Hasil untuk "Public aspects of medicine"

Hyunjae Kim, Jiwoong Sohn, Aidan Gilson et al.

Large language models (LLMs) are transforming the landscape of medicine, yet two fundamental challenges persist: keeping up with rapidly evolving medical knowledge and providing verifiable, evidence-grounded reasoning. Retrieval-augmented generation (RAG) has been widely adopted to address these limitations by supplementing model outputs with retrieved evidence. However, whether RAG reliably achieves these goals remains unclear. Here, we present the most comprehensive expert evaluation of RAG in medicine to date. Eighteen medical experts contributed a total of 80,502 annotations, assessing 800 model outputs generated by GPT-4o and Llama-3.1-8B across 200 real-world patient and USMLE-style queries. We systematically decomposed the RAG pipeline into three components: (i) evidence retrieval (relevance of retrieved passages), (ii) evidence selection (accuracy of evidence usage), and (iii) response generation (factuality and completeness of outputs). Contrary to expectation, standard RAG often degraded performance: only 22% of top-16 passages were relevant, evidence selection remained weak (precision 41-43%, recall 27-49%), and factuality and completeness dropped by up to 6% and 5%, respectively, compared with non-RAG variants. Retrieval and evidence selection remain key failure points for the model, contributing to the overall performance drop. We further show that simple yet effective strategies, including evidence filtering and query reformulation, substantially mitigate these issues, improving performance on MedMCQA and MedXpertQA by up to 12% and 8.2%, respectively. These findings call for re-examining RAG's role in medicine and highlight the importance of stage-aware evaluation and deliberate system design for reliable medical LLM applications.

Maximilien Lora, H. A. Ménard, Anastasia Nijnik et al.

ABSTRACT Low dose methotrexate (LD‐MTX) remains the gold standard in rheumatoid arthritis (RA) therapy. Multiple mechanisms on a variety of immune cells contribute to the anti‐inflammatory effects of LD‐MTX. Inflammatory signaling is deeply implicated in hematopoiesis by regulating hematopoietic stem and progenitor cell (HSPC) fate decisions; raising the question of whether HSPC are also modulated by LD‐MTX. This is the first study to characterize the effects of LD‐MTX on HSPC. CD34+ HSPC were isolated from healthy donors' non‐mobilized peripheral blood. Resting and/or cycling HSPCs were treated with LD‐MTX [dose equivalent to that used in RA patients]. Flow cytometry was performed to assess HSPC viability, cell cycle, surface abundance of reduced folate carrier 1 (RFC1), proliferation, reactive oxygen species (ROS) levels, DNA double‐strand breaks, p38 activation, and CD34+ subpopulations. HSPC clonogenicity was tested in colony‐forming cell assays. Our results indicate that in cycling HSPC, membrane RFC1 is upregulated and, following LD‐MTX treatment, they accumulate more intracellular MTX than resting HSPC. In cycling HSPC, LD‐MTX inhibits HSPC expansion by promoting S‐phase cell‐cycle arrest, increases intracellular HSPC ROS levels and DNA damage, and reduces HSPC viability. Those effects involve the activation of the p38 MAPK pathway and are rescued by folinic acid. The effects of LD‐MTX are more evident in CD34+ CD38High progenitors. In non‐cycling HSPC, LD‐MTX also reduces the proliferative response while preserving their clonogenicity. In summary, HSPC uptake LD‐MTX differentially according to their cycling state. In turn, LD‐MTX results in reduced proliferation and the preservation of HSPC clonogenicity.

P.A. López Fernández, A. Lertxundi, M. Subiza et al.

Julian Chingoma, Umberto Grandi, Arianna Novaro

We study situations where a group of voters need to take a collective decision over a number of public issues, with the goal of getting a result that reflects the voters' opinions in a proportional manner. Our focus is on interconnected public decisions, where the outcome on one or more issues has repercussions on the acceptance or rejection of other issues in the agenda. We show that the adaptation of classical justified-representation axioms to this enriched setting are always satisfiable only for restricted classes of public agendas. We adapt well-known proportional decision rules to take the structure of the public agenda into account, and we show that they match justified-representation properties in approximation on a class of expressive constraints. We also identify another path to achieving proportionality on interconnected issues via an adaptation of the notion of priceability.

