TL;DR
GenAI-assisted processes can provide rapid, actionable design mitigations that reduce error likelihood and enhance patient autonomy, establishing a replicable pipeline for producing heuristic-driven design libraries across diverse medical device contexts.
BackgroundBecause automated peritoneal dialysis (APD) systems are used in the home and operated by patients or their care partners, the human-device interface must be thoughtfully designed to facilitate intuitive operation to reduce the risk of errors that could occur during their use. User-friendly design should also help reduce anxiety to peritoneal dialysis (PD) adoption by patients incident to end-stage kidney disease (ESKD). Usability issues remain a significant barrier to PD adoption and are an important contributor to premature death, serious injuries, and PD technique failure.MethodsA summative human factors usability study was conducted on a novel, gravity-based APD device (Archimedes™) with 15 current or former PD patients and 15 dialysis nurses. Participants were trained for 2 h, followed by a training decay period, then evaluated with critical Use Scenario tasks consisting of Simulated Use tasks and Knowledge tasks reflecting tasks that could result in patient harm if performed incorrectly.ResultsOf the Simulated Use tasks evaluated, 97.3% were deemed successful across all users. This high success rate demonstrates the effectiveness of the Archimedes APD device in facilitating the tasks required for PD. For patients, success outcomes were achieved in 96.8% of Simulated Use tasks. Nurses achieved a success rate of 97.8% for Simulated Use tasks evaluated. For Knowledge tasks, success outcomes were achieved in 99.3% and 98.6% of tasks for patients and nurses.ConclusionIn this summative human factors usability study, the Archimedes APD system usability was found to be safe, accessible and easy for its intended users and use environments after the relatively short training period compared to the status quo.
Healthcare software prototypes usually have conversational help, scheduling, role-specific dashboards, and real-time data. Their evaluations focus more on feature lists than on failure modes. This paper reports a reproducible evaluation of safety and access control for Medicare. React, Node.js/Express, MongoDB, and socket.io. AI.IO educational health care platform. We froze a 20-prompt safety benchmark covering emergencies, first aid, medication use, mental health and general questions. We defined eight source-level security checks, including JSON Web Token middleware, role guards, appointment ownership, patient-only creation, password hashing, secret handling and Socket.IO room access. The baseline deterministic chatbot passed 7 out of 20 prompts (35%) and 5 out of 8 security checks. The dynamic loopback test also showed that an unauthenticated client could join the doctor telemetry room. Remediation introduced specific intent-matching, explicit high-risk routes, authenticated socket.io middleware, room-level role restrictions, and appointment ownership enforcement. Our final benchmark passed all 20 prompts and all eight source-level checks. The study does not claim clinical efficacy: no patient records, clinician participants, diagnostic model, or clinical outcomes were assessed. The Atlas network allowlist blocked database-backed exploitation and post-remediation dynamic socket tests, which were reported as unexecuted. Thus, the contribution is a transparent case study of how an ambitious student prototype can be transformed into a more testable, safety-aware system.
Traditional usability assessments and questionnaires, such as the System Usability Scale (SUS), were designed for deterministic systems with predictable, linear outputs. However, AI-enabled medical devices are inherently probabilistic and co-evolve with the user through repeated interaction, rendering traditional usability assessments insufficient for guaranteeing the long-term safety in the use of high-risk probabilistic systems. Current literature reveals a striking absence of longitudinal studies, creating significant methodological blind spots regarding how trust calibrates over time and whether automation bias intensifies with habitual use. In this position paper, we present a manifesto for a longitudinal, three-fold methodological pivot in health human-AI interaction. We propose moving beyond static satisfaction metrics towards relational metrics —Longitudinal Trust Calibration (LTC), Automation Bias Drift (ABD), and Error Recovery Velocity (ERV)—that track the maturity and resilience of the human-AI partnership. This framework provides an actionable path toward a safety-in-use paradigm that acknowledges the temporal, dynamic nature of high-risk health AI.
