Skip to main content
Download PDF

Machine learning models for dementia screening to classify brain amyloid positivity on positron emission tomography using blood markers and demographic characteristics: a retrospective observational study

Alzheimer's Research & Therapy volume 17, Article number: 25 (2025) Cite this article

Abstract

Background

Intracerebral amyloid β (Aβ) accumulation is considered the initial observable event in the pathological process of Alzheimer’s disease (AD). Efficient screening for amyloid pathology is critical for identifying patients for early treatment. This study developed machine learning models to classify positron emission tomography (PET) Aβ-positivity in participants with preclinical and prodromal AD using data accessible to primary care physicians.

Methods

This retrospective observational study assessed the classification performance of combinations of demographic characteristics, routine blood test results, and cognitive test scores to classify PET Aβ-positivity using machine learning. Participants with mild cognitive impairment (MCI) or normal cognitive function who visited Oita University Hospital or had participated in the USUKI study and met the study eligibility criteria were included. The primary endpoint was assessment of the classification performance of the presence or absence of intracerebral Aβ accumulation using five machine learning models (i.e., five combinations of variables), each constructed with three classification algorithms, resulting in a total of 15 patterns. L2-regularized logistic regression, and kernel Support Vector Machine (SVM) and Elastic Net algorithms were used to construct the classification models using 34 pre-selected variables (12 demographic characteristics, 11 blood test results, 11 cognitive test results).

Results

Data from 262 records (260 unique participants) were analyzed. The mean (standard deviation [SD]) participant age was 73.8 (7.8) years. Using L2-regularized logistic regression, the mean receiver operating characteristic (ROC) area under the curve (AUC) (SD) in Model 0 (basic demographic characteristics) was 0.67 (0.01). Classification performance was similar in Model 1 (basic demographic characteristics and Mini Mental State Examination [MMSE] subscores) and Model 2 (demographic characteristics and blood test results) with a cross-validated mean ROC AUC (SD) of 0.70 (0.01) for both. Model 3 (demographic characteristics, blood test results, MMSE subscores) and Model 4 (Model 3 and ApoE4 phenotype) showed improved performance with a mean ROC AUC (SD) of 0.73 (0.01) and 0.76 (0.01), respectively. In models using blood test results, thyroid-stimulating hormone and mean corpuscular volume tended to be the largest contributors to classification. Classification performances were similar using the SVM and Elastic Net algorithms.

Conclusions

The machine learning models used in this study were useful for classifying PET Aβ-positivity using data from routine physician visits.

Trial registration

UMIN Clinical Trials Registry (UMIN000051776, registered on 31/08/2023).

Background

Dementia is a growing public health concern given increases in life expectancy and population aging, and has serious social and economic effects on both patients and their caregivers [1]. In people over 65 years of age, Alzheimer’s disease (AD) is a major cause of dementia [1]. The accumulation of amyloid β (Aβ) in the brain, a key characteristic of AD, is widely considered the initial observable event in the AD pathological process [2]. Other neuropathologic hallmarks of AD include neurofibrillary tangles and neuronal degeneration [2]. Biomarker research has focused on tracking the progression of AD through the accumulation of Aβ and tau in the brain, which are associated with neurodegeneration and cognitive decline [3, 4].

Recently, three anti-Aβ antibodies, aducanumab, lecanemab, and donanemab, were approved by the Food and Drug Administration for the treatment of patients with mild cognitive impairment (MCI) or mild dementia due to AD (note that aducanumab is not currently on the market) [5,6,7]. According to the package inserts [7,8,9], confirmation of the presence of Aβ pathology is required prior to treatment initiation. Therefore, accurate detection of amyloid pathology in individuals with MCI is essential to maximize the benefit of these anti-Aβ antibody therapies in future clinical practice [10].

11C-Pittsburgh Compound-B (PiB) positron emission tomography (PET) and measurement of cerebrospinal fluid (CSF) Aβ levels are established amyloid biomarkers that can predict the incidence of AD before the onset of dementia [11]. Aβ was first isolated and quantified from CSF in 1992 [12]. In 2004, Klunk et al. first demonstrated PiB-PET (amyloid PET), which has been widely used in AD research for over 20 years [13]. However, both techniques are unsuitable as screening tools in clinical settings owing to the high cost, radiation exposure, and the need for invasive lumbar puncture. Considering the low prevalence of Aβ-positive MCI, which has been shown to range from 22.0% in a population-based study to 46.6% in a multicenter research study, performing PET screening on all patients with suspected AD may translate to unnecessary increased costs and patient screening burden [14, 15].

To improve early identification of patients who may benefit from AD treatments, it is important to develop a simple and efficient method for screening that uses universally available data such as demographic characteristics, routinely collected blood data, and cognitive scores [16, 17]. Blood-based biomarkers (e.g., p-tau217, p-tau181, or Aβ42/40) have shown promise in their ability to predict Aβ pathology [18,19,20,21]; however, advanced technologies such as mass spectrometry and immunoassays are needed to achieve highly precise and reproducible plasma Aβ measurements [20, 22, 23]. Previous studies have shown a possible association between the presence of Aβ and variables measured in routine blood testing [24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43]. Additionally, age is one of the most important risk factors for developing AD [44]. The development of a predictive model to detect amyloid pathology using blood data collected as part of routine medical practice would provide an important new tool that could be implemented worldwide.

In recent years, there has been a focus on using machine learning to build models that are able to predict PET Aβ-positivity, MCI, and risk of future PET Aβ-positivity [10, 45,46,47,48]. Although some progress has been made in developing predictive models for Aβ pathology and early detection of AD, studies using a combination of data collected in regular clinical practice (e.g., demographic characteristics, lifestyle factors, and blood test results) to make predictions/classifications are limited. In the present study, we used data obtainable from medical health practice and physician visits, including blood tests and cognitive test scores, to detect individuals with elevated intracerebral Aβ accumulation. To the best of our knowledge, few studies have focused on routine blood data as variables for identifying individuals at high risk for intracerebral Aβ accumulation.

This study aimed to develop machine learning models to classify PET Aβ-positivity in patients with preclinical AD, defined as a state of clinically normal cognitive function with the presence of pathologic changes of AD [49], and prodromal AD from patient demographic characteristics and routinely collected blood data accessible to family physicians.

Methods

Study design

This was a retrospective observational study to assess the classification performance of combinations of variables (demographic characteristics, blood test results, and cognitive test scores) that may be useful for screening Aβ accumulation in the brains of people with MCI or normal cognitive function using machine learning. Prior to initiation, the study protocol was approved by the local ethics committee of Oita University Hospital (approval number 2574; approval date, June 22, 2023). This study used existing patient information, and it was difficult to obtain informed consent. Thus, an information disclosure document that included information on the conduct and purpose of the study was made publicly available on the university website [50]. This document was approved by the ethics committee and authorized by the head of Oita University Hospital. The study was registered with the University hospital Medical Information Network Clinical Trials Registry (UMIN000051776 on August 31, 2023).

