98 (0.96-1.00)	1.00 (0.99-1.01)		1.01 (0.98-1.03)
	Interaction	1.00 (0.98-1.01)	1.00 (0.99-1.01)	1.01 (0.98-1.03)	1.00 (0.99-1.02)	1.00 (0.99-1.01)		0.99 (0.97-1.01)
^aRR = Relative risks estimate the change in rate of the outcome for a 1 point change in dementia care score, controlled for all pre-admission factors ^bMain effects estimate the increase in risk of the outcome in PwoD for a 1-point change in the dementia care score ^cInteractions estimate the degree to which the risk is modified in PwD ^dOutcomes in 2016/17 regressed on dementia care scores in 2012/13 p<0.05; *p<0.01

Sensitivity analysis

Findings of the sensitivity analyses were very similar across FYs so only 2016 results are reported here. Excluding hospital spells of <2 days, the fully adjusted excess length of stay for PwD increased slightly to 23% for males and 17% for females; adjusted results for other outcomes hardly changed (Table A4; Figures 3 and A1d). Exclusion of spells involving a psychiatric comorbidity increased the group difference for all outcomes, though by only a few percentage points. Results using the matched patient sample closely corresponded to the primary analysis with the exception of excess ERAs, which increased to 25% for males and 21% for females under full covariate adjustment. Use of a logistic model to analysis the binary outcomes made no appreciable difference. Use of a random-effects model produced substantially lower estimates of excess LoS, to just 9% for male PwD and 6% for female PwD under full adjustment; results from random-effects models for other outcomes closely matched the primary analysis. Sub-hazard ratios from the competing risks survival analysis estimated that after full adjustment males with dementia were 11% less likely to be discharged and females 8% less, at any point in time.

Analyses of relationships between hospital dementia care indices and patient group differences using MLMs did not substantively alter the results (Table A8).

INSERT FIG 3 HERE.

In unadjusted analysis PwD of both sexes experienced longer LoS and higher rates of ERAs and mortality across the 6-year period. However, most of the apparent disparity in outcomes could be accounted for by group differences in patient and spell characteristics. After full adjustment, LoS in 2016/17 was on average 17% longer for male PwD and 12% longer for female PwD, equivalent to an average of 1.0 and 0.8 additional days in hospital, with similar or slightly lower rates in earlier FYs. Estimated group differences in LoS were smaller still under random-effects and competing risks survival models. However, these results may include some degree of over-adjustment stemming from the adjustment for post-admission numbers of consultant episodes and discharge destination: for many patients these factors represent care for pre-existing health needs, but for some may reflect a potentially avoidable event in hospital, such as a fall or infection,[7, 36] for which additional bed days would ideally not be adjusted out.

The adjusted excess risk an ERA in 2016/17 was 17% for PwD of both sexes, with post-admission factors making little difference. These rates were reduced compared to previous FYs, though interpretation is not straight-forward since absolute ERA rates increased in 2016/17 for both groups, except less-so for PwD. PwD had a notably higher risk of dying in hospital or shortly after discharge, which after adjustment ranged close to 30% across the study period for both sexes.

On the whole, adjusted patient group differences in LoS and ERAs were small at the end of the study period, if not before, while PwD experienced a consistently higher risk of mortality throughout. Attributing the remaining group differences to the care received in hospital is questionable however. Although we controlled for many covariates, multiple factors outside of hospital control remain that we could not quantify, but which might explain most – if not all – of the remaining outcome differences. PwD are often more vulnerable upon admission, for example presenting with malnourishment,[7] delirium[37, 38] and frailty; conditions only partially captured in our analysis that increase the risk of poorer outcomes.[9, 37, 39-41] They are also less able to adapt to the hospital environment itself, resulting in significant distress and deterioration.[42, 43] Difficulties in identifying the terminal phase of dementia make PwD much more likely to be admitted rather than offered alternatives such as referral to palliative care services.[43, 44] Death or an ERA shortly after discharge may have more to do with how a patient’s dementia impacts on their post-discharge recovery, than with the care received whilst in hospital. An excessive LoS may be largely outside of hospital control:[45, 46] PwD discharged to a care home have shorter stays in areas with a greater concentration of care home beds, and stays are longer in localities with higher rates of delayed discharges for which social services are responsible.[47]

