BACKGROUND:The visual interpretation of amyloid PET scans, or visual read (VR), is the most common technique used in clinical practice to identify the presence of cerebral amyloid plaques. Amyloid status (positive or negative) determined by VR or using a Centiloid (CL) cut-off shows high overall concordance. However, discordant cases can occur where the VR is positive, but the CL is below the positivity cut-off, or vice versa. OBJECTIVES:The objective of this analysis was to evaluate the rate of discordance and explore potential causes, particularly the role of amyloid tracer uptake in the white matter (WM), when determining amyloid status using VR and CL in screening for the elenbecestat phase 3 studies in early Alzheimer's disease (AD). DESIGN:Amyloid PET scans using either Florbetapir (Amyvid™), Florbetaben (Neuraceq™) or Flutemetamol (Vizamyl™) from 3,232 participants (1507 VR- and 1725 VR+) with cognitive impairment screened for the elenbecestat phase 3 studies in early AD were visually interpreted at screening by trained neuroradiologists and quantified using CL values. SETTING:Academic and clinical centers INTERVENTION/MEASUREMENTS: Quantitatively, amyloid positivity was defined as CL > 32.21. The number of positive cortical regions was determined by counting the number of regions with a standardized uptake value ratio (SUVr) that exceeded 1.17. PET SUVr levels in the cerebral WM were measured using an eroded WM region of interest (ROI). Statistical tests were conducted to detect differences among the four concordance groups, defined by the relationship of VR and CL status (positive or negative). Additionally, tests examined the relationship between uptake in the WM and rates and type of discordance. Receiver operating characteristic (ROC) analysis and DeLong's test were also used to examine the effect of different tracers on the discordant rates. RESULT:Discordance was observed in 6.53% of cases (n=211), with VR+/CL- in 4.61% (n=149) and VR-/CL+ in 1.92% (n=62). VR+/CL- discordant cases had significantly fewer amyloid-positive cortical regions compared to both VR+/CL+ and VR-/CL+ cases. VR-/CL+ cases had a significantly higher WM uptake than VR-/CL- and VR+/CL- cases. Our findings revealed a relationship between WM uptake and rates and types of discordance. High WM uptake can erroneously lead to CL+, due to gray matter (GM) contamination from the WM, and VR- status, due to reduced contrast between WM and GM, resulting in VR-/CL+ cases. Conversely, low WM uptake can result in an underestimation of CL values, inaccurately classifying a scan as CL-, and at the same time, the increased contrast may result in a VR+, thereby increasing the occurrence of discordant VR+/CL- cases. CONCLUSION:Variations in WM uptake significantly contribute to discordances by introducing positive or negative bias in CL values and altering the GM to WM contrast, which forms the basis of the VR. Nevertheless, the rates of discordant cases are low and VR represents a robust and validated method to determine the presence of amyloid deposition. VR enables enrolling patients with amyloid beta pathology, as seen on amyloid PET scans, whereas CL scaling was developed to provide standardized units that more consistently characterize longitudinal amyloid‑β change. These findings reflect the complementary roles of VR and CL in amyloid PET evaluation, with implications for refining diagnostic accuracy and disease monitoring in AD clinical trials and practice.

Long non-coding RNAs (lncRNAs) play a significant role in the pathogenesis of Alzheimer's disease (AD). They modulate various cellular processes such as amyloid production, Tau hyperphosphorylation, neuroinflammation, and the impairment of mitochondrial and synaptic functions. Emerging studies have revealed that certain lncRNAs can encode small open reading frame-derived peptides. However, the identification and understanding of the role of lncRNA-encoded peptides in AD remain largely unexplored due to the inherent low abundance and small sizes of these peptides. Here, we leveraged the de novo peptide sequencing algorithm and a custom database to identify lncRNA-encoded peptides in cerebrospinal fluid (CSF) of demented subjects We developed an innovative strategy to identify lncRNA-encoded peptides in biofluids. A custom database of hypothetical peptides was generated by six-frame translation of all lncRNAs from LNCipedia ( www.lncipedia.org ) and integrated to human SwissProt protein entries. MS data (PRIDE archive PXD016278) from CSF of demented subjects with ( n = 29) or without ( n = 31) amyloid positivity were analyzed. Peptide raw peak area intensities were quantile normalized and log2 transformed to reduce technical variation and ensure distribution symmetry. Differentially expressed peptides were identified via analysis of covariance after adjusting for age and gender, with the significant criteria based on 20% false discovery rate (FDR). The de novo -assisted search of MS spectra identified 32,191 peptides at 0.1% global peptide -level FDR . Mapping of de novo peptides to our custom database identified 99 lncRNA-encoded peptides in CSF of demented subjects. 7/99 significantly (q<0.2; Cohens >0.8) altered lncRNA-encoded peptides were linked to amyloid positivity. These peptides were translated from non-coding regions of six lncRNA genes involved in the cellular processes relevant to AD, including tau protein aggregation and glutamatergic synapse plasticity. This is the first study identifying differentially regulated lncRNA-encoded peptides in the CSF of Ab+ (AD) and Ab- (non-AD) dementia. Further investigation of novel lncRNA-derived peptides lights a new beacon to explore their promising applications in AD diagnosis, staging, and future treatment strategies.