Participants

This study used two data sources. One data source was patients who visited Oita University Hospital (September 1, 2012, to November 30, 2017, [5 years, 3 months]), and the other was elderly residents in Usuki City, Oita Prefecture, Japan, who participated in the USUKI study (USUKI study cohort; October 1, 2015, to November 30, 2017, [2 years, 2 months]). Participants with MCI or normal cognitive function and who met the inclusion/exclusion criteria of the present study were eligible for this analysis. The USUKI study was a prospective study of community-dwelling older adults without dementia designed to explore the risk and protective lifestyle factors of cognitive decline later in life. The study design and details of the methods have been published previously [51, 52]. The inclusion criteria for the present study were (1) a clinical dementia rating (CDR) of 0 or 0.5; and (2) availability of Aβ PiB-PET data. Individuals who met the following criteria were excluded: (1) insufficient data for analysis, (2) an indication of refusal to participate in the study, or (3) deemed ineligible for participation by the principal investigator or subinvestigators (e.g., data with many missing or abnormal values or data for suspected dementia). For criterion 3), data were reviewed prior to analysis, and depending on the situation, participants with a large number of missing or abnormal values were excluded. Abnormal values were those that, as judged from a clinical perspective by an expert, could not be obtained because of measurement errors or other factors. Blood test results that exceeded the standard value were not considered abnormal if they were clinically relevant. Among people who met the eligibility criteria, those whose available data satisfied the following were included in the analysis set: (1) all measurements for the variables included in the classification analysis (demographic characteristics, blood tests, MMSE, and CDR) were collected ≤ 365 days from the outcome measurement (PiB-PET scan) date, and (2) no missing values in the study blood test results.

Study endpoint

The primary endpoint was the classification performance of the presence or absence of intracerebral Aβ accumulation using five different machine learning models. The true label for positive Aβ accumulation, or the positive presence of intracerebral Aβ accumulation, was defined as a mean PiB standardized uptake value ratio (SUVR) of ≥ 1.2.

Model construction and evaluation

The methods used for model construction were similar to those previously described by Kimura et al. [10]. Positive and negative Aβ accumulation in the brain was used as the objective variable. The classification model was developed using data that had been imputed for missing values. Missing values for continuous variables were imputed using the median value, which was calculated for the entire dataset. Missing values for categorical variables were imputed using the mode value (i.e., the value that appeared most frequently in the dataset). As a sensitivity analysis to the missing value completion method, the model was also built and evaluated using a dataset in which missing value completion was performed using the K-nearest neighbors (KNN) method (number of neighbors was set at 5). The computation of these completion values was performed using the entire dataset (i.e., calculated for the entire dataset). Three algorithms previously found by Kimura et al. to be effective in predicting Aβ accumulation were selected as the basis for constructing the models: L2-regularized logistic regression, kernel Support Vector Machine (SVM), and Elastic Net [10]. The radial basis function kernel was used as the kernel function in the SVM algorithm in the present study. Thirty-four variables were pre-selected for use in the classification modeling based on existing literature [24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43], which included 12 demographic characteristics, 11 blood test results, and 11 cognitive test results. The 12 demographic characteristics comprised age, sex, years of education, body mass index (BMI), current alcohol consumption, smoking status (past or present), and six medical history variables (hypertension, hyperlipidemia, heart disease [e.g., heart failure, arrhythmia], stroke, diabetes mellitus, and thyroid disease). The 11 blood test results were apolipoprotein E4 (ApoE4) phenotyping (E4+/−), liver function markers (aspartate aminotransferase [AST] to alanine transaminase [ALT] ratio, alkaline phosphatase [ALP], gamma-glutamyl transferase [γ-GTP], total bilirubin), renal function markers (estimated glomerular filtration rate [eGFR]), thyroid-stimulating hormone (TSH), white blood cell (WBC) count, hemoglobin concentration, mean corpuscular volume (MCV), and platelet count. The 10 markers other than ApoE4 phenotyping were typically measured during routine physician visits. The cognitive test results included 11 Mini Mental State Examination (MMSE) subscores. Note that in this study, the MMSE total score was not used for classification, but rather the subscores were explored for their individual classification value.

For these pre-selected variables, continuous variables were transformed into quantile scales [53], and categorical variables were used as dummy variables using one-hot encoding. The conversion to quantile scales was used because it was previously found to be an effective method for improving prediction accuracy [10]. To avoid the overestimation of classification performance, the machine learning models were trained and validated using a 5-fold cross-validation scheme. For each round of validation, participants were randomly divided into five groups. Four groups of data were used as learning sets to train the models, and the classification performance of the trained model was evaluated using the remaining data set (validation set). This process was repeated for each of the groups (i.e., data from each of the five groups were used as the validation set). To reduce bias in the data split, 15 different split data values were created based on 15 seed values for which the 5-fold cross-validation process was performed. The analysis outputs classification performance results for five folds for each of the 15 seeds, totaling 75 values, and these values are averaged for each seed to obtain 15 cross-validated mean values. This method was applied across five models and three classification algorithms. Finally, the mean (standard deviation [SD]) of the classification performance output values (15 cross-validated mean values) was calculated. Classification performance was evaluated using receiver operating characteristic (ROC) area under the curve (AUC) values, sensitivity, specificity, positive predictive value, negative predictive value, and accuracy. The Youden index calculated from learning sets was used to determine the cut-off point for calculating these evaluation indices in 5-fold cross-validation. The Boruta method was used as the variable selection technique to eliminate variables that did not contribute to the classification performance of the model [54]. In the Boruta method, the contribution of each variable was calculated as permutation importance [55]. Variables were selected in the training models without the use of the validation set to avoid overfitting and biased evaluation of classification performance. Hyper parameter tuning was conducted by grid search for each learning set, in which Log-Loss was used as the objective function.

Model 0 used four participant demographic characteristic variables: age, sex, years of education, and BMI. These were the most easily obtained variables and were used as benchmarks in Model 0 to evaluate classification performance. Note that the Boruta method was not used in Model 0 because it included only four variables. Model 1 variables were those included in Model 0 plus MMSE subscores. Model 2 variables were those included in Model 0 plus blood test results (excluding ApoE4 phenotype) and the other demographic characteristics (medical history, current alcohol consumption, and smoking status). Model 3 variables were those included in Model 2 plus MMSE subscores. Model 4 variables were those included in Model 3 plus ApoE4 phenotype. ApoE4 phenotype is not measured in routine clinical practice; therefore, we note that classifications by Model 4 cannot be made using routine medical data alone. Using the classification models generated as described above, a diagram of the variable importance calculated based on Shapley Additive exPlanations (SHAP) and a SHAP summary plot [56, 57] were developed. LinearExplainer was used to calculate SHAP values. The variable importance was calculated as an average of the absolute SHAP values.

Statistical analysis

Considering study feasibility, the number of cases was set so that as many patients as possible could be enrolled. The target number of cases was set at 200, and the maximum number of cases was pre-planned to be 283.

The analysis of the primary endpoint was performed on the analysis set (262 records representing 260 unique participants). To assess the distribution balance of participant background characteristics between the Aβ positive and negative groups, Welch’s two-sample two-sided t-test was used for continuous variables and the chi-square test was used for categorical variables. R version 4.3.1 with the tableone package 0.13.2 (The R Foundation for Statistical Computing, Vienna, Austria) was used to calculate descriptive statistics for participant background characteristics. Python version 3.7.12 with the scikit-learn 1.0.2 and shap 0.41.0 packages (Python Software Foundation, Wilmington, DE, USA) was used for model construction and evaluation.

