An interpretable nomogram for positive surgical margin risk after laparoscopic and robot-assisted radical prostatectomy: a single-centre development and internal validation study

Abstract

Positive surgical margins (PSM) after radical prostatectomy are associated with biochemical recurrence, although margin status reflects a combination of tumour-related, anatomical, and operative factors. We aimed to derive and internally validate an interpretable nomogram that accommodates non-linear predictor effects and estimates postoperative PSM probability after laparoscopic radical prostatectomy (LRP) or robot-assisted radical prostatectomy (RARP). We retrospectively analysed 376 patients with prostate cancer who underwent LRP or RARP and had a documented postoperative margin status. The candidate set included demographic characteristics, indices of preoperative tumour burden, operative variables, comorbid conditions, and postoperative pathological findings. Least absolute shrinkage and selection operator (LASSO) regression was used for initial predictor reduction. Correlation matrices, variance inflation factors, and sensitivity analyses across Top-N predictor sets were used to examine redundancy, stability, and model size. Restricted cubic splines (RCS) were applied to continuous predictors to assess departures from linearity. The selected predictors were entered into multivariable logistic regression. Model performance was evaluated by receiver operating characteristic analysis, calibration plots, Brier scores, and decision curve analysis. A total of 376 patients with prostate cancer were included, of whom 155 (41.2%) had positive surgical margins (PSMs). Multiple margin sites were the most common pattern of PSM involvement (n = 91), followed by the apex (n = 35), base (n = 15), and lateral or posterolateral margins (n = 10). The PSM rate was 33.9% (83/245) in patients with pT2 disease and 68.8% (66/96) in those with pT3 disease. The corresponding PSM rates were 34.8% (49/141) for laparoscopic radical prostatectomy (LRP) and 45.1% (106/235) for robot-assisted radical prostatectomy (RARP). LASSO selection, feature-subset optimization, and subsequent multivariable modeling indicated that perineural invasion, the proportion of positive biopsy cores, seminal vesicle invasion, surgical approach, clinical T stage, lymphovascular invasion, age, and preoperative PSA-related variables contributed to PSM risk prediction. In the final multivariable model, perineural invasion (OR = 2.609, 95% CI: 1.328–5.124; P < 0.05) and the proportion of positive biopsy cores (OR = 1.020, 95% CI: 1.005–1.035; P < 0.05) were significantly associated with PSM. Restricted cubic spline analysis suggested a possible nonlinear association between preoperative log(PSA + 1) and the risk of PSM. The model achieved areas under the receiver operating characteristic curve of 0.801 in the training set and 0.824 in the validation set. The corresponding Brier scores were 0.1778 and 0.1697, respectively. Calibration curves showed overall agreement between predicted probabilities and observed outcomes. Decision curve analysis further indicated that the model provided a positive clinical net benefit across a range of threshold probabilities. The internally validated nomogram separated patients with and without PSM reasonably well, produced acceptably calibrated risk estimates, and may assist postoperative risk stratification after LRP or RARP. Independent validation is necessary before clinical implementation.