Understanding health-care outcomes of older people with cognitive impairment and/or dementia admitted to hospital: a mixed-methods study

Protocol and registration

A protocol for the review was developed and registered with PROSPERO in 2015 (PROSPERO CRD42015024492) before work was started on defining search terms. 37

Eligibility criteria

An inclusive approach was adopted in the original search. Table 1 contains the full list of inclusion and exclusion criteria. The exposure of interest was CSD and how it was measured or diagnosed.

Randomised controlled trials, intervention studies, quality improvement initiatives, before-and-after designs and narrative reviews were excluded as they do not provide general population (unselected) prevalence or outcome data. Systematic reviews were retained for review of their reference lists.

Other exclusions were made to remove non-general hospital settings, such as community hospitals, intensive and post-acute care units, rehabilitation hospitals, outpatient clinics, primary care, mixed settings (i.e. outpatients and inpatients) and inpatients who had been discharged home before data collection.

Outcomes of interest included mortality (in-hospital and at follow-up), length of hospital stay, hospital re-admission, admission to long-term care (nursing homes, etc.), health or social care costs, physical function [activities of daily living (ADL)], QoL and change in cognitive function. Any articles that did not report prevalence data or outcome data on any of the outcomes of interest were excluded.

No restrictions were made for date of publication in the search. Conference abstracts were included in the screening and a search was carried out based on title and first and last author to identify any subsequent full-text publication for inclusion in the review. Systematic reviews were not included in the full-text review. The results were restricted to publications available in the English language.

Information sources

The following databases were searched between 29 January and 1 February 2016:

• Ovid MEDLINE In-Process & Other Non-Indexed Citations and Ovid MEDLINE (date range searched: 1946 to present)

• Ovid EMBASE (date range searched: 1980 to 2016 week 4)

• EBSCOhost Cumulative Index to Nursing and Allied Health Literature (CINAHL) Plus

• Ovid PsycINFO (date range searched: 1806 to January week 4 2016)

• Cochrane Database of Systematic Reviews.

See Appendix 2, Table 25.

Search

Initial scoping searches were undertaken in MEDLINE using keywords and medical subject heading (MeSH) terms both separately and combined in order to comprehend their coverage. The search was formed using three distinct concepts: (1) CSD (made up of searches for dementia, delirium and cognitive impairment), (2) hospital inpatient and (3) prevalence and/or outcomes. Search terms were combined within the concepts using OR and the three concepts were combined using AND.

Search strategy development drew on a range of existing published sources. The reviews conducted by Mukadam and Sampson13 and Siddiqi et al. 38 were consulted and search strategies were shared with the review team. The Cochrane Dementia and Cognitive Improvement Group search strategies for delirium and dementia were used to inform search strategy development. 39 Terms for cognitive impairment were expanded. Additional assistance and advice was given by an information specialist from the University of Stirling.

Multiple iterations of the search were tested to ensure that it identified all papers on a list of target publications constructed by the research team. The search strategy was circulated to the External Advisory Board for comment and then finalised. A copy of the complete search strategy is available in Appendix 2.

Study selection

All articles were uploaded into the online systematic review tool for screening and review (Covidence, VIC, Australia). Direct deduplication was carried out by the software. Titles and abstracts were screened independently by pairs of reviewers, one of whom was a senior reviewer, and all conflicts were resolved through discussion between two of the senior reviewers.

Prior to full-text screening, two senior reviewers reassessed all titles and abstracts and removed those that were not in the general or geriatric medical setting in order to limit the heterogeneity of study methods resulting from disease-specific hospital settings. A spreadsheet was developed to classify these so-called specialist populations for future use by one reviewer, and each entry was checked by a second reviewer to ensure consistency in classification. Any studies that did not meet the review exclusion criteria were removed at this stage with the agreement of the two reviewers.

Full-text screening of 422 articles was undertaken independently by pairs of experienced reviewers. Of these articles, 73 were reviewed twice independently by pairs of reviewers and consensus was reached over any conflicts. Full-text screening for the remaining 349 articles was completed by at least one independent senior reviewer. Eighty per cent of the 349 articles reviewed were cross-checked by two senior reviewers to ensure consistency and evaluation of judgements about relevance. There was strong consistency in judgements between both reviewers, and any conflicts between the reviewers were resolved. The final number of included articles was 146. Sixty-four foreign-language papers were not considered for review.

An exclusion hierarchy was developed for full-text screening, recognising that there may be multiple reasons to exclude a single study:

1. non-human study

2. paediatric population (aged < 18 years)

3. duplicate record

4. specialist population

5. wrong study design

6. wrong setting

7. wrong population

8. conference abstract (with no subsequent full text)

9. no prevalence/outcome data reported

10. foreign-language publication

11. insufficient information to evaluate.

Data collection

A data extraction form was developed using Google Docs (Google Inc., Mountain View, CA, USA). This was piloted with the data extraction team to improve consistency of approach. It was developed in line with Stirling University Literature Review and Evaluation Methodology. Data extraction was carried out by two independent reviewers. To ensure consistency of data extraction, all articles were cross-checked by each reviewer. Particular attention was given to articles for which a second opinion was sought. All conflicts were resolved by consensus between the two reviewers and did not require the opinion of a third reviewer.

Data items

Data were extracted on the following items: emerging issues, population, setting, type of study, age range, sample size, sex of participants, coverage, the country that the study was conducted in, inclusion and exclusion criteria, start date of the study and study duration. For each CSD covered in the article, data items included the definition of CSD used in the paper, assessment tools or diagnostic criteria, number of cases, size of underlying population, quoted prevalence and quoted incidence. For associated outcomes, the following data items were extracted: LoS; QoL; mortality; nursing/care home admission; functional status/ADL; change in cognitive status; health-care costs; hospital re-admission; other outcomes; and covariates. All outcomes were reported on.

Quality assessment of studies

Quality assessment was conducted based on the tool developed by Boyle40 and adapted by Mukadam and Sampson. 13 This was integrated into the data extraction form.

A maximum quality score of 18 could be assigned in the context of each diagnosed condition. When evaluation questions were not applicable to the study (e.g. when studies did not address all conditions), the maximum score was adjusted to reflect this and to ensure that quality was measured equitably for all studies.

Risk of bias across studies

To ensure consistency of judgements about the quality of evidence, the second independent reviewer assessed 20% of the included studies; this has been found in previous work to be sufficient to ensure consistency. These studies were identified randomly and any identified disagreements were resolved.

Summary measures

Studies were included if they reported quantitative data on the prevalence of any of the CSDs in an unselected hospital population and/or if they reported quantitative data on the outcomes of interest, based on CSD category or comparing a CSD with no CSD.

Study selection

The initial search identified 23,000 records after initial deduplication. Following title and abstract screening, 2646 records remained. Specialist population removal removed a further 1553 articles (Table 2). In addition, 671 identified articles did not comply with the original study eligibility criteria and so were removed, resulting in 422 for full-text review. A total of 277 records were excluded from the review on full-text screening (see the exclusion hierarchy in Study selection). The search was re-run to identify any conference abstracts available as full-text articles, which yielded one additional full text for inclusion. A total of 141 articles were included in the review (Figure 1).

FIGURE 1.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram.

Studies excluded from the review

A total of 283 studies were excluded. Studies were excluded for the following reasons: foreign language (n = 64), duplicates (n = 11), specialist population (n = 6), insufficient evidence to evaluate (n = 1), no prevalence or outcome data (n = 6), wrong setting (n = 102), wrong population (n = 11), wrong study design (n = 40), conference abstract (n = 10), abstract identified and no full text available (n = 9), response letters (n = 4) and full text not obtainable (n = 19).

Included study characteristics

A total of 141 studies were included in the review. A summary of the characteristics of the included studies is provided in Appendix 3, Table 26.

The sample size varied significantly, from 1841 to 1,135,42342 participants. Fifty-five per cent of the studies were conducted in Europe. A total of 127 (89%) of the included studies were reported as cohort designs, 12 of which were retrospective and 113 of which were prospective. There was one descriptive cohort study.

Five retrospective studies were reported as secondary analyses and two were reported as case–control studies. There were seven cross-sectional studies. 19,43–48

Participants in one study had a mean age of < 65 years; 13 studies did not report the average age of participants. 49–52

Study duration varied, with data ranging from 1 month to 17 years. Seventy-three studies represented the acute general setting, with 65 in the acute/geriatric setting, and three studies encompassed both acute general and geriatric settings. 53–55

Twelve studies evaluated participants for both cognitive impairment and dementia. Sixteen studies evaluated dementia alone and 21 studies evaluated only cognitive impairment. Delirium was evaluated in 89 studies, DSD was evaluated in 18 studies, 46 studies screened for both delirium and dementia, and delirium, dementia and cognitive impairment were screened for in 11 studies.

There was heterogeneity in terminology used to describe the acute hospital setting, which encompassed terminology including ‘teaching hospital’, ‘university hospital’ and ‘internal medicine’.

