Answers are grounded in this paper and cite their sources.
Spontaneous language offers a scalable, noninvasive approach for identifying biological indicators of preclinical Alzheimer's disease (AD). We examined prediction of APOE4 carrier status and cerebral amyloid-beta burden using Cookie Theft picture descriptions from two cohorts of Hebrew- and English-speaking participants (IRAP, N=134; WRAP, N=1,541). Amyloid-beta burden was classified using lowest-quartile, median, and highest-quartile cutoffs, capturing accumulation below conventional amyloid-positivity thresholds. We compared embeddings from pretrained large language models (LLMs) with standard cognitive test scores, reflecting conventional clinical assessment, and theory-driven linguistic features, reflecting traditional language analysis. Embedding-based models achieved the highest F1 scores among constituent models in five of eight tasks involving APOE4 status and global amyloid-beta burden. High prediction disagreement suggested complementary signals across approaches. An any-positive ensemble significantly outperformed every constituent model and a prior-informed random baseline across all eight tasks, and significantly outperformed all constituent models in 21 of 24 regional amyloid-beta tasks. These findings highlight the complementarity of LLM-based, theory-driven, and cognitive measures for predicting AD-related genetic risk and amyloid pathology, and support further investigation of their potential for early screening approaches.
No takes yet. Share an insight, caveat, or question.
Zadok et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: