Search Results (1,705)

Previously, food processing and nutrition sciences operated as distinct scientific domains; however, increasing health concerns have highlighted the need for integrated perspectives and practical food classifications. This study aimed to assess participants’ knowledge of food processing classification according to the NOVA system and investigate its association with ultra-processed food (UPF) consumption patterns and nutrient intake. A cross-sectional study of 403 healthy students (18–30 years) was conducted. Participants were asked to classify each food according to its level of processing. Dietary intake was assessed in a subsample of 190 participants who completed the 24-h dietary recall assessment and had complete dietary data available. Higher food processing knowledge was associated with lower UPF consumption (B = −0.241, 95% CI: −0.414 to −0.068; p = 0.006). Participants with lower knowledge exhibited higher mean consumption of bread and bakery products, breakfast cereals, savory snacks, dairy products, ready-to-eat meals, and processed meats. After adjustment for age, no significant differences in nutrient intake were observed between knowledge groups. Full article

The overall prevalence of dyslipidemia among adults continues to rise. However, limited attention has been devoted to the association between free sugar intake and dyslipidemia. This study aimed to explore the association between free sugar intake and dyslipidemia, as well as its subtypes, in Chinese adults. A total of 38,656 adults from the Chinese Food Consumption Survey (CFCS) and 3974 from the China Health and Nutrition Survey (CHNS) were included in the final analysis. Dietary intake data were collected by 24 h dietary recalls and the weighing of household foods and condiments. For the CHNS, multivariable Cox proportional hazards regression models were used to examine the association between free sugar intake and the risk of dyslipidemia and its subtypes. Restricted cubic spline (RCS) regression was used to evaluate the nonlinear relationship between free sugar intake and dyslipidemia. Stratified analyses and interaction tests were performed across key covariates to explore potential effect modifiers of the association between free sugar intake and dyslipidemia risk. For the CFCS, the mean free sugar intake among Chinese adults was 6.4 g/d. For the CHNS, a total of 1052 individuals were diagnosed with dyslipidemia, corresponding to an incidence rate of 26.5% during a total of 24,531 person-years of follow-up (median follow-up, 6.0 years) in the CHNS cohort. After multivariable adjustment, the risk of dyslipidemia was increased by 38% with >20 g/d free sugar intake (HR = 1.38, 95% CI: 1.04–1.84, p for trend = 0.006), with <5 g/d free sugar intake as a reference. For different subtypes of dyslipidemia, with <5 g/d free sugar intake as a reference, the risk of hypo-HDL-cholesterolemia and hyper-LDL-cholesterolemia was increased by 91% (HR = 1.91, 95% CI: 1.34–2.74, p for trend < 0.001) and 52% (HR = 1.52, 95% CI: 1.01–2.29, p for trend = 0.004) with > 20g/d free sugar intake, respectively. Stratified analyses demonstrated that the impact of high free sugar intake on the development of hypo-HDL-cholesterolemia and hyper-LDL-cholesterolemia was significantly amplified among urban residents and those with insufficient physical activity. While linear dose–response associations were observed between free sugar intake and the risk of dyslipidemia (p for nonlinearity > 0.05), a nonlinear, inverted U-shaped association emerged for hyper-LDL-cholesterolemia (p for nonlinearity = 0.007). Our study demonstrates that free sugar intake levels in Chinese adults were relatively low compared to the recommended intake level (<50 g/d) in the Chinese Dietary Guidelines 2022. Nevertheless, increased free sugar intake was positively associated with a higher risk of incident dyslipidemia, as well as incident hypo-HDL-cholesterolemia and hyper-LDL-cholesterolemia. Targeted nutrition education is recommended to promote balanced dietary patterns and enhance overall dietary quality, thereby effectively reducing the risk of dyslipidemia. Full article

