In 2022, prostate cancer accounted for around 27% of all new cancer diagnoses in the United States and was the most common cancer diagnosis among men (1). Research has focused on prostate biopsies as the gold standard for diagnosing prostate cancer in recent years. There are two methods for prostate biopsies: transrectal (TR) and transperineal (TP) (2). TR has historically carried out prostate biopsies. The burden that postoperative sepsis places on patient health and adds to the expenditures to the healthcare system, however, is the primary critique of this strategy (3–6). As a result of the declining percentage of TP biopsy in sepsis cases, there has been a nationwide movement to offer TP biopsy. AUR, or acute urinary retention, is another danger associated with prostate biopsies that may be higher than other TR techniques (7–10). AUR frequently necessitates catheterization and additional hospitalization (11, 12). While TP biopsy may lessen infection, it does not entirely prevent infection. According to studies referenced by NICE, urinary retention rates range from 1.6 to 11.4% (12).

Data evaluating possible risk factors for AUR after prostate biopsy in TP are currently hard to come by (13–15). Because medical big data is so common, machine learning (ML), the most significant artificial intelligence implementation technique, has been applied extensively to data-driven risk prediction (16, 17). Given the high dimensionality of the data, we chose to use LASSO first for preliminary screening to reduce Boruta ‘s computational burden and improve the efficiency of the overall analysis. Therefore, after screening out risk factors using LASSO regression. We further screened variables using Boruta ‘s feature screening method. We then iteratively processed random fluctuations in forest importance scores and factor interactions to screen for significant urinary retention predictors. In addition, this approach is commonly used for feature selection in diabetes mellitus (DM) studies (18, 19). In addition, Meng Zirui et al. constructed predictive models for severe novel coronavirus pneumonia using variant analysis and least absolute contraction and selection operator (LASSO) modeling along with Boruta ‘s algorithm (20). Ding Xuexuan et al. identified the core genes of asthma using five machine learning algorithms: LASSO, SVM-RFE, Boruta, XGBoost, and RF (21).

In view of the high clinical incidence of acute urinary retention (AUR) after transperineal prostate puncture (TP) and the lack of a precise predictive model for this complication, the aim of this study was to construct and validate a nomogram model that can be used to rapidly assess the risk of AUR preoperatively by integrating clinical routine indicators (such as BMI, prostate volume, etc.) using LASSO and Boruta feature selection algorithms to provide a quantitative tool for clinical decision-making in order to reduce the postoperative catheter indwelling rate and length of hospital stay.

Current studies have mostly focused on the comparison of diagnostic accuracy and complication rates between transperineal and transrectal prostate biopsy in the diagnosis of prostate cancer (27–30). Since template biopsies are carried out more frequently than TRUS biopsies, evaluating any problems and the patient’s following effects is critical. Before template biopsy, it is helpful to understand that postoperative complications, counseling, and patient consent are clinically significant risk factors linked with urine retention. According to reports, between 0.10% and 3% of TP biopsies result in infection (13, 31, 32). AUR occurs in 1.5-13% of patients having TP prostate biopsies (14, 15, 33, 34). Although the frequency of these occurrences has been reported, there is a shortage of information evaluating probable risk factors for infection and AUR following prostate biopsy for TP (13, 14). Predictive models are receiving more attention in several clinical domains, including sepsis, Kawasaki disease, and malignant tumors. Additionally, there is a growing body of pertinent research in these domains (35–38). For AUR following TP, there aren’t many prediction models available yet. Therefore, to support clinical decision-making and offer individualized treatment plans, we set out to create a predictive model for developing AUR following TP surgery.

The clinical information and laboratory parameters of 599 patients who had prostate biopsies were reviewed retrospectively. These included easily accessible laboratory and anthropological data, such as age, sex, BMI, prostate-specific antigen, diabetes, and residual urine before biopsy. We determined the independent variables for AUR in TP patients to be body mass index, prostate volume, history of diabetes, constipation, IPSS, and residual urine before biopsy using various statistical techniques. We created a straightforward and precise nomogram, verified it within the model, and demonstrated its clinically solid applicability and efficacy.

In 14.3% of our cohort, urinary retention happened. Urinary retention has an incidence ranging from 1.6 to 11.4%, according to the NICE guidelines for transperineal template biopsy of the prostate (12, 33). Pepe et al. (34) conducted a large single-center study involving 3000 patients with varied amounts of biopsies, and they found a low mean incidence of urine retention of 6.7%. Our rate was compared to other groups (39), even though it was marginally higher than the incidence reported by NICE, as reported by Merrick et al. Urinary retention occurs 1.7% of the time in patients receiving TRUS biopsy, according to a comprehensive review by Loeb et al. (11). In contrast to transperineal biopsy, which typically requires at least 24 cores, the conventional protocol for TRUS biopsy only requires 12 cores. Urinary retention can also be prevented by reducing the number of biopsies performed. Taking targeted biopsies instead of entire templates can also save operating time. However, it is contingent upon the operator’s cognitive objectives. We selected targeted + systematic access, a total of 16 needles, for patients with apparent targets on MRIs, which may be why the number of biopsy needles we used did not lower the rate of urine retention.