Lucy Jiang

Activism can take a multitude of forms, including protests, social media campaigns, and even public art. The uniqueness of public art lies in that both the act of creation and the artifacts created can serve as activism. Furthermore, public art is often site-specific and can be created with (e.g., commissioned murals) or without permission (e.g., graffiti art) of the site's owner. However, the majority of public art is inaccessible to blind and low vision people, excluding them from political and social action. In this position paper, we build on a prior crowdsourced mural description project and describe the design of one potential process artifact, ARtivism, for making public art more accessible via augmented reality. We then discuss tensions that may occur at the intersection of public art, activism, and technology.

Ahmed M Salih

Data pre-processing is a significant step in machine learning to improve the performance of the model and decreases the running time. This might include dealing with missing values, outliers detection and removing, data augmentation, dimensionality reduction, data normalization and handling the impact of confounding variables. Although it is found the steps improve the accuracy of the model, but they might hinder the explainability of the model if they are not carefully considered especially in medicine. They might block new findings when missing values and outliers removal are implemented inappropriately. In addition, they might make the model unfair against all the groups in the model when making the decision. Moreover, they turn the features into unitless and clinically meaningless and consequently not explainable. This paper discusses the common steps of the data preprocessing in machine learning and their impacts on the explainability and interpretability of the model. Finally, the paper discusses some possible solutions that improve the performance of the model while not decreasing its explainability.

Juan Miguel Lopez Alcaraz, Hjalmar Bouma, Nils Strodthoff

Background: AI-driven prediction algorithms have the potential to enhance emergency medicine by enabling rapid and accurate decision-making regarding patient status and potential deterioration. However, the integration of multimodal data, including raw waveform signals, remains underexplored in clinical decision support. Methods: We present a dataset and benchmarking protocol designed to advance multimodal decision support in emergency care. Our models utilize demographics, biometrics, vital signs, laboratory values, and electrocardiogram (ECG) waveforms as inputs to predict both discharge diagnoses and patient deterioration. Results: The diagnostic model achieves area under the receiver operating curve (AUROC) scores above 0.8 for 609 out of 1,428 conditions, covering both cardiac (e.g., myocardial infarction) and non-cardiac (e.g., renal disease, diabetes) diagnoses. The deterioration model attains AUROC scores above 0.8 for 14 out of 15 targets, accurately predicting critical events such as cardiac arrest, mechanical ventilation, ICU admission, and mortality. Conclusions: Our study highlights the positive impact of incorporating raw waveform data into decision support models, improving predictive performance. By introducing a unique, publicly available dataset and baseline models, we provide a foundation for measurable progress in AI-driven decision support for emergency care.

Ashley Williams, Kayla Schulte, Diana Varaden

Abstract Background Air pollution is of significant environmental and public health concern globally. While much research has historically focused on outdoor air pollution, indoor air pollution has been relatively under-explored despite its strong connection with health outcomes, particularly respiratory health. Studies on air pollution exposure mitigation consistently reveal a significant knowledge gap between the understanding of air pollution as a health risk among lay individuals and expert scientists. This study aimed to assess how people define and understand the concept of ‘clean air’ within their home setting. Methods We adopted a mixed-methods approach which used a guided questionnaire designed to elicit both quantitative and qualitative data, collected as digital voice notes. The total sample (n = 40) comprised data from two socially different sites of science and non-science events. We compared whether the notion of clean air inside homes differs between these two different social contexts and how views and ‘sense’ of indoor air pollution are formed. The concept of ‘place’ facilitated fluidity in our explorative analysis. Insights allowed us to assess the extent to which context mediates individuals’ perceptions of indoor air pollution and attitudes towards health risk. Results We found that individuals’ insights were embodied in repetitive day-to-day activities (e.g. cleaning and cooking). Three key themes emerged (1) Stimulative Effects, (2) Contextual Conditions, and (3) Risk Attitudes. Sensory perceptions such as sight, smell and temperature primarily motivated participants to assess air quality inside their homes. These perceptions were shaped by contextual conditions, influencing how individuals perceived their health risk and were subsequently motivated to spend personal time considering or seeking information about household air pollution, or improving their home air quality. Conclusions Our insights revealed that social, geographical, and contextual factors play a crucial role in individuals’ understandings of indoor air pollution. These dimensions should be integrated into designs of effective public health risk communication strategies. Our findings highlight that common lay perceptions and practices intended to improve air quality may pose health risks. Therefore, risk communication about household air pollution must extend beyond objective information by considering contextual factors that shape how people interpret and respond to air quality issues. Clinical trial number Not applicable.

Ziru Fu, Yu Cheng Hsu, Christian S Chan et al.