Abstract Background: Metastatic breast cancer (MBC) is a major clinical challenge. Fam-trastuzumab deruxtecan-nxki (T-DXd), an antibody-drug conjugate targeting HER2, has shown efficacy across HER2+ and HER2-low subtypes. However, predictive biomarkers beyond HER2 status are underexplored in real-world settings. We leveraged real-world data from the Tempus database1 to evaluate associations between somatic mutations and clinical outcomes in a clinically annotated cohort of patients with MBC treated with T-DXd, aiming to identify genomic correlates of response and survival to inform patient selection and therapeutic strategies. Methods: We analyzed 124 patients with MBC with baseline tumor-normal matched sequencing data (Tempus xT 2). Endpoints included real-world best overall response (rwBOR), progression-free survival (rwPFS), time to next treatment (rwTTNT), and overall survival (rwOS). Patients were classified as responders (CR/PR) or nonresponders (SD/PD) based on curated rwBOR. Oncoplots identified frequently mutated genes by rwBOR and HER2 status. Logistic regression assessed rwBOR (nonresponder as reference); Cox proportional hazard models evaluated rwPFS, rwTTNT, and rwOS, adjusting for age at T-DXd initiation, ER/PR status, care plan, sampling time, and tissue location. Models were run for all variants and for pathogenic-only subsets, stratified by HER2 status. Genes with >4% mutation frequency were included (50-60 genes for all-variant models; 11-13 genes for pathogenic-only models). Results: The most frequently mutated genes were TP53 (48%), PIK3CA (31%), and GATA3 (17%). In all-variant models, DYNC2H1 was associated with worse rwBOR (HER2+: P= 0.006; HER2-low: P=0.049), rwOS (HER2-low: P=0.014), and rwTTNT (HER2-low: P=0.004). PIK3CA mutations correlated with improved rwBOR (HER2+: P=0.016), rwPFS (all: P=0.01; HER2-low: P=0.007), and rwOS (HER2-low: P=0.002). Additional genes with consistent associations included SPEN, POLQ, MED12, MAP2K4, ARID1B, SYNE1, KMT2C, and RB1. Pathogenic-only analyses confirmed PIK3CA as a key predictor across multiple endpoints. While most models yielded FDR-adjusted P values >0.2, we prioritized genes with nominal P<0.1 and consistent prognostic direction across endpoints as indicative of potential signal. Conclusions: Mutations such as PIK3CA were consistently correlated with improved outcome across HER2 subtypes, suggesting potential as predictive/prognostic biomarkers. Conversely, DYNC2H1 mutations correlated with poorer outcomes, particularly in HER2-low patients, implicating potential resistance mechanisms. These findings support integrating genomic data into real-world evidence frameworks to enhance patient stratification, personalize treatment, and guide biomarker-driven clinical trials in MBC. References: 1. www.tempus.com 2. Tempus-xT.v4_Validation Citation Format: Abraham Apfel, Alka A. Potdar, Yuanqing Ye, Viswanath Devanarayan, Evvie Jagoda, Shelley MacNeil, Yan Zhang, Pallavi Sachdev. Integrating genomics and real-world data to predict fam-trastuzumab deruxtecan response in metastatic breast cancer across HER2 subtypes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1038.

Heterogeneity in Alzheimer's disease (AD) progression introduces variability in treatment effect assessments. Using predicted future progression as an AD prognostic covariate (APC) may reduce this variability. This study evaluates this strategy in lecanemab trials and its implications for AD trial design. Two APCs were derived at baseline for each trial participant from published models with historical controls: one with clinical features, the other adding structural MRI features. Their impact on estimating the difference in cognitive decline between the treatment and placebo arms and the time saved from delayed progression (TSDP) was assessed. Incorporating either APC reduced variance estimates by up to 19.1% across phase II and phase III trials, increased power to 90.2%, and reduced sample size by 27.2%. These APCs improved treatment effect estimates and TSDP, demonstrating broad applicability across endpoints. APCs enhance treatment effect evaluation, improve statistical power, and reduce required sample sizes in Alzheimer's trials. NCT01767311 (Lecanemab Study 201), NCT03887455 (Lecanemab Study 301; ClarityAD). Baseline prediction of future progression can serve as an APC for treatment effect assessments. These predictions can be derived from progression models developed using external controls. APC accounts for heterogeneity in progression among trial participants, improving treatment effect estimates. Enhanced accuracy and precision were observed across lecanemab phase II and phase III trials for various endpoints. This approach results in substantial increase in statistical power and reduced sample size for future AD trials.

PET quantification of brain tau pathology aids in Alzheimer’s disease staging and patient screening. This study assesses whether the phosphorylated to nonphosphorylated plasma Tau217 ratio (pTau217R) predicts regional tau PET standardized uptake value ratio (SUVR) and accurately identifies subjects with different levels of tau accumulation. Plasma pTau217 and non-phosphorylated tau217 concentrations were quantified via immunoprecipitation-mass spectrometry. Predictive models for MK6240 tau PET SUVR were developed and validated using a 60-40 random split of a clinical trial cohort comprising 242 amyloid-β positive early Alzheimer’s disease individuals. The stochastic gradient boosting algorithm was employed to construct models for concurrently predicting SUVR values across various brain regions. Additional analyses explored whether integrating additional predictors (e.g., cognitive assessments, fluid biomarkers, structural MRI, and amyloid PET) into the model could improve the prediction performance of pTau217R. Model performance was cross-validated within the training set and evaluated further in a hold-out test set. pTau217R-based models accurately predicted tau PET uptake across various brain regions, with R2 values ranging from 0.49 to 0.65. The maximum SUVR values reliably predicted across these brain regions fell within the range of 1.87 to 2.3. The area under the receiver operating characteristic curve for detecting tau presence ranged from 84% to 95% across the six Braak stages and cortical regions, maintaining consistent performance at higher tau accumulation levels (Figure 1A-B; cortical regions). Integrating additional predictors did not improve pTau217R’s performance. Therefore, using pTau217R alone in predicting tau PET SUVR reduced the need for tau PET scans by up to 65%, particularly in identifying low tau concentrations within cortical grey matter, while maintaining a 5% false negative rate. Our study demonstrates the robust predictive ability of pTau217R in estimating regional brain tau levels among Aβ+ early Alzheimer’s disease patients, accurately discerning individuals with varying degrees of tau accumulation while notably reducing the need for tau PET scans. The simultaneous prediction of continuous SUVR values across multiple brain regions enables pathological disease staging, enhances practicality, and maximizes accessibility for patients, thereby providing greater flexibility in patient screening and monitoring procedures for both clinical trials and real-world clinical settings.

Cross-validation is a standard tool for obtaining a honest assessment of the performance of a prediction model. The commonly used version repeatedly splits data, trains the prediction model on the training set, evaluates the model performance on the test set, and averages the model performance across different data splits. A well-known criticism is that such cross-validation procedure does not directly estimate the performance of the particular model recommended for future use. In this paper, we propose a new method to estimate the performance of a model trained on a specific (random) training set. A naive estimator can be obtained by applying the model to a disjoint testing set. Surprisingly, cross-validation estimators computed from other random splits can be used to improve this naive estimator within a random-effects model framework. We develop two estimators – a hierarchical Bayesian estimator and an empirical Bayes estimator – that perform similarly to or better than both the conventional cross-validation estimator and the naive single-split estimator. Simulations and a real-data example demonstrate the superior performance of the proposed method.

INTRODUCTION:Mild cognitive impairment (MCI) is underdiagnosed by primary care providers (PCPs), with detection rates as low as 6%-15%. Predictive models support the identification of individuals at risk, enabling timely intervention. METHODS:This retrospective study was conducted on 271,054 cognitively unimpaired and 14,501 confirmed MCI cohorts from electronic health records. A machine learning model was developed with a data-driven variable selection approach based on demographics and comorbidities. RESULTS:From 101 variables, 26 were chosen for the final model, achieving an overall area under the curve (AUC) of 86%. Age-stratified AUCs were 79.1% (40-49), 77.0% (50-64), 76.9% (65-79), and 74.4% (≥80), showing the highest predictive performance in younger age groups. DISCUSSION:Demographic factors and comorbidities can serve as effective predictors for the risk of MCI. The model demonstrates strong predictive performance and assists as a triage tool for PCPs, facilitating the identification of individuals with MCI for early treatment. Highlights:Predictive algorithms using electronic health records (EHRs) are useful for identifying individuals at risk for mild cognitive impairment (MCI) to triage for further clinical evaluation.A machine learning model was developed using EHR data to predict those at risk for MCI.The model identified 26 variables that were able to distinguish the MCI from non-MCI cohorts.The model accurately detected MCI across the cohort (area under the curve [AUC] = 86%) and trended best for younger age groups (AUC was 77%, 77%, and 74% in 50-64, 65-79, and ≥80 age groups, respectively).Implementation of a triage tool could be used to detect MCI across aging patient populations sooner, leading to a timelier diagnosis, intervention, and treatment management.

