JMIR Medical Informatics最新文献

Background: Cohort studies contain rich clinical data across large and diverse patient populations and are a common source of observational data for clinical research. Because large scale cohort studies are both time and resource intensive, one alternative is to harmonize data from existing cohorts through multicohort studies. However, given differences in variable encoding, accurate variable harmonization is difficult.

Objective: We propose SONAR (Semantic and Distribution-Based Harmonization) as a method for harmonizing variables across cohort studies to facilitate multicohort studies.

Methods: SONAR used semantic learning from variable descriptions and distribution learning from study participant data. Our method learned an embedding vector for each variable and used pairwise cosine similarity to score the similarity between variables. This approach was built off 3 National Institutes of Health cohorts, including the Cardiovascular Health Study, the Multi-Ethnic Study of Atherosclerosis, and the Women's Health Initiative. We also used gold standard labels to further refine the embeddings in a supervised manner.

Results: The method was evaluated using manually curated gold standard labels from the 3 National Institutes of Health cohorts. We evaluated both the intracohort and intercohort variable harmonization performance. The supervised SONAR method outperformed existing benchmark methods for almost all intracohort and intercohort comparisons using area under the curve and top-k accuracy metrics. Notably, SONAR was able to significantly improve harmonization of concepts that were difficult for existing semantic methods to harmonize.

Conclusions: SONAR achieves accurate variable harmonization within and between cohort studies by harnessing the complementary strengths of semantic learning and variable distribution learning.

Background: Health data typically include patient-generated data and clinical medical data. Different types of data contribute to disease prevention, precision medicine, and the overall improvement of health care. With the introduction of regulations such as the Health Insurance Portability and Accountability Act (HIPAA), individuals play a key role in the sharing and application of personal health data.

Objective: This study aims to explore the impact of different types of health data on users' willingness to share. Additionally, it analyzes the effect of data control and delay discounting rate on this process.

Methods: The results of a web-based survey were analyzed to examine individuals' perceptions of sharing different types of health data and how data control and delay discounting rates influenced their decisions. We recruited participants for our study through the web-based platform "Wenjuanxing." After screening, we obtained 257 valid responses. Regression analysis was used to investigate the impact of data control, delayed discounting, and mental accounting on the public's willingness to share different types of health care data.

Results: Our findings indicate that the type of health data does not significantly affect the perceived benefits of data sharing. Instead, it negatively influences willingness to share by indirectly affecting data acquisition costs and perceived risks. Our results also show that data control reduces the perceived risks associated with sharing, while higher delay discounting rates lead to an overestimation of data acquisition costs and perceived risks.

Conclusions: Individuals' willingness to share data is primarily influenced by costs. To promote the acquisition and development of personal health data, stakeholders should strengthen individuals' control over their data or provide direct short-term incentives.

Background: Prediction models have demonstrated a range of applications across medicine, including using electronic health record (EHR) data to identify hospital readmission and mortality risk. Large language models (LLMs) can transform unstructured EHR text into structured features, which can then be integrated into statistical prediction models, ensuring that the results are both clinically meaningful and interpretable.

Objective: This study aims to compare the classification decisions made by clinical experts with those generated by a state-of-the-art LLM, using terms extracted from a large EHR data set of individuals with mental health disorders seen in emergency departments (EDs).

Methods: Using a dataset from the EHR systems of more than 50 health care provider organizations in the United States from 2016 to 2021, we extracted all clinical terms that appeared in at least 1000 records of individuals admitted to the ED for a mental health-related problem from a source population of over 6 million ED episodes. Two experienced mental health clinicians (one medically trained psychiatrist and one clinical psychologist) reached consensus on the classification of EHR terms and diagnostic codes into categories. We evaluated an LLM's agreement with clinical judgment across three classification tasks as follows: (1) classify terms into "mental health" or "physical health", (2) classify mental health terms into 1 of 42 prespecified categories, and (3) classify physical health terms into 1 of 19 prespecified broad categories.

Results: There was high agreement between the LLM and clinical experts when categorizing 4553 terms as "mental health" or "physical health" (κ=0.77, 95% CI 0.75-0.80). However, there was still considerable variability in LLM-clinician agreement on the classification of mental health terms (κ=0.62, 95% CI 0.59-0.66) and physical health terms (κ=0.69, 95% CI 0.67-0.70).

Conclusions: The LLM displayed high agreement with clinical experts when classifying EHR terms into certain mental health or physical health term categories. However, agreement with clinical experts varied considerably within both sets of mental and physical health term categories. Importantly, the use of LLMs presents an alternative to manual human coding, presenting great potential to create interpretable features for prediction models.

Unlabelled: The health care industry is currently going through a transformation due to the integration of technologies and the shift toward value-based health care (VBHC). This article explores how digital health solutions play a role in advancing VBHC, highlighting both the challenges and opportunities associated with adopting these technologies. Digital health, which includes mobile health, wearable devices, telehealth, and personalized medicine, shows promise in improving diagnostic accuracy, treatment options, and overall health outcomes. The article delves into the concept of transformation in health care by emphasizing its potential to reform care delivery through data communication, patient engagement, and operational efficiency. Moreover, it examines the principles of VBHC, with a focus on patient outcomes, and emphasizes how digital platforms play a role in treatment among tertiary hospitals by using patient-reported outcome measures. The article discusses challenges that come with implementing VBHC, such as stakeholder engagement and standardization of patient-reported outcome measures. It also highlights the role played by health innovators in facilitating the transition toward VBHC models. Through real-life case examples, this article illustrates how digital platforms have had an impact on efficiencies, patient outcomes, and empowerment. In conclusion, it envisions directions for solutions in VBHC by emphasizing the need for interoperability, standardization, and collaborative efforts among stakeholders to fully realize the potential of digital transformation in health care. This research highlights the impact of digital health in creating a health care system that focuses on providing high-quality, efficient, and patient-centered care.

