The current gold standard treatment for benign prostatic hyperplasia (BPH) is transurethral resection of the prostate (TUR-P), which can lead to urethral stricture in about 10% of patients. This condition can negatively impact micturition capacity and quality of life [1]. Although the exact cause of urethral stricture formation is uncertain, it's thought to result from inflammatory processes leading to fibrosis in the urethral epithelium and subepithelial tissues. Risk factors for urethral stricture post-TUR-P include age, surgical duration, prostate volume, surgeon expertise, and infection presence [2]. Given the pivotal role of inflammation in the pathogenesis of urethral stricture, recent studies have concentrated their efforts on investigating this aspect. Despite all efforts, there is currently no established marker or method capable of predicting the inflammation related to urethral stricture development after TUR-P [3]. Based on this information and knowledge gap, we aimed to develop a machine-learning model by using preoperative laboratory data and different machine-learning algorithms to predict urethral stricture formation after TURP surgery.

A retrospective analysis of data was conducted from patients who underwent bipolar-TURP using the GyrusPlasmaKineticTM system (Gyrus ACMI, USA) between January 2015 and January 2022. Since surgeon experience may affect the results, cases performed by only two urologists with at least five years of experience in TUR-P were included in the study to minimize bias. Exclusion criteria were applied to individuals with a medical history of any urethral issues, bladder neck strictures, ongoing or prior cancer treatments (received or receiving chemotherapy and/or radiotherapy), hematologic disorders, active infections during the surgery, strictures at the location of previous resections, and those who had received blood transfusions.
The data perioperative hematological parameters, prostate-specific antigen levels, alanine aminotransferase (ALT), aspartate aminotransferase (AST), activated partial thromboplastin clotting time, prothrombin time, international normalized ratio, and age at the time of TURP from were recorded. Additionally, the following indices were calculated from the gathered data: 
•	De Ritis Ratio (DRR) = AST/ALT
•	NLR = Neutrophil Count/Lymphocyte Count
•	Mentzer Index (MI) = MCV/RBC
•	PLR= Platelet Count/ (Lymphocyte Count * 1000)
•	Lymphocyte/Monocyte Ratio (LMR) = Lymphocyte Count/Monocyte Count
•	Systemic Inflammatory Response Index (SIRI) = (Neutrophil Count * Monocyte Count) /Lymphocyte Count
•	Systemic Immune-Inflammatory Index (SII) = (Neutrophil Count * Platelet Count) / (Lymphocyte Count * 1000)
A machine learning model was developed by using the gathered data, indices, and six different machine learning algorithms (decision trees, logistic regression, random forests, support vector machines, K-nearest neighbors, and naive Bayes algorithms) (Figure 1). The data was randomly split into 80% for training and 20% for testing during the development of the machine learning models. To compare the performance of our machine learning models, we evaluated them based on different performance metrics (F1 score, model accuracy, negative predictive value, positive predictive value, sensitivity, specificity, Youden Index, ROC AUC value, and confidence interval (Figure-1). Python v3.11.5 and Python libraries were utilized in the development of the machine-learning models. 
Descriptive statistical analyses for both the groups with and without urethral strictures were conducted using Python libraries, pandas version 2.1.1, and scipy version 1.11.2. Statistical analysis included the calculation of mean ± standard deviation values.

A total of 109 patients (55 with urethral stricture and 54 without urethral stricture)  from the initial pool of 313 patients after applying the inclusion and exclusion criteria. Statistical analysis revealed no discernible statistical disparities between these two cohorts, except for preoperative Platelet Distribution Width (PDW), Mean Platelet Volume (MPV), Plateletcrit (PCT), Activated Partial Thromboplastin Time (APTT), and Prothrombin Time (PT) values. Furthermore, the median follow-up period after TURP and the median duration between the TURP procedure and the emergence of recurrence were computed at 12 months (with a range of 10 to 64 months) and 6.4 months, respectively. An analysis of the hematological parameters revealed significant disparities in PDW (16.56±2.22 vs. 11.54±2.41; p<0.001) and PT (11.11±0.92 vs. 9.12 ± 0.89; p<0.001) levels in patients who developed urethral stricture compared to those who did not exhibit this condition. In contrast, patients without urethral stricture showcased noteworthy elevations in MPV (9.85 ± 0.94 vs. 8.27 ± 1.18; p<0.001), PCT (0.25 ± 0.07 vs. 0.19 ± 0.05; p<0.001), and APTT (27.53 ± 2.60 vs. 26.09 ± 2.81; p=0.006) levels, when scrutinized against their counterparts who eventually developed urethral stricture.
Notably, the accuracy prediction scores for the diverse algorithms were as follows: decision trees (0.82), logistic regression (0.82), random forests (0.91), support vector machines (0.86), K-nearest neighbors (0.82), and naive Bayes (0.77) (Figure 2). To assess both the sensitivity and specificity of the developed models together, the Youden Index was calculated for the machine learning models. The model with the highest Youden Index was the Random Forest algorithm (0.82), and the model with the lowest Youden Index was the Naive Bayes algorithm (0.53).

The platelet indices such as MPV, PCT, and APTT were found to be lower in the urethral stricture developed group. 
The most successful model was the Random Forest algorithm (Accuracy 0.91, ROC AUC 0.96) and the model with the lowest accuracy was the Naive Bayes algorithm (Accuracy 0.77, ROC AUC 0.73). 
Models with a Youden Index greater than 0.5 are considered to discriminate successfully, and both our most successful model, Random Forest, and our least successful model, Naive Bayes, have Youden Index values greater than 0.5.

Our machine learning models showed good accuracy in predicting urethral stricture following TUR-P. However, prospective investigations incorporating additional parameters hold the potential to refine the specificity and sensitivity of machine learning models, thereby advancing their predictive capabilities.

Ivaz SL, Veeratterapillay R, Jackson MJ, Harding CK, Dorkin TJ, Andrich DE, Mundy AR. Intermittent self-dilatation for urethral stricture disease in males: A systematic review and meta-analysis. Neurourol Urodyn. 2016 Sep;35(7):759-63.
Kumar BN, Srivastava A, Sinha T. Urethral stricture after bipolar transurethral resection of prostate - truth vs hype: A randomized controlled trial. Indian J Urol. 2019 Jan-Mar;35(1):41-47.
Grimes MD, Tesdahl BA, Schubbe M, Dahmoush L, Pearlman AM, Kreder KJ, Erickson BA. Histopathology of Anterior Urethral Strictures: Toward a Better Understanding of Stricture Pathophysiology. J Urol. 2019 Oct;202(4):748-756.