Construction of a LASSO regression-based predictive model for recurrence of urinary tract stones

Introduction

Urinary tract stones are a common global health problem, with incidence rates in southern China reaching 5–10% and an annual new case rate of approximately 150–200 per 100,000 population; about 25% of these patients require hospitalization. Stone recurrence is extremely high, with a 10-year recurrence rate up to 50% and rising to 90% over the subsequent decade (1,2). Stone formation is closely associated with metabolic abnormalities (e.g., hypercalciuria, hyperuricemia) (3,4), urinary tract obstruction (5), infection (6), and genetic predisposition (7).

Despite advances in minimally invasive techniques—such as flexible ureteroscopy and percutaneous nephrolithotomy (PCNL)—stone clearance remains challenging for complex calculi (e.g., staghorn stones), which carry a low stone-free rate (~30%) and a high risk of residual fragments leading to secondary obstruction (8). Although ureteroscopic lithotripsy usage increased from 37% in 2013 to 64% in 2021, postoperative management gaps persist: approximately 19.35% of patients experience recurrence within 1 year, driven by factors such as residual fragments, infection stones, and patient age (9). At present, only a minority of patients undergo comprehensive 24-hour urine analysis, and predictive assessments rely largely on single parameters (e.g., urinary calcium or uric acid). Joint analysis of stone composition, urinary metabolic profiles, urine cultures, and inflammatory markers remains uncommon, limiting the accuracy of recurrence-prediction models.

Machine learning—a key branch of artificial intelligence—can automatically extract the most predictive features from labeled data with minimal manual intervention (10-12). It has been established as a powerful approach for identifying reliable predictors and classifying distinct stone phenotypes (13-16). In this study, we apply state-of-the-art machine-learning techniques to integrate and analyze clinical, metabolic, and inflammatory data, with the goal of uncovering patterns and risk factors for urinary stone recurrence (17-20). Such models promise sensitive identification of high-risk patients during clinical follow-up and offer clinicians a more precise tool for individualized recurrence prediction. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-987/rc).

Methods

Study population and inclusion criteria

We retrospectively reviewed 1,000 patients who underwent surgical treatment for urinary tract stones at the Department of Urology, Xuzhou Central Hospital in Jiangsu Province, between October 2018 and October 2024.

Inclusion criteria

(I) Age at surgery ≥18 years; (II) underwent stone removal via PCNL, ureteroscopy (URS), extracorporeal shock-wave lithotripsy (ESWL), or equivalent procedures; (III) radiographic confirmation of complete stone clearance within one week postoperatively [no residual stones on computed tomography (CT) or ultrasound]; (IV) provision of written informed consent by the patient or legal guardian.

Exclusion criteria

Patients were excluded if they had: (I) chronic kidney disease [estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m²]; (II) active urinary tract infection requiring >2 weeks of antibiotic therapy; (III) pregnancy or lactation; (IV) incomplete clinical data or inability to comply with study requirements (e.g., unavailable follow-up data).

Ethical approval

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Xuzhou Central Hospital in Jiangsu Province (approval No. XZXY-LK-20250407-0041) and informed consent was obtained from all individual participants. All procedures adhered to the ethical standards of the Xuzhou Human Research Ethics Committee.

Data collection

Demographic and clinical data—including sex, age, surgical date, and hypertension history—were recorded. Stone recurrence was defined as a new calculus ≥3 mm on CT or ultrasound within 1 year post-surgery. Patients were categorized into recurrence and non-recurrence groups accordingly.

Stone composition analysis

Retrieved or spontaneously passed stones were cleaned of blood and tissue, dried, and ground to a fine powder (≤50 µm). Infrared spectroscopy (FLA8100, Crystal Technology, Hangzhou, China) was performed using the KBr pellet method (scan range 4,000–400 cm⁻¹; resolution 4 cm⁻¹). Spectra were matched against the JCPDS standard library (match threshold >60%) to identify principal components (e.g., calcium oxalate, struvite) (21).

