Development and validation of a prediction model for the diagnosis of breast cancer based on clinical and ultrasonic features
Authors:
Open Access
- 5 June 2023
- journal article
- research article
- Published by AME Publishing Company in Gland Surgery
- Vol. 12 (6) , 736-748
Abstract
Background: The aim of this study was to develop a simple and effective prediction model for calculating the probability of breast cancer by selecting clinical and sonographic features associated with breast cancer. Methods: A total of 402 lesions from 304 adult females from the ultrasound department of of PLA General Hospital from March 1st, 2020 to April 1st, 2021, were prospectively collected as the development group. The validation group included 121 lesions from 98 patients in our physical examination center from April 1st, 2021 to March 1st, 2022. Least absolute shrinkage and selection operator (LASSO) was applied to select clinical and ultrasonic variables, and R language was applied to build a web version of the interactive dynamic column line graph. The prediction model was validated by the validation group and the Breast Imaging Reporting and Data System (BI-RADS) categories. Calibration, differentiation and effectiveness were evaluated by R2, receiver operating characteristic (ROC) and decision curve analysis (DCA), respectively. Results: One hundred and seventy-nine malignant lesions and 223 benign lesions were included in the development group after exclusion and follow-up, whereas 62 malignant lesions and 59 benign lesions were enrolled in the validation group. Age, bloody nipple discharge, irregular shape, irregular border, heterogeneous echo, microcalcification, attenuation effects, decreased echo in surrounding tissues, lesions in ducts, abnormal lymph node morphology, nourishing vessel and nourishing vessel’s resistance index (RI) greater than 0.70 were selected as independent risk factors. There was no significant difference in the area under the curve (AUC) of the development group between the prediction model and the BI-RADS category (0.959 vs. 0.953, P>0.05), and so as the validation group (0.952 vs. 0.932, P>0.05). For the prediction model, R2 of the development and validation group was 0.78 and 0.72. The DCA showed that the net benefits (NB) of the development group were higher than that of the validation group (0–100% vs. 0–90%). Conclusions: A prediction model was developed with the clinical and ultrasonic features for the precise and intuitive probability of breast cancer. This could provide a reliable reference for further examination.Keywords
This publication has 32 references indexed in Scilit:
- Breast cancer risk models: a comprehensive overview of existing models, validation, and clinical applicationsBreast Cancer Research and Treatment, 2017
- Survival Prediction and Feature Selection in Patients with Breast Cancer Using Support Vector RegressionComputational and Mathematical Methods in Medicine, 2016
- Breast cancer in ChinaThe Lancet Oncology, 2014
- Controversies in Breast Cancer Screening for Women Aged 40-49 YearsJapanese Journal of Clinical Oncology, 2014
- BOADICEA breast cancer risk prediction model: updates to cancer incidences, tumour pathology and web interfaceBritish Journal of Cancer, 2013
- Risk prediction models of breast cancer: a systematic review of model performancesBreast Cancer Research and Treatment, 2011
- Breast Cancer Screening and Problem Solving Using Mammography, Ultrasound, and Magnetic Resonance ImagingUltrasound Quarterly, 2011
- Assessing the Performance of Prediction ModelsEpidemiology, 2010
- Breast cancer risk-assessment modelsBreast Cancer Research, 2007
- Limitations of the Gail Model in the Specialized Breast Cancer Risk Assessment ClinicThe Breast Journal, 2002