Screening and prevalence of delirium

A total of 89 included studies reported delirium prevalence. Delirium prevalence ranged from 5% to 85.5% (see Appendix 3, Table 27), reflecting the range of diagnostic tools and methodological approaches used. 56,57 Demographic characteristics of the cohort and differences in study design may also have influenced prevalence estimates; for example, Jitapunkul and Hanvivadhanakul49 included an all-female sample. Goldberg et al. 58 used a sample that disproportionately included patients who had carers living locally and who had longer hospital stays.

Adamis et al. 59 included those with less severe delirium – in parallel with other studies demonstrating selection biases – and required participant consent, thus potentially underestimating delirium prevalence.

There were eight retrospective studies and three secondary analyses. These designs can lead to underestimates of prevalence figures and depend on quality and availability of medical data. Prevalence figures may also have been influenced by not routinely diagnosing delirium but using only a single assessment in which it is difficult to differentiate between new and existing cases of delirium (e.g. Edlund et al. 60). Four studies included only incident cases; thus, prevalence could not be reported. 61–64

The term ‘acute confusion’ was used to describe delirium in five studies. 49,50,65–67 Five studies reported prevalence of subsyndromal delirium (SSD): Bourdel-Marchasson et al. 68 (20.6% SSD), Cole et al. 69 (65% SSD), Lam et al. 70 [66.2% residual subsyndromal delirium (rSSD)], Martínez-Velilla et al. 71 (22.3% SSD) and Zuliani et al. 48 (37.9% SSD). Six studies reported prevalence figures for subtypes of delirium: mixed, hypoactive and hyperactive delirium. 19,57,60,70,72,73

The most common tools adopted to diagnose delirium (see Appendix 3, Table 27) were the Confusion Assessment Method (CAM)/Confusion Assessment Method for the intensive care unit (CAM-ICU) (41 studies) and the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) (21 studies). There was considerable heterogeneity across studies in the diagnostic tools used. Basic and Khoo74 and Basic and Hartwell75 did not specify how diagnosis was made. Díez-Manglano et al. 76 defined delirium presence from any cause during the previous hospitalisation, and relied on medical notes to diagnose delirium.

Screening and prevalence of dementia

The Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR), (21 studies) and the Mini Mental State Examination (MMSE) (21 studies) were the most commonly used tools to diagnose dementia (see Appendix 3, Table 28). There were five case-only studies. 12,77–80 Prevalence in the remainder of studies varied greatly: from 1.33%81 to 74.4%,70 probably owing to sample selection biases and heterogeneity in screening tools and methodological approaches used to diagnose dementia. Five studies did not specify the criteria used to diagnose dementia. 41,72,75,82,83 Variations in demographic characteristics of the studied population also may have affected prevalence figures; for example, Di Iorio et al. 84 conducted a study in multiple hospital sites.

Six studies appeared to use the terms ‘cognitive impairment’ and ‘dementia’ interchangeably. 59,62,68,85–87

Several studies were selective in the approach used. For example, Sampson et al. 12 excluded people with persistent delirium, so may have underestimated the prevalence of dementia, and Aminoff77 used a case-only study to include only patients with severe dementia.

Five studies reported the prevalence of subtypes of dementia. Erkinjuntti et al. 17 found vascular dementia to be the most common (72.4%), then primary degenerative dementia (PDD) (23%) and then other causes (4.6%). Erkinjuntti et al. 18 reported prevalence of PDD as 16.1% and prevalence of vascular dementia as 69.4%, and 14.5% of patients had specific causes of dementia. Jackson et al. 88 reported Alzheimer’s disease (AD) prevalence to be 66%, vascular dementia prevalence to be 26% and mixed dementia prevalence to be 6%, followed by dementia with Lewy bodies at 21%. Lorén Guerrero and Gascón Catalán44 reported AD prevalence to be 40.74%. Wancata et al. 21 reported AD prevalence to be 61.8%, multi-infarct dementia prevalence to be 21.6%, pre-senile dementia prevalence to be 2.5% and unidentified dementia prevalence to be 14.1%.

There were 11 retrospective studies, including five secondary analyses. These study designs are likely to produce an underestimate of prevalence of dementia, as estimates were derived from medical notes or discharge codes, rather than from assessment actively undertaken by a researcher. In retrospective studies, validity depends on the quality of the discharge reports used as well as access to other data, including re-admissions. Furthermore, five prospective studies relied only on medical records to diagnose dementia. 49,67,79,89,90 Six prospective studies and one retrospective study did not specify how dementia was diagnosed. 41,54,63,72,75,82,91 Hsieh et al. 92 prospectively assessed dementia using a researcher-led approach but did not specify the diagnostic tool used.

Screening and prevalence of delirium superimposed on dementia

Typically, the term ‘delirium superimposed on dementia’ is used when an acute change in mental status (e.g. fluctuating course, inattention, altered level of consciousness and disorganised thinking) occurs alongside pre-existing dementia. 93 Across all papers, there was considerable heterogeneity in the operationalisation and measurement of DSD and the diagnostic criteria applied (see Appendix 3, Table 29). Five studies did not explicitly reference the term ‘delirium superimposed on dementia’, but included patients who presented with both conditions concurrently. Faezah et al. 72 did not clearly assign prevalence estimates to differentiate the subjects in the cohort under study.

Prevalence varied widely, from 0.5% to 76%, which probably reflects the range of diagnostic approaches and the variation in the populations assessed. McCusker et al. 94 reported the highest prevalence of DSD across all prospective studies, at 76% (based on 164/217 patients).

The DSM-IV was used in nine studies in conjunction with other tools to diagnose DSD, but there was no consensus approach to measure DSD. International Classification of Diseases, Ninth Edition (ICD-9), coding was used in two studies: Bellelli et al. ,55 who reported the lowest prevalence of DSD, and Gallerani et al. 95 Bellelli et al. 55 did not document pre-existing dementia using a separate, validated tool, which limited the ability to differentiate deficits associated with delirium from those in dementia. Rockwood83 did not specify how dementia assessment took place. Several papers did not distinguish prevalent and incident cases of delirium as delirium assessment was undertaken on the day of admission only. 60,79,96,97 Single assessments also make it difficult to differentiate delirium from dementia as they share many symptoms. McCusker et al. 26 assessed prevalent and incident delirium separately but did not report the relative number of cases.

Screening and prevalence of cognitive impairment

A total of 21 studies screened for cognitive impairment but did not report prevalence estimates.

The range of prevalence was 8.9%59 to 80%. 62 The majority of studies defined cognitive impairment as distinct from ‘dementia’, and, in 45 studies, the MMSE was used to diagnose cognitive impairment (see Appendix 3, Table 30). There was heterogeneity in the assessment tools used across studies, potentially biasing prevalence estimates.

Five studies focused on ‘mild cognitive impairment (MCI)’. Bickel et al. 98 reported ‘mild cognitive impairment’ as patients diagnosed using International Working Group on MCI criteria and fulfilling criteria for cognitive impairment but not dementia; Orsitto et al. 46,99,100 used the Petersen criteria101 to diagnose ‘mild cognitive impairment’. Jackson et al. 88 defined MCI as when the person is neither normal nor demented with some evidence of cognitive decline, and ADL largely intact.

Two studies did not describe the diagnostic approach used to screen cognitive impairment. 102,103 There was some heterogeneity in the terminology used to define cognitive impairment, with six studies appearing to use the terms ‘cognitive impairment’ and ‘dementia’ interchangeably. 59,62,68,85–87

Freedberg et al. 104 used ICD-9 coding to diagnose delirium and/or dementia under the umbrella term ‘cognitive impairment’, and did not report separate prevalence figures for delirium or dementia.

Esmayel et al. 43 evaluated a cohort with a high level of illiteracy and did not adequately distinguish between educational attainment levels that would potentially affect MMSE scores.

Dementia outcomes

Delirium outcomes

Cognitive impairment outcomes

Delirium superimposed on dementia outcomes

Appendix 3, Table 34, outlines outcomes for DSD. Two studies examined outcomes for DSD. 79,94 McCusker et al. 94 showed that those presenting with both delirium and dementia had a poorer cognitive status and were more likely to be admitted to long-term care than those with neither condition. Lang et al. 79 identified DSD as a marker for prolonged hospital stay.

Quality of studies

Variation in the quality of studies in a systematic review is a limitation. From the assessment of the quality of the 141 selected studies, we would recommend that future studies in related areas ensure that, from the outset, they give a clear description of the population, the method of sampling and the condition(s) studied, with adjustments for any confounding factors; that a standardised tool for diagnosis of dementia and other risk factors is used; and, finally, that the study clearly explains statistical methods and clinically significant associations.

We defined a high-quality study as one that dropped no more than 1 point in our assessment. From the 141 studies, 63 studies12,21,45–48,53,56,61,64,65,69,70,74,84,98,103–105,107,110,112–115,118,120,121,123,124,128–132,134,136–162 scored high in our quality assessment. A further 22 studies57,58,62,67,68,73,78,88,94,100,109,111,119,126,133,163–169 were classified as good, which we defined as dropping 2 or 3 points. The remaining 56 studies17–20,26,41–44,49–52,54,55,59,60,63,66,71,72,75–77,79–83,85–87,89–92,95–97,99,102,108,116,117,122,125,127,135,170–177 dropped ≥ 4 points, scoring low in our quality assessment. A review of these studies showed that the most common study deficiency is a lack of or insufficient description of presence of condition and/or adjustment for confounding factors. Twenty-four studies did not contain this description and 16 studies gave only partial information about this. Apart from this, the study factors most commonly in need of improvement were clear explanation of statistical methods and clinically significant associations (10 did not and 22 only partially); use of a structured/standardised tool for definition of dementia and other risk factors (11 did not and eight only partially); clear description of the process of sampling and selecting patients (five did not and 26 only partially); and clear and detailed description of the characteristics of the population (five did not and 23 only partially).

