Association of C-Reactive Protein With Cognitive Impairment FREE

Creactive protein (CRP) is known to be elevated in those with risk factors common to stroke and dementia, including diabetes,^{1 - 2} obesity,³ and smoking,⁴ and is associated with adverse risk of cardiovascular disease,⁵ increased risk of primary stroke,⁶ and increased stroke severity.⁷ High-sensitivity CRP (hsCRP) is increasingly used in clinical practice as a marker of cardiovascular risk and to guide therapy.⁸ Cardiovascular disease⁹ and inflammation may also be important in Alzheimer disease,¹⁰ and hsCRP has been suggested to be an Alzheimer disease biomarker.¹¹ We sought to investigate the cross-sectional association between hsCRP and impairment in specific cognitive domains among a multiethnic nondemented elderly population in northern Manhattan.

The source sample was 2776 participants from a prospective study of aging and dementia in Medicare-eligible northern Manhattan residents aged 65 years and older (Washington/Hamilton Heights-Inwood Columbia Aging Project [WHICAP] II). The WHICAP II cohort represents a combination of continuing members of a cohort originally recruited in 1992 (WHICAP I; n = 602) and members of a new cohort recruited between 1999 and 2001 (n = 2174). The sampling strategies and recruitment outcomes of these 2 cohorts have been described in detail elsewhere.¹² The population from which participants were drawn is composed of individuals from 3 broadly defined ethnic categories (ie, Hispanic, African American, and white). Participants have been followed up at approximately 18-month intervals with similar assessments at each interval. Ethnic group was determined by self-report using the format of the 2000 US Census.¹³ All individuals were first asked to report their race (ie, American Indian/Alaska Native, Asian, Native Hawaiian or other Pacific Islander, black or African American, or white), then, in a second question, were asked whether or not they were Hispanic. Recruitment, informed consent, and study procedures were approved by the institutional review boards of Columbia Presbyterian Medical Center and Columbia University Health Sciences and the New York State Psychiatric Institute.

High-sensitivity CRP was measured in 2008 in frozen plasma obtained in 1999-2001 in the WHICAP II baseline examination. Participants without prevalent dementia at baseline and at least 1 follow-up examination were chosen. Of the total 2776 participants, 356 were excluded owing to prevalent dementia, 544 were excluded owing to lack of follow-up, and 542 were excluded for lack of a frozen plasma sample; hsCRP was measured in 1352 persons. The group with hsCRP measurements was younger than those excluded, had a lower proportion of women than those with prevalent dementia and without blood measurements, had a higher proportion of white individuals than those with prevalent dementia, had a lower proportion of black individuals than those without blood measurements, and had a lower proportion of Hispanic individuals than those with prevalent dementia (eTable 1). Of the 1352 persons with hsCRP measured, 1330 had sufficient information for memory scores, 1328 had sufficient information for visuospatial scores, 1331 had sufficient information for language and executive scores, 906 had sufficient information for Color Trails Test 1, and 854 had sufficient information for Color Trails Test 2.

C-reactive protein was measured with an ultra-sensitive enzyme-linked immunosorbent assay. Sociodemographic covariates included age, sex, race/ethnicity using the format of the 1990 census (subjects were categorized as Hispanic, white, or black), and education (recorded as a continuous variable and categorized for the purposes of these analyses as 0-6, 7-12, 13-16, and >16 years of education).¹³ APOE genotypes were determined as described by Hixson and Vernier¹⁴ with slight modification.¹⁵ We classified persons by the presence (homozygous or heterozygous) or absence of the APOE ε4 allele.

