Due to high cost of amyloid imaging, its use of amyloid imaging to confirm amyloid pathology is limited in clinical practice. It is of importance to develop a model to predict cerebral amyloid positivity using clinical data obtained from a memory clinic.
A total of 410 participants who had symptom of subjective cognitive decline and underwent amyloid PET and apolipoprotein ε (APOE) genotyping were retrospectively enrolled from January 2016 to January 2019. Models for cerebral amyloid positivity prediction were developed in all subjects, mild cognitive impairment (MCI) subjects, and Alzheimer’s disease (AD) dementia subjects through multivariate logistic regression analysis. The performance of the models was assessed using receiver operating characteristic (ROC) curve analysis and the area under the curve (AUC) values.
Age, sex, years of education, body mass index (BMI), APOE4, and mini mental state examination score (MMSE) were selected for the final model for all subjects. The AUC value of the ROC curve was 0.775. Age, sex, years of education, BMI, and APOE4 were selected for the final model for MCI subjects. The AUC value was 0.735. Age, sex, years of education, BMI, APOE4, MMSE, and history of hypertension were selected for the final model for AD dementia subjects. The AUC value was 0.845.
This study found that models using clinical data can predict cerebral amyloid positivity according to cognitive status. These models can be useful as a screening tool predict cerebral amyloid deposition in cognitively impaired patients in a memory clinic.
The global population of Alzheimer’s disease (AD) patients is about 35 million people and expected to be 115.3 million people in 2050 [
AD has been diagnosed based on clinical diagnostic criteria in most clinical practice. However, the diagnostic method in AD based on clinical diagnostic criteria has limitations, showing low sensitivity (71% to 81%) and specificity (about 70%), since confirmation of AD is based on the histopathologic findings [
In 2013, the Alzheimer’s Association and the Society of Nuclear Medicine and Molecular Imaging convened the Amyloid Imaging Taskforce to gather empirical evidence to support the clinical utility of amyloid imaging. Patients who have persistent unexplained mild cognitive impairment (MCI), patients with atypical presentations of AD, and patients who develop progressive neurocognitive disorder (NCD) symptoms at an atypically early age are more likely to benefit from amyloid imaging [
A total of 410 patients with various diagnosis, 410 participants with MCI or AD were enrolled in this study from a cohort of cognitive impairment in the Seoul St. Mary’s Hospital, Catholic Medical Center. This study included participants who underwent both amyloid PET and apolipoprotein ε (APOE) genotyping from January 2016 to January 2019.
Detailed history taking, comprehensive physical examinations, including body weight, height, and blood pressure (BP), laboratory tests, and brain imaging, such as magnetic resonance imaging (MRI), or computed tomography (CT), were performed in all of the subjects. With regard to clinical history, age, sex, years of education, family history of dementia, and past medical history, including diabetes mellitus (DM), hypertension (HTN) were collected. Body mass index (BMI) was calculated by the body weight divided by the square of the body height. To exclude other causes of cognitive impairment, a thyroid function test (thyroid stimulating hormone), lipid profile (total cholesterol), and vitamin B12 were assessed using blood. APOE genotype was determined by PCR using blood samples in EDTA tubes. Subjects with at least one ε4 allele were classified as APOE4 positive. Mini mental state examination (MMSE) and sum of the box score for clinical dementia rating (CDR-SB) were used to assess global cognition in all participants. In addition, a neuropsychological battery such as the Seoul Neuropsychological Screening Battery (SNSB) [
Clinical diagnosis was made based on the clinical diagnostic criteria for subjective memory complaint (SMC), MCI, Alzheimer’s dementia, frontotemporal dementia (FTD), vascular dementia (VaD), and dementia with Lewy body (DLB) [
Amyloid PET/CT scans were performed on a Discovery PET/CT 710 (GE Medical Systems, Milwaukee, WI, USA) or a Biograph TruePoint 40 PET/CT (Siemens Medical Solutions USA, Knoxville, TN, USA). Here, 18F-florbetaben and 18F-flutemetamol were used as 18F-labeled Aβ targeting tracers. Patients received an intravenous injection of 8.0 mCi of 18F-florbetaben or 5.0 mCi of 18F-flutemetamol and rested in a waiting room. Image acquisition started approximately 90 minutes after the injection, with a scan duration of 20 minutes.