Prevalence of delirium and delirium superimposed on dementia

There was considerable heterogeneity in the diagnostic tools used to assess delirium. In addition, some studies did not distinguish between prevalent and incident cases as assessment of delirium was conducted in one session rather than routinely. 60 This variation in approaches and tools used can affect the reliability of prevalence estimates.

Delirium has previously been reported to be under-recognised in older hospitalised patients. 93 The reliance on discharge codes in retrospective studies also typically leads to a higher underestimation risk. Underdiagnosis of delirium may also arise from lack of awareness of the fluctuative course of delirium and its potential overlap with dementia; thus, comprehensive cognitive assessment is necessary for distinguishing disorders with overlapping symptoms. Forty-nine studies that screened for dementia also screened for delirium, increasing the overall reliability of prevalence estimates, but when studies did not separately screen for each condition, an underestimate of delirium may have arisen in favour of dementia diagnosis.

Delirium superimposed on dementia is typically characterised by premorbid dementia followed by an acute mental change in which delirium is suspected. Previously, studies reported that delirium is underdetected in older hospitalised patients. 93 The potential risk factors for under-recognition of delirium by nursing staff are dementia and the hypoactive form of delirium, the onset of which does not necessarily elicit a distinctly recognisable change in mental status. 178

The variation in assessment tools used to detect delirium and dementia also influences the wide range of prevalence estimates for DSD reflected in the included studies. For example, Bellelli et al. 55 reported that the ICD-9 has poor diagnostic accuracy for delirium, and Johnson et al. 179 found the Diagnostic and Statistical Manual of Mental Disorders, Third Edition (DSM-III), to have greater diagnostic accuracy than the ICD-9. Of the included papers reviewed in which DSD was formally documented, one study did not assess the premorbid existence of dementia using a validated tool, such as the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE), which meant that neurocognitive deficits attributed to delirium could not be reliably differentiated from those in dementia. 55

Although the majority of studies used validated tools to assess delirium, there was no comment specifically on the diagnostic accuracy of those tools in the context of patients with dementia. In addition, when studies did not continually assess delirium to record incident cases, it was difficult to assess the epidemiological impact of delirium on those with dementia had there been serial cognitive testing.

In summary, to recognise cases of DSD it is important to conduct a baseline assessment of dementia, whereby medical staff have full clarity and recognition of the symptoms pertinent to dementia. This must be supplemented with the ability to differentiate between this baseline status and symptoms characteristic of an acute mental change attributed to delirium. Dementia is a risk factor for delirium and the co-existence of these conditions leads to poor prognostic outcomes. It is thus imperative that consensual, comprehensive assessment approaches are undertaken using validated tools in the context of DSD in order to improve diagnostic accuracy in these groups.

Prevalence of dementia

Of 80 studies that screened for dementia, 49 also screened for delirium. The prevalence of dementia may be overestimated in studies not also screening for delirium.

Studies not including patients who were too ill to give informed consent or without proxies may exclude those at most risk of having pre-existing dementia, leading to selection bias, which affects prevalence estimates.

The high prevalence of mixed and vascular dementia reported in studies can be attributed to hospital setting; for example, more stroke patients are likely to be admitted to acute hospital settings or exhibit higher cardiovascular comorbidity. 13

Prevalence of cognitive impairment

Seven papers used inconsistent terminology to distinguish ‘cognitive impairment’ and ‘dementia’, used the terms interchangeably or did not make separate diagnoses for each condition. According to the majority of studies screening for cognitive impairment, there is a consensus that cognitive impairment is diagnostically exclusive of dementia or delirium, despite the fact that the three conditions share common characteristics and thus present with overlapping symptoms. Inconsistencies in the terminology adopted across studies can thus give rise to unreliable estimates of prevalence for cognitive impairment.

There is overlap in the assessment of cognitive impairment and dementia given their shared characteristics and, when studies explicitly excluded subjects with dementia from their analyses, a more reliable prevalence estimate for cognitive impairment could be achieved. When studies did not explicitly report cut-off scores in the MMSE, prevalence estimates for cognitive impairment could be biased as criteria by which cognitive impairment is diagnosed are not reported. 62 Other factors that can lead to overestimates of prevalence are the potentially inadequate conditions to appropriate detection inside a hospital setting, presence of other clinical conditions or performance difficulties not related to cognitive impairment, which can yield unreliable cognitive evaluations, for example patients failing to use glasses/hearing aids while being assessed with the MMSE. 156

Clinical implications

The population studied and the Older Persons Routine Acute Assessment data set

NHS Fife provides medical care to a varied urban and rural population of ≈ 360,000 people. From January 2011, all emergency medical admissions within the health board from any source were via a single acute medical unit (AMU) at the research hospital (the only exceptions are acute stroke and acute ST segment myocardial infarction, for which admission is via specialist services). The research hospital is a district general hospital with 640 beds and a full range of health-care specialties. After AMU admission, patients are usually discharged or stepped down to appropriate medical wards after 12–24 hours. Orthopaedic trauma patients requiring surgery are admitted via the surgical admissions unit and non-operative trauma patients are admitted via the AMU.

Starting in 2009 and funded by the Scottish Government Joint Improvement Team, this health board’s Dementia Co-ordinating Group designed and implemented the Older Persons Routine Acute Assessment (OPRAA). From 2011, OPRAA was offered routinely to all people aged ≥ 65 years admitted as an emergency to a NHS hospital in this health board. By design, individuals with a predicted LoS of < 24 hours, for whom death was expected or with an acute illness requiring critical care intervention did not undergo an OPRAA.

The cohort

The design is a cohort study of all people aged ≥ 65 years with an acute medical admission to one district general hospital in Scotland, prospectively recruited to undergo an OPRAA. Inclusion and exclusion criteria are detailed in Table 3.

Data for all emergency medical admissions of people aged ≥ 65 years were identified from Scottish Morbidity Records (SMR) 01 data, which is a validated NHS Scotland routine data set providing information on the date of admission and discharge, type of admission, admission and discharge destination and the patient’s main and other conditions [in the form of International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes]. An emergency medical admission was defined as an admission via the study hospital AMU under an acute medicine, general medicine or geriatric medicine specialty. Admission and final discharge date and the Community Health Index (CHI) number (the NHS Scotland unique patient identifier) were then used to link all eligible admissions in SMR01 to the OPRAA data set and to determine eligible admissions whereby patients underwent an OPRAA.

An incident cohort was further defined, comprising people aged ≥ 65 years who had received an OPRAA during their first acute medical admission within the study period, providing that they had not had an acute medical admission in the 6 months prior to the start of the study.

The study period was chosen to reflect a period when OPRAA was routine. Between 1 January 2012 and 30 June 2013, > 80% of all acute medical admissions for those aged ≥ 65 years underwent an OPRAA.

Data extracted

For eligible patients, discharge diagnosis (excluding dementia) from all previous admissions recorded in SMR01 were used to calculate each participant’s Charlson Comorbidity Index (CCI) at each eligible admission. 180 The CHI data set was used to define participant age, sex and postcode-defined socioeconomic status [measured using quintiles of the Scottish Index of Multiple Deprivation (SIMD)] on admission. 181 Data on all community-dispensed prescriptions were used to create an additional multimorbidity score for case-mix adjustment, calculated as the number of drugs (defined as the number of distinct British National Formulary182 subsections) prescribed to the patient in the 84 days prior to admission. 183

A number of ways of identifying whether or not patients were care home residents were examined, including SMR01 admission and discharge coding, and the CHI Institution Flag. However, none was considered reliable, so manual classification of recorded address in the CHI register was used to identify patients as care home residents by comparison with the published list of Care Inspectorate-registered residential services.

The OPRAA data set was used to identify patients with CSDs, defined as one or more of known dementia alone, delirium alone, delirium superimposed on known dementia and unspecified cognitive impairment, and functional status based on assessment of ADL. 184 Full definitions of these variables are shown in Appendix 1.

The CHI number was used to deterministically link all data sets. The SMR01 data set was linked to the CHI, OPRAA, SMR04 and community-dispensed prescribing data sets to ascertain demography on admission and mortality, the presence of CSDs on admission and functional status before and on admission.

Definitions of primary and secondary outcomes

The primary outcome was whether or not an individual was living at home 30 days after discharge (binary outcome). The primary outcome was measured only in patients who were admitted from their private home (i.e. not care home residents) and were discharged alive from hospital. A negative outcome was defined as patients living in a care home, being back in the hospital or being dead at 30 days after discharge. Master CHI, together with SMR01, was used to determine whether the person was admitted or discharged to a care home, a process that involved a thorough validation of care home addresses. SMR01 was also used to determine whether or not the person had survived the incident admission or was re-admitted within 30 days after discharge and CHI was used to ascertain mortality within the 30 days from discharge.

