Citations

Background: Incidental thyroid nodules (ITNs) detected on nonthyroid imaging contribute to thyroid cancer overdiagnosis, largely through downstream evaluation initiated by thyroid ultrasound (TUS). However, TUS follow-up rates for ITNs and their drivers in routine practice remain unclear. We quantified TUS utilization after ITN detection and identified patient, imaging, nodule, and reporting factors associated with this first step in the diagnostic cascade. Methods: We conducted a retrospective cohort study of adults without prior thyroid disease who underwent imaging that incidentally captured the thyroid at Mayo Clinic sites (2017–2023). A validated natural language processing pipeline identified ITNs and extracted nodule characteristics from radiology reports across multiple imaging modalities. The main outcome was the rate of TUS within 180 days after ITN detection. Clinical outcomes, including thyroid biopsy, thyroidectomy, and thyroid cancer diagnosis, were compared between patients who underwent TUS and those who did not. Factors associated with TUS utilization were assessed using multivariable logistic regression with penalized variable selection. Results: Among 115,683 patients who underwent thyroid-capturing imaging, 8426 (7.3%) had an ITN. These patients were predominantly female (64.6%) and White (89.7%), with a mean age of 61.7 years (standard deviation 15.2). Most nodules were detected on computed tomography (80.8%), most commonly from chest imaging (52.3%). Overall, 1677 patients with ITNs (19.9%) underwent TUS. In multivariable analysis, TUS use was independently associated with higher body mass index (odds ratio [OR], 1.20 per 10-unit increase; confidence interval [CI] [1.08–1.33]), nodule size ≥1.5 cm (OR, 1.79; CI [1.52–2.10]), and a radiologist recommendation (OR, 2.51; CI [2.12–2.98]), whereas older age was associated with lower odds of TUS (OR, 0.89 per 10-year increase; CI [0.84–0.93]). Patients undergoing TUS had higher rates of thyroid biopsy, partial thyroidectomy, and total thyroidectomy compared with those without it. All detected thyroid cancers (103 cases) occurred in the TUS group, representing 6.1% of that cohort. Conclusions: One in five patients with ITNs underwent TUS, representing a critical gateway to downstream diagnostic procedures and thyroid cancer detection. Factors associated with TUS utilization highlight opportunities to better define appropriateness of ultrasound after incidental nodule detection and to reduce unnecessary testing and treatment.

Background/Objectives: To develop and evaluate a deep learning-based quality control system for Lumbar Spinal Digital Radiographs (LSDR), designed to automate and improve their evaluation and reduce reliance on manual reviews. Methods: This retrospective study utilized a deep learning workflow comprising image segmentation, feature extraction, and a classification model. The dataset, including anteroposterior (AP) and lateral (LAT) X-ray images, was expanded through data augmentation techniques. Four U-Net-based models were assessed: standard U-Net, Swin-UNet, Attention U-Net, and Attention U-Net with the weight map, with the latter selected for its superior performance. Extracted features, such as brightness, contrast, and anatomical positioning, were used in an XGBoost classifier, which was evaluated using mean intersection over union (mIoU), accuracy, sensitivity, specificity, and AUC. Results: The Attention U-Net with weighted attention outperformed the other models, achieving high mIoU scores in both AP and LAT views. The XGBoost classifier achieved the best performance in classifying images as “qualified” or “unqualified,” with an AUC of approximately 0.9, high accuracy, and balanced sensitivity and specificity. This approach effectively addressed class imbalances and improved model accuracy compared to traditional machine learning models such as MLP and SVM. Conclusions: The developed automated quality control system demonstrated potential for enhancing image quality, enhancing diagnostic reliability, and optimizing clinical workflow efficiency.

