Undiagnosed dementia in primary care: a record linkage study

Article history

The research reported in this issue of the journal was funded by the HS&DR programme or one of its preceding programmes as project number 14/70/96. The contractual start date was in January 2016. The final report began editorial review in May 2018 and was accepted for publication in March 2019. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HS&DR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

none

Copyright statement

© Queen’s Printer and Controller of HMSO 2020. This work was produced by Aldus et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2020 Queen’s Printer and Controller of HMSO

Dementia

Dementia is a syndrome of progressive loss of both cognitive and everyday functional ability. 1 It is associated with a number of underlying brain pathologies, including Alzheimer’s disease-type pathologies, cerebrovascular disease and Lewy body disease among many others. 2,3 Memory impairment is a characteristic of Alzheimer’s disease, but other dementias, particularly those presenting at younger ages, can have different profiles of cognitive impairment and non-cognitive symptoms, including psychosis, personality changes and behaviour changes.

Dementia has an enormous social and economic impact;4,5 as it is a major cause of disability, there is a need for long-term residential care and it is increasingly recognised as a cause of death. Dementia is more common after the age of 65 years, after which its incidence roughly doubles every 5 years. 6

It is estimated that there are 50 million people living with dementia globally,1 with an economic cost of US$1T. 7 The prevalence of dementia is difficult to estimate, but as of 2018, in England, there are around 650,000 people living with dementia and around 540,000 carers of people with dementia. 8

Diagnosis of dementia in the UK and internationally

Dementia can be formally diagnosed when a patient meets certain clinical criteria. These criteria vary but typically require (1) impairment in at least two aspects of cognitive function and (2) an associated impact on daily life, such as disability or loss of occupation caused by cognitive impairment. A diagnosis of dementia is valid only where symptoms of the impairment are not reversible (e.g. symptoms arising as a result of depression or from the side effects of prescribed medication) or transient (e.g. delirium), and cognitive impairment is likely to worsen as the underlying pathology advances. There is no single accepted operational definition of dementia, with different interpretations and thresholds used to make a diagnosis in different research and clinical contexts.

The benefits and harms of receiving a clinical diagnosis for a person with dementia are not well understood and are likely to vary substantially between individuals. 9 There is usually no medical treatment that will alter the course of the neurodegeneration, but some symptomatic treatment to improve cognitive function is available. It is also suggested that a clinical diagnosis of dementia may improve quality of life by enabling people with dementia and their families and carers to receive appropriate support and information that may help with daily living, delay institutionalisation or avert health crises. 10

Diagnosing more people with dementia has been a priority in dementia care policy both in the UK and internationally. Across the developing world, the vast majority of people with dementia do not have access to services providing dementia diagnosis. 11 The World Health Organization Global Action Plan 2017–2512 set a global target (global target 4): in at least 50% of countries, 50% of the estimated number of people with dementia will be diagnosed by 2025 (as a minimum). In the UK, the 2009 National Dementia Strategy set a target that 67% of people living with dementia should have a formal diagnosis recorded in their primary care record.

Why do people with dementia remain undiagnosed?

Despite the difficulties in estimating diagnosis rates outlined above, it is certain that there are a substantial number of people living with dementia who are not diagnosed. The reasons for this among individual patients, their families and their doctors have been extensively explored in the qualitative literature. There are many complex factors influencing presentation and diagnosis. 22,23 For example, some individuals will be fearful or antagonistic towards seeking medical advice. They may deny that there is a problem and reject suggestions from family members that they should see a doctor or get help. Social and cultural factors are also important. For example, some ethnic groups may regard dementia as something to be concealed within the family network. Some of the stated ‘barriers’ to presentation are the belief that dementia symptoms are a normal aspect of ageing, the ability to manage without help, stigma, lack of knowledge around dementia, lack of treatment, as well as the possibility that a diagnosis of dementia would do harm and be a negative development for the individual and their context (possibly from observation of support to others in similar situations). GPs consistently attribute missed diagnosis to a lack of education and support for themselves and carers, their own concern about the availability of secondary or support services, financial and time constraints, and the difficulty in making the diagnosis or breaking bad news. 24

In contrast, acute events such as hospitalisation or interaction with health-care services, change in circumstances, behavioural disturbance or concerned family can act as triggers to assessment and ultimately diagnosis. 22,25,26

Our conversations with public and patient involvement (PPI) representatives have reinforced the possibility that help-seeking is inhibited by a fear of the consequences of diagnosis and a determination among some family units to cope without outside interference until help-seeking becomes unavoidable (see Chapter 12). Management of other health conditions was also suggested by our PPI representatives as a potential distraction that meant cognitive impairments are not fully explored by GPs (so-called diagnostic overshadowing) and the possibility that formal diagnoses are not made if they are seen to be likely to impact on care arrangements in which an individual and those around them are coping well.

The benefits and harms of diagnosis for people with dementia

Central to the debate around the need for timely diagnosis (hence the political imperative for increased diagnosis rate) is whether a diagnosis is ultimately helpful or harmful to a person with dementia and those around them. At present, there is no treatment that can delay the progress of any dementia. Symptomatic treatments exist for various stages of Alzheimer’s disease that have been shown to improve cognitive function and ability in some cases. Although effects are typically small and side effects can be significant, such treatments are generally regarded as offering a net benefit for a significant number of people with Alzheimer’s disease. Non-pharmacological interventions to support people with dementia and their families and carers have stronger support.

However, the fact of the existence of programmes to support people with dementia who seek diagnosis does not automatically translate into a net benefit of diagnosis.

Psychosocial and occupational benefits and harms of dementia diagnosis are also reported.

Although many people with dementia report feeling relief at understanding their symptoms and being able to plan for the future, they may also experience increased tension while negotiating their new identity, feel the need for increased social and psychological support (which may or may not be available to them) or have a heightened fear of what the future holds for them. 22

Hence, to understand the impact of diagnosis it is essential to compare people living with dementia who are diagnosed with those who are not diagnosed. At the time of writing, two epidemiological studies of the benefits or harms of dementia showed no improvement with respect to mortality or functional decline, but did suggest that those seeking referral or diagnosed earlier were more likely to transfer to institutional care. 27,28

Why is further research needed?

Globally and nationally, substantial resources are being directed towards diagnosing more people with dementia. Nationally, this has led to a significantly higher number of people with dementia being diagnosed, although there remains a substantial gap between the expected number of cases and the number of cases reported in some areas.

The James Lind Alliance priorities for dementia research include the consequences of the emphasis on early diagnosis, with questions about the impact of early diagnosis and what the appropriate role of primary care in this would be (this was rated in the top three priorities). 29

Local authorities and CCGs are tasked with examining areas where the number of patients diagnosed is significantly lower than the target based on the number expected from prevalence data. Yet local commissioning groups, GPs and the Alzheimer’s Society have expressed concern over the validity of comparing local dementia registers with national or international estimated prevalence data. Additional concerns are the possibility of overdiagnosis and the large number of people who are known to services but who are not diagnosed. 30 Without understanding any inequalities in access to dementia diagnosis, it will be difficult to design strategies to make this more accessible to those who would benefit. Lessons may be learnt from the management of other chronic long-term illnesses.

Reviews of whether or not dementia meets the internationally accepted criteria for the introduction of a screening programme also highlight the lack of evidence for the benefit of diagnosis in the context of currently available treatment, and the lack of evidence for the acceptability or effectiveness of any potential screening or case-finding programme. 31,32 Despite this, the case is often forcefully made for earlier diagnosis of dementia based on supposed social and economic benefits without any direct evidence for these,10 and targets for the numbers of people who health services are expected to diagnose remain in place.

Hence, a considerable number of questions remain unanswered. Central to these is the characterisation of the population with undiagnosed dementia and estimates of the possible social and economic benefits of diagnosis, which we begin to address in the current work.

Broad aims

Our main aim was to estimate the prevalence and distribution of dementia not recorded in primary care among the UK population aged ≥ 65 years. To achieve this, we linked data from an epidemiological study to primary care records of the study participants. The study used was CFAS II,16 which is the most recently recruited cohort study representative of the primary care population of England to study dementia and which underpins current UK dementia prevalence estimates. CFAS II is fully described at www.cfas.ac.uk/ (accessed 11 February 2019).

In particular, we linked and compared objectively and independently conducted CFAS II dementia assessments with records of cognitive complaints and dementia diagnoses recorded in primary care.

Crucially, this means that we were able to establish the primary care dementia diagnosis status of a population-representative cohort of people who independently met an objective, standardised and validated assessment of dementia (henceforth ‘people with dementia’). This enables us to estimate the prevalence, correlates and consequences of undiagnosed dementia in the older population of England.

This study is reported in accordance with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

Specific research questions

This linkage was used to address the following specific questions.

Study design and objectives

This study, henceforth the Cognitive Function and Ageing Study II Dementia Diagnosis Study (CADDY), is a record linkage study of a population-representative cohort of older people in England. In short, we linked data from GP records with epidemiologically determined dementia diagnoses from a population-representative cohort of older people to estimate the prevalence of undiagnosed dementia and the causes and consequences of diagnosis among people living with dementia.

To answer our research questions our study required a sample of people with dementia who:

• are representative of the population with dementia

• are known to have dementia but with their dementia status determined independently of health-care services

• had primary care records that we could access and extract data from.

Recruiting this sample specifically for this research was prohibitively expensive, and CFAS II is the only study in England that meets these criteria. Although other English studies of older people, such as the English Longitudinal Study on Ageing, do include people with dementia, dementia is typically not systematically assessed using validated algorithms (www.elsa-project.ac.uk/; accessed 14 June 2019). Epidemiological studies do not often include the population in residential care settings and often it is not possible to access health records.

CFAS II consented participants for medical record linkage but did not attempt to collect any data from primary care beyond that needed for recruitment into the study and ongoing contact with participants. Hence, the main objectives of CADDY were to:

• identify a subsample of the CFAS II participants for inclusion in the current study, the ‘CADDY’ sample

• contact the GPs of these participants (their GP at the time of study recruitment) and extract data on recorded cognitive concerns, referral and dementia diagnoses

• encode and link data extracted from primary care medical records to Cognitive Function and Ageing Study (CFAS) assessment and interview data, and additional data supplied by CFAS on mortality and area-level deprivation from the Office for National Statistics (ONS)

• conduct an analysis of the linked data set to estimate each of the research questions above.

