Trajectories of cognitive decline in individuals with the most (A) and least (B) dementia pathologic burden (based on Braak stage, cortical amyloid plaques, infarcts, and Lewy bodies) according to delirium status. Individuals with delirium and more dementia pathologic burden have the fastest decline (line A), whereas individuals with no delirium and little dementia pathologic burden have slowest decline (line D). For some individuals, cognitive decline is driven by dementia pathologic burden (no delirium, high pathologic burden) (line B). For other individuals, cognitive decline is associated with delirium (delirium, little pathologic burden) (line C), and this condition is distinct from, but contributory to, classic dementia pathologic burden. P < .001 for line A vs B, P < .001 for line C vs D, and P = .01 for line A vs C. MMSE indicates Mini-Mental State Examination.
eAppendix. Neuropsychological Evaluation of the Cohorts
Customize your JAMA Network experience by selecting one or more topics from the list below.
Davis DHJ, Muniz-Terrera G, Keage HAD, et al. Association of Delirium With Cognitive Decline in Late Life: A Neuropathologic Study of 3 Population-Based Cohort Studies. JAMA Psychiatry. 2017;74(3):244–251. doi:10.1001/jamapsychiatry.2016.3423
What is the association among delirium, the pathologic processes of dementia, and cognitive decline in older persons?
In this cohort of 987 autopsied brains from 3 population-based cohort studies, delirium and the pathologic processes of dementia were associated with cognitive decline; however, the combination of delirium and the pathologic processes of dementia interacted to give the fastest trajectory of cognitive decline.
During cognitive decline in the oldest-old, delirium appears to act independently and multiplicatively to the neuropathologic processes of classic dementia.
Delirium is associated with accelerated cognitive decline. The pathologic substrates of this association are not yet known, that is, whether they are the same as those associated with dementia, are independent, or are interrelated.
To examine whether the accelerated cognitive decline observed after delirium is independent of the pathologic processes of classic dementia.
Design, Setting, and Participants
Harmonized data from 987 individual brain donors from 3 observational cohort studies with population-based sampling (Vantaa 85+, Cambridge City Over-75s Cohort, Cognitive Function and Ageing Study) performed from January 1, 1985, through December 31, 2011, with a median follow-up of 5.2 years until death, were used in this study. Neuropathologic assessments were performed with investigators masked to clinical data. Data analysis was performed from January 1, 2012, through December 31, 2013. Clinical characteristics of brain donors were not different from the rest of the cohort. Outcome ascertainment was complete given that the participants were brain donors.
Delirium (never vs ever) and pathologic burden of neurofibrillary tangles, amyloid plaques, vascular lesions, and Lewy bodies. Effects modeled using random-effects linear regression and interactions between delirium and pathologic burden were assessed.
Change in Mini-Mental State Examination (MMSE) scores during the 6 years before death.
There were 987 participants (290 from Vantaa 85+, 241 from the Cambridge City Over-75s Cohort, and 456 from the Cognitive Function and Ageing Study) with neuropathologic data; mean (SD) age at death was 90 (6.4) years, including 682 women (69%). The mean MMSE score 6 years before death was 24.7 points. The 279 individuals with delirium (75% women) had worse initial scores (−2.8 points; 95% CI, −4.5 to −1.0; P < .001). Cognitive decline attributable to delirium was −0.37 MMSE points per year (95% CI, −0.60 to −0.13; P < .001). Decline attributable to the pathologic processes of dementia was −0.39 MMSE points per year (95% CI, −0.57 to −0.22; P < .001). However, the combination of delirium and the pathologic processes of dementia resulted in the greatest decline, in which the interaction contributed an additional −0.16 MMSE points per year (95% CI, −0.29 to −0.03; P = .01). The multiplicative nature of these variables resulted in individuals with delirium and the pathologic processes of dementia declining 0.72 MMSE points per year faster than age-, sex-, and educational level–matched controls.
