Predicted Mini-Mental State Examination (MMSE) scores by apolipoprotein E ε4 (APOE ε4) allele group.
Predicted Physical Self-Maintenance Scale (PSMS) scores by apolipoprotein E ε4 (APOE ε4) allele group.
Predicted Instrumental Activities of Daily Living (IADL) scores by apolipoprotein E ε4 (APOE ε4) allele group.
Predicted Alzheimer Disease Assessment Scale (ADAS) scores by apolipoprotein E ε4 (APOE ε4) allele group.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Hoyt BD, Massman PJ, Schatschneider C, Cooke N, Doody RS. Individual Growth Curve Analysis of APOE ε4–Associated Cognitive Decline in Alzheimer Disease. Arch Neurol. 2005;62(3):454–459. doi:10.1001/archneur.62.3.454
The apolipoprotein E ε4 (APOE ε4) allele is associated with an increased risk of developing Alzheimer disease (AD). However, findings regarding an association between the APOE ε4 allele and the rate of decline in AD have been mixed.
To examine the relationship between the APOE ε4 allele and the rate of cognitive and functional decline in AD using individual growth curve analyses.
Longitudinal cohort study.
Alzheimer Disease Research Center at Baylor College of Medicine.
A total of 189 patients meeting NINCDS-ADRDA (National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association) criteria for probable AD at baseline who underwent annual follow-up evaluations for at least 2 years.
Main Outcome Measures
Individual growth curve parameters derived from baseline and follow-up performance on global and specific measures of cognitive and functional abilities.
Patients with 2 APOE ε4 alleles exhibited a slower rate of decline on measures of global cognitive functioning and functional abilities. No significant association was detected between the APOE ε4 allele and the rate of decline on measures of specific cognitive functions.
Although the APOE ε4 allele is associated with an increased risk of developing AD, it seems that having 2 APOE ε4 alleles is associated with a slower clinical course. These findings are consistent with hypotheses that the biological processes contributing to the onset of AD are different from those involved in determining its clinical course.
The apolipoprotein E ε4 (APOE ε4) allele is overrepresented in Alzheimer disease (AD) and has been associated with a 2-fold risk in heterozygotes and a 5-fold risk in homozygotes.1 If neuropathologic processes controlling disease onset are also associated with symptom progression, then the rate of cognitive decline may be greater for patients possessing an APOE ε4 allele.2 In the “preclinical” phase of AD, decreased memory performance and faster decline have been associated with the APOE ε4 allele in some samples3 but not in others.4 In AD samples, on global measures of cognitive function, a faster rate of cognitive decline has been associated with the APOE ε4 allele in studies by Craft et al5 and Dal Forno et al,6 whereas a slower rate of cognitive decline has been associated with the APOE ε4 allele in studies by Frisoni et al7 and Stern et al.8
Although various regression models9 and nonlinear growth curve models10 have examined cognitive decline in AD, many such techniques pose rigid methodological requirements that exclude a considerable amount of available data. This study sought to improve on previous longitudinal studies by using the individual growth curve analysis method specified by Francis et al,11 which is ideally suited to longitudinal investigations because it allows for the use of all available data for each patient.
Individual growth curves were constructed for each patient on each measure, and the relationship between the number of APOE ε4 alleles and parameters that characterize these curves was examined. It was hypothesized that having the APOE ε4 allele would be associated with a slower rate of cognitive and functional decline. These findings were expected to confirm those of the only other longitudinal study8 to date using growth curve methods in a large sample.
A total of 189 longitudinal research participants were selected from a database at the Alzheimer Disease Research Center at Baylor College of Medicine. Informed consent was obtained from each patient or caregiver. The protocol and informed consent were approved by the Baylor College of Medicine Committee on Research Including Human Subjects. All participants met the NINCDS-ADRDA (National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association) criteria for probable AD12 at baseline and had completed at least 3 evaluations (baseline plus at least 2 follow-up evaluations). A total of 99 patients completed 3 neuropsychologic evaluations, 42 completed 4 evaluations, 32 completed 5 evaluations, 10 completed 6 evaluations, 4 completed 7 evaluations, and 2 completed 8 evaluations. These data were used in developing the unconditional models. Apolipoprotein E genotypes were available for 151 patients, and these data were used in developing the conditional models. Forty-seven patients had no APOE ε4 alleles, 82 had 1 APOE ε4 allele, and 22 had 2 APOE ε4 alleles. The frequency of the APOE ε4 allele was 42%, consistent with previous estimates in AD samples.