Figure 1 illustrates the relationship between the pre-existing CFAS II (shaded boxes) and CADDY (white boxes).

FIGURE 1.

The design of CADDY with respect to the CFAS II.

CFAS II design

CFAS II is an epidemiological study of ageing and dementia conducted in three geographic regions of England. The study has been described in detail previously,16 but, in short, the study aimed to recruit 7500 individuals aged ≥ 65 years, selected at random from primary care registers in three geographic areas, with samples of around 2500 individuals aged ≥ 65 years in each area (Nottingham, Newcastle and rural Cambridgeshire). The sample was stratified to recruit equal numbers of those aged 65–74 years and those aged ≥ 75 years.

Those with severe cognitive impairment or living in care homes were included in the study. The only exclusion criterion was presence of a terminal illness resulting in life expectancy of < 6 months at time of sampling, as assessed by the participant’s GP.

Each participant received a letter informing them about the study and was then visited up to three times by a CFAS interviewer. Once consent had been ascertained, the participant and, where appropriate, an informant, or, in the case of those who could not give consent, a proxy, were interviewed using standardised assessments as described in Chapter 5 (with full details available at www.cfas.ac.uk; accessed 26 June 2019).

The W1 interview data were available for 7762 participants of the 14,242 participants who were approached. The W1 assessments were conducted between 2008 and 2011. Before carrying out a 2-year follow-up assessment, each participant’s primary care practice was contacted to determine their vital status. Each surviving participant was then invited to undertake a similar assessment approximately 2 years after W1 [wave 2 (W2), conducted between 2011 and 2013]. A total of 5156 participants took part at W2.

At the W1 assessment, consent for the research team to access medical records and to link this to study data was sought. Although consent to the CFAS II could be made by proxy, medical record consent was sought only if the participant had sufficient mental capacity to consent personally at W1, meaning that those with moderate or severe dementia are under-represented among those with record linkage at W1.

Primary care data collection

For all participants we requested, from their general practice, information from their primary care records about different aspects of dementia and cognitive impairment. The case report form (CRF) is appended (see Appendix 2). This was to be completed by practice staff either electronically via a web-form or by hand and posted back to the study team.

Data collection

CADDY outcome measures

Dementia status

Each participant is defined as living with dementia or not having dementia at each wave, as defined by the CFAS II classification. So, based on this and the definition of diagnosis status above, we classify people at each wave as having:

• diagnosed dementia if they meet the CFAS II criteria for dementia and have a record of a dementia diagnosis in their GP notes on or before the interview date, or

• undiagnosed dementia if they meet the CFAS II criteria for dementia but have no record of a dementia diagnosis on or before the interview date.

Note that this definition treats the CFAS II diagnosis as a gold standard, against which we are comparing the accuracy of GP diagnoses. Although the CFAS II study diagnosis has been validated against the DSM-IIIR diagnosis of dementia, we acknowledge that it may not be sensitive, that is there are likely to be people with dementia who are not classified as such. However, the CFAS II diagnosis is thought to be specific and to have a high positive predictive value, in that those reported as having dementia using this algorithm are very likely to have dementia. Hence, it is well suited to the purpose of identifying a cohort of people living with dementia. The CFAS study assessment is also administered in a standardised way across all participants and so will not introduce a bias when estimating the effects of factors potentially associated with diagnosis.

Selection and coding of covariates

CFAS II assessed many social, demographic, clinical and health service use factors for each participant. Many were included as potential predictors or consequences of dementia diagnosis. These were selected based on a review of previous literature, theoretical considerations and discussions with PPI advisory groups. Data collection methods are fully described in Chapter 3, CFAS II design.

Research question 1a: distribution of diagnosis among people with dementia

The distribution of the four levels of primary care diagnosis status among participants with dementia and the characteristics of the undiagnosed population were estimated separately for each wave, using inverse probability weights to correct for missing diagnosis status.

Each potential covariate was tabulated across each outcome group for all people with dementia, combining records from W1 and W2. Inverse probability weights were applied to estimate proportions (see Missing data).

Research question 1b: diagnosis among people without dementia

The number of people at W1 and W2 who did not have dementia (by the CFAS II criteria) but did have a diagnosis in their primary record was reported. The sample used for this research question included participants selected as control and participants selected because they had dementia at W2 but did not have dementia at W1. For people without dementia but with a diagnosis at W1, their W2 dementia status was also described. Numbers were very small and no further analysis was conducted with this group.

Research question 2: correlates of diagnosis

Research question 3: the time to diagnosis

To estimate the average time between dementia incidence and clinical diagnosis, a time-to-event analysis was performed. The analysis was restricted to patients who had developed dementia between W1 and W2 and therefore for whom an approximate date of onset was known. Patients were entered into the analysis 4 years prior to the date of their W2 interview, and were included until the earliest of death, new diagnosis, or being censored on the date that their practice submitted data to CADDY. The timescale used was years since entry into the analysis. The cumulative incidence of new diagnosis, and hence the median time to diagnosis (time at which 50% of cases had been diagnosed) was then estimated, with death prior to diagnosis considered as a competing risk.

To estimate predictors of new clinical diagnosis among those undiagnosed, we defined an ‘index assessment’ as the first CFAS II assessment at which a participant met the study criteria for dementia. In the case of patients who were undiagnosed at their index assessment, Cox proportional hazards regression was used to estimate predictors of new diagnosis over the subsequent 2 years, using the same set of covariates as for research question 2 (see Research question 2: correlates of diagnosis).

Research question 4: the consequences of diagnosis

To estimate the effect of diagnosis on people with dementia (compared with remaining undiagnosed), we selected outcomes that reflected hypothesised benefits or harms of clinical diagnosis based on previous literature and discussions with carers of people with dementia. These were mortality, anxiety, health-related worry, social contact, new institutional placement and stays in hospital.

Common issues for all analyses

Missing data

Missing outcome variable

At W1, participants were asked to consent to medical record linkage only if they had capacity to do so personally. Figure 2 shows the link between measures of dementia severity and diagnosis data availability at each wave. Participants without consent for linkage had more severe dementia at W1 and were also the least likely to have been able to complete cognitive tests or directly answer other interview questions. However, those with severe dementia at W2 were often consented at W1, at which time they did have capacity and so are included in the study. We correct for cases without consent, either through the use of multiple imputation or by the use of inverse probability weights, depending on the analysis (the generation of weights is discussed in Chapters 7–11 and multiple imputation is discussed in Multiple imputation by chained equations).

FIGURE 2.

The proportion of people with dementia at CFAS II W1 and W2 for whom diagnosis data were available in CADDY, stratified by BDS quintile (higher value = more severe dementia). (a) W1 and (b) W2.

Missing predictor variables

Proxies were available to answer some but not all questions. Variables are typically missing when either (1) the respondent is not able to answer and there is no proxy or (2) the questions rely on an informant assessment and no informant exists or is available.

Multiple imputation by chained equations

Multiple imputation by chained equations (MICE) was used to infer the study diagnosis status and other covariate data when they were missing. 42 For continuous, binary and ordinal variables, predictive mean matching was used to impute new values; for nominal values with more than two levels, multiple logistic regression models were used. In short, MICE works by imputing missing cases of each variable in turn using a regression model based on all other variables in the planned analysis, including the outcome. When predictive mean matching is used, the regression model is applied to all individuals and for each missing value an imputed value is selected from the complete cases with similar fitted values. When multinomial regression is used, imputed values are sampled directly using the predicted probabilities from the model. MICE allows imputed values for predictors to be used and iterates imputations on each variable a fixed number of times before imputed values are recorded for analysis. In this way, any pattern of missing data, including the primary outcome of interest, can be imputed.

For imputation models, data from W1 and W2 are combined. In practice, this means that the study diagnosis status among those not able to consent at W1 is imputed based on the relationship between diagnosis status and other predictor variables at W2 and among W1 cases when consent was taken. As patients with severe dementia at W2 were able to give consent to medical record linkage on recruitment to the study, this means that the imputation model for diagnosis study in severe dementia can be estimated and applied to W1 participants. The number of imputed cases is specified alongside each analysis. Only data for those who are alive and participated in the relevant CFAS II wave are imputed.

Clustering

Eighty-five participants received study diagnoses of dementia and were included in both W1 and W2. In all inferential analyses (regression models) in which waves are combined, we corrected for this using robust standard errors to account for within-participant correlation, by specifying the ‘cluster’ option in Stata^® 14.1 (StataCorp LP, College Station, TX, USA). 43

Software

All analysis was conducted using Stata 14.1.

Primary care data collection

Our research relied heavily on persuading GP practices to extract and supply data from patient records. This was successful, but at considerable cost in terms of researcher time, and was dependent on building and maintaining good relationships with our contributing practices. The remainder of this chapter is dedicated to a quantitative and qualitative description of the reality of the process of collecting data from primary care, along with insights and suggestions arising from our experience that should be of help to future researchers undertaking similar tasks.

Participant data collection statistics

Participant data were returned by general practice staff (n = 411), CRN representatives at general practices that did not have capacity to participate (n = 126) or in the form of PCSE-returned hard-copy records (n = 30). Records for 31 patients were not returned.

Consent to link medical records and primary care response rate among people with dementia

Of 7762 participants in CFAS II, 461 met the study criteria for dementia at the W1 assessment. Of those, 285 (62%) gave consent to have their medical records accessed by the study team and linked to study data. At W2, 5265 participants remained in the study, of whom 274 were assessed to have dementia at the W2 assessment, and of this number 249 (91%) had given consent to link their medical records.

Table 3 shows the breakdown of diagnosis status at the time of the W1 and W2 interviews for each study participant who took part in assessment at that wave, as well as the number with dementia in each wave who could not be consented for medical record linkage or for whom primary care data were not available.