Cattle producers must balance animal welfare with productivity, but regular observation of animals in extensively managed operations is often impractical. Artificial intelligence (AI) integrated with computer vision offers an automated alternative, but few studies compare object detection architectures within the same dataset or focus on pastured cattle. Groups (n = 2–7) of heterogeneous beef cattle were recorded on pasture with nine solar trail cameras. Footage (~132 h) was curated in VideoLAN; annotated in Computer Vision Annotation Tool (CVAT) with bounding boxes and behavioral classes; and split 62/21/17% into training (24,508 frames), validation (8231 frames), and testing (6625 frames) sets. Four architectures were trained: Faster R-CNN (ResNet-50 FPN), Single Shot MultiBox Detector (SSD300, VGG-16), RetinaNet (ResNet-50 FPN with focal loss), and YOLOv8 nano (Ultralytics). With validation at 0.50 confidence and 0.50 IoU, Faster R-CNN achieved the highest overall F1 (0.79) and best per-class balance; RetinaNet was intermediate (peak F1 = 0.72); SSD300 saturated at F1 = 0.40; and YOLOv8 nano achieved some minority class recall at lower confidence. Each model detected the classes “grazing” and “hay feeding” accurately but confused cattle with the visually similar “normal” class. Datasets, checkpoints, and analysis scripts are provided to support further refinement of AI-enabled monitoring of extensive cattle systems. Full article

Photovoltaic (PV) power interval forecasting models are commonly trained on data dominated by non-ramp samples, which may weaken uncertainty characterization during rapid power changes. This study proposes a ramp-aware quantile regression Temporal Fusion Transformer (RQR-TFT) that jointly estimates PV power quantiles and the probability of a future ramp event. Ramp labels are constructed from the normalized power change between adjacent sampling instants. A shared Temporal Fusion Transformer (TFT) encoder extracts temporal representations from historical PV power and meteorological variables, and two output branches perform quantile forecasting and ramp-event identification. Ramp-sample-weighted quantile loss and positive-class-weighted classification loss are jointly optimized to increase the influence of minority ramp samples. The proposed method is evaluated for 4 h ahead forecasting using measurements collected from a 50 MW PV power station during 2019–2020. For the nominal 90% prediction interval, RQR-TFT achieves a ramp-sample prediction interval coverage probability (${\mathrm{PICP}}_R$) of 0.864, an overall prediction interval normalized average width (PINAW) of 0.209, and an overall normalized interval score (NIS) of 0.365. The area under the precision–recall curve for ramp-event identification is 0.906. The results demonstrate improved ramp-sample coverage and overall interval quality, although ramp-sample coverage remains below the nominal level. Full article

Background/Objectives: Coronary artery disease (CAD) is influenced by diet; however, the role of micronutrients remains underexplored. This study aimed to evaluate differences in dietary micronutrient intake between patients with and without CAD. Methods: A cross-sectional study was conducted with 107 participants (51 CAD, 56 non-CAD). Dietary intake was assessed using habitual 24-h dietary recalls and analyzed for micronutrient content. Biochemical, anthropometric, and clinical data were also evaluated. Results: CAD patients exhibited lower adequate intake of vitamins A, C, and K, as well as antioxidants such as lycopene and lutein, compared to the non-CAD group. While micronutrient inadequacy was prevalent in both groups, it was significantly more pronounced in subjects with CAD. Conclusions: Subjects with CAD had lower micronutrient intake than those without CAD, highlighting the potential importance of adequate micronutrient intake in the prevention and management of CAD. Full article

Background: Accurate identification and classification of white blood cell (WBC) subtypes in peripheral blood smears (PBS) is essential for the diagnosis and monitoring of hematological malignancies, including leukemia. Conventional manual microscopy, although clinically established, is labor-intensive and subject to inter- and intra-observer variability. Deep learning-based object detection offers a route to automation, yet most prior studies are limited by small datasets, restricted cell taxonomies, or single-microscope acquisition. This study evaluates a YOLOv11-large (YOLOv11L) detector for simultaneous localization and classification of 13 leukemia-relevant WBC subtypes plus an artifact class (14 classes total), trained on the large-scale, multi-domain, open-source LeukemiaAttri dataset. Methods: From the LeukemiaAttri dataset, 18,664 annotated images (67,347 objects) acquired at 40× and 100× magnification were partitioned by stratified sampling into training (70%), validation (15%), and test (15%) sets. The training set was expanded to 65,785 images through extensive geometric, photometric, and AugMix augmentation. A YOLOv11L model pretrained on MS COCO was fine-tuned for 250 epochs (640 × 640 input) on a single NVIDIA H200 SXM GPU, using an auto-selected optimizer (momentum 0.9; weight decay 5 × 10⁻⁴), automatic mixed precision (AMP), and mosaic augmentation for the first 240 epochs. Results: On an internal held-out test set, the model achieved an mAP50 of 93.9%, mAP50-95 of 77.9%, precision of 94.1%, recall of 88.8%, and an F1 score of 0.913, with similar performance in the validation and test sets. Class-wise average precision (AP) ranged from 89.3% (monocyte) to 98.2% (monoblast), confirming consistent detection across morphologically diverse subtypes. Conclusions: The YOLOv11L detector achieved high performance across all 14 categories on the internal test set, with metrics exceeding those previously reported for subset-specific baselines. These findings support further evaluation of the model as a decision-support tool for peripheral blood smear analysis. External validation is required to determine its clinical utility and generalizability. Full article