INTRODUCTION:Cognitive decline in asymptomatic preclinical Alzheimer's disease (AD) is slow and variable, limiting detection of treatment effects. This study developed models to forecast trajectories and improve trial efficiency. METHODS:Models were trained on longitudinal Preclinical Alzheimer's Cognitive Composite (PACC) data up to 240 weeks from the Phase III A4 study of solanezumab. Baseline inputs included demographics, apolipoprotein E (APOE) ε4, clinical scores, amyloid positron emission tomography (PET), plasma pTau217, magnetic resonance imaging (MRI), and tau PET (sub-study). Stochastic gradient boosting was used, with evaluation via cross-validation and trial simulations. RESULTS:The best model without tau PET used pTau217, clinical, and MRI data (R2 = 0.32; area under the receiver operating characteristic curve (AUROC) for classifying a 0.5-point PACC decline = 78.6%). Replacing MRI with tau PET improved performance (R2 = 0.42; AUROC = 83.1%). Predicted trajectories as a prognostic covariate reduced sample sizes by 35% and increased power from 80% to 94.7%. DISCUSSION:Prognostic models can predict decline in preclinical AD and improve trial efficiency. CLINICALTRIALS: GOV IDENTIFIERS:NCT02008357 (Clinical Trial of Solanezumab for Older Individuals Who May be at Risk for Memory Loss (A4)) HIGHLIGHTS: Models forecast 4.5-year cognitive decline in amyloid-positive preclinical Alzheimer's disease (AD). Plasma pTau217 and tau positron emission tomography (PET) standardized uptake value ratios (SUVRs) in early-accumulating regions are key predictors. Tau PET improves prediction beyond plasma, magnetic resonance imaging (MRI), and clinical measures. Forecasted decline as a prognostic covariate improves power and cuts sample size in trial simulations. Alternative models underperform yet retain practical utility when tau PET or pTau217 is unavailable.

INTRODUCTION:Tau PET informs Alzheimer's disease (AD) staging but is limited by cost and access. Plasma phosphorylated-to-nonphosphorylated tau217 ratio (pTau217R) predicts amyloid and tau positron emission tomography (PET), while cerebrospinal fluid (CSF) MTBR-tau243 (tryptic form) and its endogenous form (eMTBR-tau243) reflect tau pathology. We evaluated whether combining them improves tau PET prediction in amyloid-positive early AD. METHODS:In the phase III lecanemab trial, plasma pTau217R was available for 242 participants, CSF markers for 80, and both for 57. Tau PET used [18F]MK6240. Stochastic gradient boosting models included demographics and apolipoprotein E (APOE) ε4, with performance assessed via repeated cross-validation. RESULTS:Plasma pTau217R matched CSF eMTBR-tau243 and outperformed MTBR-tau243. Combining pTau217R with either CSF marker improved prediction (R2 to 0.74; AUROC to 0.94) and could reduce PET use by 58%-80% for prescreening at 10% false negatives. DISCUSSION:Plasma pTau217R plus CSF eMTBR-tau243 predict regional tau burden and may reduce reliance on tau PET in early AD trials. CLINICALTRIALS: GOV IDENTIFIERS:NCT03887455 (ClarityAD) HIGHLIGHTS: Combining plasma pTau217R and CSF eMTBR‑tau243 improves tau PET prediction. The model predicts regional tau PET SUVRs with high accuracy across key regions. Single-marker models show pTau217R and eMTBR-tau243 outperform MTBR-tau243. PET‑sparing triage reduces scan burden by up to 80% at 10% false negatives. Stochastic gradient boosting enables region-specific tau PET prediction in one unified model.

Real-world evidence (RWE) can complement clinical trials by addressing gaps in how approved anti-amyloid therapies for early Alzheimer's disease (AD) are used in everyday practice. This article outlines strategies to generate RWE that bridge three key challenges in AD care: low detection rates of mild cognitive impairment (MCI), limited data on long-term safety and effectiveness, and a lack of personalized treatment strategies. With MCI detection rates among primary care providers as low as 6%-15%, we propose cost-effective triage tools using electronic health records to enhance early diagnosis and intervention. We also highlight the importance of understanding anti-amyloid therapy outcomes in diverse, real-world populations. Supported by FDA initiatives, pragmatic trials and observational studies using real-world data (RWD) can help develop predictive models that incorporate biomarkers and support precision medicine. These approaches aim to move AD care beyond one-size-fits-all treatment, guiding more tailored, effective strategies for patients.

