共找到 20 条结果
This study is a cross-sectional analysis aimed at systematically evaluating the information quality and reliability of popular science content related to drug-induced liver injury (DILI) on the two major Chinese video platforms, TikTok (Douyin) and BiliBili, and analyzing its content characteristics. On December 20, 2025, searches were conducted in the Chinese versions of the TikTok (Douyin) and BiliBili mobile applications using "" as the single keyword. Exclude content that does not meet the requirements based on the exclusion criteria, and ultimately retain the top 100 videos from each platform that meet the standards for analysis (N = 200). Two trained reviewers independently performed blinded assessments using the Global Quality Score (GQS), Journal of the American Medical Association (JAMA) benchmark criteria, and the modified DISCERN (mDISCERN) tool. Non-parametric tests were used to compare differences between groups, and Spearman correlation analysis was employed to explore the relationship between video characteristics and quality scores. User engagement metrics (likes, favorites, shares) for TikTok (Douyin) videos were significantly higher than those for BiliBili (p < 0.001). In terms of information quality, TikTok (Douyin) videos scored significantly higher than BiliBili on the GQS, JAMA, and mDISCERN scales (p < 0.001). There were differences in quality across content types: videos on "medication knowledge" received the highest mDISCERN reliability scores and "disease knowledge" videos scored higher in GQS practicality. Correlation analysis showed a weak positive correlation between user engagement metrics and mDISCERN scores. Among the 107 videos mentioning liver injury-related drugs, chemical drugs (antibacterial, chemotherapeutic, anti-tuberculosis drugs, and anti-inflammatory drugs) and traditional Chinese medicines (such as He Shou Wu) were mentioned most frequently. In this cross-sectional sample, TikTok (Douyin) videos demonstrated higher quality scores and user engagement than those on BiliBili, while professionals outperformed general users only on the JAMA criteria. Although some videos mentioned medications associated with liver injury, the information was generally oversimplified and biased toward trending topics. Hence, active information seekers should critically appraise the scientific soundness of medical short videos on platforms like TikTok and BiliBili before making healthcare decisions.
Diagnostic and treatment delays in infantile epileptic spasms syndrome (IESS) increase the risk of poor neurodevelopmental outcomes. Early clinical recognition of IESS is essential, especially in regions lacking expedited access to electroencephalograms (EEG). This study aimed to determine clinicians' accuracy at recognizing infantile epileptic spasms (ES) based on smartphone videos, and the impact of brief IESS education on accuracy, diagnostic confidence, and willingness to treat without EEG. This multicenter prospective cohort study took place over seven sessions globally from 2022 to 2023. Smartphone videos of children from the US and South Africa with EEG-confirmed diagnoses of IESS (6 videos) and non-epileptic ES-mimickers (3 videos) were obtained. Staff physicians and trainee participants from multiple subspecialties worldwide viewed videos three times: (1) baseline viewing, (2) after brief IESS training, and (3) with clinical history. Surveys on diagnosis and management were completed after each viewing. Of 187 participants who attended a session and initiated a survey, 180 (80 trainees [44%]) met the inclusion criteria. Initial diagnostic accuracy averaged 64% (95% confidence interval [CI]: 62-66%) and improved to 72% (69-74%) after IESS training and clinical history (V + T + CHx). Area under the curve for diagnostic performance of smartphone videos was 0.80 (0.78-0.82), and sensitivity was 0.85 (0.83-0.88) after V + T + CHx. The odds of making a correct diagnosis increased by 86% (OR 1.86, CI 1.59-2.18, p < 0.001) after V + T + CHx. Diagnostic confidence and clinician comfort level treating ES without EEG also improved significantly after V + T + CHx (by 0.36 points and 0.45 points, respectively, on 5-point Likert scales, p < 0.001). Diagnostic accuracy correlated strongly with increased diagnostic confidence and increased clinician comfort level managing patients without an EEG (p < 0.001). Staff physicians had a 24% higher likelihood of making a correct diagnosis than trainees. Smartphone videos, especially when enhanced by brief IESS training, can facilitate triage and early identification of infantile ES, reducing diagnostic delays in this time-sensitive condition. Infantile epileptic spasms syndrome is associated with severe developmental impacts, which can be worsened by delayed treatment. Rapid diagnosis is critical, especially in resource-limited settings lacking specialists and timely access to diagnostic tests. Our study found that clinician participants identified epileptic spasms, the hallmark seizure type of this condition, based on video alone with moderately high accuracy, and accuracy improved after education and clinical information. Thus, smartphone videos, particularly when enhanced by brief training, may be an effective tool to triage movements concerning for epileptic spasms, potentially improving resource allocation and reducing diagnostic delays in this urgent childhood epilepsy condition.