Primary care diagnosis status was received for 269 (94%) CFAS II participants with dementia who consented to linkage at W1, representing 58% of all those with dementia at W1. Data were received for 96% of those with dementia at W2 and consent to linkage, representing 87% of those with dementia at W2 (Table 3). Table 3 shows the breakdown of diagnosis status at the time of the W1 and W2 interviews for each study participant who took part in assessment at that wave, as well as the number with dementia in each wave who could not be consented for medical record linkage or for whom primary care data were not available.

At W1, those who gave consent were less severely functionally impaired. At W2, the link between dementia severity and consent to linkage was greatly reduced, as many participants with severe dementia at W2 had been able to consent at W1. There is an expected strong relationship between dementia severity and primary care diagnosis status at both W1 and W2. Previous studies have also shown that people with more severe dementia are more likely to be diagnosed, so missing more severe cases might cause the proportion with a diagnosis to be underestimated.

An inverse probability weight was generated at each wave using a logistic regression model to estimate the probability of inclusion in the data set, given the BDS score, age, sex, living in institutional care and the centre that each patient was recruited from. Modelled probability of participation matched the observed probability of participation well (Hosmer–Lemeshow goodness-of-fit tests: p = 0.96 for W1 and p = 0.52 for W2). At each wave, the weight was multiplied by the supplied CFAS II sampling weight and then each weight was bounded above at 20 to prevent any particular individuals from having an extreme influence on the study findings. The distribution of the W1 and W2 weights are shown in Figure 3.

FIGURE 3.

The distribution of inverse probability weights of participants with dementia at (a) W1 and (b) W2.

Figure 4 shows the flow of CADDY participants through the CFAS II, showing the numbers of dementia cases and those who consented to medical record linkage.

FIGURE 4.

Flow of CADDY participants through the CFAS II, showing the numbers of dementia cases and those who consented to medical record linkage. Not shown: three patients assessed to have dementia at W1 but not at W2.

Prevalence of diagnosis, concern or referral among people with dementia

Weighted estimates of the proportion of participants diagnosed, referred or with cognitive concern noted are shown in Table 4. The proportion diagnosed increased from 35% [95% confidence interval (CI) 28% to 43%] at W1 to 43% (95% CI 36% to 51%) at W2. At both waves, around 10% of the sample had been referred to specialist services but had not yet received a diagnosis of dementia, and 11% had a record of a cognitive concern being raised but had no subsequent referral or diagnosis. Hence, the proportion of the population with dementia whose GP had no record of any cognitive concern was 46% (95% CI 38% to 53%) at W1 and 35% (95% CI 29% to 42%) at W2.

Characteristics of the population with diagnosed versus undiagnosed dementia

The weighted distribution of demographic, social and clinical characteristics of participants with diagnosed compared with undiagnosed dementia for both waves combined is shown in Appendix 3, Table 11. Here, we discuss these numbers from the point of view of comparing the population that had a GP diagnosis with the population that did not. For a multivariable analysis of these data, we aimed to identify statistically significant independent effects of predictors of help-seeking and diagnosis (see Chapter 9). The following relates only to the group with a CFAS II diagnosis of dementia.

Clinical characteristics

The median cognitive function was considerably lower among those with diagnosed dementia (CAMCOG score: median 28, IQR = 1–54) than among the other three dementia groups (CAMCOG score for the ‘unknown’ group: median 55, IQR = 38–67). Nevertheless, more than 25% of those without a recorded cognitive concern have CAMCOG scores < 40 (including those with missing scores); this suggests a significant degree of impairment within the undiagnosed dementia groups (Figure 5).

FIGURE 5.

Cambridge Cognitive Examination (CAMCOG) scores among people with dementia stratified by GP diagnosis status, combining W1 and W2.

Similarly, informants for those with a diagnosis reported considerably more memory and non-memory symptoms than informants of the ‘unknown’ group, although, as with objectively measured cognitive impairment, a substantial proportion of informants (often carers) across all diagnosis groups endorsed most or all of the symptoms that were assessed. This indicates that symptoms are recognised by carers and are potentially problematic in many cases, even where no concern has been raised with a GP.

There was little difference in the levels of depression and anxiety across groups, but those with diagnosed dementia had substantially fewer diagnosed cardiovascular comorbidities than those in other groups.

Research question 1b: prevalence and characteristics of the cohort without dementia but with a dementia diagnosis

Of the 296 participants without dementia at the W1 interview, 22 had a GP-recorded diagnosis of dementia on or before that date. Of the 16 participants from this group who survived to W2, 14 met the CFAS criteria for dementia at W2, suggesting that in most such patients the GP had diagnosed dementia before the CFAS assessment and not that these patients constituted a substantial population with a misdiagnosis of dementia. Of the 78 participants without dementia at W2, three had a GP diagnosis of dementia recorded. There were no assessments beyond W2, so we do not know whether or not these participants would have met the study criteria for dementia at a later date, but two of the three had MMSE scores of ≤ 21 at W2, suggesting a substantial cognitive impairment, whereas one had MMSE score of 28. Hence, our study did not reveal a substantial number of GP diagnoses of dementia that might be considered false positive; we did not investigate this group further.

Predictors of diagnosis of dementia in general practitioner records

Having demonstrated that the population with diagnosed dementia differs in several characteristics from the population with undiagnosed dementia, here we estimate the probability of diagnosis conditional on patient characteristics. We then develop multivariable models to establish which factors independently predict if a patient seeks help and the factors that predict diagnosis conditional on help-seeking.

Age, sex, educational attainment, area-level deprivation and living arrangement

Appendix 3, Table 10, shows primary care diagnosis status by levels of each potential predictor.

The description of the diagnosed versus undiagnosed population provided in Chapter 8 suggests that demographic factors, such as age, sex, socioeconomic status (as measured by educational attainment) and area-level deprivation and current living arrangement, might independently predict whether or not a person with dementia seeks and receives a diagnosis.

Figure 6 shows that people with dementia are most likely to have a primary care diagnosis if they live with a long-term partner or live in residential care, compared with those who live alone or with others (not spouse). Women are more likely to be diagnosed than men, and there is a relationship with age such that the youngest (aged 65–74 years) and oldest (aged ≥ 90 years) groups are less likely to be diagnosed than other age groups (aged 75–89 years).

FIGURE 6.

The proportion of people with dementia who are diagnosed, stratified by (a) sex; (b) age; (c) living arrangement; (d) area-level deprivation; (e) years in education; and (f) cognitive function measured by CAMCOG score. Error bars correspond to 95% CIs.

Cognitive function

The proportion of people with dementia who have a primary care diagnosis increases with poorer cognition as measured by CAMCOG score. Figure 7 clearly shows that people with more severe cognitive impairment are more likely to be diagnosed, with only 4 of 63 (6%) people with a CAMCOG score > 73 and 17 of 66 (25%) of those with a CAMCOG score between 66 and 72 being diagnosed. However, > 40% of people with the most severe cognitive impairment did not have a GP record of dementia.

FIGURE 7.

The proportion of people with dementia who are diagnosed, stratified by number of reported signs and symptoms. (a) Memory symptoms; (b) non-memory symptoms; (c) depression; (d) anxiety; (e) cardiovascular comorbidity; (f) other comorbidity; (g) frequency of social contact; and (h) number of social activities. Error bars correspond to 95% CIs.

The relationship between subdomains of cognitive function, split into memory/orientation and other subdomains, and stratified by diagnosis status, is shown in Figure 8. Memory and non-memory scores are strongly associated, as would be expected. A logistic regression model was estimated to predict diagnosis based on the memory/orientation and non-memory domains, as well as the interaction effect between the two. This model shows that, once memory/orientation is taken into account, there is no further effect of cognitive function on the likelihood of having a primary care diagnosis and no significant interaction. This is illustrated by the contour lines overlying Figure 8, which indicate how the predicted probability of diagnosis under this model varies with both domains. The fact that the lines are broadly horizontal indicates that the probability of diagnosis is determined strongly by the level of impairment in memory and orientation, and not at all by other domains.

FIGURE 8.

Diagnosed and undiagnosed people with dementia in CFAS II, plotted with respect to their CAMCOG score in memory/orientation and non-memory/orientation domains. Contour lines indicate equal probability of diagnosis given the level of memory and non-memory function.

To explore this further and to identify specifically which domains of impairment were linked to diagnosis, a regression analysis was conducted, regressing diagnosis on all 10 CAMCOG subdomains simultaneously. The absolute difference in probability of diagnosis independently associated with a 1-SD change in each domain is shown in Figure 9. Only impairments in memory and orientation are positively linked to diagnosis.

FIGURE 9.

Association between CAMCOG domain scores (standardised) and the absolute probability of diagnosis. Positive scores indicate more impairment.

This analysis was repeated removing the cluster of cases with very low CAMCOG scores (CAMCOG < 10 or missing). This led to no appreciable difference in results (not shown).

Predictors of raising cognitive concern and of diagnosis once a concern is raised

This multivariable model was then disaggregated into two: (1) a model for the probability of help-seeking or referral among the whole population with dementia; and (2) a model for the probability of diagnosis among the subset with at least help-seeking or referral.

Appendix 3, Table 13 shows the association between each factor and (1) help-seeking and (2) diagnosis conditional on help-seeking. There is no clear distinction between the factors that affect either aspect of the diagnostic process. Subjective and objective memory impairment, comorbidity and living arrangement each appear to significantly affect both the chance of having a cognitive concern recorded and the probability of having a diagnosis once the concern is raised.

The time to diagnosis for people with incident dementia

This chapter describes the time to primary care diagnosis for participants with incident dementia between waves (i.e. those who did not meet the criteria for dementia at the W1 assessment but did at the W2 assessment).

There were 152 participants in CADDY with incident dementia in whom primary care diagnosis status at the time of their W2 interview was known (164 consented, of whom eight did not have data returned and four were censored). Of these participants, 14 (9%) had a diagnosis of dementia recorded in primary care before the W1 interview (see Chapter 8, Research question 1b: prevalence and characteristics of the cohort without dementia but with a dementia diagnosis) and seven (5%) had a diagnosis recorded more than 4 years before the W2 interview. A further 36 participants with incident dementia at W2 (24%) had a first primary care diagnosis between W1 and W2, leaving 102 of the 152 participants with incident dementia alive and without clinical diagnosis at the time of the W2 interview. Between the W2 interview date and the date at which primary care records were ascertained (median 3.9 years, IQR 3.5–4.6 years), 38 of these participants received a diagnosis, 26 died without ever having a dementia diagnosis and 38 remained alive and undiagnosed.

Figure 10 shows the cumulative incidence of dementia diagnosis up to 4.5 years after the W2 assessment for participants whose index assessment with dementia was at W2. The cumulative incidence of death without diagnosis is also shown. Approximately 4.5 years after the index assessment (hence, approximately 5.5 years from the point when the participant would meet CFAS II criteria for a dementia diagnosis), around 60% of participants had a diagnosis in their GP record. A further 20% had died with dementia but without a GP-recorded diagnosis, leaving a further 20% of participants with dementia alive but undiagnosed at that time.

FIGURE 10.

The cumulative incidence of diagnosis (black dashed line) and death without diagnosis (green dotted line) for participants who developed dementia between W1 and W2.

The median time to GP record of a diagnosis (time by which 50% of participants have a diagnosis) is 2.35 years from the W2 interview. The exact date of dementia onset for these participants (the date at which a participant would consistently meet the clinical criteria for dementia) is unknown, but if we assume it is at the mid-point between the W1 and W2 interviews (around 1.1 years before W2 interview), then this suggests that the median time to a diagnosis being recorded in primary care from dementia onset is approximately 3.5 years.

A further implication is that the rate of new GP diagnoses among those undiagnosed slows considerably after the period around dementia onset. Figure 11 shows the hazard function for diagnosis, that is the rate of new diagnoses among those undiagnosed as a function of time since the W2 interview. The rate of diagnosis increases sharply after a low start during the period around CFAS dementia onset, until it peaks at around 0.2 diagnoses per year (that is, the chance of getting a GP diagnosis over the course of a year for somebody without a GP diagnosis peaks at about 20%). Following this period, the rate of new GP diagnoses among the undiagnosed falls rapidly, to around half the maximum rate of 3–4 years after the W2 interview. This suggests that, if a person with dementia does not receive a diagnosis around the time of dementia onset (when they would meet the criteria for dementia), then they are increasingly unlikely ever to do so.

FIGURE 11.

The smoothed hazard (probability of being diagnosed over the next year) as a function of the time since the W2 assessment for people with incident dementia between W1 and W2.

Factors affecting the rate of new diagnosis

To understand which patient characteristics affected the probability of a new diagnosis among those who were undiagnosed, we conducted an analysis among the CADDY sample with dementia but no GP diagnosis. We conducted a time-to-event analysis, including people from the point of their index assessment, that is the first date on which they were assessed to have dementia in CFAS II, and included only people without a GP diagnosis on that date.

There were 285 participants with dementia first seen at either W1 or W2 who did not have a primary care diagnosis recorded at the time of their CFAS assessment. In this analysis, we compare the characteristics of those who were diagnosed with the characteristics of those who were not diagnosed over the subsequent 2 years.

This is in contrast with the analysis in Chapter 9, in which we considered which factors independently predicted which participants had or had not received a diagnosis up to the point of their CFAS assessment (could be considered prevalent diagnosis). Here we consider which factors can predict which of the participants who are undiagnosed at the time of their CFAS assessment go on to receive a diagnosis in the subsequent 2 years (incident diagnosis). Two years is chosen to allow a sufficient number of participants to receive a diagnosis for this to be a meaningful analysis, but also to ensure that patient characteristics are unlikely to change substantially.

Does a general practitioner-recorded diagnosis affect subsequent mortality?

As in the time-to-event analysis reported in Chapter 10, Factors affecting the rate of new diagnosis, survival was estimated from the first CFAS interview at which a participant met the study criteria for dementia (the index assessment, reflecting prevalent dementia cases at W1 and incident dementia cases at W2). There were 445 such CADDY participants in total, with GP diagnosis status known for 421 participants (136 with a GP-recorded diagnosis at the time of the index CFAS assessment, 285 without a GP-recorded diagnosis at that time). In the current analysis, survival was estimated up to 6 years post index assessment. Those who do not meet the CFAS II criteria for dementia are not included in this analysis.

A total of 278 participants died within 6 years of their index assessment, of whom 267 had known diagnosis status at their index assessment. The characteristics of those who died and those who did not are shown in Appendix 3, Table 16.

Figure 12 shows Kaplan–Meier plots comparing the survival from index assessment between those who did and those who did not have a diagnosis recorded in primary care at that time. There is little difference in the survival curves (Kaplan–Meier test, p = 0.083). The hazard ratio (HR) associated with diagnosis is 1.25 (95% CI 0.97 to 1.61; p = 0.083), suggesting a slightly increased mortality rate among the diagnosed that is not statistically significant.

FIGURE 12.

The proportion of diagnosed (green dotted line) and undiagnosed (black line) participants surviving up to 6 years following their index interview.

However, we know from previous chapters that those who are diagnosed tend to have more severe dementia, different living circumstances and other factors that could predispose to mortality, and these factors might suppress any benefit of diagnosis on survival.

The estimates of the effect of diagnosis and covariates on subsequent mortality, estimated using a multivariable Cox proportional hazard regression with multiple imputation used to recover missing covariate data are shown in Appendix 3, Table 17. After adjusting for all potential prognostic indicators that could also be linked to diagnosis status in a multiple imputation model, the HR associated with diagnosis is unchanged from the univariable analysis (HR 1.25, 95% CI 0.91 to 1.71; p = 0.169). Other factors significantly associated with increased mortality were increased age; living in residential care or with other family members compared with living alone or with a partner; increasing number of subjectively reported non-memory impairments; and lower educational attainment.

When a complete-case-only analysis with 236 participants was conducted, the results did not qualitatively change (HR for diagnosis 1.07, 95% CI 0.69 to 1.66). The proportional hazard model fitted well. The global test for proportional hazards assumption was χ432=41.0; p=0.557.

Social contact and mood

Changes in frequency of social contact and mood (generalised anxiety and health-related worry) following diagnosis were measured using data from participants who had both W1 and W2 assessments, and who had dementia at W2 (n = 232). Patients were divided into three groups depending on their diagnosis status at W1 and W2: (1) diagnosed at baseline (n = 42); (2) newly diagnosed between waves (n = 51); and (3) remained undiagnosed (n = 139). Mean average levels of social activity, health-related worry and anxiety among these groups at both waves are shown in Table 5. Only participants for whom data are available at both assessments are included. The proportions improving or worsening with respect to each outcome are plotted in Figure 13.

FIGURE 13.

The proportions of participants improving or worsening between waves with respect to (a) anxiety; (b) health-related worry; and (c) social activity; stratified by diagnosis status at W1 and W2.

Table 6 shows the numbers of participants whose data suggested improvement or worsening with respect to each outcome and ORs from multinomial regression estimating the effects of new diagnosis or existing diagnosis on the chance of improving or worsening each outcome, adjusted for age, sex, living arrangement and cognitive function.

In general, there are no clear patterns of benefit or harms associated with diagnosis in this data set. Although those with a new diagnosis are more likely than those in the other two groups to be reported to have fewer social activities after their diagnosis (OR for reducing social activity associated with new diagnosis = 3.09, 95% CI 1.26 to 7.59), this is an analysis which emerges from a large number of independent hypothesis being tested. Similarly, although more newly diagnosed participants had an increase in anxiety between waves compared with the other two groups, numbers were small and the difference was not statistically significant. Both of these possible effects should be tested in larger dedicated studies.

Note these data are limited owing to small numbers of participants, but are included for completeness and transparency as this analysis was specified in the original study plan. There is a large amount of missing data particularly for the items that rely on self-report; we did not feel that information on worry or anxiety would be meaningful among those who were not able to provide their own responses.

Transfer to residential care setting

There were 218 participants living in their own home at W1 for whom we have primary care diagnosis data and who had dementia at W2. Between W1 and W2, 36 participants were moved to a care setting. Of the 181 participants who were undiagnosed at W1, 15% (n = 27) were institutionalised between waves [compared with 24% (n = 9) of those with a diagnosis]. However, we know from previous chapters that those with a diagnosis are more likely to have more advanced dementia, which would also predispose to a move to a care setting, and so we conducted a multivariable regression analysis to control for this difference. Without adjustment, the OR for new institutional placement associated with being diagnosed with dementia is 1.83 (95% CI 0.77 to 4.31). After adjustment for baseline cognitive function, age and sex, this OR becomes stronger at 2.15 (95% CI 0.82 to 5.59).

If we restrict this analysis to participants with dementia at W1 and W2 (n = 72), then the pattern of results is similar. Seven of 26 (27%) participants with a diagnosis moved to a care setting during follow-up [compared with 10 of 46 (22%) who did not have a diagnosis]. The OR for the association between diagnosis and move to a care setting is 1.32 (95% CI 0.44 to 4.04), which, after adjusting for cognitive function, age and sex, becomes 1.95 (95% CI 0.52 to 7.29).

In summary, although numbers of participants available for this analysis are low, there is no evidence that being diagnosed prevents or delays a move to a care home among people with dementia not already in a care setting.

Inpatient hospital visits

There is also no clear evidence that being diagnosed affects the chance of a subsequent inpatient hospital stay for a person with dementia. Table 7 shows the number of patients with dementia at W2 who had hospital stays in the year prior to W2, stratified by the diagnosis status at W1. Of 193 patients with no diagnosis, 12% (n = 23) had one hospital admission and 8% (n = 15) more than one [compared with 7% (n = 3) and 19% (n = 8) of the 43 patients who were already diagnosed at W1].

The corresponding rate ratio for hospital stays associated with diagnosis was 1.19 (95% CI 0.63 to 2.23); when adjusted for age, sex, overall cognitive function and living arrangement at W1, this reduced to 0.72 (95% CI 0.40 to 1.13). When the analysis is restricted to those with a study diagnosis of dementia at W1, the pattern is similar (Table 8), with a rate ratio of 1.37 (95% CI 0.53 to 3.54) attenuated to 0.63 (95% CI 0.19 to 2.07) in multivariable analysis.

The distribution of length of stay across groups is also shown in Tables 7 and 8. Among all participants with dementia at W2, being diagnosed at W1 appears to be associated with a shorter length of stay among those with at least one admission [mean 29.7 days for undiagnosed compared with mean 18.5 days for diagnosed participants; adjusted rate ratio 0.37 (95% CI 0.16 to 0.84) days]. However, this pattern is reversed if the analysis is restricted to those with dementia at W1, with a much higher average length of stay among participants with a diagnosis than among participants who are undiagnosed. Figure 14 shows how individual data points may account for this discrepancy. When all participants are included (see Figure 14a), then the mean length of stay is affected by several large values (seven participants each with > 75 nights in hospital in the year before W2). If medians rather than means are compared, then it appears those who are diagnosed have longer stays than those who are undiagnosed (median of nine nights in hospital for the undiagnosed, compared with 17 for those diagnosed). This is consistent with the pattern seen when restricting to those with dementia at W1 (see Figure 14b), although that group includes only 17 participants.

FIGURE 14.

The total number of nights in hospital in the year before W2 among participants with at least one overnight stay, stratified by W1 diagnosis status. (a) All participants with dementia at W2; and (b) participants with dementia at W1 and W2.

Few strong conclusions can be drawn from these data regarding the effect of diagnosis on the rate of subsequent hospital visits or length of stay because of the small numbers; however, no evidence emerges from this analysis to suggest that receiving a diagnosis prevents hospitalisation or costs associated with overnight stays among people with dementia.

Introduction

Public and patient involvement in research is helpful to ensure that it is well conducted and to realise its potential society benefit.

Public and patient involvement in CADDY conceptualisation and development

Patient and public views shaped CADDY from the outset. Improving dementia diagnosis and understanding its impact has been consistently raised as a priority by patient and public groups. The National Institute for Health Research (NIHR) Dementia and Neurodegeneration Research Network Primary Care Clinical Studies Group identified diagnosis of dementia as a top five research priority in 2012. Alzheimer’s Society lay members contributed to the 2013 James Lind Alliance priority-setting exercise for dementia and identified ‘What is the impact of an early diagnosis of dementia and how can primary care support a more effective route to diagnosis?’ as a top 10 priority for dementia research. Delayed dementia diagnosis was consistently raised by members of the Norfolk Older Peoples Strategic Partnership Forum in 2013. The concept for CADDY was developed with the support of the Public and Patient Involvement in Research (PPIRes) group; this chapter describes the PPI throughout the course of the study.

In the development of the CADDY funding bid, we consulted lay members of the Alzheimer’s Society and met with and consulted with two local PPI groups (inspire and PPIRes), who all echoed the need for this study. PPI group members identified the role of comorbid mental and physical health in their own experience as predictors of dementia diagnosis; the huge fear of diagnosis for many people; the individuality of the decision to seek help; and the importance of considering those less likely to participate in a survey. We refined the research questions and methods accordingly and when the study was running, developed an advisory group of 11 PPI members and invited three members to join the Study Steering Committee (SSC) to ensure that the views of our PPI partners were included at all stages.

Public and patient involvement during the study

Public and patient involvement members joined the CADDY advisory group and SSC from three PPI associations:

1. PPIRes (this group, co-ordinated by NHS South Norfolk CCG, includes a panel of volunteers from Norfolk and Suffolk with a range of experiences as patients and carers of people with dementia)

2. Inspire (this group, co-ordinated by NHS Norfolk and Suffolk Foundation Trust, is an initiative to involve service users and carers in mental health research)

3. the Alzheimer’s Society Research Network (this group includes a large team of carers, former carers and people living with dementia).

Members were invited to join the CADDY Public and Patient Involvement Advisory Group (PPIAG) via organisers of their respective PPI associations. Five members joined from PPIRes, five from Inspire and one from the Alzheimer’s Society Research Network. One representative from each PPI association also joined the SSC to represent the views of the PPIAG and to provide PPI oversight to the study.

Ten of the 11 PPIAG members had direct experience of dementia care and one member had indirect experience through supporting friends who have parents with dementia. Nine members had been directly involved in the dementia diagnosis process and six had previously been participants of a research study. As well as experience of caring for people with dementia, PPI SSC members brought a range of professional experience from public, private, health and social care organisations, which was also helpful in terms of providing a managerial oversight.

Public and patient involvement activities

The PPIAG met twice, in February 2016 and in July 2017. George Savva chaired and Clare Aldus facilitated the meetings. Members received reimbursement for their time and travel costs in accordance with INVOLVE guidelines. The PPIAG was updated through newsletters.

The CFAS II national co-ordinator (Linda Barnes) provided training in the CFAS II data set for PPI SSC members. PPI members of the SSC met George Savva and Clare Aldus ahead of each SSC meeting to ensure that PPI and research team members worked together effectively and had the opportunity ask questions in an informal setting ahead of each meeting.

The first PPI advisory group (PPIAG 1) meeting was held while the study protocol was being developed, before our application for ethics approval was made. We consulted PPIAG 1 with respect to ethical issues and their experiences of dementia diagnosis.

The PPIAG met again in July 2017 (PPIAG 2), following the completion of data collection, preliminary analysis and early presentation of the results at a public seminar. PPIAG 2 reflected on the early results of the study and the presentation style. Key themes were identified, summarised and shared with our SSC and research team to develop a final analysis plan and to refine presentation.

Eleven members of the advisory group attended the first meeting (February 2016) and five attended the second meeting (July 2017). Each meeting ran for approximately 2 hours. Each PPIAG meeting was audio-recorded and transcribed. Transcripts were anonymised, key themes were identified and these were returned to each PPIAG for comment.

Influence of PPIAG on CADDY

Facilitating meaningful participation

At the outset, we had proposed to train PPI SSC members in the necessary epidemiological methods required to understand the study. However, on discussion with PPIAG members, it became apparent that this was not necessary as the important aspects of study design were easily understood and the range of professional backgrounds meant that any generic training would not have been useful. Instead, pre-SSC meetings (held on the same day as SSC meetings) were used to address any questions or concerns that PPI SSC members wanted to clarify ahead of the meeting. Following each meeting, George Savva and Clare Aldus checked that PPI SSC members were satisfied that they had been able to participate. The PPI SSC members have been proactive in raising questions regarding the scientific and procedural aspects of the study, which have led to substantial changes in the study. We are therefore confident that their contributions have been both genuine and meaningful.

Public and patient involvement in epidemiology

The CADDY PPI contributions have been valuable in many respects, specifically:

• in support of our ethics application

• insight with respect to the variety of patient experiences in health services, and how varied and complex decisions surrounding help-seeking and diagnosis are, and the effects of diagnosis for individuals

• revision of the study materials from the point of view of non-researchers

• support with interpretation of data in light of their experience

• guidance on presentation of data

• support for the design and implementation of dissemination activity

• identification of key messages arising from findings

• oversight and suggestions as to the management of the study

• assurance from those directly affected by the issues being studied that the work is valuable beyond academic boundaries.

Public and patient involvement is underdeveloped in epidemiological research, and many volunteers and researchers are unsure of the value of their participation because epidemiological research is most often focused on data analyses and does not usually include recruitment of ‘real’ participants. Nevertheless, CADDY has benefited enormously from this interaction, often in unexpected ways, owing to the diversity of the personal and professional expertise that was brought into the study through the PPI group. As PPI members are independent of the research team, they provide a level of oversight to an ongoing research study that may not otherwise be present depending on the specific requirements for composition of the SSC.

There have been some challenges, particularly around PPI members feeling initially nervous about contributing to discussions regarding work that is mainly statistical in nature, and, at the outset, some PPIAG members not understanding that we are largely unable to alter the nature of the data available to us or the experience of the participants of the wider CFAS II. However, these challenges were overcome and as the public become more aware of ‘big data’ research and PPI groups participate more often in epidemiological studies, we believe that the value of PPI in epidemiology will be better understood.

Reflection from a public and patient involvement Study Steering Committee member

The following is reflection on the PPI elements of CADDY from one of the study’s PPI SSC members:

It is impossible to summarise just how good the PPI aspects of the CADDY project have been. Right from the start, George and Clare recognised the importance of involving ‘real’ people, with direct experience of dementia, at every stage of the project. CADDY was refined to reflect the comments and concerns expressed by us all (not just those who spoke the loudest!), and, more importantly, we could actually see how our suggestions and comments were being incorporated/used to shape the project as it moved forward – particularly when it came to investigating and reporting back on the ‘why’ aspects of the research data and findings. Statisticians may love their numbers, and of course they are an important part of the picture (it is often useful to know how often something happens), but ordinary, ‘real’ people usually want to know why something happens: what can I/we learn from that? How is that relevant to me/my loved ones/our everyday lives? The CADDY findings and final report make it ‘real’. As one of the PPI representatives on the project’s Steering Committee, I was also able to contribute to the project on a more strategic level (in terms of governance, ethics and transparency, etc.) and once again my comments were well received and acted upon, and so that also demonstrates true partnership working. Whilst it is clear that the CADDY project has undoubtedly benefited from its PPI involvement, I think it is also fair to say that being able to contribute so meaningfully to a project has also benefited the PPI people involved: being able to share our insights, experiences and concern(s), and then actually be able to see that they have been taken on board/incorporated into such an important piece of research work, is very empowering and uplifting. I have been involved in PPI for almost 15 years now, and without doubt the CADDY project, and working with the CADDY team, is the best PPI experience I have ever had: if ever there was a ‘model of best PPI practice’ the CADDY project is it.

Introduction

The CADDY dissemination plan was designed to lead to both academic and societal impact from research. Specific activities planned or completed are academic publications, participation in academic conferences and an event to encourage broad discussion on dementia diagnosis across stakeholder groups. We are also working closely with the public policy section of the Alzheimer’s Society.

