Health data science最新文献

Background: A neuroimaging-derived biomarker termed the brain age is considered to capture the degree and diversity in the aging process of the brain, serving as a robust indicator of overall brain health. The impact of different levels of physical activity (PA) intensities on brain age is still not fully understood. This study aimed to investigate the associations between accelerometer-measured PA and brain age. Methods: A total of 16,972 eligible participants with both valid T ₁-weighted neuroimaging and accelerometer data from the UK Biobank was included. Brain age was estimated using an ensemble learning approach called Light Gradient-Boosting Machine (LightGBM). Over 1,400 image-derived phenotypes (IDPs) were initially chosen to undergo data-driven feature selection for brain age prediction. A measure of accelerated brain aging, the brain age gap (BAG) can be derived by subtracting the chronological age from the estimated brain age. A positive BAG indicates accelerated brain aging. PA was measured over a 7-day period using wrist-worn accelerometers, and time spent on light-intensity PA (LPA), moderate-intensity PA (MPA), vigorous-intensity PA (VPA), and moderate- to vigorous-intensity PA (MVPA) was extracted. The generalized additive model was applied to examine the nonlinear association between PA and BAG after adjusting for potential confounders. Results: The brain age estimated by LightGBM achieved an appreciable performance (r = 0.81, mean absolute error [MAE] = 3.65), which was further improved by age bias correction (r = 0.90, MAE = 3.03). We found that LPA (F = 2.47, P = 0.04), MPA (F = 6.49, P < 1 × 10^-300), VPA (F = 4.92, P = 2.58 × 10^-5), and MVPA (F = 6.45, P < 1 × 10^-300) exhibited an approximate U-shaped relationship with BAG, demonstrating that both insufficient and excessive PA levels adversely impact brain aging. Furthermore, mediation analysis suggested that BAG partially mediated the associations between PA and cognitive functions as well as brain-related disorders. Conclusions: Our study revealed a U-shaped association between accelerometer-measured PA and BAG, highlighting that advanced brain health may be attainable through engaging in moderate amounts of objectively measured PA irrespectively of intensities.

Background: Since the launch of drug regulatory reform in 2015, China has substantially increased the availability of new cancer therapies. However, the efficacy evidence criteria for modified new anticancer drugs have not been evaluated. This cross-sectional study aimed to assess the pivotal trials supporting the indication approvals of innovative and modified new chemical anticancer drugs in China. Methods: The characteristics of indications, regulatory aspects, and pivotal trial designs were extracted and described. The primary efficacy endpoints of the pivotal clinical trials, including overall survival (OS) and progression-free survival (PFS), were quantitatively assessed by meta-analysis. Results: Between 2016 and 2022, 77 cancer therapeutics for 107 indications were approved in China based on 128 pivotal trials. Among the 107 indications, 64 (59.8%) were classified as innovative anticancer drugs, and 43 (40.2%) as modified new anticancer drugs. The study found that pivotal trials for innovative approvals tended to be single-arm trials, while modified approvals were more likely to employ randomized clinical trials with larger sample sizes and rigorous designs. Despite innovative drugs often receiving more expedited regulatory designations, there were no statistically significant differences in clinical benefit of OS or PFS outcomes between innovative and modified approvals. Conclusions: These results suggest that the current regulatory framework may prioritize the speed of approval for innovative drugs over the strength of supporting evidence. These findings align with the strategic trends of pharmaceutical companies and regulatory inclinations that aim to expedite the approval of innovative anticancer drugs with a high unmet need, thereby accelerating patients' accessibility to treatment.

Background: Hearing loss (HL) is one major cause of disability and can lead to social impairments. However, the relationship between loneliness and the risk of incident HL remains unclear. Our study aimed to investigate this association among adults in the UK. Methods: This cohort study was based on data from the UK Biobank study. Loneliness was assessed by asking participants if they often felt lonely. Incident HL was defined as a primary diagnosis, ascertained via linkage to electronic health records. Cox proportional hazard regression models were used to examine the association between loneliness and risk of incident HL. Results: Our analyses included 490,865 participants [mean (SD) age, 56.5 (8.1) years; 54.4% female], among whom 90,893 (18.5%) reported feeling lonely at baseline. Over a median follow-up period of 12.3 years (interquartile range, 11.3 to 13.1), 11,596 participants were diagnosed with incident HL. Compared to non-lonely participants, lonely individuals exhibited an increased risk of HL [hazard ratio (HR), 1.36; 95% confidence interval (CI), 1.30 to 1.43]. This association remained (HR, 1.24; 95% CI, 1.17 to 1.31) after adjusting for potential confounders, including age, sex, socioeconomic status, biological and lifestyle factors, social isolation, depression, chronic diseases, use of ototoxic drugs, and genetic risk of HL. The joint analysis showed that loneliness was significantly associated with an increased risk of incident HL across all levels of genetic risks for HL. Conclusions: Loneliness was associated with the risk of incident HL independent of other prominent risk factors. Social enhancement strategies aimed at alleviating loneliness may prove beneficial in HL prevention.