With the increasing demand for video processing in both human perception and machine vision applications, enhancing heavily compressed video has become a critical problem in practical multimedia systems. In many real-world scenarios, video data acquired by image sensors are often compressed for efficient transmission and storage, which introduces compression artifacts and degrades both visual quality and downstream task performance. This issue is especially significant in sensor-based systems such as surveillance cameras and mobile imaging devices. To address these challenges, we propose a novel joint human-machine video enhancement framework for compressed video enhancement that jointly targets human perceptual quality and machine vision performance. The framework integrates four complementary components: a Spatio-Temporal Fusion Module that leverages inter-frame correlations, a High-Frequency Semantic Fusion module for recovering structurally important details relevant to machine tasks, a Texture-Guided Model that enhances low-level visual features, and a Refined Attention Residual Quality Enhancement Module that adaptively emphasizes salient regions. By progressively combining these modules, the framework effectively restores compressed content while preserving task-relevant semantics. The experimental results demonstrate that our method consistently outperforms existing approaches, achieving higher PSNR and SSIM as well as improved object detection and video object segmentation performance. These results highlight the framework's practical applicability for compressed video enhancement in sensor-based systems, including intelligent surveillance and autonomous imaging platforms.
Long COVID, characterized by persistent symptoms such as fatigue and cognitive impairment, continues to pose a global public health burden. Bilibili and TikTok are major platforms for public health information; however, the lack of systematic evaluation contributes to low-quality information and may hinder effective health management. We aimed to evaluate the quality and reliability of long COVID-related videos on these platforms. This cross-sectional study analyzed long COVID-related videos from Bilibili and TikTok. Inter-rater reliability was assessed using Cohen's kappa coefficient. Quality and reliability were evaluated using the Global Quality Score (GQS) and modified DISCERN (mDISCERN). Multivariate linear regression was performed to identify the independent predictors of video quality. The median GQS and mDISCERN scores were both significantly higher for Bilibili than for TikTok (P = 0.016 and P = 0.039, respectively). Professional-source videos showed significantly higher quality than non-professional ones (P < 0.001). Regression analyses revealed that a professional source was the strongest independent predictor of both GQS and mDISCERN scores (all P < 0.001). Video duration and shares were significantly, albeit weakly, associated with the GQS, whereas other engagement metrics were not. The overall quality of long COVID videos was suboptimal. Professional source was the primary independent predictor of higher quality, while engagement and duration showed limited influence. Strengthening platform review mechanisms and promoting evidence-based information are needed to improve public health communication.
Oral diseases have an estimated incidence of 35.5% worldwide, yet access to dental care is often hindered by a shortage of dental professionals and limited diagnostic resources. While Artificial Intelligence (AI) has shown promise in dental diagnostics, most existing models rely on high-quality static images captured in controlled clinical settings. This study aimed to develop and validate Deep Learning (DL) models capable of detecting dental findings and identifying tooth types using video data from a handheld, toothbrush-mounted intraoral device. In this validation study, 708 videos were captured using an electronic toothbrush integrated with a miniature intraoral camera. A custom python script extracted 16,552 image frames, which were filtered to 7,963 eligible frames for the dental-findings model and 3,799 frames for the tooth-type model. Two dental experts annotated findings, including decay, fillings, plaque, and staining as well as tooth types using bounding boxes. The YOLOv8s architecture was employed for both the models. The dental-findings model achieved an overall Mean Average Precision (mAP) of 83.0% and an F1-score of 77.9%. It performed best in detecting amalgam fillings (94.5% mAP) and decay (84.2% mAP). The tooth-type model demonstrated an overall mAP of 88.6% and F-1 score of 86.2%. The highest mAP was observed for molars (97.7%) and premolars (96.7%). Despite the lack of a standardized video capture sequence, the models maintained strong performance across diverse visual variations. This study establishes a proof of concept for embedding AI diagnostics into everyday oral care devices. By accurately detecting clinical findings from user-captured video, this technology has the potential to facilitate remote monitoring, support early intervention, and optimize healthcare resource allocation, particularly in underserved regions.
