Dae-Soon Son

Medical AI is often evaluated using aggregate measures of discrimination, calibration, and accuracy. However, these measures can obscure clinically important variation across patient groups, institutions, devices, and workflows. This viewpoint defines refined exclusion as a governance condition in which an AI system appears successful in aggregate, while uncertainty, error, or reduced clinical reliability is concentrated in populations that are insufficiently represented, measured, validated, or monitored. The concept does not replace algorithmic fairness, hidden stratification, dataset shift, or subgroup performance analysis. It connects these mechanisms to a distinct consequence: an unequal distribution of safety that remains inadequately detected or corrected. Drawing on purposively selected, illustrative evidence from population health management, chest radiography, dermatology, computational pathology, medical foundation models, and clinical measurement, we distinguish model-level disparity, patient safety signals, and documented patient harm. We then frame data justice as a complementary governance approach with distributional, procedural, and substantive dimensions. The proposed lifecycle decision gates address intended use, subgroup learnability, data provenance, validation, procurement, local deployment, monitoring, updates, and patient feedback. Each gate links minimum evidence to decision authority and 1 of 4 actions: proceed, enrich or validate, restrict use, or pause or retire. Governance intensity should be proportionate to clinical risk and evidentiary uncertainty. By linking subgroup evidence gaps to institutional decisions and corrective action, the framework shifts attention from whether a model performs well on average to whether its safety is demonstrable for the populations and settings in which it will be used.

Aim: This study addressed two research questions. First, which sociodemographic, behavioral, and psychological factors are independently associated with elevated depressive symptoms among Korean women aged 20–59 years? Second, does subjective perception of one’s own body size retain an independent association with depressive symptoms after these factors are taken into account, and does this pattern differ between younger and midlife women? The primary outcome was elevated depressive symptoms, defined as a PHQ-9 score of 5 or above. Design: This cross-sectional study used pooled data from the four Korea National Health and Nutrition Examination Survey cycles in which the PHQ-9 was administered (2014, 2016, 2018, and 2020). All estimates incorporate the complex survey design and therefore refer to the national population of Korean women aged 20–59 years. Methods: We analysed data from Korean women aged 20–59 years who completed the PHQ-9 (n = 7595). Elevated depressive symptoms were defined as a PHQ-9 score of 5 or above, indicating mild or greater symptomatology. Design-based multivariable logistic regression models were constructed in sequential blocks: demographic variables (Model 1); behavioural and psychological factors added (Model 2); and perceived body size, entered as indicator contrasts against average body size, added (Model 3). Analyses were stratified by age group (20–39 and 40–59 years), and age-group interaction terms were fitted in a pooled model to test formally whether associations differed across the female lifespan. Results: The weighted prevalence of elevated depressive symptoms (PHQ-9 ≥ 5) was 24.1% (95% CI 22.8–25.3). In bivariate analyses, perceived body size differed significantly between women with and without elevated depressive symptoms (p < 0.001). After adjustment for behavioural and psychological factors, neither perceived thinness nor perceived obesity was associated with depressive symptoms in either age group (20–39 years, thin: OR 1.25, 95% CI 0.92–1.69; obese: OR 1.05, 95% CI 0.85–1.30; 40–59 years, thin: OR 1.21, 95% CI 0.84–1.73; obese: OR 0.99, 95% CI 0.80–1.22), and the joint two-degree-of-freedom test was non-significant in both groups. Perceived stress (OR 5.15 and 4.93), poor subjective health (OR 2.11 and 2.55), and lifetime smoking experience (OR 1.75 and 1.79) showed the strongest and most consistent associations in both groups (all p < 0.001). Lower household income was associated with depressive symptoms among women aged 40–59 years and heavy episodic drinking among those aged 20–39 years; in formal interaction testing, household income showed only a non-significant trend towards effect modification by age (p = 0.060), so the age differences are described rather than established. Conclusions: The crude association between perceived body size and depressive symptoms did not persist after adjustment for behavioral and psychological factors. Because perceived stress and subjective health may lie on a causal pathway from body size perception to depressive symptoms, this attenuation is consistent with either confounding or an indirect pathway, and the present cross-sectional design cannot distinguish between them. Perceived stress and subjective health showed the strongest associations across age groups, while lower household income remained an independent correlate in midlife. This finding points to a material condition that individual-level screening cannot by itself address.