Results

Participants

Of the 855 patients in the USUKI cohort and the 230 patients who visited Oita University Hospital, 703 and 71, respectively, were excluded because they did not have amyloid PiB-PET data, and 8 and 13, respectively, because they had a CDR > 0.5; additionally, 1 case in the USUKI cohort was excluded because of an MMSE score < 20, indicating dementia, and 4 in Oita University Hospital because they did not have CDR data. No participants were excluded because of a large number of missing or abnormal values. The participant with the most missing values had 11 of 34 variables missing (11 MMSE subscores). Prior to analysis, it was determined that exclusion of this participant because of the high number of missing values was not necessary. Note that out of 285, two participants overlapped across the two cohorts, and therefore the dataset screened for eligibility included 283 unique participants. The overlapping records for the two participants were treated as separate records because the PiB-PET scan dates were approximately 4 years apart between the two cohorts. Overall, 285 records of participants with MCI or normal cognitive function were screened for eligibility and 262 (260 unique participants) were included in the analysis set; 22 were excluded because there was a > 365-day period between their PiB-PET and the collection of data related to other variables, and one patient in the USUKI cohort was excluded because of missing blood test records (Fig. 1).

Fig. 1
figure 1

Participant disposition

a262 records from a total of 260 unique participants CDR clinical dementia rating, MCI mild cognitive impairment, PiB-PET11C-Pittsburgh Compound-B positron emission tomography

Full size image

Participant background characteristics before imputing missing data are shown in Table 1. In total, 38.5% (101/262) of participant records were Aβ positive, 79.8% (209/262) had MCI (CDR = 0.5), and 20.2% (53/262) were cognitively normal (CDR = 0). The mean (SD) age was 73.8 (7.8) years, 51.9% (136/262) were male, and the mean (SD) MMSE score was 26.3 (2.4). Blood test results showed that 19.5% (50/262) of participant records were ApoE4 positive. Values for the AST to ALT ratio, total bilirubin, TSH, hemoglobin, and MCV were significantly different between the two groups (p < 0.05). Background characteristics after imputing missing data are shown in Table S1 (Additional file 1). There were a total of 12 records with missing values for any of the variables.

Table 1 Background characteristics of study participants
Full size table

Outcomes

Classification performance findings using the L2-regularized logistic regression are shown in Table 2; Fig. 2. The mean (SD) ROC AUC for Model 0, which used the demographic characteristics of age, sex, BMI, and years of education, was 0.67 (0.01). Performance improved with the use of the 11 MMSE subscores plus the four basic variables (age, sex, BMI, and years of education) of demographic characteristics (Model 1; mean [SD] ROC AUC, 0.70 [0.01]). Performance similarly improved when the demographic characteristics and blood test results (except for ApoE4 phenotype) were used (Model 2; mean [SD] ROC AUC, 0.70 [0.01]). An increase in performance was found when the 11 MMSE subscores were combined with the blood test results (except for ApoE4 phenotype) and demographic characteristics (Model 3; mean [SD] ROC AUC, 0.73 [0.01]) (Fig. 3). Finally, the best performance was observed when ApoE4 phenotype was added to the variables used in Model 3 (Model 4; mean [SD] ROC AUC, 0.76 [0.01]).

Table 2 Performance of the amyloid β positivity classification models using L2-regularized logistic regression
Full size table
Fig. 2
figure 2

Cross-validated area under the curves of amyloid β positivity classification models: L2-regularized logistic regression

Model 0: demographic characteristics (age, sex, body mass index, years of education) Model 1: Model 0 plus all MMSE subscores Model 2: Model 0 plus blood test results (excluding ApoE4 phenotype) and the other demographic characteristics (medical history, current alcohol consumption, and smoking status) Model 3: Model 2 plus all MMSE subscores Model 4: Model 3 plus ApoE4 phenotype Horizontal lines are median values, upper and lower box edges show the first quartile (Q1) and the third quartile (Q3), and upper and lower whiskers represent the 1st quartile − 1.5 × the IQR and 3rd quartile + 1.5 × the IQR, respectively. Circle point is an outlier beyond the whisker range ApoE4 apolipoprotein E4, AUC area under the curve, IQR interquartile range, MMSE Mini Mental State Examination

Full size image
Fig. 3
figure 3

Receiver operating characteristic curve of amyloid β positivity classification Model 3: L2-regularized logistic regression

AUC area under the curve, SD standard deviation

Full size image

The classification performance results when using the kernel SVM and Elastic Net algorithms were not different from those reported for the L2-regularized logistic regression algorithm (Table S2 in Additional file 1). The Elastic Net results mirrored those using the L2-regularized logistic regression, while the ROC AUC values using the kernel SVM tended to be lower, with the exception of those reported for Model 2 (mean [SD] ROC AUC, 0.72 [0.01]). The results when missing values were imputed by KNN were not notably different from the results when missing values were imputed by median value or mode value (Table S3 in Additional file 1). Figure 4 shows the variable importance and SHAP summary plot results for Model 3 trained by the L2-regularized logistic regression algorithm. MMSE subscores for delayed recall and orientation place, along with age, TSH, and MCV, were the five most important variables (Fig. 4A). The SHAP summary plot shows the direction and magnitude of the contribution of each variable to the classification (Fig. 4B). As an example, TSH contributes to the model in that the higher the value, the lower the positive probability, whereas with MCV, the higher the value, the higher the positive probability. Variable importance for Models 0–2 and 4 using the L2-regularized logistic regression algorithm is shown in Figure S1 (Additional file 1).

Fig. 4
figure 4

Variable importance (A) and Shapley Additive exPlanations summary plot (B): L2-regularized logistic regression Model 3

ALP alkaline phosphatase, BMI body mass index, eGFR estimated glomerular filtration rate, MCV mean corpuscular volume, MMSE Mini Mental State Examination, SHAP Shapley Additive exPlanations, TSH thyroid-stimulating hormone, WBC white blood cell count

Full size image

Considering the blood test variables, TSH and MCV tended to be the largest contributors to the classification across Models 2, 3, and 4. Trends for TSH and MCV were consistent across the models, which was not the case for the other variables. Hemoglobin, ALP, total bilirubin, AST to ALT ratio, eGFR, and WBC were selected and made moderate contributions to some models. Some demographic characteristic variables consistently showed large contributions, including age, BMI, and years of education in Models 2–4. The medical history variables of hyperlipidemia, diabetes, and heart disease were consistent contributors. Delayed recall, orientation place, and orientation time were consistently ranked as the top three contributors (in that order) for MMSE subscores (Models 1, 3, and 4). Language writing, language repetition, attention, and copy were also noted in Models 1, 3, and 4 but had a lower contribution than the top three contributors for MMSE subscores.

Discussion

This is the first study to develop a model using machine learning techniques to classify amyloid PET positivity using only the results of blood tests routinely performed in dementia diagnosis. Our results provide several novel insights into the social implementation of blood biomarkers for classifying amyloid PET positivity. First, the AUC of the machine learning model using demographic and general blood test data for classifying PET Aβ-positivity was 0.70. Second, the classification models included several variables associated with anemia, vascular risk factors, thyroid hormone, and liver and kidney function. Our results suggest that machine learning models using demographic and general blood test data can supplement amyloid PET and CSF analysis.

Many studies have reported on plasma biomarkers for AD, which can be measured with high accuracy and reproducibility using advanced technologies such as liquid chromatography mass spectrometry and immunoassay methods such as Single Molecule Array (SIMOA) assays. Such plasma biomarkers may include metabolic changes, neurofilament light, p-tau, glial fibrillary acidic protein, signaling proteins related to hematopoiesis, immune response, apoptosis, and neuroprotection, and others [58,59,60,61,62,63,64,65,66]. Although blood-based AD biomarkers have great potential in the diagnostic work-up of AD, they are not currently useful in routine clinical practice. Despite the numerous studies on various AD-related blood biomarkers in recent years, few have evaluated the classification value of a combination of participant demographic, lifestyle, and general blood test data in routine medical practice for classifying PET Aβ-positivity.

