Predicting Risk of Cognitive Decline in Very Old Adults Using Three Models: The Framingham Stroke Risk Profile; the Cardiovascular Risk Factors, Aging, and Dementia Model; and Oxi-Inflammatory Biomarkers

Stephanie L. Harrison, PhD; Anton J. M. de Craen, PhD; Ngaire Kerse, PhD; Ruth Teh, PhD; Antoneta Granic, PhD; Karen Davies, PhD; Keith A. Wesnes, PhD; Wendy P. J. den Elzen, PhD; Jacobijn Gussekloo, PhD; Thomas B. L. Kirkwood, PhD; Louise Robinson, MD; Carol Jagger, PhD; Mario Siervo, PhD; Blossom C. M. Stephan, PhD

Disclosures

J Am Geriatr Soc. 2017;65(2):381-389.

Discussion

Three prospective studies of 85-year-old individuals free of stroke or dementia at baseline from the United Kingdom, the Netherlands, and New Zealand found that higher FSRP and CAIDE risk scores and higher oxi-inflammatory load derived from a cumulative score of three cardiovascular biomarkers were associated with incident global impairment. Furthermore, incorporating oxi-inflammatory load scores into the FSRP or CAIDE model improved the ability of the risk models to predict incident global cognitive impairment, although this could only be determined using two of the three studies (Newcastle 85+ Study, Leiden 85-plus Study) because the LiLACS NZ Study did not have the homocysteine measures needed to determine oxi-inflammatory load.

Several longitudinal studies have found a positive association between higher scores from the Framingham and CAIDE risk models and greater risk of cognitive decline or dementia (for a review of studies see^[21]), but when investigating specific cognitive domains, the results have been inconsistent with regard to which cognitive domains higher cardiovascular risk may affect.^{[5–8,11,33]} Furthermore, the majority of studies have focused on midlife and younger old populations, and no study has previously examined the association between cardiovascular risk models and cognitive function in very old adults.

Studies in younger populations have found that homocysteine, IL-6, and CRP predict cognitive decline.^[13–15] Previous findings in the Newcastle 85 + Study have found cross-sectional associations between these biomarkers and global cognitive impairment measured using the MMSE.^[29] Similarly, previous findings in the Leiden 85-plus Study have found cross-sectional associations between homocysteine and cognitive impairment, but this association was not found with rate of cognitive decline.^[34]

In the current study, higher levels of biomarkers for oxidative stress and inflammation were longitudinally associated with greater risk of developing global and domain-specific cognitive (speed and attention) impairment. Biomarkers of cardiovascular risk may be useful for identifying individuals at risk of future cognitive impairment, because oxidative stress and inflammation are implicated in the pathophysiology of dementia. Higher levels of oxidative stress, impaired cellular function linked to abnormal protein accumulation, and modification of molecular structures may have direct effects on neuronal structure and integrity, affecting cognitive function.^[35] Furthermore, inflammation is thought to be important in neurodegeneration, contributing to the development of some of the classic hallmarks of Alzheimer's pathology such as amyloid-beta plaques.^[36] These findings have been formalized in the theory of inflammaging as a critical factor in the pathogenesis of age-related chronic cardiovascular and neurodegenerative diseases.^[37,38]

Current dementia risk prediction models are not sufficient to detect those at greatest risk of developing cognitive impairment or dementia.^[16] Factors incorporated into current dementia risk prediction models include demographic factors (e.g., age, sex, ethnicity), subjective cognitive complaints, functioning (as measured using activity of daily living scales), neuropsychological test scores, health-related measures (e.g., history of CVD, body mass index), lifestyle measures (e.g., smoking status, alcohol intake), dietary factors (e.g., folic acid and fish intake), magnetic resonance imaging results (e.g., white matter disease), and others (e.g., family history of dementia). The best models were described as those that incorporated a diverse range of risk factors, but it is most likely that there will not be one model that is suitable for all populations, and different dementia risk models may need to be developed for different age groups.^[39] No such dementia risk model has been validated in a very old population. The development of a highly accurate model for discriminating those at high risk of future dementia from those at medium and low risk would be required before screening the older population for future risk of dementia in primary care practice could become a possibility. Incorporation of the biomarkers investigated in this study into the classical cardiovascular risk factors of the FSRP or CAIDE models may be useful to investigate when developing a dementia risk prediction model for very old adults.

There are strengths and limitations to this study. The Newcastle 85+, Leiden 85-plus, and LiLACS NZ cohorts are prospective longitudinal cohort studies of very old adults, and this is the first study that has aggregated data from all three cohorts.

There are some limitations. The LiLACS NZ Study did not have homocysteine levels needed to determine oxi-inflammatory load, so findings for oxi-inflammatory load are based on the Newcastle 85+ and Leiden 85-plus studies only.

Results were not always consistent across the studies; these discrepancies may be due to differences in the assays used to determine CRP and IL-6 between the two studies, because the Leiden 85-plus Study used less-sensitive assays, which led to the attribution of values of 0 for results lower than the limit of detection (number of 0 values: CRP = 82, IL-6 = 119); differences in the cohorts themselves (e.g., education levels; the Leiden 85+ Study did not have years of education, which was needed to calculate CAIDE scores, the LiLACS Study did not have the apolipoprotein E4 measurement, which was required for the CAIDE model); differences in the cognitive tests used to assess domain-specific cognitive function; the smaller sample size of the Leiden 85-plus and LiLACS NZ studies; or a cohort effect related to differences in birth year between cohorts (1913–1915 (Leiden 85-plus Study), 1921 (Newcastle 85+ Study), 1925 (LiLACS NZ Study)).

Further research in very old adults is needed to gain a full understanding of the association between the Framingham models, the CAIDE models, or cardiovascular biomarkers and cognitive decline, in particular with respect to which cognitive domains may be most likely to be affected.

There is no recommended tool for identifying very old individuals at risk of developing cognitive impairment or dementia.^[16] Combining oxi-inflammatory load with the FSRP or the CAIDE model may improve the ability of these models to predict cognitive changes. Biomarkers would be relatively easy to measure in a clinical setting and could potentially provide clinicians with an overview of an individual's cardiovascular health in addition to their future risk of cognitive impairment. Intervention strategies to reduce oxi-inflammatory load could potentially target improvements in cardiovascular health and cognitive function.

1 2 3 4

Next Section

Acknowledgments
The Newcastle 85+ Study was supported by the National Institute for Health Research Newcastle Biomedical Research Centre based at Newcastle upon Tyne Hospitals National Health Service Foundation Trust and Newcastle University. We acknowledge the operational support of the North of England Commissioning Support Unit and of the local general practitioners and their staff.

We thank Professor Peter MacFarlane for assistance with left ventricular hypertrophy coding. We also thank the research, management, and clerical team for outstanding work throughout, as well as many colleagues for their expert advice. Thanks are due especially to the study participants and, where appropriate, their families and caregivers.

LiLACS NZ: We acknowledge the expertise of the Western Bay of Plenty Primary Health Organisation, Ngā Matāpuna Oranga Kaupapa Māori Primary Health Organisation, Te Korowai Aroha Trust, Te Rūnanga o Ngāti Pikiao, Rotorua Area Primary Health Services, Ngāti Awa Research & Archives Trust, Te Rūnanga o Ngāti Irapuaia and Te Whānau ā Apanui Community Health Centre in conducting the study through the Bay of Plenty and Rotorua. We thank all participates and their Whānau (extended family) for their participation and the local organizations that promoted the study.

The Newcastle 85+ Study was approved by the Newcastle and North Tyneside 1 Research Ethics Committee. The Leiden 85-plus Study was approved by the Medical Ethics Committee of the Leiden University Medical Centre. The Northern X Regional Ethics Committee of New Zealand granted ethical approval for the LiLACS NZ Study in December 2009.

Sponsor's Role
The funding sources had no role in the preparation of this paper.

