Development and validation of a clinical-bimodal nomogram to predict the malignancy of pathologic nipple discharge

Pathologic nipple discharge (PND) is a prevalent presentation in breast clinics, yet its accurate diagnosis remains challenging. There is still a lack of practical, non-invasive models integrating specific imaging features with clinical data to improve individualized malignancy risk assessment. This study aimed to develop a nomogram with clinical, mammography (MG) and ultrasonography (US) parameters to identify the malignancy of PND.

A total of 377 patients diagnosed with PND from January 2013 to December 2023 at two hospitals were recruited retrospectively in this study. Patients (n=280) from hospital 1 formed the training cohort. Patients (n=97) from hospital 2 formed an external validation cohort. Clinical-pathologic and imaging data were collected. Logistic regression analyses were used to construct predictive models based on MG alone or a combination of MG and US (MG-US), using the respective independent parameters. The discriminatory and calibration ability of both models were evaluated by the area under the receiver operating characteristic curve (AUC) and the calibration curve. The optimal model was selected by comparing the AUCs with De-long test. The internal validation was conducted with bootstrap method (resampling 1,000 times). The decision curve analysis (DCA) was used to determine its clinical utility.

Based on multivariate logistic regression analysis, the independent malignancy risk factors of PND were age, color of discharge, palpable mass, suspicious calcifications on MG, suspicious mass and intraductal mass/debris on US. The discriminatory ability of the MG-US model was significantly better than that of the MG model in training cohort, with an AUC of 0.851 [95% confidence interval (CI): 0.802-0.901] compared to 0.789 (95% CI: 0.729-0.850) (P=0.014). DCA also suggested that the clinical value of the MG-US model was better than that of MG model. The superior model was utilized to develop the clinical-bimodal nomogram, yielding an AUC of 0.879 (95% CI: 0.808-0.951) in the validation cohort.

A clinical-bimodal nomogram including MG-US parameters was established to predict the probability of malignancy in patients with PND, which may enable clinicians to identify the patients with a low risk of malignancy and formulate the individualized management strategies.