JMIR AI最新文献

Background: Global pandemics like COVID-19 put a high amount of strain on health care systems and health workers worldwide. These crises generate a vast amount of news information published online across the globe. This extensive corpus of articles has the potential to provide valuable insights into the nature of ongoing events and guide interventions and policies. However, the sheer volume of information is beyond the capacity of human experts to process and analyze effectively.

Objective: The aim of this study was to explore how natural language processing (NLP) can be leveraged to build a system that allows for quick analysis of a high volume of news articles. Along with this, the objective was to create a workflow comprising human-computer symbiosis to derive valuable insights to support health workforce strategic policy dialogue, advocacy, and decision-making.

Methods: We conducted a review of open-source news coverage from January 2020 to June 2022 on COVID-19 and its impacts on the health workforce from the World Health Organization (WHO) Epidemic Intelligence from Open Sources (EIOS) by synergizing NLP models, including classification and extractive summarization, and human-generated analyses. Our DeepCovid system was trained on 2.8 million news articles in English from more than 3000 internet sources across hundreds of jurisdictions.

Results: Rules-based classification with hand-designed rules narrowed the data set to 8508 articles with high relevancy confirmed in the human-led evaluation. DeepCovid's automated information targeting component reached a very strong binary classification performance of 98.98 for the area under the receiver operating characteristic curve (ROC-AUC) and 47.21 for the area under the precision recall curve (PR-AUC). Its information extraction component attained good performance in automatic extractive summarization with a mean Recall-Oriented Understudy for Gisting Evaluation (ROUGE) score of 47.76. DeepCovid's final summaries were used by human experts to write reports on the COVID-19 pandemic.

Conclusions: It is feasible to synergize high-performing NLP models and human-generated analyses to benefit open-source health workforce intelligence. The DeepCovid approach can contribute to an agile and timely global view, providing complementary information to scientific literature.

Background: Youth experiencing homelessness face substance use problems disproportionately compared to other youth. A study found that 69% of youth experiencing homelessness meet the criteria for dependence on at least 1 substance, compared to 1.8% for all US adolescents. In addition, they experience major structural and social inequalities, which further undermine their ability to receive the care they need.

Objective: The goal of this study was to develop a machine learning-based framework that uses the social media content (posts and interactions) of youth experiencing homelessness to predict their substance use behaviors (ie, the probability of using marijuana). With this framework, social workers and care providers can identify and reach out to youth experiencing homelessness who are at a higher risk of substance use.

Methods: We recruited 133 young people experiencing homelessness at a nonprofit organization located in a city in the western United States. After obtaining their consent, we collected the participants' social media conversations for the past year before they were recruited, and we asked the participants to complete a survey on their demographic information, health conditions, sexual behaviors, and substance use behaviors. Building on the social sharing of emotions theory and social support theory, we identified important features that can potentially predict substance use. Then, we used natural language processing techniques to extract such features from social media conversations and reactions and built a series of machine learning models to predict participants' marijuana use.

Results: We evaluated our models based on their predictive performance as well as their conformity with measures of fairness. Without predictive features from survey information, which may introduce sex and racial biases, our machine learning models can reach an area under the curve of 0.72 and an accuracy of 0.81 using only social media data when predicting marijuana use. We also evaluated the false-positive rate for each sex and age segment.

Conclusions: We showed that textual interactions among youth experiencing homelessness and their friends on social media can serve as a powerful resource to predict their substance use. The framework we developed allows care providers to allocate resources efficiently to youth experiencing homelessness in the greatest need while costing minimal overhead. It can be extended to analyze and predict other health-related behaviors and conditions observed in this vulnerable community.

Background: The cost of health care in many countries is increasing rapidly. There is a growing interest in using machine learning for predicting high health care utilizers for population health initiatives. Previous studies have focused on individuals who contribute to the highest financial burden. However, this group is small and represents a limited opportunity for long-term cost reduction.

Objective: We developed a collection of models that predict future health care utilization at various thresholds.

Methods: We utilized data from a multi-institutional diabetes database from the year 2019 to develop binary classification models. These models predict health care utilization in the subsequent year across 6 different outcomes: patients having a length of stay of ≥7, ≥14, and ≥30 days and emergency department attendance of ≥3, ≥5, and ≥10 visits. To address class imbalance, random and synthetic minority oversampling techniques were employed. The models were then applied to unseen data from 2020 and 2021 to predict health care utilization in the following year. A portfolio of performance metrics, with priority on area under the receiver operating characteristic curve, sensitivity, and positive predictive value, was used for comparison. Explainability analyses were conducted on the best performing models.

