Wearable-based Fair and Accurate Pain Assessment Using Multi-Attribute Fairness Loss in Convolutional Neural Networks

TL;DR

The paper tackles biases in AI-driven pain assessment from wearable and EHR data, which risk differential treatment across protected attributes. It introduces Multi-attribute Fairness Loss (MAFL) integrated into a 1D CNN (MAFL-CNN) to enforce fairness across five protected attributes while predicting pain status from Fitbit-derived time-series features and EHR ground truths. Using the NIH All-of-Us dataset (868 individuals over ~1500 days) with ground-truth pain progression labels, the approach extracts >60 time-series features and evaluates bias with five fairness metrics, focusing on Statistical Parity Difference and Disparate Impact for mitigation. In comprehensive experiments, MAFL-CNN outperforms standard baselines and other AIF360 mitigations, achieving 75–85% accuracy with improved fairness across age, gender, race, ethnicity, and cognitive ability, demonstrating a practical path toward fair wearable-based pain assessment. The work highlights the feasibility and importance of fairness-aware clinical decision support and motivates extending the approach to other datasets and architectures.

Abstract

The integration of diverse health data, such as IoT (Internet of Things), EHR (Electronic Health Record), and clinical surveys, with scalable AI(Artificial Intelligence) has enabled the identification of physical, behavioral, and psycho-social indicators of pain. However, the adoption of AI in clinical pain evaluation is hindered by challenges like personalization and fairness. Many AI models, including machine and deep learning, exhibit biases, discriminating against specific groups based on gender or ethnicity, causing skepticism among medical professionals about their reliability. This paper proposes a Multi-attribute Fairness Loss (MAFL) based Convolutional Neural Network (CNN) model designed to account for protected attributes in data, ensuring fair pain status predictions while minimizing disparities between privileged and unprivileged groups. We evaluate whether a balance between accuracy and fairness is achievable by comparing the proposed model with existing mitigation methods. Our findings indicate that the model performs favorably against state-of-the-art techniques. Using the NIH All-Of-US dataset, comprising data from 868 individuals over 1500 days, we demonstrate our model's effectiveness, achieving accuracy rates between 75% and 85%.

Figures (6)

• Figure 1: Proposed Wearables and Electronic Health Record (EHR) based Fair Pain Assessment Automation Pipeline
• Figure 2: The architecture of 1D deep Convolutional Neural Network (CNN) for pain level assessment
• Figure 3: Rich Subgroups for Each Protected Attribute
• Figure 4: Pain Assessment Performance for Classification Models
• Figure 5: Statistical Parity Difference (SPD) and Disparate Impact (DI) before and after applying pre-processing algorithms for different protected attributes. The black bars indicate the extent of $\pm 1$ standard deviation. The ideal fair value of SPD is 0 and DI is 1.