In line with our findings, several studies have found a positive correlation between the frequency of urine retention and the number of needle biopsies performed; however, this relationship was not statistically significant (33, 34). Urinary retention was linked to a prostate volume of more than 68 milliliters, according to Willis et al. (22). Increased prostate volume has also been found in several studies to be an independent predictor of AUR diagnosis after TP (7, 14). The ratio of the transitional zone volume to the total prostate volume and a greater IPSS are two additional prostate volume markers that may predispose patients to urine retention and worsen LUTS (11). Urinary retention was significantly predisposed to the severity of LUTS in our study. Furthermore, it has been suggested that more enormous prostate volumes and higher baseline IPSS values may be indicators of LUTS and AUR following PB (40). Consequently, discussing risk with patients could be considered a preoperative risk factor. Constipation can be regarded as a significant and trustworthy predictor of AUR in individuals receiving transrectal ultrasonography-guided prostate biopsy, according to research by Cahit Sahin et al. (41). Acute postoperative urine retention has also been described in our community and is also connected with diabetes mellitus and aging (11). One known consequence of diabetes is neuroautonomic dysfunction, which has been linked to an increased risk of urine retention. According to other research, diagnosing chronic urine retention (CUR) is typically more challenging. It is generally related to higher levels of postvoid residual urine (PVR) (42), and our data imply that PVR before biopsy is a separate risk factor for AUR following TP. Ultimately, our research revealed that AUR patients’ BMIs were greater than non-AUR patients. This appears to be a discovery. No study has been done on how BMI affects AUR following TP biopsy. BMI is a little-studied topic, and while we show some intriguing findings, further research is needed to draw firm conclusions.

Existing studies mostly focus on complications after transrectal puncture (TR), there are very few studies on AUR after TP, and most of them are univariate analysis (such as only focusing on prostate volume or number of puncture needles); a few studies involving TP do not establish prediction models and do not include machine learning algorithms to optimize variable screening (29, 43, 44). As Francesca Kum et al. found factors for urinary retention after transperineal template biopsy of the prostate, but only a univariate association, no predictive model was constructed (14). Sabri Cavkaytar et al. (45) used logistic regression to analyze risk factors for postpartum urinary retention, but given the high dimensionality of the data, no further dimensionality reduction was performed using machine learning algorithms. Samuel L Malnik et al. used machine learning to develop a postoperative predictive model for urinary retention after lumbar spine surgery, but only LASSO regression models were used, but often this was not able to eliminate random fluctuations in factor interactions (46). Moreover, the AUC of our two machine learning algorithms was more significant than that of Ding Xuexuan et al., who identified the core genes of asthma using five machine learning algorithms (21).

The clinical nomogram of this study integrates biomarkers and clinical characteristics, includes factors such as constipation and BMI in the prediction model of AUR after TP for the first time, and improves the rigor of variable screening through a two-feature selection algorithm to provide a personalized assessment of whether to continue Foley catheter or take additional safety measures to prevent AUR in TP patients. The nomogram constructed in this study is based on six easily accessible clinical measures (body mass index, prostate volume, history of diabetes, constipation, International Prostate Symptom Score, preoperative residual urine volume) that can be directly integrated into the preoperative assessment process of transperineal prostate biopsy (TP). Clinicians achieve precise stratified management by rapidly calculating the patient ‘s acute urinary retention (AUR) risk score: for example, high-risk patients (score > 70%): prophylactic use of alpha-blockers to relax the bladder neck preoperatively, or prolonged monitoring to 72 hours postoperatively, timely detection of signs of urinary retention, and reduction in the risk of complications such as emergency catheterization rate and overdistension of the bladder; for low-risk patients (score < 30%): use a standardized discharge regimen to reduce unnecessary inpatient observation and reduce medical costs while ensuring safety.

The current study has some limitations. It is a single-center cross-sectional study with a limited sample size that may introduce selection bias. In addition, we only performed internal validation on nomogram models, and subsequent studies also required external validation. Furthermore, potential influencing factors such as the number of puncture needles and operator experience were not included, and subsequent studies could expand the range of variables. Subsequent studies can be conducted in a large sample, multicenter, prospective study to find more risk factors for AUR complications in TP patients so that relevant measures can be taken early to avoid repeated catheterization and further hospitalization and improve patient discomfort.