Patients underwent neuropsychologic testing at approximate annual intervals until their participation in the study ended. Because of changes in the test battery and characteristics of individual patients over time, not all measures were administered to all patients during all evaluations. However, the individual growth curve method used allowed for use of all available data. Measures selected for analysis were the Mini-Mental State Examination (MMSE); the Alzheimer Disease Assessment Scale (ADAS) Cognitive Scale, Immediate Verbal Free Recall, and Immediate Recognition (discriminability); the Wechsler Memory Scale–Revised Visual Reproduction–Immediate Recall; the Verbal Series Attention Test (VSAT); the Boston Naming Test (BNT); Letter Fluency; Category Fluency (Animals); the Lawton Physical Self-Maintenance Scale (PSMS); the Lawton Instrumental Activities of Daily Living (IADL); and the Geriatric Depression Scale (patient report). Owing to floor effects at baseline, delayed recall measures were not included in the analyses.
The APOE genotype was determined in serum using the method described by Hixson and Vernier.13 Genomic DNA was prepared from peripheral leukocytes. Restriction isotyping was performed by polymerase chain reaction amplification of APOE DNA sequences containing amino acids at the 112 and 158 positions in a DNA thermal regulator using oligonucleotide primers F4 and F6. The amplified products were then digested with HhaI and subjected to electrophoresis on polyacrylamide gels. The pattern of migration of digested fragments indicated the APOE genotype.
The individual growth curve method specified by Francis et al11,14 shifts the focus of study from group differences in change to individual differences. The growth curve parameters used to describe change at the individual level are thus patient-specific rather than group means. Patient characteristics correlating with change will relate systematically to the parameters of these individual growth curves, allowing for the development of models that predict change at the group level.11
In contrast to other approaches to growth curve analyses, the present analyses use hierarchical linear modeling methods, thus allowing for parameters at the individual and group levels to be estimated simultaneously. At the individual level, unconditional models are developed using within-subject analyses to estimate the parameters of growth curves that best describe change for each individual patient. Variables that allow for increasingly more complex forms of growth (ie, linear and quadratic) are added into the equation describing each patient’s data until they no longer make a statistically significant contribution to explaining the variance. At this stage, the goal is to find the best estimate of change for each individual using as few parameters as possible. Variables may be fixed or random across all patients, depending on whether freeing the parameter increases the fit of the model enough to justify the decrease in parsimony. At the group level, conditional models are simultaneously developed using between-subject analyses in which the growth parameters for individual patients derived from the unconditional model are treated as dependent variables to be predicted by a set of independent variables (ie, number of APOE ε4 alleles). The primary advantage of this specific technique is that all available data for each patient can be used in estimating growth curve parameters. In addition, it allows for differential weighting of data from each patient so that more weight is given to patients whose parameters have been estimated more precisely.
Sample characteristics were analyzed using 1-way analysis of variance and χ2 analyses. No statistically significant group differences were detected for any of these variables (Table 1). However, patients with 2 APOE ε4 alleles seemed to be younger, to have a greater proportion of women, and to have lower baseline MMSE scores than the other groups.
Table 2 details means (SEs) and parameter variance for parameters contributing statistically significantly to unconditional models for each neuropsychologic measure and indicates which parameters had enough systematic variance to allow estimation of separate values for individual patients in the conditional models (random parameters). Variables that lacked enough systematic variance were set at an equal value for all patients in the conditional models (fixed parameters).
The unconditional models indicated curvilinear patterns of decline for 5 of the neuropsychologic measures (ADAS error score, VSAT, BNT, Letter Fluency, and PSMS). For the BNT and Letter Fluency, there was enough systematic variance in intercept, slope, and quadratic terms to allow estimation of separate terms for each patient, resulting in random intercept–random slope–random quadratic models. For the ADAS error score, VSAT, and PSMS, there was not enough systematic variance to allow for estimation of separate quadratic terms for each patient, resulting in random intercept–random slope–fixed quadratic models.
Linear patterns of decline were noted for the remaining 6 measures (MMSE, ADAS Immediate Recall, ADAS Immediate Recognition [discriminability], Visual Reproduction–Immediate Recall, Category Fluency, and IADL). For the MMSE, ADAS Immediate Recall, Category Fluency, and IADL, there was enough systematic variance in intercept and slope terms to allow estimation of separate terms for each patient, resulting in random intercept–random slope models. For ADAS Immediate Recognition [discriminability] and Visual Reproduction–Immediate Recall, there was not enough systematic variance to allow for estimation of separate slope terms for each patient, resulting in random intercept–fixed slope models.
Conditional models were developed to assess the effect of the APOE ε4 allele on cognitive and functional decline over time. Table 3 lists the predicted values of growth curve parameters for each group on each measure. Statistically significant differences in intercept, slope, and quadratic terms represent between-group differences in baseline performance, rate of linear decline over time, and change in rate of decline over time, respectively.
