Homocysteine and Cognitive Function in a Population-based Study of Older Adults

Jyme H. Schafer, MD, MPH; Thomas A. Glass, PhD; Karen I. Bolla, PhD; Margaret Mintz, MS; Anne E. Jedlicka, MS; Brian S. Schwartz, MD, MS


J Am Geriatr Soc. 2005;53(3):381-388. 

In This Article


The Baltimore Memory Study, one of the National Institutes of Health's disparities initiative grants, is a multilevel cohort study of risk factors for cognitive decline in residents aged 50 to 70 of specific, targeted neighborhoods in Baltimore. The methods are described elsewhere.[21] The selected neighborhoods were chosen to provide areas with a broad range of SES and large numbers of whites and African Americans. A cross-sectional analysis of first-visit data was performed.

Sampling and recruitment have been previously described.[21] In brief, individual dwellings in the study area were linked to telephone numbers, and households with telephones were randomly selected for recruitment. Eligibility was then determined on 2,351 subjects (aged 50-70, living at selected households, lived in Baltimore at least 5 years); 1,403 (60.8%) of these subjects were scheduled for an enrollment visit. Of these, 1,140 (81.3%) were enrolled and subsequently tested. The Committee for Human Research of the Johns Hopkins Bloomberg School of Public Health approved the study. All participants provided written, informed consent before testing and were paid $50 for their participation.

Trained research assistants collected all data for this study at the study clinic during 2001-2003. Data were collected in this order: neurobehavioral testing, blood pressure (three measurements sitting using a random zero sphygmomanometer), height, weight, spot urine collection, structured interview, venipuncture, and a satisfaction survey about the visit.

A structured interview included the following self-reported information: demographics, medical history, chronic conditions, and medications (current and historical use). Data on psychosocial and behavioral factors was also gathered, which included depressive symptoms using the Center for Epidemiologic Studies Depression Scale (CESD)[22] and history of alcohol and tobacco consumption. SES was measured using a new assessment tool consisting of 110 questions.[21] Household wealth was assessed using information on income, transfers, and assets. Educational attainment included measures of self-reported years of education completed and credentials acquired (degrees, certificated, trade school). Information from years of education and credentialing was used to create a nine-level ordinal index of educational status.[21]

A trained phlebotomist drew a 10-mL blood specimen into a no-anticoagulant tube; the specimen was clotted, centrifuged, and stored at -20°C within 1 hour. Samples were transported to the Johns Hopkins Bloomberg School of Public Health and stored at -70°C. A commercial laboratory measured serum homocysteine using fluorescence polarization immunoassay (Abbott AxSYM, Abbott Park, IL). The coefficient of variation (CV) of the assay ranged from 2.2% to 3.6%. Lead was measured in the metals laboratory of the Kennedy Krieger Institute (Baltimore) from the first study visit whole blood specimen using anodic stripping voltammetry.[23] A commercial laboratory measured serum cholesterol and high-density lipoprotein cholesterol (HDL-C) using an Olympus AU5200 or AU600 (Tokyo, Japan), with the CV ranging from 2.15% to 2.28% for total cholesterol and 5.20% to 7.06% for HDL-C. Serum triglycerides were measured using an AU5200 (CV from 2.88% to 3.32%), and low-density lipoprotein cholesterol (LDL-C) values were calculated.

Apolipoprotein E (ApoE) genotyping was performed in the Malaria Research Institute Gene Array Core Facility at the Johns Hopkins Bloomberg School of Public Health. Genomic deoxyribonucleic acid (DNA) was extracted from frozen whole blood using the Flexigene DNA Kit (Qiagen, Valencia, CA). The genomic sequence of ApoE in the population under study was confirmed in a subset of individuals using published polymerase chain reaction (PCR) conditions,[24] with forward primer 5'-gacgagaccatgaaggagttgaa and reverse primer 5'-tgctccttcacctcgtcca followed by automated sequencing. For determination of the ApoE Arg112Cys (nucleotide C to T) polymorphism, allelic discrimination using TaqMan MGB probes (Applied Biosystems, Foster City, CA) was employed using a nested approach. Samples were first amplified as described above. Subsequently, PCR amplification was performed in a GeneAmp 9700 PCR machine (Applied Biosystems) with forward primer 5'-gctgggcgcggacat, reverse primer 5'-acctcgccgcggtactg, probe C (Vic)-aggacgtgcgcgg, and probe T (6Fam)-aggacgtgtgcggc using 1 µL of the first amplification product and manufacturer's recommended protocol. After amplification, plate reads were performed in the Prism 7000 Sequence Detection System, and genotypes were determined by manual clustering using the Prism 7000 SDS software version 1 (Applied Biosystems). For determination of the ApoE Arg158Cys (nucleotide C to T) polymorphism, allelic discrimination using TaqMan MGB probes was used (Applied Biosystems). PCR amplification was performed in a GeneAmp 9700 PCR machine (Applied Biosystems) with forward primer 5'-tccgcgatgccgatgac, reverse primer 5'-ccccggcctggtacac, probe C (Vic)-caggcgcttctgc, and probe T (6Fam)-caggcacttctgc using 20 ng genomic DNA and the manufacturer's recommended protocol. Plate reads were performed as described above, and haplotypes were assigned according to a previously described method.[25]

Fasting was not requested of the subjects because study visits were scheduled at all times of the day for logistical reasons. Homocysteine was measured from a sample obtained at the first study visit in the majority of subjects, but 254 subjects only provided plasma at the first visit and so had serum obtained at the second visit for homocysteine measurement. Associations between homocysteine levels and neurobehavioral test scores were evaluated with and without these subjects in the analysis.