Secondary outcomes:

• Mortality. The CHI data set was used to ascertain mortality, defined as time to death from admission with a 2-year follow-up period.

• Re-admission. This was defined as the time to the first emergency re-admission from discharge for the patients who survived the incident admission; this was calculated from SMR01 with 2-year follow-up from discharge. Mortality following discharge was another possible outcome acting as competing risk for re-admission, and master CHI was used to calculate the time to death from discharge within the 2-year follow-up time.

• Length of stay. This was defined as the full length of incident admission (in days) and was calculated from SMR01 based on the difference between discharge and admission dates. For patients who were admitted and discharged on the same day, the LoS was corrected to 1 day rather than 0 days.

Missing data

Data on delirium diagnosis (either CAM positive or clinical delirium) were missing in 3.7% of cases within the incident cohort. Based on the OPRAA alone, 9.8% of participants in the incident cohort were recorded as having a known dementia and 20.3% of cases had missing data for known dementia. After adding information on dementia from SMR01, SMR04 and prescribing data sets, the percentage of people with known dementia increased to 15.3%, with the remainder of cases being treated as absent of dementia. A total of 20.9% of cases had a missing Abbreviated Mental Test (AMT) score within the OPRAA, of which 15.5% had neither delirium nor dementia; these were classified as not having any CSD.

Twenty-seven per cent of ADL scores within the OPRAA had missing values. Multiple imputation was used to impute the missing values in terms of presence and absence of a persistently low ADL score or changed ADL score and a sensitivity analysis was conducted to assess the effect of missing ADL scores on the survival analysis results.

Ethics considerations

Data provision and initial management including linkage was carried out by the University of Dundee Health Informatics Centre (HIC), with analysis of anonymised data carried out in an ISO27001 and Scottish Government-accredited secure safe haven. The University of Dundee HIC standard operating procedures (SOPs) have been reviewed and approved by the regional NHS Research Ethics Service and consent for this study was obtained from the health board’s Caldicott Guardian. 185

Statistical analysis

Summary statistics based on proportions and their confidence intervals (CIs) were initially used to describe prevalence of CSDs (known dementia alone, delirium alone, DSD, unspecified cognitive impairment and no CSD) in older people admitted to an AMU and how this varied with sex, age, socioeconomic deprivation (SIMD quintiles) and whether or not they were admitted from a care home. The characteristics of older people in the different CSD groups were examined in terms of CCI (with four groups: 0, 1, 2–5 and ≥ 6), the number of drugs prescribed in the 84 days prior to admission (with four groups: 0, 1–5, 6–10 and ≥ 11 drugs) and ADL function (persistently low ADL score, changed ADL score or persistently high ADL score; see Appendix 1).

Descriptive statistics of the primary and secondary outcomes based on proportion and their CIs were generated prior to any modelling exercise. As described above, the exact cohort of patients included in analysis depended on the outcome. For example, for the primary outcome of whether or not a patient was living at home 30 days after discharge, the cohort comprised people admitted to hospital from their own home (because patients living in care homes very rarely move out of the care home so have a fixed negative outcome) and who survived to discharge.

Statistical analysis of the primary outcome

Associations between presence of different types of CSD and the primary outcome (whether or not the person is living at home at 30 days from discharge) were analysed, with logistic regression unadjusted and adjusted for baseline variables at admission, such as sex, age, deprivation status, comorbidity and number of drugs. A logistic regression model adjusted for functional status was used to explain how much of the poor outcome in patients with CSDs was explained by their functional ability. The results of the logistic regression were reported in terms of odds ratios (ORs) and their CIs. The c-statistic was estimated as a measure of predictive ability.

Statistical analysis of time to death from admission

Analysis of time to death from admission was initially assessed with Kaplan–Meier survival plots and log-rank tests for association considering the explanatory variables listed above. A 2-year follow-up time from admission was considered in the survival analysis and Cox proportional hazards models were first implemented to investigate the effect of CSDs on survival. Assessment of the proportional hazards assumption showed that some of the Cox model covariates did not meet this assumption, so a non-proportional Cox model with time-varying coefficients was implemented. 186,187 Time-varying coefficients were modelled based on a piecewise constant model function, where the 2-year follow-up time was split into five clinically meaningful time intervals: up to 1 month (implemented as up to 30 days), 1–3 months (31–90 days), 3–6 months (91–180 days), 6 months to 1 year (181–365 days) and 1–2 years (366–730 days). Akaike information criterion (AIC)-based model selection was then implemented to optimally choose the time points (within the 2-year follow-up period) when a change in hazard ratio (HR) was supported by the data. The effect of CSDs on survival was estimated in terms of unadjusted HRs and HRs adjusted for demographics and comorbidity variables, as well as HRs additionally adjusted for ADL functional status to specifically determine how much of the increase in HR in people with CSDs can be explained by their functional status. Variable selection in the adjusted models was conducted based on best fit evaluated using the AIC.

Finally, the log-likelihood ratio test statistic together with AIC scores were used to test whether or not different types of CSD were associated with mortality risk. Specifically, we tested for a difference in mortality risks between DSD and delirium alone or dementia alone, and unspecified cognitive impairment and dementia alone. This was done by comparing model performance between the model based on four CSD categories and a reduced model in which two types of CSD were grouped together depending on the hypothesis we wanted to test.

Statistical analysis of time to re-admission from discharge

Time to re-admission from discharge under the competing risk of death was initially assessed with cumulative incidence function (CIF) plots, and Gray’s188 test for subdistribution hazard was used to compare CIFs of time to re-admission among the groups of patients depending on their CSD type or demographics, comorbidity and functional status. Fine and Gray’s189 regression model was used to analyse the effect of CSD type on time to re-admission under the competing risk of death with a 2-year follow-up time. Assessment of the proportional subdistribution hazards assumption indicated that some of the Fine and Gray189 model covariates did not meet this assumption, so a model with time-varying coefficients was fitted to the data using piecewise constant coefficients, where the 2-year follow-up time was split into four clinically meaningful time intervals: up to 30 days, 30–90 days, 90 days to 1 year and 1–2 years. AIC-based model selection was then implemented to determine the time points (either 30 days, 90 days or 1 year, or all of them) when a change in subdistribution HR was supported by the data. The effect of CSDs on re-admission was estimated in terms of both unadjusted HRs and HRs adjusted for demographics and comorbidity variables, with variable selection based on best fit evaluated using the AIC.

Finally, a non-proportional Fine and Gray189 subdistribution hazard model for CSDs adjusted for ADL functional status (in addition to the other covariates) was fitted to the data to help determine how much of the increase in subdistribution HR in people with CSDs can be explained by their functional status.

Statistical analysis of length of stay

Patients admitted to the AMU would generally experience a short to moderate length of hospital stay, with only a small number having long hospital admissions, resulting in a positively skewed distribution of the LoS data. Therefore, a generalised linear model assuming a Gamma distribution with log-link function was used to analyse the LoS data. 190 Again, three models were fitted to the LoS data: the unadjusted model (with different types of CSD vs. no CSD in the model), the model adjusted for baseline variables at admission (sex, age, deprivation status, comorbidity and number of drugs) and the final model that was further adjusted for functional status. The variables CCI and number of drugs prescribed in the 84 days prior to admission appeared to be linearly related to LoS and, therefore, were introduced in the gamma model as numerical covariates. The results of the generalised gamma linear model were reported in terms of LoS rate ratios (RRs) between the category of interest and the reference category.

Changes to the National Institute for Health Research protocol

The statistical analysis plan (SAP) followed the protocol developed as part of this National Institute for Health Research (NIHR) project (Health Services and Delivery Research 13/54/55). As part of the modelling framework, Cox proportional hazards models and Fine and Gray189 competing risk models were proposed in the original application to analyse mortality and re-admission data. Non-proportional hazard models were developed to address violations of the Cox proportional hazards model assumption and Fine and Gray189 subdistribution proportional hazard assumptions.

The original SAP also proposed the use of propensity scores in the regression models to reduce the potential bias introduced by the fact that not all of the acute admissions will have an OPRAA completed. However, calculation of the propensity scores would have relied on the assumption that those with no OPRAA also had no CSDs. Examination of ICD-10 codes from the SMR01 data and examination of the prescribing data revealed that, among people with no OPRAA, some had dementia, and it is also possible that some of those who were terminally ill might have had delirium. As a result, it was agreed that assuming that people with no OPRAA also had no CSDs was not appropriate and so propensity score methods based on this assumption would have been an invalid approach to reduce bias.

All data analysis was carried out using SAS^® 9.4 software (SAS Institute Inc., Cary, NC, USA). (SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ^® indicates USA registration.)

The cohort

A 2-year study period between 1 January 2012 and 31 December 2013 was defined. During this time, there were a total of 17,151 acute medical emergency admissions to the study hospital, of which 12,378 had an OPRAA, giving a coverage of 72.2%. A total of 9331 were incident admissions, of which 6724 had an OPRAA, representing 72.1% of all incident admissions during this time. The 6724 people with an incident OPRAA admission between 1 January 2012 and 31 December 2013 are the core cohort under investigation in this study.

