Development and internal validation of a multivariable prediction model for Gleason score upgrading after radical prostatectomy in patients with biopsy ISUP Grade Group 1 prostate cancer

Abstract

Background

Biopsy-confirmed International Society of Urological Pathology (ISUP) Grade Group 1 prostate cancer (PCa) is often considered suitable for active surveillance (AS). However, postoperative Gleason score upgrading (GSU) remains a major source of uncertainty in selected patients who proceed to radical prostatectomy (RP). This study aimed to develop and internally validate a multivariable model for predicting postoperative GSU in men with biopsy-confirmed Gleason score 3+3=6 disease who underwent RP.

Methods

We retrospectively reviewed patients who underwent transperineal prostate biopsy followed by RP at Northern Jiangsu People’s Hospital Affiliated to Yangzhou University between October 2022 and May 2025. Patients with biopsy Gleason score 3+3=6 and no preoperative evidence of metastasis were included; patients with prior androgen-deprivation therapy or chemotherapy, incomplete key data, coexisting malignancy, or active inflammatory/infectious disease were excluded. Candidate predictors included demographic, laboratory, imaging, pathological, and clinical variables. The reference standard for GSU was final RP pathology, with GSU defined as ISUP Grade Group 2 or higher. Patients were randomly divided into training and validation cohorts at a 7:3 ratio. Variables associated with GSU in preliminary analysis and considered clinically plausible were entered into least absolute shrinkage and selection operator (LASSO) regression, followed by multivariable logistic regression for effect estimation and nomogram construction. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration plots, the Hosmer-Lemeshow test, decision curve analysis (DCA), threshold-based classification metrics, and bootstrap internal validation.

Results

Among 312 patients, 143 (45.9%) had postoperative GSU. The median age was 72 years, and the median prostate-specific antigen (PSA) level was 9.00 ng/mL. In multivariable analysis, age [odds ratio (OR) =1.06, 95% confidence interval (CI): 1.02–1.11], percentage of positive biopsy cores (OR =1.05, 95% CI: 1.03–1.08), and Prostate Imaging Reporting and Data System (PI-RADS) score ≥4 (OR =2.26, 95% CI: 1.22–4.21) were independently associated with GSU. The nomogram showed AUCs of 0.78 (95% CI: 0.72–0.84) in the training cohort and 0.73 (95% CI: 0.63–0.83) in the validation cohort. At the Youden-index cutoff, sensitivity/specificity were 0.87/0.53 in the training cohort and 0.90/0.40 in the validation cohort, with acceptable calibration and net clinical benefit across clinically relevant threshold ranges.

Conclusions

This internally validated model provides a quantitative adjunct for estimating the probability of postoperative GSU in biopsy ISUP Grade Group 1 patients undergoing RP. Because the cohort was derived from a single center and lacked external validation, the model should be interpreted as exploratory and requires prospective multicenter validation before routine clinical use.

Highlight box.

Key findings

• Age, percentage of positive biopsy cores, and Prostate Imaging Reporting and Data System (PI-RADS) score ≥4 were independently associated with postoperative Gleason score upgrading in patients with biopsy International Society of Urological Pathology (ISUP) Grade Group 1 disease who underwent radical prostatectomy.

• The internally validated nomogram showed moderate discrimination and acceptable calibration in the present single-center cohort.

What is known and what is new?

• Active surveillance is standard management for many patients with low-risk prostate cancer, but biopsy undergrading remains a persistent source of uncertainty.

• This study focuses on a biopsy Grade Group 1 surgical cohort, including men with variable active surveillance (AS) suitability and clinical concern, and integrates clinical, pathological, and magnetic resonance imaging (MRI)-derived variables into a transparent prediction model.

What is the implication, and what should change now?

• The model may help clinicians discuss individualized upgrading risk with patients, but it should not replace guideline-based AS eligibility assessment or multidisciplinary clinical judgment. External validation is required before clinical implementation.