Variability in cognitive decline in preclinical Alzheimer’s disease (AD) presents a significant challenge in evaluating treatment effects in clinical trials. This study developed models to forecast cognitive progression and assessed their potential to improve the precision of treatment effect estimates in future trials. Data were from the Phase III Anti-Amyloid Treatment trial of Solanezumab in amyloid-positive asymptomatic preclinical AD. Due to no significant cognitive decline differences between groups, the Solanezumab arm (n=549) was used for training, and the placebo arm (n=559) for validation. Cognitive decline was assessed as 24-week changes in Preclinical Alzheimer's Cognitive Composite (PACC) score over 216 weeks. A predictive model for PACC decline was developed using demographics, APOE ε4 status, and baseline clinical assessments (PACC composite score, its components, and CDR-SB), employing Stochastic Gradient Boosting. The added value of amyloid PET Centiloid (CL), plasma pTau217 (electrochemiluminescence assay), MRI morphometrics, and tau PET measures (375-participant substudy) were evaluated. Model performance was optimized via cross-validation and validated in the placebo group. The impact of baseline PACC decline predictions as an Alzheimer’s Prognostic Covariate (APC) on treatment effect assessments was evaluated in trial simulations. A model with demographics, CL, and clinical assessments explained 16% of PACC decline at week 216 (R²=0.16), improving over CL or clinical assessments alone (p<0.05). Adding MRI increased R² to 0.23 (p<0.05); replacing CL with plasma pTau217 or tau PET raised it to 0.25 and 0.42, respectively. Without tau PET in the model, key baseline predictors were pTau217, CL, PACC, MMSE, inferior temporal area, and entorhinal volume. When tau PET was included, it became the strongest predictor, with fusiform, inferior temporal, and supramarginal regions being the most predictive. Simulations showed baseline prediction of PACC decline as APC can reduce treatment effect variance by 20.3%, increase power from 80% to 88%, or reduce sample size by 21.5%. Incorporating tau PET further reduces variance by 37%, increases power to 94%, and decreases sample size by 35%. Baseline tau PET was the strongest predictor of PACC decline. Using baseline-predicted PACC decline as APC can enhance treatment effect estimates and trial efficiency in preclinical AD.

Timely identification of mild cognitive impairment (MCI) is key to early intervention. While primary care providers are the most likely entry point to detect early signs of MCI, their detection rates are low. Building upon a published study, we used electronic health records (EHR) to develop a clinically enhanced MCI risk prediction algorithm. This retrospective analysis utilized the Optum EHR database, which includes records for more than 110M individuals across the United States. MCI and non-MCI cohorts were identified from January 2016 – March 2021. Index date was first MCI diagnosis for MCI cohort and randomly selected diagnosis record for non-MCI cohort. All individuals were observed 2 years before index date. Derived from EHR, potential predictors of MCI included health status (e.g., body mass index [BMI]), pre-existing conditions identified with EHR diagnosis records and abstracted note records derived via natural language processing. Predictors of MCI will be identified using logistic regression, and an MCI risk prediction algorithm will be established using machine-learning approaches based on those predictors. Our sample included 21,059 and 631,770 individuals with and without MCI and mean age 71.1 and 59.1 years, respectively (p<0.001). Selected comorbidities in Table 1 are statistically significantly associated with MCI. Current work explores adjusting for age and adding health status indicators: nearly 90% of individuals had BMI data, with mean 28.8 and 30.5 in the MCI and non-MCI cohorts, respectively (p<0.001); 80% of individuals had smoking status data, 37.6% and 46.9% of the MCI cohort were previous smokers and never smokers, respectively, versus 29.2% and 54.2% in the non-MCI cohort (all p<0.001). Consistent with a recent published study, this analysis finds higher unadjusted odds of hypertension, hyperlipidemia, stroke, atherosclerosis, diabetes, weight loss, depression, hearing loss insomnia, sleep apnea, chronic obstructive pulmonary disease, and kidney disease in the MCI cohort. We aim to use the prediction model from this study to improve the ability to discriminate between MCI and non-MCI individuals and, ultimately, to develop a triage tool for PCPs to identify individuals at elevated MCI risk for further workup, based on routinely available data.