UNSTRUCTURED Medical artificial intelligence (AI) systems are often evaluated through aggregate performance metrics and output-level fairness measures. However, clinically meaningful harms may remain hidden when systems perform well on average while underperforming for data-poor, underrepresented, or structurally marginalized populations. This Viewpoint uses the concept of refined exclusion to synthesize a recurring pattern in medical AI: systems may appear technically successful at the population level while transferring uncertainty, misclassification, delayed recognition, or reduced clinical reliability to groups that are less visible within training data, validation cohorts, proxy definitions, and deployment workflows. Drawing on representative cases from population health management, chest radiograph AI, dermatology, computational pathology, and foundation model applications, we argue that refined exclusion should not be treated merely as algorithmic bias or a defect of model outputs. Rather, it reflects a data governance failure with direct implications for patient safety. Moving beyond output-centered algorithmic fairness, we propose data justice as a governance foundation for medical AI, organized across distributional, procedural, and substantive dimensions. We further outline operational checkpoints across the medical AI lifecycle, including subgroup learnability assessment, data provenance documentation, local validation, procurement-stage accountability, explainability-based proxy audits, post-deployment subgroup monitoring, and patient participation. Reframing refined exclusion as a patient safety problem shifts the central governance question from “Is this model accurate on average?” to “For whom is this system safe, reliable, and clinically accountable?”

Artificial intelligence (AI) in health care is increasingly defined not by static algorithms but by adaptive intelligence—systems that evolve over time through interactions with data, clinicians, and clinical environments. This adaptive capacity creates a structural mismatch with regulatory frameworks built for technologies whose behavior remains static. As AI models drift, recalibrate, or degrade in real-world contexts, they dissolve the linear boundaries between design, deployment, and clinical interpretation. These temporal, epistemic, and organizational frictions expose responsibility gaps that cannot be resolved through incremental modifications to legacy oversight structures. Regulators across major jurisdictions are beginning to respond to these challenges, though with differing orientations. The United States advances mechanisms for predictable adaptation, including Predetermined Change Control Plans, real-world evidence frameworks, and life cycle–oriented quality management reforms. The European Union emphasizes precautionary, rights-based governance through the European Union Artificial Intelligence Act (AI Act) and modernized liability rules. South Korea, operating within a hyperconnected digital health ecosystem, has introduced the Digital Medical Products Act (DMPA), one of the world’s first comprehensive statutory frameworks for learning medical AI. Despite philosophical differences, these regulatory trajectories converge on a shared insight: learning AI systems cannot be governed by static rules or episodic evaluation. This viewpoint proposes Co-Lifecycle Governance as a conceptual framework to synchronize regulatory oversight with adaptive intelligence. Rather than treating oversight as a discrete event, Co-Lifecycle Governance frames regulation as a continuous, synchronized process grounded in 4 pillars: continuous validation, agile change management, proactive performance surveillance, and distributed accountability. Each pillar functions as a structural antidote to the responsibility frictions that arise when AI systems evolve faster than expectations surrounding them. Together, these pillars provide a governance grammar capable of supporting safe, iterative model improvement while maintaining system-level trust. Drawing from the strengths of US predictability, European Union accountability, and Korean scalability, this paper outlines a hybrid convergence pathway that synthesizes predictability, accountability, and operational feasibility. Learning AI will not wait for governance to catch up; oversight must evolve in lockstep with adaptive intelligence. Co-Lifecycle Governance offers a foundation for regulatory systems that not only regulate learning AI but also learn with it—at the speed at which adaptive intelligence actually changes.

