Inflammation indexes and machine-learning algorithm in predicting urethroplasty success
- Author:
Emre TOKUC
1
;
Mithat EKSI
;
Ridvan KAYAR
;
Samet DEMIR
;
Ramazan TOPAKTAS
;
Yavuz BASTUG
;
Mehmet AKYUZ
;
Metin OZTURK
Author Information
- Publication Type:Original Article
- From: Investigative and Clinical Urology 2024;65(3):240-247
- CountryRepublic of Korea
- Language:English
-
Abstract:
Purpose:To assess the predictive capability of hematological inflammatory markers for urethral stricture recurrence after primary urethroplasty and to compare traditional statistical methods with a machine-learning-based artificial intelligence algorithm.
Materials and Methods:Two hundred eighty-seven patients who underwent primary urethroplasty were scanned. Ages, smoking status, comorbidities, hematological inflammatory parameters (neutrophil-lymphocyte ratios, platelet-lymphocyte ratios [PLR], systemic immune-inflammation indexes [SII], and pan-immune-inflammation values [PIV]), stricture characteristics, history of previous direct-visual internal urethrotomy, urethroplasty techniques, and grafts/flaps placements were collected. Patients were followed up for one year for recurrence and grouped accordingly. Univariate and multivariate logistic regression analyses were conducted to create a predictive model. Additionally, a machine-learning-based logistic regression analysis was implemented to compare predictive performances. p<0.05 was considered statistically significant.
Results:Comparative analysis between the groups revealed statistically significant differences in stricture length (p=0.003), localization (p=0.027), lymphocyte counts (p=0.008), PLR (p=0.003), SII (p=0.003), and PIV (p=0.001). In multivariate analysis, stricture length (odds ratio [OR] 1.230, 95% confidence interval [CI] 1.142–1.539, p<0.0001) and PIV (OR 1.002, 95% CI 1.000–1.003, p=0.039) were identified as significant predictors of recurrence. Classical logistic regression model exhibited a sensitivity of 0.76, specificity of 0.43 with an area under curve (AUC) of 0.65. However, the machine-learning algorithm outperformed traditional methods achieving a sensitivity of 0.80, specificity of 0.76 with a higher AUC of 0.82.
Conclusions:PIV and machine-learning algorithms shows promise on predicting urethroplasty outcomes, potentially leading to develop possible nomograms. Evolving machine-learning algorithms will contribute to more personalized and accurate approaches in managing urethral stricture.