Introduction

Prostate cancer (PCa) remains one of the most frequently diagnosed malignancies among men worldwide. Global cancer statistics estimated more than 1.46 million new PCa cases in 2022, underscoring the importance of accurate diagnostic and risk-stratification pathways (1). Contemporary management relies on the combined interpretation of prostate-specific antigen (PSA), clinical stage, multiparametric magnetic resonance imaging (mpMRI), biopsy findings, and patient preference (2). Among these factors, pathological grade remains central because the Gleason score and International Society of Urological Pathology (ISUP) Grade Group classification are closely associated with tumor aggressiveness and treatment selection (3).

Patients with biopsy Gleason score 3+3=6, corresponding to ISUP Grade Group 1, are often considered candidates for active surveillance (AS), particularly when PSA density, disease volume, clinical stage, imaging findings, and life expectancy are favorable (2,4). However, AS decisions remain challenging because biopsy grade can underestimate final pathology. In addition to sampling limitations, PCa is frequently multifocal and spatially heterogeneous. Small foci of Gleason pattern 4 may be missed even when systematic and targeted biopsy are performed. Recent AS and surgical series continue to identify grade reclassification or upgrading as a key driver of treatment conversion and patient anxiety (5,6).

Several studies have investigated predictors of upgrading in low-risk or biopsy Grade Group 1 PCa. Age, PSA density, tumor extent in biopsy cores, positive targeted biopsy, and suspicious MRI findings have all been implicated (7-9). For example, Prostate Imaging Reporting and Data System (PI-RADS) 4–5 lesions have been associated with a markedly higher probability of ISUP upgrading after radical prostatectomy (RP) among men initially diagnosed with ISUP Grade Group 1 disease (8). Other recent models have incorporated inflammatory markers, clinical parameters, or imaging features to improve upgrading prediction (10). Established tools such as the Partin Tables and Memorial Sloan Kettering Cancer Center (MSKCC) nomograms are useful for staging and postoperative outcome counseling, but they were not designed specifically to quantify the probability of upgrading from biopsy Grade Group 1 to higher-grade disease in the subgroup of men who choose RP despite an initial low-grade biopsy (11-13).

The present study was therefore designed to address a specific clinical scenario rather than to redefine AS eligibility. We retrospectively evaluated men with biopsy-confirmed ISUP Grade Group 1 PCa who underwent RP at Northern Jiangsu People's Hospital Affiliated to Yangzhou University after multidisciplinary counseling and shared decision-making. Some patients had features that may make strict AS eligibility less certain, such as higher positive-core burden, elevated PSA density, or PI-RADS 4–5 lesions. We aimed to develop and internally validate a transparent multivariable model for estimating the risk of postoperative Gleason score upgrading (GSU), while explicitly acknowledging that external validation is required before clinical implementation. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0213/rc).

Methods

Study design, setting, and participants

This retrospective prediction-model study was conducted at Northern Jiangsu People’s Hospital Affiliated to Yangzhou University, a tertiary referral and teaching hospital in Jiangsu Province, China. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of Northern Jiangsu People’s Hospital Affiliated to Yangzhou University (approval No. 2025ky195), and the requirement for informed consent was waived because of the retrospective design.

We screened consecutive patients diagnosed with PCa who underwent transperineal prostate biopsy followed by RP between October 2022 and May 2025. The inclusion criteria were: (I) biopsy-confirmed Gleason score 3+3=6 (ISUP Grade Group 1); (II) transperineal prostate biopsy performed at Northern Jiangsu People's Hospital Affiliated to Yangzhou University before RP; (III) RP performed at the same institution; (IV) no preoperative evidence of metastasis on available imaging; and (V) no prior androgen-deprivation therapy (ADT), chemotherapy, or radiotherapy. The exclusion criteria were: (I) incomplete key clinical, laboratory, pathological, or imaging data; (II) concomitant pelvic floor surgery during biopsy or RP; (III) coexisting malignancy; (IV) autoimmune or chronic inflammatory disease; and (V) acute or chronic infection before surgery.

AS eligibility was not used as an inclusion criterion. Although all patients had biopsy Grade Group 1 disease, the cohort included men with variable disease volume, PSA density, MRI findings, and patient preference. Patients underwent counseling regarding AS and definitive treatment, and the decision to undergo RP was made through shared decision-making. Therefore, the study population should be interpreted as a biopsy Grade Group 1 surgical cohort rather than a strictly AS-eligible cohort.