Urinary metabolic parameter assessment

Ten milliliters of midstream morning urine were collected pre- and postoperatively. Automated urinalysis measured pH, specific gravity, bilirubin, urobilinogen, ketones, glucose, nitrite, protein, creatinine, albumin, albumin-to-creatinine ratio, occult blood, leukocyte esterase, and crystal morphology. Sediment microscopy quantified red blood cells, white blood cells, bacteria, and casts.

Urine culture and pathogen identification

Under sterile conditions, 10 mL of midstream urine was inoculated onto blood agar and MacConkey agar (Beyotime Biotechnology, Shanghai) and incubated at 37 ℃ for 24–48 h. Colony morphology was assessed, and positive cultures underwent species identification via the VITEK 2 Compact system.

Inflammatory marker measurement

Peripheral blood analyses included red blood cell count, white blood cell count (WBC), platelet count, hemoglobin concentration, neutrophil percentage, high-sensitivity C-reactive protein (hs-CRP), procalcitonin (PCT), interleukin-6 (IL-6), interleukin-8 (IL-8), and interleukin-10 (IL-10).

Least absolute shrinkage and selection operator (LASSO) feature selection

LASSO regression was employed for feature selection. Using 10-fold cross-validation to minimize mean squared error, the optimal penalty parameter (λ) was selected (22). Features with nonzero coefficients at this λ were retained to construct a clinical risk score as a weighted linear combination of predictors. Modeling was performed with scikit-learn v1.3.0 (https://scikit-learn.org/stable/index.html).

Machine learning model development

Retained features were used to train six algorithms: support vector machines (SVM), random forest (RF), k-nearest neighbors (KNN), eXtreme gradient boosting (XGBoost), LightGBM, and extra randomized trees (ExtraTrees). Five-fold cross-validation optimized model hyperparameters. All analyses were conducted in Python 3.11.4 (23) (https://www.python.org/).

Model evaluation

The hyperparameters were optimized using 5-fold cross-validation grid search with area under curve (AUC) as the main evaluation metric. Model performance was reported as AUC with 95% confidence intervals calculated using the DeLong method. Sensitivity, specificity, and accuracy were determined at the optimal threshold identified by Youden’s index. Model calibration was evaluated using calibration curves, and the calibration intercept and slope were reported. Decision curve analysis (DCA) assessed clinical utility across threshold probabilities.

Statistical analysis

The data was stratified and randomly divided into training set and test set in an 8:2 ratio (random_state =0). Data analysis employed GraphPad Prism 8.3.0. Normality was tested by the Shapiro-Wilk method. Continuous variables following a normal distribution are presented as mean ± standard deviation and compared by independent-samples t-test; non-normal data are reported as median (P25, P75) and compared by the Wilcoxon-Mann-Whitney test. Categorical variables were expressed as n (%) and compared using Chi-squared or Fisher’s exact test. A two-tailed P<0.05 was considered statistically significant.

Results

Clinical characteristics

A total of 800 patients completed 1-year follow-up with no losses to follow-up. Of these, 226 patients (28.25%) experienced stone recurrence and were assigned to the recurrence group; the remaining 574 patients formed the non-recurrence group. In the non-recurrence group (n=574), there were 430 males and 144 females, with a mean age of 48.71±13.81 years. In the recurrence group (n=226), there were 162 males and 64 females, with a mean age of 51.42±13.46 years. There were no significant differences between groups in sex distribution, operative time, hypertension prevalence, or diabetes prevalence (P>0.05). The mean age of patients in the recurrence group (51.42±13.46 years) was significantly older than that of the non-recurrence group (48.7±13.81 years) (P=0.01) (Table 1).

Stone composition analysis

The proportions of uric acid and calcium phosphate stones did not differ significantly between groups (both P>0.05). However, the incidence of calcium oxalate stones and magnesium ammonium phosphate (struvite) stones in the recurrent group was 84.07% and 8.85%, respectively, significantly higher than 74.39% (P=0.001) and 3.48% (P=0.002) in the non-recurrent group (Table 2).