Background: Postpartum depression (PPD) is a prevalent mental health issue with significant impacts on mothers and families. Exploring reliable predictors is crucial for the early and accurate prediction of PPD, which remains challenging.

Objective: This study aimed to comprehensively collect variables from multiple aspects, develop and validate machine learning models to achieve precise prediction of PPD, and interpret the model to reveal clinical implications.

Methods: This study recruited pregnant women who delivered at the West China Second University Hospital, Sichuan University. Various variables were collected from electronic medical record data and screened using least absolute shrinkage and selection operator penalty regression. Participants were divided into training (1358/2055, 66.1%) and validation (697/2055, 33.9%) sets by random sampling. Machine learning-based predictive models were developed in the training cohort. Models were validated in the validation cohort with receiver operating curve and decision curve analysis. Multiple model interpretation methods were implemented to explain the optimal model.

Results: We recruited 2055 participants in this study. The extreme gradient boosting model was the optimal predictive model with the area under the receiver operating curve of 0.849. Shapley Additive Explanation indicated that the most influential predictors of PPD were antepartum depression, lower fetal weight, elevated thyroid-stimulating hormone, declined thyroid peroxidase antibodies, elevated serum ferritin, and older age.

Conclusions: This study developed and validated a machine learning-based predictive model for PPD. Several significant risk factors and how they impact the prediction of PPD were revealed. These findings provide new insights into the early screening of individuals with high risk for PPD, emphasizing the need for comprehensive screening approaches that include both physiological and psychological factors.

Background: Studies suggest that less than 4% of patients with pulmonary embolisms (PEs) are managed in the outpatient setting. Strong evidence and multiple guidelines support the use of the Pulmonary Embolism Severity Index (PESI) for the identification of acute PE patients appropriate for outpatient management. However, calculating the PESI score can be inconvenient in a busy emergency department (ED). To facilitate integration into ED workflow, we created a 2023 Epic-compatible clinical decision support tool that automatically calculates the PESI score in real-time with patients' electronic health data (ePESI [Electronic Pulmonary Embolism Severity Index]).

Objective: The primary objectives of this study were to determine the overall accuracy of ePESI and its ability to correctly distinguish high- and low-risk PESI scores within the Epic 2023 software. The secondary objective was to identify variables that impact ePESI accuracy.

Methods: We collected ePESI scores on 500 consecutive patients at least 18 years old who underwent a computerized tomography-pulmonary embolism scan in the ED of our tertiary, academic health center between January 3 and February 15, 2023. We compared ePESI results to a PESI score calculated by 2 independent, medically-trained abstractors blinded to the ePESI and each other's results. ePESI accuracy was calculated with binomial test. The odds ratio (OR) was calculated using logistic regression.

Results: Of the 500 patients, a total of 203 (40.6%) and 297 (59.4%) patients had low- and high-risk PESI scores, respectively. The ePESI exactly matched the calculated PESI in 394 out of 500 cases, with an accuracy of 78.8% (95% CI 74.9%-82.3%), and correctly identified low- versus high-risk in 477 out of 500 (95.4%) cases. The accuracy of the ePESI was higher for low-risk scores (OR 2.96, P<.001) and lower when patients were without prior encounters in the health system (OR 0.42, P=.008).

Conclusions: In this single-center study, the ePESI was highly accurate in discriminating between low- and high-risk scores. The clinical decision support should facilitate real-time identification of patients who may be candidates for outpatient PE management.

Background: Residents of facilities for older people are vulnerable to COVID-19 outbreaks. Nevertheless, timely recognition of outbreaks at facilities for older people at public health centers has been impossible in Japan since May 8, 2023, when the Japanese government discontinued aggressive countermeasures against COVID-19 because of the waning severity of the dominant Omicron strain. The Facility for Elderly Surveillance System (FESSy) has been developed to improve information collection.

Objective: This study examined FESSy experiences and effectiveness in two public health center jurisdictions in Japan.

Methods: This study assessed the use by public health centers of the detection mode of an automated AI detection system (ie, FESSy AI), as well as manual detection by the public health centers' staff (ie, FESSy staff) and direct reporting by facilities to the public health centers. We considered the following aspects: (1) diagnoses or symptoms, (2) numbers of patients as of their detection date, and (3) ultimate numbers of patients involved in incidents. Subsequently, effectiveness was assessed and compared based on detection modes. The study lasted from June 1, 2023, through January 2024.

Results: In both areas, this study examined 31 facilities at which 87 incidents were detected. FESSy (AI or staff) detected significantly fewer patients than non-FESSy methods, that is, direct reporting to the public health center of the detection date and ultimate number of patients.

Conclusions: FESSy was superior to direct reporting from facilities for the number of patients as of the detection date and for the ultimate outbreak size.