Background: The aim of this study was to examine the relationships between dietary intake and two key components of sarcopenia in older, sex- and age-matched men and women: muscle strength and physical performance. Methods: The study included 620 carefully selected community-dwelling adults aged 60 years and more (310 men and 310 women) matched by sex and age (±1 year; median age: 66 years). Body mass and body mass index were assessed. Dietary intake was evaluated using 24 h dietary recall questionnaire and analyzed with Dieta 5.0 software. Handgrip strength (HGS) was measured using a hydraulic hand dynamometer, while physical performance using the Timed Up and Go test (TUG). Results: In men, but not in women, higher intakes of multiple dietary components, especially proteins, were associated with greater HGS. TUG was inversely related to animal protein, some amino acids, folate and vitamin C in men, whereas in women to copper, magnesium, vitamins C and E, riboflavin, and PUFA intakes. After multivariable adjustment, no nutrient remained associated with HGS in both sexes. In women, vitamin C intake remained inversely associated with TUG. Age was the strongest predictor of HGS, while age, physical activity (PA) and chronic diseases were significant factors associated with TUG in both sexes. Conclusions: Our findings show multifactorial nature of muscle physiology, suggesting that in preventive strategies adequate dietary intake integrated with PA and effective management of chronic diseases is required, as nutrition alone appears insufficient to preserve muscle function. The sex-stratified associations suggest that personalized dietary recommendations may be beneficial. Full article

Entity-aware quality assessment may reduce false interpretations of electronic health record (EHR) data, but evidence from small, rule-aligned benchmarks cannot establish operational effectiveness. We revised H-StreamQ as a proof-of-concept framework and evaluated its laboratory component using the complete MIMIC-IV v3.1 labevents file (158,374,764 events; 313,442 patients). Ten thousand patients were sampled across laboratory-activity quintiles and split at patient level into training (6000), threshold-calibration (2000), and test (2000) groups. The independent test set contained 918,651 numeric laboratory events. Without excluding naturally alerted records, 54,788 mutually exclusive defects were introduced using subtle value shifts, unit/scale errors, mapping errors, delayed records, and patient-clustered correlated defects. Rules, a context-aware Isolation Forest, their union (Hybrid), a context-free Isolation Forest, Local Outlier Factor (LOF), and linear and radial-basis-function (RBF) One-Class support vector machines (OCSVMs) were compared at a threshold fixed by a 2.5% calibration alert budget. Patient-cluster bootstrap intervals and event-micro and patient-macro results were reported. Rules alone achieved the highest event-micro F1-score (0.637; 95% confidence interval [CI] 0.547–0.722), followed by Hybrid (0.576; 0.484–0.668) and RBF One-Class SVM (0.559; 0.433–0.670). Hybrid increased recall over rules by only 0.004 (95% CI 0.003–0.006) while reducing F1 by 0.061 and increasing the background-alert rate by 0.015. Context conditioning did not improve aggregate Isolation Forest performance. In six batch-level drift simulations, an exponentially weighted moving average (EWMA) and a fixed-window monitor detected 97–100% and 98–100% of changes, respectively, whereas a custom Hoeffding adaptive-window detector was more conservative and often missed smaller or recurrent changes. These results support H-StreamQ as an explainable research framework, not as a validated clinical or production system. Patient-macro F1, which weights every patient equally, was substantially lower than event-micro F1 for every method (rules 0.395 versus 0.637; Hybrid 0.320 versus 0.576), indicating that event-level performance is weighted towards high-activity patients. Precision and F1 are computed relative to injected synthetic labels and are not clinically adjudicated estimates. The entity-aware architecture spans patients, admissions, diagnoses, transfers, and dictionaries, but the quantitative detection benchmark evaluates the numeric laboratory component only; other entities are used for linkage and contextual attachment and are audited descriptively rather than evaluated against labels. Full article