Urinary metabolic parameters

Urinary metabolic findings are summarized in Table 3 and Table S1. In urine sediment analysis, the recurrence group showed significantly higher levels of red blood cell count (U-RBC, 2876.00±621.70 urine µL), white blood cell count (U-WBC, 610.70±155.70 µL), bacterial count (BACT, 2,002.00±578.20) and cast count (CAST, 0.86±1.43) compared with the non-recurrence group (855.00±195.40 µL, 121.60±23.56 µL, 618.10±85.36, 0.55±0.71). The positive detection rate of struvite crystals in the recurrence group was 10.18%, significantly higher than 3.83% in the non-recurrence group. No significant intergroup differences were observed in the prevalence of calcium phosphate or uric acid crystals, urinary specific gravity, bilirubin, urobilinogen, ketones, or occult blood (all P>0.05). Urinary creatinine (UCr) levels in the recurrence group were significantly lower than in the non-recurrence group. Compared with non-recurrence patients, those with recurrence exhibited significantly higher rates of: urinary pH (P=0.03), glucose (U-GLU) positivity (P=0.03), nitrite (NIT) positivity (P<0.001), protein (U-PRO) positivity (P<0.001), albumin (U-ALB) positivity (P<0.001), albumin-to-creatinine ratio (UACR) positivity (P<0.001), leukocyte esterase (LEU) positivity (P<0.001).

Urine culture and pathogen identification

There were no significant differences in the isolation rates of Klebsiella pneumoniae, Proteus mirabilis, Enterococcus faecalis, or Candida albicans (all P>0.05). The isolation frequency of Escherichia coli in the recurrence group was 11.95%, which was significantly higher than 4.70% in the non-recurrence group (P<0.001) (Table 4).

Inflammatory markers

No significant differences were observed for platelet count (PLT), total WBC, neutrophil percentage (NEUT), or interleukin-10 (IL-10) (all P>0.05). The red blood cell count in patients in the recurrence group was [RBC, (4.42±0.62)×10¹²/L] and haemoglobin (HGB, 133.00±19.48 g/L), significantly lower than [RBC, (5.33±1.43) ×10¹²/L] and (HGB, 141.50±17.04 g/L) in the non-recurrence group (P<0.001), while hs-CRP (10.76±25.48 mg/L, P=0.001), PCT (1.61±5.56 ng/mL, P=0.03), IL-6 (5.91±15.90 pg/mL, P=0.02) and IL-8 (13.09±14.35 pg/mL, P=0.005) levels were significantly elevated compared with the non-recurrence group (4.27±17.68 mg/L, 0.80±1.02 ng/mL, 3.36±3.03 pg/mL, 10.22±8.52 pg/mL) (Table 5).

LASSO feature selection

Twenty-four variables showing statistical significance in univariate analysis were entered into a LASSO logistic regression. As the tuning parameter log λ increased, many coefficients shrank toward zero. In the training set, 10-fold cross-validation identified 15 features with nonzero coefficients (Figure 1A,1B). These features and their corresponding coefficients at the optimal λ (0.0063) are displayed in Figure 2. The resulting recurrence score formula is: recurrence of calculus = 0.2825 + 0.004898 × Age + 0.003888 × pH + 0.027998 × NIT + 0.004187 × U-PRO − 0.010329 × UCr + 0.078322 × U-ALB + 0.029470 × UACR +0.051199 × LEU + 0.033426 × U-RBC (urinary sediment) + 0.017001 × U-WBC (urinary sediment) + 0.004695 × ammoniomagnesium phosphate crystal − 0.081725 × RBC + 0.037508 × PCT + 0.022687 × IL-6 + 0.015027 × IL-8.

Figure 1 The coefficient and MSE are cross-validated by 10 times. (A) A coefficient of 10 times cross-validation. Different colors represent different characteristics. (B) 10× cross-validation. HGB, haemoglobin; hs-CRP, high-sensitivity C-reactive protein; IL-6, interleukin-6; IL-8, interleukin-8; LEU, leukocyte esterase; MSE, mean standard error; NIT, nitrite; PCT, procalcitonin; RBC, red blood cell; U-GLU, urinary glucose; U-PRO, urinary protein; UACR, albumin-to-creatinine ratio; UCr, urinary creatinine; WBC, white blood cell.