The model developed in the present study (Model 2) was able to classify the AD pathology of amyloid accumulation in the brain using data obtained during routine dementia care, including participant demographic and general blood test data. Model 2 had a classification performance of AUC = 0.70, which was comparable to the performance of Model 1 utilizing MMSE subscores (AUC = 0.70). Therefore, screening for Aβ-positive PET can be achieved using blood test data measured in routine clinical practice. Amyloid PET and CSF testing are expensive and highly invasive, thus limiting their usefulness in clinical practice. In search of more clinically useful options, previous studies have developed amyloid-positive prediction/classification models using cost-effective variables (e.g., demographic characteristics, ApoE genotype, imaging tests, and blood biomarkers) and machine learning that have achieved AUC values > 0.70. However, these models incorporate markers and other variables that are not measured in usual clinical practice, such as ApoE genotype, imaging data, chemokine ligand 13, pancreatic polypeptide, and vascular cell and adhesion protein-1 [18, 67,68,69,70]. We note that the AUC of these previously reported models is higher than that of Model 2 in the present study. However, our model, which uses combinations of readily available data, should be a suitable and successful screening tool for amyloid positivity in the primary care setting. When comparing Model 4 with previously reported models using ApoE4 data as a predictive variable, Model 4 had an AUC of 0.76, which demonstrated better performance than some (Petersen et al. AUC, 0.73 [69] and Kim et al. AUC, 0.71 [70]), but not all previously reported models (Palmqvist et al. AUC, 0.80–0.82 [67]). No studies were identified that developed models with substantially better performance than Model 2, which was constructed using only items measured in routine practice. It is important to note that plasma Aβ can be more accurately and reproducibly measured using mass spectrometry and immunoassays [22] than using a machine learning model. One review reported that, among studies using an immunoprecipitation-mass spectrometry assay of Aβ42/Aβ40 with a PET reference, the weighted average of AUC values across 21 cohorts was 0.834 [71]. However, there are limitations in the diagnostic value of the Aβ42/Aβ40 ratio without PET imaging due to overlap between Aβ-positive and -negative patients [72, 73]. Our machine learning model may be useful for addressing this, as it could be used for population-based screening to detect Aβ-positive individuals, which would reduce testing costs and the number of patients who undergo unnecessary amyloid PET imaging.

Our classification models included several variables associated with anemia (MCV and hemoglobin), vascular risk factors (lipid abnormalities, smoking, diabetes, BMI, and ischemic heart disease), and measures of thyroid (TSH), liver (ALP and bilirubin), and kidney (eGFR) function, all of which are associated with Aβ accumulation. In particular, variables used in our models that were shown to be important using L2-regularized logistic regression analysis include age, years of education, vascular risk factors, heart disease, liver and renal function, anemia, TSH, and MMSE score. The prevalence of Aβ-positivity with normal cognitive function generally increases with age. When considering educational level, a proxy of cognitive reserve, Aβ accumulation is more common in highly educated individuals than in less educated individuals [24]. There is growing evidence linking AD pathology to cerebrovascular and cardiovascular disease, with common risk factors including hypertension in middle age, obesity in middle age, diabetes, hypercholesterolemia, and smoking [25]. Furthermore, the two-hit vascular hypothesis of AD proposes that aging and vascular risk factors lead to dysregulation of the neurovascular unit and are associated with impaired Aβ clearance and increased Aβ production [26, 27]. Although the relationship between cardiovascular disease and AD pathogenesis remains unclear, several studies of autopsy cases suggest that coronary artery disease is associated with intracerebral Aβ accumulation [28]. Regarding the association of liver and renal function with AD, the liver and kidney physiologically remove Aβ from the blood and regulate Aβ levels in the brain; an age-related decline in hepatic Aβ clearance may be involved in the development of AD [29,30,31]; and eGFR is positively correlated with Aβ42 and Aβ40 and negatively correlated with total-tau levels in CSF [32,33,34]. An association between thyroid hormones and brain Aβ load has been reported both in vitro and in vivo [35]. In particular, triiodothyronine negatively regulates gene expression of amyloid precursor protein [36]. There are few reports on the association of AD with anemia-related characteristics such as MCV and hemoglobin. MCV has been suggested as a factor associated with cognitive function [37, 38] and previous reports have shown that AD causes changes in erythrocyte morphology. The mechanisms for this include relatively low folate and vitamin B12 concentrations in patients with AD [74], binding of Aβ to erythrocytes, which increases the permeability of the cell membrane and causes fluid accumulation [75,76,77], changes in the levels of protein components of the cytoskeleton and those linked to the erythrocyte membrane, and changes in the phospholipid composition of the erythrocyte membrane [78, 79]. Low blood hemoglobin (anemia) in older people is associated with cognitive decline via reduced cerebral metabolism, but this appears to be unrelated to aspects of AD-specific protein pathology or cerebrovascular disease as reflected in PET and magnetic resonance imaging measurements [39]. The inclusion of MCV and hemoglobin levels is a novel aspect of our models.

Limitations

This study had several limitations that should be noted. First, because this was a retrospective observational study, the models were cross-validated internally and have not been validated using external data. Additional studies using external data are needed to verify the classification performance of our models. Second, the model developed in this study is based on two datasets, but relies on data from a limited number of geographical locations and has not been validated using an independent dataset, so external validity and generalizability cannot be guaranteed. Third, because the study models were analyzed in a population that included both MCI and normal cognition, the performances of our developed models are not guaranteed outside of this population. It would be of future interest to investigate individuals with normal cognition with a focus on early detection of Aβ accumulation. Furthermore, the Aβ-positive rate in this study was 39% (101/262), which was lower than existing reports of Aβ-positive rates [80]. Fourth, the lack of data for T3 and T4 levels limits our ability to interpret the association between TSH and amyloid levels. Finally, the model presented in this study showed a moderate performance, so future studies are needed to identify other easily measurable variables (e.g., wearable sensor data [10]) that can be combined with it to further improve performance.

Conclusions

This study developed machine learning models that are useful for screening PET Aβ-positivity in people with MCI and preclinical AD, providing some novel and interesting insights into the utility of plasma biomarkers in classifying brain amyloid accumulation. Participant demographic characteristics and blood test data accessible to primary care physicians were successfully used to develop a machine learning model to classify PET Aβ-positivity, with a performance of AUC = 0.70. The model using this data combination was more accurate than a model based on basic patient demographic characteristics alone (age, sex, BMI, and years of education) and as accurate as a model based on basic patient demographic characteristics plus MMSE score. Furthermore, adding ApoE4, MMSE score, or both to the blood data slightly improved the classification performance. Unlike other screening models that require specialized tests, we anticipate that our model will be useful for Aβ screening by primary care physicians during routine medical practice because it is easy to use and classifies amyloid PET positivity using routine blood test, interview, and laboratory data. The use of this model as a screening tool is expected to reduce unnecessary PET scans and CSF testing and enable earlier detection and treatment of AD.

Data availability

The data used in this study cannot be shared publicly owing to ethical restrictions. The consent form signed by the participants states that their data are provided in an anonymized form only to research institutions. The raw data utilized in this study contain sensitive and personally identifiable information that could negatively affect the privacy of the study participants. However, the data supporting the results of this study are available upon ethical approval by the local ethics committee of Oita University Hospital. The ethics committee of Oita University Hospital can be contacted to obtain these data (rinrikenkyu@oita-u.ac.jp).