Background: Benzodiazepines and non-benzodiazepine hypnotics (Z-drugs) act at the benzodiazepine binding site of GABA_A receptors but differ in their receptor subtype selectivity and behavioural profiles. Diazepam acts as a non-selective positive allosteric modulator of benzodiazepine-sensitive GABA_A receptors, whereas zolpidem and indiplon display preferential affinity for α1-containing receptor subtypes. Although these compounds have been investigated individually in experimental models of exploratory behaviour and cognition, comparative behavioural data obtained using the same experimental protocol remain limited. Methods: This study evaluated the behavioural effects of diazepam, zolpidem and indiplon in male Wistar rats using a two-trial Y-maze paradigm with a 24 h retention interval. Animals received one of three dose levels of each compound before the acquisition trial. Recall-phase behaviour was evaluated using conventional exploration measures together with time- and entry-based preference indices for the previously inaccessible arm. Primary analyses were performed separately for each compound using one-way ANOVA with appropriate post hoc comparisons. Complementary exploratory factorial analyses were subsequently conducted to examine overall behavioural patterns across drug and nominal dose categories. Results: Diazepam produced marked dose-dependent reductions in recall-phase exploration of the previously inaccessible arm and in both exploratory preference indices. In contrast, zolpidem and indiplon reduced exploratory activity during the acquisition phase without significantly affecting recall-phase exploratory behaviour. The complementary factorial analyses identified significant Drug × Nominal Dose Categories interactions across all recall-related endpoints, indicating that behavioural responses across the nominal dose categories differed among the three compound-specific experimental series. The strongest interaction effects were observed for Time Index % (F(6,72) = 10.522, p < 0.001, ηp² = 0.467) and Entries Index % (F(6,72) = 6.870, p < 0.001, ηp² = 0.364). Conclusions: Diazepam produced more pronounced dose-dependent alterations in recall-phase exploratory behaviour than zolpidem or indiplon in the experimental conditions employed. The complementary factorial analyses support the interpretation that the three compounds exhibited different behavioural patterns across the nominal dose categories evaluated, while these exploratory comparisons should be interpreted within the constraints of the study design. Overall, the findings contribute to the comparative behavioural characterization of benzodiazepine-site modulators in the two-trial Y-maze paradigm. Full article

Visible fracture detection in underground roadways is challenging because fracture targets are often elongated, weakly contrasted, irregularly distributed, and easily confused with complex rock-wall textures. In addition, uneven illumination, dust interference, and blurred boundaries further reduce the reliability of conventional crack detection and segmentation methods. To improve fracture instance segmentation under such conditions, this study proposes YOLOv13n-seg-crack, an improved lightweight instance segmentation model based on a self-constructed YOLOv13n-seg baseline. The proposed model introduces three main improvements. First, a C2f-CA module is embedded into the backbone to enhance spatial-position perception and directional feature representation for elongated fractures. Second, a shallow high-resolution branch and auxiliary feature paths, denoted as B2 + H2 + P2, are constructed to strengthen the transmission of fine edge and texture information for small and discontinuous fracture targets. Third, an Edge-aware SIoU (EA-SIoU) loss is designed by adding edge-consistency and aspect-ratio constraints, thereby improving bounding-box localization for narrow and irregular fracture regions. Experiments were conducted on the public Crack Segmentation Dataset and an expanded self-built underground roadway dataset collected at the Woniushan Experimental Base. On the public dataset, YOLOv13n-seg-crack achieved detection Precision, Recall, mAP50, and mAP50:95 of 84.56%, 65.49%, 71.51%, and 52.72%, respectively, and mask Precision, Recall, mAP50, and mAP50:95 of 74.94%, 60.38%, 59.52%, and 21.99%, respectively. Compared with YOLOv13n-seg, the detection mAP50 and mask mAP50 increased by 1.91 and 3.19 percentage points, respectively, while the model maintained an inference speed of 168.73 FPS. Repeated-seed experiments, ablation studies, and degraded-image tests further demonstrate the stability and robustness of the proposed improvements. On the self-built underground roadway dataset containing 100 images and 118 annotated fracture instances, YOLOv13n-seg-crack improved detection mAP50 from 68.72% to 73.36% and mask mAP50 from 30.76% to 33.74%. These results indicate that the proposed method provides an effective and lightweight solution for visible fracture detection and instance segmentation in complex underground roadway scenes. Full article

Carotenoids are bioactive compounds abundant in fruits and vegetables that may influence gut microbial composition through effects on the intestinal environment; however, evidence in emerging adults remains limited. This pilot study examined associations between carotenoid intake and gut microbial composition through a cross-sectional analysis of 50 Latino emerging adults (18–25 years). Dietary intake was assessed using three 24 h recalls, and carotenoid intake was modeled continuously and by tertiles. Microbial profiling was conducted using 16S rRNA gene sequencing (V3–V4 region) from stool samples. α-Diversity was analyzed using multivariable linear regression and β-diversity using PERMANOVA (9999 permutations), adjusting for relevant covariates. Participants were 20.3 ± 2.19 years old and 74.0% female. Continuous carotenoid intake was not associated with α- or β-diversity. A nominal association was observed between higher carotenoid intake tertiles and greater Chao1 richness (B = 7.06; 95% CI: 0.40, 13.7; p = 0.038); however, this association did not remain statistically significant after Benjamini–Hochberg false discovery rate correction. In sensitivity analyses additionally adjusted for dietary fiber and fruit and vegetable intake, carotenoid tertiles were associated with unweighted UniFrac (R² = 0.08; p = 0.009). Exploratory differential abundance analyses identified taxa that varied across intake groups. Although carotenoid intake was not associated with global microbial diversity, exploratory differences in richness and taxonomic composition were observed across intake levels, suggesting that carotenoids may be associated with specific features of gut microbial community structure in Latino emerging adults. These preliminary findings from this pilot study warrant confirmation in larger studies. Full article

On-device plant disease perception must reconcile two competing demands: enough spatial detail to localise diseased tissue within a field image and a memory and compute budget small enough for microcontroller-class hardware. Single-network solutions that jointly learn a pixel-wise mask and a fine-grained disease label tend to oversubscribe both the Flash and activation SRAM of such devices. This paper presents Chroma-Sense 2.0, a two-stage lightweight pipeline that decouples the two subproblems and sizes each stage for its own budget. The two stages run sequentially on the same frame: Stage 1 is a per-channel convolutional classifier, derived from Chroma-Sense, that names the disease, and Stage 2 is a compact ESPNet segmenter that produces a binary diseased-versus-healthy mask localising it. Because the stages run one after the other rather than concurrently, the peak working memory of the pipeline is the maximum of the two stages rather than their sum. We evaluate the pipeline on the in-the-wild PlantSeg dataset using a curated 10-species, 34-class subset and a leakage-controlled protocol in which all training crops are derived from PlantSeg’s official training images and all reported metrics are measured on a held-out test set of 5002 crops built from the official test images. The segmentation stage attains a mean foreground recall of 0.97 (mean foreground IoU of 0.49; 0.53 pooled over pixels), a deliberately recall-oriented operating point. Against Fast-SCNN, a small U-Net, LR-ASPP, and DeepLabV3+, ESPNet is the smallest-footprint model (140k parameters, 193 KB Int8 Flash) while retaining the highest foreground recall; the per-channel classifier reaches accuracies comparable to much larger ImageNet-pretrained backbones (MobileNetV3 and EfficientNet) using 10–13× fewer parameters. End to end, the coupled pipeline classifies the disease correctly on 87.8% of the test crops. An on-device profile on the OpenMV H7 and H7 Plus shows that the binding constraint at 256 × 256 is the segmenter’s ≈4 MB contiguous activation arena, rather than parameter Flash: Even on the 32 MB-SDRAM H7 Plus, the usable interpreter heap is only about 4 MB, and the arena cannot be allocated as a single contiguous block from it, so on the tested firmware, the classifier runs on microcontrollers while the segmenter does not; the full pipeline instead fits the gigabyte-scale single-board-computer tier (for example, Raspberry Pi or NVIDIA Jetson Nano), and enabling the segmenter to run on microcontrollers remains the open gap. Full article