Table 1. Baseline Characteristics of Participants in the Newcastle 85 + Study, Leiden 85-plus Study, and Life and Living in Advanced Age, a Cohort Study in New Zealand (LiLACS NZ) Used in This Analysis (Excluding Those with Dementia or Stroke at Baseline)
Characteristic	Newcastle 85+ Study, n = 616	Leiden 85-plus Study, n = 444	LiLACS NZ Study, n = 396
Risk model components
Male, %	39.9	34.6	48.0
Systolic blood pressure, mmHg, mean ± SD	151.4 ± 23.5	155.2 ± 18.7	151.6 ± 22.8
Total cholesterol, mg/dL, mean ± SD	192.0 ± 48.0	220.8 ± 43.7	199.8 ± 41.8
Body mass index, kg/m², mean ± SD	24.5 ± 4.4	27.1 ± 4.5	26.5 ± 4.3
Blood pressure–lowering drugs, %	70.8	43.9	60.4
Current smoker, %	5.3	16.0	5.1
Diabetes mellitus, %	18.2	13.9	13.1
Atrial fibrillation, %	14.2	9.1	12.2
History of cardiovascular disease, %	38.7	33.6	42.2
Physically active, %	44.0	29.0	37.7
Education <10 years or low category, %	64.0	72.0	27.0
Risk models
Framingham Stroke Risk Profile score, mean ± SD (range)	19.5 ± 3.6 (11–33)	19.1 ± 3.8 (10–33)	18.9 ± 3.7 (10–31)
Low, %	42.2	34.3	33.7
Medium, %	25.0	34.8	36.2
High, %	32.5	30.9	30.1
CAIDE score, mean ± SD (range)	8.5 ± 1.9 (4–13)	10.2 ± 1.7 (4–15)	7.9 ± 1.8 (4–13)
Low, %	48.7	59.3	48.4
Medium, %	24.3	16.8	34.2
High, %	27.0	23.9	17.4
Biomarkers, median (IQR)
Homocysteine, μmol/L	16.7 (7.9)	12.5 (5.5)	—
Interleukin 6, ng/mL	19.7 (20.6)	10.0 (53.0)	2.3 (1.0)
C-reactive protein, mg/L	2.5 (4.9)	3.0 (7.0)	2.9 (4.2)
Oxi-inflammatory load, mean ± SD	−0.1 ± 2.0	−0.1 ± 2.0	0.1 ± 1.7
Cognitive function scores, median (IQR)
Global cognitive function	27.3 (2.8)	27.0 (4.0)	27.0 (3.2)
Memory, median (IQR)	0.6 (0.2)	0.2 (1.8)	—
Attention, median (IQR)	1,521.8 (290.3)	74.8 (37.9)	—
Speed, median (IQR)	417.6 (149.6)	16.0 (9.0)	—

The Cardiovascular Risk Factors, Aging, and Incidence of Dementia (CAIDE) score for LiLACS NZ does not include apolipoprotein E4; the oxi-inflammatory load for the study is based only on interleukin-6 and C-reactive protein. Education for the Leiden 85-plus Study is based not on years of education but is determined according to the categorizations of education used in that study.
In the Newcastle 85+ Study, memory was measured using the sensitivity index for recognition ability, attention was measured using power of attention, and speed was measured using simple reaction time—all part of the Cognitive Drug Research computerized assessment system.
In the Leiden 85-plus Study, memory was measured using the Word-Learning Test, Immediate and Delayed Recall (based on sum of z scores); attention was measured using the Stroop Test Part 3; and speed was measured using the Letter Digit Coding Test. Numbers exclude people with dementia or stroke at baseline.
IQR = interquartile range; SD = standard deviation.

Table 2. Results of Cox Proportional Hazards Model of Association Between Framingham Stroke Risk Profile (FSRP) Model, Cardiovascular Risk Factors, Aging, and Incidence of Dementia (CAIDE) Model, Oxi-Inflammatory Load, and Incident Global Cognitive Impairment After Follow-Up for the Newcastle 85+ Study, Leiden 85-plus Study, and Life and Living in Advanced Age, a Cohort Study in New Zealand (LiLACS NZ) Cohorts
Global Cognitive Function	Meta-analysis	Newcastle 85+ Study	Leiden 85-plus Study	LiLACS NZ Study
Global Cognitive Function	Hazard Ratio (95% Confidence Interval) P-Value
FSRP (reference low)
Medium	—	0.95 (0.57–1.59) .85	0.99 (0.65–1.53) .97	1.56 (0.64–3.77) .33
High	1.46 (1.08–1.98) .01	1.15 (0.71–1.86) .56	1.44 (0.93–2.22) .10	3.53 (1.44–8.63) .01
CAIDE (reference low)
Medium	—	0.83 (1.50–0.46) .54	1.84 (1.14–2.96) .01	2.43 (1.13–5.22) .02
High	1.53 (1.09–2.14) .01	1.14 (0.64–2.03) .65	1.64 (1.04–2.58) .03	2.87 (0.97–8.47) .06
Oxi-inflammatory load (reference medium)
Low	—	1.02 (0.53–1.96) .95	0.65 (0.31–1.35) .25	—
High	1.73 (1.04–2.88) .04	2.13 (1.08–4.20) .03	1.32 (0.61–2.86) .48	—
CAIDE + oxi-inflammatory load (reference low)
Medium	—	1.38 (0.77–2.46) .28	1.28 (0.83–1.97) .26	—
High	1.93 (1.39–2.67) <.001	1.96 (1.27–3.42) .02	1.89 (1.18–3.02) .008	—
FSRP + oxi-inflammatory load (reference low)
Medium	—	1.20 (0.75–1.93) .45	1.36 (0.88–2.09) .17	—
High	1.65 (1.17–2.33) <.001	1.59 (0.94–2.69) .08	1.70 (1.08–2.68) .02	—