Abbreviations

γ-GTP:

gamma-glutamyl transferase

Aβ:

amyloid β

AA:

amino acid

AD:

Alzheimer’s disease

ALP:

alkaline phosphatase

ALT:

alanine transaminase

ApoE4:

apolipoprotein E4

AST:

aspartate aminotransferase

AUC:

area under the curve

BMI:

body mass index

CDR:

clinical dementia rating

CSF:

cerebrospinal fluid

eGFR:

estimated glomerular filtration rate

KNN:

K-nearest neighbors

MCI:

mild cognitive impairment

MCV:

mean corpuscular volume

MMSE:

Mini Mental State Examination

NC:

normal cognitive function

NPV:

negative predictive value

PET:

positron emission tomography

PiB:

11C-Pittsburgh Compound-B

PPV:

positive predictive value

ROC:

receiver operating characteristic

SD:

standard deviation

SHAP:

Shapley Additive exPlanations

SMD:

standardized mean difference

SUVR:

standardized uptake value ratio

SVM:

Support Vector Machine

TSH:

thyroid-stimulating hormone

WBC:

white blood cell

References

  1. 2022 Alzheimer’s disease facts and figures. Alzheimers Dement. 2022;18(4):700–89. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/alz.12638.

    Article  Google Scholar 

  2. Long JM, Holtzman DM. Alzheimer disease: an update on pathobiology and treatment strategies. Cell. 2019;179(2):312. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.cell.2019.09.001.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  3. Jack CR Jr, Bennett DA, Blennow K, Carrillo MC, Dunn B, Haeberlein SB, et al. NIA-AA research framework: toward a biological definition of Alzheimer’s disease. Alzheimers Dement. 2018;14(4):535–62. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.jalz.2018.02.018.

    Article  PubMed  Google Scholar 

  4. Scheltens P, De Strooper B, Kivipelto M, Holstege H, Chételat G, Teunissen CE, et al. Alzheimer’s disease. Lancet. 2021;397(10284):1577–90. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/S0140-6736(20)32205-4.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  5. Dhillon S. Aducanumab: First Approval. Drugs. 2021;81(12):1437-43. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s40265-021-01569-z. Erratum in: Drugs. 2021;81(14):1701. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s40265-021-01590-2.

  6. Hoy SM. Lecanemab: first approval. Drugs. 2023;83(4):359–65. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s40265-023-01851-2.

    Article  CAS  PubMed  Google Scholar 

  7. Eli Lilly and Company. KISUNLA (donanemab-azbt) prescribing information. 2024. Available from: https://www.accessdata.fda.gov/drugsatfda_docs/label/2024/761248s000lbl.pdf. Accessed 29 Aug 2024.

  8. Eisai Inc. LEQEMBI® (lecanemab-irmb) prescribing information. 2023. Available from: https://www.accessdata.fda.gov/drugsatfda_docs/label/2023/761269Orig1s001lbl.pdf. Accessed 29 Aug 2024.

  9. Biogen Inc. and Eisai Inc. ADUHELM® (aducanumab-avwa) injection prescribing information. 2022. Available from: https://www.accessdata.fda.gov/drugsatfda_docs/label/2022/761178s005lbl.pdf. Accessed 29 Aug 2024.

  10. Kimura N, Aota T, Aso Y, Yabuuchi K, Sasaki K, Masuda T, et al. Predicting Positron emission tomography brain amyloid positivity using interpretable machine learning models with wearable sensor data and lifestyle factors. Alzheimers Res Ther. 2023;15(1):212. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-023-01363-x.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  11. Young CB, Mormino EC. Prevalence rates of amyloid positivity–updates and relevance. JAMA Neurol. 2022;79(3):225–7. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/jamaneurol.2021.5225.

    Article  PubMed  Google Scholar 

  12. Seubert P, Vigo-Pelfrey C, Esch F, Lee M, Dovey H, Davis D, et al. Isolation and quantification of soluble Alzheimer’s beta-peptide from biological fluids. Nature. 1992;359(6393):325–7. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/359325a0.

    Article  CAS  PubMed  Google Scholar 

  13. Klunk WE, Engler H, Nordberg A, Wang Y, Blomqvist G, Holt DP, et al. Imaging brain amyloid in Alzheimer’s disease with Pittsburgh Compound-B. Ann Neurol. 2004;55(3):306–19. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/ana.20009.

    Article  CAS  PubMed  Google Scholar 

  14. Fleisher AS, Chen K, Liu X, Roontiva A, Thiyyagura P, Ayutyanont N, et al. Using positron emission tomography and florbetapir F18 to image cortical amyloid in patients with mild cognitive impairment or dementia due to Alzheimer disease. Arch Neurol. 2011;68(11):1404–11. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/archneurol.2011.150.

    Article  PubMed  Google Scholar 

  15. Roberts RO, Aakre JA, Kremers WK, Vassilaki M, Knopman DS, Mielke MM, et al. Prevalence and outcomes of amyloid positivity among persons without dementia in a longitudinal, population-based setting. JAMA Neurol. 2018;75(8):970–9. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/jamaneurol.2018.0629.

    Article  PubMed  PubMed Central  Google Scholar 

  16. Contador J, Vargas-Martínez AM, Sanchez-Valle R, Trapero-Bertran M, Lladó A. Cost-effectiveness of Alzheimer’s disease CSF biomarkers and amyloid-PET in early-onset cognitive impairment diagnosis. Eur Arch Psychiatry Clin Neurosci. 2023;273(1):243–52. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s00406-022-01439-z.

    Article  PubMed  Google Scholar 

  17. Wittenberg R, Knapp M, Karagiannidou M, Dickson J, Schott JM. Economic impacts of introducing diagnostics for mild cognitive impairment Alzheimer’s disease patients. Alzheimers Dement Transl Res Clin Interv. 2019;5:382–7. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.trci.2019.06.001.

    Article  Google Scholar 

  18. Burnham SC, Faux NG, Wilson W, Laws SM, Ames D, Bedo J, et al. Alzheimer’s Disease Neuroimaging Initiative; Australian imaging, biomarkers and Lifestyle Study Research Group. A blood-based predictor for neocortical Aβ burden in Alzheimer’s disease: results from the AIBL study. Mol Psychiatry. 2014;19(4):519–26. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/mp.2013.40.

    Article  CAS  PubMed  Google Scholar 

  19. Janelidze S, Berron D, Smith R, Strandberg O, Proctor NK, Dage JL, et al. Associations of plasma phospho-tau217 levels with tau positron emission tomography in early Alzheimer disease. JAMA Neurol. 2021;78(2):149–56. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/jamaneurol.2020.4201.

    Article  PubMed  Google Scholar 

  20. Janelidze S, Teunissen CE, Zetterberg H, Allué JA, Sarasa L, Eichenlaub U, et al. Head-to-head comparison of 8 plasma amyloid-β 42/40 assays in Alzheimer disease. JAMA Neurol. 2021;78(11):1375–82. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/jamaneurol.2021.3180. Erratum. In: JAMA Neurol. 2023;80(4):422.

    Article  PubMed  Google Scholar 

  21. Udeh-Momoh C, Zheng B, Sandebring-Matton A, Novak G, Kivipelto M, Jönsson L, et al. Blood derived amyloid biomarkers for Alzheimer’s disease prevention. J Prev Alzheimers Dis. 2022;9(1):12–21. https://doiorg.publicaciones.saludcastillayleon.es/10.14283/jpad.2021.70.

