JMIR Medical Informatics最新文献

Background: Although an increasing number of bedside medical devices are equipped with wireless connections for reliable notifications, many nonnetworked devices remain effective at detecting abnormal patient conditions and alerting medical staff through auditory alarms. Staff members, however, can miss these notifications, especially when in distant areas or other private rooms. In contrast, the signal-to-noise ratio of alarm systems for medical devices in the neonatal intensive care unit is 0 dB or higher. A feasible system for automatic sound identification with high accuracy is needed to prevent alarm sounds from being missed by the staff.

Objective: The purpose of this study was to design a method for classifying multiple alarm sounds collected with a monaural microphone in a noisy environment.

Methods: Features of 7 alarm sounds were extracted using a mel filter bank and incorporated into a classifier using convolutional and recurrent neural networks. To estimate its clinical usefulness, the classifier was evaluated with mixtures of up to 7 alarm sounds and hospital ward noise.

Results: The proposed convolutional recurrent neural network model was evaluated using a simulation dataset of 7 alarm sounds mixed with hospital ward noise. At a signal-to-noise ratio of 0 dB, the best-performing model (convolutional neural network 3+bidirectional gate recurrent unit) achieved an event-based F1-score of 0.967, with a precision of 0.944 and a recall of 0.991. When the venous foot pump class was excluded, the classwise recall of the classifier ranged from 0.990 to 1.000.

Conclusions: The proposed classifier was found to be highly accurate in detecting alarm sounds. Although the performance of the proposed classifier in a clinical environment can be improved, the classifier could be incorporated into an alarm sound detection system. The classifier, combined with network connectivity, could improve the notification of abnormal status detected by unconnected medical devices.

Background: Predicting serious hematological adverse events (SHAEs) from poly (adenosine diphosphate ribose) polymerase inhibitors (PARPis) would allow us to prioritize patients with ovarian cancer at higher risk for more intensive care, ultimately lowering morbidity and preventing them from premature termination of medication.

Objective: This study aimed to explore the risk factors for SHAEs in patients with ovarian cancer receiving PARPi treatment and develop a risk prediction model for such events.

Methods: Prospective clinical data were collected on patients with ovarian cancer who received PARPi treatment at the Guangxi Medical University Affiliated Tumor Hospital from December 2018 to August 2024. They were divided into a SHAE group and a no-SHAE group based on the occurrence of SHAEs. Variable differences were screened using the chi-square test or Fisher exact test. Multivariate logistic regression was used to determine independent factors influencing SHAEs in patients with ovarian cancer. A predictive model for serious blood-related complications in ovarian cancer treatment was developed from identified independent risk factors using the R software. The model's clinical utility was assessed through decision curve analysis (net benefit), calibration (calibration curve), and discrimination (receiver operating characteristic curve).

Results: A total of 70 patients with ovarian cancer receiving PARPi treatment were included in this study. Of these 70 patients, 16 (23%) experienced SHAEs, with decreases in red blood cell (RBC) count and hemoglobin levels being the most common. Multiple logistic regression analysis identified 4 independent predictors of PARPi-associated SHAEs in patients with ovarian cancer: lymph node metastasis (odds ratio [OR] 6.733, 95% CI 1.197-37.873; P=.03), creatinine clearance rate of ≤60 mL per minute (OR 23.722, 95% CI 3.121-180.303; P=.002), RBC count of ≤3.3×10¹² per liter (OR 4.847, 95% CI 1.020-23.041; P=.047), and combination therapy with vascular endothelial growth factor inhibitors (OR 6.749, 95% CI 1.313-34.689; P=.02). The internal validation yielded an area under the curve of 0.874 (95% CI 0.793-0.955), indicating moderate clinical utility and accuracy for the risk prediction model incorporating these predictors.

Conclusions: Lymph node metastasis, creatinine clearance rate of ≤60 mL per minute, RBC count of ≤3.3×10¹² per liter, and combination therapy with vascular endothelial growth factor inhibitors are independent risk factors for PARPi SHAEs in patients with ovarian cancer. The risk prediction model established based on these factors demonstrated moderate predictive value.

Background: Delayed extubation after general anesthesia increases complications and can lead to longer hospital stays and higher mortality. Current risk assessments often rely on subjective judgment or simple tools, whereas machine learning offers potential for real-time evaluation, though research is limited and typically uses single-algorithm models.

Objective: The aims of this study were to identify risk factors for delayed extubation after general anesthesia in the sample and to construct a risk prediction model for delayed extubation in this population.

Methods: Data from 4779 patients admitted to the postanesthesia care unit between September 2023 and May 2024 were used to develop prediction models for delayed extubation using k-nearest neighbor, decision tree, extreme gradient boosting, random forest, a light gradient boosting machine, and an artificial neural network. Model performance was assessed by calculating the area under the receiver operating characteristic curve, sensitivity, specificity, accuracy, F1-score, and Brier score. Calibration performance was evaluated using calibration curves generated with 100-bin quantile calibration and Loess smoothing to provide bias-corrected and smoothed visual assessment. Additionally, the Hosmer-Lemeshow goodness-of-fit test was performed to quantitatively evaluate calibration, with P values >.05 indicating good calibration.