Figure 2 Rad score histogram based on selected features. IL-6, interleukin-6; IL-8, interleukin-8; LEU, leukocyte esterase; NIT, nitrite; PCT, procalcitonin; RBC, red blood cell; U-ALB, urinary albumin; U-PRO, urinary protein; UACR, albumin-to-creatinine ratio; UCr, urinary creatinine; WBC, white blood cell.

Model comparison

The 15 selected features were used to train six classifiers—SVM, RF, KNN, XGBoost, LightGBM, and ExtraTrees—using 100 iterations of stratified five-fold cross-validation. Performance metrics in the test set are summarized in Table 6. The XGBoost model in the test group demonstrated good discriminative ability, with an AUC value of 0.87 (95% confidence interval: 0.81–0.92; sensitivity: 0.87; specificity: 0.74). Figure 3A shows the AUC for each model in testing cohorts. We have attached the AUC results of the training and test sets for the main model (Figure 3B-3F). Among all models (Figure 4A-4F), XGBoost demonstrated the good calibration (intercept =−0.003, slope =1.045), followed by RF. The SVM classifier showed acceptable calibration with slight overconfidence (slope =1.252), indicating predicted probabilities were more extreme than observed outcomes (Table 7). The DCA results of each model are shown in Figure 5A-5F.

Figure 3 ROC curve analysis of different models on Rad features. (A) Test cohort. (B) Model XGBoost. (C) Model RF. (D) Model SVM. (E) Model ExtraTrees. (F) Model KNN. AUC, area under the curve; CI, confidence interval; ExtraTrees, extra randomized trees; KNN, k-nearest neighbor; LightGBM, gradient boosting decision tree; NA, not available; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machines; XGBoost, extreme gradient boosting.

Figure 4 Calibration curves of all models. (A) Model XGBoost. (B) Model RF. (C) Model SVM. (D) Model ExtraTrees. (E) Model KNN. (F) Model LightGBM. ExtraTrees, extra randomized trees; KNN, k-nearest neighbor; LightGBM, gradient boosting decision tree; RF, random forest; SVM, support vector machines; XGBoost, extreme gradient boosting.

Figure 5 Decision curves of all models. (A) Model XGBoost. (B) Model LightGBM. (C) Model ExtraTrees. (D) Model SVM. (E) Model RF. (F) Model KNN. DCA, decision curve analysis; ExtraTrees, extra randomized trees; KNN, k-nearest neighbor; LightGBM, gradient boosting decision tree; RF, random forest; SVM, support vector machines; XGBoost, extreme gradient boosting.

Discussion

Urinary tract stone recurrence remains a major challenge in urological practice because recurrence is common even after apparently successful stone removal, and current follow up strategies are often insufficiently individualized. We are dedicated to constructing a machine learning model to predict the recurrence of urinary tract stones in patients. In this study, we employed a LASSO logistic regression model to analyse 45 features, including the patients’ basic clinical information, urinary stone composition, urinary metabolic parameters, urine culture results, and inflammatory indicators. Ultimately, 15 features were selected to construct the machine learning model.