Global cognitive function measured using the Mini-Mental State Examination (MMSE) in all studies. Oxi-inflammatory load is a sum score of the three biomarkers; standardized z-scores for each log-transformed biomarker were calculated, and the sums of these scores were determined to characterize oxi-inflammatory load. Categories were based on tertiles for the FSRP and CAIDE models and for oxi-inflammatory load: low is <10th percentile, middle is 10th–90th percentile, and high is >90th percentile. Impairment is ≤25 points on the MMSE for global cognitive function. Oxi-inflammatory load models adjusted for sex, years of education, current alcohol consumption, and smoking status.

Table 3. Linear Mixed Model Results of the Association Between the Framingham Stroke Risk Profile (FSRP); the Cardiovascular Risk Factors, Aging, and Incidence of Dementia (CAIDE) Model; Oxi-Inflammatory Load and Cognitive Decline After Follow-Up for the Newcastle 85+ Study, Leiden 85-plus Study, and Life and Living in Advanced Age, a Cohort Study in New Zealand (LiLACS NZ) Cohorts
Global Cognitive Function	Newcastle 85+ Study			Leiden 85-plus Study			LiLACS NZ Study
	Cross-Sectional	B (Standard Error) P-Value	Change Due to the Risk Model	Cross-Sectional	Change over Time	Change Due to the Risk Model	Cross-Sectional	Change over Time	Change Due to the Risk Model
	Cross-Sectional	Change over Time	Change Due to the Risk Model	Cross-Sectional	Change over Time	Change Due to the Risk Model	Cross-Sectional	Change over Time	Change Due to the Risk Model
FSRP (reference low)
Medium	−0.026 (0.098) .32	0.109 (0.019) <.001	0.011 (0.030) .71	−0.033 (0.111) .77	0.129 (0.015) <.001	−0.031 (0.021) .15	0.029 (0.091) .75	0.018 (0.029) .54	−0.069 (0.041) .09
High	−0.083 (0.093) .37	0.109 (0.019) <.001	0.024 (0.031) .43	0.159 (0.115) .17	0.129 (0.015) <.001	−0.015 (0.023) .53	−0.052 (0.097) .59	0.018 (0.029) .54	0.043 (0.046) .36
CAIDE (reference low)
Medium	0.071 (0.105) .50	0.109 (0.021) <.001	−0.018 (0.035) .60	0.126 (0.128) .33	0.102 (0.012) <.001	0.057 (0.025) .02	0.053 (0.088) .54	0.008 (0.026) .76	−0.011 (0.040) .99
High	0.123 (0.103) .23	0.109 (0.021) <.001	0.053 (0.037) .15	0.493 (0.112) <.001	0.102 (0.012) <.001	0.001 (0.022) .99	0.348 (0.109) .001	0.008 (0.026) .76	−0.032 (0.054) .54
Oxi-inflammatory load (reference medium)
Low	−0.148 (0.138) .28	0.118 (0.014) <.001	0.022 (0.041) .60	0.060 (0.147) .68	0.118 (0.010) <.001	−0.023 (0.031) .45	—	—	—
High	0.20 (0.121) .10	0.118 (0.014) <.001	−0.018 (0.047) .70	0.320 (0.127) .01	0.118 (0.010) <.001	0.007 (0.033) .83