No outcome exhibited much association with our NAD-derived hospital indices of dementia care. However, the reliability of the OC data is unknown and the survey items mainly tap hospital policies rather than the actual care delivered to patients. Nonetheless, there is little doubt that hospitals, charities and other organisations put considerable effort and resources into introducing initiatives and improving staff training. It seems this was either not accurately captured in the OC returns or had little impact on our study outcomes. It is also possible that the high-level outcomes of LoS, ERA and mortality are simply not responsive enough to the changes that have been implemented, subject as these outcomes are to a multiplicity of political, policy and local influences besides the care delivered to individual patients. Another possibility is that the 6-year period under study was too short for the changes to become sufficiently embedded to impact on these outcomes.[48] It is also worth noting that many initiatives have sought to improve the patient and carer hospital experience and reduce potential distress and discomfort, benefits that may not feed through into clinical outcomes. This is an important aspect of care not captured within the APC and hence not available to our study.

Implications for policy and practice

Both policy and practice should give less emphasis to reducing LoS and ERAs for PwD, for which covariate adjusted disparities were small, and focus more on achieving appropriate, rather than short, lengths of stay. Higher hospital mortality rates for PwD may be more effectively addressed by keeping PwD out of hospital, particularly by identifying individuals at terminal stages before admission, with a shift of emphasis to advance care planning,[49] admission avoidance,[50] and referral to palliative care services. This may require some relocation of resources into the community. Hospital staff could facilitate earlier discharge back to home through specific training in recognising the terminal phase of the condition and on end-of-life care in dementia.

Implications for research

Better understanding is needed of the higher hospital mortality rates for PwD, especially the roles played by pre- and post-admission factors such as unrecognised terminal illness and post-discharge community support. Research is also needed into how best to use hospital avoidance approaches and advance care planning to minimise unnecessary hospital admissions. Comparison of hospital outcomes between PwD and PwoD require comprehensive case-mix adjustment to produce accurate results and avoid mis-interpretation. Better alternatives to LoS, ERAs and mortality as metrics for outcomes of care are needed. Our analysis needs to be repeated using more recent data.

Comparison to previous literature

Mollers et al in a systematic review found that 52 out of 60 observational studies reported a longer hospitalisation time for people with dementia, with mean stays longer by up to 22 days. However, very few studies controlled for confounding factors.[10] An integrative review also reported LOS to be typically longer for PwD, though not in every study, while in the same review 8 out of 11 studies found PwD to be at higher risk of mortality in hospital, though this varied by various factors including age, presence of delirium, and disease condition.[7] Another more recent systematic review drew attention to wide methodological heterogeneity between studies but concluded that the majority report poorer outcomes for PwD, though the authors refrained from quantifying the differences.[38] A systematic review of cohort and case-control studies of ERA rates concluded that any increased rate due to dementia per se was modest at best, in the range 3% to 13%,[51] close to our present estimate of 17%.

Few UK-based studies have controlled for confounders. The UK Care Quality Commission (CQC) analysed national APC data on EAs from 2008 to 2012, using patient samples matched on age, gender, primary diagnosis and co-morbidities.[24] PwD had higher rates for all outcomes across the period though the gap decreased over time: excess LoS declined from 32% to 26%; ERAs from 21% to 18% and risk of death from 49% to 36%. Rates in the present study adjusted for pre-admission factors are comparable but suggest a slowing, or even reversal, in trend.

Fogg et al analysed 21,399 incident EAs between 2014 and 2017 at one large acute English hospital, with patients screened for cognitive impairment upon admission.[22] Covariates included age, gender, primary diagnosis, Charlson co-morbidity index, illness severity, malnutrition, admission route and discharge specialty. Survival analysis treating death as a competing risk estimated that at any given time PwD were 20% (95%CI 17% to 24%) less likely to be discharged. PwD also had an increased risk of readmission within 30 days (OR=1.21; 95% CI 1.04 to 1.40) and of death (OR=1.66; 95%CI 1.48 to 1.86). An investigation of incident EAs (n=6724), at one large Scottish district general hospital in 2012 and 2013 [38, 52], compared four “cognitive spectrum disorder” (CSD) groups (dementia, delirium, both, unspecified) to non-CSD patients. Adjusted for baseline demographic and comorbidity covariates, LoS for PwD was 73% (95%CI 54% to 94%) longer than for those without CSD and risk of re-admission was 33% (16% to 52%) higher, but the risk of death within three months of admission was similar (HR=1.04; 95%CI 0.84 to 1.29). Effect sizes vary considerably between these two single-site studies, and being based on incident admissions rather than individual spells, cannot be directly compared to our own.

Strengths and weaknesses

Our patient samples were many times larger than any previous study and we included all English acute hospitals of any substantial size for three different financial years. We also took account of a wider number and range of confounding factors than any previous study. We analysed individual hospital spells (also called prevalent admissions) to facilitate estimation of effect sizes adjusted for spell-specific covariates. We deemed this more appropriate to our objectives than an incident admission approach, in which sequential spells are combined including time outside hospital between spells and adjustment is limited to initial-spell covariates.[22, 52] We focused on patients with dementia as this was the central policy issue over the period. Similar issues pertain to patient groups with other cognitive impairments, most notably delirium, and broadly similar patterns of outcomes reported.[38, 52] Delirium is greatly under-recorded in the APC, with one UK study finding that routine coding detected only around 10% of the delirium cases diagnosed via a gold standard clinical assessment.[53] Our findings were mostly highly stable across FYs and a variety of sensitivity analyses. They were also very similar for males and females analysed separately.

Our findings are dependent upon the reliability and completeness of the hospital record, which can be variable.[54] Dementia status was defined by the presence of a related code in the hospital record during the previous five years. Compared against a large mental health registry, APC dementia diagnoses have been found reasonably accurate, with sensitivity of 78% and specificity of 92% over the span of a hospital record.[55] Sensitivity analysis excluding spells involving comorbidities likely to be confused with dementia, produced mostly only small increases in the relative risk estimates. However, we lacked information on stage of dementia. Hospital admission and discharge dates (used to calculate length of stay and ERAs) are expected to be accurate and deaths were based on Office for National Statistics mortality data. We suspect that the coding of discharge destination may under-represent discharges to residential homes.

We adjusted for many covariates but were limited by what was available in the hospital record. Non-mandatory comorbidities are only recorded if deemed clinically relevant and some HRG subchapters are very broad, implying a potential for uncaptured confounding that might further account for group differences. In view of model complexity and computational overheads we analysed each FY separately and performed no direct tests for trends or differences across FYs. Sample sizes were such that most RRs were estimated with a 95% CI of plus/minus 2% or 3%, hence even small differences would reach statistical significance.

The accuracy of Hospital Trust responses to the NAD OC is unknown and question items changed over time making it not possible to construct fully consistent care indices across FYs. The item weights used for computing scores were subjective; however, the lack of associations to outcomes makes it unlikely that alternative weighting schemes would have changed the results.

Over the six-year period, covariate-adjusted hospital LoS, ERA rates and mortality rates for PwD showed little beneficial change relative to similar patients without dementia, despite the many national and local efforts to improve hospital care for this patient group. Group differences after covariate adjustment were mostly small and may reflect factors beyond hospital control. Greater emphasis is needed on avoiding unnecessary hospitalisations for PwD.

APC Admitted Patient Care

CQC Care Quality Commission

EA Emergency admission

ERA Emergency readmission

FCE Finished consultant Episode

FY Financial year

HES Hospital Episode Statistics

HRG Health Resource Group

IRR Incident rate ratio

LoS Length of stay

MLM Multi-level model

NAD National Audit of Dementia

NHS National Health Service

OC Occupational Checklist

PwD People living with dementia

PwoD People without dementia

RR Relative Risk

SHR Sub-hazard ratio

T2DM Type 2 Diabetes Mellitus

Ethics approval and consent to participate

Although specific Ethics approval was not required for the secondary analysis of anonymised administrative data reported in this article, the wider Work Programme 5, within which this study sits, was reviewed and approved by the Faculty of Health and Medicine Research Ethics Committee (FHMREC 17056) at the University of Lancaster and the Health Research Authority (Integrated Research Approval System (IRAS) ID 242166: REC reference 18/HRA/1198). As the study uses routinely collected data which has been pseudonymised for the purposes of research, informed consent was not required. Approval for release of secondary care and ONS data through the NHS Digital Data Access Request Service (DARS) was given by the Independent Group Advising on the Release of Data (IGARD) on 6/4/2017 (ref NIC 33318).

Consent for publication

Not applicable

Availability of data and materials

The Hospital Episode Statistics (HES) and ONS data that support the findings of this study are available from NHS-Digital (https://digital.nhs.uk/data-and-information/data-tools-and-services/data-services/hospital-episode-statistics), but restrictions apply to the availability of these data, which were used following approvals and data sharing agreements for the current study, and so are not publicly available. We are not permitted to share this data. The National Audit of Dementia Occupational Checklist data that were used are available from the Royal College of Psychiatrists. The hospital-level occupational checklist data for 2012/13 and 2016/17 can be downloaded from https://www.rcpsych.ac.uk/improving-care/ccqi/national-clinical-audits/national-audit-of-dementia/nad-reports-and-resources. Restrictions apply to access to the 2010/11 occupational checklist dataset, which were used for the present study under approvals and we are not permitted to share this data.

Competing interests

The authors declare that they have no competing interests.

Funding

This study was funded jointly by the Economic and Social Research Council (ESRC) and the National Institute for Health Research (NIHR). ESRC is part of UK Research and Innovation. The views expressed are those of the author(s) and not necessarily those of the ESRC, UKRI, NHS, the NIHR or the Department of Health and Social Care. This work forms part of the ESRC/NIHR Neighbourhoods and Dementia: a mixed methods study [https://sites.manchester.ac.uk/neighbourhoods-and-dementia/]. This paper relates to work programme 5: Developing the evidence base for dementia training in acute NHS hospitals (DEMTRAIN). ESRC Grant reference: ES/L001772/1. The funding bodies covered the costs of access to the HES dataset and supported FH, HM, FA and MH when working on the study. The funding bodies had no role in the study design, implementation, analysis, or interpretation of the data.

Authors' contributions

Contributors DR, JK, SR conceived the study idea and methodology with assistance from HM, FA, MH, FH, LD AND IL. FH, DR and MH analysed the data. DR, FH SR, JK, and HM wrote the first draft of the manuscript. All authors edited the drafts of this article and approved the final version. DR is the guarantor of the article.

Acknowledgements

We would like to thank the DEMTRAIN study Advisory Group, that included people living with dementia and was Chaired by Dr Dawne Garratt, who was the RCN Professional Lead for Older People and Dementia Care, who have given feedback throughout the study. We particularly want to thank and acknowledge Chloe Hood, Programme Manager, and the National Audit of Dementia team at the Royal College of Psychiatry for their support and advice on matters related to the NAD surveys and datasets. Further thanks go to the members of our Expert Panel (Asan Akpan, Neil Pendleton and Emma Vardy) for their advice on clinical aspects of dementia, related ICD10 codes, and the understanding and interpretation of Hospital Episode Statistics data. We would also like to thank Professor Alistair Burns in his role as National Clinical Director for Dementia and Older People's Mental Health at NHS England and NHS Improvement for his help and assistance in the work programme.