Medical artificial intelligence (AI) is transitioning from static, rule-based systems into adaptive models capable of continuous learning and iterative refinement. Such adaptivity expands the utility and performance of clinical AI systems across diverse patient populations and real-world conditions. However, these properties challenge regulatory paradigms originally designed for fixed-function medical devices. Although the United States and the European Union share goals of ensuring safety, accountability, and trustworthy performance, their regulatory architectures diverge due to underlying legal-philosophical traditions. The United States employs a common-law, evidence-driven approach centered on the Total Product Life Cycle, using predetermined change-control mechanisms and real-world observational data to support iterative improvement under controlled risk. In contrast, the European Union adopts a civil-law, precautionary model operationalized through the Artificial Intelligence Act, the Medical Device Regulation, and the revised Product Liability Directive, emphasizing ex-ante duties, transparency, traceability, and accountability. Understanding these distinct regulatory DNAs is critical for aligning lifecycle governance of adaptive AI across jurisdictions and ensuring safe, context-responsive innovation.

UNSTRUCTURED Artificial intelligence (AI) in health care is increasingly defined not by static algorithms but by adaptive intelligence—systems that evolve over time through interactions with data, clinicians, and clinical environments. This adaptive capacity creates a structural mismatch with regulatory frameworks built for technologies whose behavior remains static. As AI models drift, recalibrate, or degrade in real-world contexts, they dissolve the linear boundaries between design, deployment, and clinical interpretation. These temporal, epistemic, and organizational frictions expose responsibility gaps that cannot be resolved through incremental modifications to legacy oversight structures. Regulators across major jurisdictions are beginning to respond to these challenges, though with differing orientations. The United States advances mechanisms for predictable adaptation, including Predetermined Change Control Plans (PCCPs), real-world evidence frameworks, and lifecycle-oriented quality management reforms. The European Union emphasizes precautionary, rights-based governance through the AI Act and modernized liability rules. South Korea, operating within a hyper-connected digital health ecosystem, has introduced the Digital Medical Products Act (DMPA), one of the world’s first comprehensive statutory frameworks for learning medical AI. Despite philosophical differences, these regulatory trajectories converge on a shared insight: learning AI systems cannot be governed by static rules or episodic evaluation. This Viewpoint proposes Co-Lifecycle Governance as a conceptual framework to synchronize regulatory oversight with adaptive intelligence. Rather than treating oversight as a discrete event, Co-Lifecycle Governance frames regulation as a continuous, synchronized process grounded in four pillars: continuous validation, agile change management, proactive performance surveillance, and distributed accountability. Each pillar functions as a structural antidote to the responsibility frictions that arise when AI systems evolve faster than expectations surrounding them. Together, these pillars provide a governance grammar capable of supporting safe, iterative model improvement while maintaining system-level trust. Drawing from the strengths of U.S. predictability, EU accountability, and Korean scalability, this paper outlines a global hybrid pathway that synthesizes predictability, accountability, and operational feasibility. Learning AI will not wait for governance to catch up; oversight must evolve in lockstep with adaptive intelligence. Co-Lifecycle Governance offers a foundation for regulatory systems that not only regulate learning AI, but also learn with it—at the speed at which adaptive intelligence actually changes.

Diabetes is a growing health concern among young adults; yet, most studies focus on older populations. This study aimed to examine the factors associated with prediabetes development in young adults. We carried out a cross-sectional study using data from the Korean National Health and Nutrition Examination Survey (KNHANES) for the period 2013-2021. Young adults were categorized into prediabetes (n = 2306) and normal glucose (n = 2306) groups. Propensity score matching was applied to control for age and sex, and multiple logistic regression was performed to identify significant predictors for prediabetes. The results showed that glycated hemoglobin had the highest adjusted Odds Ratio (OR = 2.27, p < 0.001), confirming its strong association with pre diabetes. Age group (OR = 2.27, p < 0.001), also showed a substantial effect, Sex (OR = 1.95, p < 0.001), physical activity (OR = 1.69, p < 0.001), Overweight (OR = 1.39, p < 0.0034) and Fasting blood sugar (OR = 1.33, p < 0.0034) were also significant predictors. However, alcohol consumption and underweight were not statistically significant. Metabolic indicators along with physical inactivity are critical factors influencing prediabetes in young adults. Early screening and targeted lifestyle interventions are essential in preventing diabetes progression, and reducing the risk of future health burdens.