Sound is a core component of digital games, and its integration is assumed to support learning, motivation, and positive emotions. However, empirical evidence on the role of sound in educational video games remains limited, particularly in narrative-driven educational adventure games such as digital history games. In a laboratory experiment, university students (N = 111) either played an educational history video game without sound or with additional sound features (ambient audio, character voices, and narrated codex entries providing additional historical information). Post-test measures assessed factual knowledge, triggered and maintained situational interest, and academic emotions (enjoyment and boredom). Engagement with optional supplemental historical information provided through in-game codex entries was measured using behavioral log data. We analyzed differences between the two conditions while controlling for relevant pre-test variables. Participants in the add-on sound condition did not score significantly higher on the knowledge test than those in the no-sound condition. Likewise, no statistically significant differences emerged in situational interest, enjoyment, boredom, or codex engagement between conditions. Additional analyses indicated that participants' interactions with codex entries positively predicted knowledge test performance, indicating that voluntary engagement with supplemental content contributed to learning. Our findings suggest that the presence of sound alone may not enhance academic outcomes in a narrative-driven educational video game. Additionally, our findings indicate that learning outcomes depended strongly on learners' engagement with in-game codex entries. Overall, our results on the inclusion of sound highlight the importance of examining specific design features within educational history video games.
Action recognition in videos is an important task in computer vision, widely used in sports, healthcare, and human-computer interaction. Existing methods often struggle to balance global motion understanding and local detail extraction, especially when dealing with rapid transitions or combinations of multiple actions. This paper proposes a new action recognition model, ViT-ConvGAN, which integrates Video Transformer (ViT), 3D Convolutional Neural Networks (CNN), and Conditional Generative Adversarial Networks (CGAN). Input video frames are first processed by ViT, which captures long-term temporal dependencies via spatiotemporal attention to generate global spatiotemporal features; these global features are then fed into the 3D CNN, which refines local motion details and fuses them with the global features to form a comprehensive feature map; finally, the fused feature map is transmitted to the CGAN-where the generator optimizes feature representation for more discriminative action characteristics, and the discriminator enhances the distinction between different action categories to improve prediction accuracy. ViT models long-term temporal dependencies through spatiotemporal attention, 3D CNN extracts local motion features, and CGAN optimizes action predictions to enhance the reliability of classification results. The model excels in capturing both global and local motion patterns, especially for complex action sequences. Experiments on the UCF101 and Kinetics-400 datasets show that ViT-ConvGAN achieves 87.3% and 95.2% Top-1 accuracy, respectively, with strong performance on Kinetics-400, surpassing several state-of-the-art models. Ablation studies confirm the contribution of each module, particularly the critical role of ViT and 3D CNN in feature extraction. ViT-ConvGAN provides an efficient solution, improving complex action recognition performance and offering new insights for model architecture design in action analysis.
The limited availability of diverse and representative training data poses a critical barrier to the development of clinically relevant computational tools for intraoperative surgical decision support. Surgical procedures are not routinely recorded, and data annotation requires domain expertise, resulting in a scarcity of open-access surgical video datasets with high-quality annotations. Existing datasets are typically limited to single institutions and specific procedures, such as cholecystectomy, and rarely comprise patient-level metadata like demographic characteristics, disease history, or laboratory parameters. The Appendix300 dataset comprises 330 laparoscopic surgery recordings, including 325 full-length laparoscopic appendectomies and 5 control recordings from non-appendectomy procedures in pediatric and adult patients treated at five German centers. The dataset includes patient-level clinical metadata (demographics, medical history, clinical symptoms, preoperative laboratory parameters, and histopathologic findings, as well as standardized expert annotations of the laparoscopic grade of appendicitis). This dataset enables novel validation tasks for computer vision in laparoscopic surgery and facilitates simulation of decentralized learning approaches, overall enhancing the breadth and translational relevance of AI-based surgical video analysis.
