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

Liver transplantation often faces fairness challenges across subgroups defined by sensitive attributes such as age group, gender, and race/ethnicity. Machine learning models for outcome prediction can introduce additional biases. Therefore, we introduce Fairness through the Equitable Rate of Improvement in Multitask Learning (FERI) algorithm for fair predictions of graft failure risk in liver transplant patients. FERI constrains subgroup loss by balancing learning rates and preventing subgroup dominance in the training process. Our results show that FERI maintained high predictive accuracy with AUROC and AUPRC comparable to baseline models. More importantly, FERI demonstrated an ability to improve fairness without sacrificing accuracy. Specifically, for the gender, FERI reduced the demographic parity disparity by 71.74%, and for the age group, it decreased the equalized odds disparity by 40.46%. Therefore, the FERI algorithm advanced fairness-aware predictive modeling in healthcare and provides an invaluable tool for equitable healthcare systems.

Cancer outcomes are poor in resource-limited countries owing to high costs and insufficient pathologist-population ratio. The advent of digital pathology has assisted in improving cancer outcomes, however, Whole Slide Image scanners are expensive and not affordable in low-income countries. Microscope-acquired images on the other hand are cheap to collect and can be more viable for automation of cancer detection. In this study, we propose LCH-Network, a novel method to identify the cancer mitotic count from microscope-acquired images. We introduced Label Mix, and also synthesized images using GANs to handle data imbalance. Moreover, we applied progressive resolution to handle different image scales for mitotic localization. We achieved F1-Score of 0.71 and outperformed other existing techniques. Our findings enable mitotic count estimation from microscopic images with a low-cost setup. Clinically, our method could help avoid presumptive treatment without a confirmed cancer diagnosis.

Amyloid imaging has been widely used in Alzheimer's disease (AD) diagnosis and biomarker discovery through detecting the regional amyloid plaque density. It is essential to be normalized by a reference region to reduce noise and artifacts. To explore an optimal normalization strategy, we employ an automated machine learning (AutoML) pipeline, STREAMLINE, to conduct the AD diagnosis binary classification and perform permutation-based feature importance analysis with thirteen machine learning models. In this work, we perform a comparative study to evaluate the prediction performance and biomarker discovery capability of three amyloid imaging measures, including one original measure and two normalized measures using two reference regions (i.e., the whole cerebellum and the composite reference region). Our AutoML results indicate that the composite reference region normalization dataset yields a higher balanced accuracy, and identifies more AD-related regions based on the fractioned feature importance ranking.

Recently, hospitals and healthcare providers have made efforts to reduce surgical site infections as they are a major cause of surgical complications, a prominent reason for hospital readmission, and associated with significantly increased healthcare costs. Traditional surveillance methods for SSI rely on manual chart review, which can be laborious and costly. To assist the chart review process, we developed a long short-term memory (LSTM) model using structured electronic health record data to identify SSI. The top LSTM model resulted in an average precision (AP) of 0.570 [95% CI 0.567, 0.573] and area under the receiver operating characteristic curve (AUROC) of 0.905 [95% CI 0.904, 0.906] compared to the top traditional machine learning model, a random forest, which achieved 0.552 [95% CI 0.549, 0.555] AP and 0.899 [95% CI 0.898, 0.900] AUROC. Our LSTM model represents a step toward automated surveillance of SSIs, a critical component of quality improvement mechanisms.

Symptom information is primarily documented in free-text clinical notes and is not directly accessible for downstream applications. To address this challenge, information extraction approaches that can handle clinical language variation across different institutions and specialties are needed. In this paper, we present domain generalization for symptom extraction using pretraining and fine-tuning data that differs from the target domain in terms of institution and/or specialty and patient population. We extract symptom events using a transformer-based joint entity and relation extraction method. To reduce reliance on domain-specific features, we propose a domain generalization method that dynamically masks frequent symptoms words in the source domain. Additionally, we pretrain the transformer language model (LM) on task-related unlabeled texts for better representation. Our experiments indicate that masking and adaptive pretraining methods can significantly improve performance when the source domain is more distant from the target domain.

Per-/poly-fluoroalkyl substances (PFAS) are a group of manmade compounds with known human toxicity and evidence of contamination in drinking water throughout the US. We augmented our electronic health record data with geospatial information to classify PFAS exposure for our patients living in New Jersey. We explored the utility of three different methods for classifying PFAS exposure that are popularly used in the literature, resulting in different boundary types: public water supplier service area boundary, municipality, and ZIP code. We also explored the intersection of the three boundaries. To study the potential for bias, we investigated known PFAS exposure-disease associations, specifically hypertension, thyroid disease and parathyroid disease. We found that both the significance of the associations and the effect size varied by the method for classifying PFAS exposure. This has important implications in knowledge discovery and also environmental justice as across cohorts, we found a larger proportion of Black/African-American patients PFAS-exposed.

Learning about diagnostic features and related clinical information from dental radiographs is important for dental research. However, the lack of expert-annotated data and convenient search tools poses challenges. Our primary objective is to design a search tool that uses a user's query for oral-related research. The proposed framework, Contrastive LAnguage Image REtrieval Search for dental research, Dental CLAIRES, utilizes periapical radiographs and associated clinical details such as periodontal diagnosis, demographic information to retrieve the best-matched images based on the text query. We applied a contrastive representation learning method to find images described by the user's text by maximizing the similarity score of positive pairs (true pairs) and minimizing the score of negative pairs (random pairs). Our model achieved a hit@3 ratio of 96% and a Mean Reciprocal Rank (MRR) of 0.82. We also designed a graphical user interface that allows researchers to verify the model's performance with interactions.

Real-world data (RWD) like electronic health records (EHR) has great potential for secondary use by health systems and researchers. However, collected primarily for efficient health care, EHR data may not equitably represent local regions and populations, impacting the generalizability of insights learned from it. We assessed the geospatial representativeness of regions in a large health system EHR data using a spatial analysis workflow, which provides a data-driven way to quantify geospatial representation and identify adequately represented regions. We applied the workflow to investigate geospatial patterns of overweight/obesity and depression patients to find regional "hotspots" for potential targeted interventions. Our findings show the presence of geospatial bias in EHR and demonstrate the workflow to identify spatial clusters after adjusting for bias due to the geospatial representativeness. This work highlights the importance of evaluating geospatial representativeness in RWD to guide targeted deployment of limited healthcare resources and generate equitable real-world evidence.

Hormonal therapy is an important adjuvant treatment for breast cancer patients, but medication discontinuation of such therapy is not uncommon. The goal of this paper is to conduct research on the modeling of clinic communications, which have shown value in understanding medication discontinuation, to predict the discontinuation of hormonal therapy medications. Notably, we leveraged the Hypergraph Neural Network to capture the hidden connections of patients that were inferred from clinical communications. Combining the content of clinical communications as well as the demographics, insurance, and cancer stage information, our model achieved an AUC of 67.9%, which significantly outperformed other baselines such as Graph Convolutional Network (65.3%), Random Forest (62.7%), and Support Vector Machine (62.8%). Our study suggested that incorporating the hidden patient connections encoded in clinical communications into prediction models could boost their performance. Future research would consider combining structured medical records and clinical communications to better predict medication discontinuation.