Conferences were selected to enable discussion of preliminary results with the international dementia and UK primary care research communities. Hence, in 2017, early findings were presented at the Alzheimer’s Association International Conference, the Alzheimer’s Society Research and Practice Conference, the Society of Academic Primary Care and the Dementias 2018 conference in London. Early findings were also shared locally through an Alzheimer’s Research UK public event held at UEA and with PPI groups. CADDY has also been presented at conferences: AgeNet, University of Hertfordshire, UK, 2018; Alzheimers Europe, Barcelona, Spain, 2018; Dementias 2019, London, UK, 2019; Contemporary Populations and Dementia, What Have we Learnt and Where are we Going? Alzheimer’s Disease Forum, Tasmania, Australia, 2019; Dementia Care Research Centres & Dementia Training Australia (part of the Government’s Australia Dementia Research Institute), Webinar, Australia, 2019.

Impact objectives

Broad impact objectives are summarised in Table 9.

Dissemination event

An event was held in London in December 2017 with the main aim of promoting the study findings among relevant stakeholder groups, including patients and the public, health-care professionals, local and national policy-makers and managers, and academic researchers in related areas. The event was participatory to ensure that the views of each stakeholder group and not just researchers were aired, and so that research teams could gauge reaction to their early findings from a diverse audience to help dissemination of final results. This event was called ‘Dementia Diagnosis in Primary and Secondary Care: Policy and Practice’ and was held at the Royal College of General Practitioners, London.

To maximise engagement with key stakeholders, the event was widened to other research studies beyond the CADDY. Speakers were invited to represent three other active studies on dementia diagnosis, in order to build an attractive and engaging forum. The findings of each study were complementary and this promoted (1) good exposure for each study to a wider audience; (2) more incentive and better value for delegates attending; (3) a research community networking opportunity; and (4) opportunity for each study to inform the others.

Practical considerations to encourage attendance were (1) using a central London venue; (2) securing influential speakers; (3) making attendance free of charge, with some support available for travel for specific delegates; (4) providing a coherent dementia diagnosis focus; (5) securing continuing professional development accreditation; and (6) employing social media, personal and professional contacts to promote the event.

Evaluation of the event

Summary and future impact activities

We have used conference presentation and our own dissemination event to begin to generate interest in the study and refine our final analyses and outputs. Dissemination of CADDY findings and resulting translation into public awareness, policy and practice will take place following the conclusion of the study. Future research will be generated by the CADDY data set that will be made available to researchers.

One of our impact objectives is to evidence the national conversation and policy around timely diagnosis of dementia. The Dementia Diagnosis in Primary and Secondary Care: Policy and Practice meeting highlighted the lack of quantitative evidence for the supposed benefits of diagnosis. Since then we have had an ongoing conversation with NHSE regarding dementia screening in hospitals via the Dementia Assessment and Referral data collection, a national statistic (formerly a CQUIN), under which three measures are reported: the number and proportion of patients aged ≥ 75 years admitted as an emergency for more than 72 hours who (1) have a diagnosis of dementia or delirium or to whom case finding is applied, (2) if identified as potentially having dementia or delirium, are appropriately assessed, and (3) where the outcome was positive or inconclusive, are referred on to specialist services. NHSE have since received all the necessary internal approvals to retire the Dementia Assessment and Referral subject to public consultation (summer 2019).

We continue to liaise with dementia advocacy groups including Alzheimer’s Research UK and the Alzheimer’s Society with a view to raising public, clinical and policy-maker awareness of non-memory symptoms of dementia through public awareness campaigns.

Our next steps are to translate the study findings into academic papers, liaise with charities to translate the findings into local and national policy, and prepare secondary outputs that are accessible to all stakeholders.

Summary of findings

We successfully ascertained the details of dementia diagnoses, cognitive concerns and referrals to memory services recorded in primary care of a cohort of 449 people independently assessed to have dementia between 2008 and 2011 (W1) and 2011 and 2013 (W2).

Strengths and limitations

This study was based on a large population-representative cohort with an objective and independent dementia status determined using a validated algorithm, and with data on a wide range of other social- and health-related characteristics. It is the largest study to examine prevalence, causes and consequences of diagnosis among people with dementia. However, the study does have several important limitations.

Although the response rate from general practices was very high, a substantial proportion of participants with moderate to severe dementia at W1 could not consent to the use of their medical records and CFAS II itself has a response rate of 56%. Weights and multiple imputation were used, as appropriate, to reduce the impact of these missing data on estimates.

Diagnosis status was extracted from primary care records by practice staff or CRN nurses. Although the data were validated with respect to the correct individual being identified, data entry was not independently verified.

Few participants fell into the ‘referred’ and ‘cognitive concern’ groups that were not diagnosed but had some record of cognitive concern in their primary care record, and so it was difficult to draw strong conclusions about these groups, or to identify separate predictors of help-seeking and diagnosis conditional on help-seeking. Likewise, the sample size for estimation of predictors for those with undiagnosed dementia but diagnosis in the subsequent 2 years is also small (based on only 66 new diagnoses).

For some analyses, the numbers of people with dementia falling into many of the categories were small; hence, the power to detect subtle associations is limited, particularly for risk factors that were rare in the data (e.g. very high levels of comorbidity or social participation).

The CFAS assessments covered a wide range of personal circumstances. However, with the exception of cognitive testing, each of the risk factors we tested relied on the report of the participant themselves, a proxy or an informant for their ascertainment. People with dementia may not have answered all questions correctly or have been able to provide any response at all. In some cases, only limited cognitive testing is available, with informant questionnaire providing much of the missing data where possible. Some of the variables of interest were taken from the informant questionnaire and were missing for many people. The CFAS assessment covered many aspects of health and well-being, but many areas were not covered in depth, including, for example, the quality and nature of social interaction.

A strength of CFAS II is its inclusivity, meaning that those with very severe cognitive impairments and people living in care settings are included. However, people with severe dementia are likely to be unable to complete cognitive assessments or interviews and so covariate data are often missing for this group.

We assumed throughout that the CFAS II classification of dementia status truly represents whether or not a person had dementia on the date of their interview. Although the CFAS II algorithm is validated against DSM-IIIR diagnostic criteria, it does not include a complete clinical workup and it is possible that some individuals are misclassified. A substudy of the first MRC CFAS study was conducted to validate this algorithm (Lu Gao, personal communication) and found that CFAS diagnosis of dementia is specific, in that those classified as having dementia are very likely to have dementia, but may not be sensitive, in that some cases of dementia might not be detected. For our purpose, this specificity is important, in that we are confident that the vast majority those identified in CADDY as having dementia do so. However, our ‘control’ sample of those without a CFAS diagnosis may have included some people with dementia and our control sample was not large enough to detect a small prevalence of false-positive dementia diagnoses. In any case, the threshold for dementia varies considerably with definition and the severity of symptoms fluctuates within individuals, so to attempt to be definitive about whether or not a person has dementia at any particular time is not meaningful, particularly in the milder stages.

Similarly, we have assumed that the primary care record represents what is known in health-care services with respect to a person’s cognitive status. There is evidence that some dementia is diagnosed and recorded in hospital records that does not appear in primary care record. It is also possible that cognitive impairment is recognised and managed without being coded in a patient’s notes. However, maintaining a dementia register has been an aspect of the GP contract monitored through the Quality and Outcomes Framework and there is a financial incentive for its maintenance, so it is likely to be a relatively good reflection of a GP’s knowledge of a patient.

Comparison with previous literature

Prevalence of diagnosis

A systematic review of the global prevalence and determinants of undetected dementia from worldwide studies discovered 23 studies, published between 1988 and 2015, linking standardised assessments of dementia to diagnoses recorded in primary care, 21 of which were conducted in Europe or North America. 11 The average proportion of people with dementia who were undiagnosed across studies was 61.7% (95% CI 55% to 68%), although heterogeneity across studies was high (I² = 96.4%) (Figure 15). Only two single-site studies had been conducted in England: one in Cambridge in 1998, among 208 people with dementia recruited as part of the Cambridge City over 75 Study;17 and the other in Newcastle in 2009, among 105 people from six general practices determined to have dementia on the basis of MMSE score alone (that is without using a validated dementia assessment), recruited as part of the Newcastle 85+ study. 44 Both studies are consistent with our results. Most recently, the National Health and Aging Trends Study45 in the USA found that, in 2011, in a study of 585 people with dementia, 39.5% were undiagnosed (based on diagnoses in their Medicare records) and 19.2% were diagnosed but were unaware of their diagnosis.

FIGURE 15.

Meta-analysis of estimates of the proportion of people with dementia who do not have a recorded diagnosis in primary care. See Lang et al. 11 for full reference details of these studies. Reproduced with permission from Lang et al. 11 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Reproduced with permission from Lang et al. 11 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Official estimates of the proportion of people with dementia who are diagnosed are usually calculated by dividing the number of recorded cases by the expected number of dementia cases in the population, which is in turn derived using CFAS II estimates for age- and sex-specific prevalence. A comparison of the proportion calculated in this way across the years from 2006 to 2012 in England and Wales shows a stable proportion of around 36% between 2006 and 2009, then a steady increase of 2% per year between 2009 and 2012. 46 This is consistent with our estimate of 35% in 2008–11 and 43% in 2011–13.

Prevalence of concern and referral

Our results for the proportion with undiagnosed dementia are consistent with official statistics for the same periods, but official statistics do not include data on the group of people with no diagnosis but with some note of a cognitive concern or a referral in their record. This was partly addressed by O’Connor et al.,17 who reported the proportion of cases with possible as well as definite dementia, and found that, although only 22% of people with dementia were reported as having ‘definite dementia’ by their GP, this number rose to 58% of people when ‘possible dementia’ records are also included. It is possible that recent apparent increases in the proportion of people diagnosed with dementia are at least in part attributable to a reclassification, to ‘definite dementia’, of people who would otherwise be recorded as possibly or probably living with dementia.