Regarding staging, computed tomography (CT) was reviewed when performed as part of institutional preoperative evaluation or when clinical concern existed. CT was not intended to imply that all biopsy Grade Group 1 patients require metastatic staging. Bone scan was not routinely performed in asymptomatic biopsy Grade Group 1 patients because contemporary guidelines do not recommend additional staging imaging for low-risk and favorable intermediate-risk localized PCa, while bone scan or cross-sectional staging is generally reserved for unfavorable intermediate-risk, high-risk, locally advanced, symptomatic, or otherwise clinically suspicious cases (2).

Candidate predictors and data collection

Perioperative data were obtained from telephone follow-up, outpatient records, and the electronic medical record system. Laboratory tests were performed after 8–12 hours of fasting within 1–2 days before surgery. Candidate predictors were selected before modeling through multidisciplinary discussion among urologists, radiologists, pathologists, and clinical researchers and were based on prior literature and clinical availability.

Candidate predictors included demographic variables (age and body mass index), laboratory variables (PSA and PSA density), radiological variables (prostate volume and PI-RADS score on mpMRI), biopsy pathology variables (number of positive biopsy cores and percentage of positive biopsy cores), and clinical variables (clinical T stage, surgical method, and operative route). Percentage of positive biopsy cores was calculated as the number of cancer-positive cores divided by the total number of biopsy cores. Cancer length and percentage involvement per individual core were not consistently available in structured reports during the entire study period and therefore could not be included reliably; this limitation is addressed in the Discussion.

MRI acquisition and reporting

MRI was performed before prostate biopsy using a 1.5-T scanner equipped with a pelvic phased-array coil, in accordance with the recommendations of the European Society of Urogenital Radiology and the PI-RADS framework (14,15). MRI examinations were interpreted by a board-certified radiologist experienced in prostate imaging, who routinely evaluates more than 100 prostate MRI examinations annually. PI-RADS was dichotomized as <4 versus ≥4 for modeling because PI-RADS 4 and 5 lesions indicate a high or very high likelihood of clinically significant cancer.

Transperineal prostate biopsy and RP

All biopsies were performed via a transperineal approach by a single experienced urologist. Under transrectal ultrasound guidance, local infiltration anesthesia was administered to the perineal region and deeper periprostatic tissue. Needles were inserted at the 1, 3, 5, 7, 9, and 11 o’clock positions relative to the prostate, and local anesthetic was injected around the prostatic neurovascular bundles as previously described (16). The prostate was divided into 11 anatomical regions, and 1–3 biopsy cores were obtained from each region. MRI-ultrasound fusion-targeted biopsy was performed for MRI-visible lesions.

Among the 312 included patients, 125 (40.1%) underwent laparoscopic RP and 187 (59.9%) underwent robot-assisted radical prostatectomy (RARP). The surgical approach was selected after discussion among the surgeon, patient, and family. All RARP procedures were performed by the same senior urologist using a fourth-generation Da Vinci surgical system.

Reference standard and outcome definition

The reference standard for GSU was the final RP specimen. Biopsy and prostatectomy specimens were independently reviewed by a board-certified urological pathologist with more than 10 years of experience, according to ISUP recommendations (17). The pathologist was blinded to the statistical model and study hypothesis during outcome classification. Complete blinding to all clinical information was not feasible in routine retrospective clinical practice because pathology review was performed as part of standard care; this is acknowledged as a limitation.

All enrolled patients had biopsy-confirmed Gleason score 3+3=6 (ISUP Grade Group 1). GSU was defined as a final RP Gleason score of ≥3+4, corresponding to ISUP Grade Group 2 or higher.

Sample size and missing data

Because this was a retrospective study including all eligible patients during the study period, a formal a priori sample size calculation was not performed. To evaluate model stability, we assessed the events-per-variable (EPV) ratio. In the training cohort, 101 GSU events were available for seven candidate predictors entered into LASSO after univariate screening, corresponding to approximately 14.4 events per candidate predictor. The final three-predictor model had approximately 33.7 events per predictor in the training cohort. In the validation cohort, 42 GSU events were available, corresponding to 14.0 events per final predictor. These values were considered adequate for exploratory model development but do not remove the need for external validation (18).

