Scholarlink

Background <p>Postoperative atrial fibrillation (POAF) is a frequent complication following cardiac surgery, linked to higher risks of death, stroke, and extended hospital stays. Given that glycemic control metrics—such as stress hyperglycemia ratio (SHR), glycemic variability (GV), and hemoglobin glycation index (HGI)—have been linked to adverse cardiovascular outcomes, it is important to understand their role in POAF. However, their comparative predictive value for POAF remains unclear. Therefore, this study aimed to evaluate the predictive value of SHR, GV, and HGI for POAF and to develop a machine learning-based model for POAF risk prediction.</p> Methods <p>We retrospectively analyzed 2,177 cardiac surgery patients from the MIMIC-IV database (median age 69 years [IQR 60–76]; 69.4% male). SHR, GV, and HGI were calculated from postoperative glucose measurements. Associations with POAF were examined using multivariable logistic regression, restricted cubic splines, threshold effect, and subgroup analyses. Thirteen machine learning algorithms were compared to develop a prediction model, with variable importance assessed using SHapley Additive exPlanations (SHAP). The final model was implemented as an interactive Shiny web application.</p> Results <p>POAF incidence was 39.8%. SHR and HGI were identified as independent predictors (SHR: OR = 1.39, 95% CI 1.01–1.93, <i>P</i> = 0.049; HGI: OR = 0.90, 95% CI 0.82–0.98, <i>P</i> = 0.017). Restricted cubic spline analysis demonstrated a negative linear association between HGI and POAF risk (P for overall = 0.039; P for nonlinearity = 0.375). In contrast, SHR showed a nonlinear relationship with POAF, with risk increasing above an inflection point of 0.9067 (P for overall = 0.019; P for nonlinearity = 0.042). GV also exhibited a significant nonlinear association with POAF risk (<i>P</i> = 0.025; P for nonlinearity = 0.049). The final AdaBoost model achieved an AUC of 0.74 (95% CI% 0.684–0.778), demonstrating moderate discrimination. In contrast, the SHR-only model yielded an AUC of 0.557 (95% CI% 0.512–0.602). These findings indicate that integrating multidimensional clinical variables substantially improves predictive performance compared with a single metabolic indicator. However, as the model was developed using a single database with internal validation only, external validation in independent cohorts is required before clinical implementation.</p> Conclusion <p>SHR and HGI are independent predictors of POAF, and GV shows a threshold effect. An AdaBoost-based web tool enables individualized, rapid perioperative POAF risk estimation, potentially supporting tailored prevention strategies.</p> Graphical Abstract <p></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

Prognostic value of stress hyperglycemia ratio, hemoglobin glycation index, and glycemic variability for postoperative atrial fibrillation: a machine learning-based prediction model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月24日

查看全文
  • Runjia Liu,
  • Chenglong Yao,
  • Hongfan Qiu,
  • Jiatong Li,
  • Ling Yao,
  • Dongdong Su,
  • Haixia Li

摘要

Background

Postoperative atrial fibrillation (POAF) is a frequent complication following cardiac surgery, linked to higher risks of death, stroke, and extended hospital stays. Given that glycemic control metrics—such as stress hyperglycemia ratio (SHR), glycemic variability (GV), and hemoglobin glycation index (HGI)—have been linked to adverse cardiovascular outcomes, it is important to understand their role in POAF. However, their comparative predictive value for POAF remains unclear. Therefore, this study aimed to evaluate the predictive value of SHR, GV, and HGI for POAF and to develop a machine learning-based model for POAF risk prediction.

Methods

We retrospectively analyzed 2,177 cardiac surgery patients from the MIMIC-IV database (median age 69 years [IQR 60–76]; 69.4% male). SHR, GV, and HGI were calculated from postoperative glucose measurements. Associations with POAF were examined using multivariable logistic regression, restricted cubic splines, threshold effect, and subgroup analyses. Thirteen machine learning algorithms were compared to develop a prediction model, with variable importance assessed using SHapley Additive exPlanations (SHAP). The final model was implemented as an interactive Shiny web application.

Results

POAF incidence was 39.8%. SHR and HGI were identified as independent predictors (SHR: OR = 1.39, 95% CI 1.01–1.93, P = 0.049; HGI: OR = 0.90, 95% CI 0.82–0.98, P = 0.017). Restricted cubic spline analysis demonstrated a negative linear association between HGI and POAF risk (P for overall = 0.039; P for nonlinearity = 0.375). In contrast, SHR showed a nonlinear relationship with POAF, with risk increasing above an inflection point of 0.9067 (P for overall = 0.019; P for nonlinearity = 0.042). GV also exhibited a significant nonlinear association with POAF risk (P = 0.025; P for nonlinearity = 0.049). The final AdaBoost model achieved an AUC of 0.74 (95% CI% 0.684–0.778), demonstrating moderate discrimination. In contrast, the SHR-only model yielded an AUC of 0.557 (95% CI% 0.512–0.602). These findings indicate that integrating multidimensional clinical variables substantially improves predictive performance compared with a single metabolic indicator. However, as the model was developed using a single database with internal validation only, external validation in independent cohorts is required before clinical implementation.

Conclusion

SHR and HGI are independent predictors of POAF, and GV shows a threshold effect. An AdaBoost-based web tool enables individualized, rapid perioperative POAF risk estimation, potentially supporting tailored prevention strategies.

Graphical Abstract