BackgroundSentiment analysis is a significant yet difficult task in natural language processing. The linguistic peculiarities of Cantonese, including its high similarity with Standard Chinese, its grammatical and lexical uniqueness, and its colloquialism and multilingualism, make it different from other languages and pose additional challenges to sentiment analysis. Recent advances in models such as ChatGPT offer potential viable solutions. ObjectiveThis study investigated the efficacy of GPT-3.5 and GPT-4 in Cantonese sentiment analysis in the context of web-based counseling and compared their performance with other mainstream methods, including lexicon-based methods and machine learning approaches. MethodsWe analyzed transcripts from a web-based, text-based counseling service in Hong Kong, including a total of 131 individual counseling sessions and 6169 messages between counselors and help-seekers. First, a codebook was developed for human annotation. A simple prompt (“Is the sentiment of this Cantonese text positive, neutral, or negative? Respond with the sentiment label only.”) was then given to GPT-3.5 and GPT-4 to label each message’s sentiment. GPT-3.5 and GPT-4’s performance was compared with a lexicon-based method and 3 state-of-the-art models, including linear regression, support vector machines, and long short-term memory neural networks. ResultsOur findings revealed ChatGPT’s remarkable accuracy in sentiment classification, with GPT-3.5 and GPT-4, respectively, achieving 92.1% (5682/6169) and 95.3% (5880/6169) accuracy in identifying positive, neutral, and negative sentiment, thereby outperforming the traditional lexicon-based method, which had an accuracy of 37.2% (2295/6169), and the 3 machine learning models, which had accuracies ranging from 66% (4072/6169) to 70.9% (4374/6169). ConclusionsAmong many text analysis techniques, ChatGPT demonstrates superior accuracy and emerges as a promising tool for Cantonese sentiment analysis. This study also highlights ChatGPT’s applicability in real-world scenarios, such as monitoring the quality of text-based counseling services and detecting message-level sentiments in vivo. The insights derived from this study pave the way for further exploration into the capabilities of ChatGPT in the context of underresourced languages and specialized domains like psychotherapy and natural language processing.

Himeshwari Verma, Devpriya Lakra, Vyom Agarwal

Introduction: HIV (Human Immunodeficiency Virus) continues to be a major global public health issue with no cure. Vitamin D is a fat-soluble hormone that is majorly involved in the classical function of calcium and phosphorus hemostasis and bone mineralization as well as non-classical functions of immune modulation in various viral and autoimmune diseases. A combination of both traditional risk factors, HIV- specific and antiretroviral therapy (ART)-specific contributors leave HIV-infected persons (PLHIV) at a greater risk for low 25-OH-Vitamin D levels and frank vitamin D deficiency. Aims and Setting: The current study was conducted to assess and characterize the prevalence of Vitamin D deficiency in PLHIV-on-ART attending a tertiary care hospital and assess the factors that may be affecting it. Methods: 95 PLHIV registered at an ART center were selected over a period of 6 months based on Inclusion and Exclusion criteria. Flow cytometry estimation of CD4 count and ELISA based quantitative assessment of serum 25-OH Vitamin D3 were done along with detailed clinical examination. P<0.05 was considered to be statistically significant. Results: About half of the PLHIV assessed were deficient in vitamin D. Severe vitamin D deficiency was noted in one-fourth of subjects. Serum vitamin D levels were significantly less in subjects on ZLN regime compared to TLE regime. No significant difference was found between vitamin D deficiency and duration of treatment, different treatment regimens or differing CD4 counts. No significant association of serum levels of Vitamin D with duration of treatment or varying CD4 count was found. Conclusion: There is greater prevalence of subnormal levels of Vitamin D in PLHIV on ART. ZLN regime appears to have a negative impact on Vitamin D levels in comparison to TLE regimen. More research needs to be done to further evaluate the physiology of Vitamin D in PLHIV on ART.

Mingli Pang, Jieru Wang, Mingyue Zhao et al.