    Article  CAS  PubMed  Google Scholar 

  22. Benedet AL, Brum WS, Hansson O, Karikari TK, Zimmer ER, Zetterberg H, et al. Alzheimer’s disease neuroimaging initiative. The accuracy and robustness of plasma biomarker models for amyloid PET positivity. Alzheimers Res Ther. 2022;14(1):26. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-021-00942-0.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  23. Zicha S, Bateman RJ, Shaw LM, Zetterberg H, Bannon AW, Horton WA, et al; Alzheimer's Disease Neuroimaging Initiative (ADNI); Foundation for the National Institutes of Health (FNIH) Biomarkers Consortium Plasma Aβ as a Predictor of Amyloid Positivity in Alzheimer's Disease Project Team. Comparative analytical performance of multiple plasma Aβ42 and Aβ40 assays and their ability to predict positron emission tomography amyloid positivity. Alzheimers Dement. 2022:10.1002/alz.12697. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/alz.12697. Epub ahead of print.

  24. Jansen WJ, Ossenkoppele R, Knol DL, Tijms BM, Scheltens P, Verhey FR, Amyloid Biomarker Study Group, et al. Prevalence of cerebral amyloid pathology in persons without dementia: a meta-analysis. JAMA. 2015;313(19):1924–38. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/jama.2015.4668.

  25. Santos CY, Snyder PJ, Wu WC, Zhang M, Echeverria A, Alber J. Pathophysiologic relationship between Alzheimer’s disease, cerebrovascular disease, and cardiovascular risk: a review and synthesis. Alzheimers Dement (Amst). 2017;7:69–87. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.dadm.2017.01.005.

    Article  PubMed  Google Scholar 

  26. Zlokovic BV. Neurovascular pathways to neurodegeneration in Alzheimer’s disease and other disorders. Nat Rev Neurosci. 2011;12(12):723–38. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/nrn3114.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Soto-Rojas LO, Pacheco-Herrero M, Martínez-Gómez PA, Campa-Córdoba BB, Apátiga-Pérez R, Villegas-Rojas MM, et al. The neurovascular unit dysfunction in Alzheimer’s disease. Int J Mol Sci. 2021;22(4): 2022. https://doiorg.publicaciones.saludcastillayleon.es/10.3390/ijms22042022.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Reed BR, Marchant NL, Jagust WJ, Stocker H, Holleczek B, Beyreuther K, et al. Coronary risk correlates with cerebral amyloid deposition. Neurobiol Aging. 2012;33(9):1979–87. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.neurobiolaging.2011.10.002.

    Article  CAS  PubMed  Google Scholar 

  29. Cheng Y, He CY, Tian DY, Chen SH, Ren JR, Sun HL, et al. Physiological β-amyloid clearance by the liver and its therapeutic potential for Alzheimer’s disease. Acta Neuropathol. 2023;145(6):717–31. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s00401-023-02559-z.

    Article  CAS  PubMed  Google Scholar 

  30. Bassendine MF, Taylor-Robinson SD, Fertleman M, Khan M, Neely D. Is Alzheimer’s disease a liver disease of the brain? J Alzheimers Dis. 2020;75(1):1–14. https://doiorg.publicaciones.saludcastillayleon.es/10.3233/JAD-190848.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  31. Wang YR, Wang QH, Zhang T, Liu YH, Yao XQ, Zeng F, et al. Associations between hepatic functions and plasma amyloid-beta levels-implications for the capacity of liver in peripheral amyloid-beta clearance. Mol Neurobiol. 2017;54(3):2338–44. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s12035-016-9826-1.

    Article  CAS  PubMed  Google Scholar 

  32. Tian DY, Cheng Y, Zhuang ZQ, He CY, Pan QG, Tang MZ, et al. Physiological clearance of amyloid-beta by the kidney and its therapeutic potential for Alzheimer’s disease. Mol Psychiatry. 2021;26(10):6074–82. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/s41380-021-01073-6.

    Article  CAS  PubMed  Google Scholar 

  33. Sun HL, Yao XQ, Lei L, Jin WS, Bai YD, Zeng GH, et al. Associations of blood and cerebrospinal fluid Aβ and tau levels with renal function. Mol Neurobiol. 2023;60(9):5343–51. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s12035-023-03420-w.

    Article  CAS  PubMed  Google Scholar 

  34. Ren JR, Wang Z, Cheng Y, He CY, Jian JM, Fan DY, et al. Associations between plasma klotho with renal function and cerebrospinal fluid amyloid-β levels in Alzheimer’s disease: the Chongqing ageing & dementia study. J Alzheimers Dis. 2023;92(2):477–85. https://doiorg.publicaciones.saludcastillayleon.es/10.3233/JAD-221107.

    Article  CAS  PubMed  Google Scholar 

  35. Kim JH, Lee HS, Kim YH, Kwon MJ, Kim JH, Min CY, et al. The association between thyroid diseases and Alzheimer’s disease in a national health screening cohort in Korea. Front Endocrinol (Lausanne). 2022;13: 815063. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fendo.2022.815063.

    Article  PubMed  Google Scholar 

  36. Belandia B, Latasa MJ, Villa A, Pascual A. Thyroid hormone negatively regulates the transcriptional activity of the beta-amyloid precursor protein gene. J Biol Chem. 1998;273(46):30366–71. https://doiorg.publicaciones.saludcastillayleon.es/10.1074/jbc.273.46.30366.

    Article  CAS  PubMed  Google Scholar 

  37. Chen Y, Ma CN, Luo L, Yin J, Gao Z, Yu Z, et al. The cross-sectional association between mean corpuscular volume level and cognitive function in Chinese over 45 years old: evidence from the China Health and Retirement Longitudinal Study. PLoS ONE. 2020;15(12): e0243227. https://doiorg.publicaciones.saludcastillayleon.es/10.1371/journal.pone.0243227.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Qiang YX, Deng YT, Zhang YR, Wang HF, Zhang W, Dong Q, et al. Associations of blood cell indices and anemia with risk of incident dementia: a prospective cohort study of 313,448 participants. Alzheimers Dement. 2023;19(9):3965–76. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/alz.13088.

    Article  PubMed  Google Scholar 

  39. Kim JW, Byun MS, Yi D, Lee JH, Jeon SY, Ko K, et al. Blood hemoglobin, in-vivo Alzheimer pathologies, and cognitive impairment: a cross-sectional study. Front Aging Neurosci. 2021;13: 625511. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fnagi.2021.625511.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Livingston G, Sommerlad A, Orgeta V, Costafreda SG, Huntley J, Ames D, et al. Dementia prevention, intervention, and care. Lancet. 2017;390(10113):2673–734. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/S0140-6736(17)31363-6.

    Article  PubMed  Google Scholar 

  41. Livingston G, Huntley J, Sommerlad A, Ames D, Ballard C, Banerjee S, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet. 2020;396(10248):413–46. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/S0140-6736(20)30367-6 Erratum in: Lancet. 2023;402(10408):1132. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/S0140-6736(23)02043-3

    Article  PubMed  PubMed Central  Google Scholar 

  42. Shah RC, Buchman AS, Wilson RS, Leurgans SE, Bennett DA. Hemoglobin level in older persons and incident Alzheimer disease: prospective cohort analysis. Neurology. 2011;77(3):219–26. https://doiorg.publicaciones.saludcastillayleon.es/10.1212/WNL.0b013e318225aaa9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  43. Huang LT, Zhang CP, Wang YB, Wang JH. Association of peripheral blood cell profile with Alzheimer’s disease: a meta-analysis. Front Aging Neurosci. 2022;14: 888946. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fnagi.2022.888946.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  44. Riedel BC, Thompson PM, Brinton RD, Age. APOE and sex: triad of risk of Alzheimer’s disease. J Steroid Biochem Mol Biol. 2016;160:134–47. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.jsbmb.2016.03.012.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  45. Moradi E, Prakash M, Hall A, Solomon A, Strange B, Tohka J. Alzheimer’s disease neuroimaging initiative. Machine learning prediction of future amyloid beta positivity in amyloid-negative individuals. Alzheimers Res Ther. 2024;16(1):46. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-024-01415-w.

