Enhancing Fairness in Diabetes Prediction Systems through Smart User Interface Design

Artificial intelligence (AI) in chronic disease prediction often exhibits algorithmic biases, hindering equitable healthcare delivery. This study aims to develop and evaluate a Smart User Interface (Smart UI) framework that enhances fairness in diabetes prediction systems by operationalizing fairness at the human-computer interaction level, a dimension frequently overlooked in AI fairness research. We employed a nine-metric fairness evaluation framework across four demographically diverse diabetes datasets (Kaggle, Pima Indian, Azure Open, CDC Health Indicators). The Smart UI integrates contextual adjustment tools, dynamic visualizations, real-time alerts, and transparent reporting, combining structured EHR data, wearable sensor inputs, and unstructured clinical notes via natural language processing. The framework was evaluated on a clinical dataset to assess fairness and performance improvements. The Smart UI significantly reduced disparities: for age, the equal opportunity difference (EOD) improved from 0.35 to 0.25, with accuracy rising from 90.52% to 91.83%; for BMI, EOD decreased from 0.56 to 0.38, with the F1-score increasing from 83.89% to 86.37%. These outcomes highlight the framework’s ability to enhance fairness without altering underlying algorithms. While the Smart UI demonstrates promise as a model-agnostic, scalable solution for equitable AI deployment, challenges such as data privacy, usability, and real-time processing persist. The framework’s reliance on diverse data sources and user-centered design underscores its potential, though validation in broader clinical settings is needed. The Smart UI offers a replicable blueprint for embedding fairness in healthcare AI through interface design. Future research should focus on multicenter trials and applications to other chronic diseases to advance inclusive digital health solutions.