Diagnostic Prediction Model for De Novo Metastatic Breast Cancer

Breast Cancer, Metastatic Invasive Breast Cancer

de novo metastasis, breast cancer, predictive modelling

The goal of this observational study is to create a tool to estimate the risk of macroscopic distant tumor spread in women with newly diagnosed breast cancer. The main question it aims to answer is: • Can a model, which combines routine medical information collected at diagnosis, accurately predict if a patient has metastasis large enough to be found during systemic imaging? Researchers will review the past medical records of participants who were treated at a university hospital. Because this study looks at past data, participants will not be asked to do any new tasks, take new tests, or change their medical care.

BACKGROUND AND RATIONALE Baseline risk assessment traditionally relies on anatomical extent, while intrinsic tumour aggressiveness is established as a key determinant of distant spread. In addition, the systemic inflammatory response is increasingly recognized as a driver of tumour progression. The evaluation of these diverse and often conflicting factors complicates early clinical decision-making. OBJECTIVE The objective of this study is to develop and internally validate a multivariable diagnostic prediction model using routinely available baseline parameters to estimate the individualised probability of macroscopic distant metastasis among patients with newly diagnosed invasive breast cancer. STUDY DESIGN AND SETTING This is an investigator-initiated, single-centre, retrospective cohort study adhering to the TRIPOD+AI statement. The study is conducted at a university teaching hospital, where comprehensive baseline systemic staging is the institutional standard for all newly diagnosed breast cancer patients. PREDICTORS The model integrates routinely available baseline parameters: clinical tumor size, clinical lymphnode status, the Ki-67 proliferation marker, and a composite systemic inflammatory marker (Pan-immune-inflammation value). DATA COLLECTION AND QUALITY ASSURANCE Data extraction is performed independently by two multidisciplinary teams. Any discrepancies are resolved through a formal adjudication process by an expert panel not involved in data collection (senior oncologist, radiologist, pathologist, and surgeon). SAMPLE SIZE AND MISSING DATA Based on the criteria proposed by Riley et al., the study requires a minimum of 925 participants and 93 events to accommodate maximum model complexity. Missing data will be handled using complete-case analysis if the rate is \<5%, or multiple imputation if \>5%. STATISTICAL ANALYSIS AND MODEL DEVELOPMENT For the multivariable logistic regression model, all pre-specified predictors are entered simultaneously. Continuous variables are retained in their continuous form. Non-linear relationships modeled using restricted cubic splines. Model performance is evaluated via discrimination and calibration metrics, while clinical utility is assessed through decision curve analysis. Internal validation is conducted using bootstrapping. To correct for model optimism, a global shrinkage factor based on the bootstrap calibration slope is applied. Instability plots are used to illustrate the stability of predictions, calibration, and net benefit. PATIENT AND PUBLIC INVOLVEMENT (PPI) To initiate Patient and Public Involvement, an inaugural patient engagement event will be held to discuss the clinical acceptability of a diagnostic prediction model and its potential role in shared decision-making. This event will serve as the foundation to establish a voluntary Patient Advisory Group (PAG). In subsequent project stages, the newly formed PAG will collaborate to co-produce a Plain Language Summary and explore potential pathways for the tool's future clinical application.

No related papers found yet.

None listed

Hungary