JMIR Medical Informatics

Balancing Privacy Protection and Reproducible Entity Extraction in Real-World Data in the Development of a Framework for Deidentified Japanese Electronic Health Record Narratives Using BERT: Retrospective Observational Study

2026-11-06T15:30:22-05:00

Background: Pharmacoepidemiologic studies using real-world data often lack clinical context embedded in free-text electronic health records (EHRs). However, sharing EHR narratives is restricted by privacy requirements, and conventional deidentification approaches may reduce analytic utility and limit auditability. Objective: This study aims to develop and evaluate a framework that (1) extracts structured cancer-related entities from Japanese EHR free text using BERT (Bidirectional Encoder Representations from Transformers)–based natural language processing and (2) generates deidentified analytic text that preserves contextual information for post hoc auditing under privacy constraints. Methods: This retrospective observational study used the DATuM IDEA database, which integrates EHR narratives with structured records (claims, prescriptions, and procedure/surgery records) from 2 hospitals in Japan from January 1, 2019, through June 30, 2025. Deterministic linkage was performed within ICI (Integrated Clinical Care Informatics Inc) prior to deidentification. Downstream recoverability analyses were conducted within a governance-constrained analyzable text cohort derived after deidentification and preprocessing. We quantified deidentification using masking rates and evaluated downstream recoverability of TNM (tumor-node-metastasis) or stage-related mentions, internal logical consistency (M1 vs Stage IV), and proxy-based concordance using structured data treatment proxies (Anatomical Therapeutic Chemical [ATC] code L for systemic anticancer therapy and procedure-name keywords for cancer-directed interventions). A stratified manual audit of 400 deidentified narratives primarily inspected for apparent residual direct identifiers and secondarily for retention of TNM and stage outputs. Results: Among 10,214 patients with neoplasms ( [] C00-D48), linked progress notes were available for 4383 patients (82,863 records). The analyzable text cohort (Layer C) comprised 3689 patients and 38,841 documents, of which 1339 patients (1339/3689, 36.3%) and 11,914 documents (11,914/38,841, 30.7%) had at least one recoverable TNM or stage mention (Layer D). Marginal document-level recoverability in Layer C was 24.2% for T, N, and M elements and 23.6% for stage. Conditional on Layer D, it was 78.8% for T, 78.7% for N, 78.8% for M, and 76.8% for stage. Median masking rates ranged from 9.9% to 15% across major cancer categories. M1 and Stage IV mention indicators showed an overall agreement of 94.7%, positive percent agreement of 54.8%, and negative percent agreement of approximately 99%. Proxy-based concordance analysis showed positive predictive values of 0.700 for Stage IV using the systemic therapy proxy and 0.031 for early-stage classification using the procedure proxy. Manual audit demonstrated retention of intended TNM/stage-related outputs after normalization and identified no apparent residual direct identifiers in the audited deidentified narratives. Conclusions: This framework generates deidentified Japanese EHR narratives that preserve contextual structure for auditing while supporting structured entity extraction, addressing the trade-off between privacy protection and reproducibility in real-world data research.

Correction: A Machine Learning–Based Model to Predict Overactive Bladder Risk Among US Women: Evidence From the National Health and Nutrition Examination Survey 2011-2018

