Demographic and Physical Determinants of Unhealthy Food Consumption in Polish Long-Term Care Facilities

Nutrients, Journal Year: 2025, Volume and Issue: 17(6), P. 1008 - 1008

Published: March 13, 2025

Background: Unhealthy food consumption in long-term care facilities (LTCFs) contributes to poor health outcomes among residents. This study aimed assess its prevalence, identify demographic and physical risk factors, propose targeted interventions. Methods: A mixed-methods (2017–2021) analyzed data from 1000 Polish LTCF residents (aged 35–105 years). Anthropometric measurements, bioimpedance analyses, dietary assessments, activity records were collected. Food items classified as “healthy” or “unhealthy” using an AI-based Large Language Model (LLM), applying WHO guidelines the NOVA classification system. Logistic regression chi-square tests assessed associations between unhealthy marital status, education level, mobility aid use, portion control. Results: prevalence was 15.6%. Married had significantly higher rates than unmarried individuals (22.6% vs. 14.3%, p < 0.01). Lower educational attainment correlated with increased (partial primary education: 34.7% tertiary 8.1%). Mobility users exhibited elevated (cane: 34.6%; walker: 22.6%). Poor control showed strongest association (OR = 3.2, 95% CI: 1.8–5.7). Conclusions: Marital disparities, limitations, key modifiable factors. Findings suggest need for nutrition programs, caregiver education, policy reforms improve literacy meal portioning. Future research should validate methods, intervention outcomes, expand studies diverse settings. These findings align Poland’s National Health Programme provide actionable insights global populations.

Language: Английский

Assessment of large language models’ performances and hallucinations for Chinese postgraduate medical entrance examination

Discover Education, Journal Year: 2025, Volume and Issue: 4(1)

Published: March 13, 2025

This study evaluates Large language models (LLMs)' performance on Chinese Postgraduate Medical Entrance Examination (CPGMEE) as well the hallucinations produced by LLMs and investigate their implications for medical education. We curated 10 trials of mock CPGMEE to evaluate performances 4 (GPT-4.0, ChatGPT, QWen 2.1 Ernie 4.0). Each question was inputted into LLMs, responses were independently reviewed three experienced graders determine accuracy using a three-tier scale (poor, borderline, good). The hallucination rates LLMs' also evaluated. chose GPT-4.0 4.0 further analysis since these two achieved better among four. outperformed in overall (76.2% vs. 69.1%, p < 0.0001), achieving higher 'good' (70.0% 64.6%, 0.01) lower 'poor' (25.2% vs 32.3%, rating. Factuality most prevalent type (9.7% 14.7% GPT-4 respectively). exhibited factual fabrication (6.0% 7.8%, = 0.033), instruction inconsistency (2.3% 5.4%, 0.0001) logical (3.7% 5.7%, 0.005) than GPT-4.0.Our results underscore promising potential both assisting preparation enhancing postgraduate education programs.

Language: Английский

Can ChatGPT Function as a Virtual Multidisciplinary Team? A Proof-of-Concept Study in Vascular Malformation Syndromes (Preprint)

Published: March 15, 2025

Sturge-Weber syndrome (SWS) and Klippel-Trenaunay (KTS) are complex vascular malformation syndromes that require multidisciplinary team (MDT) management. However, the traditional MDT approach faces challenges such as time coordination, geographical barriers, inefficiencies in cross-disciplinary communication. This study aims to evaluate ChatGPT’s potential simulating decision-making for SWS KTS by comparing its diagnostic treatment recommendations with conclusions. A case-based proof-of-concept design was employed, retrospectively analyzing records of two patients. Clinical data, imaging, laboratory results were input into ChatGPT, outputs evaluated dermatology experts using a 1-5 Likert scale across five dimensions: accuracy, completeness, appropriateness, insight, safety. ChatGPT performed well most dimensions, particularly but showed occasional uncertainty handling or rare cases, gene-phenotype associations. Inter-rater reliability ranged from negligible moderate (ICC -0.00 0.63), no significant differences observed between experts’ ratings (p > 0.05). shows strong tool decision-making, completeness recommendations, insights. it has limitations managing cases ensuring feasibility recommendations. Future studies larger sample sizes multi-center validation needed fully assess clinical value.

Language: Английский