Global cognitive function was measured using the Mini-Mental State Examination (MMSE) in all studies. Cognitive tests that were not normally distributed were transformed. Oxi-inflammatory load is a sum score of the three biomarkers; standardized z-scores for each log-transformed biomarker were calculated, and the sums of these scores were determined to characterize oxi-inflammatory load. Categories were based on tertiles for the FSRP and CAIDE models and for oxi-inflammatory load: low is <10th percentile, middle is 10th–90th percentile, and high is >90th percentile. For the non-MMSE measurements, impairment was defined as a score 1.5 standard deviations or more below (or above where higher scores indicate worse performance) the mean score. If scores were not normally distributed, this was calculated as the 93rd percentile for scores for which higher numbers reflect worse cognitive function and the 7th percentile for scores for which lower numbers reflect worse cognitive function. Oxi-inflammatory load models adjusted for sex, years of education, current alcohol consumption, and smoking status.

Authors and Disclosures

Stephanie L. Harrison, PhD^a, Anton J. M. de Craen, PhD^b, Ngaire Kerse, PhD^c, Ruth Teh, PhD^c, Antoneta Granic, PhD^d,e, Karen Davies, PhD^d,e, Keith A. Wesnes, PhD^f,g, Wendy P. J. den Elzen, PhD^h,i, Jacobijn Gussekloo, PhDⁱ, Thomas B. L. Kirkwood, PhD^j, Louise Robinson, MD^a,d, Carol Jagger, PhD^d, Mario Siervo, PhD^d,k,* and Blossom C. M. Stephan, PhD^a,d,*

^aInstitute of Health and Society, Newcastle University, Newcastle upon Tyne, United Kingdom; ^bDepartment of Gerontology and Geriatrics, Leiden University Medical Center, Leiden, the Netherlands; ^cDepartment of General Practice and Primary Health Care, School of Population Health, University of Auckland, Auckland, New Zealand; ^dInstitute of Ageing, Campus for Ageing and Vitality, Newcastle University; ^eAgeing, Geriatrics and Epidemiology, Institute of Neuroscience, National Institute for Health Research Newcastle Biomedical Research Centre in Ageing and Chronic Disease, Newcastle University and Newcastle upon Tyne NHS Foundation Trust, Newcastle upon Tyne; ^fWesnes Cognition Ltd, Streatley on Thames; ^gDepartment of Psychology, Northumbria University, Newcastle upon Tyne, United Kingdom; ^hDepartment of Clinical Chemistry and Laboratory Medicine; ⁱDepartment of Public Health and Primary Care, Leiden University Medical Center, Leiden, the Netherlands; ^jInstitute of Cell and Molecular Biosciences; and ^kInstitute of Cellular Medicine, Institute of Ageing, Campus for Ageing and Vitality, Newcastle University, Newcastle upon Tyne, United Kingdom.

*Joint final authors (equal contribution).

Address correspondence to
Stephanie Harrison, Institute of Health and Society, Newcastle University, Baddiley-Clark, Richardson Road, Newcastle upon Tyne, United Kingdom NE2 4AX. E-mail: s.harrison@newcastle.ac.uk

Conflict of Interest
KAW has specified relationships with Bracket Global and Wesnes Cognition Ltd that might have an interest in the submitted work in the previous 3 years.

Author Contributions
SLH: conception, design, drafting of manuscript, statistical analysis, interpretation of results. AJMD: co-principal investigator of the Leiden 85-plus Study, statistical analysis, interpretation of results, critical review of manuscript. NK: leader of LiLACS NZ Study, critical review of manuscript. RT: CVD aspects of LiLACS NZ Study, critical review of manuscript. AG: critical review of manuscript. KD: design of the Newcastle 85 + Study, critical review of the manuscript. KAW: development of CDR system, supply of system to Newcastle 85+ Study, critical review of manuscript. WPJD: critical review of manuscript. JG: co-principal investigator of Leiden 85-plus Study, critical review of manuscript. TBLK: principal investigator of Newcastle 85+ Study, critical review of manuscript. LR: critical review of manuscript. CJ: design of Newcastle 85+ Study, conception and design of manuscript, interpretation of results, critical review of manuscript. MS: conception and design of manuscript, interpretation of results, critical review of manuscript. BCMS: conception and design of manuscript, interpretation of results, critical review of manuscript.