Identification and correction of risk factors are the cornerstone of preventing urinary tract stone recurrence (9,24). Urinary tract infection stones risk factors are the cornerstone of preventing urinary tract stone recurrencend infurea into ammonia and carbon dioxide, which elevates urine pH and promotes the precipitation of magnesium ammonium phosphate (25). In this study, most patients with urinary tract stones were under 60 years of age. The majority of stone samples (73.4%) were mixed compositions, with calcium oxalate stones accounting for the highest proportion (77.3%), followed by uric acid stones (9.7%) and infectious stones (5.0%). Aizezi et al. (26) retrospectively analyzed the association between stone composition and patient clinical characteristics, revealing a continuous decline in uric acid content during the 15-year study period, while infectious stone content gradually increased. Additionally, factors influencing urinary stone formation included gender, age, body mass index (BMI), albumin, creatinine, WBC, urine pH, nitrite, and urine culture, which similar to our research findings. Although infectious stone composition was not included as a significant predictor of stone recurrence in this study (a marked increase in infectious stones among recurrent patients), this may be attributed to the inclusion criteria. The study only followed recurrence for 1 year, overlooking the gradual increase in stone content over time (27,28). A notable strength of this study is that the final model did not rely solely on conventional stone related variables, but rather incorporated features that reflect several biologically plausible pathways involved in recurrence. The selected predictors can be broadly interpreted within three domains. First, urinary pH, NIT, LEU, WBC (urinary sediment), and ammonium magnesium phosphate crystals suggest a role for infection related urinary microenvironmental changes. Second, U-PRO, U-ALB, UACR, UCr, and RBC (urinary sediment) may reflect ongoing renal tubular or urothelial irritation, microinjury, or persistent local vulnerability after stone events. Third, circulating inflammatory markers including PCT, IL-6, IL-8, and RBC count may indicate a systemic inflammatory milieu that is associated with recurrent stone activity. Taken together, these findings support the concept that stone recurrence is not driven by a single metabolic abnormality, but by the interaction of infection, urinary biochemical disturbance, local tissue injury, and host inflammatory response (7,29,30).

The model also highlighted markers of urinary tract injury and renal microenvironmental disturbance. U-PRO, U-ALB, UACR, RBC (urinary sediment), and UCr were retained in the final feature set. These variables may reflect persistent epithelial irritation, subtle tubular dysfunction, or ongoing inflammatory exudation after a stone episode. Rather than being viewed as isolated abnormalities, they should be interpreted as part of a broader pattern indicating incomplete restoration of urinary homeostasis. From a clinical perspective, this observation is important because it suggests that recurrence risk may be detectable not only through classic metabolic workup, but also through routinely available urinalysis parameters.

Another clinically relevant finding is the contribution of systemic inflammatory markers. PCT, IL-6, and IL-8 were retained by LASSO, and recurrent stone formers also showed evidence of a more pronounced inflammatory profile. Previous studies have suggested that inflammatory activity is closely linked to crystal retention, papillary injury, and stone related renal tissue responses (31,32). Our findings do not establish a causal relationship, but they support the biological plausibility that inflammatory activation is associated with an increased risk of recurrence. This aspect may partly distinguish our model from previous recurrence models that relied primarily on imaging or 24-hour urine data. It also raises the possibility that recurrence prediction could be improved by combining traditional metabolic assessment with markers reflecting host response.

Of course, ML is increasingly being used in the study of predicting urinary stone recurrence (33-35). Homayoun et al. (36) screened the clinical data, demographics, and CT results of 4,246 patients and, after evaluating six machine learning classifiers, found that the RF model was the best model for predicting stone recurrence, with an AUC of 0.64. Geraghty et al. (37) applied a multicentre, externally validated algorithm to routine urine biochemistry tests and found that XGBoost could accurately classify stone types but could not reliably predict recurrence. Shee et al. (17) prospectively evaluated the performance of seven classifiers in 24-hour urine analysis, identifying elastic net regularized logistic regression predicts stone recurrence with moderate accuracy. One likely explanation is that recurrence is a heterogeneous outcome with multiple underlying mechanisms, and models built on a single data source may fail to capture this complexity. By integrating stone related, urinary, microbiological, and inflammatory dimensions, our model may better reflect the real world pathophysiology of recurrent disease. At the same time, differences in recurrence definition, follow up duration, feature availability, patient selection, and validation strategy across studies make direct comparisons difficult. Therefore, our results should be interpreted as encouraging rather than definitive evidence of superior model performance. In this study, we used the Lasso regression method to identify 15 high-value predictive factors, among which the feature coefficients of urinary data ALB, LEU, RBC, UACR and WBC were ranked highest. We then compared six ML classifiers and found that XGBoost offered better discrimination. The XGBoost model, demonstrated good performance (AUCce0.87 in the test set). Meanwhile, XGBoost demonstrated the good calibration (intercept =−0.003, slope =1.045). By integrating a variety of clinical and laboratory characteristics, our model demonstrates relatively better performance compared to the early prediction methods that rely solely on urine biochemical indicators. However, information bias and standardization differences across studies have led to divergent outcomes. Future research on urinary stone recurrence models should further control for confounding factors to optimize model accuracy and enhance its clinical applicability (38).