Results: When trained with random oversampling, 4 models, that is, logistic regression, multivariate adaptive regression splines, boosted trees, and multilayer perceptron consistently achieved high area under the receiver operating characteristic curve (>0.80) and sensitivity (>0.60) across training-validation and test data sets. Correcting for class imbalance proved critical for model performance. Important predictors for all outcomes included age, number of emergency department visits in the present year, chronic kidney disease stage, inpatient bed days in the present year, and mean hemoglobin A_1c levels. Explainability analyses using partial dependence plots demonstrated that for the best performing models, the learned patterns were consistent with real-world knowledge, thereby supporting the validity of the models.

Conclusions: We successfully developed machine learning models capable of predicting high service level utilization with strong performance and valid explainability. These models can be integrated into wider diabetes-related population health initiatives.

Background: Brief message interventions have demonstrated immense promise in health care, yet the development of these messages has suffered from a dearth of transparency and a scarcity of publicly accessible data sets. Moreover, the researcher-driven content creation process has raised resource allocation issues, necessitating a more efficient and transparent approach to content development.

Objective: This research sets out to address the challenges of content development for SMS interventions by showcasing the use of generative artificial intelligence (AI) as a tool for content creation, transparently explaining the prompt design and content generation process, and providing the largest publicly available data set of brief messages and source code for future replication of our process.

Methods: Leveraging the pretrained large language model GPT-3.5 (OpenAI), we generate a collection of messages in the context of medication adherence for individuals with type 2 diabetes using evidence-derived behavior change techniques identified in a prior systematic review. We create an attributed prompt designed to adhere to content (readability and tone) and SMS (character count and encoder type) standards while encouraging message variability to reflect differences in behavior change techniques.

Results: We deliver the most extensive repository of brief messages for a singular health care intervention and the first library of messages crafted with generative AI. In total, our method yields a data set comprising 1150 messages, with 89.91% (n=1034) meeting character length requirements and 80.7% (n=928) meeting readability requirements. Furthermore, our analysis reveals that all messages exhibit diversity comparable to an existing publicly available data set created under the same theoretical framework for a similar setting.

Conclusions: This research provides a novel approach to content creation for health care interventions using state-of-the-art generative AI tools. Future research is needed to assess the generated content for ethical, safety, and research standards, as well as to determine whether the intervention is successful in improving the target behaviors.

Infodemics pose significant dangers to public health and to the societal fabric, as the spread of misinformation can have far-reaching consequences. While artificial intelligence (AI) systems have the potential to craft compelling and valuable information campaigns with positive repercussions for public health and democracy, concerns have arisen regarding the potential use of AI systems to generate convincing disinformation. The consequences of this dual nature of AI, capable of both illuminating and obscuring the information landscape, are complex and multifaceted. We contend that the rapid integration of AI into society demands a comprehensive understanding of its ethical implications and the development of strategies to harness its potential for the greater good while mitigating harm. Thus, in this paper we explore the ethical dimensions of AI's role in information dissemination and impact on public health, arguing that potential strategies to deal with AI and disinformation encompass generating regulated and transparent data sets used to train AI models, regulating content outputs, and promoting information literacy.

Ambient scribe technology, utilizing large language models, represents an opportunity for addressing several current pain points in the delivery of primary care. We explore the evolution of ambient scribes and their current use in primary care. We discuss the suitability of primary care for ambient scribe integration, considering the varied nature of patient presentations and the emphasis on comprehensive care. We also propose the stages of maturation in the use of ambient scribes in primary care and their impact on care delivery. Finally, we call for focused research on safety, bias, patient impact, and privacy in ambient scribe technology, emphasizing the need for early training and education of health care providers in artificial intelligence and digital health tools.

Background: Women have been underrepresented in clinical trials for many years. Machine-learning models trained on clinical trial abstracts may capture and amplify biases in the data. Specifically, word embeddings are models that enable representing words as vectors and are the building block of most natural language processing systems. If word embeddings are trained on clinical trial abstracts, predictive models that use the embeddings will exhibit gender performance gaps.

Objective: We aim to capture temporal trends in clinical trials through temporal distribution matching on contextual word embeddings (specifically, BERT) and explore its effect on the bias manifested in downstream tasks.

Methods: We present TeDi-BERT, a method to harness the temporal trend of increasing women's inclusion in clinical trials to train contextual word embeddings. We implement temporal distribution matching through an adversarial classifier, trying to distinguish old from new clinical trial abstracts based on their embeddings. The temporal distribution matching acts as a form of domain adaptation from older to more recent clinical trials. We evaluate our model on 2 clinical tasks: prediction of unplanned readmission to the intensive care unit and hospital length of stay prediction. We also conduct an algorithmic analysis of the proposed method.

Results: In readmission prediction, TeDi-BERT achieved area under the receiver operating characteristic curve of 0.64 for female patients versus the baseline of 0.62 (P<.001), and 0.66 for male patients versus the baseline of 0.64 (P<.001). In the length of stay regression, TeDi-BERT achieved a mean absolute error of 4.56 (95% CI 4.44-4.68) for female patients versus 4.62 (95% CI 4.50-4.74, P<.001) and 4.54 (95% CI 4.44-4.65) for male patients versus 4.6 (95% CI 4.50-4.71, P<.001).

Conclusions: In both clinical tasks, TeDi-BERT improved performance for female patients, as expected; but it also improved performance for male patients. Our results show that accuracy for one gender does not need to be exchanged for bias reduction, but rather that good science improves clinical results for all. Contextual word embedding models trained to capture temporal trends can help mitigate the effects of bias that changes over time in the training data.

Background: Physicians spend approximately half of their time on administrative tasks, which is one of the leading causes of physician burnout and decreased work satisfaction. The implementation of natural language processing-assisted clinical documentation tools may provide a solution.

Objective: This study investigates the impact of a commercially available Dutch digital scribe system on clinical documentation efficiency and quality.

Methods: Medical students with experience in clinical practice and documentation (n=22) created a total of 430 summaries of mock consultations and recorded the time they spent on this task. The consultations were summarized using 3 methods: manual summaries, fully automated summaries, and automated summaries with manual editing. We then randomly reassigned the summaries and evaluated their quality using a modified version of the Physician Documentation Quality Instrument (PDQI-9). We compared the differences between the 3 methods in descriptive statistics, quantitative text metrics (word count and lexical diversity), the PDQI-9, Recall-Oriented Understudy for Gisting Evaluation scores, and BERTScore.

Results: The median time for manual summarization was 202 seconds against 186 seconds for editing an automatic summary. Without editing, the automatic summaries attained a poorer PDQI-9 score than manual summaries (median PDQI-9 score 25 vs 31, P<.001, ANOVA test). Automatic summaries were found to have higher word counts but lower lexical diversity than manual summaries (P<.001, independent t test). The study revealed variable impacts on PDQI-9 scores and summarization time across individuals. Generally, students viewed the digital scribe system as a potentially useful tool, noting its ease of use and time-saving potential, though some criticized the summaries for their greater length and rigid structure.

Conclusions: This study highlights the potential of digital scribes in improving clinical documentation processes by offering a first summary draft for physicians to edit, thereby reducing documentation time without compromising the quality of patient records. Furthermore, digital scribes may be more beneficial to some physicians than to others and could play a role in improving the reusability of clinical documentation. Future studies should focus on the impact and quality of such a system when used by physicians in clinical practice.

Background: A significant proportion of young at-risk patients and nonsmokers are excluded by the current guidelines for lung cancer (LC) screening, resulting in low-screening adoption. The vision of the US National Academy of Medicine to transform health systems into learning health systems (LHS) holds promise for bringing necessary structural changes to health care, thereby addressing the exclusivity and adoption issues of LC screening.

Objective: This study aims to realize the LHS vision by designing an equitable, machine learning (ML)-enabled LHS unit for LC screening. It focuses on developing an inclusive and practical LC risk prediction model, suitable for initializing the ML-enabled LHS (ML-LHS) unit. This model aims to empower primary physicians in a clinical research network, linking central hospitals and rural clinics, to routinely deliver risk-based screening for enhancing LC early detection in broader populations.

Methods: We created a standardized data set of health factors from 1397 patients with LC and 1448 control patients, all aged 30 years and older, including both smokers and nonsmokers, from a hospital's electronic medical record system. Initially, a data-centric ML approach was used to create inclusive ML models for risk prediction from all available health factors. Subsequently, a quantitative distribution of LC health factors was used in feature engineering to refine the models into a more practical model with fewer variables.

Results: The initial inclusive 250-variable XGBoost model for LC risk prediction achieved performance metrics of 0.86 recall, 0.90 precision, and 0.89 accuracy. Post feature refinement, a practical 29-variable XGBoost model was developed, displaying performance metrics of 0.80 recall, 0.82 precision, and 0.82 accuracy. This model met the criteria for initializing the ML-LHS unit for risk-based, inclusive LC screening within clinical research networks.

Conclusions: This study designed an innovative ML-LHS unit for a clinical research network, aiming to sustainably provide inclusive LC screening to all at-risk populations. It developed an inclusive and practical XGBoost model from hospital electronic medical record data, capable of initializing such an ML-LHS unit for community and rural clinics. The anticipated deployment of this ML-LHS unit is expected to significantly improve LC-screening rates and early detection among broader populations, including those typically overlooked by existing screening guidelines.