Patients with missing values for key variables required for model development were excluded before analysis. Because the proportion of missing key data was limited and the study design was retrospective, complete-case analysis was used.

Statistical analysis and model development

Statistical analyses were performed using R version 4.4.1. Continuous variables were evaluated for normality using the Shapiro-Wilk test. Normally distributed variables are presented as mean ± standard deviation and compared using the independent-samples t-test. Non-normally distributed variables are presented as median with interquartile range and compared using the Mann-Whitney U test. Categorical variables are presented as frequencies and percentages and compared using Pearson’s Chi-squared test or Fisher’s exact test, as appropriate. Two-sided P values <0.05 were considered statistically significant.

Patients were randomly assigned to a training cohort and an internal validation cohort at a 7:3 ratio using a computer-generated random seed [1234]. Baseline balance between cohorts was assessed before model development. In the training cohort, univariate logistic regression was first performed as an exploratory screening step to reduce the dimensionality of candidate predictors in this retrospective cohort; it was not used as the final variable-selection method. Variables associated with GSU in this preliminary analysis and considered clinically plausible were then entered into least absolute shrinkage and selection operator (LASSO) regression using the glmnet package in R. The penalty parameter lambda was selected by 10-fold cross-validation using the lambda.min criterion. To address concerns regarding model instability, no stepwise logistic regression was used. Variables with non-zero LASSO coefficients were subsequently entered into a conventional multivariable logistic regression model to estimate odds ratios and 95% confidence intervals (CIs), and the independently associated predictors were used to construct the nomogram.

Model discrimination was evaluated using receiver operating characteristic (ROC) curve analysis and area under the receiver operating characteristic curve (AUC) with 95% CIs. Calibration was assessed using calibration plots and the Hosmer-Lemeshow goodness-of-fit test. Clinical utility was evaluated using decision curve analysis (DCA) (19). For descriptive classification-performance reporting, the optimal cutoff was determined using the Youden index, and accuracy, sensitivity, specificity, F1 score, and Brier score were calculated in the training and validation cohorts. The Youden-index cutoff and the corresponding sensitivity and specificity were interpreted descriptively and were not used to claim clinical adoption. Internal validation was performed using 1,000 bootstrap resamples.

Results

Patient characteristics and GSU status

A total of 674 patients who underwent transperineal saturated prostate biopsy and RP during the study period were screened. After excluding 362 patients, 312 eligible men with biopsy ISUP Grade Group 1 disease were included in the final analysis (Figure 1). The median age was 72 years, the median PSA level was 9.00 ng/mL, and the median prostate volume was 48.00 mL. GSU occurred in 143 patients (45.9%). Among upgraded cases, 108 (75.5%) were upgraded to ISUP Grade Group 2, 18 (12.6%) to Grade Group 3, 9 (6.3%) to Grade Group 4, and 8 (5.6%) to Grade Group 5.

Figure 1.

Flowchart of participant selection. ADT, androgen-deprivation therapy; GSU, Gleason score upgrading; RP, radical prostatectomy.

According to D’Amico risk classification, 144 patients (46.1%) were classified as low risk, 92 (29.5%) as intermediate risk, and 76 (24.4%) as high risk. This distribution further supports that the cohort represents a biopsy Grade Group 1 surgical population with heterogeneous AS suitability rather than a strictly low-volume AS cohort.

Construction of training and validation cohorts

The 312 eligible patients were randomly assigned to a training cohort (n=218) and an internal validation cohort (n=94). Baseline characteristics were comparable between the two cohorts (Table 1), supporting use of the training cohort for model development and the validation cohort for internal performance assessment.

Table 1. Comparison of baseline characteristics between the training and validation cohorts.