Abstract Context Incidental thyroid findings (ITFs) are increasingly detected on imaging performed for non-thyroid indications. Their prevalence, features, and consequences remain undefined. Objective To develop, validate, and deploy a natural language processing (NLP) pipeline to identify ITFs in radiology reports and assess their prevalence, features, and clinical outcomes. Design Retrospective cohort study. Setting Mayo Clinic sites (Rochester, Arizona, Florida, Mayo Clinic Health System). Participants Adults without prior thyroid disease undergoing thyroid-capturing imaging from July 1, 2017, to September 30, 2023. A transformer-based NLP pipeline identified ITFs and extracted nodule characteristics from image reports from multiple modalities and body regions. Outcomes ITF prevalence, downstream thyroid ultrasound, biopsy, thyroidectomy, and cancer diagnosis. Logistic regression identified demographic and imaging-related factors. Results Among 115 683 patients (mean age, 56.8 [SD 17.2]; 52.9% women), 9077 (7.8%) had an ITF (92.9% nodular). ITFs were more likely in women, older adults, higher BMI, and in imaging ordered by specialties different from Emergency Medicine. Compared with chest CT, ITFs were more likely via neck CT, PET, and nuclear medicine scans. Nodule characteristics were poorly documented, with size reported in 44% and other features in fewer than 15%. Compared with patients without ITFs, those with ITFs had higher odds of thyroid nodule diagnosis (OR 45, 95%CI 41.1-49.3), biopsy (OR 46.8, 95%CI 39.0-56.2) thyroidectomy (OR 55.8, 95%CI 31.3-99.3), and thyroid cancer diagnosis (OR 61.7, 95%CI 38.6-98.5). Most cancers were papillary (88.5%) and larger when detected after ITFs (2 cm SD 1.4) vs no ITF (1.3 cm SD 0.8). Conclusion ITFs were common and strongly associated with cascades leading to the detection of small, low-risk cancers, highlighting their role in overdiagnosis and the need for standardized reporting and more selective follow-up.

Abstract Background Integrating advanced artificial intelligence (AI) into clinical decision-support often requires the sharing of sensitive patient data with external services, raising privacy concerns. Homomorphic encryption (HE) allows computing directly on encrypted data, without revealing the underlying patient information. Objectives To develop a large language model (LLM)-assisted diagnosis framework while preserving patient privacy in the clinical text analysis, by leveraging HE and using rare disease (RD) diagnosis as a representative application. To demonstrate HE does not hinder the system performance. Materials and Methods Texts from patient histories and a RD knowledge base were embedded by LLMs into vectors, then encrypted using HE to obscure private information while retaining the semantic nuances. Diagnostic recommendations were generated by computing and ranking the similarities between the patient history and RD vectors in the encrypted space. The system was evaluated using 50 synthetic case reports (5 RDs, each with 10 reports). Results Applying HE did protect private information from reverse-embedding attacks. HE imposed little disruption to the diagnostic accuracy, with normalized discounted cumulative gains (nDCG) of 0.6108 ± 0.3412 (encrypted) versus 0.6083 ± 0.3415 (unencrypted). The accuracy and computational performance were tunable and consistent, as demonstrated across five different LLMs. Discussion Our privacy-preserving framework opens tremendous opportunities toward hosting and serving powerful AI solutions across institution boundaries, which would remove the need for local deidentification and incentivize users to access secure external decision-support services. Conclusions Integrating HE with LLM retrieval can promote the dissemination of nonredundant, high-capacity AI services by preserving both privacy and accuracy.

Abstract Pain medicine triage plays a crucial role in ensuring patients receive timely and appropriate care by scheduling them to the most suitable treatment path. However, the absence of standardized triage protocols in pain medicine often leads to inefficiencies, including delay of care and wastage of healthcare resources. This study aims to develop a rule-based automated referral triage system leveraging information from patients' medical notes for scheduling patients to specific procedures in the pain medicine department. The proposed triage system, grounded in the knowledge and expertise of clinical providers, processed referral order comments and referring provider notes by iteratively refining the Natural Language Processing (NLP) rules and post-processing rules through intensively reviewing 76 patients. A post-processing regression model was incorporated to further enhance the accuracy. To ensure alignment with real-world practices, the system was integrated into an electronic health record (EHR) platform for real-time application, streamlining scheduling workflows and enhancing usability in daily clinical settings. After three iterations, the proposed NLP and post-processing rules improved accuracy from 76.3 to 80.3% compared to machine learning (ML) approaches in the preliminary study. The post-processing model further increased accuracy to 84.2%. The implementation accuracy of 200 cases for the first 3 months was consistent with our prediction at 83.5%, which concluded that the improvement over ML models (p-value = 0.018) was statistically significant at 95% significance level. This study demonstrates the feasibility and benefits of a knowledge-driven approach to referral triage in specialized medical fields. It lays a foundation for others in building similar triaging solutions to other specialties.