2026-11-06T13:00:27-05:00

In "A Machine Learning-Based Model to Predict Overactive Bladder Risk Among US Women: Evidence From the National Health and Nutrition Examination Survey 2011-2018" [1], the authors noted the following corrections. First, the Figure 2B legend has been corrected. Original: “Seven predictor variables corresponding to the log(λ) value with the minimum MSE were retained.” Revised: “Fourteen predictor variables corresponding to the log(λ) value with the minimum MSE (λ = 0.00550) were retained.” At the penalty selected under the minimum criterion (λ = 0.00550), the least absolute shrinkage and selection operator (LASSO) model retained 14 non-zero coefficients, and all 14 variables were entered into model construction. The error was confined to the figure legend text, and no numerical result is affected. Second, the Figure 4A legend has been clarified. The legend now states that Figure 4A displays the 7 continuous predictors among the 14 variables retained by LASSO regression. SHAP beeswarm and dependence plots require a continuous colour gradient and therefore cannot display categorical predictors, which are presented separately; no figure content is affected by this clarification. Third, the sentence reporting the significant predictors in the Results section of the Abstract has been corrected. Original: “Five variables were identified as significant predictors.” Revised: “Five of these 14 variables were identified as the most significant clinical predictors.” Fourth, the sentence characterising the predictive accuracy of the model in the Conclusions has been corrected. Original: “The RF model demonstrated good predictive accuracy and practical clinical applicability, serving as a noninvasive, cost-effective tool for early screening and risk stratification in community health and primary outpatient settings.” Revised: “The RF model demonstrated moderate predictive accuracy and practical clinical applicability, serving as a noninvasive, cost-effective tool for early screening and risk stratification in community health and primary outpatient settings.” These corrections affect the wording of the figure legends, the Abstract, and the Conclusions only. All analyses were performed on the 14 predictors retained by LASSO regression, and every reported metric derives from them; no analytical result is affected. The correction will appear in the online version of the paper on the JMIR Publications website together with the publication of this correction notice. Because this was made after submission to PubMed, PubMed Central, and other full-text repositories, the corrected article has also been resubmitted to those repositories.

Expert Perspectives on Transparency and Trust in AI-Based Clinical Decision Support Systems for Chest X-Ray Diagnosis: Qualitative Study

2026-11-05T16:00:30-05:00

Background: The integration of AI-based Clinical Decision Support Systems (AI-CDSS) into clinical practice has expanded rapidly, particularly in medical imaging. Beyond algorithmic performance, growing evidence suggests that clinicians’ engagement with AI-CDSS is shaped by how transparently systems communicate their outputs, and how trust is formed through human-AI interaction. However, a limited understanding of how these concepts are perceived by end users and developers continues to hinder real-world implementation. Objective: This study aims to explore expert perspectives on transparency and trust in AI-CDSS for chest X-ray diagnosis, with a focus on the two-dimensional lens: user interface and users’ mental models. Methods: A qualitative study was conducted using semistructured interviews with a total of 21 experts, including 11 radiologists and 10 AI experts experienced in AI-CDSS. Data were analyzed thematically to explore factors perceived to influence transparent communication and trust. Results: Three themes emerged: (1) perspective of transparency and trust, (2) user interface, and (3) users’ mental models. Findings reveal that trust is a dynamic calibration process developed through repeated use, feedback loops, and alignment with clinicians’ mental models, rather than algorithmic accuracy alone. Significant discrepancies were identified between how developers design system transparency and how clinicians interpret AI outputs. Conclusions: This study demonstrates that transparency and trust are achieved through interpretable outputs, clear communication of system limitations, and alignment with clinical workflows. Designing trustworthy AI-CDSS requires integrating transparent interface design with consideration of users’ knowledge, experience, and attitudes to support effective human-AI interaction.

Development and Content Validation of a Self-Assessment Checklist for Electronic Medical Record Use Certification in Korea: Convergent Mixed Methods Study