BACKGROUND:This study investigated the potential of phosphorylated plasma Tau217 ratio (pTau217R) and plasma amyloid beta (Aβ) 42/Aβ40 in predicting brain amyloid levels measured by positron emission tomography (PET) Centiloid (CL) for Alzheimer's disease (AD) staging and screening. METHODS:Quantification of plasma pTau217R and Aβ42/Aβ40 employed immunoprecipitation-mass spectrometry. CL prediction models were developed on a cohort of 904 cognitively unimpaired, preclinical and early AD subjects and validated on two independent cohorts. RESULTS:Models integrating pTau217R outperformed Aβ42/Aβ40 alone, predicting amyloid levels up to 89.1 CL. High area under the receiver operating characteristic curve (AUROC) values (89.3% to 94.7%) were observed across a broad CL range (15 to 90). Utilizing pTau217R-based models for low amyloid levels reduced PET scans by 70.5% to 78.6%. DISCUSSION:pTau217R effectively predicts brain amyloid levels, surpassing cerebrospinal fluid Aβ42/Aβ40's range. Combining it with plasma Aβ42/Aβ40 enhances sensitivity for low amyloid detection, reducing unnecessary PET scans and expanding clinical utility. CLINICALTRIALS: GOV IDENTIFIERS:NCT02956486 (MissionAD1), NCT03036280 (MissionAD2), NCT04468659 (AHEAD3-45), NCT03887455 (ClarityAD) HIGHLIGHTS: Phosphorylated plasma Tau217 ratio (pTau217R) effectively predicts amyloid-PET Centiloid (CL) across a broad spectrum. Integrating pTau217R with Aβ42/Aβ40 extends the CL prediction upper limit to 89.1 CL. Combined model predicts amyloid status with high accuracy, especially in cognitively unimpaired individuals. This model identifies subjects above or below various CL thresholds with high accuracy. pTau217R-based models significantly reduce PET scans by up to 78.6% for screening out individuals with no/low amyloid.

BACKGROUND:This study examines whether phosphorylated plasma Tau217 ratio (pTau217R) can predict tau accumulation in different brain regions, as measured by positron emission tomography (PET) standardized uptake value ratio (SUVR), for staging Alzheimer's disease (AD). METHODS:Plasma pTau217R was measured using immunoprecipitation-mass spectrometry. Models for predicting tau PET SUVR, developed with 144 early AD individuals using [18F]MK6240, were validated in two validation sets, VS1 (98 early AD) and VS2 (47 preclinical/early AD with a different tracer, flortaucipir (Tauvid)), all amyloid-beta positive (Aβ+). RESULTS:The pTau217R-based model predicted tau levels up to an SUVR of 2 in multiple brain regions, effectively assessing tau status at different tau levels with receiver operating characteristic (ROC) curve areas of 0.84-0.95 in VS1 and 0.71-0.88 in VS2 (using a different tracer). It reduced PET scan needs by 65% while maintaining 95% sensitivity. DISCUSSION:PTau217R reliably predicts regional tau accumulation in early AD, reducing reliance on tau PET scans and broadening its clinical application. CLINICAL TRIAL REGISTRATION NUMBER:NCT03887455 (ClarityAD) HIGHLIGHTS: Developed a model using plasma pTau217R to predict tau levels across brain regions. pTau217R model outperformed models based on clinical, MRI, and other blood biomarkers. The model reliably predicted tau levels exceeding tau positivity and higher thresholds. Screening with pTau217R could reduce tau PET scans by 65% at 95% sensitivity. pTau217R model aids in disease staging and monitoring in early AD.

AbstractPlasma pTau181, a marker of amyloid and tau burden, was evaluated as a prognostic predictor of clinical decline and Alzheimer's disease (AD) progression of amyloid‐positive (Aβ+) patients with mild cognitive impairment (MCI). The training cohort for constructing the Bayesian prediction models comprised 135 Aβ+ MCI clinical trial placebo subjects. Performance was evaluated in two validation cohorts. An 18‐month ≥1 increase in the Clinical Dementia Rating Sum of Boxes was the clinical decline criterion. Baseline plasma pTau181 concentration matched clinical assessments’ prediction performance. Adding pTau181 to clinical assessments significantly improved the prediction of an 18‐month clinical decline and the 36‐month progression from Aβ+ MCI to AD. The area under the receiver operating characteristic curve for the latter increased from 71.8% to 79%, and the hazard ratio for time‐to‐progression improved from 2.26 to 3.11 (p < 0.0001). Baseline plasma pTau181 has the potential for identifying Aβ+ MCI subjects with faster clinical decline over time.Highlights This study assessed pTau181 as a prognostic predictor of 18‐month clinical decline and extended progression to Alzheimer's disease (AD) in amyloid‐positive patients with mild cognitive impairment (Aβ+ MCI). The research findings underscore the promise of baseline plasma pTau181 as a screening tool for identifying Aβ+ MCI individuals with accelerated clinical decline within a standard 18‐month clinical trial period. The predictive accuracy is notably enhanced when combined with clinical assessments. Similar positive outcomes were noted in forecasting the extended progression of Aβ+ MCI subjects to AD.

Amyloid PET VR is the most common method to determine Aβ pathology in clinical practice. Amyloid positivity by VR or CL cut-off generally shows a good concordance, however, it is possible to have discordant cases whereby the VR is positive, but CL is below the cut-off of amyloid positivity, or vice versa. The objective of this analysis was to assess the rate and cause of discordance in defining amyloid pathology using VR and CL in Eisai’s Elenbecestat MissionAD Phase 3 program. Florbetaben, Florbetapir, and Flutemetamol amyloid PET scans from 3412 participants with MCI due to AD or mild AD dementia were visually read at screening by trained neuroradiologists and quantitated using CLs. VR classification of the amyloid status is based on the higher or equal grey matter (GM) uptake relative to uptake in the white matter (WM) in at least one area, as per manufacturer’s guidance. Quantitatively, amyloid negativity was defined as CL<30. The number of positive cortical regions (SUVR>1.17) was also calculated. The positivity concordance between VR and CL was 92%, while the negativity concordance was 94.6%. VR+/CL- discordance was 8% (n = 1814 VR+, n = 145 CL-) whereas VR-/CL+ discordance was 5.4% (n = 1598 VR-, n = 86 CL+). VR+/CL- discordant cases had significantly fewer positive cortical regions than both VR+/CL+ and VR-/CL+ cases (Figure 1). VR-/CL+ had a significantly higher WM uptake than both VR-/CL- and VR+/CL- cases (Figure 2). VR+/CL- discordant cases show fewer amyloid positive cortical regions, computing CLs results in below cut-off values due to a dilution effect. These cases are considered VR+ as per manufacturer’s guidelines as they show at least one area of amyloid accumulation. On the other hand, VR-/CL+ discordant cases result from an increased WM uptake, reducing GM/WM contrast required to determine visual read positivity. VR represents a robust and validated method to determine the presence of amyloid deposition and enables enrolling patients with amyloid beta pathology, as visible on amyloid PET scans, while CLs are optimal for assessing longitudinal changes over time and are a more sensitive measure to assess disease progression.

Background Identifying individuals with mild cognitive impairment (MCI) who are likely to progress to Alzheimer’s disease and related dementia disorders (ADRD) would facilitate the development of individualized prevention plans. We investigated the association between MCI and comorbidities of ADRD. We examined the predictive potential of these comorbidities for MCI risk determination using a machine learning algorithm. Methods Using a retrospective matched case-control design, 5185 MCI and 15,555 non-MCI individuals aged ≥50 years were identified from MarketScan databases. Predictive models included ADRD comorbidities, age, and sex. Results Associations between 25 ADRD comorbidities and MCI were significant but weakened with increasing age groups. The odds ratios (MCI vs non-MCI) in 50–64, 65–79, and ≥ 80 years, respectively, for depression (4.4, 3.1, 2.9) and stroke/transient ischemic attack (6.4, 3.0, 2.1). The predictive potential decreased with older age groups, with ROC-AUCs 0.75, 0.70, and 0.66 respectively. Certain comorbidities were age-specific predictors. Conclusions The comorbidity burden of MCI relative to non-MCI is age-dependent. A model based on comorbidities alone predicted an MCI diagnosis with reasonable accuracy.

BACKGROUND:Models for forecasting individual clinical progression trajectories in early Alzheimer's disease (AD) are needed for optimizing clinical studies and patient monitoring. METHODS:Prediction models were constructed using a clinical trial training cohort (TC; n = 934) via a gradient boosting algorithm and then evaluated in two validation cohorts (VC 1, n = 235; VC 2, n = 421). Model inputs included baseline clinical features (cognitive function assessments, APOE ε4 status, and demographics) and brain magnetic resonance imaging (MRI) measures. RESULTS:The model using clinical features achieved R2 of 0.21 and 0.31 for predicting 2-year cognitive decline in VC 1 and VC 2, respectively. Adding MRI features improved the R2 to 0.29 in VC 1, which employed the same preprocessing pipeline as the TC. Utilizing these model-based predictions for clinical trial enrichment reduced the required sample size by 20% to 49%. DISCUSSION:Our validated prediction models enable baseline prediction of clinical progression trajectories in early AD, benefiting clinical trial enrichment and various applications.