Multi-view video recordings are increasingly used to capture the 3D movements of animals in experimental settings, yet extracting rich 3D representations from these recordings remains challenging. Supervised pose estimation requires extensive manual annotation, while general-purpose 3D reconstruction models trained on generic scene datasets fail on the specialized imagery and sparse-view setting of laboratory experiments. We address these limitations with BEAST3D, a self-supervised pretraining framework that learns 3D visual representations from unlabeled, calibrated multi-view video. BEAST3D uses a vision transformer to predict 3D Gaussian splats that reconstruct held-out views through differentiable rendering, while simultaneously segmenting the animal from the background. BEAST3D reconstructs 3D structure with as few as four views by conditioning directly on known camera parameters--unlike general-purpose models, which must estimate camera geometry from dense overlapping viewpoints that are seldom available in lab settings. Through comprehensive evaluation across four species, we demonstrate that BEAST3D produces rich, viewpoint-invariant features that transfer effectively to three downstream tasks: novel view synthesis, which validates the quality of the learned 3D representations; multi-view pose estimation, which provides the sparse keypoint trajectories widely used in behavioral analysis; and neural encoding, which relates 3D behavioral features to simultaneously recorded neural activity. BEAST3D thus establishes a versatile framework for behavioral analysis that leverages 3D structure in modern multi-view laboratory recordings.
暂无摘要（点击查看详情）
暂无摘要（点击查看详情）
暂无摘要（点击查看详情）
Personalized feedback improves the clinical pediatric behavior guidance performance of students but is prohibitively time-consuming to provide. Large language models (LLMs) can automate the process of evaluating clinical sessions but are limited to text-only input and consistency issues. This study compared the use of text-only transcripts against the use of video recordings for evaluating the clinical behavior guidance performance of dental students. Additionally, the consistency and accuracy of LLMs in evaluating the transcripts were compared against a human assessor. This study was conducted by using 40 video-recorded clinical encounters involving final-year dental students who were managing patients aged between 4 and 12 years at the Faculty of Dentistry, National University of Singapore. The videos were scored by using a previously validated pediatric behavior guidance scale. Clinical encounters were transcribed verbatim and scored by a study member using a modified version of the scale (nonverbal components removed). The time taken to rate the transcripts was recorded. Video scores were compared with transcript scores. Both the free-to-use version and the paid version of the ChatGPT LLM were also used to score the transcripts; consistency was evaluated and compared against the human assessor. The average time taken to rate the transcripts (mean 12, range 3-25 min) was significantly (P<.001) lower than the average video length (mean 73, range 37-120 min). Comparing transcript scores with video scores resulted in a consistency intraclass correlation coefficient of 0.830 (95% CI 0.679-0.910; P<.001), demonstrating good reliability. Comparing transcript scores with the free-to-use LLM's and paid LLM's scores yielded an absolute agreement intraclass correlation coefficient of 0.729 (95% CI 0.475-0.859; P<.001) and 0.670 (95% CI 0.377-0.825; P<.001), respectively, demonstrating moderate agreement. The LLMs were inconsistent, producing variable scores with the same prompt. The free-to-use and paid versions produced the same score for all 3 runs in only 7 (18%) and 4 (10%) of the 40 clinical encounters, respectively. Using transcripts to evaluate students' clinical behavior guidance was time-saving for faculty, demonstrated good agreement with video-based evaluation, and could improve clinical teaching. Although LLMs can automate the task, improvements are needed to improve their consistency and accuracy.
Health literacy is a resource that enables individuals to make health related choices to promote and protect their health and that of those around them. For parents, health literacy is essential to access and use information and services in ways that support their child's health. Immigrant parents may face health literacy challenges due to language barriers, differing approaches to managing child health and parenting, unfamiliar services, and divergent expectations of services and staff. Since parental health literacy is linked to child health outcomes, addressing the needs of immigrant parents may help prevent avoidable inequities in child health. Few studies have developed and tested interventions to promote health literacy among parents with immigrant backgrounds. Based on results from a needs assessment conducted in a culturally and linguistically diverse population in Oslo, Norway, we aimed to co-create an action to promote parental health literacy. We undertook a two-phase co-creation process drawing on methods from the Optimising Health Literacy and Access (Ophelia) Process and the James Lind Alliance Priority Setting Partnership. In phase one, we collected action ideas from a broad range of stakeholders; analysed and synthesised the ideas; facilitated prioritisation workshops with user representatives; and selected one idea for co-design. In phase two, we co-designed the action with user representatives; and conducted quality-improvement cycles in the clinical setting. In phase one, 14 immigrant parents and 59 staff from different disciplines generated 302 action ideas. Analysis reduced these to a short-list of 22 ideas which were prioritised by user representatives (parents and staff) resulting in two Top-10 lists. Five priorities overlapped and one of these was selected for development: improving communication on services provided by the family health clinic. In phase two, we operationalised this idea by co-designing short, multilingual, animated videos about follow-up at the clinic. The videos were refined through five iterative quality improvement cycles with input from 43 end users (parents and staff). We successfully engaged user representatives, stakeholders and end users across multiple stages of co-creation and co-designed a health literacy action. The videos developed were completed to the stage of feasibility testing in the clinical setting.