The 18F-florbetaben PET and flutemetamol PET scans were visually rated as either amyloid-positive or amyloid-negative by the trained nuclear medicine physician. Interpretation of the 18F-florbetaben PET images was achieved by visually comparing the activity in the cortical gray matter with the activity in the adjacent cortical white matter. Brain regions of the—lateral temporal cortex, frontal cortex, posterior cingulate cortex/precuneus, and parietal cortex—should be systematically and cortical tracer uptake at any of the cortical target regions, while a negative scan indicates good grey-white matter contrast, with no tracer uptake at target regions [
The following variables were used for developing amyloid prediction models; age, sex, years of education, BMI, family history of dementia, history of hypertension or DM, MMSE, CDR-SB, APOE4 genotype, total cholesterol, TSH, vitamin B12, blood pressure and clinical diagnosis. These variables were included because they are well-known risk factors for dementia.
Continuous variables, including age, education (years), BMI, the MMSE, the CDR-SB, systolic and diastolic BP, TSH, total cholesterol, and vitamin B12, are presented with the mean and standard error (SE). Independent t-test was used to compare subjects with amyloid deposition and those without amyloid deposition. Categorical variables, including sex, family history of dementia, DM, HTN, APOE genotype, and clinical diagnosis, are presented with number and percentage. Chi-square test was used to compare between group with amyloid deposition and group without amyloid deposition. p value<0.05 was considered as statistically significant.
Through multivariate logistic regression analysis using a forward conditional method, we developed final models for cerebral amyloid positivity prediction. Age, sex, and year of education were entered as fixed variables. Sex, the categorical variable, was coded as 1 for male and 2 for female. Other variables were sequentially entered into the model using the forward likelihood ratio (LR) method. Odds ratios (ORs) and 95% confidence interval (CIs) were used to identify associations between clinical variables and amyloid positivity. We developed three cerebral amyloid prediction models according to cognitive status for all subjects, MCI subjects, and AD dementia subjects. The discriminative power of the cerebral amyloid prediction models was assessed using receiver operating characteristic (ROC) curve analysis and the area under the curve (AUC) values. Also, we produced simple models for cerebral amyloid prediction using logistic regression that included age, sex, years of education, and one of the other variables in the final model to compare the final models with each of the simple models. Additionally, we explored how the Youden index, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) increases from 0.1 to 0.9 by 0.1 unit and presented the optimal cutoff point for probability for being amyloid positive (
The Institutional Review Board of the Catholic Medical Center approved this study protocol (KC20RESASI0639).
The characteristics of the 410 subjects according to amyloid positivity are summarized in
We developed three final models using multivariate logistic regression analysis in all subjects, MCI subjects, and AD dementia subjects.
P=probability for being cerebral amyloid positive
Pcase=(exp [Logitcase])/(1+exp [Logitcase])
Logitall=β0+β1 (age)+β2 (sex)+β3 (years of education)+β4 (APOE4 positivity)+β5 (MMSE score) (β0=0.158, β1=0.029, β2=0.247, β3=0.034, β4=1.937, β5=-0.155)
LogitMCI=β0+β1 (age)+β2 (sex)+β3 (years of education)+β4 (APOE4 positivity) (β0=-3.668, β1=0.037, β2=-0.016, β3=0.004, β4=1.915)
LogitAD=β0+β1 (age)+β2 (sex)+β3 (years of education)+β4 (APOE4 positivity)+β5 (MMSE score)+β6 (history of hypertension) (β0=-0.012, β1=0.039, β2=1.043, β3=0.075, β4=2.631, β5=-0.198, β6=-1.615)
The first model was to predict cerebral amyloid positivity for all subjects with subjective cognitive decline in our memory clinic. Age, sex, years of education, BMI, APOE4, and MMSE score were selected for the final model for all subjects (
The major findings of the present study were as follows. First, we found that older age, female sex, lower BMI, APOE4 genotype, lower MMSE score, higher CDR-SB score, and elevated blood total cholesterol independently predicted cerebral amyloid positivity in all subjects. Second, we developed models to predict the risk of cerebral amyloid positivity in all subjects, MCI subjects, and AD dementia subjects. Additionally we found superiority of ability to predict amyloid positivity by comparing final models and simple models.
