Scholarlink

Objective <p>This retrospective cohort study aimed to systematically evaluate the impact of different mechanical ventilation strategies on postoperative pulmonary complications (PPCs) in thoracic surgery and to establish a risk prediction model that facilitates high-risk patient identification and perioperative management.</p> Methods <p>A total of 300 patients undergoing thoracic surgery at a tertiary hospital were enrolled. Demographic data, perioperative indicators, and mechanical ventilation parameters were collected. Potential risk factors were initially identified by univariate analyses and then entered into multivariable logistic regression and machine learning models (random forest and XGBoost) for model training and evaluation. The primary outcome was the incidence and types of PPCs, and the predictive performances of different models were compared.</p> Results <p>PPCs occurred in 38.7% of the 300 patients included in the final analysis. Both univariate and multivariate analyses indicated that higher tidal volume, lower positive end-expiratory pressure (PEEP), higher plateau pressure, and prolonged ventilation duration were positively associated with an increased risk of PPCs. Conversely, pulmonary protective ventilation strategies and recruitment maneuvers had a significant protective effect. The logistic regression model demonstrated an AUC of 0.60, with key variables including tidal volume (OR = 2.119, 95%CI 1.362–3.294), PEEP level (OR = 1.570, 95%CI 1.180–2.088), and plateau pressure (OR = 1.984, 95%CI 1.202–3.276) as significant predictors. Compared with traditional logistic regression and random forest models, the XGBoost model showed moderate performance with potential clinical utility, with a sensitivity and specificity of approximately 0.70 and 0.48, respectively, and an AUC of 0.56. Feature importance analyses revealed that ventilation-related variables (e.g., tidal volume, PEEP, plateau pressure, and total ventilation time) carried substantial weight in predicting PPCs.</p> Conclusion <p>Adopting pulmonary protective ventilation strategies––such as low tidal volume, appropriate PEEP, and recruitment maneuvers––significantly reduces the risk of PPCs in patients undergoing thoracic surgery. Machine learning algorithms like XGBoost may assist in identifying high-risk patients and tailoring individualized ventilation management. Further applications of multicenter big data and real-time monitoring technologies may help optimize mechanical ventilation strategies and improve patient outcomes.</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

Application of risk prediction model to evaluate the effect of mechanical ventilation on postoperative pulmonary complications in thoracic surgery

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月19日

查看全文
  • Li Wang,
  • Xiaoyun He,
  • Minxi Zhao,
  • Jun Liu

摘要

Objective

This retrospective cohort study aimed to systematically evaluate the impact of different mechanical ventilation strategies on postoperative pulmonary complications (PPCs) in thoracic surgery and to establish a risk prediction model that facilitates high-risk patient identification and perioperative management.

Methods

A total of 300 patients undergoing thoracic surgery at a tertiary hospital were enrolled. Demographic data, perioperative indicators, and mechanical ventilation parameters were collected. Potential risk factors were initially identified by univariate analyses and then entered into multivariable logistic regression and machine learning models (random forest and XGBoost) for model training and evaluation. The primary outcome was the incidence and types of PPCs, and the predictive performances of different models were compared.

Results

PPCs occurred in 38.7% of the 300 patients included in the final analysis. Both univariate and multivariate analyses indicated that higher tidal volume, lower positive end-expiratory pressure (PEEP), higher plateau pressure, and prolonged ventilation duration were positively associated with an increased risk of PPCs. Conversely, pulmonary protective ventilation strategies and recruitment maneuvers had a significant protective effect. The logistic regression model demonstrated an AUC of 0.60, with key variables including tidal volume (OR = 2.119, 95%CI 1.362–3.294), PEEP level (OR = 1.570, 95%CI 1.180–2.088), and plateau pressure (OR = 1.984, 95%CI 1.202–3.276) as significant predictors. Compared with traditional logistic regression and random forest models, the XGBoost model showed moderate performance with potential clinical utility, with a sensitivity and specificity of approximately 0.70 and 0.48, respectively, and an AUC of 0.56. Feature importance analyses revealed that ventilation-related variables (e.g., tidal volume, PEEP, plateau pressure, and total ventilation time) carried substantial weight in predicting PPCs.

Conclusion

Adopting pulmonary protective ventilation strategies––such as low tidal volume, appropriate PEEP, and recruitment maneuvers––significantly reduces the risk of PPCs in patients undergoing thoracic surgery. Machine learning algorithms like XGBoost may assist in identifying high-risk patients and tailoring individualized ventilation management. Further applications of multicenter big data and real-time monitoring technologies may help optimize mechanical ventilation strategies and improve patient outcomes.