The rising tide of dementia deaths: triangulation of data from three routine data sources using the Clinical Practice Research Datalink

There is increased recording of dementia diagnoses across both primary and secondary care between 2001 and 2015, and this is reflected in an increased number of deaths recording dementia, both as a contributory or underlying cause, a trend also seen in national data [3]. By 2015, the vast majority (97.7 %) of deaths with dementia mentioned on the death certificate had existing evidence of dementia recorded in primary and/or secondary care records. However, over the same period, the percentage of people dying with dementia, but not of dementia has substantially increased. In particular, while the percentage of deaths with dementia mentioned on the death certificate, but not as underlying cause, has increased, this has not kept pace with the rise in dementia diagnoses in primary and secondary care records. Thus by 2015 only half of those dying with dementia have it mentioned on the death certificate as a contributory cause. While not surprising, it does mean that death certification provides an inadequate picture of the prevalence of dementia at the time of death.

One of the major strengths of this study is the breadth of geographical coverage and basis in an unselected population from a large, validated longitudinal primary care database that is broadly representative of the population [13]. The time period assessed includes key dementia related policy changes to QOF [8] and in contrast to some other studies, we have restricted to practices that contributed data over the entire study period for consistency.

In addition to the comprehensive data linkage between the databases [13], HES has universal coverage in England, which is a key strength [15], although due to variation in quality of coding, data about co-morbidities such as dementia may be entered inconsistently across sites. As well as issues with missing data HES does not include outpatient data so any patients with dementia who attended a dementia/memory assessment clinic will not have been coded in HES, though diagnoses from such letters should have been coded in primary care records and therefore included in CPRD. HES-CPRD data first became available from April 1997 [15], which may impact on the results as deaths from earlier in the study period would have a maximum of three years of prior hospital data. The time period assessed includes policy change related to hospital recording of dementia introduced in 2012 [11].

We were constrained to using data for England only, as CPRD linkage to HES and ONS is limited to this area [14]. Data quality in CPRD is reliant on accurate coding by primary care staff and there may be variability in coding across sites, as well as issues with the Read codes associated with dementia as coding changes were introduced during the study period. This is illustrated by the sharp increase in recorded GP dementia awareness after 2006 which corresponded with the introduction of new dementia related codes incentivising QOF after 2006 [7, 8]. There was also a coding change affecting ONS records in 2010, with a modification in the rules on death certification, aimed at improving the accuracy of recording of the underlying cause of death [21]. Whilst this did not impact greatly on the all-cause dementia rates (Supplementary Fig. 2), it has affected the subtypes of dementia recorded, as a correction was made to the coding of vascular dementia and previously coded cerebrovascular disease was corrected to vascular dementia [22]. As a result of these coding changes there may have been some changes in the proportions of underlying cause within the any-mention of dementia group at the 2010 to 2011 transition. After this coding change the number of deaths with any mention of dementia on the death certificate continued to rise.

A recent systematic review [23] of the validity of dementia diagnoses in routinely collected UK electronic health records reported generally high validity, although all studies included were at significant risk of bias. Individual studies have also previously addressed recording of dementia across different data sources. Brown et al. found higher levels of agreement between GP and HES data (95 %) than we reported, but in a small sample of women only (n = 340) [24]. Perera et al. linked dementia diagnoses from a London mental health database to ONS mortality data from 2006 to 2013 and reported an increase in dementia recording over this period, consistent with our findings [25]. A much larger study (47,386 people with dementia) examined the diagnostic validity of dementia in CPRD records from 1998 to 2010, also using HES and ONS records, as we did [26]. Whilst not directly comparable to our study, as they did not examine trends over time and we provide more recent data; they found that 8 % had evidence of a dementia diagnosis across all three sources during this period, compared to our figures of 1.5 % in 2001 and 10.4 % in 2015 [26].

In UK, the Cognitive Functioning in Ageing Study (CFAS) provided important data on trends over time (1989–2016), in accuracy of death certification of dementia, from their large cohort study of patients aged > 65 years (n = 26,699) [4, 27]. They demonstrated an increase in overall unadjusted prevalence of dementia on death certificates (5.3 to 25.9 %) and an increase in sensitivity of dementia reporting on death certificates compared to a gold-standard of study clinical diagnosis of dementia (from 21 % in CFAS I to 45.2 % in CFAS II) [4, 27]. These are consistent with our findings, indicating a similar trend of increased recording of dementia on death certificates, but with a recognition that death certificates still give an inadequate picture of the number of people dying with dementia.

It is not generally well understood what influences decisions regarding diagnoses recorded in the death certificate [28]. During our study period we observed a steady increase in dementia appearing on death certificates between 2001 and 2015, while at the same time age-specific incidence rates of dementia were falling in the UK [27]. This trend has been replicated in other cohort studies in Europe and North America, suggesting that over the last 25 years the incidence rate may have declined by 13 % per decade [29]. It seems plausible that the improvements in the management of cardiovascular disease over time has contributed to this reduction, particularly for vascular dementia [30]. This might also result in a deferral of dementia to older age, however with people living longer, age at death will have increased over this time too, so the likelihood of having dementia by the time of death would almost certainly have risen as well. This could explain the increasing numbers of deaths from dementia observed in this study. Other potential contributing factors over the study period could include the increased public and medical awareness due to activity by campaigning groups and government, as well as increased referral access to memory clinics. The National Dementia Strategy, launched in 2009, has previously been shown to be associated with a significant increase in dementia diagnosis rates and prescriptions of antidementia drugs [31]. Dementia deaths in this study with a prior diagnosis of dementia in both electronic GP and hospital records has increased from 76.3 % to 2001 to 95.7 % in 2015 which may reflect a change in clinicians’ attitudes and less reluctance in entering a diagnosis of dementia due to concerns about stigma and increased case finding and confidence in management [32].

Changes in QOF in primary care and in policy relating to hospital inpatients, both designed to increase the diagnosis rate of dementia, have led to a greater awareness of dementia in clinical practice and increased recording of dementia on death certificates as a primary or contributory cause. Nevertheless, only half of those dying with a prior diagnosis of dementia had dementia recorded on their death certificate in 2015. The high and increasing proportion of people recognized to be dying “with” dementia (in addition to those dying “of” dementia) has important cost and workforce implications for both health and social care service planning for care of older people.

The introduction of Independent Medical Examiners (IME) from April 2019 [33] could impact on dementia recording rates, as the role of the IME extends to discussions with relatives before deciding on cause of death. Repeating the study in the future and comparing it to current findings may give further insights into dementia detection and recording in primary care. Previous work has demonstrated significant regional variations in dementia recording [34] this could also be investigated as a potential explanatory factor.