Several previous studies have made prediction models for cerebral amyloid positivity. Bahar-Fuchs et al. [
In the models we developed, the AUC values were 0.775 for all subjects, 0.735 for MCI subjects, and 0.845 for AD dementia subjects. This means that the models can predict the positive rates of cerebral amyloid PET in 77.5%, 73.5%, and 84.5%, respectively. The all subjects model was applied to patients whose objective cognitive function level was not identified. The possibility of amyloid PET positivity was 77.5% in the subjects who were elderly, female, and had a low MMSE score, and had the APOE4 gene. The MCI model was applied to patients diagnosed with MCI. The possibility of amyloid PET positivity was 73.5% in subjects who were elderly, and had the APOE4 gene. These patients may need to undergo an amyloid PET or prepare for an early preventive intervention. In addition, the AD model was applied to patients probable AD dementia by NIA-AA diagnostic guideline. The possibility of amyloid PET positivity was 84.5% in subjects who were elderly, had low MMSE score, the APOE4 gene, and no history of hypertension. This result can help us develop additional diagnostic or therapeutic plans.
In the present study, APOE4 status was repeatedly associated with cerebral amyloid positivity in the three models for all subjects, MCI subjects, and AD dementia subjects. Inheritance of the ε4 allele of the APOE gene is the strongest genetic risk factor for AD. The interaction between APOE and Aβ plays a key role in AD pathogenesis. Previous studies revealed that the APOE-Aβ interaction regulates Aβ aggregation and clearance and therefore directly influences development of amyloid plaques, congophilic amyloid angiopathy, and the subsequent tau-related pathology [
This study has several limitations. First, although our study included patients with vascular dementia, Lewy body dementia, frontotemporal dementia, and other types of dementia, we could not develop models to predict cerebral amyloid positivity for each type of dementia due to the small number of subjects. Second, it is possible that clinicians have applied amyloid PET to patients who are likely to be cerebral amyloid-positive. Third, since the same neuropsychological battery such as SNSB or CERAD-K was not applied uniformly, we could not include it in the analysis except for MMSE and CDR-SB results. Fourth, two types of 18F-labeled amyloid-β targeting tracers were applied to the study subjects, but these were not considered when developing the models. Nevertheless, this is the first study to developed three cerebral amyloid prediction models according to the level of cognitive function by integrating clinically obtained demographic, hematologic, and neuropsychiatric tests data.
In conclusion, we developed cerebral amyloid prediction models with variables that are routinely collected in clinical practice for subjects complaining of cognitive decline. This study demonstrated that integration of demographic variables, neuropsychological measurements, and blood-based markers significantly improved prediction accuracy. Furthermore, our developed models have the potential to be widely used as screening tools, because we created cerebral amyloid prediction models with data obtained routinely in clinical settings. We hope that early prediction of cerebral amyloid positivity through data commonly used in a memory clinic, will be helpful for screening cognitively impaired patients with high probability of cerebral amyloid deposition.
The online-only Data Supplement is available with this article at
Youden index, sensitivity, specificity, PPV, and NPV in all subjects
Youden index, sensitivity, specificity, PPV, and NPV in MCI subjects
Youden index, sensitivity, specificity, PPV, and NPV in AD subjects
The datasets generated or analyzed during the study are available from the corresponding author on reasonable request.
The authors have no potential conflicts of interest to disclose.
Conceptualization: Chang Uk Lee. Data curation: Soo Hyun Joo. Formal analysis: Chang Uk Lee. Investigation: Soo Hyun Joo. Methodology: Soo Hyun Joo. Resources: Soo Hyun Joo. Software: Soo Hyun Joo. Supervision: Chang Uk Lee. Validation: Chang Uk Lee. Visualization: Soo Hyun Joo. Writing—original draft: Soo Hyun Joo. Writing—review & editing: Chang Uk Lee, Soo Hyun Joo.
None.