In critical infrastructure environments, cybersecurity situation-awareness platforms generate large volumes of alerts, including substantial numbers of false positives, placing a considerable burden on security analysts. At present, alert correlation methods mainly rely on rule-based matching or statistical clustering, and large language models often lack the domain-specific threat intelligence required for reliable security analysis. This paper proposes GRAL, which is an AI-driven framework that combines graph neural networks (GNN) for cross-asset temporal alert correlation, retrieval-augmented generation (RAG) for dynamic threat intelligence enrichment, and large language models (LLM) for semantic reasoning and verdict generation. A temporal heterogeneous graph attention network constructs alert-relation graphs within a 72 h sliding window, and temporal decay and multi-relational dependencies are captured. Powered by bge-m3 embeddings and a dense vector index, the RAG module retrieves the most relevant threat intelligence entries above a cosine similarity threshold of 0.75. A domain-specific dataset of 1000 annotated security alerts from a nuclear power operational environment was built, and Cohen’s Kappa reached 0.87. The experiments show that GRAL achieves a macro-averaged precision of 87.0%, a macro-averaged recall of 97.0%, and a binary false-positive rate of 9.1%, together with 92.5% alert compression. Generalisation capability is confirmed by cross-dataset evaluation on CICIDS2017 (93.0% accuracy and 92.5% F1-score) and UNSW-NB15 (89.4% accuracy and 89.8% F1-score). Full article

Background/Objectives: There is continuing debate regarding the association between added sugars (AS) consumption and overall diet quality, which remains a research gap within the Canadian population. This study aimed to assess the association between intakes of added sugars and the Healthy Eating Food Index (HEFI) 2019 that measures adherence to Canada’s Food Guide as an indicator of diet quality. Methods: The first 24 h dietary recalls of Canadians 2 years and older (n = 19,532) from the 2015 Canadian Community Health Survey—Nutrition were used. Added sugars intakes were calculated based on a systematic 10-step algorithm. Descriptive statistics were obtained for total HEFI and component scores across different AS intake categories. The association between AS categories and HEFI scores was tested using a generalized linear model adjusted for age, sex, education level, type of smoker, and misreporting status. Results: The association between continuous AS intake and HEFI scores was non-linear. In the categorical model, AS intake categories were inversely associated with HEFI scores among Canadians (R² = 0.32, p < 0.001). Large overlaps in HEFI scores were observed among individuals with similar AS intakes for the whole population and each age group. Among high AS consumers (>15% EI), lower HEFI scores were largely due to higher proportions of sweetened beverages (i.e., low beverage scores) and lower intakes of vegetables and fruit and plant-based proteins. Scores for protein food and whole grains were low across all AS intake categories. Conclusions: Large variations in overall HEFI scores were observed amongst those with similar AS intakes, suggesting that added sugars alone may not fully capture overall diet quality or adherence to Canada’s Food Guide. Full article

Background/Objectives: Romania faces a shortage of family physicians, particularly in rural areas. Undergraduate Family Medicine rotations may influence students’ perceptions before residency selection, but the contribution of rural versus urban training settings remains insufficiently documented. This study examined differences between post-rotation perceptions and students’ recalled pre-rotation perceptions, and explored whether these differences varied according to placement setting. Methods: This single-centre study used a single-administration retrospective pretest–posttest observational design. After completing the mandatory one-month Family Medicine rotation, 141 sixth-year medical students completed a questionnaire containing recalled pre-rotation and post-rotation ratings on 12 five-point Likert items. Paired comparisons used Wilcoxon signed-rank tests; multiple-testing corrections were applied within the defined item families. A baseline-adjusted model with robust standard errors examined rural placement. Results: The composite score increased from 3.62 ± 0.84 to 4.10 ± 0.74 (mean change 0.483; W = 602.5; Z = −7.990; p < 0.001; r = 0.673). All 12 paired item comparisons remained significant after Bonferroni correction. No rural–urban item comparison remained significant after correction. In the adjusted model, rural placement was not associated with a statistically significant difference in the post-rotation score (β = 0.197, 95% CI −0.002 to 0.397; p = 0.053). Among the 86 rurally exposed students, the mean willingness to practice rurally was 2.87/5. The apparent association with career intention was not retained when excluding the overlapping career-likelihood item (11-item composite: H = 6.695, p = 0.153). Conclusions: Students reported more favourable perceptions after the rotation than they recalled having before it, but their willingness to practice rurally remained limited. The uncontrolled, retrospective design does not permit causal attribution, and the data do not establish a rural-placement advantage or subsequent specialty choice. Full article