Epilepsy, a chronic neurological disorder, is triggered by various insults, including traumatic brain injury and stroke. Acid sphingomyelinase (ASMase), an enzyme that hydrolyzes sphingomyelin into ceramides, is implicated in oxidative stress, neuroinflammation, and neuronal apoptosis. Ceramides, which have pro-apoptotic properties, contribute to oxidative damage and lysosomal dysfunction, exacerbating neuronal injury. This study investigates the role of ASMase in epilepsy, hypothesizing that seizure activity upregulates ASMase, increasing ceramide levels, DNA damage, and neuronal apoptosis. We employed a pilocarpine-induced rat seizure model and examined the effects of imipramine, an ASMase inhibitor, administered intraperitoneally (10 mg/kg) for four weeks post-seizure induction. Histological and cognitive analyses showed that while imipramine did not prevent early neuronal death within the first week, it significantly reduced markers of neuronal apoptosis by four weeks. Imipramine also promoted hippocampal neurogenesis and preserved cognitive function, which is often impaired following seizures. These findings suggest that ASMase inhibition could mitigate neuronal apoptosis and improve cognitive recovery after seizures. Imipramine may serve as a promising therapeutic strategy for epilepsy-associated neuronal damage and cognitive deficits. Further studies should delineate the molecular mechanisms of ASMase inhibition and evaluate its long-term efficacy in addressing epilepsy-related neurodegeneration and functional impairments.

Brain ischemia, a condition in which the brain is deprived of blood flow, can lead to a stroke due to blocked or unstable blood vessels. Global cerebral ischemia (GCI), characterized by an interruption in blood flow, deprives the brain of oxygen and nutrients, producing reactive oxygen species (ROS) that trigger cell death, which kills nerve cells. Microplastics (MPs), tiny environmental pollutants, can enter the human body through contaminated food, water, disposable items, cosmetics, and more. Once in the brain, MPs can increase neuroinflammation by overstimulating inflammatory factors such as microglia. MPs can also damage neurons by scratching myelin and microtubules, slowing signal transduction, causing cognitive impairment, and leading to neuronal death. Furthermore, microtubule damage may result in the release of phosphorylated tau proteins, potentially linked to Alzheimer’s disease. We hypothesized that MPs could exacerbate neuroinflammation and microtubule destruction after GCI, leading to increased neuronal death. To test this hypothesis, we administered MPs (0.5 µm) orally at a dose of 50 mg/kg before and after inducing GCI. Staining techniques such as Fluoro-Jade B (FJB), ionized calcium-binding adaptor molecule 1 (Iba-1), cluster of differentiation 68 (CD68), myelin basic protein (MBP), and microtubule-associated protein 2 (MAP2) were used, along with Western blot analysis for interleukin-6 (IL-6), TNF-α, tau-5, and phospho-tau (S396) to evaluate the effects of MPs on neuronal cell death, neuroinflammation, and microtubule destruction. The results showed that MP accumulation significantly increased neuroinflammation, microtubule disruption, and neuronal cell death in the GCI-MP group compared to the GCI-vehicle group. Therefore, this study suggests that MP accumulation in daily life may contribute to the exacerbation of the disease, potentially leading to severe neuronal cell death after GCI.

Background/Objectives: We investigated the association between chronic rhinosinusitis (CRS) and cancer risk in an adult Korean population. Methods: This retrospective cohort study used data from the Korean National Health Insurance Service database. To ensure comparability between the groups, adjustments were made for potential confounding factors, including sex, age, residence, household income, diabetes, hypertension, and chronic kidney disease. The primary endpoint was the presence of newly diagnosed cancer. Results: Among 1,337,120 individuals in the nationally representative cohort database, 10,567 patients with CRS were identified and matched with 42,268 control subjects without CRS. Patients with CRS had a significantly higher risk of overall cancer events than controls. The adjusted hazard ratio (HR) for cancer in the CRS group was 1.16 (95% confidence interval [CI]: 1.05–1.28). Notably, female patients with CRS had an elevated risk of incident cancer events. Furthermore, patients with CRS without nasal polyps exhibited a significantly increased risk of cancer, whereas those with CRS with nasal polyps did not show a similar association. Conclusions: These findings underscore the need for physicians to carefully monitor patients with CRS for potential cancer progression and develop appropriate therapeutic strategies to mitigate the impact of this condition.