    Article  PubMed  PubMed Central  Google Scholar 

  46. Tan WY, Hargreaves C, Chen C, Hilal S. A machine learning approach for early diagnosis of cognitive impairment using population-based data. J Alzheimers Dis. 2023;91(1):449–61. https://doiorg.publicaciones.saludcastillayleon.es/10.3233/JAD-220776.

    Article  PubMed  PubMed Central  Google Scholar 

  47. Diao Y, Lanz B, Jelescu IO. Subject classification and cross-time prediction based on functional connectivity and white matter microstructure features in a rat model of Alzheimer’s using machine learning. Alzheimers Res Ther. 2023;15(1):193. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-023-01328-0.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Momota Y, Bun S, Hirano J, Kamiya K, Ueda R, Iwabuchi Y, et al. Amyloid-β prediction machine learning model using source-based morphometry across neurocognitive disorders. Sci Rep. 2024;14(1):7633. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/s41598-024-58223-3.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, Fagan AM, et al. Toward defining the preclinical stages of Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):280–92. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.jalz.2011.03.003.

    Article  PubMed  Google Scholar 

  50. Oita University. Department of Neurology. Faculty of Medicine. FY2023 3rd Medical School Ethics Committee (Rapid review). Report of deliberation results and summary of proceedings. [In Japanese] https://www.med.oita-u.ac.jp/joho/rinri/giji_gaiyo/rinri/2023/jin-no03.pdf. Accessed 6 July 2024.

  51. Kimura N, Aso Y, Yabuuchi K, Ishibashi M, Hori D, Sasaki Y, et al. Modifiable lifestyle factors and cognitive function in older people: a cross-sectional observational study. Front Neurol. 2019;10: 401. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fneur.2019.00401.

    Article  PubMed  PubMed Central  Google Scholar 

  52. Kimura N, Aso Y, Yabuuchi K, Ishibashi M, Hori D, Sasaki Y, et al. Association of modifiable lifestyle factors with cortical amyloid burden and cerebral glucose metabolism in older adults with mild cognitive impairment. JAMA Netw Open. 2020;3(6):e205719. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/jamanetworkopen.2020.5719.

    Article  PubMed  PubMed Central  Google Scholar 

  53. Scikit-learn. QuantileTransformer. 2023. https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.QuantileTransformer.html. Accessed 6 Jul 2024.

  54. Kursa MB, Rudnicki WR. Feature selection with the Boruta package. J Stat Softw. 2010;36(11):1–13. https://doiorg.publicaciones.saludcastillayleon.es/10.18637/jss.v036.i11.

    Article  Google Scholar 

  55. Fisher A, Rudin C, Dominici F. All models are wrong, but many are useful: learning a variable’s importance by studying an entire class of prediction models simultaneously. J Mach Learn Res. 2019;20:177.

    PubMed  PubMed Central  Google Scholar 

  56. Lundberg SM, Lee SI. A unified approach to interpreting model predictions. In: Advances in neural information processing systems. 2017. https://papers.nips.cc/paper_files/paper/2017/hash/8a20a8621978632d76c43dfd28b67767-Abstract.html. Accessed 6 Jul 2024.

  57. Molnar C. Interpretable machine learning: a guide for making black box models explainable, 2nd ed. 2022. https://christophm.github.io/interpretable-ml-book/. Accessed 6 Jul 2024.

  58. Mendes AJ, Ribaldi F, Lathuiliere A, Ashton NJ, Zetterberg H, Abramowicz M. Comparison of plasma and neuroimaging biomarkers to predict cognitive decline in non-demented memory clinic patients. Alzheimers Res Ther. 2024;16(1):110. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-024-01478-9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  59. Alcolea D, Beeri MS, Rojas JC, Gardner RC, Lleó A. Blood biomarkers in neurodegenerative diseases: implications for the clinical neurologist. Neurology. 2023;101(4):172–80. https://doiorg.publicaciones.saludcastillayleon.es/10.1212/WNL.0000000000207193.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  60. Koivumäki M, Ekblad L, Lantero-Rodriguez J. Blood biomarkers of neurodegeneration associate differently with amyloid deposition, medial temporal atrophy, and cerebrovascular changes in APOE ε4-enriched cognitively unimpaired elderly. Alzheimers Res Ther. 2024;16(1):112. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-024-01477-w.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  61. Bun S, Ito D, Tezuka T, Kubota M, Ueda R, Takahata K, et al. Performance of plasma Aβ42/40, measured using a fully automated immunoassay, across a broad patient population in identifying amyloid status. Alzheimers Res Ther. 2023;15(1):149. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-023-01296-5.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  62. Ray S, Britschgi M, Herbert C, Takeda-Uchimura Y, Boxer A, Blennow K, et al. Classification and prediction of clinical Alzheimer’s diagnosis based on plasma signaling proteins. Nat Med. 2007;13(11):1359–62. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/nm1653.

    Article  CAS  PubMed  Google Scholar 

  63. Angelopoulou E, Paudel YN, Shaikh MF, Piperi C, Flotillin. A promising biomarker for Alzheimer’s disease. J Pers Med. 2020;10(2):20. https://doiorg.publicaciones.saludcastillayleon.es/10.3390/jpm10020020.

    Article  PubMed  PubMed Central  Google Scholar 

  64. Abdullah M, Kimura N, Akatsu H, Hashizume Y, Ferdous T, Tachita T, et al. Flotillin is a novel diagnostic blood marker of Alzheimer’s disease. J Alzheimers Dis. 2019;72(4):1165–76. https://doiorg.publicaciones.saludcastillayleon.es/10.3233/JAD-190908.

    Article  CAS  PubMed  Google Scholar 

  65. Abe K, Shang J, Shi X, Yamashita T, Hishikawa N, Takemoto M, et al. A new serum biomarker set to detect mild cognitive impairment and Alzheimer’s disease by peptidome technology. J Alzheimers Dis. 2020;73(1):217–27. https://doiorg.publicaciones.saludcastillayleon.es/10.3233/JAD-191016.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  66. Bayoumy S, Verberk IMW, den Dulk B, Hussainali Z, Zwan M, van der Flier WM, et al. Clinical and analytical comparison of six simoa assays for plasma P-tau isoforms P-tau181, P-tau217, and P-tau231. Alzheimers Res Ther. 2021;13(1):198. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-021-00939-9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  67. Palmqvist S, Insel PS, Zetterberg H, Blennow K, Brix B, Stomrud E. Alzheimer’s Disease Neuroimaging Initiative; Swedish BioFINDER study. Accurate risk estimation of β-amyloid positivity to identify prodromal Alzheimer’s disease: cross-validation study of practical algorithms. Alzheimers Dement. 2019;15(2):194–204. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.jalz.2018.08.014.

    Article  PubMed  Google Scholar 

  68. Ezzati A, Harvey DJ, Habeck C, Golzar A, Qureshi IA, Zammit AR, et al. Alzheimer’s Disease Neuroimaging Initiative. Predicting amyloid-β levels in amnestic mild cognitive impairment using machine learning techniques. J Alzheimers Dis. 2020;73(3):1211–9. https://doiorg.publicaciones.saludcastillayleon.es/10.3233/JAD-191038.