Medication Administration Errors (MAEs) represent a persistent global threat to patient safety, contributing significantly to patient morbidity, prolonged hospitalisations, and substantial financial strain on healthcare infrastructure. In the context of an accelerating global population ageing trend, the complexity of managing polypharmacy in home-based elderly care environments has reached an unprecedented peak, rendering traditional manual clinical workflows and human verification protocols increasingly insufficient. This comprehensive review synthesises contemporary literature from 2020 to 2026 regarding the design, architectural deployment, operational efficacy, and clinical outcomes of medication administration safety and error prevention systems leveraging intelligent nursing support technologies. A systematic appraisal of high-impact literature across databases including PubMed, IEEE Xplore, ScienceDirect, and MDPI was conducted, focusing on the convergence of the Internet of Things (IoT), wearable sensors, Artificial Intelligence (AI), Machine Learning (ML), smart home environments, and mobile health (mHealth) frameworks. The findings demonstrate that integrated closed-loop medication management infrastructures substantially reduce MAEs by automating patient identification, prescription label parsing via computer vision, and real-time physiological response tracking. Furthermore, machine learning models provide predictive clinical decision support that alerts nursing professionals to potential adverse drug events before administration. However, broad translation from controlled pilot settings to ubiquitous clinical practice faces critical challenges regarding systemic interoperability, data security, and ethical issues surrounding patient autonomy. Ultimately, intelligent nursing support technologies represent a paradigm shift in healthcare delivery, transforming medication safety from a reactive, human-dependent verification process into a proactive, data-driven, and preventive system. Future research must target scalable, secure, and human-centred frameworks to fully harmonise machine intelligence with professional nursing practice.
Nearly half of adults struggle to understand written health information, making medical communication a persistent barrier to effective care. While artificial intelligence has potential to improve health communication, few patient-facing tools have undergone systematic validation for personalized medical explanation. Patiently AI is a mobile application designed to clarify clinician-authored medical notes using large language models with audience-specific adaptations (child, teenager, adult, carer) and tone variations (friendly, informative, reassuring). A three-phase mixed-methods evaluation was conducted: (1) computational readability analysis of 210 AI-generated explanations using established metrics; (2) expert review by 15 healthcare professionals assessing medical accuracy, safety, and communication quality; and (3) a patient survey of 54 participants evaluating preferences, comprehension, and acceptance. AI-generated explanations demonstrated consistent improvements in readability, with mean Flesch–Kincaid Grade Level decreasing by 2.96 levels (10.57–7.61), Flesch Reading Ease increasing by 31.9 points (37.7–69.6), and Gunning Fog Index decreasing by 4.09 points (14.5–10.4); all improvements were statistically significant (all P ≤ 0.002). Readability gains were greatest for younger audiences (child: 4.25 grade-level reduction; adult: 1.80). Expert reviewers rated outputs highly for medical accuracy (4.49 ± 0.83/5), clarity (4.53 ± 0.77/5), and trustworthiness (4.37 ± 0.90/5), with 87.3% assessed as clinically safe. Inter-rater agreement across the 15 reviewers was substantial (Gwet's AC1 = 0.72 for safety assessments). Among patients, 70.0% of responses preferred AI-generated explanations ( P < 0.001), with 98.1% comprehension accuracy and high ratings for clarity (4.58 ± 0.65/5) and confidence in care (4.19 ± 0.85/5). Overall, 70.4% indicated a likelihood of using the application. This mixed-methods evaluation suggests that a deliberately constrained, language-focused AI system can improve the accessibility of medical notes while preserving clinical accuracy and safety. Patiently AI demonstrates a scalable approach to supporting health literacy and patient engagement without extending into clinical interpretation.
Low-resource healthcare systems in South Asia and Africa face severe constraints in infrastructure, connectivity, data availability, and digital literacy that shape how artificial intelligence can be deployed. This paper examines human–AI interaction in these settings, focusing on patients, community health workers, and clinicians rather than model performance alone. Drawing on empirical studies and case examples from Bangladesh, India, and Nigeria, it shows that AI systems designed for high-income contexts often fail when transferred without adaptation. Key challenges include language diversity, absence of electronic health records, limited AI literacy, trust deficits, and unresolved ethical and liability concerns. The analysis demonstrates that localized, multilingual, human-in-the-loop AI can meaningfully augment care when integrated into existing workflows and mediated by trusted health workers. Overall, the paper synthesizes design principles for human-centered AI that emphasize localization, explainability, training, and accountability, arguing that effectiveness depends more on interaction design than technical sophistication.
Earth AI
Piloting the world's first double-blind AI evaluations
The “CrysVCD” tool developed at MIT could cut the huge amounts of time and money spent on screening out chemically unstable designs.