Background: The automatic coding of clinical text documents by using the International Classification of Diseases, 10th Revision (ICD-10) can be performed for statistical analyses and reimbursements. With the development of natural language processing models, new transformer architectures with attention mechanisms have outperformed previous models. Although multicenter training may increase a model’s performance and external validity, the privacy of clinical documents should be protected. We used federated learning to train a model with multicenter data, without sharing data per se. Objective: This study aims to train a classification model via federated learning for ICD-10 multilabel classification. Methods: Text data from discharge notes in electronic medical records were collected from the following three medical centers: Far Eastern Memorial Hospital, National Taiwan University Hospital, and Taipei Veterans General Hospital. After comparing the performance of different variants of bidirectional encoder representations from transformers (BERT), PubMedBERT was chosen for the word embeddings. With regard to preprocessing, the nonalphanumeric characters were retained because the model’s performance decreased after the removal of these characters. To explain the outputs of our model, we added a label attention mechanism to the model architecture. The model was trained with data from each of the three hospitals separately and via federated learning. The models trained via federated learning and the models trained with local data were compared on a testing set that was composed of data from the three hospitals. The micro F1 score was used to evaluate model performance across all 3 centers. Results: The F1 scores of PubMedBERT, RoBERTa (Robustly Optimized BERT Pretraining Approach), ClinicalBERT, and BioBERT (BERT for Biomedical Text Mining) were 0.735, 0.692, 0.711, and 0.721, respectively. The F1 score of the model that retained nonalphanumeric characters was 0.8120, whereas the F1 score after removing these characters was 0.7875—a decrease of 0.0245 (3.11%). The F1 scores on the testing set were 0.6142, 0.4472, 0.5353, and 0.2522 for the federated learning, Far Eastern Memorial Hospital, National Taiwan University Hospital, and Taipei Veterans General Hospital models, respectively. The explainable predictions were displayed with highlighted input words via the label attention architecture. Conclusions: Federated learning was used to train the ICD-10 classification model on multicenter clinical text while protecting data privacy. The model’s performance was better than that of models that were trained locally.

Early enteral nutrition (EN) and a nutrition target >60% are recommended for patients in the intensive care unit (ICU), even for those with acute respiratory distress syndrome (ARDS). Prolonged prone positioning (PP) therapy (>48 h) is the rescue therapy of ARDS, but it may worsen the feeding status because it requires the heavy sedation and total paralysis of patients. Our previous studies demonstrated that energy achievement rate (EAR) >65% was a good prognostic factor in ICU. However, its impact on the mortality of patients with ARDS requiring prolonged PP therapy remains unclear. We retrospectively analyzed 79 patients with high nutritional risk (modified nutrition risk in the critically ill; mNUTRIC score ≥5); and identified factors associated with ICU mortality by using a Cox regression model. Through univariate analysis, mNUTRIC score, comorbid with malignancy, actual energy intake, and EAR (%) were associated with ICU mortality. By multivariate analysis, EAR (%) was a strong predictive factor of ICU mortality (HR: 0.19, 95% CI: 0.07–0.56). EAR >65% was associated with lower 14-day, 28-day, and ICU mortality after adjustment for confounding factors. We suggest early EN and increase EAR >65% may benefit patients with ARDS who required prolonged PP therapy.

Early and prolonged prone positioning (PP) therapy improve survival in advanced ARDS; however, the predictors of mortality remain unclear. The study aims to identify predictive factors correlated with mortality and build-up the prognostic score in patients with severe ARDS who received early and prolonged PP therapy. A total of 116 patients were enrolled in this retrospective cohort study. Univariate and multivariate regression models were used to estimate the odds ratio (OR) of mortality. Factors associated with mortality were assessed by Cox regression analysis and presented as the hazard ratio (HR) and 95% CI. In the multivariate regression model, renal replacement therapy (RRT; OR: 4.05, 1.54–10.67), malignant comorbidity (OR: 8.86, 2.22–35.41), and non-influenza-related ARDS (OR: 5.17, 1.16–23.16) were significantly associated with ICU mortality. Age, RRT, non-influenza-related ARDS, malignant comorbidity, and APACHE II score were included in a composite prone score, which demonstrated an area under the curve of 0.816 for predicting mortality risk. In multivariable Cox proportional hazard model, prone score more than 3 points was significantly associated with ICU mortality (HR: 2.13, 1.12–4.07, p = 0.021). We suggest prone score ≥3 points could be a good predictor for mortality in severe ARDS received PP therapy.