Conclusions and Relevance
Delirium in the presence of the pathologic processes of dementia is associated with accelerated cognitive decline beyond that expected for delirium or the pathologic process itself. These findings suggest that additional unmeasured pathologic processes specifically relate to delirium. Age-related cognitive decline has many contributors, and these findings at the population level support a role for delirium acting independently and multiplicatively to the pathologic processes of classic dementia.
Understanding the pathologic basis of cognitive impairment in whole populations is a prerequisite to mitigating the increasing public health burden of dementia.1 Many strands of investigation presuppose that Alzheimer, vascular, and Lewy body pathologic features are the predominant causes of dementia. This paradigm has directed the search for biomarkers, treatments, and potential prevention strategies. However, evidence indicates that these classic pathologic processes do not fully account for the clinical syndrome,2 especially in unselected populations of the oldest-old.3,4 For example, older people may have a large burden of the pathologic processes of classic dementia but no associated clinical dementia and vice versa.
Delirium is a syndrome of acute brain dysfunction characterized by inattention and other mental status impairments. It is a major public health problem that affects at least 20% of older inpatients and has well-documented adverse associations.5 An emerging literature reveals that delirium is a strong predictor of new-onset dementia and acceleration of existing cognitive decline.6-10 These results are consistent across several different settings: after hospitalization,11 in those with dementia,6,12 in postoperative patients,13 and in a community population.8 In multiple animal models of neurodegeneration, triggers of acute cognitive dysfunction, such as systemic inflammation, also exacerbate the pathologic processes14,15 and accelerate functional decline during longer periods.16,17 This finding implies that delirium and/or its causes can contribute to the overall burden of dementia. Moreover, research indicates that 3 of 10 cases are preventable,18 which in turn suggests that delirium interventions might reduce at least some cognitive decline and dementia.
Although delirium is now established as a strong predictor of cognitive decline in older adults,6,8,12 whether it accounts for additional, interrelated, or unexplained pathologic injury that contributes to dementia has not previously been examined. It is possible that when dementia follows delirium it has a different pathologic profile compared with dementia that develops without delirium. Therefore, understanding how delirium affects the evolution of dementia in the context of a particular burden of pathologic findings may offer new insights into independent mechanisms that explain cognitive decline after delirium.
In this study, the challenge was to examine a key hypothesis: that faster cognitive decline associated with delirium would act independently of the cognitive decline associated with the pathologic processes of classic dementia. Accordingly, we investigated the extent to which delirium and the pathologic processes of classic dementia contributed to associated cognitive decline in 3 unselected, population-based cohort studies with neuropathologic autopsy data: the Medical Research Council Cognitive Function and Ageing Study (CFAS), the Cambridge City Over-75s Cohort (CC75C), and the Vantaa 85+ study. These studies represent the entirety of such studies conducted in Europe and provide a unique opportunity to increase the understanding of the clinical significance of delirium and its interrelation with the pathologic processes of dementia in the general population.
The individual studies have previously been described in detail,19-21 and participant-level data have been harmonized as the Epidemiological Clinicopathological Studies in Europe (EClipSE) collaboration.22 Briefly, participants were sampled from general practitioners’ registers (CFAS [1991-2011] and CC75C [1985-2011] in the United Kingdom) and the Population Register Centre (Vantaa 85+ in Finland [1991-2001]) from January 1, 1985, through December 31, 2011. Data analysis was performed from January 1, 2012, through December 31, 2013. The CFAS recruited persons 65 years or older, the CC75C recruited persons 75 years or older, and Vantaa 85+ recruited persons 85 years or older. Individuals were assessed mostly at 2- to 4-year intervals, with some subsamples having annual evaluation. The Mini-Mental State Examination (MMSE)23 was performed in all 3 studies. Additional neuropsychological batteries were also performed, with some differences among the studies (eAppendix in the Supplement). Table 1 summarizes the characteristics of each cohort. Previous work found that participants in the brain donor programs had no systematic differences in clinical characteristics compared with other participants in the cohorts,24 although donors in the CFAS were selected by stratified random sampling, weighted to those who were older and cognitively impaired. Each study had local ethical approval (CFAS centers: Cambridge: North West Anglia Health Authority Local Research Ethics Committee [Peterborough]; Huntingdon Local Research Ethics Committee; Cambridge Local Research Ethics Committee; Gwynedd: Gwynedd Hospitals National Health Service Trust–North West Health Authority Research Ethics Committee (West); Liverpool: Liverpool Local Research Ethics Committee; Newcastle: Newcastle & North Tyneside Health Authority–Joint Ethics Committee; Northumberland and Tyne & Wear Health Authority–Local Research Ethics Committee; Nottingham: Queen’s Medical Centre National Health Service Trust Ethics Committee; Nottingham University Medical School Ethical Committee; City Hospital Ethics Committee; Oxford: Oxfordshire Health Authority: Central Oxford Research Ethics Committee; CC75C: Cambridge Research Ethics Committee; and Vantaa 85+: Ethics Committee of the City of Vantaa). Written informed consent was provided for each study, and all analyses were conducted with deidentified data.
