Scholarlink

<p>With the rapid advancement of society and healthcare, life expectancy has steadily increased, accompanied by a growing aging population. Among elderly health concerns, cardiovascular and cerebrovascular diseases are the most prevalent and deadly. Due to complex physiological and pathological conditions, elderly patients require precise and rational medication, yet inappropriate prescriptions remain common and burdensome for physicians to evaluate individually. This study proposes a machine learning–based medication warning model for cardiovascular and cerebrovascular diseases in the elderly. Using NHANES data from 2013 to 2018, 1,054 samples with 48 indicators were extracted, including 31 input features (e.g., age, gender, race) and 17 output medication labels coded by ICD-10-CM. Feature selection yielded 12 physiological indicators (e.g., LDL, HDL, cell counts) and 20 comorbidity factors (e.g., hypertension, kidney disease). All features were standardized using zero-mean and Min-Max normalization. Three machine learning algorithms–Multilayer Perceptron, XGBoost, and Polynomial Naive Bayes–were applied to build six multi-target probability prediction models. Among them, the XGBoost model based on physiological features demonstrated the highest accuracy (96.91%) and sensitivity. A novel medication scoring algorithm was further developed, integrating the predicted probabilities and feature weights via the coefficient of variation method. By mapping probability thresholds and aggregating scores, the model enables efficient identification of inappropriate prescriptions. This study systematically integrates data preprocessing, feature engineering, model construction, evaluation, and interpretability analysis to develop an elderly medication safety warning system. It supports accurate medication decisions and enhances prescription safety.</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

Construction of drug safety early warning model for elderly cardiovascular and cerebrovascular diseases based on machine learning

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • Ruiquan Chen,
  • Kai Qu,
  • Zhequn Zhao,
  • Hui Cao,
  • Jie Deng,
  • Xinghui Li

摘要

With the rapid advancement of society and healthcare, life expectancy has steadily increased, accompanied by a growing aging population. Among elderly health concerns, cardiovascular and cerebrovascular diseases are the most prevalent and deadly. Due to complex physiological and pathological conditions, elderly patients require precise and rational medication, yet inappropriate prescriptions remain common and burdensome for physicians to evaluate individually. This study proposes a machine learning–based medication warning model for cardiovascular and cerebrovascular diseases in the elderly. Using NHANES data from 2013 to 2018, 1,054 samples with 48 indicators were extracted, including 31 input features (e.g., age, gender, race) and 17 output medication labels coded by ICD-10-CM. Feature selection yielded 12 physiological indicators (e.g., LDL, HDL, cell counts) and 20 comorbidity factors (e.g., hypertension, kidney disease). All features were standardized using zero-mean and Min-Max normalization. Three machine learning algorithms–Multilayer Perceptron, XGBoost, and Polynomial Naive Bayes–were applied to build six multi-target probability prediction models. Among them, the XGBoost model based on physiological features demonstrated the highest accuracy (96.91%) and sensitivity. A novel medication scoring algorithm was further developed, integrating the predicted probabilities and feature weights via the coefficient of variation method. By mapping probability thresholds and aggregating scores, the model enables efficient identification of inappropriate prescriptions. This study systematically integrates data preprocessing, feature engineering, model construction, evaluation, and interpretability analysis to develop an elderly medication safety warning system. It supports accurate medication decisions and enhances prescription safety.