Patients with an incident OPRAA admission were, on average, aged 79.2 (95% CI 79.0 to 79.4) years, 56.3% (95% CI 55.1% to 57.5%) were women and 7.4% (95% CI 6.8% to 8.1%) were admitted from a care home. A total of 20.5% (95% CI 19.6% to 21.5%) of patients lived in the most deprived fifth of areas and 14.6% (95% CI 13.8% to 15.5%) lived in the most affluent fifth.

One or more CSDs were present in 35.4% (95% CI 34.3% to 36.6%) of the incident OPRAA admissions. Delirium alone was present in 15.8% (95% CI 14.9% to 16.7%) of admissions, known dementia alone was present in 7.8% (95% CI 7.2% to 8.5%) of admissions, DSD was present in 7.6% (95% CI 7.0% to 8.3%) of admissions and unspecified cognitive impairment was present in 4.2% (95% CI 3.7% to 4.2%) of admissions.

Table 4 compares people with and without CSDs in terms of their demographics, comorbidities and functional status. People with CSDs were significantly older than those without CSDs (mean age 82.1 years vs. 77.6 years, difference 4.4 years, 95% CI 4.1 to 4.7 years), with 41.5% of people with CSDs aged ≥ 85 years versus 21.1% of those without (difference 20.4%, 95% CI 18.1% to 22.7%). A total of 59.2% of patients with CSDs were women, compared with 54.6% of those without CSDs (difference 4.6%, 95% CI 2.1% to 7.1%). A total of 17.9% of people with CSDs were admitted from a care home, compared with only 1.7% of those without (difference 16.2%, 95% CI 14.6% to 17.8%), with 29.9% of people with dementia alone and 34.1% of people with DSD residing in a care home. There were no significant differences by socioeconomic deprivation status.

A total of 4846 (73%) people had an ADL assessment recorded. In general, the presence of any CSD was strongly associated with low functional ability, with 81% of patients with CSDs having a persistently low ADL or changed ADL; the corresponding figure for patients without a CSD was 58.2% (difference 22.8%, 95% CI 20.3% to 25.3%). Patterns of ADL score varied by CSD, with > 50% of patients with known dementia having a low ADL score prior to as well as on admission (persistently low ADL score), whereas almost 50% of patients admitted with delirium alone had a reduction in ADL score on admission (changed ADL score).

Primary outcome: being at home 30 days from discharge

The primary outcome of being at home 30 days from discharge was relevant to only those patients who were not admitted from a care home and who were discharged alive. Of the 6724 people admitted to an AMU with an OPRAA, 5570 were eligible for primary outcome analysis.

A total of 90.0% of people were living at home at 30 days from discharge. Table 5 shows the distribution of the primary outcome depending on whether or not patients had CSDs on admission and the type of CSD. The proportion of people living at home at 30 days was significantly lower in patients with CSDs than in patients without CSDs (81.7% vs. 93.4%, difference 11.7%, 95% CI 9.9% to 13.6%). Among the types of CSD, DSD had the poorest outcome, with only 69.1% of people in this group living at home at 30 days, mainly due to the fact that 25.6% of them were living in care homes.

Unadjusted OR estimates of the logistic regression showed that people with CSDs were less likely to be living at home 30 days from discharge than people without CSDs (Table 6). After adjustment for demographics and comorbidity (see Table 6), the model also showed that people with any form of CSD had a significantly lower chance of living at home 30 days from discharge, with ORs being particularly low for people with dementia (with or without delirium) (OR 0.33, 95% CI 0.25 to 0.46 for dementia alone and OR 0.18, 95% CI 0.16 to 0.25 for DSD) and slightly higher for delirium alone (OR 0.48, 95% CI 0.38 to 0.46) and unspecified cognitive impairment (OR 0.51, 95% CI 0.34 to 0.75). In addition, the pairwise multiple comparison test indicated that people with DSD were significantly less likely to be living at home at 30 days than people with other forms of CSD (p-values of < 0.017). The other pairwise comparisons for living at home at 30 days associated with delirium alone, dementia alone and unspecified cognitive impairment were not significant (p-values of > 0.242).

Increased age was significantly associated with a reduced chance of living at home 30 days from discharge (OR 0.76, 95% CI 0.71 to 0.81), as was having a CCI score of ≥ 6 (OR 0.32, 95% CI 0.23 to 0.45). Sex, socioeconomic deprivation and number of community-dispensed drugs were not associated with living at home at 30 days from discharge in univariate analysis and were not included in the adjusted logistic regression model based on not improving model fit assessed using the AIC (see Table 6).

Regardless of their cognitive status, patients with persistently low ADL score and changed ADL score were significantly less likely to be living at home 30 days from discharge than those with a persistently high ADL score (see Table 6) (OR 0.43, 95% CI 0.33 to 0.57 for persistently low ADL score and OR 0.65, 95% CI 0.57 to 0.83 for changed ADL score). Reflecting the strong correlation between CSD presence and worse ADL score, adjustment by ADL score somewhat attenuated associations between CSDs and living at home 30 days from discharge. However, patients with CSDs continued to have a significantly lower chance of living at home 30 days from discharge than patients without CSDs. After adjustment for functional ability, patients were significantly less likely to be living at home at 30 days if they had delirium (OR 0.57, 95% CI 0.45 to 0.79), dementia (either alone or superimposed on delirium) [OR 0.43 (95% CI 0.31 to 0.59) for dementia alone and OR 0.25 (95% CI 0.18 to 0.33) for dementia superimposed on delirium] or unspecified cognitive impairment (OR 0.59, 95% CI 0.40 to 0.89) (see Table 6). The sensitivity analysis for the complete-case ADL scores (see Appendix 4, Table 36) was in agreement with the analysis of data after multiple imputation for missing ADL score.

Mortality

Mortality outcomes were measured in in the entire OPRAA cohort from the date of admission. Mortality was very high in older people with emergency medical admissions, with > 10% of them dying within 30 days of admission and 30% of them dying within 1 year of admission. Mortality was particularly high in people with CSDs, with 40% of them dying within 1 year of admission. Among patients with CSDs, the highest mortality rate was recorded for patients in the DSD group, with 43.9% dying within 1 year of admission (see Appendix 4, Table 35).

Survival analysis and the Cox proportional hazards model

Survival time at 2-year follow-up was significantly lower in patients with CSDs, with 47.4% surviving at 2 years, compared with 66.5% of those without (Figure 2a, log-rank test p < 0.001). Increasing age was significantly associated with lower survival (see Figure 2b, p < 0.001), as was sex, with males being at higher risk (see Figure 2c, p < 0.001). There was significantly poorer survival for people admitted from a care home (see Figure 2d, p < 0.001) and people with a high comorbidity index (CCI ≥ 6) (see Figure 2e, p < 0.001). Survival in patients with persistent low ADL score was generally poor, with only 38.2% of the cohort patients surviving for the 2-year follow-up time; the corresponding figure for those with a changed ADL score and persistently high ADL score was 54.4% and 71.6% (see Figure 2f).

FIGURE 2.

Kaplan–Meier survival functions. (a) CSD groups; (b) age groups; (c) sex; (d) residential status; (e) CCI score; and (f) ADL functional status.

The results of the Cox proportional hazards model are shown in Appendix 4, Table 37. However, the assumption of proportional hazards over time was violated for several covariates, as indicated by the proportional hazard test, and reflected in the crossing of the survival curves in Figure 2a. We therefore concluded that the Cox model was misspecified.

Differences in mortality risks among the different types of cognitive spectrum disorder

The log-likelihood test statistics together with the AIC scores indicated that mortality risks associated with DSD are significantly different from the risk associated with delirium alone (Figure 3) (p = 0.002 for model B adjusted for demographics and comorbidity and p = 0.017 for model C additionally adjusted for ADL status), but it was not significantly different from the risk associated with dementia alone (p = 0.587 for model B and p = 0.257 for model C). At the same time, mortality risks in people admitted with unspecified cognitive disorder are significantly different from those for people with dementia alone (p = 0.017 for model B and p = 0.032 for model C), confirming that people with unspecified cognitive impairment become at an increased risk of mortality later on after admission compared with people with dementia alone.

FIGURE 3.

Cumulative incidence function for time to re-admission and mortality as competing risks within the 2-year follow-up time from discharge. The solid lines represent time to re-admission and the dotted lines represent mortality. (a) CSD groups; (b) age groups; (c) residential status; (d) CCI score; (e) ADL groups; and (f) number of drugs prescribed in the 84 days prior to admission.

Re-admission

Re-admission outcomes can be assessed only in people discharged from hospital alive. Table 10 shows in-hospital mortality for the incident OPRAA cohort broken down by CSD status and CSD type. Of the 6724 people admitted to an AMU with an OPRAA, 745 (11.1%, 95% CI 10.4% to 11.9%) died in hospital, with the remaining 5978 being discharged alive either to their private home or to a care home.