Abstract Purpose Overuse of thyroid ultrasound (TUS) has contributed to rising thyroid cancer diagnoses and is projected to increase US healthcare costs from $1.5 billion to $3.5 billion by 2030. This study evaluated the healthcare cost of inappropriately ordered TUS in a national multicenter academic system. Methods This is a secondary cost analysis of a retrospective cohort study across 4 Mayo Clinic sites (Rochester, MN; Jacksonville, FL; Scottsdale, AZ; and the Midwest Mayo Clinic Health System). Adult patients (≥18 years) undergoing their first TUS between January 1, 2017, and December 31, 2021, with at least 1 year of follow-up were included. TUS indications were classified as appropriate or inappropriate using a guideline-based natural language processing algorithm. The primary outcome was a comparison of adjusted 1-year all-cause healthcare costs. A secondary analysis calculated the direct procedural costs of the inappropriate TUS cascade. Results Among 6984 patients (mean age 56 [SD 16.4]; 76.2% female; 90.9% White), 546 (7.8%) underwent TUS for inappropriate indications. These patients were younger (mean age 53.0 vs 56.3 years, P < .0001) but otherwise demographically similar. Adjusted total healthcare costs over 90 days and 1 year were comparable: $4842 vs $5794 and $13 748 vs $14 257 for inappropriate vs appropriate TUS, respectively. The inappropriate TUS cascade, including an estimated 56 subsequent biopsies and 22 thyroidectomies, resulted in a minimum of $576 134 in direct procedural costs. Conclusion While adjusted total costs were similar, inappropriate TUS represents potentially avoidable spending and remains a viable target for cost-reduction strategies. Reducing low-value imaging remains a critical target for cost-saving interventions.

Abstract Objectives To enable users with modest technical background to perform biomedical natural language processing (NLP). Materials and Methods We developed medspacyV using the Python graphical programming tkinter library, following the model-view-controller (MVC) design pattern. The interface wraps around a rule-based pipeline for sentence splitting, section segmentation, concept identification, and negation detection. Results The primary window allows the user to configure the project and NLP rules, execute the pipeline, and save the outputs into a table. A separate annotation viewer window can be launched to inspect the immediate or previous NLP outputs. Discussion We developed medspacyV with three rationales: controllability, explainability, and economy. The rule-based approach is sufficient for many NLP use cases. Conclusion The medspacyV program is publicly available at https://github.com/medspacy/medspacyV, targeting use by healthcare professionals and researchers in their NLP projects.

Abstract Objective: Electronic Health Record (EHR) data are critical for advancing translational research and AI technologies. The ENACT network offers access to structured EHR data across 57 CTSA hubs. However, substantial information is contained in clinical narratives, requiring natural language processing (NLP) for research. The ENACT NLP Working Group was formed to make NLP-derived clinical information accessible and queryable across the network. Methods: We established the ENACT NLP Working Group with 13 sites selected based on criteria including clinical notes access, IT infrastructure, NLP expertise, and institutional support. We divided sites into five focus groups targeting clinical tasks within disease contexts. Each focus group consisted of two development sites and two validation sites. We extended the ENACT ontology to standardize NLP-derived data and conducted multisite evaluations using the Open Health Natural Language Processing (OHNLP) Toolkit. Results: The working group achieved 100% site retention and deployed NLP infrastructure across all sites. We developed and validated NLP algorithms for rare disease phenotyping, social determinants of health, opioid use disorder, sleep phenotyping, and delirium phenotyping. Performance varied across sites (F1 scores 0.53–0.96), highlighting data heterogeneity impacts. We extended the ENACT common data model and ontology to incorporate NLP-derived data while maintaining Shared Health Research Informatics NEtwork (SHRINE) compatibility. Conclusion: This demonstrates feasibility of deploying NLP infrastructure across large, federated networks. The focus group approach proved more practical than general-purpose approaches. Key lessons include the challenge of data heterogeneity and importance of collaborative governance. This work also provides a foundation that other networks can build on to implement NLP capabilities for translational research.