Non-invasive methods for detecting neurofibrillary tangles that spread throughout the brain are needed for efficiently screening patients in Alzheimer’s disease (AD) clinical trials. This can be accomplished via machine-learning models using baseline clinical characteristics. Adding plasma pTau181 and brain region volumes may improve detection accuracy. Regional MK6240 Tau-PET SUVR data were available for 312 Amyloid-positive patients from a clinical trial (182 MCI, 130 mild AD, 131 Tau-negative, 181 Tau-positive). Over 90% of the Tau-positive subjects were in Braak 3-6. Plasma pTau181 was measured using Simoa assay. Data were split randomly into 70%-30% for training and testing purposes. Tau positivity for each Braak stage was defined via the one-sided upper 99% confidence limit of a subset of 70 subjects with background SUVR. Signatures for detecting Tau positivity in Braak 3-6 and for further discriminating Braak 3-4 and 5-6 were derived via the stochastic gradient boosting algorithm and Bayesian ordinal logistic regression respectively. Demographics (age, gender, BMI), ApoE4 status, and cognitive assessments were included, and the added value of plasma pTau181 and volumetric MRI measures were assessed. Prediction performance was assessed via 10-fold cross-validation in the training set followed by evaluation in the test set. Signatures comprising psychometric data, ApoE4 status, and demographics achieved 76% accuracy (ROC AUC = 80.4%) for detecting Tau positivity in Braak 3-6, and 60.8% accuracy for discriminating Braak 3-4 versus 5-6. Adding plasma pTau181 significantly improved the accuracy to 82.2% ( ROC AUC = 86.6% , p<0.05) in Braak 3-6, and to 67.6% accuracy (p<0.05) for Braak 3-4 versus 5-6. Besides plasma pTau181, key predictors were delayed word recall, ADAS-cog-14, BMI, CDR-SB, and ApoE4 genotype. Detection accuracy of Tau positivity in Braak 3-4 versus the spread to Braak 5-6 improved significantly from 67.6% to 74.6% (p<0.05) when combining MRI data with plasma pTau181, ApoE4, and demographics. Key MRI predictors were the thickness and volume of the inferior parietal cortex. Adding cognitive assessments did not improve the detection accuracy. These results demonstrate the potential of non-invasive investigations such as plasma pTau181, cognitive performance, and MRI data for detecting Tau deposition throughout the brain along the Braak stages.

Prediction of longitudinal cognitive decline for patients with mild dementia using baseline characteristics will be useful for designing optimal clinical trials and real-world patient monitoring. This can be accomplished via machine-learning models using baseline clinical characteristics. Adding plasma pTau181 and brain region volumes may improve the predictions. The training cohort (TC) included 905 amyloid-positive (A+) patients with mild dementia from two clinical trials. The validation cohort (VC) included 230 A+ patients from another clinical trial. Over 85% of these patients had MCI. The longitudinal cognitive decline was defined by the change from baseline in CDR-SB at months 3, 6, 9, 12, 15, and 18. Plasma pTau181 was measured using Simoa assay in a subset of 159 patients in TC. Structural brain network (SBN) modules and hubs were derived using 207 regional volumetric MRI measures in TC via an algorithm from genomics called “multiscale embedded gene co-expression network analysis”. A signature for predicting longitudinal cognitive trajectory for each patient was first derived using baseline cognitive assessments and demographics within TC via the Stochastic Gradient Boosting algorithm. The added value of SBNs and plasma pTau181 was then evaluated within this framework. Prediction performance was evaluated in VC via the correlation between predicted versus observed cognitive trajectory. Predictions of the longitudinal cognitive trajectory using baseline cognitive assessments achieved 43.4% and 42.5% correlation with observed values at months 12 and 18 respectively. Key baseline cognitive function predictors were ADAS-Cog-14, CDR-SB, and sub-scores such as ideational praxis and word recall. Adding baseline SBNs significantly improved the correlation between predicted versus observed cognitive trajectory to 46.3% and 49.7% at months 12 and 18 respectively (p<0.05). Key baseline SBN predictors were middle and inferior temporal, inferior parietal, and superior frontal cortex, along with a module comprising the entorhinal cortex and temporal pole. Adding baseline plasma pTau181 to cognitive assessments yielded similar improvements, but adding both pTau181 and SBNs did not improve the predictions further. Longitudinal cognitive decline of A+ patients with mild dementia can be predicted using baseline cognitive assessments. Adding specific structural brain networks or plasma pTau181 significantly improves the predictions.