Patients within the incident OPRAA cohort who were discharged alive were, on average, aged 78.9 (95% CI 78.7 to 79.1) years, 56.7% (95% CI 55.4% to 58.0%) were female and 10.7% (95% CI 9.9% to 11.5%) were discharged to a care home. In addition, 20.4% (95% CI 19.4% to 21.4%) of patients lived in the most deprived fifth of areas (SIMD 1), whereas 14.8% (95% CI 13.9% to 15.7%) of them lived in the most affluent fifth (SIMD 5).

One or more CSDs were present in 33.6% (95% CI 32.4% to 34.8%) of the incident OPRAA admissions discharged alive, delirium alone was present in 14.8% (95% CI 13.9% to 15.7%) and known dementia alone was present in 7.6% (95% CI 7.0% to 8.8%). DSD was present in 7.1% (95% CI 6.5% to 7.8%) of the incident OPRAA admissions discharged alive, and a further 4.1% (95% CI 3.6% to 4.6%) were patients with unspecified cognitive impairment (AMT score of < 8 points in the absence of delirium or known dementia).

A total of 4307 (72%) people discharged alive had an ADL score recorded. In general, the presence of any CSD was strongly associated with low functional ability, with 79% of patients with CSDs having a persistently low ADL score or changed ADL score; the corresponding figure for patients without a CSD was 38.5% (difference 40.5%, 95% CI 37.7% to 43.1%). Patterns of ADL score varied by CSD condition, with > 50% of patients with dementia (either alone or superimposed on delirium) having a low ADL score prior to admission (persistently low ADL score), whereas almost 50% of patients admitted with delirium alone had a change in ADL score at admission (changed ADL).

Length of stay

People aged ≥ 65 years admitted to an AMU had an average LoS of 16.4 days (95% CI 15.7 to 17.2 days) (Table 18), with LoS being more than doubled for people with CSDs compared with those without CSDs (24.8 vs. 11.8 days; RR 2.1, 95% CI 1.97 to 2.24).

Unadjusted RR estimates of the gamma regression model indicated that people with CSDs had a LoS that was significantly longer than that for people without CSDs (Table 19).

After adjustment for demographics and comorbidity (see Table 19), the model showed that hospital stay for people with DSD is more than three times longer than for people without CSDs (RR 3.10, 95% CI 2.76 to 3.48), almost twice as long for people with delirium alone or cognitive impairment (RR 1.92, 95% CI 1.77 to 2.09, and RR 1.95, 95% CI 1.69 to 2.26, respectively) and significantly increased to a lesser degree for dementia alone (RR 1.75, 95% CI 1.56 to 1.96). In addition, the pairwise multiple comparison test indicated that LoS for people with DSD was significantly higher than for people with other forms of CSD (p-values < 0.001). The rest of the pairwise comparisons between delirium alone, dementia alone and unspecified cognitive impairment in terms of LoS were not significant (p > 0.508).

Sex was significantly associated with LoS, with men having a significantly shorter LoS than women (RR 0.90, 95% CI 0.84 to 0.95), and increased age was significantly associated with an increase in LoS (RR 1.18, 95% CI 1.15 to 1.20, per 5-year increase in age). Patients admitted from a care home had a significantly shorter LoS that those admitted from a private home (RR 0.38, 95% CI 0.33 to 0.42), as did patients living in the most deprived areas compared with those living in the least deprived areas (RR 0.87, 95% CI 0.78 to 0.96). Increased CCI was significantly associated with longer LoS (RR 1.08, 95% CI 1.06 to 1.10 per 1-unit increase in CCI), whereas an increase in the number of drugs dispensed in the 84 days prior to admission was significantly associated with shorter LoS (RR 0.89, 95% CI 0.86 to 0.92).

Length of stay was significantly associated with ADL functional status. Further model adjustment by ADL status showed that patients with persistently low ADL score and changed ADL score had a LoS that was significantly longer than for those with a persistently high ADL score (see Table 19) [RR 2.58 (95% CI 2.31 to 2.88) for persistently low ADL score vs. persistently high ADL score, and RR 2.64 (95% CI 2.45 to 2.84) for changed ADL score vs. persistently high ADL score]. Reflecting the strong correlation between CSD presence and worse ADL score (see Table 4), adjustment by ADL score attenuated associations between CSDs and LoS. However, patients with CSDs still had a significantly longer LoS than patients without CSDs. After adjustment for functional ability, patients had a significantly longer LoS if they were delirious (RR 1.53, 95% CI 1.41 to 1.68), if they had dementia [either alone (RR 1.51, 95% CI 1.35 to 1.70) or superimposed on delirium (RR 2.52, 95% CI 2.24 to 2.82)] or if they were cognitively impaired in the absence of delirium or dementia (OR 1.66, 95% CI 1.44 to 1.91).

The sensitivity analysis for the complete-case ADL scores (see Appendix 4, Table 42) was in agreement with the analysis of data after multiple imputations.

Coverage

By design, the OPRAA was not carried out in patients with brief admissions to exclude serious illness such as myocardial infarction in people with chest pain, who required immediate escalation to critical care or who were admitted for palliative care. OPRAA coverage was, therefore, 79.0% of all admissions and 77.3% of incident admissions. However, this compares favourably with most consented research cohorts including those with the highest coverage, such as Sampson et al. ,12 who, in their study of dementia prevalence, screened 88.2% of people aged ≥ 70 years admitted for ≥ 48 hours, and included 76.7% of patients (617 patients in total) after exclusions. For comparison, 88.3% of all admissions of > 48 hours in those aged > 70 years were included in this analysis.

Accuracy of brief assessment tools

The OPRAA used relatively simple instruments suitable for identifying delirium and cognitive impairment in a routine clinical context, which may not always match assessment using gold standard research instruments, although the OPRAA was carried out by trained, experienced specialist nurses. The sensitivities of the screening tools used in OPRAA have been discussed in the literature.

Only 31% of people diagnosed with delirium in this data set were CAM positive. This contrasts with the literature comparing CAM with a gold standard assessment of delirium, where CAM sensitivity ranges from 46% to 100%. 191 This probably reflects the difference between assessments carried out by dedicated staff during research studies and assessments such as OPRAA carried out in routine clinical practice where high workload and competing clinical demands constrain when assessments can be undertaken, and make it difficult to repeatedly return to carry out an optimal assessment (e.g. with an informant present). During the period of the study, the nurses applied the original scoring for the CAM in terms of CAM positivity requiring an acute and fluctuating course, which the CAM developers have since recognised is often difficult to assess when using the CAM in routine clinical practice. The CAM manual was updated in 2014 to allow two methods of scoring this criterion. 192 It states that the original scoring (‘and’) maximises specificity but reduces sensitivity in clinical use, and suggests using a course that fluctuates or is acute in order to maximise sensitivity at the cost of specificity. In addition, delirium by its nature is fluctuant, and others have found that CAM positivity varies over time in people with delirium, with, for example, 35% of assessments being CAM negative in people with hip fracture who were ever CAM positive. 193 As implemented in this study, CAM would therefore be expected to be highly specific but less sensitive, which is consistent with the observed patterns and with the conclusion of a recent systematic review of the CAM that ‘the use of these tools should not replace clinical judgement’. 194

Similar discussions are present for the AMT in the literature. Initial reports of the accuracy of the AMT in screening for cognitive impairment suggested that ‘The best cut-off point was 8, with less than 8 suggesting abnormal cognitive function’. 195 A recent systematic review and meta-analysis examines its accuracy when used as an instrument to screen for dementia. 196 In this meta-analysis, with a cut-off point of < 7 points, pooled analysis of the AMTs showed a sensitivity of 81% and a specificity of 84% for a diagnosis of dementia. As noted in this paper,196 a cut-off point of < 8 points is considered more usual in clinical practice. In the current study, we use a cut-off point of < 8 points to report unspecified cognitive impairment.

The cross-sectional nature of the assessment

The OPRAA was carried out within the first 24 hours of admission, and therefore captures prevalent cases of CSD at time of admission. Any changes in a patient’s cognitive status during the course of admission are not captured in the study design. For example, patients admitted to hospital with no CSDs or with known dementia alone may develop incident delirium through the course of their admission, and their outcomes will be narrowing the divide between the CSD subgroups in the reported analyses.

Lack of full dementia diagnostic workup

Data on results of further diagnostic workups for definitive diagnoses of dementia are not included. For this reason, the categories of the CSDs are based on the diagnoses that were known about at the time of admission (i.e. known dementia), along with diagnoses that can be attributed as a result of the brief assessment. It is therefore most likely that those patients with a low AMT score (unspecified cognitive impairment) are those with undiagnosed dementia.

Differences between admission and incident cohorts

Two cohorts were examined for the analysis. Within the admission (prevalence) cohort, each hospital episode is featured and therefore an individual may be counted a number of times with each re-admission to hospital. The incident cohort differs from the admission cohort in that it identifies individuals at the beginning of their interaction with acute health-care services and follows them through that journey, capturing all re-admissions and mortality. Outcomes reported from this incident cohort are therefore applicable to individual patients. Data from the admission (prevalence) cohort can be seen as reporting the impact that this population has on the acute hospital.

Lack of adjustment for other factors

The analysis reported here is unadjusted for other factors that may be associated with the outcomes, including physical health, function and nutrition. The OPRAA did not include evaluation of nutrition. It did include an assessment of ADL and variation in function may explain some of the observed associations. This is an area that requires further in-depth analysis because declines in ADL may reflect physical and/or cognitive impairment, making adjustment complicated, and any interaction between cognitive status and ADL may vary with time.

Descriptive analysis

We start with the cost of incident admission, defined as the first hospital admission between January 2012 and December 2013 given that the patient had not been admitted to a hospital in the 6 months prior to this admission. As shown in Figure 5, on average, patients with DSD incur a higher cost than patient groups. We can also observe the greatest variation in costs among this group of patients. By contrast, non-CSD patients have both the lowest average cost and the lowest variation in stay length.

FIGURE 5.

Incident hospital admission costs, by CSD conditions.

Patient cost is primarily driven by the length of hospital stay. Figure 6 shows the group comparison of the LoS. The trend is very similar to the patterns of cost. Patients with DSD have the longest stay, followed by patients with unspecified cognitive impairment, patients with delirium and patients with dementia. Patients without any CSDs have the shortest stay, as well as the smallest variation.

FIGURE 6.

Length of stay for the incident admission, by CSD conditions.

However, it is important to note that LoS is not the only factor that influences the total cost. The specialty that a patient was admitted or transferred to also plays a role. In order to examine the influence of specialty on costs, we selected 10 of the most commonly used specialties by our cohort of patients and grouped the rest into an ‘other’ category, resulting in 11 groups of specialties. We ranked these specialties by their day costs from the highest to the lowest: coronary care unit, high-dependency unit, ‘other’ specialties, general/acute medicine, communicable diseases, nephrology, gastroenterology, cardiology, respiratory medicine, geriatric medicine and geriatric long stay.

Figure 7 shows the cost ratio of each specialty to the least costly specialty (geriatric long stay), with the size of the marker indicating the specialty usage by patients, which is weighted by the total number of days that patients spent in a specific specialty. For example, we see that the total number of patient hospital days of geriatric long stay is about 19 times higher than that of coronary care unit, whereas the day costs of coronary care unit and high-dependency unit are over three times higher than the cost of geriatric long stay. However, these two units are among the least used specialties. The most commonly used are geriatric medicine, general/acute medicine and geriatric long stay. These three specialties account for > 77% of patient days for incident admissions.

FIGURE 7.

Ratio of specialty day cost to the lowest cost (geriatric long stay). 1, Coronary care unit; 2, high-dependency unit; 3, other specialty; 4, general/acute medicine; 5, communicable diseases; 6, nephrology; 7, gastroenterology; 8, cardiology; 9, respiratory medicine; 10, geriatric medicine; 11, geriatric long stay.

The distribution of hospital days among specialties is very different between CSD and non-CSD patients, as shown in Figure 8. Generally speaking, non-CSD patients are equally likely to be admitted or transferred to high- and low-cost specialties, whereas CSD patients, those with DSD in particular have a tendency to use low-cost hospital services. For CSD patients, 77% of hospital stays are in geriatric medicine or geriatric long stay, compared with 40% for patients with no CSDs. The specialty distributions for patient groups with different CSDs are very similar.

FIGURE 8.

Distribution of days by specialty and CSD conditions. The darker the colour, the higher the specialty’s day cost. From left to right, coronary care unit, high-dependency unit, other, acute/general medicine, infectious medicine, renal medicine, gastroenterology, cardiology, respiratory medicine, geriatric medicine, and geriatric long stay.

Figure 9 shows the pathways of individual patients’ moves between different specialties, omitting patients who had more than four episodes (5%). The individual-level data confirm what we observe in Figure 7. Nearly all patients, regardless of their CSD, were initially admitted to the specialty of general/acute medicine. However, CSD patients were more likely to have multiple episodes than non-CSD patients. In addition, they were more likely to be transferred to geriatric medicine or geriatric long stay, in which the day costs are significantly lower.

FIGURE 9.

Specialty pathways for individual patients by CSD group. (a) No CSD; (b) DSD; (c) delirium; (d) dementia; and (e) unspecified cognitive impairment.

To sum up, there are two main factors that influence the cost of incident admission for each patient: LoS and the specialties in which the patient was staying. We see that CSD patients tend to stay longer, which drives up their hospital costs. However, they are more likely to stay in specialties with lower day costs, which suppresses their costs.

Hospital costs over time

Given that the OPRAA cohort was followed for 2 years since the incident admission, we are able to look at hospital costs from a longitudinal perspective. This allows us to model the growth of patient costs over time across the cohort groups.

Descriptive analysis

Figure 10 shows the average total cost per patient for five groups of the OPRAA patients over different time periods following their incident admissions. In the first 12 months, the group with delirium and dementia had the highest average cost. However, the cost grows more steeply for the unspecified cognitive impairment group, so that it exceeds the delirium and dementia group. The delirium group also catches up by the end of the follow-up period. The flattening of the delirium and dementia group’s average total cost is likely to be explained by the significantly higher mortality rate among this group after 1 year.

FIGURE 10.

Cumulated total cost over time, by CSD conditions. This graph shows the average total cost per patient at five time points: 30 days, 90 days, 12 months, 18 months and 24 months from index admission. It is calculated by adding the costs of all hospital days from the index admission to the respective time point, divided by the number of patients.

Gamma regression models of the cumulative costs over time

In this section, we model the cumulative cost as a function of individual characteristics in order to estimate cost differences between different groups, after controlling for the differences in the composition of patient groups. The results are presented in Table 21, with the gamma models estimated over different time periods. We can see that the cumulative total costs for patients with any CSD are significantly higher than those for non-CSD patients if we constrain the time period to up to 6 months (models I and II). If we prolong the follow-up period to 2 years, only patients with delirium alone cost significantly more than those in the non-CSD group. There is no difference in costs for patients with other CSDs compared with the non-CSD group. The estimated coefficient for patients with DSD is significantly higher than the coefficients for patients with other CSDs in model I only. There appears to be no significant difference between patients with different CSDs if the follow-up period is > 3 months.

We can plot the estimated log costs and their CIs obtained from the gamma regression models based on different lengths of time. Figure 11a illustrates how patients with DSD differ from patients with no CSDs across models. It is clear that the gap between these two groups gets narrower, becoming insignificant in model IV, which accounts for all costs that were cumulated within the 2-year period. Figure 11b compares the delirium alone with the delirium and dementia group. Although patients with DSD start with a significantly higher cost, their costs seem to accumulate at a lower rate over time than patients with delirium alone.

FIGURE 11.

Cost estimates and CIs from the gamma regression models. (a) Patients with DSD vs. patients with no CSDs; (b) patients with delirium alone vs. patients with delirium and dementia.

Longitudinal growth model

The gamma regression models provide snapshots of the cumulative costs over different follow-up periods, giving a description of how the influence of CSDs on the cumulative cost changes over time. To gain a better understanding, we can model the growth trajectory of the cumulative costs by using a growth modelling approach. The results are presented in Table 22.

The unadjusted model is simply a longitudinal random slope model, in which we allow the influence of time (months) to vary across patients, and allow the CSD variable to influence both the intercept and the growth rate. The adjusted model shows the longitudinal estimates from a joint random-coefficient model in which the longitudinal model is jointly fitted with a survival function (results are presented in Appendix 5, Table 43). This takes into account the fact that the growth trajectory of hospital cost is related to mortality. Generally speaking, the estimates from these two models are fairly close; however, we do see some differences in the estimated influence of the CSD variable on the growth rate.

Figure 12 shows the predicted linear growth trajectories estimated by the unadjusted and adjusted models for different CSD groups. The predicted lines are plotted separately in four subfigures to show the differences more clearly. Figure 12a shows the trajectories for the delirium and dementia and no-CSD groups. We see that the estimated starting costs and the growth rates for both groups from the adjusted model are higher than the costs from the unadjusted model. Moreover, the magnitude of the cost difference between these two groups shrinks over time in both models, but the adjusted model has a higher rate whereby the cost difference becomes non-significant by the end of the 2-year follow-up period. This is consistent with the estimates of the gamma regression model in Gamma regression models of the cumulative costs over time (see model IV in Table 21). In contrast, the cost difference in the unadjusted model is still statistically significant (p < 0.05). Figure 12b shows the comparison between the unspecified cognitive impairment and no-CSD groups. According to the unadjusted model, the cumulative costs for these two groups grow almost in parallel. However, in the adjusted model, the no-CSD group clearly has a significantly higher rate, allowing it to quickly catch up with the unspecified cognitive impairment group.

FIGURE 12.

Predicted linear growth lines of the cumulative costs. (a) Delirium and dementia vs. no CSD; (b) unspecified cognitive impairment vs. no CSD; (c) delirium alone, dementia alone and unspecified cognitive impairment; (d) delirium and dementia vs. delirium alone and dementia alone.