The potential translational value of this model lies in postoperative risk stratification. In clinical practice, a tool based on routinely available perioperative data could help identify patients who may benefit from intensified surveillance, early metabolic evaluation, stricter infection control, and more individualized counseling regarding hydration, diet, and preventive medication (29,39,40). Such an approach is particularly relevant because comprehensive metabolic workup is not uniformly performed in all stone formers, despite guideline recommendations emphasizing recurrence prevention and metabolic assessment (9). A practical prediction model may therefore serve as a bridge between routine postoperative care and more personalized secondary prevention.

Nevertheless, there are limitations in this study. First, this was a retrospective single center study, and selection bias cannot be excluded. Patients with incomplete data were excluded, which may also have introduced bias. Second, the study outcome was recurrence within 1 year after surgery. Although this endpoint is clinically relevant, it may not fully capture the longer term dynamics of stone recurrence. Third, several established recurrence related variables were not available or were not incorporated, including body mass index, dietary intake, fluid intake, prior stone history, family history, residual fragment burden, medication based prevention, and 24-hour urine analysis. The absence of these factors may have affected both model performance and interpretability. Fourth, the present study performed internal testing using a random training and test split, but did not include external validation. As a result, the apparent predictive performance may be optimistic, and the generalizability of the model remains uncertain. Fifth, although XGBoost achieved the best performance, the interpretability of the model remains limited. Future studies should incorporate explainable machine learning methods, such as SHAP based analyses, to better clarify the relative contribution and interaction of individual predictors.

Future research should focus on prospective multicenter validation, longer follow up, and broader integration of imaging, metabolic, and molecular biomarkers. In addition, converting the model into a clinically usable tool, such as a web based calculator or electronic medical record integrated risk score, would enhance its practical value. A further important direction is to determine whether model guided surveillance and targeted preventive interventions can actually reduce recurrence, rather than merely predict it.

Conclusions

In conclusion, we developed a machine learning based model for predicting postoperative recurrence of urinary tract stones and found that XGBoost provided the good performance among the tested classifiers in the internal test cohort. The model incorporated multidimensional predictors reflecting urinary microenvironment, infection related features, and systemic inflammatory status, and showed promising discrimination and calibration. These findings suggest that integrated machine learning approaches may improve recurrence risk stratification after stone surgery.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-987/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-987/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-987/prf