The neurobehavioral battery has been previously described and required approximately 90 minutes to complete.[21] The 20 neurobehavioral tests (Appendix) were selected to evaluate a wide range of cognitive domains on which performance was known to vary by age, race/ethnicity, SES, and physical environmental exposures of interest (e.g., lead, mercury). The test session was recorded on audiotape, and a random sample was reviewed routinely as part of quality control. After the tests were completed, two technicians scored them, and a final score for each test was determined by review and agreement.

The main objectives of this analysis were to evaluate the relationships between homocysteine levels and test scores on a comprehensive neurobehavioral battery, controlling for age, race/ethnicity, sex, and other potential confounding variables; and to evaluate whether these relationships were modified by age, sex, race/ethnicity, blood lead levels, or ApoE haplotypes. Multiple linear regression was used to examine the relationships between neurobehavioral tests and homocysteine levels, controlling for covariates. All statistical analysis was performed using Stata version 8.0 (Stata Corp., College Station, TX).

Seventy-eight of the 1,140 enrolled subjects were missing homocysteine data; three were without total cholesterol levels, five were missing body mass index (BMI) data, and seven were missing alcohol consumption data. Regression models thus included a maximum of 1,047 subjects. (Because of performance difficulties, some subjects were unable to complete all neurobehavioral tests.) Subjects with and without homocysteine data did not differ ( P >.05) by age, race/ethnicity, educational level, or neurobehavioral test scores.

Before linear regression modeling, neurobehavioral test scores were standardized (by Z-transformation-subject score minus sample mean divided by sample standard deviation) so that the magnitude of the associations could be easily compared across tests with varying metrics. Regression models were first constructed by including known covariates of neurobehavioral function (age, sex, race/ethnicity, education level, neurobehavioral testing technician) and homocysteine (Model 1, base model). Associations between homocysteine and test scores were evaluated for linearity with squared and cubic terms and with splines (including squared and cubic splines) for homocysteine. Results are presented by cognitive domain, with beta coefficients and 95% confidence intervals (CIs).

Other potential confounders were included sequentially and retained if the homocysteine coefficient changed more than 5% as averaged over the 20 tests. Variables that were evaluated included total cholesterol, HDL-C, LDL-C, BMI, alcohol and tobacco intake, blood lead, systolic and diastolic blood pressures, household income and household assets, medication use (current use of antihypertensives, antidepressants, hormone replacement therapy, or anxiolytics), and CESD score (as a surrogate for depression). Results from two additional models are presented. Model 2 added confounding variables to the base model that met the aforementioned criteria and that are not likely to be in the causal pathway between homocysteine and cognitive function, including alcohol (drinks per month in volume-partitioned quartiles), tobacco use (cigarettes per day by quantity-partitioned quartiles), BMI (kg/m2), and total cholesterol level (mg/dL). Model 3 included several additional important risk factors for cognitive dysfunction-self-reported history (yes vs no) of known or suspected stroke, heart disease, hypertension, diabetes mellitus, and kidney disease, but because homocysteine may be in the multiple-cause pathway for some of these diseases, their inclusion could underestimate associations between homocysteine and test scores, so results of these models are presented separately.

Effect modification by age (divided into quartiles), sex, race/ethnicity, LDL-C (divided into quartiles), blood lead levels (divided into quartiles), and ApoE genotype (at least one ε4 allele vs none) was evaluated by including cross-product terms with homocysteine levels (e.g., to evaluate effect modification by ApoE genotype, a cross-product of ApoE and homocysteine was included). Final models were checked for the assumptions of linear regression and model fit using influence and leverage diagnostic procedures, with examination of residuals and added variable plots.

The magnitude of the associations between homocysteine levels and test scores was examined in two ways. First, from the base linear regression model, the average increase in age at first study visit that was equivalent in its associations with neurobehavioral test scores to an increase from the 25th to the 75th percentile of homocysteine levels (an arbitrarily chosen increase to illustrate the magnitude of the homocysteine relations) was determined. The average expected decline in test scores was calculated; this average decline was then used to calculate the expected increase in age at the first study visit that would result in the same average decline in test scores.

Next, logistic regression was used to model the odds of being in the lowest 25th percentile for each neurobehavioral test score, compared with the upper three quartiles. Odds ratios (ORs) are presented for homocysteine, age, education, and sex. A unique model is presented for each variable to minimize residual confounding, rather than an approach in which each variable was simply dichotomized in a single model. To determine the OR for homocysteine, levels were divided into quartiles to compare the top quartile of levels (≥11.3 µmol/L) with the lowest quartile (<7.6 µmol/L), with age as a continuous variable and education (in nine categories). For the age OR, age was divided into quartiles (<55, 55-<60, 60-65, and >65), and the OR for those in the oldest quartile (compared with those in the youngest quartile reference group) is presented, controlling for homocysteine (continuous variable) and education. For the education OR, subjects with at least a baccalaureate degree were compared with those with less than high school diploma attainment and without completion of trade school, with homocysteine and age included as continuous variables in this model. These logistic regression models included the above variables as well as current tobacco use (yes vs no), alcohol use (a drink in the previous month, yes vs no), BMI (kg/m2), and total cholesterol (mg/dL).


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.