In the CFAS and CC75C, delirium symptoms were a feature of the standardized interview schedules administered by trained interviewers to participants and informants. These schedules assigned diagnostic groups based on validated, structured algorithms for psychiatric disorders, themselves based on DSM-III-R or related classifications.25 Questions included the following: “Were there brief episodes during the 24 hours when s/he seemed much worse and then times when quite clear?” “Were there marked fluctuations in his/her level of attention or alertness?” “Could a physical illness…be sufficient explanation for the subject's mental or psychiatric symptoms (eg, delirious due to acute infection)?” A full list of relevant questions is given in the eAppendix in the Supplement.
At each interview in the Vantaa 85+ study, the examining neurologists assessed participants and informant(s) for a history of any episodes of delirium, with reference to a checklist of DSM-III-R criteria for delirium diagnosis.26 The reported history was corroborated with medical case records that were available at the time of assessment such that the study ascertainment of delirium was retrospectively derived from multiple sources and the overall diagnosis accepted if the examining neurologists judged there was sufficient evidence from participant and informant recall and/or indication in the medical records.
Parafﬁn-embedded brain tissue samples were used to assess neuropathologic markers with investigators masked to clinical data. Each study reported Braak stage as a semiquantitative measure of τ neurofibrillary tangles and neocortical amyloid plaque burden from the Consortium to Establish a Registry for Alzheimer’s Disease protocol.27 The presence of infarcts (>10 mm), lacunes, and hemorrhage was histologically assessed using hematoxylin-eosin. Lewy bodies in the substantia nigra were assessed with hematoxylin-eosin but also included immunohistochemical staining against α-synuclein (or ubiquitin in some of the earlier CC75C specimens) (eAppendix in the Supplement).
Consistent with previous approaches, delirium exposure was operationalized as never or ever.8 Change in MMSE score before death was modeled using a time-to-death random-effects (random slopes) model.28 We were interested in estimating the final trajectory toward death because this approach makes associations with pathologic data easier to define. The mean time from the start of the trajectory identified by the model to death was 5.2 years; therefore, the start point (intercept) for this trajectory was set (centered) at 6 years. This start point is not so near the point of death that rates of change (slopes) cannot be estimated yet not so far from death that the pathologic findings at autopsy might not plausibly be related to the estimated parameters. Six years before death is also comparable to start points from change-point models of the final trajectory of cognitive decline29-31 and in the range observed in other analyses (3-8 years).32 Models were adjusted for baseline MMSE score, age at death (centered at a mean age of 90 years), sex (0 for men, 1 for women), years of education (0-3, 4-7, 8-11, or ≥12), and study. Missing data were assumed to be missing at random given that outcome ascertainment was essentially complete in this brain donor cohort.