Results: Among the 6 models evaluated, the extreme gradient boosting model demonstrated the best performance, with an area under the receiver operating characteristic curve of 0.750 (95% CI 0.703-0.796), a sensitivity of 0.734 (95% CI 0.635-0.827), and a specificity of 0.647 (95% CI 0.623-0.673). The model calibration was acceptable, with a Brier score of 0.0505 and a nonsignificant Hosmer-Lemeshow goodness-of-fit test (χ²6=7.3; P=.287), indicating good calibration. Shapley additive explanations were used to rank feature importance.

Conclusions: These machine learning models enable early identification of delayed extubation risk, supporting personalized clinical decisions and optimizing postanesthesia care unit resource allocation.

Background: The diagnosis of sleep disorders presents a challenging landscape, characterized by the complex nature of their assessment and the often divergent views between objective clinical assessment and subjective patient experience. This study explores the interplay between these perspectives, focusing on the variability of individual perceptions of sleep quality and latency.

Objective: Our primary goal was to investigate the alignment, or lack thereof, between subjective experiences and objective measures in the assessment of sleep disorders.

Methods: To study this, we developed an aspect-based sentiment analysis method for clinical narratives: using large language models (Falcon 40B and Mixtral 8X7B), we are identifying entity groups of 3 aspects related to sleep behavior (day sleepiness, sleep quality, and fatigue). To phrases referring to these aspects, we are assigning sentiment values between 0 and 1 using a BERT-BiLSTM-based approach (accuracy 78%) and a fine-tuned GPT-2 sentiment classifier (accuracy 87%).

Results: In a cohort of 100 patients with complete subjective (Karolinska Sleepiness Scale [KSS]) and objective (Multiple Sleep Latency Test [MSLT]) assessments, approximately 15% exhibited notable discrepancies between perceived and measured levels of daytime sleepiness. A paired-sample t test comparing KSS scores to MSLT latencies approached statistical significance (t99=2.456; P=.06), suggesting a potential misalignment between subjective reports and physiological markers. In contrast, the comparison using text-derived sentiment scores revealed a statistically significant divergence (t99=2.324; P=.047), indicating that clinical narratives may more reliably capture discrepancies in sleepiness perception. These results underscore the importance of integrating multiple subjective sources, with an emphasis on narrative free text, in the assessment of domains such as fatigue and daytime sleepiness-where standardized measures may not fully reflect the patient's lived experience.

Conclusions: Our method has potential in uncovering critical insights into patient self-perception versus clinical evaluations, which enables clinicians to identify patients requiring objective verification of self-reported symptoms.

Background: Extracting genetic phenotype mentions from clinical reports and normalizing them to standardized concepts within the human phenotype ontology are essential for consistent interpretation and representation of genetic conditions. This is particularly important in fields such as dysmorphology and plays a key role in advancing personalized health care. However, modern clinical named entity recognition methods face challenges in accurately identifying discontinuous mentions (ie, entity spans that are interrupted by unrelated words), which can be found in these clinical reports.

Objective: This study aims to develop a system that can accurately extract and normalize genetic phenotypes, specifically from physical examination reports related to dysmorphology assessment. These mentions appear in both continuous and discontinuous lexical forms, with a focus on addressing challenging discontinuous entity spans.

Methods: We introduce DiscHPO, a 2-phase pipeline consisting of a sequence-to-sequence named entity recognition model for span extraction, and an entity normalizer that uses a sentence transformer biencoder for candidate generation and a cross-encoder reranker for selecting the best candidate as the normalized concept. This system was tested as part of our participation in Track 3 of the BioCreative VIII shared task.

Results: For overall performance on the test set, the top-performing model for entity normalization achieved an F₁-score of 0.723, while the best span extraction model reached an F₁-score of 0.665. Both scores surpassed those of 2 baseline models using the same dataset, indicating superior efficacy in handling both continuous and discontinuous spans. On the validation set, we were able to demonstrate our system's ability to recognize these mentions, with the model achieving an F₁-score of 0.631 for exact match on discontinuous spans only.

Conclusions: The findings suggest that exact extraction of entity spans may not always be necessary for successful normalization. Partial mention matches can be sufficient as long as they capture the essential concept information, supporting the system's utility in clinical downstream tasks.

Background: Large language models (LLMs) are increasingly explored in nursing education, but their capabilities in specialized, high-stakes, culturally specific examinations, such as the Chinese National Nurse Licensure Examination (CNNLE), remain underevaluated, making rigorous evaluation crucial before their adoption in nursing training and practice.

Objective: This study aimed to evaluate the performance, accuracy, repeatability, confidence, and robustness of 4 LLMs on the CNNLE.

Methods: Four LLMs (Sider Fusion [Vidline Inc], GPT-4o [OpenAI], Gemini 2.0 Pro [Google DeepMind], and DeepSeek V3) were tested on 237 multiple-choice questions from the 2024 CNNLE. Accuracy and repeatability were assessed using 2 prompting strategies. Confidence was evaluated via self-ratings (1-10 scale) and robustness via repeated adversarial prompting.