Variables	Total (n=312)	Validation (n=94)	Training (n=218)	P value
BMI, kg/m2	24.22±2.95	24.03±3.07	24.30±2.90	0.45
Age, years	72.00 (67.00, 76.00)	73.00 (68.25, 77.00)	71.00 (66.00, 76.00)	0.052
PSA, ng/mL	9.00 (6.51, 13.22)	9.14 (6.18, 14.34)	8.80 (6.60, 12.79)	0.76
Volume, mL	48.00 (36.30, 64.35)	46.80 (33.16, 65.00)	48.91 (36.40, 64.17)	0.65
PSA density, ng/mL2	0.18 (0.13, 0.30)	0.18 (0.13, 0.30)	0.18 (0.12, 0.30)	0.95
Number of positive biopsy cores	3.00 (2.00, 6.00)	3.00 (2.00, 5.00)	3.00 (2.00, 6.00)	0.18
Percentage of positive biopsy cores	14.29 (8.33, 26.14)	14.29 (8.33, 25.00)	14.72 (8.00, 29.13)	0.74
PI-RADS				0.80
<4	126 (40.38)	37 (39.36)	89 (40.83)
≥4	186 (59.62)	57 (60.64)	129 (59.17)
Clinical T stage				0.57
<T2b	185 (59.29)	58 (61.70)	127 (58.26)
≥T2b	127 (40.71)	36 (38.30)	91 (41.74)
Surgical method				0.67
RP	125 (40.06)	36 (38.30)	89 (40.83)
RARP	187 (59.94)	58 (61.70)	129 (59.17)
Operative route				0.95
Transperitoneal	132 (42.31)	40 (42.55)	92 (42.20)
Extraperitoneal	180 (57.69)	54 (57.45)	126 (57.80)

Data are presented as mean ± standard deviation, median (interquartile range), or n (%). AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; GSU, Gleason score upgrading; OR, odds ratio; PI-RADS, Prostate Imaging Reporting and Data System; PSA, prostate-specific antigen; RARP, robot-assisted radical prostatectomy; RP, radical prostatectomy; T, tumor.

Baseline characteristics according to GSU status in the training cohort

Within the training cohort, 117 patients were classified into the non-GSU group and 101 into the GSU group. Age, PSA, PSA density, number of positive biopsy cores, percentage of positive biopsy cores, PI-RADS score, and clinical T stage differed significantly between groups (Table 2).

Table 2. Comparison of baseline characteristics according to GSU status in the training cohort.

Variables	Total (n=218)	Non-GSU (n=117)	GSU (n=101)	P value
BMI, kg/m2	24.30±2.90	24.14±2.87	24.48±2.93	0.39
Age, years	71.00 (66.00, 76.00)	69.00 (65.00, 75.00)	73.00 (68.00, 76.00)	0.006
PSA, ng/mL	8.80 (6.60, 12.79)	7.95 (6.26, 10.90)	10.10 (7.40, 16.55)	<0.001
Volume, mL	48.91 (36.40, 64.17)	51.00 (39.00, 65.21)	44.80 (32.76, 62.71)	0.050
PSA density, ng/mL2	0.18 (0.12, 0.30)	0.15 (0.11, 0.22)	0.25 (0.15, 0.39)	<0.001
Number of positive biopsy cores	3.00 (2.00, 6.00)	2.00 (2.00, 4.00)	5.00 (2.00, 8.00)	<0.001
Percentage of positive biopsy cores	14.72 (8.00, 29.13)	10.53 (7.69, 18.18)	21.74 (11.76, 36.59)	<0.001
PI-RADS				<0.001
<4	89 (40.83)	60 (51.28)	29 (28.71)
≥4	129 (59.17)	57 (48.72)	72 (71.29)
Clinical T stage				0.001
<T2b	127 (58.26)	80 (68.38)	47 (46.53)
≥T2b	91 (41.74)	37 (31.62)	54 (53.47)
Surgical method				0.94
RP	89 (40.83)	48 (41.03)	41 (40.59)
RARP	129 (59.17)	69 (58.97)	60 (59.41)
Operative route				0.86
Transperitoneal	92 (42.20)	50 (42.74)	42 (41.58)
Extraperitoneal	126 (57.80)	67 (57.26)	59 (58.42)

Data are presented as mean ± standard deviation, median (interquartile range), or n (%). AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; GSU, Gleason score upgrading; OR, odds ratio; PI-RADS, Prostate Imaging Reporting and Data System; PSA, prostate-specific antigen; RARP, robot-assisted radical prostatectomy; RP, radical prostatectomy; T, tumor.

