A retrospective study used machine learning to predict treatment response and survival rates in breast cancer patients from an underserved, inner-city population. The study examined the impact of tumor subtypes and socioeconomic factors on outcomes.
- Machine learning models reveal no association between pathological complete response and patient demographics but highlight the significance of tumor subtypes and imaging characteristics.
- Tumor subtypes and specific imaging features are key predictors of pathological complete response and overall survival, with socioeconomic status playing a lesser role.
- The study underscores the potential of machine learning to improve personalized treatment strategies for breast cancer, particularly in diverse populations.
Breast cancer’s complexity and its treatment outcomes can be significantly influenced by both biological and socioeconomic factors. In an effort to bridge the gap in predictive accuracy for patients undergoing neoadjuvant chemotherapy, a study published in the journal Breast Cancer Research employed four machine learning models to forecast pathological complete response (pCR) and overall survival (OS) in a demographically diverse cohort from an underserved inner-city area. This approach not only aims to personalize treatment plans but also to understand the nuanced effects of socioeconomic disparities on health outcomes.
Harnessing Machine Learning for Better Predictions
The study analyzed data from 475 breast cancer patients, focusing on demographics, tumor characteristics, and socioeconomic status, using logistic regression, neural networks, random forest, and gradient boosted regression models for predictions. Interestingly, pCR was not linked to patient demographics but was significantly influenced by tumor subtypes and imaging features, such as tumor size and background parenchymal enhancement (BPE). This indicates that biological markers play a more critical role in predicting treatment response than previously acknowledged demographic factors.
The Impact of Tumor Subtypes and Imaging on Treatment Outcomes
Tumor subtypes emerged as a pivotal factor, with ER−/HER2+ showing the highest rate of pCR, (56.5%) followed by triple-negative (31%), and ER+/HER2+ (8.5%) subtypes. Machine learning models effectively used tumor subtype information alongside imaging data to predict outcomes, demonstrating the importance of a well-rounded approach that includes both biological and technological data inputs.
Socioeconomic Factors and Survival Rates
While the study found associations between insurance status and OS, indicating that socioeconomic factors do play a role in survival probabilities, these were less significant than the biological markers. This highlights a critical area of focus for improving healthcare delivery and outcomes in underserved populations, emphasizing the need for equitable healthcare access and treatment options.
Implications for Clinical Practice
This research underscores the utility of machine learning in refining the prediction of treatment responses and survival rates in breast cancer patients, emphasizing the importance of tumor biology over demographic factors. For healthcare providers, these findings suggest a move toward more individualized treatment plans that consider the unique tumor characteristics of each patient.
Additionally, the study highlights the ongoing need to address healthcare disparities to improve outcomes for all patients, regardless of socioeconomic status. By integrating complex data analysis techniques with a comprehensive understanding of patient demographics and tumor biology, clinicians can better navigate the complexities of breast cancer treatment, leading to more effective and personalized care strategies.
Source:
Dell’Aquila, K., Vadlamani, A., Maldjian, T., Fineberg, S., Eligulashvili, A., Chung, J., Adam, R., Hodges, L., Hou, W., Makower, D., & Duong, T. Q. (2024). Machine learning prediction of pathological complete response and overall survival of breast cancer patients in an underserved inner-city population. Breast Cancer Research, 26(1). https://doi.org/10.1186/s13058-023-01762-w