Background: Large language models (LLMs) have shown promise in educational applications, but their performance on high-stakes admissions tests, such as the Dental Admission Test (DAT), remains unclear. Understanding the capabilities and limitations of these models is critical for determining their suitability in test preparation. Methods: This study evaluated the ability of 16 LLMs, including general-purpose models (e.g., GPT-3.5, GPT-4, GPT-4o, GPT-o1, Google's Bard, mistral-large, and Claude), domain-specific fine-tuned models (e.g., DentalGPT, MedGPT, and BioGPT), and open-source models (e.g., Llama2-7B, Llama2-13B, Llama2-70B, Llama3-8B, and Llama3-70B), to answer questions from a sample DAT. Quantitative analysis was performed to assess model accuracy in different sections, and qualitative thematic analysis by subject matter experts examined specific challenges encountered by the models. Results: GPT-4o and GPT-o1 outperformed others in text-based questions assessing knowledge and comprehension, with GPT-o1 achieving perfect scores in the natural sciences (NS) and reading comprehension (RC) sections. Open-source models such as Llama3-70B also performed competitively in RC tasks. However, all models, including GPT-4o, struggled substantially with perceptual ability (PA) items, highlighting a persistent limitation in handling image-based tasks requiring visual-spatial reasoning. Fine-tuned medical models (e.g., DentalGPT, MedGPT, and BioGPT) demonstrated moderate success in text-based tasks but underperformed in areas requiring critical thinking and reasoning. Thematic analysis identified key challenges, including difficulties with stepwise problem-solving, transferring knowledge, comprehending intricate questions, and hallucinations, particularly on advanced items. Conclusions: While LLMs show potential for reinforcing factual knowledge and supporting learners, their limitations in handling higher-order cognitive tasks and image-based reasoning underscore the need for judicious integration with instructor-led guidance and targeted practice. This study provides valuable insights into the capabilities and limitations of current LLMs in preparing prospective dental students and highlights pathways for future innovations to improve performance across all cognitive skills assessed by the DAT.

Background: Multimodal large language models (LLMs) have shown potential in various health-related fields. However, many healthcare studies have raised concerns about the reliability and biases of LLMs in healthcare applications. Methods: To explore the practical application of multimodal LLMs in skin disease identification, and to evaluate sex and age biases, we tested the performance of 2 popular multimodal LLMs, ChatGPT-4 and LLaVA-1.6, across diverse sex and age groups using a subset of a large dermatoscopic dataset containing around 10,000 images and 3 skin diseases (melanoma, melanocytic nevi, and benign keratosis-like lesions). Results: In comparison to 3 deep learning models (VGG16, ResNet50, and Model Derm) based on convolutional neural network (CNN), one vision transformer model (Swin-B), we found that ChatGPT-4 and LLaVA-1.6 demonstrated overall accuracies that were 3% and 23% higher (and F1-scores that were 4% and 34% higher), respectively, than the best performing CNN-based baseline while maintaining accuracies that were 38% and 26% lower (and F1-scores that were 38% and 19% lower), respectively, than Swin-B. Meanwhile, ChatGPT-4 is generally unbiased in identifying these skin diseases across sex and age groups, while LLaVA-1.6 is generally unbiased across age groups, in contrast to Swin-B, which is biased in identifying melanocytic nevi. Conclusions: This study suggests the usefulness and fairness of LLMs in dermatological applications, aiding physicians and practitioners with diagnostic recommendations and patient screening. To further verify and evaluate the reliability and fairness of LLMs in healthcare, experiments using larger and more diverse datasets need to be performed in the future.

Background: Acute ischemic stroke is a leading cause of death in the United States. Identifying patients with stroke at high risk of mortality is crucial for timely intervention and optimal resource allocation. This study aims to develop and validate machine learning-based models to predict in-hospital mortality risk for intensive care unit (ICU) patients with acute ischemic stroke and identify important associated factors. Methods: Our data include 3,489 acute ischemic stroke admissions to the ICU for patients not discharged or dead within 48 h from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database. Demographic, hospitalization type, procedure, medication, intake (intravenous and oral), laboratory, vital signs, and clinical assessment [e.g., Glasgow Coma Scale Scores (GCS)] during the initial 48 h of admissions were used to predict in-hospital mortality after 48 h of ICU admission. We explored 3 machine learning models (random forests, logistic regression, and XGBoost) and applied Bayesian optimization for hyperparameter tuning. Important features were identified using learned coefficients. Results: Experiments show that XGBoost tuned for area under the receiver operating characteristic curve (AUC ROC) was the best performing model (AUC ROC 0.86, F1 0.52), compared to random forests (AUC ROC 0.85, F1 0.47) and logistic regression (AUC ROC 0.75, F1 0.40). Top features include GCS, blood urea nitrogen, and Richmond RASS score. The model also demonstrates good fairness for males versus females and across racial/ethnic groups. Conclusions: Machine learning has shown great potential in predicting in-hospital mortality risk for people with acute ischemic stroke in the ICU setting. However, more ethical considerations need to be applied to ensure that performance differences across different racial/ethnic groups will not exacerbate existing health disparities and will not harm historically marginalized populations.

