The integration of large language models in medical applications (Med LLMs) has raised critical questions about fairness and potential biased behaviors, which require rigorous evaluation. Due to the high complexity, existing work in this domain has primarily relied on manually crafted datasets for bias evaluation. In this study, we present a new method to scale up such bias evaluations by generating evaluation cases based on rigorous medical evidence. We use knowledge graphs and medical ontologies, coupled with a base LLM in a generative pipeline, to generate red-teaming evaluation scenarios. Such scenarios are commonly used to evaluate Med LLMs and often include medical question-answering cases, in which changes in an LLM's response to the same question with only a sensitive attribute flipped (e.g., a patient with vs. without insurance) can indicate unfair behavior. We specifically target the challenges of domain-specific bias characterization (e.g., in prevention vs. diagnosis), hallucination, and the dependencies between health outcomes and sensitive attributes (e.g., gender-dependent diseases). Through extensive experiments, we show that the test cases generated by our proposed method are reliable and can effectively reveal bias patterns in Med LLMs.Additionally, we publish a bias evaluation dataset, providing a comprehensive platform for testing and improving the fairness of medical LLMs, available at https://github.com/healthylaife/autofair.
Explainable artificial intelligence (XAI) is increasingly used in clinical decision support systems (CDSS) to improve the interpretability and trustworthiness of machine learning (ML) predictions in healthcare. While prior work has examined how healthcare professionals (HCPs) interpret various explanation methods, limited research has investigated how these explanations function in systems that incorporate novel data sources like game-based digital biomarkers. This paper extends our previous work by presenting an improved, interactive web-based CDSS designed to support HCPs in assessing the risk of Mild Cognitive Impairment (MCI). The system integrates ML predictions from gameplay data with data-centric, feature-importance, and counterfactual explanations to assist diagnosis and case-finding. We conducted a mixed-methods study with 51 HCPs to evaluate the system's understandability (objective and perceived), perceived usefulness, perceived ease of use, and trust. Participants gave consistently positive ratings across the self-reported perceived constructs, which were found to be strongly correlated. HCPs generally perceived the explanations as useful for clinical reasoning and reported that they understood and trusted the system, while acknowledging an initial learning curve. However, the evaluation of objective understanding was more mixed and correlated only weakly with the perceived constructs. In particular, participants encountered challenges in understanding game-based digital biomarkers, interpreting feature importances, combining local and global explanations, and understanding details of the underlying ML models. These findings highlight the importance of explanation designs that foster both positive user perceptions and accurate comprehension.
With the widespread adoption of digital pathology, computational pathology has facilitated the transition from traditional manual interpretation to intelligent-assisted diagnosis through the analysis of whole-slide images (WSIs). Nevertheless, WSIs are gigapixel in size and contain sparsely distributed diagnostically relevant information, while large portions of tissue may lack diagnostic relevance or even confound the model's decision-making. In addition, the heterogeneity of tissue structures scatters diagnostic features, necessitating models that capture long-range dependencies and integrate contextual information effectively. In this paper, we propose a novel reinforcement learning-guided Discriminative Region Sampling Framework (DRSF) for weakly supervised histopathology WSI classification. Central to DRSF is the proposed discriminative region sampling strategy, in which a reinforcement learning (RL)-based policy network sequentially samples the patches deemed most diagnostically relevant, thereby reducing redundant or interfering information. The classification module utilizes a bidirectional Mamba backbone to effectively model long-range semantic dependencies among dispersed sampled patches, while accommodating variable-length input sequences. Experimental results on WSI datasets of different staining types demonstrate that DRSF outperforms representative multiple-instance learning (MIL) frameworks. Moreover, DRSF enables direct visualization of the intermediate results of selected patches and sampling trajectories, thereby assisting pathologists in rapidly locating key regions and enhancing the interpretability of automated diagnosis.
Digital technologies now mediate much of health and care work, but their benefits are unevenly distributed, and research documenting this reaches practice slowly. One reason is that HCI and health evaluate work differently, so papers at their intersection are judged by whichever criteria a reviewer brings from their home community. In this introduction to the ACM HEALTH Continuous Special Section on Human-Centered Computing in Healthcare, we state the criteria we used and where they came from: which people a paper accounts for, whether the health context changes what the paper contributes to computing, whether its methods and any theory suit its question, and what its findings change for design, adoption, or governance. We introduce the 18 papers in this first installment.
Cervical cancer is a major global public health issue and ranks as the fourth most common cancer among women worldwide. The World Health Organization (WHO) has set an ambitious target to eliminate it by 2030. Yet, in low- and middle-income countries like India, the burden remains alarmingly high—it is the second most common cancer among women. Despite being largely preventable through vaccination and early screening, the disease’s prevalence continues to be a significant challenge. This study explores the factors contributing to the high prevalence of cervical cancer in India, assesses awareness levels, and identifies potential technological interventions to address this issue. We conducted this study using a mixed-method approach, including surveys (n = 177) and interviews (n = 29) with key stakeholders such as the general public, doctors, women who had undergone cervical cancer screening, and patients. Our findings revealed a concerning lack of awareness about cervical cancer along with a pervasive disregard for women’s health, underscoring the need for technological interventions that could help bridge these gaps. From this research, we are suggesting ways to alleviate the cervical cancer burden and identify avenues for Human–Computer Interaction (HCI) researchers to pursue further research, thereby enhancing the body of knowledge in women’s health research.