Regression analyses and predictor selection

In the preliminary univariate logistic regression analysis, age, PSA, PSA density, number of positive biopsy cores, percentage of positive biopsy cores, PI-RADS score, and clinical T stage were associated with GSU. These variables were then entered into LASSO regression for penalized predictor selection, and five variables with non-zero coefficients were retained (Figure 2). In the subsequent multivariable logistic regression model, age, percentage of positive biopsy cores, and PI-RADS score ≥4 remained independently associated with postoperative GSU (Table 3).

Figure 2.

LASSO regression for predictor selection. (A) Coefficient profile plot against the log(lambda) sequence. (B) Cross-validation plot showing the lambda value corresponding to the minimum mean cross-validated error and the one-standard-error criterion. LASSO, least absolute shrinkage and selection operator.

Table 3. Univariate and multivariate logistic regression analyses for GSU in the training cohort.

Variables	Univariate				Multivariate
	OR	95% CI	P value		OR	95% CI	P value
Age	1.06	1.02–1.10	0.006		1.06	1.02–1.11	0.006
PSA	1.04	1.01–1.08	0.01		–	–	–
PSA density	5.88	1.54–22.50	0.01		–	–	–
Number of positive biopsy cores	1.23	1.12–1.34	<0.001		–	–	–
Percentage of positive biopsy cores	1.06	1.04–1.08	<0.001		1.05	1.03–1.08	<0.001
PI-RADS
<4	Reference				Reference
≥4	2.61	1.49–4.59	<0.001		2.26	1.22–4.21	0.01
Clinical T stage
< T2b	Reference				–	–	–
≥ T2b	2.48	1.43–4.31	0.001		–	–	–

AUC, area under the receiver operating characteristic curve; CI, confidence interval; GSU, Gleason score upgrading; OR, odds ratio; PI-RADS, Prostate Imaging Reporting and Data System; PSA, prostate-specific antigen; T, tumor.

Nomogram construction and internal validation

A nomogram was constructed using the three independent predictors: age, percentage of positive biopsy cores, and PI-RADS score ≥4 (Figure 3). Each variable contributes points to a total score, which is then mapped to an estimated probability of postoperative GSU.

Figure 3.

Nomogram for predicting postoperative Gleason score upgrading. PI-RADS, Prostate Imaging Reporting and Data System.

The AUC was 0.78 (95% CI: 0.72–0.84) in the training cohort and 0.73 (95% CI: 0.63–0.83) in the validation cohort (Figure 4A). The optimal Youden-index cutoff was 0.34 in both cohorts. At this cutoff, the model showed high sensitivity in the training and validation cohorts (0.87 and 0.90, respectively), whereas specificity was modest (0.53 and 0.40, respectively). The detailed classification metrics, including accuracy, F1 score, and Brier score, are summarized in Table 4. These threshold-based metrics describe practical discrimination and should not be interpreted as a prespecified treatment threshold. DCA showed net clinical benefit over treat-all and treat-none strategies across threshold probabilities of 0.21–0.95 in the training cohort and 0.17–0.82 in the validation cohort (Figure 4B,4C). Calibration plots showed close agreement between predicted and observed GSU probabilities, and the Hosmer-Lemeshow test indicated acceptable fit in the training cohort (P=0.65) and validation cohort (P=0.45) (Figure 4D,4E). Bootstrap internal validation using 1,000 resamples supported acceptable calibration stability.

Figure 4.

Model performance. (A) ROC curves in the training and validation cohorts. (B) DCA curve in the training cohort. (C) DCA curve in the validation cohort. (D) Calibration plot in the training cohort. (E) Calibration plot in the validation cohort. AUC, area under the receiver operating characteristic curve; CI, confidence interval; DCA, decision curve analysis; ROC, receiver operating characteristic.

Table 4. Classification performance of the nomogram at the Youden-index cutoff.

Cohort	AUC (95% CI)	Acc. (95% CI)	Sens. (95% CI)	Spec. (95% CI)	F1 score	Brier	Cutoff
Training	0.78 (0.72–0.82)	0.69 (0.62–0.75)	0.87 (0.81–0.94)	0.53 (0.44–0.62)	0.72	0.20	0.34
Validation	0.73 (0.63–0.83)	0.63 (0.52–0.73)	0.90 (0.82–0.99)	0.40 (0.27–0.54)	0.68	0.23	0.34

The cutoff indicates the predicted-probability threshold selected by the Youden index. Acc., accuracy; AUC, area under the receiver operating characteristic curve; CI, confidence interval; Sens., sensitivity; Spec., specificity.