Background: There are few data on the comorbidity of diabetes in Chinese patients with depression. We aimed to calculate the prevalence and explore risk factors of type 2 diabetes mellitus (T2DM) among depression inpatients from 2005 to 2018 in Beijing. Methods: This study is a cross-sectional study. The data collected from 19 specialized psychiatric hospitals in Beijing were analyzed. The prevalence of T2DM and its distribution were analyzed. The multivariable logistic regression was performed to explore the risk factors of T2DM. Results: A total of 20,899 depression inpatients were included. The prevalence of T2DM was 9.13% [95% confidence interval (CI), 8.74% to 9.52%]. The prevalence of T2DM showed an upward trend with year (P for trend < 0.001) and age (P for trend < 0.001). The prevalence of T2DM was higher among readmitted patients (12.97%) and patients with comorbid hypertension (26.16%), hyperlipidemia (21.28%), and nonalcoholic fatty liver disease (NAFLD) (18.85%). The prevalence of T2DM in females was lower than in males among patients aged 18 to 59 years, while the prevalence of T2DM in females was higher than in males among patients aged ≥60 years. T2DM was associated with older age [adjusted odds ratios (aORs) ranged from 3.68 to 29.95, P < 0.001], hypertension (aOR, 3.01; 95% CI, 2.70 to 3.35; P < 0.001), hyperlipidemia (aOR, 1.69; 95% CI, 1.50 to 1.91; P < 0.001), and NAFLD (aOR, 1.58; 95% CI, 1.37 to 1.82; P < 0.001). Conclusions: The prevalence of T2DM among depression inpatients from 2005 to 2018 in Beijing was high and increased with the year. Depression inpatients who were older and with hypertension, hyperlipidemia, and NAFLD had a higher prevalence and risk of T2DM.

Background: It is still uncertain how multimorbidity patterns affect transitions between fall states among middle-aged and older Chinese. Methods: Data were obtained from China Health and Retirement Longitudinal Study (CHARLS) 2011-2018. We utilized latent class analysis to categorize baseline multimorbidity patterns, Markov multi-state model to explore the impact of multimorbidity characterized by condition counts and multimorbidity patterns on subsequent fall transitions, and Cox proportional hazard models to assess hazard ratios of each transition. Results: A total of 14,244 participants aged 45 years and older were enrolled at baseline. Among these participants, 11,956 (83.9%) did not have a fall history in the last 2 years, 1,054 (7.4%) had mild falls, and 1,234 (8.7%) had severe falls. Using a multi-state model, 10,967 transitions were observed during a total follow-up of 57,094 person-times, 6,527 of which had worsening transitions and 4,440 had improving transitions. Among 6,711 multimorbid participants, osteo-cardiovascular (20.5%), pulmonary-digestive-rheumatic (30.5%), metabolic-cardiovascular (22.9%), and neuropsychiatric-sensory (26.1%) patterns were classified. Multimorbid participants had significantly higher risks of transitions compared with other participants. Among 4 multimorbidity patterns, osteo-cardiovascular pattern had higher transition risks than other 3 patterns. Conclusions: Multimorbidity, especially the "osteo-cardiovascular pattern" identified in this study, was associated with higher risks of fall transitions among middle-aged and older Chinese. Generally, the effect of multimorbidity is more significant in older adults than in middle-aged adults. Findings from this study provide facts and evidence for fall prevention, and offer implications for clinicians to target on vulnerable population, and for public health policymakers to allocate healthcare resources.

Background: The electrocardiogram (ECG) is a valuable, noninvasive tool for monitoring heart-related conditions, providing critical insights. However, the interpretation of ECG data alongside patient information demands substantial medical expertise and resources. While deep learning methods help streamline this process, they often fall short in integrating patient data with ECG readings and do not provide the nuanced clinical suggestions and insights necessary for accurate diagnosis. Methods: Although recent advancements in multi-modal large language modeling have propelled their application scope beyond the natural language processing domain, their applicability to ECG processing remains largely unexplored, partly due to the lack of text-ECG data. To this end, we develop ECG-Language Model (ECG-LM), the first multi-modal large language model able to process natural language and understand ECG signals. The model employs a specialized ECG encoder that transforms raw ECG signals into a high-dimensional feature space, which is then aligned with the textual feature space derived from the large language model. To address the scarcity of text-ECG data, we generated text-ECG pairs by leveraging detailed ECG pattern descriptions from medical guidelines, creating a robust dataset for pre-training ECG-LM. Additionally, we fine-tune ECG-LM with public clinical conversation datasets and build an additional supervised fine-tuning dataset based on real clinical data from the hospital, aiming to provide a more comprehensive and customized user experience. Results: ECG-LM outperforms existing few-shot and zero-shot solutions in cardiovascular disease detection across all 3 tasks (diagnostic, rhythm, and form) while also demonstrating strong potential in ECG-related question answering. Conclusions: The results across various tasks demonstrate that ECG-LM effectively captures the intricate features of ECGs, showcasing its versatility in applications such as disease prediction and advanced question answering.