    Article  PubMed  PubMed Central  Google Scholar 

  69. Petersen KK, Lipton RB, Grober E, Davatzikos C, Sperling RA, Ezzati A. Predicting amyloid positivity in cognitively unimpaired older adults: a machine learning approach using A4 data. Neurology. 2022;98(24):e2425-2435. https://doiorg.publicaciones.saludcastillayleon.es/10.1212/WNL.0000000000200553.

    Article  PubMed  PubMed Central  Google Scholar 

  70. Kim JP, Kim J, Jang H, Kim J, Kang SH, Kim JS, et al. Predicting amyloid positivity in patients with mild cognitive impairment using a radiomics approach. Sci Rep. 2021;11(1):6954. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/s41598-021-86114-4.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  71. Brand AL, Lawler PE, Bollinger JG, Li Y, Schindler SE, Li M, et al. The performance of plasma amyloid beta measurements in identifying amyloid plaques in Alzheimer’s disease: a literature review. Alzheimers Res Ther. 2022;14(1):195. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-022-01117-1.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  72. Verberk IMW, Slot RE, Verfaillie SCJ, Heijst H, Prins ND, van Berckel BNM, et al. Plasma amyloid as prescreener for the earliest alzheimer pathological changes. Ann Neurol. 2018;84(5):648–58. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/ana.25334.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  73. Schindler SE, Bollinger JG, Ovod V, Mawuenyega KG, Li Y, Gordon BA, et al. High-precision plasma β-amyloid 42/40 predicts current and future brain amyloidosis. Neurology. 2019;93(17):e1647-1649. https://doiorg.publicaciones.saludcastillayleon.es/10.1212/WNL.0000000000008081.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  74. Chan LN, Mike LA. The science and practice of micronutrient supplementations in nutritional anemia: an evidence-based review. JPEN J Parenter Enter Nutr. 2014;38(6):656–72. https://doiorg.publicaciones.saludcastillayleon.es/10.1177/0148607114533726.

    Article  CAS  Google Scholar 

  75. Lan J, Liu J, Zhao Z, Xue R, Zhang N, Zhang P, et al. The peripheral blood of Aβ binding RBC as a biomarker for diagnosis of Alzheimer’s disease. Age Ageing. 2015;44(3):458–64. https://doiorg.publicaciones.saludcastillayleon.es/10.1093/ageing/afv009.

    Article  PubMed  Google Scholar 

  76. Engström I, Ronquist G, Pettersson L, Waldenström A. Alzheimer amyloid beta-peptides exhibit ionophore-like properties in human erythrocytes. Eur J Clin Invest. 1995;25(7):471–6. https://doiorg.publicaciones.saludcastillayleon.es/10.1111/j.1365-2362.1995.tb01732.x.

    Article  PubMed  Google Scholar 

  77. Rogers J, Li R, Mastroeni D, Grover A, Leonard B, Ahern G, et al. Peripheral clearance of amyloid beta peptide by complement C3-dependent adherence to erythrocytes. Neurobiol Aging. 2006;27(12):1733–9. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.neurobiolaging.2005.09.043.

    Article  CAS  PubMed  Google Scholar 

  78. Mohanty JG, Shukla HD, Williamson JD, Launer LJ, Saxena S, Rifkind JM. Alterations in the red blood cell membrane proteome in alzheimer’s subjects reflect disease-related changes and provide insight into altered cell morphology. Proteome Sci. 2010;8: 11. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/1477-5956-8-11.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  79. Oma S, Mawatari S, Saito K, Wakana C, Tsuboi Y, Yamada T, et al. Changes in phospholipid composition of erythrocyte membrane in Alzheimer’s disease. Dement Geriatr Cogn Dis Extra. 2012;2(1):298–303. https://doiorg.publicaciones.saludcastillayleon.es/10.1159/000341603.

    Article  PubMed  PubMed Central  Google Scholar 

  80. Iwatsubo T, Iwata A, Suzuki K, Ihara R, Arai H, Ishii K, et al. Japanese Alzheimer’s Disease Neuroimaging Initiative. Japanese and north American Alzheimer’s disease neuroimaging initiative studies: harmonization for international trials. Alzheimers Dement. 2018;14(8):1077–87. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.jalz.2018.03.009.

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

The authors thank Sarah Bubeck, PhD, of Edanz (www.edanz.com) for providing medical writing support, which was funded by Eisai Co., Ltd., in accordance with Good Publication Practice (GPP 2022) guidelines (https://www.ismpp.org/gpp-2022).

Funding

This study was funded by Eisai Co., Ltd. This study used data from the USUKI study [51, 52], which was funded by the Japan Agency for Medical Research and Development (grant number JP 18he1402004, awarded to Etsuro Matsubara).

Author information

Authors and Affiliations

  1. Department of Neurology, Faculty of Medicine, Oita University, 1-1 Idaigaoka, Hasama-machi, Yufu, Oita, 879-5593, Japan

    Noriyuki Kimura, Teruaki Masuda, Takuya Ataka & Etsuro Matsubara

  2. Human Biology Integration Foundation, Deep Human Biology Learning, Eisai Co., Ltd, 4-6-10 Koishikawa, Bunkyo-ku, Tokyo, 112-8088, Japan

    Kotaro Sasaki

  3. Neurology Department, Medical Headquarters, Eisai Co., Ltd, 3-7-1 Nishi Shinjuku, Shinjuku-ku, Tokyo, 163-1023, Japan

    Mariko Matsumoto, Mika Kitamura & Yosuke Nakamura

Authors
  1. Noriyuki Kimura
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Kotaro Sasaki
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Teruaki Masuda
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Takuya Ataka
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Mariko Matsumoto
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Mika Kitamura
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Yosuke Nakamura
    View author publications

    You can also search for this author inPubMed Google Scholar

  8. Etsuro Matsubara
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

NK drafted the manuscript and contributed to the design of the study and the acquisition, analysis, and interpretation of data. MM and YN contributed to the design of the study. TM and TA contributed to data acquisition. KS performed the statistical analyses. KS and MK drafted the manuscript. EM supervised the study. All authors substantively revised the manuscript and approved the final version of the manuscript.

Corresponding authors

Correspondence to Noriyuki Kimura or Kotaro Sasaki.

Ethics declarations

Ethics approval and consent to participate

The study protocol was approved by the local ethics committee of Oita University Hospital. Given the difficulty of obtaining informed consent for existing patient information, an information disclosure document (approved by the ethics committee and authorized by the head of Oita University Hospital) that included information on the conduct of this study and the study purpose was made publicly available on the university website. The study was registered with the University hospital Medical Information Network Clinical Trials Registry (UMIN000051776).

Consent for publication

Not applicable.

Competing interests

NK has received consultancy/speaker fees from Takeda Pharmaceutical Co., Ltd., Daiichi Sankyo Co., Ltd., Eisai Co., Ltd., Sumitomo Pharma Co., Ltd., PDRadiopharma Inc., Otsuka Pharmaceutical Co., Ltd., and Eli Lilly Japan K.K. KS, MM, MK and YN are employees of Eisai Co., Ltd. TM, TA, and EM have no competing interest to declare.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Kimura, N., Sasaki, K., Masuda, T. et al. Machine learning models for dementia screening to classify brain amyloid positivity on positron emission tomography using blood markers and demographic characteristics: a retrospective observational study. Alz Res Therapy 17, 25 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-024-01650-1

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13195-024-01650-1

Keywords

Alzheimer's Research & Therapy

ISSN: 1758-9193