BackgroundDriven by the accelerated aging of the population of China, the number of older adults has increased rapidly in the country. Meanwhile, following children, migrant older adults (MOA) have emerged as a vulnerable group in the process of fast urbanization. Existed studies have illustrated the association between social support and loneliness and the relationship between sleep disturbance and loneliness; however, the underlying mechanisms and the migrant-local difference in the association between social support, sleep disturbance, and loneliness have not been identified. ObjectiveThis study aimed to clarify the migrant-local difference in the relationship between social support, sleep disturbance, and loneliness in older adults in China. MethodsMultistage cluster random sampling was used to select participants: 1205 older adults (n=613, 50.9%, MOA and n=592, 49.1%, local older adults [LOA]) were selected in Weifang City, China, in August 2021. Loneliness was assessed with the 6-item short-form University of California, Los Angeles Loneliness Scale, social support was evaluated with the Social Support Rating Scale, and sleep disturbance was measured with the Pittsburgh Sleep Quality Index. The chi-square test, t test, and structural equation modeling (SEM) were adopted to explore the migrant-local difference between social support, sleep disturbance, and loneliness among the MOA and LOA. ResultsThe mean score of loneliness was 8.58 (SD 3.03) for the MOA and 8.00 (SD 2.79) for the LOA. SEM analysis showed that social support exerts a direct negative effect on both sleep disturbance (standardized coefficient=–0.24 in the MOA and –0.20 in the LOA) and loneliness (standardized coefficient=–0.44 in the MOA and –0.40 in the LOA), while sleep disturbance generates a direct positive effect on loneliness (standardized coefficient=0.13 in the MOA and 0.22 in the LOA). ConclusionsBoth MOA and LOA have a low level of loneliness, but the MOA show higher loneliness than the LOA. There is a negative correlation between social support and loneliness as well as between social support and sleep disturbance among the MOA and LOA (MOA>LOA), while loneliness is positively associated with sleep disturbance in both populations (MOA<LOA). Measures should be taken by the government, society, and families to increase social support, decrease sleep disturbance, and further reduce the loneliness among older adults, especially the MOA.

Evgeny S. Saveliev, Mihaela van der Schaar

TemporAI is an open source Python software library for machine learning (ML) tasks involving data with a time component, focused on medicine and healthcare use cases. It supports data in time series, static, and eventmodalities and provides an interface for prediction, causal inference, and time-to-event analysis, as well as common preprocessing utilities and model interpretability methods. The library aims to facilitate innovation in the medical ML space by offering a standardized temporal setting toolkit for model development, prototyping and benchmarking, bridging the gaps in the ML research, healthcare professional, medical/pharmacological industry, and data science communities. TemporAI is available on GitHub (https://github.com/vanderschaarlab/temporai) and we welcome community engagement through use, feedback, and code contributions.

Zhong Zhang, Jian-liang Wu, Cun-quan Qu et al.

The dissemination of information and the development of public opinion are essential elements of most social media platforms and are often described as distinct, man-made occurrences. However, what is often disregarded is the interdependence between these two phenomena. Information dissemination serves as the foundation for the formation of public opinion, while public opinion, in turn, drives the spread of information. In our study, we model the co-evolutionary relationship between information and public opinion on heterogeneous multiplex networks. This model takes into account a minority of individuals with steadfast opinions and a majority of individuals with fluctuating views. Our findings reveal the equilibrium state of public opinion in this model and a linear relationship between mainstream public opinion and extreme individuals. Additionally, we propose a strategy for regulating public opinion by adjusting the positions of extreme groups, which could serve as a basis for implementing health policies influenced by public opinion.

Nicholas Vincent, David Bau, Sarah Schwettmann et al.

Can governments build AI? In this paper, we describe an ongoing effort to develop ``public AI'' -- publicly accessible AI models funded, provisioned, and governed by governments or other public bodies. Public AI presents both an alternative and a complement to standard regulatory approaches to AI, but it also suggests new technical and policy challenges. We present a roadmap for how the ML research community can help shape this initiative and support its implementation, and how public AI can complement other responsible AI initiatives.

Adelaida Lozano Polo, Pedro Simón Cayuela Fuentes, Josep Sánchez Monfort et al.