Discussion

In this retrospective single-center prediction-model study, GSU was observed in 45.9% of men with biopsy ISUP Grade Group 1 disease who underwent RP. Age, percentage of positive biopsy cores, and PI-RADS score ≥4 were independently associated with upgrading. The resulting nomogram demonstrated moderate discrimination, acceptable calibration, and potential clinical net benefit in internal validation. The Youden-index analysis showed high sensitivity but modest specificity, suggesting that the model may be more useful for risk discussion and for identifying patients who may warrant closer evaluation than for ruling in upgrading or directing definitive treatment on its own. These findings support the clinical value of integrating routinely available preoperative variables, while also reinforcing that the model remains exploratory until validated in independent cohorts.

The relatively high upgrading rate should be interpreted in light of patient selection. The study population was not a pure AS cohort. It included patients who underwent surgery after counseling and shared decision-making, and some had features that may reduce strict AS suitability, including higher biopsy tumor burden, elevated PSA density, clinical concern, or PI-RADS 4–5 lesions. This design reflects a practical clinical scenario: patients with biopsy Grade Group 1 disease who nevertheless proceed to definitive surgery because initial low-grade pathology is discordant with other clinical findings or with patient preference. At the same time, this design introduces selection bias and limits generalizability to very-low-risk, low-volume patients managed conservatively. The model should therefore be viewed as a risk-estimation adjunct for similar surgical biopsy Grade Group 1 populations rather than as an AS eligibility tool.

The importance of MRI in our model is biologically plausible. PCa is frequently multifocal, and a systematic biopsy can miss small or spatially restricted Gleason pattern 4 components. A PI-RADS score ≥4 indicates a high likelihood of clinically significant cancer and may therefore mark lesions in which the biopsy grade underrepresents the dominant pathology. Zhou et al. reported that MRI-target positivity, together with age and free/total PSA (f/tPSA), predicted upgrading in low-risk patients after transperineal biopsy and RP (7). Dekalo et al. similarly found that PI-RADS 4–5 lesions were associated with ISUP upgrading and adverse pathology in men with biopsy ISUP Grade Group 1 disease (8). Zhang et al. further showed that combining clinical variables with mpMRI findings improved prediction of Gleason score changes between biopsy and prostatectomy (9). These observations are consistent with earlier MRI-based studies showing that PI-RADS and periprostatic fat features may help identify upgrading risk among men with biopsy Gleason score 3+3 disease (20,21).

The percentage of positive biopsy cores also remained independently associated with upgrading. This variable captures biopsy tumor extent and is a practical surrogate for disease burden. In clinical counseling, a biopsy Grade Group 1 diagnosis with a high proportion of positive cores should not be interpreted in the same way as very-low-volume Grade Group 1 disease. The European Association of Urology (EAU) guideline emphasizes that both the number of cancer-positive cores and the extent of cancer involvement within each core should be reported because they are associated with tumor volume, pathological stage, surgical margin status, and recurrence-related outcomes (2). Our inability to include maximum cancer length or percentage involvement per individual core is therefore important. Recent work has also shown that biopsy core length can affect the discrepancy between biopsy and RP Gleason scores, underscoring the need for standardized core-level reporting in future models (22).

Age was another independent predictor. Prior studies have reported associations between older age and postoperative upgrading (23,24). Several mechanisms may contribute, including age-related changes in immune surveillance, stromal-tumor interactions, and the accumulation of biologically more aggressive subclones. Recent pathological evidence also suggests that adverse architectural patterns may have age-dependent prognostic implications after RP (25). Nevertheless, age alone should not drive treatment selection. Its effect is modest and may interact with PSA density, prostate volume, MRI findings, comorbidity, and life expectancy. In the present model, age should therefore be interpreted as one component of a multiparametric assessment rather than as an isolated decision criterion.