Findings from the Three-City Study47 of 9294 participants from Bordeaux, Dijon and Montpellier showed that, in 2005, 34.4% of 253 people with incident dementia had not consulted any physician about their condition, 34% had consulted only their GP and 31.6% had consulted a specialist (in France referral to a consultant is usually via a GP), indicating that 65.6% of people were in the diagnostic process (including those diagnosed). Our results indicated that 65% of people with dementia were in the diagnostic process (including diagnosed). Around 21% of people with dementia were in the diagnostic process but had not received a diagnosis. Of these, half had been referred to specialist services. These findings are also broadly consistent with our findings that, in 2011–13, approximately one-third (37%) of people with dementia in our study were unknown to their GP, with respect to dementia or cognitive impairment.

Severity and nature of cognitive impairment

Dementia severity is consistently positively associated with the chance of being diagnosed. 11,23 To our knowledge, only one previous study45 has disaggregated domains of cognitive impairment to explore which specific impairments are most likely to lead to diagnosis, or which dementia subtypes are least likely to be diagnosed. Diagnostic accuracy and awareness of different symptoms may play a role in the association between the severity and nature of cognitive function and diagnosis; a meta-analysis of physician accuracy identified severity of dementia as the main predictor for accurate diagnosis. 48

Age

We found a consistent gradient such that older people with dementia were less likely to be diagnosed than younger people with dementia, with the exception of the youngest group studied (age 65–70 years), who were also relatively unlikely to have a diagnosis. Many previous studies also suggest that older people with dementia are less likely to be diagnosed than younger people. 11

Socioeconomic status

Low socioeconomic status is a known risk factor for dementia, as well as for other chronic diseases and mortality. We found no evidence that the chance of diagnosis among people with dementia differed across groups classified by educational attainment or by area-level deprivation (as measured by Townsend index). NHS Digital reports a higher ‘diagnosis rate’ among those in the lowest decile of deprivation (74.4%, 95% CI 66.9% to 80.6%) than those in the highest decile of deprivation (64.4%, 95% CI 58% to 69.8%),13,49 but this could reflect a failure to correctly adjust the expected prevalence across areas to take into account socioeconomic status. Direct estimates, such as those reported in the current study and based on cohorts known to have dementia, would not suffer the same bias, as such studies consistently show no effect of educational attainment (a useful proxy for socioeconomic status) on dementia diagnosis. 11 There is no difference in the level of health-care services received by people with diagnosed dementia across regions of the UK stratified by Townsend deprivation level. 50 Together, this suggests that there is no substantial social inequality in access to dementia diagnoses or dementia services in England.

Living arrangement

Those living with a long-term partner are more likely to be diagnosed than those living alone or with others; this finding is repeated in other studies of dementia and other chronic diseases. 11,51 Within residential care settings in England, dementia prevalence is around 65%. 16 Our results have shown that many such cases did not have a formal record of dementia in their primary care notes, and, once in a residential care setting, having a new diagnosis recorded is unlikely. Although it is likely that dementia is recognised and managed in many of these cases, a study in south-east England found that nursing home staff did not recognise dementia among their residents in more than half of cases,52 suggesting that much of the undiagnosed dementia in residential care settings will be unrecognised by staff. The recent review by Lang et al. 11 stratified studies into those conducted in the community and those conducted in care settings. Although none of the included studies of the prevalence of undetected dementia in care setting was conducted in the UK, globally those studies conducted in care homes report an average proportion diagnosed of 50% compared with 36% in community settings.

Comorbidity

Dementia is often complicated by frailty and multiple comorbidities, particularly among the oldest. Our findings suggest that cardiovascular comorbidity was inversely related to diagnosis, with those reporting three or more cardiovascular conditions much more unlikely to have received a diagnosis than others. However, no such association was seen for other comorbidities, and very few people with dementia reported three or more cardiovascular conditions. The association between dementia diagnosis and comorbidity has not been extensively studied elsewhere;11 a study of a cohort with cognitive impairment referred to a functional impairment support programme in the USA found a higher rate of hypertension among the non-diagnosed, but no other difference in comorbidity profile. 53

Carers

The presence of neuropsychiatric symptoms in older people and people with dementia is related to increased stress54 and distress in caregivers. 55 We found a positive association between informant reports of both cognitive and non-cognitive symptoms of dementia with the chance of being diagnosed.

Meaning and implications

Our estimate for the proportion of people with dementia who have a GP-recorded diagnosis align closely with those reported by NHS Digital in the same time periods. This suggests that the indirect approach taken by the NHS to calculating the dementia ‘diagnosis rate’ was broadly accurate at the national level. However, the indirect approach does appear to suggest a socioeconomic gradient in diagnosis that was not seen in epidemiological studies (including CADDY), suggesting that, at the regional level, the estimate of the numbers of people with dementia might not be accurate with respect to comparison between regions.

An important implication of this study is that people with dementia but without the memory impairments characteristic of Alzheimer’s disease are less likely to have a primary care-recorded diagnosis than those whose dementia primarily affects their memory. Our analyses also highlight some particular demographic circumstances that may delay dementia diagnosis. The youngest and oldest age groups appear less likely to be diagnosed, as are those not living with a spouse or other long-term partner. Our findings also suggest that dementia may go undiagnosed among those with many other comorbid conditions. There was no evidence for a difference in the factors that determine the presence of a cognitive concern in a patient’s record or a diagnosis given that cognitive concern.

There are many possible reasons for GPs being unaware of cognitive impairment among their patients. Many psychosocial factors among patient, carer and health-care professionals contribute to a diagnosis of dementia. 22 Even when patients do present to their GPs, GPs’ decisions to diagnose and to communicate a diagnosis are influenced by their own beliefs, patient circumstances, health system access and cultural norms. 61 For example, a multicentre randomised controlled trial of an intervention to improve GP diagnosis rates by empowering people from the general population in the south-east of England aged ≥ 70 years to seek help from their GP if they were concerned found that the intervention did not increase diagnosis rate. 62

Evidence suggests that some GPs recognise and manage patients as having dementia without recording them in primary care record systems as having dementia. 19,48 It is possible that recent increases in the number of people diagnosed with dementia simply reflect an improvement in the coding of existing knowledge; further ongoing epidemiological studies such as CFAS and CADDY are needed to monitor this.

This work is also closely aligned with recommendations of the roadmap to advance dementia research in prevention, diagnosis, intervention and care by 2025,63 most specifically those of goal 2, which aims to maximise the benefits to people living with dementia and their families of seeking and receiving a diagnosis of dementia by helping to define what a ‘timely’ and ‘quality’ diagnosis of dementia means; to identify unmet needs for support and help; and to clarify differences in diagnosis rates and service usage.

In recent years there have been a number of initiatives to promote dementia recognition and management in primary care. 64 However, initiatives and financial incentivisation to increase diagnosis rates can lead to concerns about overdiagnosis or misdiagnosis. 30,65,66 Our study data collection took place before the introduction of incentives for dementia diagnosis, and we found no substantial evidence of people without dementia receiving dementia diagnosis. However it is important, with continuing pressure on local health authorities to diagnose more cases, that the risks and potential harm associated with overdiagnosis are monitored.

It has been suggested that missed diagnosis of dementia might contribute to an earlier move to residential care, or, conversely, that diagnosing more people with dementia could delay admission to residential care services. 10 Yet the only evidence in support of this claim is the fact that some psychosocial interventions, if optimally applied, have been found to delay a move to care settings, followed by the supposition that increased diagnosis would automatically increase provision and optimal uptake of these interventions by the people who might benefit from them. It would also be necessary to assume that the benefits of diagnosis would apply equally among those who had sought help and those who had not. On the other hand, claims about the possible psychosocial or practical harms of dementia diagnosis are also often made without a direct quantitative evidence base to support them. This report highlights the lack of such evidence and the difficulty in obtaining it.

We found that people with dementia living in care homes were also likely to not have a formal diagnosis in their GP record. It is acknowledged that the usual routes to formal dementia diagnosis in England, that is memory clinics or specialist mental health services, may not be accessible or appropriate for people with more severe impairments living in residential care, which may have contributed to lack of formal diagnosis among this group. More emphasis has been placed on diagnosis in care homes in recent years and so it is likely that diagnosis rates have increased in this setting alongside those in the community. However, in common with the wider community of people living with dementia, the benefits and harms of diagnosis in long-term care settings remain poorly understood.

Although calls for dementia screening intensified around a decade ago, lately there has been an acknowledgement that early diagnosis of dementia might not be optimal for everybody and that more research is needed into when a diagnosis is timely (i.e. made at the right time for an individual). Recently, the World Alzheimer Report 2016 has acknowledged that screening for dementia cannot be recommended at present, owing to a lack of evidence of benefit and low uptake of screening programmes where they have been implemented at regional level. 67 To our knowledge, the only screening trial for dementia is yet to report; 4000 older people have been recruited to the Indiana University Cognitive Health Outcomes Investigation of the Comparative Effectiveness of dementia screening (IU-CHOICE) trial (ClinicalTrials.gov identifier: NCT01699503)68 and have undertaken a short cognitive function test, with onward referral where indicated and diverse outcomes, including quality of life, to be measured. In any case, it is likely that whether there is a net benefit or harm of diagnosis will depend enormously on the individuals being screened, how the screening and diagnosis process works and the quality of post-diagnostic support available. Rightly, it is now acknowledged that establishing good care and support for those already known to have dementia, and establishing the benefits of diagnosis, should be prerequisite ahead of expensive programmes aimed at detecting more cases. 31,69

Generalisability of the findings

This work is directly applicable to England. Although CFAS II was conducted in three diverse centres across England, the dementia diagnosis rate is likely to vary substantially across the country and so our estimates of the proportion without a diagnosis are unlikely to be directly applicable within regions. Although health services vary in terms of diagnostic pathways, incentives for diagnosis and awareness of dementia among the public and clinicians, many of the factors that lead people to seek help for symptoms of dementia, and the effects on people of receiving a diagnosis, are likely to be similar across the Western world. Policy changes, GP and public attitudes and awareness of dementia have changed in recent years and, as such, the overall prevalence of undiagnosed dementia and numbers of people with a false diagnosis are very likely to have changed in the years since the CFAS II interview data were collected. However, our findings on the benefits and harms of diagnosis and of the predictors of diagnosis are likely to be applicable today.

Conclusion

Dementia diagnosis rates identified under CADDY for the period 2008–13 are similar to those reported at the time nationally through NHS Digital. This provides confidence in the data provided to policy-makers and commissioners through NHS Digital, at least at the national level. We identified no evidence of overdiagnosis of patients with dementia and that around 20% of patients with dementia were not diagnosed but were in the diagnostic pathway at any one time.

Despite the limitations of this analysis, the associations between diagnosis and cognitive function appear robust. The major implication is that people with dementia but without the memory impairments characteristic of Alzheimer’s disease are much more likely to be undiagnosed than those whose dementia primarily affects their memory.

Our analysis also highlights some particular demographic circumstances that may delay dementia diagnosis, and so suggests areas in which health education and awareness campaigns could be directed if benefits of diagnosis were identified. The youngest and oldest age groups appear less likely to be diagnosed, as are those who have completed fewer years of education and those not living with a spouse or other long-term partner. Our findings also suggest that dementia may go undiagnosed in those with many other comorbid conditions.

Implications for practice

• There is a gap for many people between having manifest signs of dementia and having a diagnosis of dementia recorded in primary care.

• Non-memory presentations of dementia appear to be more readily overlooked. This could be evidence of a gap in public and clinical awareness of other symptoms of dementia. Campaigns to raise awareness of non-memory symptoms of dementia could be considered.

• People with dementia aged ≥ 90, those living alone and those with comorbidity may also be underdiagnosed, and practitioners should be aware of the possibility of undiagnosed dementia among these groups.

Recommendations for research

• At present, there is little objective quantitative evidence that obtaining a diagnosis confers any benefit on the individual. This may reflect the shortcomings of current post-diagnostic support, but also the lack of studies specifically designed to quantitatively explore the diagnostic pathway and estimate the effect of dementia diagnosis on patient outcomes. Further research with longer and more systematically measured follow-up post-diagnostic support than the current study is required to bridge this evidence gap.

• Researchers should be keenly aware of the large average lag and large variation in the distribution of the time between dementia incidence and the time to diagnosis of dementia in primary care, particularly when first record of a dementia diagnosis is used as the basis for recruitment or as a proxy for dementia incidence in epidemiological studies.

Study governance

The study was carried out at the UEA and the sponsor for this study was South Norfolk CCG.

The SSC met on four occasions. It was not independent of the research team. It was chaired by Professor Frances Bunn (University of Hertfordshire; independent member). Ordinary members were study co-applicants: George Savva, Antony Arthur, Carol Brayne, Tom Dening, Chris Fox, Fiona Matthews, Louise Robinson and Blossom Stephan. Other members were Judy Henwood (sponsor representative), Katy Blakely (independent local commissioner), Gavin Terry (Alzheimer’s Society Policy Division), Paul Millac, Abi Dennington-Price and Peter Richmond (PPI representatives) and Clare Aldus (researcher).

The Management Committee met on six occasions. Membership comprised study co-applicants (George Savva, Antony Arthur, Carol Brayne, Tom Dening, Chris Fox, Fiona Matthews, Louise Robinson and Blossom Stephan), representatives from the Norwich CTU (Martin Pond and Antony Colles), a sponsor representative (Clare Symms) and a researcher (Clare Aldus).

Ethics considerations and approvals

We sought data for patients who had dementia or cognitive impairment. We were mindful of the extremely sensitive nature of the data we sought and of the fact that many of the patients who had given permission for the use of their data had since died. To mitigate risk to patient data we discussed ethics issues with the SSC, our PPI advisory group and the sponsor representative, and we designed a system of data management to ensure confidentiality and accuracy of data.

The study received favourable opinion at first review on 29 April 2016 from the London-Bromley Research Ethics Committee, with HRA approval on 6 May 2016 (National Research Ethics Service 16/LO/0667; Integrated Research Application System Project ID 186510; protocol v1.0 08032016).

We gained HRA approval for a minor amendment (CADDY minor amendment 1; protocol v1.1 5 January 2017), to enable members of the CRN teams to retrieve records that had been returned to the patient record repository at PCSE (from 8 February 2017).

Study registration

This study is registered with the NIHR CRN Central Portfolio Management System (CPMS 30655).

Funding

This study was funded by the NIHR under their Health Services and Delivery Research programme. This study constitutes independent research and the views expressed in this report are those of the authors and not necessarily those of the NHS, NIHR or the Department of Health and Social Care.

Contributions of authors

Clare F Aldus (Research Fellow, Health Sciences, UEA) was study manager (January 2016–September 2017) and chief investigator (October 2017–October 2018). She contributed to research design, managed the research and administrative team, led data collection processes and liaised with general practices. She also managed governance, including ethics approval, dissemination and impact creation, and PPI activity. She contributed to statistical design and analyses.

Antony Arthur (Professor of Nursing Science, UEA) developed the study concept with George Savva. He was co-applicant, contributed to study design and oversaw leadership of the study in the period October 2017 to March 2018. He was a member of study management and steering groups, contributing to all aspects of governance, research, dissemination and reporting. He contributed to the research design, statistical design and reporting.

Abi Dennington-Price (patient and service user representative) was a member of the SSC in a PPI advisory role and a member of the CADDY PPIAG. She contributed to governance, research design, dissemination and reporting.

Paul Millac (patient and service user representative) was a member of the SSC in a PPI advisory role and a member of the CADDY PPIAG. He contributed to governance, research design, dissemination and reporting.

Peter Richmond (patient and service user representative) was a member of the SSC in a PPI advisory role and a member of the CADDY PPIAG. He contributed to governance, research design, dissemination and reporting.

Tom Dening (Professor of Dementia Research, University of Nottingham) was a co-applicant and contributed to study design. He was a member of study management and steering groups, contributing to all aspects of governance, research and reporting. He contributed to research design, dissemination, statistical design and reporting.

Chris Fox (Professor in Psychiatry, UEA) was a co-applicant and contributed to study design and the funding application. He was a member of study management and steering groups, contributing to all aspects of governance, research and reporting. He contributed to research design, dissemination, statistical design and reporting.

Fiona E Matthews (Professor of Epidemiology, University of Newcastle) was a co-applicant and contributed to study design and the funding application. She was a member of study management and steering groups, contributing to all aspects of governance, research and reporting. She contributed to research design, dissemination and reporting, with particular focus on statistical design.

Louise Robinson (Professor of Primary Care and Ageing, Regius Professor of Ageing, University of Newcastle) was a co-applicant and contributed to study design and the funding application. She was a member of study management and steering groups, overseeing all aspects of governance, research and reporting. She contributed to research design, dissemination, statistical design and reporting, with particular focus on dementia in primary care.

Blossom CM Stephan (Senior Lecturer, University of Newcastle) was a co-applicant and contributed to study design. She was a member of study management and steering groups, overseeing all aspects of governance, research and reporting. She contributed to research design, dissemination, statistical design and reporting.

Carol Brayne (Professor of Public Health Medicine, University of Cambridge, and Principal Investigator for CFAS) was responsible for developing the concept of CFAS II, which was pivotal to this work. She was a co-applicant and contributed to study design and the funding application. She was a member of study management and steering groups, overseeing all aspects of governance, research and reporting. She contributed to research design, dissemination, statistical design and reporting.

George M Savva (Institute Statistician; Quadram Institute Bioscience, formerly Senior Lecturer in Applied Statistics, UEA) was chief investigator for the study in the period January 2016 to September 2017. He developed the study concept, and led study design and the funding application. He built and led the study team, overseeing all aspects of governance, research and reporting. He led the statistical design and analyses for this study.

Data-sharing statement

We aim to share data with the scientific community with as few restrictions as possible. CADDY data are held by the CFAS secretariat and will be made available to researchers on request, subject to the approval of the CFAS II Oversight and Management Committee, in accordance with the process for data sharing current at the time of application. In the first instance, data requests should be submitted to the corresponding author. Access to anonymised data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health and Social Care.

Criteria for selection of the CADDY control sample

Select a sample of around 150 people without dementia, but weighted towards those with cognitive impairment. Include all people with prescription of an anti-dementia drug at W1 or W2, if they did not meet the criteria for dementia at that time.

For each practice:

• sample at least one W1 control (person without dementia at W1) when there is a W1 case

• sample at least one W2 control (person without dementia at W2) if there is a W2 incident case (person who first meets the criteria for dementia at W2)

• allow up to two controls for each practice when the number of cases is ≥ 3

• allow up to three controls for each practice when the number of cases ≥ 4.

Algorithm for selection of the control sample

Begin with the list of practices from which there is a CADDY participant and at least one other participant, merged with the list of patients without dementia who have consented to use of their medical records.

Stages:

• exclude participants who could not give any interview information (as likely dementia)

• exclude participants for whom dementia status at W1 is uncertain

• for 40 iterations, select one participant at random from the group with the following criteria:

  • their MMSE group at W1 = 1

  • they are not currently selected as a control

  • the practice they come from does not have the maximum number of controls selected

• for 40 further iterations, select one participant at random from the group with the following criteria:

  • their MMSE group at W1 = 2

  • they are not currently selected as a control

  • the practice they come from does not have the maximum number of controls selected

• for 30 further iterations, select one participant at random from the group with the following criteria:

  • their MMSE group at W1 = 3

  • they are not currently selected as a control

  • the practice they come from does not have the maximum number of controls selected

• select any participant not already selected who reports use of a dementia-specific medication at W1

• for 30 further iterations, select one participant at random from the group with the following criteria:

  • their MMSE group at W1 = 4

  • they are not currently selected as a control

  • the practice they come from does not have the maximum number of controls selected

• finally, select all participants remaining from any practice with a case (person with dementia) but not a control.

The total number of selected cases using this algorithm was 153. When the CFAS core team reviewed the paper records of consent for medical records linkage for these selected participants, four were found to be incomplete and so the number of people without dementia included in the study was reduced to 149.