In a recent study titled “Fairness of Machine Learning Algorithms for Predicting Foregone Preventive Dental Care for Adults,” researchers investigated the effectiveness of machine learning (ML) in predicting foregone preventive dental care among adults and assessed the fairness of these predictive models across various sociodemographic subgroups. The study was based on a comprehensive analysis of data from the US Medical Expenditure Panel Survey (MEPS) conducted from 2016 to 2019 and involved 32,234 adult participants.
Key Findings:
- High Predictive Performance: The study found that tree-based ensemble prediction models, a type of machine learning algorithm, demonstrated high overall predictive performance with an area under the receiver operating characteristic curve (AUC) of 0.84, indicating the models’ ability to accurately predict foregone preventive dental care.
- Fairness Challenges: However, the study also revealed that the ML models’ performance varied across different sociodemographic subgroups. The models performed less effectively for individuals from racial or ethnic minority groups, low-income individuals, and younger adults. Specifically, the models showed lower precision in discriminating the outcomes for these underrepresented minority groups.
- Importance of Fairness Evaluation: The study emphasized the importance of evaluating algorithmic fairness during the development and testing of ML models to prevent exacerbating existing biases in healthcare.
Background:
Dental diseases affect a significant portion of the global population, with most oral health conditions being preventable through early identification and treatment. Routine access to preventive dental care can prevent acute dental problems and improve overall health. However, a substantial number of adults, particularly those from disadvantaged socioeconomic and racial or ethnic minority groups, still do not receive adequate preventive dental care due to various barriers, including cost, transportation, and limited access to dental insurance.
The Role of Machine Learning:
Machine learning approaches have the potential to improve healthcare delivery and identify individuals at risk of foregoing preventive dental care, directing resources and prevention efforts toward high-risk populations. ML algorithms use statistical modeling to predict future outcomes based on existing data. However, if these algorithms are trained with biased data, they may perpetuate disparities in healthcare access, which raises ethical and fairness concerns.
Fairness and Bias Concerns:
The study pointed out that the ML algorithms used in the research demonstrated bias against underrepresented sociodemographic groups, including Asian, Black, Hispanic, low-income, and younger adults. The potential deployment of these discriminatory models in practice could further exacerbate existing inequalities in healthcare outcomes.
Recommendations and Future Research:
The study suggests the need for standard benchmarks for evaluating fairness, mitigation, and reporting in ML models. Identifying and addressing biases in healthcare algorithms is crucial to ensuring equitable access to care and improving health outcomes for all patients.
This research highlights the complex challenges in developing ML models for healthcare and emphasizes the importance of fairness, transparency, and ethical considerations in the development and deployment of these technologies.