Authors information

Not applicable

Knapp M, Prince M: Dementia UK: The full report. In. London: Alzheimer's Society; 2007.
Hsiao F-Y, Peng L-N, Wen Y-W, Liang C-K, Wang P-N, Chen L-K. Care Needs and Clinical Outcomes of Older People with Dementia: A Population-Based Propensity Score-Matched Cohort Study. PLoS ONE. 2015;10(5):e0124973.
Shepherd H, Livingston G, Chan J, Sommerlad A. Hospitalisation rates and predictors in people with dementia: a systematic review and meta-analysis. BMC Med. 2019;17(1):130.
Boaden A. Fix Dementia Care: Hospitals. In. Alzheimer's Society; 2016.
Lakey L: Counting the cost. Caring for people with dementia on hospital wards. In. London: Alzheimer's Society; 2009.
Digby R, Lee S, Williams A. The experience of people with dementia and nurses in hospital: an integrative review. J Clin Nurs. 2017;26(9–10):1152–71.
Fogg C, Griffiths P, Meredith P, Bridges J. Hospital outcomes of older people with cognitive impairment: An integrative review. Int J Geriatr Psychiatry. 2018;33(9):1177–97.
Bail K, Berry H, Grealish L, Draper B, Karmel R, Gibson D, Peut A. Potentially preventable complications of urinary tract infections, pressure areas, pneumonia, and delirium in hospitalised dementia patients: retrospective cohort study. BMJ Open. 2013;3(6):e002770.
Reynish EL, Hapca SM, De Souza N, Cvoro V, Donnan PT, Guthrie B. Epidemiology and outcomes of people with dementia, delirium, and unspecified cognitive impairment in the general hospital: prospective cohort study of 10,014 admissions. BMC Med. 2017;15(1):140.
Möllers T, Stocker H, Wei W, Perna L, Brenner H. Length of hospital stay and dementia: A systematic review of observational studies. Int J Geriatr Psychiatry. 2019;34(1):8–21.
Banerjee S, Owen J. Living Well with Dementia: A National Dementia Strategy. In. London: Department of Health; 2009.
Department of Health: Delivering high quality, effective, compassionate care: Developing the right people with the right skills and right values. A mandate from the Government to Health Education England: April 2013 to March. 2015. In. London: Department of Health; 2013.
Ahmed F, Morbey H, Harding A, Reeves D, Swarbrick C, Davies L, Hann M, Holland F, Elvish R, Leroi I. Developing the evidence base for evaluating dementia training in NHS hospitals (DEMTRAIN): a mixed-methods study protocol. BMJ open. 2020;10(1):e030739.
National Health Service. Five year forward view. In. NHS England; 2014.
Dementia Roadmap [https://dementiaroadmap.info/about/].
Skills for Health. Dementia Training Standards Framework. In.
The Butterfly Scheme: reaching out to people with dementia and to anyone needing memory support. http://butterflyscheme.org.uk/about-us/.
John's Campaign: for the right to stay with people with dementia. https://johnscampaign.org.uk/.
Mukadam N, Sampson EL. A systematic review of the prevalence, associations and outcomes of dementia in older general hospital inpatients. Int Psychogeriatr. 2010;23(3):344–55.
Zhao Y, Kuo T-C, Weir S, Kramer MS, Ash AS. Healthcare costs and utilization for Medicare beneficiaries with Alzheimer's. BMC Health Serv Res. 2008;8(1):108.
Natalwala A, Potluri R, Uppal H, Heun R. Reasons for Hospital Admissions in Dementia Patients in Birmingham, UK, during 2002–2007. Dement Geriatr Cogn Disord. 2008;26(6):499–505.
Fogg C, Meredith P, Culliford D, Bridges J, Spice C, Griffiths P. Cognitive impairment is independently associated with mortality, extended hospital stays and early readmission of older people with emergency hospital admissions: A retrospective cohort study. Int J Nurs Stud. 2019;96:1–8.
Hospital. Episode Statistics. https://digital.nhs.uk/data-and-information/data-tools-and-services/data-services/hospital-episode-statistics.
Care Quality Commission. The state of health care and adult social care in England Technical Annex 3: Dementia thematic review. In.; 2012.
National Audit of Dementia. Guidance for acute hospital sites (updated). In.; 2016.
Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care. 1998;36(1):8–27.
Bottle R, Gaudoin R, Goudie R, Jones S, Aylin P. Can valid and practical risk-prediction or casemix adjustment models, including adjustment for comorbidity, be generated from English hospital administrative data (Hospital Episode Statistics)? A national observational study. Health Services and Delivery Research 2014, 2(40).
ISB 0070:. Health care Resource Groups (HRGs). https://digital.nhs.uk/data-and-information/information-standards/information-standards-and-data-collections-including-extractions/publications-and-notifications/standards-and-collections/isb-0070-healthcare-resource-groups-hrgs.
National Audit of Dementia. [https://.
Williams RL. A Note on Robust Variance Estimation for Cluster-Correlated Data. Biometrics. 2000;56(2):645–6.
Stapleton LM, McNeish DM, Yang JS. Multilevel and Single-Level Models for Measured and Latent Variables When Data Are Clustered. Educational Psychol. 2016;51(3–4):317–30.