Not all clinically relevant variables were retained in the final model. PSA density, clinical tumor (T) stage, and prostate volume differed between groups or were clinically plausible, but they did not remain final predictors after penalized selection and multivariable modeling. This may reflect shared information with MRI suspicion and biopsy tumor burden in a relatively small single-center cohort. Smaller prostate volume has been associated with upgrading in prior studies and meta-analyses, possibly because benign enlargement can dilute PSA and increase sampling uncertainty (26). In our cohort, however, prostate volume did not add independent predictive information after the other variables were considered. This finding should not be interpreted as evidence that these variables are unimportant; rather, it emphasizes that variable selection in a moderate-sized retrospective model is sample-dependent and requires external confirmation.

To place the present nomogram in the context of existing models, it should be distinguished from both established PCa risk calculators and recent upgrading-focused studies. The Partin Tables and MSKCC nomograms remain widely used for preoperative staging or postoperative outcome counseling, but their primary outputs differ from the specific endpoint of upgrading from biopsy ISUP Grade Group 1 to Grade Group 2 or higher after RP (11-13). Recent studies have increasingly incorporated MRI findings, biopsy tumor burden, inflammatory indices, or dynamic AS variables to improve prediction of upgrading or grade reclassification (5,10,20). The main distinction of the present model is its focus on a biopsy ISUP Grade Group 1 surgical cohort and its use of routinely available preoperative clinical, pathological, and MRI-derived variables. We do not suggest that this model is superior to existing instruments. Its moderate discrimination, single-center derivation, and absence of external validation indicate that it should be regarded as a complementary exploratory risk-estimation tool requiring multicenter validation and recalibration before clinical implementation.

The performance pattern of the model also deserves careful interpretation. The validation AUC of 0.73 indicates moderate discrimination, not a level of accuracy sufficient for stand-alone treatment decisions. In addition, the Youden-index cutoff produced high sensitivity but modest specificity, meaning that the model may identify many patients at possible risk of upgrading but may also classify a considerable number of non-upgraded patients as higher risk. For this reason, the model is better suited to supporting shared decision-making, discussing uncertainty, and selecting patients for closer diagnostic review than to replacing guideline-based AS criteria. DCA provides complementary information because net benefit evaluates whether model-guided decisions could provide more benefit than treat-all or treat-none strategies across a range of clinically relevant thresholds (19).

Several limitations warrant emphasis. First, this was a single-center retrospective study, and the surgical cohort introduces selection bias. The model may overestimate upgrading risk when applied to the broader population of biopsy Grade Group 1 patients, particularly those meeting strict AS criteria. Second, although internal validation was performed using a separate validation cohort and bootstrap resampling, no external validation was available. Third, key variables such as prostate weight, MRI lesion volume, extracapsular extension suspicion, maximum cancer length, and percentage involvement per core were not consistently available and could not be included. Fourth, complete blinding of the pathologist to all clinical information was not feasible in routine clinical practice. Fifth, CT was reviewed when available as part of preoperative evaluation, but routine additional staging imaging is not recommended for all low-risk patients. Future work should prospectively validate the model in multicenter cohorts, incorporate standardized biopsy core-length metrics and quantitative MRI features, and evaluate whether recalibration is required in AS-eligible patients and different practice settings. More granular risk-stratification frameworks, such as the Cambridge Prognostic Group classification and EAU risk groups, may also help define clinically meaningful subgroups for future upgrading-specific models (27,28).

Conclusions

This study developed and internally validated a multivariable nomogram for estimating the risk of postoperative GSU in patients with biopsy-confirmed ISUP Grade Group 1 PCa who underwent RP. Age, percentage of positive biopsy cores, and PI-RADS score ≥4 were independently associated with upgrading. The model may support individualized risk discussion, but its moderate discrimination, single-center design, mixed AS suitability, and lack of external validation preclude routine clinical implementation at this stage. Prospective multicenter validation and incorporation of more granular biopsy and imaging variables are required.

Supplementary

The article’s supplementary files as

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 Medical Ethics Committee of Northern Jiangsu People’s Hospital Affiliated to Yangzhou University (approval No. 2025ky195), and the requirement for informed consent was waived because of the retrospective design.

Data Sharing Statement

Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0213/dss

tau-15-06-213-dss.pdf ^{(65.9KB, pdf)}

DOI: 10.21037/tau-2026-0213