Results: DeepSeek V3 and Gemini 2.0 Pro demonstrated significantly higher overall accuracy (ranging from 199/237 to 209/237; >83%) compared to GPT-4o and Sider Fusion (ranging from 151/237 to 166/237; <71%). However, all LLMs showed suboptimal repeatability (highest at 206/237; <87% consistency). Critically, poor confidence calibration was evident; most models showed high confidence often mismatching actual accuracy (Sider Fusion: P=.01; GPT-4o: P=.03; and Gemini 2.0 Pro: P=.049). A stability-flexibility trade-off paradox was also observed.

Conclusions: While some LLMs show promising accuracy on the CNNLE, fundamental reliability limitations (poor confidence calibration and inconsistent repeatability) hinder safe application in nursing education and practice. Future LLM development must prioritize trustworthiness and calibrated reliability over surface accuracy.

Background: Pelvic organ prolapse (POP) and stress urinary incontinence (SUI) often concurrently exist. The incontinence in some patients with POP resolves after POP surgery, but it persists in others. Some patients without SUI before surgery may develop de novo SUI. It is unclear whether a concomitant anti-incontinence procedure should be performed at the time of POP surgery to prevent postoperative incontinence. A prediction model is needed to guide clinical decision-making.

Objective: This study aimed to analyze the risk factors and develop prediction models for SUI after POP surgery based on machine learning to provide new tools for evaluating and predicting postoperative SUI.

Methods: Sample size calculation was performed using the Riley 4-step method. Data of patients undergoing prolapse surgery in Shanxi Bethune Hospital were prospectively collected from August 2022 to February 2025 and were retrospectively collected from January 2021 to August 2022. General clinical data, relevant laboratory test results, urodynamic examination findings, and pelvic floor ultrasound findings were collected. Lasso regression, univariate analysis, and logistic analysis were used to screen the predictors of SUI after prolapse surgery. Data were split randomly in a 7:3 ratio into training and validation sets. The training set was used to develop the prediction model involving Lasso regression, random forest, support vector machine (SVM), extreme gradient boosting (XGBoost), classification and regression tree (CART), and logistic regression, and the validation set was used for internal verification. The final implementation was achieved by developing a Shiny-based application for model deployment.

Results: A total of 286 patients were enrolled in this study, and 91 patients had postoperative SUI. The following 6 risk factors were identified through univariate, logistic, and Lasso regression analyses: preoperative SUI, urge urinary incontinence, urodynamic occult SUI, anti-incontinence surgery, genital hiatus, and anterior colporrhaphy. Five prediction models were constructed by using logistic regression, random forest, XGBoost, SVM, and CART. Based on a comprehensive evaluation of model discrimination, calibration, and clinical utility, the SVM model demonstrated optimal overall performance, with an area under the curve of 0.821 in the training set and 0.846 in the validation set.

Conclusions: This study developed 5 prediction models for postoperative SUI following prolapse surgery, which demonstrated good performance in internal validation. Among them, the SVM prediction model appeared to be the most promising. However, further external validation data are required to assess its generalizability. This model has the potential to become a high-quality clinical risk prediction tool for postoperative SUI in patients with prolapse, guiding clinical decisi

Background: Depression is a critical psychological disorder necessitating urgent assessment and treatment, given its strong association with increased suicide risk (SR). Effective management hinges on promptly identifying individuals with high depression severity (DS) and SR. While machine learning and deep learning have advanced the identification of DS and SR, research focusing on both aspects simultaneously remains limited and requires further refinement.

Objective: This study aimed to evaluate whether our proposed methods, which integrate multitask learning (MTL), multimodal learning, and transfer learning, enhance the efficacy of deep learning models in the joint classification of DS and SR.

Methods: This study proposed a multitask framework employing a multimodal fusion strategy for pretrained audio and text embeddings to concurrently assess DS and SR. Data encompassing Chinese audio recordings and clinical questionnaire scores from 100 patients with depression and 100 healthy controls were used. Preprocessed audio and text data were transformed into pretrained embeddings and integrated using concatenation and hard parameter sharing. Single-task learning (STL) models (DS and SR tasks) were evaluated with different embeddings and further compared with the MTL models.

Results: The STL models demonstrated exceptional DS prediction (area under the curve [AUC]=0.878) using wav2vec 2.0 combined with ERNIE-health, and SR prediction (AUC=0.876) using HuBERT combined with ERNIE-health. The MTL models significantly improved SR prediction over DS prediction, achieving the highest DS classification (AUC=0.887) with wav2vec 2.0 combined with ERNIE-health, and SR classification (AUC=0.883) with HuBERT combined with ERNIE-health.

Conclusions: The findings of this study underscore the effectiveness of the proposed MTL models using specific pretrained audio and text embeddings in enhancing model performance. However, we advocate for cautious implementation of MTL to mitigate potential negative transfer effects. Our research presents a method that is both promising and effective, offering an objective approach for accurate clinical decision support in the parallel diagnosis of DS and SR.