The 4 neuropathologic variables of classic dementia that contribute the greatest population-attributable risk for dementia4 were examined: Braak stage (neurofibrillary tangles), neocortical amyloid plaques, vascular pathologic findings (large artery infarcts, lacunes, or hemorrhage), and Lewy bodies in the substantia nigra. In keeping with previous methods, neuropathologic variables were dichotomized (0, none to mild; 1, moderate to severe).3,8,24 This approach allows for simpler interpretation and is more likely to be robust. Individuals were assigned a pathologic burden score based on the number of times they scored in the higher category for each of the 4 markers. Therefore, the overall pathologic burden score ranged from 0 to 4 (ie, being in the lower category for all markers [pathologic burden score of 0], in the upper category of all 4 markers [pathologic burden score of 4], or some combination). Finally, interactions between delirium and pathologic burden ([delirium history] × [pathologic score]) in terms of their effect on both the start point (−6 years before death) and rate of change of MMSE scores were calculated. Full details are given in the eAppendix in the Supplement. All analyses were conducted with STATA statistical software, version 12.1 (StataCorp). P values were calculated though tests of maximum likelihood, where P < .05 was considered significant. All tests were 2-tailed.
There were 987 participants (290 from Vantaa 85+, 241 from the CC75C, and 456 from the CFAS) with neuropathologic data (mean [SD] age at death, 90 [6.4] years; 472 females [67%] without delirium and 210 [75%] with delirium). Table 2 describes the characteristics of the sample. Persons with delirium were slightly older, more likely to be women, and more likely to have more years of education. Neocortical amyloid plaques, vascular pathologic findings, or Lewy bodies were not significantly different in individuals with and without a history of delirium.
Results from the random-effects models that described delirium and cognitive decline are presented in Table 3. The median number of longitudinal observations for participants in the model was 2 (interquartile range, 1-4). In the fully adjusted model (including delirium and pathologic burden), the start point was estimated at 24.7 MMSE points. The start point should be interpreted as the estimated MMSE score 6 years before death in persons in whom all covariates are in the reference category (eg, youngest age, no delirium). For the typical 90-year-old, the mean base rate of decline was 0.35 points per year (base rate indicates all covariates in the reference category, eg, no delirium, lowest pathologic score). There was no significant influence of study source (Vantaa 85+, CC75C, or CFAS) on the model estimates (eAppendix in the Supplement).
Delirium was associated with a mean 2.8-point lower MMSE score (P < .001) 6 years before death. For these persons, the rate of change was an additional 0.37 points per year (P < .001). These coefficients are additive. Therefore, for the typical individual aged 90 years at death with delirium, the estimated MMSE score is 24.7 points (baseline) with −2.8 points equaling 21.9 MMSE points, declining at 0.35 points (base rate) with −0.37 (attributable to delirium) equaling 0.72 points per year.
An increasing pathologic burden score was associated with a lower MMSE score (−0.7 for 1 instance of high dementia pathologic marker, −2.2 point for 2 markers, and −4.4 for 3 or more markers; P < .001). Pathologic burden conferred an additional 0.39-point decline in MMSE score over and above the effects of age and delirium (P < .001).
A significant interaction between delirium and pathologic burden estimated an additional decline of 0.16 MMSE points per year (P = .01). Therefore, individuals with delirium and high dementia pathologic burden had estimated rates of decline of −0.35 points (base rate), −0.37 points (attributable to delirium), −0.39 points (attributable to pathologic burden), and −0.16 points (attributable to interaction), which equals 1.27 points per year. In comparison, the independent effect of age alone on the rate of MMSE score change was 0.01 points per year (ie, MMSE score difference of 0.05 between the ages of 85 and 90 years).
The Figure shows how the rate of cognitive decline varies by delirium and pathologic status. The slowest decline was seen in persons with no history of delirium and least dementia pathologic burden. The fastest decline was seen in persons with a history of delirium and most dementia pathologic burden. Intermediate rates of decline were observed in individuals with delirium but least dementia pathologic burden and in those with no delirium history but most dementia pathologic burden.