Baseline scores did not differ statistically significantly among groups for any of the neuropsychologic measures. Patients with 2 APOE ε4 alleles exhibited a statistically significantly slower rate of linear decline than those with either 0 or 1 APOE ε4 allele on 1 measure of global cognitive functioning, the MMSE (Figure 1), and on 2 measures of functional abilities, the PSMS (Figure 2) and IADL (Figure 3). Patients with 2 APOE ε4 alleles also exhibited a statistically significantly slower rate of linear decline than those with 1 APOE ε4 allele on another measure of global cognitive functioning, the ADAS (Figure 4). No statistically significant differences in the rate of linear decline were observed for any of the measures of specific neuropsychologic functions. Finally, no statistically significant differences were found for acceleration of rate of cognitive decline between groups for the 5 neuropsychologic measures for which a curvilinear pattern of decline was observed (ADAS error score, VSAT, BNT, Letter Fluency, and PSMS).
The hypothesis that the APOE ε4 allele would be associated with a slower rate of cognitive decline was partially supported for measures of global cognitive functioning and functional abilities but not for measures of specific cognitive functions. Slower rates of decline were found in patients with 2 APOE ε4 alleles than in those with either 0 or 1 APOE ε4 allele on the MMSE and 2 measures of functional abilities. On the ADAS, patients with 2 APOE ε4 alleles also demonstrated a slower rate of decline than those with 1 APOE ε4 allele. Although not statistically significant, similar patterns of decline were noted on specific measures of neuropsychologic functioning. Specifically, patients with 2 APOE ε4 alleles were noted to exhibit the slowest rate of decline on 4 of the 5 measures in which separate slope terms could be estimated (VSAT, ADAS Immediate Recall, BNT, and Category Fluency). These findings parallel those of 2 previous studies,7,8 both of which found that patients with AD and 1 or more APOE ε4 alleles declined at a slower rate on global cognitive measures. The present study found that patients with AD and 2 APOE ε4 alleles declined at a slower rate than those with either 0 or 1 APOE ε4 allele. Although findings regarding the role of the APOE ε4 allele in the rate of cognitive and functional decline in AD have been mixed, the present study and that by Stern et al8 are the only studies, to our knowledge, that have examined large samples of patients with AD over an extended period using growth curve techniques ideally suited for longitudinal analyses. Therefore, the best evidence indicates that progression is slowest in patients with 2 APOE ε4 alleles.
The exact role of the APOE ε4 allele in the clinical course of AD remains unclear. Although the APOE ε4 allele has been associated with an increased risk of developing AD,1 increased amyloid deposition and plaque formation,15 and abnormal neurite maintenance,16 cognitive impairment has not been found to correlate with plaque density but rather with synaptic loss and number of neurofibrillary tangles.17,18 Combining these findings with the present results and those of Frisoni et al7 and Stern et al,8 it seems likely, as Corder et al19 postulated, that the biological processes that lead to the onset of AD are different from those involved in determining its clinical course. That is not to say that the APOE ε4 allele has no effect on the clinical course of AD, as has been suggested by some researchers.20 Instead, although the risk of developing AD is greater with increasing APOE ε4 allele dose, it seems that having 2 APOE ε4 alleles is associated with a slower clinical course.
Strengths of the present study include the large, single-site sample of patients with AD with APOE genotypes and successful follow-up of 2 years or more; the battery, which includes global and specific measures of cognition and function; the inclusion of the Cognitive Scale of the ADAS, a measure used in clinical trials; and a novel analysis approach appropriate for curvilinear and linear patterns of decline that makes use of all available data.
Some methodological factors may have obscured potential statistically significant relationships. The median number of observations per patient was 3 (baseline plus 2 annual follow-up evaluations), allowing only a 2-year “window” into a disease process that can last many years. Also, the number of patients with 2 APOE ε4 alleles (n = 22) was relatively small, which may have reduced statistical power.
Correspondence: Brian D. Hoyt, PhD, Division of Psychosocial Medicine, National Jewish Medical and Research Center, 1400 Jackson St, Room A103, Denver, CO 80206 (HoytB@NJC.org).
Accepted for Publication: May 27, 2004.
Author Contributions:Study concept and design: Hoyt and Massman. Acquisition of data: Cooke and Doody. Analysis and interpretation of data: Hoyt, Massman, Schatschneider, and Doody. Drafting of the manuscript: Hoyt. Critical revision of the manuscript for important intellectual content: Massman, Schatschneider, Cooke, and Doody. Statistical analysis: Hoyt, Massman, and Schatschneider. Obtained funding: Doody. Administrative, technical, and material support: Cooke and Doody. Study supervision: Massman and Cooke.
Funding/Support: This research was partially supported by Alzheimer’s Disease Research Center grant AG08664 from the National Institute on Aging, Bethesda, Md; a Zenith Award from the Alzheimer’s Association, Chicago, Ill (Dr Doody); and the Andrew Goodman Fellowship at National Jewish Medical and Research Center (Dr Hoyt).