Abstract
BACKGROUND: The primary criteria for diagnosing mild cognitive impairment (MCI) due to Alzheimer's Disease (AD) or probable mild AD dementia rely partly on cognitive assessments and the presence of amyloid plaques. Although these criteria exhibit high sensitivity in predicting AD among cognitively impaired patients, their specificity remains limited. Notably, up to 25% of non-demented patients with amyloid plaques may be misdiagnosed with MCI due to AD, when in fact they suffer from a different brain disorder. The introduction of anti-amyloid antibodies complicates this scenario. Physicians must prioritize which amyloid-positive MCI patients receive these treatments, as not all are suitable candidates. Specifically, those with non-AD amyloid pathologies are not primary targets for amyloid-modifying therapies. Consequently, there is an escalating medical necessity for highly specific blood biomarkers that can accurately detect pre-dementia AD, thus optimizing amyloid antibody prescription. OBJECTIVES: The objective of this study was to evaluate a predictive model based on peripheral biomarkers to identify MCI and mild dementia patients who will develop AD dementia symptoms in cognitively impaired population with high specificity. DESIGN: Peripheral biomarkers were identified in a gene transfer-based animal model of AD and then validated during a retrospective multi-center clinical study. SETTING: Participants from 7 retrospective cohorts (US, EU and Australia). PARTICIPANTS: This study followed 345 cognitively impaired individuals over up to 13 years, including 193 with MCI and 152 with mild dementia, starting from their initial visits. The final diagnoses, established during their last assessments, classified 249 participants as AD patients and 96 as having non-AD brain disorders, based on the specific diagnostic criteria for each disorder subtype. Amyloid status, assessed at baseline, was available for 82.9% of the participants, with 61.9% testing positive for amyloid. Both amyloid-positive and negative individuals were represented in each clinical group. Some of the AD patients had co-morbidities such as metabolic disorders, chronic diseases, or cardiovascular pathologies. MEASUREMENTS: We developed targeted mass spectrometry assays for 81 blood-based biomarkers, encompassing 45 proteins and 36 metabolites previously identified in AAV-AD rats. METHODS: We analyzed blood samples from study participants for the 81 biomarkers. The B-HEALED test, a machine learning-based diagnostic tool, was developed to differentiate AD patients, including 123 with Prodromal AD and 126 with mild AD dementia, from 96 individuals with non-AD brain disorders. The model was trained using 70% of the data, selecting relevant biomarkers, calibrating the algorithm, and establishing cutoff values. The remaining 30% served as an external test dataset for blind validation of the predictive accuracy. RESULTS: Integrating a combination of 19 blood biomarkers and participant age, the B-HEALED model successfully distinguished participants that will develop AD dementia symptoms (82 with Prodromal AD and 83 with AD dementia) from non-AD subjects (71 individuals) with a specificity of 93.0% and sensitivity of 65.4% (AUROC=81.9%, p<0.001) during internal validation. When the amyloid status (derived from CSF or PET scans) and the B-HEALED model were applied in association, with individuals being categorized as AD if they tested positive in both tests, we achieved 100% specificity and 52.8% sensitivity. This performance was consistent in blind external validation, underscoring the model's reliability on independent datasets. CONCLUSIONS: The B-HEALED test, utilizing multiomics blood-based biomarkers, demonstrates high predictive specificity in identifying AD patients within the cognitively impaired population, minimizing false positives. When used alongside amyloid screening, it effectively identifies a nearly pure prodromal AD cohort. These results bear significant implications for refining clinical trial inclusion criteria, facilitating drug development and validation, and accurately identifying patients who will benefit the most from disease-modifying AD treatments.