This is the first report, to our knowledge, that people with delirium and higher levels of pathologic processes of classic dementia have the greatest cognitive decline. Delirium in the presence of dementia-related neuropathologic processes was associated with cognitive decline beyond that expected for delirium or the neuropathologic process itself. This finding means that delirium may be independently associated with pathologic processes that drive cognitive decline, which are different from the pathologic processes of classic dementia. These findings suggest new possibilities regarding the pathologic correlates of cognitive impairment, positioning delirium, and/or its precipitants as a critically interrelated mechanism.
These results are in keeping with other studies identified in a systematic review33 reporting that delirium is associated with faster trajectories of cognitive decline.6,8,13 A previous report8 from the Vantaa 85+ study raised the possibility that the pathologic processes of classic dementia might not mediate the observed association between delirium and dementia, although the analysis was underpowered. In this study, the larger sample size and the more precise determination of cognitive change in the 6 years before death allow us to be more conclusive about the interrelated effect of delirium on clinicopathologic correlations in dementia. Experimental data from mouse models suggest that delirium may arise through the interaction between systemic or central nervous system inflammation and existing neurodegenerative pathologic processes,34 and acute exacerbation of inflammation clearly leads to neuronal death,14 synaptic changes,35 and accelerated decline.36 These changes occur independently of increased extracellular amyloid. However, we now need to know whether individuals with delirium superimposed on dementia have different patterns of inflammation, synaptic loss, axonal pathologic findings, and/or differential loss of key neuronal populations of the hippocampus and cortex and of cholinergic and noradrenergic projection areas.37
This analysis has a number of strengths. It focuses on a major question arising from the prevalence of cognitive impairment and aging. In terms of study design, the 3 cohorts have high generalizability for the oldest-old populations, who are underrepresented in dementia research despite having the highest prevalence of dementia.38 This is also the first article, to our knowledge, to examine delirium and the pathologic correlates of cognitive decline at the end of life in the general population; the other analysis comes from a leading study in this area, the Religious Orders cohort study,30 which is, however, focused on specific populations. Modeling change in cognitive outcomes as continua rather than simply the presence or absence of dementia allows for an exploration of the effect of delirium across the whole spectrum of cognitive function (ie, from no baseline impairment through mild cognitive impairment to more severe dementia severity). The power to assess such effects as interactions between delirium and neuropathologic processes is unique.
A number of limitations should be taken into account. Delirium was retrospectively ascertained and by slightly different methods. In the Vantaa 85+ study, assessments for history of delirium occurred at each visit, using information from participants, informants, and medical records. Ascertainment of data in the CFAS and CC75C relied on diagnostic interviews at each study visit, but these data are likely to underestimate delirium in the intervening period. The diagnostic classification criteria also varied, although the different diagnostic schedules for delirium have good agreement with DSM-III-R.39 Despite these differences, the results appear to be consistent across the cohorts. The implication, either way, is that core symptoms in delirium—acute fluctuating change in attention in association with acute illness—represent an adverse state for subsequent cognitive trajectories regardless of the exact methods for operationalizing the syndrome. As with other prospective cohort data, the possibility remains that residual confounding contributes to these observed associations. Another consideration is that only a limited range of pathologic markers and comorbidities could be examined in this harmonized data set. Finally, although recent research based on neuroimaging and neuropathologic examination suggests that insults in earlier life can also be malignant,40-42 this hypothesis could not be examined within this study.
Our results indicate that delirium interacts with underlying pathologic processes of classic dementia and so represents a potential independent but interrelated pathologic pathway to chronic cognitive impairment and dementia. If delirium prevention could lead to consequent prevention of dementia,43,44 it will be essential to understand whether certain dimensions of the delirium syndrome might have a greater effect on cognitive trajectories than others. For example, duration, severity, and/or cause (eg, medications vs acute illness, surgery vs sepsis) may be differently important. The degree of preexisting multimorbidity or frailty may have a significant bearing. Animal studies modeling different causes and severities have some scope to elucidate some of these questions, but greater clarity on these issues must also come from careful prospective studies in representative populations. Nonetheless, our findings indicate that clinicians need to be alert to older people’s cognitive changes during acute episodes and in follow-up across all settings and therefore support wider implementation of best practice in delirium prevention.45
Corresponding Author: Daniel H. J. Davis, PhD, MRCP, Medical Research Council Unit for Lifelong Health and Ageing at University College London, 33 Bedford Pl, London WC1B 5JU, England (email@example.com).
