A recent large-scale retrospective study conducted by researchers at the Fleury Group in Brazil has shed light on the potential of using complete blood count (CBC) as a risk stratification tool for breast cancer using machine learning. The study, approved by the Fleury Group’s Research Ethics Committee and conducted in compliance with Brazilian legislation and data protection laws, analyzed CBC test results from nearly 400,000 women aged 40-70 who were screened for breast cancer between 2004 and 2022.
Here are some key findings from the study:
– Researchers collected CBC test results from 396,848 women across eight Brazilian states.
– The case group included women diagnosed with breast cancer or highly suspected of having it, while the control group consisted of women with negative imaging results.
– Various histologic subtypes of breast cancer were identified, with ductal carcinoma being the most common among invasive cases.
– The study utilized two machine learning models, ridge regression, and LightGBM, to analyze CBC biomarkers and derived ratios for predicting breast cancer risk.
– The models were trained and evaluated using the Area Under the Curve (AUC) metric, with feature selection and hyperparameter tuning conducted to optimize performance.
The study also employed the SHapley Additive exPlanations (SHAP) approach for explainability analysis and categorized the population into four risk groups based on the models’ probability outputs.
Overall, this groundbreaking study provides valuable insights into the potential of utilizing CBC as a risk stratification tool for breast cancer, highlighting the importance of leveraging machine learning algorithms for predictive modeling in healthcare.
Source: [Scientific Reports](insert link)
