Answers are grounded in this paper and cite their sources.
Background: Mixed-format lifestyle questionnaires combine structured responses with free-text narratives, but the predictive value of compact longitudinal representations for amyotrophic lateral sclerosis (ALS)-versus-control classification remains uncertain. This study evaluated the contribution of temporal feature representations. Methods: We analyzed 103 participants (60 controls, 43 ALS cases), including 95 eligible for the primary temporal analysis. GPT-6 Sol extracted compact lifestyle descriptors. Three feature pools were compared: baseline variables (Pool1), additional T1 structured and text-derived features (Pool2), and Pool2 plus 24 revised T1–2 change descriptors (Pool3). Logistic regression, linear support vector classification (SVC), and Random Forest were evaluated using ten repetitions of nested five-fold cross-validation. Results: Pool3 achieved mean macro-F1 scores of 0.681 ± 0.031, 0.659 ± 0.031, and 0.647 ± 0.034, respectively. Pooled Pool3-Pool2 improvements were +0.117, +0.143, and +0.099 (Holm-adjusted exploratory p-values: 0.044, 0.009, and 0.255). The advantage persisted directionally in 36 duration-matched pairs, although no matched contrast survived multiplicity correction. Removing correlation filtering did not reverse this advantage. Conclusion: Compact longitudinal representations improved internal ALS-versus-control classification relative to T1-only enrichment, supporting further investigation and external validation before clinical application.
No takes yet. Share an insight, caveat, or question.
Nova et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: