AMIA Joint Summits on Translational Science proceedings. AMIA Joint Summits on Translational Science最新文献

The process of patients waiting for diagnostic examinations after an abnormal screening mammogram is inefficient and anxiety-inducing. Artificial intelligence (AI)-aided interpretation of screening mammography could reduce the number of recalls after screening. We proposed a same-day diagnostic workup to alleviate patient anxiety by employing an AI-aided interpretation to reduce unnecessary diagnostic testing after an abnormal screening mammogram. However, the potential unintended consequences of introducing this workflow in a high-volume breast imaging center are unknown. Using discrete event simulation, we observed that implementing the AI-aided screening mammogram interpretation and same-day diagnostic workflow would reduce daily patient volume by 4%, increase the time a patient would be at the clinic by 24%, and increase waiting times by 13-31%. We discuss how changing the hours of operation and introducing new imaging equipment and personnel may alleviate these negative impacts.

Accurate prediction of future clinical events such as discharge from hospital can not only improve hospital resource management but also provide an indicator of a patient's clinical condition. Within the scope of this work, we perform a comparative analysis of deep learning based fusion strategies against traditional single source models for prediction of discharge from hospital by fusing information encoded in two diverse but relevant data modalities, i.e., chest X-ray images and tabular electronic health records (EHR). We evaluate multiple fusion strategies including late, early and joint fusion in terms of their efficacy for target prediction compared to EHR-only and Image-only predictive models. Results indicated the importance of merging information from two modalities for prediction as fusion models tended to outperform single modality models and indicate that the joint fusion scheme was the most effective for target prediction. Joint fusion model merges the two modalities through a branched neural network that is jointly trained in an end-to-end fashion to extract target-relevant information from both modalities.

Using physiological data from wearable devices, the study aimed to predict exercise exertion levels by building deep learning classification and regression models. Physiological data were obtained using an unobtrusive chest-worn ECG sensor and portable pulse oximeter from healthy individuals who performed 16-minute cycling exercise sessions. During each session, real-time ECG, pulse rate, oxygen saturation, and revolutions per minute (RPM) data were collected at three intensity levels. Subjects' ratings of perceived exertion (RPE) were collected once per minute. Each 16-minute exercise session was divided into eight 2-minute windows. The self-reported RPEs, heart rate, RPMs, and oxygen saturation levels were averaged for each window to form the predictive features. In addition, heart rate variability (HRV) features were extracted from the ECG for each window. Different feature selection algorithms were used to choose top-ranked predictors. The best predictors were then used to train and test deep learning models for regression and classification analysis. Our results showed the highest accuracy and F1 score of 98.2% and 98%, respectively in training the models. For testing the models, the highest accuracy and F1 score were 80%.

Many of the existing composite social determinant of health indices, such as Area Deprivation Index, are constrained by their reliance on geographic approximations and American Community Survey data. This study builds on the body of literature around deprivation indices to construct an individual socioeconomic deprivation index (ISDI) within the NIH's All of Us Data Network by using weighted multiple correspondence analysis on SDOH data elements collected at the participant level. In this study, the correlation between ISDI and another area-approximated index is assessed to the extent possible, along with the changes in an AI models performance due to stratified sampling based on ISDI quintiles. Individual level deprivation indices may have a wide range of utility particularly in the context of precision medicine in both centralized and distributed data networks.

This study addresses rising incidence of pediatric venous thromboembolism by validating a VTE phenotype and developing a polygenic risk score (PRS) using UK Biobank data. Our findings demonstrate predictive value of the PRS, enhancing VTE risk assessment in clinical settings. Future steps involve integrating the PRS into risk stratification models.

Transgender and nonbinary (TGNB) individuals have an increased risk of certain mental health outcomes, such as depression and suicide attempts. This population skews younger in the United States and prior studies have not included TGNB patients for the entire pediatric age range in an emergency department (ED) setting. The present study aimed to examine gender identity documentation in the electronic health record and then use that information to identify and further characterize the pediatric TGNB population presenting to a psychiatric emergency service. Preliminary findings include a greater percentage of TGNB patients compared to non-TGNB individuals who had repeat visits to the ED for high acuity psychiatric concerns. A larger portion of TGNB patients also had at least one evaluation that included suicidal ideation. These results call for increased attention on the quality of mental healthcare for TGNB youth both inside and outside of the ED.

Mental health challenges are significant global public health concerns, affecting millions of people and impacting individuals, families, and communities alike. Therapists play a crucial role in supporting those with mental health issues by providing emotional, practical, and financial assistance, as well as facilitating access to treatment and services. Utilizing one-to-one interviews is an effective approach that yields valuable transcripts for further study. In this paper, we focus on interview transcripts between therapists and caregivers with family members suffering from dementia. We propose a method to efficiently handle long interview transcripts for classification. Then we employ the Shapley-value based interpretability technique to identify important contents that significantly contribute to classification results and build a corpus containing sentences potentially beneficial to the therapy. This approach offers valuable insights for enhancing the treatment of mental health issues.

Clinical imaging is an important diagnostic test to diagnose non-ischemic cardiomyopathies (NICM). However, accurate interpretation of imaging studies often requires readers to review patient histories, a time consuming and tedious task. We propose to use time-series analysis to predict the most likely NICMs using longitudinal electronic health records (EHR) as a pseudo-summary of EHR records. Time-series formatted EHR data can provide temporality information important towards accurate prediction of disease. Specifically, we leverage ICD-10 codes and various recurrent neural network architectures for predictive modeling. We trained our models on a large cohort of NICM patients who underwent cardiac magnetic resonance imaging (CMR) and a smaller cohort undergoing echocardiogram. The performance of the proposed technique achieved good micro-area under the curve (0.8357), F1 score (0.5708) and precision at 3 (0.8078) across all models for cardiac magnetic resonance imaging (CMR) but only moderate performance for transthoracic echocardiogram (TTE) of 0.6938, 0.4399 and 0.5864 respectively. We show that our model has the potential to provide accurate pre-test differential diagnosis, thereby potentially reducing clerical burden on physicians.

Class imbalance issues are prevalent in the medical field and significantly impact the performance of clinical predictive models. Traditional techniques to address this challenge aim to rebalance class proportions. They generally assume that the rebalanced proportions are derived from the original data, without considering the intricacies of the model utilized. This study challenges the prevailing assumption and introduces a new method that ties the optimal class proportions to model complexity. This approach allows for individualized tuning of class proportions for each model. Our experiments, centered on the opioid overdose prediction problem, highlight the performance gains achieved by this approach. Furthermore, rigorous regression analysis affirms the merits of the proposed theoretical framework, demonstrating a statistically significant correlation between hyperparameters controlling model complexity and the optimal class proportions.

Subpopulation models have become of increasing interest in prediction of clinical outcomes because they promise to perform better for underrepresented patient subgroups. However, the personalization benefits gained from these models tradeoff their statistical power, and can be impractical when the subpopulation's sample size is small. We hypothesize that a hierarchical model in which population information is integrated into subpopulation models would preserve the personalization benefits and offset the loss of power. In this work, we integrate ideas from ensemble modeling, personalization, and hierarchical modeling and build ensemble-based subpopulation models in which specialization relies on whole group samples. This approach significantly improves the precision of the positive class, especially for the underrepresented subgroups, with minimal cost to the recall. It consistently outperforms one model for all and one model for each subgroup approaches, especially in the presence of a high class-imbalance, for subgroups with at least 380 training samples.