Scholarlink

Objective <p>Risk of Cardiovascular Disease (CVD) in Obstructive Sleep Apnea (OSA)patients is a major health concern as it elevates the cardiovascular strain, leading to severe complications. Machine learning algorithms have been predominantly engaged for traditional prediction tasks, but struggled with long-term sequential data. This study proposes a hybrid deep learning framework combining Self-Attention Mechanism-Based Long Short-Term Memory (SA-LSTM) network and eXtreme Gradient Boosting (XGBoost) to predict the risk of CVD in individuals diagnosed with OSA.</p> Methods <p>With 6411 subjects from Sleep Heart Health Study (SHHS), the model uses sequential ECG data and extracted Heart Rate Variability (HRV) to analyze cardiac activities. Our proposed prediction model uses a self-attention mechanism based LSTM segment to capture the long-term temporal dependencies with minimum recursive iterations. Meanwhile, XGBoost is implemented on the data to surpass baseline boosting techniques, where the weak features learners are integrated to form a strong learner. Prediction results of both individual models are combined by the weighted average method.</p> Results <p>Experimental results show how the proposed hybrid model outperforms other individual baseline models and demonstrates its effectiveness in CVD risk prediction. The prediction results are evaluated with MSE of 0.98, which outperformed the standalone XGBoost model (MSE = 2.77) by 64.62% and SA-LSTM model (MSE = 1.39) by 29.4% error reduction.</p> Conclusion <p>The results confirm the efficacy of our proposed model for evaluating the potential CVD risk under various data scales. The prediction model holds promise as it doesn’t transform the original signals, hence it can be applied for the diagnosis of other OSA-related disorders.</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

A Hybrid Self-Attention LSTM-XGBoost Model for Cardiovascular Disease Risk Prediction in Patients with Obstructive Sleep Apnea Using Sleep Heart Rate Variability Analysis

  • 期刊论文

  • 发表日期：

    2026年2月13日

查看全文
  • Prateek Pratyasha,
  • Aditya Prasad Padhy

摘要

Objective

Risk of Cardiovascular Disease (CVD) in Obstructive Sleep Apnea (OSA)patients is a major health concern as it elevates the cardiovascular strain, leading to severe complications. Machine learning algorithms have been predominantly engaged for traditional prediction tasks, but struggled with long-term sequential data. This study proposes a hybrid deep learning framework combining Self-Attention Mechanism-Based Long Short-Term Memory (SA-LSTM) network and eXtreme Gradient Boosting (XGBoost) to predict the risk of CVD in individuals diagnosed with OSA.

Methods

With 6411 subjects from Sleep Heart Health Study (SHHS), the model uses sequential ECG data and extracted Heart Rate Variability (HRV) to analyze cardiac activities. Our proposed prediction model uses a self-attention mechanism based LSTM segment to capture the long-term temporal dependencies with minimum recursive iterations. Meanwhile, XGBoost is implemented on the data to surpass baseline boosting techniques, where the weak features learners are integrated to form a strong learner. Prediction results of both individual models are combined by the weighted average method.

Results

Experimental results show how the proposed hybrid model outperforms other individual baseline models and demonstrates its effectiveness in CVD risk prediction. The prediction results are evaluated with MSE of 0.98, which outperformed the standalone XGBoost model (MSE = 2.77) by 64.62% and SA-LSTM model (MSE = 1.39) by 29.4% error reduction.

Conclusion

The results confirm the efficacy of our proposed model for evaluating the potential CVD risk under various data scales. The prediction model holds promise as it doesn’t transform the original signals, hence it can be applied for the diagnosis of other OSA-related disorders.