AMIA ... Annual Symposium proceedings. AMIA Symposium最新文献

Randomized Clinical Trials (RCTs) measure an intervention's efficacy, but they may not be generalizable to a desired target population if the RCT is not equitable. Thus, representativeness of RCTs has become a national priority. Synthetic Controls (SCs) that incorporate observational data into RCTs have shown great potential to produce more efficient studies, but their equity is rarely considered. Here, we examine how to improve treatment effect estimation and equity of a trial by augmenting "on-trial" concurrent controls with SCs to form a Hybrid Control Arm (HCA). We introduce FRESCA - a framework to evaluate HCA construction methods using RCT simulations. FRESCA shows that doing propensity and equity adjustment when constructing the HCA leads to accurate population treatment effect estimates while meeting equity goals with potentially less "on-trial" patients. This work represents the first investigation of equity in HCA design that provides definitions, metrics, compelling questions, and resources for future work.

Enhancing diversity and inclusion in clinical trial recruitment, especially for historically marginalized populations including Black, Indigenous, and People of Color individuals, is essential. This practice ensures that generalizable trial results are achieved to deliver safe, effective, and equitable health and healthcare. However, recruitment is limited by two inextricably linked barriers - the inability to recruit and retain enough trial participants, and the lack of diversity amongst trial populations whereby racial and ethnic groups are underrepresented when compared to national composition. To overcome these barriers, this study describes and evaluates a framework that combines 1) probabilistic and machine learning models to accurately impute missing race and ethnicity fields in real-world data including medical and pharmacy claims for the identification of eligible trial participants, 2) randomized controlled trial experimentation to deliver an optimal patient outreach strategy, and 3) stratified sampling techniques to effectively balance cohorts to continuously improve engagement and recruitment metrics.

Breast cancer is the second leading cause of cancer death for women in the United States. While breast cancer screening participation is the most effective method for early detection, screening rate has remained low. Given that understanding health perception is critical to understand health decisions, our study utilized the Health Belief Model-based deep learning method to predict and examine public health beliefs in breast cancer and its screening behavior. The results showed that the trends in public health perception are sensitive to political (i.e., changes in health policy), sociological (i.e., representation of disease and its preventive care by public figure or organization), psychological (i.e., social support), and environmental factors (i.e., COVID-19 pandemic). Our study explores the roles social media can play in public health surveillance and in public health promotion of preventive care.

Physicians spend a large amount of time with the electronic health record (EHR), which the majority believe contributes to their burnout. However, there are limitedstandardized measures of physician EHR time. Vendor-derived metrics are standardized but may underestimate real-world EHR experience. Investigator-derived metrics may be more reliable but not standardized, particularly with regard to timeout thresholds defining inactivity. This study aimed to enable standardized investigator-derived metrics using conversion factors between raw event log-derived metrics and Signal (Epic System's standardized metric) for primary care physicians. This was an observational, retrospective longitudinal study of EHR raw event logs and Signal data from a quaternary academic medical center and its community affiliates in California, over a 6-month period. The study evaluated 242 physicians over 1370 physician-months, comparing 53.7 million event logs to 6850 Signal metrics, in five different time based metrics. Results show that inactivity thresholds for event log metric derivation that most closely approximate Signal metrics ranged from 90 seconds (Visit Navigator) to 360 seconds ("Pajama time") depending on the metric. Based on this data, conversion factors for investigator-derived metrics across a wide range of inactivity thresholds, via comparison with Signal metrics, are provided which may allow researchers to consistently quantify EHR experience.

Viewing laboratory test results is patients' most frequent activity when accessing patient portals, but lab results can be very confusing for patients. Previous research has explored various ways to present lab results, but few have attempted to provide tailored information support based on individual patient's medical context. In this study, we collected and annotated interpretations of textual lab result in 251 health articles about laboratory tests from AHealthyMe.com. Then we evaluated transformer-based language models including BioBERT, ClinicalBERT, RoBERTa, and PubMedBERT for recognizing key terms and their types. Using BioPortal's term search API, we mapped the annotated terms to concepts in major controlled terminologies. Results showed that PubMedBERT achieved the best F1 on both strict and lenient matching criteria. SNOMED CT had the best coverage of the terms, followed by LOINC and ICD-10-CM. This work lays the foundation for enhancing the presentation of lab results in patient portals by providing patients with contextualized interpretations of their lab results and individualized question prompts that they can, in turn, refer to during physician consults.

Complete and accurate race and ethnicity (RE) patient information is important for many areas of biomedical informatics research, such as defining and characterizing cohorts, performing quality assessments, and identifying health inequities. Patient-level RE data is often inaccurate or missing in structured sources, but can be supplemented through clinical notes and natural language processing (NLP). While NLP has made many improvements in recent years with large language models, bias remains an often-unaddressed concern, with research showing that harmful and negative language is more often used for certain racial/ethnic groups than others. We present an approach to audit the learned associations of models trained to identify RE information in clinical text by measuring the concordance between model-derived salient features and manually identified RE-related spans of text. We show that while models perform well on the surface, there exist concerning learned associations and potential for future harms from RE-identification models if left unaddressed.

Professional medical publications writers (PMWs) cover a wide range of biomedical writing activities that recently includes translation of biomedical publications to plain language summaries (PLS). The consumer health informatics literature (CHI) consistently describes the importance of incorporating health literacy principles in any natural language processing (NLP) app designed to communicate medical information to lay audiences, particularly patients. In this stepwise systematic review, we searched PubMed indexed literature for CHI NLP-based apps that have the potential to assist PMWs in developing text based PLS. Results showed that available apps are limited to patient portals and other technologies used to communicate medical text and reports from electronic health records. PMWs can apply the lessons learned from CHI NLP-based apps to supervise development of tools specific to text simplification and summarization for PLS from biomedical publications.

Obtaining reliable data on patient mortality is a critical challenge facing observational researchers seeking to conduct studies using real-world data. As these analyses are conducted more broadly using newly-available sources of real-world evidence, missing data can serve as a rate-limiting factor. We conducted a comparison of mortality data sources from different stakeholder perspectives - academic medical center (AMC) informatics service providers, AMC research coordinators, industry analytics professionals, and academics - to understand the strengths and limitations of differing mortality data sources: locally generated data from sites conducting research, data provided by governmental sources, and commercially available data sets. Researchers seeking to conduct observational studies using extant data should consider these factors in sourcing outcomes data for their populations of interest.

Prior work has shown that analyzing the use of first-person singular pronouns can provide insight into individuals' mental status, especially depression symptom severity. These findings were generated by counting frequencies of first-person singular pronouns in text data. However, counting doesn't capture how these pronouns are used. Recent advances in neural language modeling have leveraged methods generating contextual embeddings. In this study, we sought to utilize the embeddings of first-person pronouns obtained from contextualized language representation models to capture ways these pronouns are used, to analyze mental status. De-identified text messages sent during online psychotherapy with weekly assessment of depression severity were used for evaluation. Results indicate the advantage of contextualized first-person pronoun embeddings over standard classification token embeddings and frequency-based pronoun analysis results in predicting depression symptom severity. This suggests contextual representations of first-person pronouns can enhance the predictive utility of language used by people with depression symptoms.

This study aims to develop machine learning (ML) algorithms to predict exercise exertion levels using physiological parameters collected from wearable devices. Real-time ECG, oxygen saturation, pulse rate, and revolutions per minute (RPM) data were collected at three intensity levels during a 16-minute cycling exercise. Parallel to this, throughout each exercise session, the study subjects' ratings of perceived exertion (RPE) were gathered once per minute. Each 16-minute exercise session was divided into a total of eight 2-minute windows. Each exercise window was labeled as "high exertion," or "low exertion" classes based on the self-reported RPEs. For each window, the gathered ECG data were used to derive the heart rate variability (HRV) features in the temporal and frequency domains. Additionally, each window's averaged RPMs, heart rate, and oxygen saturation levels were calculated to form all the predictive features. The minimum redundancy maximum relevance algorithm was used to choose the best predictive features. Top selected features were then used to assess the accuracy of ten ML classifiers to predict the next window's exertion level. The k-nearest neighbors (KNN) model showed the highest accuracy of 85.7% and the highest F1 score of 83%. An ensemble model showed the highest area under the curve (AUC) of 0.92. The suggested method can be used to automatically track perceived exercise exertion in real-time.