Vascular risk factors have been found to be predictors of cognitive impairment in our cohort.¹⁶ Thus, we included them as covariates. Vascular covariates included type 2 diabetes, hypertension, heart disease, stroke, smoking, and use of lipid-lowering medications. Determination of the presence of diabetes was based on self-report or by medications indicated for the treatment of diabetes. Hypertension was also based on self-report or medication use, but also on blood pressure measurements. Based on standardized criteria,¹⁷ hypertension was defined by a systolic blood pressure greater than 140 mm Hg or diastolic blood pressure greater than 90 mm Hg. Blood pressure measurements did not significantly affect the predictive value over self-report, and results included in this study only reflect self-report and/or medication use. Heart disease was defined by a history of atrial fibrillation or other arrhythmias, coronary artery disease including myocardial infarction or angina pectoris, or congestive heart failure. Stroke was defined according to the World Health Organization criteria¹⁸ and was based on questioning of the participant or relatives, supplemented by a neurological examination or review of medical records. Smoking was also determined by self-report, and individuals were classified as current smokers, former smokers, or nonsmokers. Lipid-lowering medications significantly reduce hsCRP.⁵ Thus, we included them as a covariate. The use of lipid-lowering medications was based on self-report or review of medications.

Learning and memory were assessed with the Selective Reminding Test,¹⁹ including delayed recall and delayed recognition, and with recognition from the Benton Visual Retention Test.^{20 - 21} Visuospatial ability was assessed with the Rosen Drawing Test²² and the Benton Visual Retention Test–Matching.²³ Language was assessed with the Boston Naming Test,²⁴ the Controlled Oral Word Association Test,²⁵ and Category Fluency Test.²⁶ Psychomotor speed was assessed with the Color Trails Test 1.²⁷ Executive Functioning was assessed with the Similarities subtest from the Wechsler Adult Intelligence Scale–Revised²⁸ and the Color Trails Test 2.

The normative sample used to define cognitive impairment was selected from participants recruited in 1992 and 1999 by means of the robust norms approach.²⁹ Details of the normative sample are published elsewhere.³⁰ Demographically corrected T scores were developed on the basis of the Heaton et al³¹ regression method. Influences of age, years of education, race/ethnicity, and sex on each cognitive test score were assessed by performing multiple linear regression analyses. Racial/ethnic group and language (ie, Spanish vs English) were highly related, since most of the Hispanic individuals were tested in Spanish and all of the white and African American individuals were tested in English; therefore, we did not add language into the model. Each of the 4 cognitive domain scores were included as dependent variables: memory (average composite of total raw scores for immediate recall and delayed recall from the Selective Reminding Test and Benton Visual Retention Test–Recognition); language (average composite of total correct responses on the 15-item Boston Naming Test, number of phrases correctly produced on the Boston Diagnostic Aphasia Examination repetition, and number of correct responses on Boston Diagnostic Aphasia Examination comprehension); executive function (average composite of total correct responses on the Mattis Identities and Oddities test, raw score on the Wechsler Adult Intelligence Scale–Revised Similarities subtest, and mean number of words generated during three 60-second trials for category and letter fluency); and visuospatial skill (average composite of number correct on the Rosen Drawing Test and Benton Visual Retention Test–Matching). Continuous predictors were age and years of education. Sex was a categorical predictor, as was racial/ethnic classification. For each of the 4 regression analyses, we first included all 4 predictors in the model, retaining only the variables that significantly contributed to prediction of cognitive test score. The β weights of each of these predictors in the final model as well as the standard error of each regression model were used to calculate predicted scores on each test. These predicted scores were subtracted from each participant's actual composite score to calculate residual scores. Residual scores were converted to T scores according to the following formula: T score = [(residual score/standard error of the estimate for the regression equation) × 10] + 50. T scores have a mean of 50 and a standard deviation (SD) of 10, allowing a T score of 35 to be the −1.5-SD mark for each of the 4 composite scores. We defined cognitive impairment as scores in specific cognitive domains of less than 1.5 SD below these demographically corrected means,³⁰ as previously done in our cohort for the diagnosis of mild cognitive impairment.³² This definition of cognitive impairment differs from the definition of mild cognitive impairment in that it did not require the memory complaint or the functional impairment criteria. The Color Trails Tests 1 and 2 were not originally part of the cognitive scores in our cohort³⁰ and were not available in all participants. Thus, we defined impairment in the Color Trails Tests 1 and 2 separately, also using the 1.5-SD cutoff and conducted secondary analyses with these tests.

