Scholarlink

<p>Early dementia risk stratification remains challenging despite advances in biomarker research. We evaluated machine learning approaches for predicting 10-year incident dementia among older adults using routinely collected demographic, cardiometabolic, and cognitive measures from the Sydney Memory and Ageing Study. From 1037 community-dwelling Australians aged ≥70 years at baseline, 432 participants were alive and assessed at approximately 10-year follow-up, of whom 119 developed dementia. We compared logistic regression, LASSO regression, random forest, and XGBoost models using baseline predictors. Models were trained on 70% of participants and evaluated on a 30% held-out test set. LASSO regression demonstrated the highest discrimination (AUC = 0.752), outperforming logistic regression (AUC = 0.707), random forest (AUC = 0.657), and XGBoost (AUC = 0.589), retaining four predictors: age, global cognition, fasting glucose, and cardiovascular disease risk score. At the Youden-optimal threshold, sensitivity was 0.698 and specificity was 0.736. Decision-curve analysis indicated greater net clinical benefit across a range of plausible risk thresholds. These findings indicate that a parsimonious model using four accessible variables can support dementia risk stratification up to a decade before diagnosis among older adults who survive to follow-up. The model is intended for use in research and specialist clinical settings where structured cognitive assessment is available; external validation is required prior to broader clinical deployment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning models for dementia risk prediction: evidence from the Sydney Memory and Ageing Study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月27日

查看全文
  • Rebecca A. Chalmers,
  • Matti Cervin,
  • Carol Choo,
  • Katya Numbers,
  • Karen A. Mather,
  • Henry Brodaty,
  • Nicole A. Kochan,
  • Perminder S. Sachdev,
  • Oleg N. Medvedev

摘要

Early dementia risk stratification remains challenging despite advances in biomarker research. We evaluated machine learning approaches for predicting 10-year incident dementia among older adults using routinely collected demographic, cardiometabolic, and cognitive measures from the Sydney Memory and Ageing Study. From 1037 community-dwelling Australians aged ≥70 years at baseline, 432 participants were alive and assessed at approximately 10-year follow-up, of whom 119 developed dementia. We compared logistic regression, LASSO regression, random forest, and XGBoost models using baseline predictors. Models were trained on 70% of participants and evaluated on a 30% held-out test set. LASSO regression demonstrated the highest discrimination (AUC = 0.752), outperforming logistic regression (AUC = 0.707), random forest (AUC = 0.657), and XGBoost (AUC = 0.589), retaining four predictors: age, global cognition, fasting glucose, and cardiovascular disease risk score. At the Youden-optimal threshold, sensitivity was 0.698 and specificity was 0.736. Decision-curve analysis indicated greater net clinical benefit across a range of plausible risk thresholds. These findings indicate that a parsimonious model using four accessible variables can support dementia risk stratification up to a decade before diagnosis among older adults who survive to follow-up. The model is intended for use in research and specialist clinical settings where structured cognitive assessment is available; external validation is required prior to broader clinical deployment.