2026-11-04T17:45:20-05:00

Background: Electronic medical record (EMR) certification programs have traditionally focused on product-level requirements, such as interoperability, security, and core functionality. However, growing evidence suggests that many patient safety risks associated with EMRs arise after implementation and are closely related to system configuration, workflow integration, and real-world use. Despite the presence of a national EMR product certification system in Korea, a systematic evaluation of how certified systems are implemented and used in clinical practice is still lacking. Objective: This study aimed to develop and validate a self-assessment checklist for EMR use certification, focused on patient safety-critical functions, and to evaluate its content validity and expert-assessed practical applicability across health care institutions of different sizes. Methods: A convergent mixed methods design was used. A self-assessment checklist comprising 10 patient safety–related EMR functions was developed by adapting the US Office of the National Coordinator for Health Information Technology self-assessment templates to the Korean EMR certification context. Content validity was quantitatively assessed using the Content Validity Index across 4 dimensions (relevance, usefulness, clarity, and importance) by a panel of 24 experts with experience in EMR implementation and certification. Qualitative feedback was collected through open-ended survey questions to examine feasibility, anticipated challenges, and differentiation between product and use certification across tertiary referral hospitals, general hospitals, and clinics. Results: The overall Scale-Level Content Validity Index was 0.921, indicating strong content validity for the proposed checklist. Although most items demonstrated high agreement, lower ratings were observed for certain functions, particularly in clinic-level settings, reflecting structural and resource-related constraints rather than limited recognition of patient safety priorities. Qualitative findings highlighted challenges related to staffing, system governance, and the feasibility of assembling multidisciplinary assessment teams in smaller institutions. Conclusions: The self-assessment checklist demonstrated high content validity and was perceived as relevant for evaluating patient safety–critical aspects of EMR use. However, expert-assessed feasibility varied across health care institution types, particularly in clinic-level settings. Further empirical testing is needed to assess the checklist’s reliability, usability, and applicability in real-world certification processes.

Impact of Ambient Scribes on Primary Care Communication Pathways: Qualitative Observation and Interview Study

2026-11-04T10:00:18-05:00

Background: The administrative burden of manual documentation has been a primary driver of physician burnout and a barrier to patient-centered care. Ambient scribes have emerged as a promising solution that uses passive audio capture and natural language processing to automate clinical note-taking. While their impact on clinical efficiency is frequently studied, there is a lack of research on how these tools have impacted interpersonal communication during clinical encounters. Objective: This study investigates the communication pathways and interactional shifts that occur when ambient scribes are integrated into primary care, with a focus on how physicians adapt their behavior to accommodate an ambient digital listener. Methods: We conducted a qualitative study with 17 licensed family practice physicians in the Greater Toronto Area who use ambient scribes in their daily practice. Data collection included observations from in-person clinical shadowing (n=5) and think-aloud walk-throughs (n=12), each followed by semistructured interviews; reported findings are predominantly drawn from interview data. We analyzed the combined dataset using iterative thematic analysis and open coding to identify affordances and constraints along 3 communication pathways: physician-to-patient, physician-to-patient-and-scribe, and physician-to-note. Results: Along the physician-to-patient pathway, participants reported that ambient scribes increased their capacity to demonstrate active listening, with some reporting greater attentiveness to nonverbal cues, yet introducing the technology to patients to obtain consent consumed limited consultation time, particularly when addressing sensitive health concerns. Along the physician-to-patient-and-scribe pathway, participants appreciated how ambient scribes acted as a forcing function for physicians to verbalize their clinical reasoning to enhance patient comprehension, as well as their scribes’ ability to follow and organize nonlinear conversations. However, they reported that scribes frequently omitted implicit information (eg, deictic gestures) or created unintended documentation artifacts (eg, casual remarks about family members) that required post hoc corrections. Along the physician-to-note pathway, participants reported that scribes optimized their note-taking workflows and enabled note manipulation via spoken words for some physicians but recalling patient information to fix transcription errors and support patient follow-ups proved challenging. Conclusions: Integrating ambient scribes introduces a triadic interactional dynamic into traditionally dyadic patient consultations. Clinicians must continuously navigate the boundaries between patient-facing empathy and scribe-facing precision, deciding when to articulate clinical reasoning aloud and when to exercise verbal discretion regarding sensitive disclosures. While ambient scribes effectively reduce in-encounter computer screen time, they do not eliminate documentation overhead—they redistribute it from real-time typing to postencounter cognitive recall and verbal note curation. To prevent delayed cognitive fatigue from postencounter error verification, we suggest that future ambient scribes should move beyond passive transcription to incorporate interactive recall scaffolding and context-aware error prevention.

Informatics for the Planet: Integrating Species, Ecosystems, and Geophysical Data to Manage the Health of Earth Systems

2026-11-03T16:00:31-05:00

Biomedical and health informatics (BMHI) has traditionally focused on human-centric clinical data to support individual patient care and health systems. However, contemporary health challenges driven by climate change, biodiversity loss, and ecological degradation demand an expanded paradigm that addresses interdependencies among human, animal, and environmental health. This editorial introduces a new JMIR Medical Informatics section on “Planetary Health Informatics” (PHI), an essential evolution extending informatics praxis beyond human healthcare toward Earth’s interconnected systems. JMIR Medical Informatics is launching a call for papers for a new section dedicated to advancing PHI theory, methods, and applications for a sustainable future shaped by human impact on Earth. We invite authors to submit work that explores how PHI integrates species, ecosystem, and geophysical data to address critical global challenges, including metagenomic disease surveillance, climate-related hazard forecasting, environmental pollution monitoring, and health infrastructure decarbonization. By synthesizing broader ecological frameworks, we present a roadmap for transforming fragmented data streams into planetary-scale intelligence.

Implementation Evaluation of an AI Scribing Tool in the Ambulatory Setting: Single-Site Qualitative Implementation Evaluation Study

2026-11-03T15:15:19-05:00

Background: AI scribing tools to assist health care providers by generating clinical notes are rapidly spreading across health care settings. Although these tools may reduce documentation burden, successful implementation depends on the contextual factors influencing their integration into clinical practice. Implementation evaluations are needed to inform practice integration, guide future adoption, and enhance effective use of this novel technology. Objective: This study aims to explore contextual factors influencing the implementation of an AI scribe for physicians in the ambulatory setting. Methods: This study was a single-site qualitative implementation evaluation at a metropolitan university health system where the AI scribe was available across primary and specialty care clinics. The Practical, Robust Implementation and Sustainability Model (PRISM) and the Sociotechnical Systems framework guided the evaluation. Using maximum variation sampling, we recruited attending physicians with varied levels of AI scribe use. Semistructured interviews were conducted with attending physicians working in the ambulatory setting. Data collection and analysis occurred iteratively until thematic saturation was achieved. Interviews were analyzed using thematic analysis with a constant comparative method, incorporating both deductive and inductive coding. Resulting subthemes were mapped onto corresponding domains of the PRISM and Sociotechnical Systems frameworks. Results: A total of 25 physicians were interviewed, including 7 who had never used the AI scribe, 9 who used it minimally, and 9 who used it frequently. Participants represented a broad range of specialties. A total of 5 themes and 15 subthemes emerged. First, workflow integration influenced adoption. Limitations in customization were a barrier to adoption, while flexibility among users was critical to fit. Second, ease of tool usability facilitated initial adoption, whereby users tried the tool without paying close attention to training materials. However, this lack of training engagement hindered widespread sustained use due to limited knowledge of advanced functionalities. Third, attending physicians balanced trust and confidence in AI-generated documentation with mistrust and professional accountability. Many described a “trust but verify” approach. Also, some concerns about others’ inadequate review of AI-generated content weakened physician-physician trust. Fourth, attending physicians viewed the ubiquitous presence of AI scribing as inevitable, driven by organizational priorities related to productivity, cost, and burnout. Interviewees disagreed about the appropriate use of AI scribes by trainees. Finally, identical mechanisms were viewed as both benefits and drawbacks. Increased efficiency from AI documentation was perceived by some as reducing burden and cognitive load, but perceived by others as potentially intensifying workload expectations. Similarly, some thought AI scribes facilitated more patient-centered communication, while others thought it had the opposite impact. Conclusions: Implementation of AI scribes in ambulatory care is shaped by complex sociotechnical and contextual factors. Adoption is influenced by workflow alignment, trust, and preexisting beliefs about the tool’s benefits and drawbacks, illustrating how clinical context affects AI scribe implementation.

Generating Population-Linked Case Vignettes From Nationwide Triage Data: Framework Development and Validation Study

2026-11-03T15:00:30-05:00

Background: Digital decision-support tools such as triage systems and symptom checkers support millions of health-related decisions each year. Their quality and safety are commonly evaluated using textual patient cases, known as case vignettes. However, expert-written vignette sets cover only a limited spectrum of real-world patient presentations and lack population weights, which would allow extrapolating evaluation results to the underlying patient population. Objective: This study aimed to develop a data-driven framework for automatically generating a human-manageable set of case vignettes from nationwide triage data that capture broad presentation diversity and link each vignette to a quantitative weight reflecting the number of underlying patient assessments. Methods: From 3.2 million triage assessments conducted over 1 year using structured triage software in the German medical on-call service (telephone triage and online self-triage) and at joint contact points of the outpatient emergency care service and hospital emergency departments, we randomly sampled 50,000 cases. Triage questionnaires were converted into semantic embeddings using a German Sentence Transformer model and grouped by agglomerative clustering. For clusters containing sufficient assessments, we generated one representative assessment using a 2-phase simulated annealing optimization that minimized the distance from the cluster centroid while maximizing the number of answered triage questions. Each representative assessment was assigned the size of its source cluster as its sample-based weight. A similarity-based sensitivity analysis assessed whether these weights were preserved in the full 1-year population. Finally, question-answer pairs of the representative assessments were converted into textual case vignettes using controlled prompting of a large language model. Results: The cluster analysis yielded 514 included clusters covering 96.8% of the sampled 50,000 assessments. The generated representatives showed strong agreement with the majority treatment urgency recommendation of their source cluster (Spearman ρ=0.78; <.001) and contained, on average, 4.3 more answered triage questions than the original assessments within their clusters. When weighted by cluster size, the representatives approximated the sample distributions of treatment urgency, demographics, and symptoms, although some systematic deviations remained, most notably an overrepresentation of female cases (13.5%), patients aged 14‐49 (8.0%) years, and the urgency category “As soon as possible” (6.6%). Of 121 recorded symptoms, 101 (83.5%) were covered by the representatives; the rest each occurred in <0.5% of the sample. Cluster-based vignette weights were strongly correlated with similarity-based population weights (Spearman ρ=0.77; <.001), and 90.1% of assessments in the full 1-year population matched at least one vignette. Conclusions: We present a data-driven framework for deriving a manageable set of population-weighted case vignettes from nationwide triage data. The resulting vignettes captured broad presentation diversity, approximated key sample characteristics, and provided an explicit quantitative link to the number of underlying assessments. After medical expert review and refinement, the vignettes may support more population-aware evaluation and quality assurance of digital decision-support tools.

Machine Learning–Based Detection of Prophylactic Indication Mentions for Stress-Related Mucosal Disease in Chinese Clinical Progress Notes: Text Classification Study

2026-11-02T16:00:28-05:00

Background: Machine learning has been extensively applied in medical prediction; however, existing machine-assisted pharmacotherapy review systems predominantly rely on structured, codifiable data while neglecting unstructured natural language clinical narratives, thereby allowing latent medication risks to remain undetected. Objective: To address the limitations of structured data–based review, this study proposes and validates a machine learning framework grounded in text classification for the automated identification of “prophylactic” vs “nonprophylactic” medication indications within clinical narratives. As a result, it aims to assist in the automated detection of prophylactic indication mentions for stress-related mucosal disease in clinical narratives as a preliminary step toward comprehensive medication-use review. Methods: A total of 3695 progress notes were independently annotated by 2 clinical pharmacists to construct a binary dataset for classifying prophylactic medication indication for stress-related mucosal disease, yielding an interannotator agreement κ of 0.8934. Comparative experiments were conducted using traditional machine learning models, including logistic regression, support vector machine, random forest, CatBoost, and LightGBM, alongside TextCNN and a Chinese pretrained BERT model. The dataset was split into training, validation, and test sets on a patient-wise basis, and model evaluation was conducted on the independent test set. SHAP (Shapley additive explanations) analysis was used to achieve model interpretability and elucidate decision-making processes, ultimately enabling automated indication detection of prophylactic proton pump inhibitor administration. Results: All models and strategies demonstrated efficacy in medical text classification tasks. In a single-center cohort of 66 patients (3695 notes, 11.0% positive), the CatBoost model with balanced class weights achieved a note-level -score of 0.727 (precision=0.693; recall=0.765) and a patient-level -score of 0.889 (precision=0.800; recall=1.000) on the independent test set. At the note level, CatBoost achieved a higher -score than TextCNN (-score=0.718) and BERT (-score=0.628). Conclusions: In summary, the text classifier constructed through machine learning and natural language processing effectively extracts prophylactic medication indications for stress-related mucosal disease from unstructured progress notes. Combined with patient-level data partitioning, standardized annotation protocols, and an ethics-compliant framework, this approach provides an interpretable preliminary screening tool for identifying potential prophylactic indications to support subsequent pharmacist adjudication. Trial Registration: Chinese Clinical Trial Registry ChiCTR2600126317; https://tinyurl.com/yjzjd443

Exchange of High-Frequency Medical Monitoring Data Using Fast Healthcare Interoperability Resources for Interoperable Integration Into Clinical Practice and Research: Methodological Development Study

2026-11-02T16:00:28-05:00

Background: Continuous patient monitoring generates vast volumes of high-frequency monitoring data, such as electrocardiogram, electroencephalography, and invasive blood pressure tracings, that are often persisted in heterogeneous, noninteroperable formats, limiting integration with electronic health records and secondary use. Although Fast Healthcare Interoperability Resources (FHIR) supports general health data exchange, native support for high-frequency medical monitoring data, including associated metadata and data quality metrics, remains limited. Objective: This study aims to develop a modality-agnostic FHIR package implementing a pattern to enable scalable, semantically precise representation of high-frequency medical monitoring data, decoupling structured clinical metadata from raw signal storage. Methods: This study provides the FHIR package Time Series Integration and Data in Endpoints (TIDE), a collection of FHIR Release 4 artifacts for interoperable exchange and the standardization of high-frequency monitoring clinical data, using electroencephalography as the primary use case while remaining applicable to generic medical monitoring data. The core TIDE profile separates technical metadata and analytical summaries stored directly in FHIR from high-volume raw data referenced via resources. Quantitative results are standardized with pattern slicing, including sampling frequency, statistical summaries, and data quality metrics. Semantic interoperability was achieved through a hybrid terminology binding, using Systematized Nomenclature of Medicine Clinical Terms, Logical Observation Identifiers Names and Codes for modalities, and custom for technical metadata, waveform modalities, channel identifiers, and connection types not covered by standard terminologies. All profiles were authored in FHIR Shorthand and compiled with SUSHI to produce a validated Implementation Guide conformant with the FHIR Release 4 core specification. Results: The resulting framework provides a modality-agnostic structure for high-frequency monitoring data, comprising 7 primary FHIR artifacts that define 16 distinct component slices for metadata and signal analysis results. resources reference associated and instances, encapsulate acquisition parameters, and link to full-time-series repositories. Controlled vocabularies ensure unambiguous exchange, thereby enabling downstream use in research and decision support. The architecture decouples the metadata layer from the storage layer via a standardized extension, enabling efficient querying of signals without retrieving large amounts of raw high-frequency monitoring data. Conclusions: TIDE provides a reusable, standards-compliant framework for integrating continuous high-frequency monitoring data into FHIR-based ecosystems. By separating semantically rich, queryable metadata from raw signal storage, TIDE enables longitudinal patient monitoring and automated signal quality-based triage without requiring modification of existing signal storage infrastructure.