Introduction Nurses have an essential role in smoking cessation. The Murcia Society of Family and Community Nursing (SEAPREMUR) launched a research project (GRUPALTAB-SEAPREMUR) to analyze the effectiveness of various group interventions to quit smoking, based on health education techniques and whith the App S’Acabó designed by the Spanish Society of Tobacco Specialists (SEDET). Nurses were trained in the protocol used in the interventions. Objective To compare the effectiveness of group interventions to quit smoking in the short and long term. Analyze the utility of the App to encourage smoking cessation. Material and Methods Multicenter randomized clinical trial conducted in Primary Care Center (PCC) in the Region of Murcia from 2018 to 2020 in two phases (P). Inclusion criteria: Being over 18 years old, wanting to quit smoking, speaking Spanish, having Internet access. Exclusion: Polydrug use, pregnant women or psychiatric pathology. The sample size was calculated and randomly assigned to the type of intervention. Interventions: Workshop of 2 to 4 hours inly session vs Course of 4 sessions of 2 hours duration for 1 month. The smoking abstinence at three months and one year (prevalence and OR; 95%CI) is calculated with SPSSV21, comparing the population that uses or not the App. Results In 2018, the study (P1) started in 8 PPC 228 participants: 54.2% women. They were followed-up for one year: 83 (46.1%). In 2019, the second phase (P2) was carried out in 16 PCC, with a shorter workshop (2:30h) and the same course (296 participants; 59.2% women) Global abstinence at 3 months (P1: 23.8%; P2: 20.9%) and at 12 months (P1: 31.7%). No significant statistical differences were observed in smoking cessation by sex, social class, or type of intervention, although abstinence was lower in the workshop: - P1 Workshop vs Course. OR at 3 months: 0.89 (95%CI: 0.38-2.08) OR at 12 months: 0.83 (95%CI: 0.32-2.57). - P2 Workshop vs Course. OR at 3 months: 0.61 (95%CI: 0.29-1.29). Use of App (P1: 42.1%; P2: 42.6%). An increase in quit attempts was observed in those who used the App compared to those who did not (P1 at 12 months: (37.7% vs 14.3%; p=0.047); P2 at 3 months: (54.5% vs 45.5%, p=0.01). Conclusions 1. Group smoking cessation interventions conducted by nurses are effective. 2. No significant differences were observed by type of group intervention. 3. The use of the App promotes quit attempts.

Alvaro Andres Alvarez Peralta, Priya Desai, Somalee Datta

This manuscript explores linking real-world patient data with external death data in the context of research Clinical Data Warehouses (r-CDWs). We specifically present the linking of Electronic Health Records (EHR) data for Stanford Health Care (SHC) patients and data from the Social Security Administration (SSA) Limited Access Death Master File (LADMF) made available by the US Department of Commerce's National Technical Information Service (NTIS). The data analysis framework presented in this manuscript extends prior approaches and is generalizable to linking any two cross-organizational real-world patient data sources. Electronic Health Record (EHR) data and NTIS LADMF are heavily used resources at other medical centers and we expect that the methods and learnings presented here will be valuable to others. Our findings suggest that strong linkages are incomplete and weak linkages are noisy i.e., there is no good linkage rule that provides coverage and accuracy. Furthermore, the best linkage rule for any two datasets is different from the best linkage rule for two other datasets i.e., there is no generalization of linkage rules. Finally, LADMF, a commonly used external death data resource for r-CDWs, has a significant gap in death data making it necessary for r-CDWs to seek out more than one external death data source. We anticipate that presentation of multiple linkages will make it hard to present the linkage outcome to the end user. This manuscript is a resource in support of Stanford Medicine STARR (STAnford medicine Research data Repository) r-CDWs. The data are stored and analyzed as PHI in our HIPAA-compliant data center and are used under research and development (R&D) activities of STARR IRB.

Arman Rahmim, Tyler J. Bradshaw, Irène Buvat et al.

The SNMMI Artificial Intelligence (SNMMI-AI) Summit, organized by the SNMMI AI Task Force, took place in Bethesda, MD on March 21-22, 2022. It brought together various community members and stakeholders from academia, healthcare, industry, patient representatives, and government (NIH, FDA), and considered various key themes to envision and facilitate a bright future for routine, trustworthy use of AI in nuclear medicine. In what follows, essential issues, challenges, controversies and findings emphasized in the meeting are summarized.

Wenhao Qi, Fang Liu, Tian Zhang et al.

The New Rural Cooperative Medical System (NRCMS) is one of the essential systems for ensuring public health in rural China. This paper investigates the effect of farmers' participation in the NRCMS on their subjective well-being and its mechanisms using data from the Chinese General Social Survey 2017. The results show that farmers' participation in the NRCMS significantly enhances their subjective well-being, and these results remain robust after regression with the instrumental variables method and propensity score matching method. Further analysis of the mechanisms suggests that participation in the NRCMS can enhance farmers' subjective well-being by increasing their consumption levels other than medical consumption. Moreover, medical consumption levels play a negative role in participating in the NRCMS on farmers' subjective well-being, which can be explained as the “masking effect.” The regression results of the subsamples show that the higher a farmer's income is, the less his or her participation in the NRCMS enhances subjective well-being. And the effect of participation in the NRCMS on farmers' subjective well-being is not significant if their health status is too high or too low.

Chi Zhang, Chi Zhang, Wei Fang Liao et al.