ImportanceThyroid cancer incidence has tripled in the past 3 decades, largely driven by increased detection of small, indolent papillary thyroid cancers. Overuse of thyroid ultrasonography (TUS) contributes to overdiagnosis, leading to unnecessary biopsies, procedures, and potential patient harm.ObjectiveTo evaluate the frequency of and factors associated with inappropriate TUS (iTUS) orders and assess related clinical outcomes.Design, Setting, and ParticipantsThis retrospective cohort study analyzed adult patients 18 years and older who underwent their first TUS from January 1, 2017, to December 30, 2021, at 4 Mayo Clinic sites. Patients with previous thyroid nodule, thyroid cancer, thyroid surgery, or TUS were excluded. Based on documented clinical indications, a validated natural language processing model classified TUS orders as appropriate or inappropriate. Data were analyzed from April 2024 to May 2025.ExposuresBaseline characteristics of patients, clinicians, and clinical encounters investigated for their association with an iTUS order.Main Outcomes and MeasuresThe primary outcome was the proportion of iTUS orders. Secondary outcomes included factors associated with iTUS (sociodemographic, clinical, order-specific, and clinician-related factors) and clinical outcomes following iTUS, including the detection of thyroid nodules, thyroid procedures, and thyroid cancer diagnoses.ResultsOf 11 442 included patients, 8422 (73.6%) were female, and the mean (SD) age was 57.3 (15.8) years with a mean (SD) body mass index (calculated as weight in kilograms divided by height in meters squared) of 29.7 (7.0). Among 11 442 TUS orders, 866 (7.6%) were classified as inappropriate. In multivariable analysis, younger age (age of 18 to 54 years vs older than 65 years; odds ratio [OR], 1.86; 95% CI, 1.56-2.22), hyperthyroidism (OR, 9.04; 95% CI, 6.75-12.11), ordering by nonendocrinology specialties (eg, oncology/hematology; OR, 3.43; 95% CI, 2.47-4.76), and orders not linked to an in-person appointment (eg, portal messages; OR, 2.42; 95% CI, 2.08-2.82) were strongly associated with increased odds of iTUS. Compared with appropriate TUS, iTUS led to lower rates of thyroid nodule detection (202 of 866 [23.3%] vs 6885 of 10 576 [65.1%]; OR, 0.16; 95% CI, 0.14-0.19), biopsies (89 [10.3%] vs 2647 [25.0%]; OR, 0.34; 95% CI, 0.27-0.43), partial thyroidectomies (14 [1.6%] vs 424 [4.0%]; OR, 0.39; 95% CI, 0.23-0.67), and confirmed thyroid cancer cases (16 [1.8%] vs 425 [4.0%]; OR, 0.45; 95% CI, 0.27-0.74).Conclusions and RelevanceIn this cohort study, approximately 1 in 13 TUS orders were inappropriate. iTUS was more common in younger patients, those with thyroid dysfunction, those not seen in person, and in those referred by nonendocrinology specialties. These findings highlight the need for targeted strategies to optimize TUS use and mitigate overdiagnosis and overtreatment.

Background Electronic health records (EHRs) serve as a critical tool for clinical documentation, yet their complexity has raised concerns about readability and usability for healthcare providers and patients. Readability scores, such as the Flesch-Kincaid (FK) Grade Level, offer a quantitative method to assess documentation clarity. In this study, the authors evaluated the impact of an EHR transition on the readability of clinical notes across multiple medical specialties. Methods All clinical notes recorded between 2016 and 2019 at a large healthcare organization were evaluated across specialties. The organization transitioned EHR systems during the study period, in May 2018. FK Grade Level scores were calculated for each clinical consultation note and stratified by specialty and EHR. Results Across 10 medical specialties, 630,246 clinical notes were evaluated. The median FK grade levels from notes composed in the new EHR were similar to those composed in the previous EHR (8.228 vs 8.240; p=0.58). The variation in FK grade level as measured by standard deviation was lower in the new EHR compared with the old EHR (1.405 vs 1.764; p <0.01). Conclusions Calculation of FK grade level of clinical notes is a feasible method to assess clinical note readability. Study results showed that clinical documentation readability may not be strongly associated with underlying EHR. Migrating to a new EHR alone should not be expected to improve readability of clinical notes.