Huang FL. Multilevel Modeling and Ordinary Least Squares Regression: How Comparable Are They? J Experimental Educ. 2018;86:265–81.
Barros AJD, Hirakata VN. Alternatives for logistic regression in cross-sectional studies: an empirical comparison of models that directly estimate the prevalence ratio. BMC Med Res Methodol. 2003;3(1):21.
Zou G. A modified poisson regression approach to prospective studies with binary data. Am J Epidemiol. 2004;159(7):702–6.
Austin PC. Absolute risk reductions, relative risks, relative risk reductions, and numbers needed to treat can be obtained from a logistic regression model. J Clin Epidemiol. 2010;63(1):2–6.
Hartley P, Gibbins N, Saunders A, Alexander K, Conroy E, Dixon R, Lang J, Luckett J, Luddington T, Romero-Ortuno R. The association between cognitive impairment and functional outcome in hospitalised older patients: a systematic review and meta-analysis. Age Ageing. 2017;46(4):559–67.
Pendlebury ST, Lovett NG, Smith SC, Dutta N, Bendon C, Lloyd-Lavery A, Mehta Z, Rothwell PM. Observational, longitudinal study of delirium in consecutive unselected acute medical admissions: age-specific rates and associated factors, mortality and re-admission. BMJ Open. 2015;5(11):e007808.
Reynish E, Hapca S, Walesby R, Pusram A, Bu F, Burton JK, Cvoro V, Galloway J, Laidlaw HE, Latimer M, et al: Understanding health-care outcomes of older people with cognitive impairment and/or dementia admitted to hospital: a mixed-methods study. 2021, 9:8.
Kagansky N, Berner Y, Koren-Morag N, Perelman L, Knobler H, Levy S. Poor nutritional habits are predictors of poor outcome in very old hospitalized patients. Am J Clin Nutr. 2005;82(4):784–91. quiz 913 – 784.
Saravay SM, Kaplowitz M, Kurek J, Zeman D, Pollack S, Novik S, Knowlton S, Brendel M, Hoffman L. How do delirium and dementia increase length of stay of elderly general medical inpatients? Psychosomatics. 2004;45(3):235–42.
Sampson EL, Leurent B, Blanchard MR, Jones L, King M. Survival of people with dementia after unplanned acute hospital admission: a prospective cohort study. Int J Geriatr Psychiatry. 2013;28(10):1015–22.
Sampson EL, Blanchard MR, Jones L, Tookman A, King M. Dementia in the acute hospital: prospective cohort study of prevalence and mortality. Br J Psychiatry. 2009;195(1):61–6.
Alzheimer’s Society and Marie Curie Cancer Care. Living and dying with dementia in England: Barriers to care. In.; 2014.
Thuné-Boyle ICV, Sampson EL, Jones L, King M, Lee DR, Blanchard MR. Challenges to improving end of life care of people with advanced dementia in the UK. Dementia. 2010;9(2):259–84.
Moore G, Hartley P, Romero-Ortuno R. Health and social factors associated with a delayed discharge amongst inpatients in acute geriatric wards: A retrospective observational study. Geriatr Gerontol Int. 2018;18(4):530–7.
Challis D, Hughes J, Xie C, Jolley D. An examination of factors influencing delayed discharge of older people from hospital. Int J Geriatr Psychiatry. 2014;29(2):160–8.
Goddard M, Kasteridis P, Jacobs R, Santos R, Mason A. Bridging the gap. J Integr Care. 2016;24(1):15–25.
Royal College of Psychiatrists. National Audit of Dementia care in general hospitals 2016–2017: Third round of audit report. In. London: Royal College of Psychiatrists; 2017.
Advance care planning. https://www.nice.org.uk/about/nice-communities/social-care/quick-guides/advance-care-planning#introducing-care-planning.
Shepperd S, Iliffe S, Doll HA, Clarke MJ, Kalra L, Wilson AD. D G-B: Admission avoidance hospital at home. Cochrane Database of Systematic Reviews 2016(9).
Pickens S, Naik AD, Catic A, Kunik ME. Dementia and Hospital Readmission Rates: A Systematic Review. Dement Geriatric Cogn Disorders Extra. 2017;7(3):346–53.
Hapca S, Guthrie B, Cvoro V, Bu F, Rutherford AC, Reynish E, Donnan PT. Mortality in people with dementia, delirium, and unspecified cognitive impairment in the general hospital: prospective cohort study of 6,724 patients with 2 years follow-up. Clin Epidemiol. 2018;10:1743–53.
Pendlebury ST, Lovett NG, Thomson RJ, Smith SC. Impact of a system-wide multicomponent intervention on administrative diagnostic coding for delirium and other cognitive frailty syndromes: observational prospective study. Clin Med. 2020;20(5):454.
CHKS. The quality of clinical coding in the NHS. In.: Department of Health and Social Care; 2014.
Sommerlad A, Perera G, Singh-Manoux A, Lewis G, Stewart R, Livingston G. Accuracy of general hospital dementia diagnoses in England: Sensitivity, specificity, and predictors of diagnostic accuracy 2008–2016. Alzheimer's & Dementia 2018, 14(7):933–943.

Additionalfile1.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Retrospective study of more than 5 million emergency hospital admissions: epidemiology and outcomes for people with dementia

Status:

Version 1

Abstract

Figures

Background

Aims

Methods

Results

Discussion

Conclusion

Abbreviations

Declarations

References

Supplementary Files

Status:

Version 1