ObjectiveCOVID-19 has caused great loss of human life and livelihoods. The dissemination of health information in online social networks increased during the pandemic's quarantine. Older people are the most vulnerable group in sudden public health emergencies, and they have the disadvantage of infection rates and online search for health information. This study explores the relationship between the health risk perception and health information search behavior of older people in social networks, to help them make better use of the positive role of social networks in public health emergencies.MethodBased on the Risk Information Search and Processing model, and in the specific context of COVID-19, this study redefines health risk perception as a second-order construct of four first-order factors (perceived probability, perceived severity, perceived controllability, and perceived familiarity), and constructs a research model of the health risk perception and health information search behavior of older people. An online survey of people over 55 years old was conducted through convenience sampling in China from February 2020 to March 2020.ResultsA total of 646 older adults completed the survey. The structural equation model showed that health risk perception is a second-order factor (H1), that health risk perception has significant positive effects on health information search behavior (H2: β = 0.470, T = 11.577, P &lt; 0.001), and that health risk perception has significant positive effects on affective response (H3: β = 0.536, T = 17.356, P &lt; 0.001). In addition, affective response has a significant positive mediating effect on information sufficiency (H4: β = 0.435, T = 12.231, P &lt; 0.001), and information sufficiency has a significant positive mediating effect on health information search behavior (H5: β = 0.136, T = 3.081, P = 0.002).ConclusionThe study results indicate that the health risk perception of older people during the COVID-19 outbreak not only directly affected their health information search behavior, but also had an indirect impact on their health information search behavior by affecting affective response and information sufficiency.

Mengbi SHEN, Zixia WANG, Jiaqi ZHENG et al.

BackgroundExercise during pregnancy is closely related to maternal and infant health. Previous studies in developed countries have linked maternal exercise during pregnancy with newborn body weight as well as subcutaneous fat thickness. However, the relevant studies in China are limited, and the conclusions remain inconsistent. ObjectiveTo investigate the effects of maternal exercise during pregnancy on neonatal weight and subcutaneous fat thickness. MethodsBased on the Shanghai Birth Cohort, 959 maternal-infant pairs were included in this study. The International Physical Activity Questionnaire was used to collect average weekly frequency and daily minutes of walking in the first and second trimesters, and entropy weight method was used to calculate the cumulative exercise index in the two trimesters. Birth weight was measured using a calibrated weigh scale. Subcutaneous fat thickness was measured at abdomen, scapula, and triceps with a Harpenden skinfold caliper for all newborns and the sum of the thickness for the three sites was then calculated. A multiple linear regression model was employed to estimate the relationships of cumulative exercise index during pregnancy with neonatal body weight and subcutaneous fat thickness. Subgroup analyses stratified by pre-pregnancy body mass index (BMI) and sex of newborns were also performed. ResultsThe mean age of pregnant women was (28.5±3.8) years, and the pre-pregnancy BMI was (21.4±3.0) kg·m−2. Newborn boys were slightly more than newborn girls (54.3% vs 45.7%), and the neonatal weight was (3374.0±427.5) g. The means of newborns' abdominal, scapular, and triceps subcutaneous fat thickness were (4.4±1.3), (5.4±1.4), and (6.0±1.5) mm, respectively, and the sum of subcutaneous fat thickness was (15.8±3.9) mm. In the first and second trimesters, 77.3% and 88.7% of pregnant women walked 4 d per week and more, respectively; the daily minutes of walking was (36.9±27.2) min and (43.3±26.3) min, respectively; the cumulative exercise index was 25.6±17.7 and 35.9±21.1, respectively. The results of multiple linear regression analysis showed that the cumulative exercise index in the second trimester was negatively associated with newborns' abdominal (b=−0.006, 95%CI: −0.010-−0.003), scapular (b=−0.005, 95%CI: −0.009-−0.002), triceps (b=−0.006, 95%CI: −0.010-−0.002), and their sum of (b=−0.018, 95%CI: −0.028-−0.007) subcutaneous fat thickness (P<0.05); in the first and second trimesters, however, the relationship between maternal cumulative exercise and newborns' body weight was not significant. The results of stratified analyses showed that the negative associations between maternal cumulative exercise index and newborns' subcutaneous fat thickness for the second trimester remained significant in the subgroups of boys and neonates whose mothers had normal pre-pregnancy BMI (P<0.05). ConclusionCumulative exercise index in the second trimester is negatively correlated with the neonatal thickness of subcutaneous fat, and the association may be altered by neonatal sexes and maternal pre-pregnancy BMI levels.