Funding: This study was supported by Scientific research project of Jiangsu Provincial Health Commission (Z2021038); Youth Medical Science and Technology Innovation Project of Xuzhou Municipal Health Commission (XWKYSL20210234); the Xuzhou Medical Key Talents Project (XWRCHT20220055) and the Youth Science and Technology Innovation Team, Affiliated Hospital of Xuzhou Medical University (XYFC202404).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-1-987/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Xuzhou Central Hospital in Jiangsu Province (approval No. XZXY-LK-20250407-0041) and informed consent was obtained from all individual participants. All procedures adhered to the ethical standards of the Xuzhou Human Research Ethics Committee.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Tan S, Yuan D, Su H, et al. Prevalence of urolithiasis in China: a systematic review and meta-analysis. BJU Int 2024;133:34-43. [Crossref] [PubMed]
2. Wang P, Zhang H, Zhou J, et al. Study of risk factor of urinary calculi according to the association between stone composition with urine component. Sci Rep 2021;11:8723. [Crossref] [PubMed]
3. Wang X, Ma Q, Xie X, et al. Relationship between the Composition of Urinary Calculi and Protein Biomarkers in Serum and Urine. Arch Esp Urol 2025;78:84-92. [Crossref] [PubMed]
4. Xu Z, Yao X, Duan C, et al. Metabolic changes in kidney stone disease. Front Immunol 2023;14:1142207. [Crossref] [PubMed]
5. Papatsoris A, Alba AB, Galán Llopis JA, et al. Management of urinary stones: state of the art and future perspectives by experts in stone disease. Arch Ital Urol Androl 2024;96:12703. [Crossref] [PubMed]
6. Trinchieri A. Urinary calculi and infection. Urologia 2014;81:93-8. [Crossref] [PubMed]
7. Wagner CA. Etiopathogenic factors of urolithiasis. Arch Esp Urol 2021;74:16-23. [PubMed]
8. Klein I, Gutiérrez-Aceves J. Preoperative imaging in staghorn calculi, planning and decision making in management of staghorn calculi. Asian J Urol 2020;7:87-93. [Crossref] [PubMed]
9. Skolarikos A, Somani B, Neisius A, et al. Metabolic Evaluation and Recurrence Prevention for Urinary Stone Patients: An EAU Guidelines Update. Eur Urol 2024;86:343-63. [Crossref] [PubMed]
10. Jordan MI, Mitchell TM. Machine learning: Trends, perspectives, and prospects. Science 2015;349:255-60. [Crossref] [PubMed]
11. Sassanarakkit S, Hadpech S, Thongboonkerd V. Theranostic roles of machine learning in clinical management of kidney stone disease. Comput Struct Biotechnol J 2023;21:260-6. [Crossref] [PubMed]
12. Shi Y, Lin J, Zhu J, et al. Predicting the Recurrence of Common Bile Duct Stones After ERCP Treatment with Automated Machine Learning Algorithms. Dig Dis Sci 2023;68:2866-77. [Crossref] [PubMed]
13. Balasubramanian A, Bhambhvani H, Lee J, et al. Artificial Intelligence and Machine Learning for Stone Management. Urol Clin North Am 2025;52:465-74. [Crossref] [PubMed]
14. Kim US, Kwon HS, Yang W, et al. Prediction of the composition of urinary stones using deep learning. Investig Clin Urol 2022;63:441-7. [Crossref] [PubMed]
15. Chew BH, Wong VKF, Halawani A, et al. Development and external validation of a machine learning-based model to classify uric acid stones in patients with kidney stones of Hounsfield units < 800. Urolithiasis 2023;51:117. [Crossref] [PubMed]
16. Guo J, Zhang J, Zhang J, et al. Construction and validation of a urinary stone composition prediction model based on machine learning. Urolithiasis 2025;53:154. [Crossref] [PubMed]
17. Shee K, Liu AW, Chan C, et al. A Novel Machine-Learning Algorithm to Predict Stone Recurrence with 24-Hour Urine Data. J Endourol 2024;38:809-16. [Crossref] [PubMed]
18. Liu X, Wang Y, Wang Y, et al. A machine learning-based nomogram model for predicting the recurrence of cystitis glandularis. Ther Adv Urol 2024;16:17562872241290183. [Crossref] [PubMed]
19. Sabuncu Ö, Bilgehan B, Kneebone E, et al. Effective deep learning classification for kidney stone using axial computed tomography (CT) images. Biomed Tech (Berl) 2023;68:481-91. [Crossref] [PubMed]