Accepted for Publication: October 16, 2016.
Published Online: January 18, 2017. doi:10.1001/jamapsychiatry.2016.3423
Author Contributions: Dr Davis had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Davis, Ince, Matthews, Ely, Brayne.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Davis, Keage, Ely, MacLullich.
Critical revision of the manuscript for important intellectual content: Davis, Muniz-Terrera, Stephan, Fleming, Ince, Matthews, Cunningham, Ely, MacLullich, Brayne.
Statistical analysis: Davis, Muniz-Terrera, Matthews.
Obtained funding: Davis, Fleming, Ince, Matthews, Brayne.
Administrative, technical, or material support: Keage, Fleming, Ince, Ely.
Study supervision: MacLullich.
Conflict of Interest Disclosures: None reported.
Funding/Support: The Cognitive Function and Ageing Study (CFAS) and Cambridge City Over-75s Cohort (CC75C) are member studies of the Cambridgeshire and Peterborough Collaboration for Leadership in Applied Health Research and Care funded by the United Kingdom’s National Institute for Health Research. The Cambridge Brain Bank Laboratory (which processed all the CC75C and some of the CFAS cases) is supported by the National Institute for Health Research Cambridge BioMedical Research Centre. The original Epidemiological Clinicopathological Studies in Europe (EClipSE) harmonization project was established through grant RHAG/094 from the Bupa UK Foundation and a European Union Marie Curie International Incoming Fellowship (Dr Keage). Dr Fleming is supported by grant RHAG/058 from the Bupa UK Foundation, which included the latest CC75C fieldwork funding. Dr Davis is supported by grants 090661/Z/09/Z and 107467/Z/15/Z from the Wellcome Trust. Drs Davis and MacLullich are members of The University of Edinburgh Centre for Cognitive Ageing and Cognitive Epidemiology, part of the cross-council Lifelong Health and Wellbeing Initiative funded by grant G0700704/84698. Funding from the Biotechnology and Biological Sciences Research Council, Engineering and Physical Sciences Research Council, Economic and Social Research Council, and the Medical Research Council is gratefully acknowledged. Dr Keage is supported by Australian National Health and Medical Research Council Fellowship grant 568890. Dr Cunningham is supported by grant 09/09/07 from the Wellcome Trust Senior Research Fellowship.
Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Group Members:Main EClipSE group: Carol Brayne, MD, University of Cambridge, Paul G. Ince, MD, University of Sheffield, Hannah A. D. Keage, PhD, University of Cambridge, Fiona E. Matthews, PhD, Medical Research Council Biostatistics Unit, Ian McKeith, MD, Newcastle University, Tuomo Polvikoski, MD, PhD, Newcastle University, and Raimo Sulkava, MD, PhD, University of Kuopio; EClipSE genetic subgroup: Carol Brayne, MD, University of Cambridge, John Hardy, PhD, University College London, Hannah A. D. Keage, PhD, University of Cambridge, Fiona E. Matthews, PhD, Medical Research Council Biostatistics Unit, David Melzer, MBBCh, MSc, PhD, University of Exeter, Liisa Myllykangas, MD, PhD, University of Helsinki, James Nicoll, BSc, MBChB, MD, FRCPath, University of Southampton, Tuomo Polvikoski, MD, PhD, University of Newcastle, Andy Singleton, PhD, National Institutes of Health, and Pentti Tienari, MD, PhD, University of Helsinki; Others involved: Anders Paetau, MD, PhD, University of Helsinki, Maarit Tanskanen, MD, PhD, University of Helsinki, Terhi Peuralinna, PhD, University of Helsinki, and Minna Oinas, MD, PhD, University of Helsinki.
Additional Contributions: We thank the CFAS, CC75C, and Vantaa 85+ teams; their study participants; and relatives.
Create a personal account or sign in to: