Scholarlink

Objective <p>To develop, validate, and compare a Traditional Multivariable Logistic Regression model with a Machine Learning-based LASSO Regression Model for predicting significant renal function recovery in adult patients undergoing surgical repair for ureteric obstruction, and to present the most practical model as a clinical nomogram.</p> Methods <p>We retrospectively analyzed data from 100 adult patients who underwent surgical repair for unilateral ureteric obstruction at our institution. The primary endpoint was significant renal function recovery, defined as a ≥ 15% increase in ipsilateral glomerular filtration rate (GFR). Two predictive models were developed: a parsimonious “Multivariable Model” using backward stepwise regression, and a “LASSO Model” using least absolute shrinkage and selection operator regression. The models’ performance was rigorously compared based on discrimination (Area Under the Curve - AUC), calibration (calibration curve), and clinical utility (Decision Curve Analysis - DCA). Internal validation was performed using 500 bootstrap resamples.</p> Results <p>The Multivariable Model identified four independent predictors: age (OR = 1.05, <i>p</i> = 0.042), preoperative ipsilateral GFR (OR = 1.20, <i>p</i> &lt; 0.001), renal atrophy (OR = 6.32, <i>p</i> = 0.028), and SFU Grade 4 (vs. Grade 2, OR = 23.8, <i>p</i> = 0.003). The LASSO model selected 12 variables. The LASSO model showed a slightly higher discrimination (AUC: 0.841 vs. 0.827) and a better optimism-corrected AUC in bootstrap validation (0.779 vs. 0.776). Decision Curve Analysis also revealed a marginally higher net benefit for the LASSO model. However, the simpler 4-variable Multivariable Model demonstrated excellent calibration and its clinical utility was very close to that of the more complex model. Given its superior balance of high performance and clinical practicality, the Multivariable Model was selected and presented as a nomogram.</p> Conclusion <p>We developed a robust 4-variable nomogram for predicting renal recovery. While a more complex LASSO model showed marginal statistical advantages, our simpler model provides a better balance of accuracy, interpretability, and clinical utility, making it a valuable tool for patient counseling and surgical decision-making.</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 nomogram for predicting renal function recovery after robotic-assisted ureteral reconstruction: development and comparative validation using traditional and machine learning models

  • 期刊论文

  • 发表日期：

    2025年11月25日

查看全文
  • Mu-Yang Xu,
  • Bin-Bin Gong,
  • Yu He,
  • Guang-Jie Ji,
  • Zheng-Yao Song,
  • Chao-Zhao Liang

摘要

Objective

To develop, validate, and compare a Traditional Multivariable Logistic Regression model with a Machine Learning-based LASSO Regression Model for predicting significant renal function recovery in adult patients undergoing surgical repair for ureteric obstruction, and to present the most practical model as a clinical nomogram.

Methods

We retrospectively analyzed data from 100 adult patients who underwent surgical repair for unilateral ureteric obstruction at our institution. The primary endpoint was significant renal function recovery, defined as a ≥ 15% increase in ipsilateral glomerular filtration rate (GFR). Two predictive models were developed: a parsimonious “Multivariable Model” using backward stepwise regression, and a “LASSO Model” using least absolute shrinkage and selection operator regression. The models’ performance was rigorously compared based on discrimination (Area Under the Curve - AUC), calibration (calibration curve), and clinical utility (Decision Curve Analysis - DCA). Internal validation was performed using 500 bootstrap resamples.

Results

The Multivariable Model identified four independent predictors: age (OR = 1.05, p = 0.042), preoperative ipsilateral GFR (OR = 1.20, p < 0.001), renal atrophy (OR = 6.32, p = 0.028), and SFU Grade 4 (vs. Grade 2, OR = 23.8, p = 0.003). The LASSO model selected 12 variables. The LASSO model showed a slightly higher discrimination (AUC: 0.841 vs. 0.827) and a better optimism-corrected AUC in bootstrap validation (0.779 vs. 0.776). Decision Curve Analysis also revealed a marginally higher net benefit for the LASSO model. However, the simpler 4-variable Multivariable Model demonstrated excellent calibration and its clinical utility was very close to that of the more complex model. Given its superior balance of high performance and clinical practicality, the Multivariable Model was selected and presented as a nomogram.

Conclusion

We developed a robust 4-variable nomogram for predicting renal recovery. While a more complex LASSO model showed marginal statistical advantages, our simpler model provides a better balance of accuracy, interpretability, and clinical utility, making it a valuable tool for patient counseling and surgical decision-making.