Given concerns that screen time may impact dietary habits, this study investigated the association between screen time and dietary intake among adolescents in the United States. We analyzed a prospective cohort (N = 6485, 47.3% female, age: 12 ± 0.7 years) from the Adolescent Brain Cognitive Development (ABCD) Study, using data from Year 2 (2018-2020) and Year 3 (2019-2021). Multinomial logistic regression models estimated the associations between participant-reported screen time (watching television shows and videos, playing video games, socializing, browsing the internet, and total screen time (hours/day)) and parent/participant-reported intake of various food/nutrient categories 1 year later (Year 3). We adjusted for age, sex, race and ethnicity, household income, parent education, average daily kilocalorie intake, respective food or nutrient, and study site (Year 2). Each additional hour of most screen time modalities was prospectively associated with higher odds of consuming fewer fruits, vegetables, whole grains, legumes, fiber, and dairy, and higher glycemic index, and higher odds of consuming more added sugars and a higher polyunsaturated fats ratio 1 year later. These findings highlight the need for parental guidance and clinical interventions to support screen time habits and promote healthy dietary choices among adolescents. This study examines the association between contemporary screen time modalities and dietary intake 1 year later in a demographically diverse U.S. sample of early adolescents. Most screen time modalities, such as total screen time and watching television shows and videos, were prospectively associated with higher odds of consuming fewer fruits, vegetables, whole grains, legumes, and fiber 1 year later. Greater total screen time and time spent socializing were prospectively associated with higher odds of a higher polyunsaturated fats ratio 1 year later.
Extraperitoneal access strategies have been successfully applied to left-sided colorectal resections, but their application to right-sided colectomy with complete mesocolic excision (CME) has not been described in accordance with the IDEAL framework. The EXPERTS (EXtraPEritoneal coloRecTal Surgery) with Complete Mesocolic Excision (CME) approach aims to utilise the superior oncological advantages of CME with the safety of extraperitoneal approach to central vascular structures. As part of IDEAL Stage 0 (preclinical development), a structured cadaveric Programme was undertaken to assess the anatomical feasibility, reproducibility and visualisation of critical structures during EXPERTS right hemicolectomy with CME. A total of 11 cadavers were dissected, including nine preserved using the Thiel method and two using fresh frozen preservation. All dissections were video recorded and analysed. In nine out of 11 cadavers (9 Thiel-preserved and 2 fresh frozen cadavers), consistent identification of key anatomical structures-including the right kidney, duodenum, superior mesenteric artery (SMA), superior mesenteric vein (SMV) and their major branches-was achieved via an extraperitoneal approach. Fresh frozen cadavers demonstrated more variable tissue quality and plane definition. High-quality operative videos were obtained for further analysis. This IDEAL Stage 0 study demonstrates the anatomical feasibility of an extraperitoneal approach to right hemicolectomy with CME and supports progression to assess its safety and effectiveness in a clinical setting.
Family mealtimes are promoted as a strategy to address childhood obesity; however, the joint associations of different mealtime elements with childhood obesity are not often considered. The current study investigated how mealtime family functioning and the nutritional quality of meals separately and jointly associate with childhood obesity. Typical mealtimes for 273 children (M age = 5.9 years, SD = 0.7 years) were video-recorded. Observed mealtime family functioning, specifically mealtime structure and behaviour management, was coded from the video-recorded mealtimes. The nutritional quality of meals was assessed using the Healthy Meal Index (HMI). Logistic regression models indicated no independent associations between mealtime family functioning or nutritional quality of meals and child obesity status. However, the nutritional quality of meals moderated the association between mealtime structure and child obesity status (Interaction beta -0.06, SE 0.03; p = 0.04). Stratified analyses indicated that greater mealtime structure was associated with a higher likelihood of obesity among children served low-nutritional-quality meals (OR = 4.17, 95% CI 1.13-15.50, p = 0.03). Greater mealtime structure may be associated with an increased risk of childhood obesity when meals have low nutritional quality. Thus, various mealtime elements may interact in association with childhood obesity.
To quantify patient preferences for key joint protection program (JPP) delivery components and identify preference segments to inform a patient-centred, technology-enabled program. Attributes and levels were developed via three focus groups with 16 people living with hand osteoarthritis (HOA). A discrete choice experiment (six attributes with two levels each; 16 choice tasks; three unlabelled alternatives; no opt-out) was fielded online. Choices were analysed using latent class choice models with effects coding. One to six class solutions were compared by Akaike Information Criterion/Bayesian Information Criterion and segment sizes. Within-class relative attribute importance was calculated from part-worth ranges and overall (class-adjusted) importance was computed using class membership probabilities. A total of 150 participants with HOA completed the survey (89% women; mean age 68 years). Latent class modelling identified five distinct preference segments (segment sizes: 34%, 18%, 10.7%, 21.3% and 16%). Across segments, clinician-led question and answer (Q&A; class-adjusted importance 23.7%) and clinician-delivered exercise demonstrations (21.5%) consistently drove choices. Other features showed marked heterogeneity: one large segment strongly preferred short (2-7 min) videos (importance 56.1%) while others were indifferent to video length; interaction mode split respondents between asynchronous forums and live monthly meetings; and lived-experience guest speakers were valued by some segments but actively traded off by others. Quizzes were conflicting, adding value in some classes and reducing it in others. Delivery preferences were heterogeneous, but clinician involvement Q&A and clinician-led demonstrations anchored preferences. The remaining features functioned as tuneable options. A configurable JPP that preserves clinician touchpoints and allows patients to select preferred formats is best aligned with revealed preferences.
In this article, we reflect on the preparatory phase of a multispecies research project focused on dog-human care relationships. Playing with creative posthumanist methodologies that seek to decenter the human, we attached light-weight videorecording devices to our companion animals' collars. As we approach a dogs-eye view of our everyday lives and interactions, we think with Jacques Derrida to ask: What does it mean to respond when met with the animal's gaze? Through unsettling our gaze, the videos take us somewhere else entirely, raising another question: What do attempts to tangibly and imaginatively see with the dog do? The unsettling of our own reflections-on the question of the animal and the gaze of the "Other"-offers a space for enacting Haraway's conception of response-ability, for moving with our impetus to respond despite our current situated involvement in neoliberalism and settler colonialism with their commodification and domestication of more-than-human life.
Non-communicable diseases (NCDs), such as hypertension, diabetes, and asthma, require continuous medication management. However, medication adherence remains suboptimal. Telepharmacy-defined as pharmacist-led care delivered remotely via telephone, video, or digital platforms-may improve adherence and clinical outcomes while addressing access barriers, but uncertainty remains regarding clinical effectiveness and generalisability. A systematic review is warranted to assess whether telepharmacy improves medication adherence, safety, and other key outcomes compared with usual care. To assess the clinical effectiveness of telepharmacy services, compared with usual care, on medication adherence and clinical outcomes in patients with NCDs in ambulatory care settings. We searched CENTRAL, MEDLINE, Embase, Global Index Medicus, and two trial registries up to 15 December 2025. We also assessed the reference lists of included studies and relevant reviews, conducted citation searching, and contacted study authors to clarify information and identify additional data. No language or publication status restrictions were applied. We included individually randomised controlled trials (RCTs) and cluster-RCTs comparing pharmacist-led telepharmacy with usual care for people with NCDs (e.g. cardiovascular disease, diabetes, and cancer) in ambulatory care settings. Critical outcomes were medication adherence, patients' satisfaction, and drug-related problems (DRPs). Important outcomes included mortality rate, worsening of NCDs, clinical measurements, laboratory values, patients' quality of life, healthcare use, and economic outcomes. We included seven outcomes in the summary of findings table. We assessed the risk of bias for the seven outcomes in the summary of findings table using the Cochrane RoB 2 tool, incorporating both individually RCTs and