Abstract
The management of hyperglycemia in hospitalized patients has a significant impact on both morbidity and mortality. This study used a large clinical database to predict the need for diabetic patients to be hospitalized, which could lead to improvements in patient safety. These predictions, however, may be vulnerable to health disparities caused by social determinants such as race, age, and gender. These biases must be removed early in the data collection process, before they enter the system and are reinforced by model predictions, resulting in biases in the model's decisions. In this paper, we propose a machine learning pipeline capable of making predictions as well as detecting and mitigating biases. This pipeline analyses clinical data, determines whether biases exist, removes them, and then make predictions. We demonstrate the classification accuracy and fairness in model predictions using experiments. The results show that when we mitigate biases early in a model, we get fairer predictions. We also find that as we get better fairness, we sacrifice a certain level of accuracy, which is also validated in the previous studies. We invite the research community to contribute to identifying additional factors that contribute to health disparities that can be addressed through this pipeline.
Abstract (translated)
URL
https://arxiv.org/abs/2206.06279