In Figure 12c, we see that the growth trajectories of the delirium alone and dementia alone groups almost overlap in both models. According to the unadjusted model, they are also very similar to the growth trajectories of the unspecified cognitive impairment group. However, if we look at the adjusted model, despite having close starting points, the unspecified cognitive impairment group has a relatively lower growth rate and a significantly lower cumulative cost by the end of the follow-up period (p < 0.05). Figure 12d shows that the delirium and dementia group has a significant higher starting cost than the delirium alone or dementia alone groups. However, given it has a lower growth rate, it will be overtaken by the other two groups, but the predicted difference between the delirium and dementia and delirium alone or dementia alone groups is not statistically significant in either model.

Care and treatment

The role of the care and treatment was referred to 59 times in reference to hospital admissions and 62 times in reference to outcomes from an admission. Box 1 contains examples.

The survey highlights that, while in hospital, carers and those living with CSDs expect to:

• be treated with dignity and receive the same level of care as other patients

• experience minimal waiting time, as this can increase anxiety

• be treated appropriately (e.g. are there problems with swallowing, will the treatment increase anxiety?)

• have access to family and carers (i.e. someone who understands and knows the patient).

It also highlights that aspects of lack of focus in care and treatment can have a negative impact:

• Medication is administrated without the carers and patient understanding why and for what.

• Inappropriate medication is administrated.

• Pain is not managed appropriately, or the patient is asked about pain when they are not able to communicate their needs.

• There is a lack of focus on rehabilitation, which leads to a reduced level of mobility and abilities post admission. Some respondents felt that there was very little focus on rehabilitation at all, as if it had no value for this group of patients.

• There is not sufficient focus on ensuring that the patient receives the right fluids and nutrition while in hospital.

• Other aspects of care not directly related to the admission were missed out, such as mouth hygiene.

A positive care situation after the admission was mentioned by several respondents as a positive outcome, including:

• The person feeling settled again after returning from hospital.

• Getting back home, this being their private home or the place in which they were cared for prior to admission. For several patients, this needed to happen sooner rather than later, although some were also concerned that they were discharged too early.

• That support was in place at discharge to ensure that the transition was successful.

• That any change in a care package was evaluated and put in place prior to the discharge.

• That they were back to the level of health they had prior to the condition that necessitated a hospital admission.

• That a proper follow-up was carried out with all involved, including family and carers, regarding any changes in care plan and medication as a result of the admission.

Many respondents shared negative outcomes from a hospital admission. These fell into categories such as:

• The person was less mobile owing to additional illness contracted as a result of the admission.

• Contracting a urinary tract infection during the admission.

• Being more confused or having delirium as a result of the admission.

• Dying as a result of the admission.

• Being on increased medication, or inappropriate medication, after an admission.

• A lack of rehabilitation after a hospital admission.

Hospital staff

The role of staff was referred to 109 times in reference to hospital admissions and eight times in reference to outcomes from an admission (Box 2 contains examples).

Staff who demonstrated the following behaviours were part of ensuring that the hospital admission was a positive experience:

• provided assurance and comfort

• respected the patient

• were trained

• recognised the family as key to the care of the patient

• were flexible

• provided assistance when eating and drinking

• were not patronising.

Staff who demonstrated the following behaviours contributed to a negative experience around a hospital admission for both carers and the person with dementia or confusion:

• did not actively encourage the patient to eat and drink

• changed a lot

• did not listen to carers relating to the individual patient’s needs, pain, etc.

• did not recognise dementia and its challenges or lacked the appropriate training

• were too busy, or too thin on the ground

• did not care enough

• did not treat the individual with respect.

Staff who demonstrated the following behaviours had a positive impact on the outcome of an admission:

• understood and recognised dementia

• engaged with carers and family and made them a part of the care

• took the time to get to know and understand the patient

• treated the patient with dignity and respect.

Staff who demonstrated the following behaviours had a negative impact on the outcome of an admission:

• labelled patients as difficult

• did not understand what dementia is

• did not engage with carers in the treatment.

Communication

Communication was referred to 57 times in reference to hospital admissions and 31 times in reference to outcomes from an admission (Box 3 gives examples).

Several respondents mentioned communication as being key to a positive experience and a positive outcome from a hospital admission: communication with the person admitted but especially communication with the family and carers of the person admitted.

The aspects of communication that are highlighted as beneficial to ensure a positive stay are:

• Communication with the patient includes talking with the patient not across the patient.

• Constant reassurance to both the patients and the carers involved.

• Involvement of carers throughout the admission through to discharge, to be kept informed of changes in treatment and/or health.

• When hospital staff consider the carers and family as part of the wider caring team and treat their knowledge with respect. In most cases, they have the most in-depth knowledge about the person.

• A focus on keeping communication simple so carers can understand, avoiding excess jargon.

Often carers commented that communication had failed, listing issues such as:

• The patient being dismissed or not being included in communication or being spoken to in a condescending way.

• Disregarding communication from carers or simply not taking the time to listen to the carers, which led to problems during admission. These could be problems around eating or taking medicine or other functional and/or emotional problems.

Communication comes through as key to a positive admission and outcome: communication throughout the admission from arrival, while in hospital and around the plan for discharge, and communication including the immediate carers and family, but also the wider care team, so that those who are involved after discharge are fully aware of any changes and needs arising from the admission.

Some of the negative experiences shared included patients with dementia being discharged without their immediate family and/or carers being informed and changes to medication while in hospital without informing carers.

Family and carers

The role of carers and family was referred to 70 times in reference to hospital admissions and 21 times in reference to outcomes from an admission (Box 4 gives examples).

A large number of carers and family members indicated a strong desire to continue to take an active part in the care while the person with dementia was in hospital. They felt that taking part would:

• help to keep a routine that the patient was used to outside hospital

• allow them to give valuable input into the care of the person

• allow them to provide some of the care, easing the situation for both the person with dementia and the hospital staff.

To enable these, there is a demand for greater access with regard to visiting hours, car parking and information (communication).

Carers and family also voiced a need for them to be reassured throughout the hospital admission and beyond, and that it is recognised that this time is stressful for them too.

Often, carers felt they were being ignored and that their lack of involvement had a detrimental effect on the hospital admission. They felt that if they were more closely involved they could also assist in the identification of signs of further illness.

Carers indicated that, after a hospital admission, their role was more difficult than before and it was therefore not always possible for the patient to return home.

Health and functionality

The role of health and functionality was referred to 19 times in reference to hospital admissions and 44 times in reference to outcomes from an admission (Box 5 gives examples).

While in hospital, the main comment around health and functionality that ensures a positive outcome relates to mobility and hydration: ensuring that the patient is kept mobile and properly hydrated. There were several mentions of problems relating to health and functionality during a hospital admission:

• Increased confusion and delirium while in hospital.

• Pain was not managed appropriately.

• Lack of stimulation.

• Lack of rehabilitation.

• Going in with one condition and contracting additional conditions during the hospital admission.

A positive outcome relating to health and functionality can be summarised as:

• Dementia has not worsened.

• The patient is back to the same or close to same level of mobility and functionality as they had prior to the admission.

However, the experiences shared by respondents indicate that often the patient’s dementia has worsened as a result of the hospital admission, their mobility is reduced and their functionality and QoL have deteriorated, so the patient cannot return to their previous living arrangements. If they do, the carer finds the situation unmanageable.

Hospital environment

The role of the hospital environment was referred to 56 times in reference to hospital admissions and once in reference to outcomes from an admission (Box 6 gives examples).

The hospital can help to ensure that a hospital admission is a positive experience for the person with dementia or confusion and their family and carers by:

• Providing flexible visiting hours. This was the most stated wish by family and carers, and several would like the ability to stay 24 hours per day/7 days per week to assist in the care of their loved one.

• Ensuring that the environment is safe and secure, so the patient feels safe and the family and carers feel at ease.

• Keeping the environment as familiar and navigable as possible. This includes not moving the patient multiple times and also keeping the staff consistent throughout the admission. This also includes the ability to make the patient feel ‘at home’ by bringing some familiar items in.

• Keeping the environment as quiet and calm as possible, even having the option of a private room.

• Easing wayfinding for the patient, such as clear access to the bathroom.

• Providing access for carers; this includes the availability and price of parking.

• Providing facilities for carers within the hospital, including a room to retreat to and a comfortable chair to sit on when spending many hours on the ward.

Several respondents stated that the person with dementia found the admission to hospital very confusing, and the environment of the hospital added to the state of confusion. Some examples, all of which would add to the distress and confusion, were:

• The patient not getting enough fluid, as fluid was left out of reach.

• The patient could easily get lost.

• The environment would be too noisy and too hot.

EMBASE

Date range searched: 1980–2016, week 4.

Date searched: 29 January 2016.

MEDLINE [Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations and Ovid MEDLINE(R) <1946 to Present>, Ovid MEDLINE(R) Daily Update <January 26, 2016>]

Date range searched: 1946 to 26 January 2016.

Date searched: 14.15 on 27 January 2016.

CINAHL Plus

Date range searched: 1946 to 26 January 2016.

Date of search: Friday 29 January 2016 at 12:05:42.

PsycINFO (via Ovid)

Date range searched: 1806 to January week 4 2016.

Date searched: 28 January 2016.

Cochrane Database of Systematic Reviews

Date range searched: 1946 to 26 January 2016.

Date searched: 1 February 2016.