cluster-RCTs. We conducted synthesis analyses using random-effects models, calculating summary risk ratios or mean differences (MDs)/standardised mean differences (SMDs) with 95% confidence intervals (CIs). For cluster-RCTs, we used adjusted estimates or applied design effect corrections. Where meta-analysis was not feasible, we used narrative synthesis. We assessed the certainty of the evidence using GRADE. We included 21 trials (17 individually RCTs and 4 cluster-RCTs) involving a total of 5440 participants with NCDs. Sample sizes ranged from 20 to 1400 participants. Studies were conducted in high-, upper-middle-, and lower-middle-income countries, across hospital, clinic, pharmacy, or insurer-based settings. Interventions targeted conditions such as diabetes, hypertension, and asthma. Telepharmacy interventions varied in delivery modes (e.g. telephone, video, and app), intensity, and components (e.g. adherence support, monitoring, and education). Follow-up durations ranged from one to 18 months, with most studies lasting 12 months or less. Telepharmacy interventions may improve medication adherence compared with usual care (SMD 0.32, 95% CI 0.10 to 0.55; 10 studies, 2978 participants; low-certainty evidence). For patients' satisfaction, the evidence is very uncertain about the effect of telepharmacy interventions compared with usual care (SMD 0.37, 95% CI -0.11 to 0.85; 3 studies, 422 participants; very low-certainty evidence). One additional study using a 5-point Likert scale reported little to no difference between groups (96.5% versus 97.5%; P = 0.68). Another study lacked a comparator group, and we excluded it from the synthesis. We did not pool the evidence for DRPs due to clinical and methodological heterogeneity. Narrative findings from individual studies showed that one study reported increased detection of DRPs. Other studies reported fewer adverse events, suggesting prevention of DRPs, while the remaining studies found no clear differences. The certainty of the evidence was low. Regarding important outcomes, two studies reported worsening of NCDs. Due to clinical heterogeneity, we did not pool the results and presented them narratively. The effect of telepharmacy on worsening of NCDs remains uncertain. For asthma control, no clear difference was observed (SMD 0.23, 95% CI -0.34 to 0.80; 2 studies, 318 participants). Telepharmacy interventions may reduce systolic blood pressure (SBP) (MD -6.82 mmHg, 95% CI -12.16 to -1.48; 5 studies, 1254 participants; low-certainty evidence) and may reduce diastolic blood pressure (DBP) (MD -2.50 mmHg, 95% CI -4.80 to -0.20; 5 studies, 1254 participants; low-certainty evidence) compared to usual care. Two additional studies reporting clinical measurements found more pain relief with telepharmacy in one study, and no clear difference in thromboembolic events in the other. For glycated haemoglobin (HbA1c), telepharmacy interventions probably have little or no effect (MD -0.10%, 95% CI -0.25 to 0.05; 5 studies, 1771 participants; moderate-certainty evidence). For LDL cholesterol, a meta-analysis of two studies showed no clear difference between the groups (MD -0.84 mg/dL, 95% CI -4.70 to 3.02; 2 studies, 444 participants). One study reported better prothrombin time-international normalised ratio (INR) control in the intervention group. Three studies assessed quality of life using different tools, but did not show consistent evidence of benefit. Hospital admissions and emergency department visits showed no clear differences between groups. Two studies evaluated economic outcomes, with one reporting cost savings and the other showing no difference in total or disease-related costs. No included studies reported data on mortality rate or adverse events attributable to telepharmacy, so potential harms remain uncertain. Low-certainty evidence suggests that telepharmacy interventions may improve medication adherence, and may reduce both SBP and DBP in patients with NCDs in ambulatory care settings compared to usual care. Moderate-certainty evidence indicates telepharmacy interventions probably have little or no effect on HbA1c. The evidence is very uncertain about the effect of telepharmacy interventions on patients' satisfaction. The evidence base is limited by short follow-up periods, variation in interventions and outcome measures, and lack of equity-related data. Telepharmacy appears promising for ambulatory care, but further high-quality trials with standardised adherence measures and longer follow-up are needed to clarify effectiveness, implementation potential, and equity impacts. Takeshi Hasegawa and Hisashi Noma were supported by a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (Grant numbers: JP24K06239 and JP23K24811). Protocol (2023): DOI 10.1002/14651858.CD015136.