20. Dai JC, Johnson BA. Artificial intelligence in endourology: emerging technology for individualized care. Curr Opin Urol 2022;32:379-92. [Crossref] [PubMed]
21. Han X, Zhang Z, Yao P, et al. Infrared spectroscopic analysis of urinary stone composition. Actas Urol Esp (Engl Ed) 2025;49:501810. [Crossref] [PubMed]
22. Kong L, Xue W, Zhao H, et al. Predicting pleural invasion of invasive lung adenocarcinoma in the adjacent pleura by imaging histology. Oncol Lett 2023;26:438. [Crossref] [PubMed]
23. Fan Z, Jiang J, Xiao C, et al. Construction and validation of prognostic models in critically Ill patients with sepsis-associated acute kidney injury: interpretable machine learning approach. J Transl Med 2023;21:406. [Crossref] [PubMed]
24. Frassetto L, Kohlstadt I. Treatment and prevention of kidney stones: an update. Am Fam Physician 2011;84:1234-42. [PubMed]
25. Flannigan R, Choy WH, Chew B, et al. Renal struvite stones--pathogenesis, microbiology, and management strategies. Nat Rev Urol 2014;11:333-41. [Crossref] [PubMed]
26. Aizezi X, Xie L, Xie H, et al. Epidemiological and clinical characteristics of stone composition: a single-center retrospective study. Urolithiasis 2022;50:37-46. [Crossref] [PubMed]
27. Nguyen LD, Nguyen TT, Mai LV, et al. The first epidemiology of urolithiasis in Northern Vietnam: Urinary stone composition, age, gender, season, and clinical features study. Urologia 2024;91:42-8. [Crossref] [PubMed]
28. Siener R, Rüdy J, Herwig H, et al. Mixed stones: urinary stone composition, frequency and distribution by gender and age. Urolithiasis 2024;52:24. [Crossref] [PubMed]
29. Wang K, Ge J, Han W, et al. Risk factors for kidney stone disease recurrence: a comprehensive meta-analysis. BMC Urol 2022;22:62. [Crossref] [PubMed]
30. Peerapen P, Thongboonkerd V. Kidney Stone Prevention. Adv Nutr 2023;14:555-69. [Crossref] [PubMed]
31. Dejban P, Wilson EM, Jayachandran M, et al. Inflammatory Cells in Nephrectomy Tissue from Patients without and with a History of Urinary Stone Disease. Clin J Am Soc Nephrol 2022;17:414-22. [Crossref] [PubMed]
32. Williams JC Jr, El-Achkar TM. Recent developments in the study of cellular inflammation in the papillae of stone formers. Urolithiasis 2025;53:34. [Crossref] [PubMed]
33. Wu Y, Mo Q, Xie Y, et al. A retrospective study using machine learning to develop predictive model to identify urinary infection stones in vivo. Urolithiasis 2023;51:84. [Crossref] [PubMed]
34. Anastasiadis A, Koudonas A, Langas G, et al. Transforming urinary stone disease management by artificial intelligence-based methods: A comprehensive review. Asian J Urol 2023;10:258-74. [Crossref] [PubMed]
35. Zheng J, Yu H, Batur J, et al. A multicenter study to develop a non-invasive radiomic model to identify urinary infection stone in vivo using machine-learning. Kidney Int 2021;100:870-80. [Crossref] [PubMed]
36. Homayoun H, Mousavirad SJ, Zareian Baghdadabad L, et al. Machine Learning-Based Prediction of Urolithiasis Recurrence Using Patient's Clinical Data, Demography, and CT Findings. Urol J 2026;22:289-300. [PubMed]
37. Geraghty RM, Wilson I, Olinger E, et al. Routine Urinary Biochemistry Does Not Accurately Predict Stone Type Nor Recurrence in Kidney Stone Formers: A Multicentre, Multimodel, Externally Validated Machine-Learning Study. J Endourol 2023;37:1295-304. [Crossref] [PubMed]
38. Doyle P, Gong W, Hsi R, et al. Machine Learning Models to Predict Kidney Stone Recurrence Using 24 Hour Urine Testing and Electronic Health Record-Derived Features. Preprint. Res Sq. 2023;rs.3.rs-3107998. doi: 10.21203/rs.3.rs-3107998/v1.10.21203/rs.3.rs-3107998/v1
39. Siener R, Stein J, Ritter M. Prevention of recurrence of urolithiasis. Urologie 2024;63:387-95. [Crossref] [PubMed]
40. Xia K, Xu Y, Qi Q, et al. Establishment of a new predictive model for the recurrence of upper urinary tract stones. Int Urol Nephrol 2023;55:2411-20. [Crossref] [PubMed]