Receiver operating characteristic curves for the final and the three prediction models for cerebral amyloid positivity. (A) All subjects, (B) mild cognitive impairment subjects, and (C) Alzheimer’s disease dementia subjects.
Clinical characteristics of all subjects according to results of amyloid PET
Amyloid PET-CT results |
p-value | ||
---|---|---|---|
Negative | Positive | ||
Number | 188 | 222 | |
Age, years | 74.48±8.54 | 76.01±8.64 | 0.073 |
Female, N (%) | 115 (61.17) | 156 (70.27) | 0.052 |
BMI, kg/m2 | 23.73±3.31 | 23.16±3.14 | 0.079 |
Education, years | 11.70±4.81 | 10.91±5.10 | 0.105 |
APOE4-positive, N (%) | 32 (17.02) | 119 (53.60) | <0.001 |
Family history of dementia, N (%) | 46 (25.41) | 64 (30.77) | 0.242 |
Hypertension, N (%) | 92 (48.94) | 102 (45.95) | 0.546 |
Diabetes mellitus, N (%) | 59 (31.38) | 53 (23.87) | 0.089 |
Diagnosis, N (%) | <0.001 | ||
MCI | 160 (85.11) | 137 (61.71) | |
AD dementia | 28 (14.89) | 85 (38.29) | |
MMSE score | 24.69±3.63 | 21.87±5.54 | <0.001 |
CDR-SB score | 2.5±2.32 | 4.09±3.64 | <0.001 |
Systolic blood pressure, mm Hg | 126.35±17.34 | 128.77±16.75 | 0.191 |
Diastolic blood pressure, mm Hg | 72.54±10.56 | 73.40±11.25 | 0.469 |
TSH, mIU/L | 2.75±7.60 | 2.64±2.89 | 0.859 |
Total cholesterol, mg/dL | 178.88±41.05 | 186.04±40.54 | 0.111 |
Vitamin B12, pg/mL | 823.89±768.07 | 751.12±552.97 | 0.344 |
PET, positron emission tomography; BMI, body mass index; APOE, apolipoprotein ε; MCI, mild cognitive impairment; AD, Alzheimer’s disease, MMSE, mini-mental state examination; CDR-SB, clinical dementia rating sum of box; TSH, thyroid stimulating hormone
The final and simpler models predicting cerebral amyloid positivity in all subjects
Variable | Beta | Wald | OR (95% CI) | p-value |
---|---|---|---|---|
Final model | ||||
Age | 0.029 | 4.374 | 1.030 (1.002, 1.058) | 0.037 |
Sex | 0.247 | 0.953 | 1.280 (0.780, 2.100) | 0.329 |
Education | 0.034 | 1.653 | 1.034 (0.983, 1.088) | 0.199 |
APOE4 | 1.937 | 55.879 | 6.939 (4.176, 11.532) | <0.0001 |
MMSE | -0.155 | 27.068 | 0.857 (0.808, 0.908) | <0.0001 |
Intercept | 0.158 | |||
R-square |
0.302 | |||
APOE4-only model | ||||
Age | 0.038 | 8.213 | 1.038 (1.012, 1.066) | 0.004 |
Sex | 0.203 | 0.694 | 1.225 (0.760, 1.974) | 0.405 |
Education | -0.010 | 0.176 | 0.990 (0.946, 1.037) | 0.675 |
APOE4 | 1.826 | 54.903 | 6.208 (3.830, 10.062) | <0.0001 |
Intercept | -3.517 | |||
R-square |
0.215 | |||
MMSE-only model | ||||
Age | 0.009 | 0.564 | 1.009 (0.985, 1.034) | 0.453 |
Sex | 0.354 | 2.344 | 1.425 (0.905, 2.243) | 0.126 |
Education | 0.021 | 0.776 | 1.021 (0.975, 1.070) | 0.378 |
MMSE | -0.138 | 25.617 | 0.871 (0.826, 0.919) | <0.0001 |
Intercept | 1.883 | 2.089 | 0.148 | |
R-square |
0.122 |
Nagelkerke R-square.
OR, odds ratio; CI, confidence interval; APOE, apolipoprotein ε; MMSE, mini-mental state examination
The final model predicting cerebral amyloid positivity in MCI subjects
Final model | Beta | Wald | OR (95% CI) | p-value |
---|---|---|---|---|
Age | 0.037 | 5.679 | 1.038 (1.007, 1.071) | 0.017 |
Sex | -0.016 | 0.003 | 0.985 (0.565, 1.714) | 0.956 |
Education | 0.004 | 0.018 | 1.004 (0.949, 1.062) | 0.893 |
APOE4 | 1.915 | 46.405 | 6.784 (3.911, 11.768) | <0.0001 |
Intercept | -3.668 | |||
R-square |
0.227 |
Nagelkerke R-square.
MCI, mild cognitive impairment; OR, odds ratio; CI, confidence interval; APOE, apolipoprotein ε
The final and simpler models predicting cerebral amyloid positivity in AD dementia subjects
Variable | Beta | Wald | OR (95% CI) | p-value |
---|---|---|---|---|
Final model | ||||
Age | 0.039 | 1.016 | 1.040 (0.964, 1.122) | 0.314 |
Sex | 1.043 | 2.680 | 2.838 (0.814, 9.896) | 0.102 |
Education | 0.075 | 1.724 | 1.077 (0.964, 1.204) | 0.189 |
APOE4 | 2.631 | 9.339 | 13.890 (2.569, 75.089) | 0.002 |
MMSE | -0.198 | 8.795 | 0.820 (0.720, 0.935) | 0.003 |
Hypertension | -1.615 | 6.443 | 0.199 (0.057, 0.692) | 0.011 |
Intercept | -0.012 | |||
R-square |
0.429 | |||
APOE4-only model | ||||
Age | -0.002 | 0.003 | 0.998 (0.939, 1.061) | 0.958 |
Sex | 1.213 | 4.271 | 3.365 (1.065, 10.634) | 0.039 |
Education | 0.050 | 0.943 | 1.051 (0.950, 1.164) | 0.332 |
APOE4 | 2.453 | 9.888 | 11.619 (2.519, 53.589) | 0.002 |
Intercept | -1.831 | |||
R-square |
0.259 | |||
MMSE-only model | ||||
Age | -0.008 | 0.065 | 0.992 (0.932, 1.056) | 0.799 |
Sex | 1.061 | 3.851 | 2.890 (1.001, 8.342) | 0.050 |
Education | 0.048 | 0.865 | 1.049 (0.948, 1.162) | 0.352 |
MMSE | -0.125 | 6.066 | 0.882 (0.798, 0.975) | 0.014 |
Intercept | 1.955 | |||
R-square |
0.160 | |||
Hypertension-only model | ||||
Age | 0.006 | 0.036 | 1.006 (0.942, 1.075) | 0.849 |
Sex | 0.987 | 3.105 | 2.684 (0.895, 8.049) | 0.078 |
Education | 0.017 | 0.115 | 1.017 (0.924, 1.119) | 0.735 |
Hypertension | -1.579 | 7.940 | 0.206 (0.069, 0.618) | 0.005 |
Intercept | -0.143 | |||
R-square |
0.185 |
Nagelkerke R-square.
AD, Alzheimer’s disease; OR, odds ratio; CI, confidence interval; APOE, apolipoprotein ε; MMSE, minimental state examination
The AUC values of ROC curves for the three prediction models for cerebral amyloid positivity
Model | AUC | Lower | Upper |
---|---|---|---|
All subjects | |||
Final model | 0.775 | 0.731 | 0.820 |
APOE4-only model | 0.730 | 0.681 | 0.779 |
MMSE-only model | 0.663 | 0.611 | 0.715 |
MCI subjects | |||
Final model | 0.735 | 0.676 | 0.793 |
AD subjects | |||
Final model | 0.845 | 0.765 | 0.925 |
APOE4-only model | 0.747 | 0.652 | 0.842 |
MMSE-only model | 0.710 | 0.603 | 0.816 |
Hypertension-only model | 0.724 | 0.615 | 0.832 |
AUC, area under the curve; ROC, receiver operating characteristic; APOE, apolipoprotein ε; MMSE, mini-mental state examination; MCI, mild cognitive impairment; AD, Alzheimer’s disease