Middle ear diseases, such as otitis media and middle ear effusion, are difficult to accurately detect in primary care. We developed an AI-powered system using Azure OpenAI’s GPT-4 Vision, the first multimodal large language model (LLM) applied to analyze otoscopic images and detect ten middle ear conditions. Built with a Model-View-Controller (MVC) architecture, the system processes images in under 5 seconds and provides bilingual (English and Chinese) detection reports with confidence scores and treatment recommendations. This scalable solution integrates advanced image analysis and natural language generation to enhance workflows in telemedicine and primary care settings.

Background: Accurate transcription of pathology gross examination dictation is important for clinical documentation, but multilingual dictation remains challenging in settings where clinicians mix Chinese and English while final pathology reports are written in English. Objective: This study aimed to evaluate whether a Whisper-based automatic speech recognition (ASR) pipeline guided by contextual system messages and combined with open-source large language models (LLMs; Qwen2:72b, Llama3.1:70b, Gemma2:27b) could improve multilingual (Chinese-English) pathology dictation transcription accuracy and generate clinically appropriate English gross description reports. Methods: We conducted a controlled proof-of-concept study using 125 simulated mixed Chinese-English pathology gross examination audio recordings created by physicians or pathologists. Audio recordings were transcribed using Whisper ASR with and without a contextual system message. The ASR transcripts were then converted into English gross description reports using 3 open-source LLMs: Qwen2:72b, Llama3.1:70b, and Gemma2:27b. Outcomes included character error rate, Bilingual Evaluation Understudy, Recall-Oriented Understudy for Gisting Evaluation (ROUGE)-1, ROUGE-2, ROUGE-L, Metric for Evaluation of Translation with Explicit Ordering, pathologist Win-Tie-Lose rankings, report-level error categories, inference time, and interrater agreement. Results: The ASR contextual system message reduced the mean character error rate from 0.344 (SD 0.176; 95% CI 0.313‐0.375) to 0.066 (SD 0.100; 95% CI 0.048‐0.084; P <.001). Qwen2:72b achieved the highest automated metric scores, including a Bilingual Evaluation Understudy of 0.644 (SD 0.307), ROUGE-1 of 0.866 (SD 0.163), ROUGE-2 of 0.771 (SD 0.235), ROUGE-L of 0.842 (SD 0.178), and Metric for Evaluation of Translation with Explicit Ordering of 0.805 (SD 0.214). Pathologist-coded total error rates were 16.8% (21/125) for Qwen2:72b, 45.6% (57/125) for Llama3.1:70b, and 92.8% (116/125) for Gemma2:27b. The exact agreement between the 2 pathologists across full ranking categories was 76.8% (96/125; Cohen κ=0.668), and agreement on the top-ranked model or tied top group was 81.6% (102/125; Cohen κ=0.722). Conclusions: In this proof-of-concept evaluation, contextual prompting improved ASR transcription accuracy, and Qwen2:72b generated the most accurate English pathology reports among the evaluated LLMs. However, the study used simulated audio recordings, a local vocabulary prompt, and report-level rather than term-level clinical annotation. LLM-generated reports should therefore be considered draft documentation requiring pathologist verification, and prospective validation in real clinical workflows is needed before clinical deployment.

This paper describes the development of Health Level Seven Fast Healthcare Interoperability Resource (FHIR) profiles for pathology reports integrated with whole slide images and clinical data to create a pathology research database. A report template was designed to collect structured reports, enabling pathologists to select structured terms based on a checklist, allowing for the standardization of terms used to describe tumor features. We gathered and analyzed 190 non–small-cell lung cancer pathology reports in free text format, which were then structured by mapping the itemized vocabulary to FHIR observation resources, using international standard terminologies, such as the International Classification of Diseases, LOINC, and SNOMED CT. The resulting FHIR profiles were published as an implementation guide, which includes 25 profiles for essential data elements, value sets, and structured definitions for integrating clinical data and pathology images associated with the pathology report. These profiles enable the exchange of structured data between systems and facilitate the integration of pathology data into electronic health records, which can improve the quality of care for patients with cancer.