First, we examined the distribution of all variables. High-sensitivity CRP was not normally distributed, but was normally distributed after log transformation. We used the χ² test in bivariate analyses of dichotomous variables and the t test for continuous variables. For multivariable analyses, we used logistic regression relating hsCRP to the presence of cognitive impairment. Models were constructed for each of the cognitive scores. We report several models for the multivariable analyses in the tables: one adjusted for age and sex, one further adjusted for education, ethnic group, and APOE ε4, and another one adjusted for vascular risk factors. We report the second model in the text unless otherwise indicated. We decided a priori that changes in the odds ratio (OR) in the model adjusted for vascular risk factors would be evidence of mediation and not confounding since vascular risk factors and hsCRP are in the same hypothetical causal pathway.³³ All analyses were conducted using SAS 9.1.

Table 1 shows the general characteristics of the sample and compares characteristics among hsCRP tertiles. Compared with the first tertile, persons in the third tertile where younger; more likely to be women, black, or Hispanic; less likely to be white and to have the APOE ε4 allele; and more likely to be current or past smokers, have hypertension, and have memory or visuospatial impairment.

We conducted multivariable analyses, first relating log-transformed hsCRP as a continuous variable with the cognitive scores. High-sensitivity CRP was associated with memory impairment (adjusted OR, 1.1; 95% confidence interval [CI], 1.0-1.3); this association was not attenuated by adjustment for vascular risk factors and stroke. Log-transformed hsCRP was also associated with visuospatial impairment (OR, 1.2; 95% CI, 1.0-1.34) after adjusting for sociodemographics and APOE, but was modestly attenuated and became nonsignificant after adjusting for vascular factors (OR, 1.1; 95% CI, 0.96-1.27). We found no association between hsCRP and language, executive, or cognitive impairment in the Color Trails Tests 1 and 2.

We explored the thresholds of the association between hsCRP tertiles and cognitive impairment (Table 2). Compared with participants in the lowest hsCRP tertile, participants in the highest hsCRP tertile had a 50% greater odds of impaired memory, which was not attenuated after adjustment for vascular risk factors. The OR for the second tertile was 1 and not statistically significant, suggesting a threshold for the association between hsCRP and memory impairment.

Subjects in the highest hsCRP tertile had a 60% greater odds of visuospatial impairment (Table 2). Additionally adjusting for vascular factors, heart disease, diabetes, hypertension, smoking, stroke, and use of lipid-lowering medications led to marked attenuation of the association between hsCRP and the visuospatial factor, suggesting that vascular factors may be mediators in this association. Similar to memory, in all models the OR for the middle tertile of hsCRP and visuospatial impairment was not significant; furthermore, crude models of hsCRP and visuospatial factors as well as models adjusted for sociodemographic factors suggested a dose-response relationship between hsCRP and visuospatial impairment.

High-sensitivity CRP was not associated with executive function language. We conducted secondary analyses in a subset of participants with data on Color Trails Tests 1 and 2, but there was no association with either. The OR for the third tertile of hsCRP of the Color Trails Test 1 was 1.1 (95% CI, 0.8-1.5; P = .62) and 0.9 for Color Trails Test 2 (95% CI, 0.7-1.3; P = .75).

We examined effect modification by APOE ε4 by constructing strata of APOE ε4 and hsCRP levels as suggested for the examination of gene-environment interactions (Table 3).¹⁵ Because only the third tertile of hsCRP was associated with memory and visuospatial impairment, we created a dichotomous variable defining high hsCRP as the third tertile. The interaction term of APOE ε4 and high hsCRP was of borderline statistical significance (P = .06). Persons with both the highest tertile of hsCRP and APOE 4ε had a higher risk of memory impairment compared with those with neither. Persons with either risk factor alone did not have a higher risk of memory impairment. These findings suggest an additive interaction. Both risk factors were associated with a higher risk of visuospatial impairment in isolation, with only a modest increase in risk with their joint presence.

In cross-sectional analyses of a large, elderly, multiethnic, community-based cohort, we found that high hsCRP was related to memory and visuospatial impairment. The association between high hsCRP and memory seemed to occur in the presence of APOE ε4.

There are several mechanisms through which inflammation could affect cognition. Endothelial function depends on the sum of all factors contributing to and attenuating atherogenesis and is an important risk factor for cardiovascular outcomes.^{34 - 35} Diseases associated with systemic inflammation may lead to impaired endothelial function, which has been associated with cerebral white matter hyperintensities,^{36 - 37} vascular dementia, and Alzheimer disease.^{38 - 40} Noninfectious systemic inflammatory markers have been independently associated with impaired cerebral blood flow,⁴¹ and animal inflammatory models suggest focal dysregulation in cerebrovascular flow in areas important to memory, such as the hippocampus.⁴² It can be postulated that high systemic inflammation could be a marker of vascular disease,⁸ but it could also directly affect the amyloid cascade.⁴³ Recent findings suggest that CRP is a marker of vascular disease, but its elevation does not have direct effects on vascular outcomes.^{44 - 45} We found that the inclusion of vascular risk factors in multivariable models attenuated the association between hsCRP and visuospatial abilities, but not memory. Acknowledging the limitations of our cross-sectional analyses, one could speculate that the association of hsCRP with visuospatial abilities, affected most by cerebrovascular disease and disruption of frontal subcortical pathways,⁴⁶ is mediated by vascular disease, while the association with memory is explained by nonvascular mechanisms. Surprisingly, we did not find an association of hsCRP with a subcortical pattern of cognitive impairment, with hallmarks of impaired executive function, processing speed, and attention as described in patients with vascular cognitive impairment.⁴⁷

Several studies have examined the association of several markers of inflammation and incident cognitive decline.^{48 - 50} Although some longitudinal studies have found associations specifically between hsCRP and incident cognitive decline,⁴⁹ others have revealed conflicting results, including minimal or no overall association with incident decline in memory,^{51 - 53} dementia,⁵⁰ or neuropsychological test performance.^{54 - 55} Compared with other studies of cognition and hsCRP, our threshold level of the highest hsCRP tertile was much higher in this population than in some other studies of dementia and memory. Studies examining hsCRP and cognition have classified high hsCRP as higher than 1.0 to 4.1 mg/L^{48 - 50 ,52 ,56 - 57} or at least 5.3 mg/L.⁵⁴ However, levels of hsCRP in our study are comparable with other reports in northern Manhattan⁷ as well as elderly Hispanic individuals in California.^{7 ,58}

Among the oldest, persons with high hsCRP and APOE ε4 carriers may be at greatest risk for impaired memory,⁵¹ but this has not previously been found in other studies with wider age ranges.^{57 - 58} Our findings suggest an additive interaction between APOE ε4 and hsCRP for memory impairment, but this observation is limited by the cross-sectional nature of the study.

High-sensitivity CRP is currently being used as a marker of cardiovascular risk⁸ and lipid-lowering treatments.⁵ Our results suggest that it could also be used as a marker of cognitive impairment in persons without dementia and could serve as the basis for interventions. This possibility requires further exploration.

Correspondence: José A. Luchsinger, MD, MPH, 630 W 168th St, PH9E-105, New York, NY 10032 (jal94@columbia.edu).

Accepted for Publication: May 20, 2009.

Author Contributions: Dr Noble had full access to all of 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: Noble, Mayeux, and Luchsinger. Acquisition of data: Manly, Schupf, and Luchsinger. Analysis and interpretation of data: Noble, Manly, Schupf, Tang, Mayeux, and Luchsinger. Drafting of the manuscript: Noble and Luchsinger. Critical revision of the manuscript for important intellectual content: Noble, Schupf, Tang, Mayeux, and Luchsinger. Statistical analysis: Noble, Schupf, Tang, and Luchsinger. Obtained funding: Manly, Mayeux, and Luchsinger. Administrative, technical, and material support: Manly and Mayeux. Study supervision: Manly, Mayeux, and Luchsinger.

Financial Disclosure: Dr Schupf has consulted for Elan Pharmaceuticals.

Funding/Support: Support for this work was provided by grant P01AG07232 from the National Institutes of Health.