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

Accelerated use of telemedicine during the COVID-19 pandemic enabled uninterrupted healthcare delivery while unmasking care disparities for several vulnerable communities. The social determinants of health (SDOH) serve as a critical model for understanding how the circumstances in which people are born, work, and live impact health outcomes. We performed semi-structured interviews to understand patients and providers' experiences with telemedicine encounters during the COVID-19 pandemic. Through a deductive approach, we applied the SDOH to determine telemedicine's role and impact within this framework. Overall, patient and provider interviews supported the use of existing SDOH domains to describe disparities in Internet access and telemedicine use, rather than reframing technology as a sixth SDOH. In order to mitigate the digital divide, we identify and propose solutions that address SDOH-related barriers that shape the use of health information technologies.

The Electronic Health Record (EHR) contains information about social determinants of health (SDoH) such as homelessness. Much of this information is contained in clinical notes and can be extracted using natural language processing (NLP). This data can provide valuable information for researchers and policymakers studying long-term housing outcomes for individuals with a history of homelessness. However, studying homelessness longitudinally in the EHR is challenging due to irregular observation times. In this work, we applied an NLP system to extract housing status for a cohort of patients in the US Department of Veterans Affairs (VA) over a three-year period. We then applied inverse intensity weighting to adjust for the irregularity of observations, which was used generalized estimating equations to estimate the probability of unstable housing each day after entering a VA housing assistance program. Our methods generate unique insights into the long-term outcomes of individuals with a history of homelessness and demonstrate the potential for using EHR data for research and policymaking.

For patients with thyroid nodules, the ability to detect and diagnose a malignant nodule is the key to creating an appropriate treatment plan. However, assessments of ultrasound images do not accurately represent malignancy, and often require a biopsy to confirm the diagnosis. Deep learning techniques can classify thyroid nodules from ultrasound images, but current methods depend on manually annotated nodule segmentations. Furthermore, the heterogeneity in the level of magnification across ultrasound images presents a significant obstacle to existing methods. We developed a multi-scale, attention-based multiple-instance learning model which fuses both global and local features of different ultrasound frames to achieve patient-level malignancy classification. Our model demonstrates improved performance with an AUROC of 0.785 (p<0.05) and AUPRC of 0.539, significantly surpassing the baseline model trained on clinical features with an AUROC of 0.667 and AUPRC of 0.444. Improved classification performance better triages the need for biopsy.

Data Augmentation is a crucial tool in the Machine Learning (ML) toolbox because it can extract novel, useful training images from an existing dataset, thereby improving accuracy and reducing overfitting in a Deep Neural Network (DNNs). However, clinical dermatology images often contain irrelevant background information,such as furniture and objects in the frame. DNNs make use of that information when optimizing the loss function. Data augmentation methods that preserve this information risk creating biases in the DNN's understanding (for example, that objects in a particular doctor's office are a clue that the patient has cutaneous T-cell lymphoma). Creating a supervised foreground/background segmentation algorithm for clinical dermatology images that removes this irrelevant information would be prohibitively expensive due to labeling costs. To that end, we propose a novel unsupervised DNN that dynamically masks out image information based on a combination of a differentiable adaptation of Otsu's Method and CutOut augmentation. SoftOtsuNet augmentation outperforms all other evaluated augmentation methods on the Fitzpatrick17k dataset (0.75% improvement), Diverse Dermatology Images dataset (1.76% improvement), and our proprietary dataset (0.92% improvement). SoftOtsuNet is only required at training time, meaning inference costs are unchanged from the baseline. This further suggests that even large data-driven models can still benefit from human-engineered unsupervised loss functions.

Self-report is purported to be the gold standard for collecting demographic information. Many entry forms include a free-text "write-in" option in addition to structured responses. Balancing the flexibility of free-text with the value of collecting data in a structured format is a challenge if the data are to be useful for measuring and mitigating health disparities. While much work has been done to improve collection of race and ethnicity information, how to best collect data related to sexual and gender minority status and military veteran status has been less commonly studied. We analyzed 3,381 patient-provided free-text responses collected via a patient portal for gender identity, sexual orientation, pronouns, and veteran experiences. We identified common responses to better understand our patient population and help improve future iterations of data collection tools.

Ensemble learning is a powerful technique for improving the accuracy and reliability of prediction models, especially in scenarios where individual models may not perform well. However, combining models with varying accuracies may not always improve the final prediction results, as models with lower accuracies may obscure the results of models with higher accuracies. This paper addresses this issue and answers the question of when an ensemble approach outperforms individual models for prediction. As a result, we propose an ensemble model for predicting patients at risk of postoperative prolonged opioid. The model incorporates two machine learning models that are trained using different covariates, resulting in high precision and recall. Our study, which employs five different machine learning algorithms, shows that the proposed approach significantly improves the final prediction results in terms of AUROC and AUPRC.

As the population of older adults grows at an unprecedented rate, there is a large gap to provide culturally tailored end-of-life care. This study describes a payor-led, informatics-based approach to identify Medicare members who may benefit from a Compassionate Care^SM Program (CCP), which was designed to provide specialized care management services and support to members who have end-stage and/or life-limiting illnesses by addressing the quintuple aim. Potential participants are identified through machine learning models whereby nurse care managers then provide tailored outreach via telephone. A retrospective, observational cohort analysis of propensity-weighted Medicare members was performed to compare decedents who did or did not participate in the CCP. This program enhanced the end-of-life care experience while providing equitable outcomes regardless of age, gender, and geography and decreased inpatient (-37%) admissions with concomitant reduced (-59%) medical spend when compared to decedents that did not utilize the end-of-life care management program.

Mental health disorders remain a significant challenge in modern healthcare, with diagnosis and treatment often relying on subjective patient descriptions and past medical history. To address this issue, we propose a personalized mental health tracking and mood prediction system that utilizes patient physiological data collected through personal health devices. Our system leverages a decentralized learning mechanism that combines transfer and federated machine learning concepts using smart contracts, allowing data to remain on users' devices and enabling effective tracking of mental health conditions for psychiatric treatment and management in a privacy-aware and accountable manner. We evaluated our model using a popular mental health dataset, which yielded promising results. By utilizing connected health systems and machine learning models, our approach offers a novel solution to the challenge of providing psychiatrists with further insight into their patients' mental health outside of traditional office visits.

Widespread adoption of electronic health records (EHR) in the U.S. has been followed by unintended consequences, overexposing clinicians to widely reported EHR limitations. As an attempt to fixing the EHR, we propose the use of a clinical context ontology (CCO), applied to turn implicit contextual statements into formally represented data in the form of concept-relationship-concept tuples. These tuples form what we call a patient specific knowledge base (PSKB), a collection of formally defined tuples containing facts about the patient's care context. We report the process to create a CCO, which guides annotation of structured and narrative patient data to produce a PSKB. We also present an application of our PSKB using real patient data displayed on a semantically oriented patient summary to improve EHR navigation. Our approach can potentially save precious time spent by clinicians using today's EHRs, by showing a chronological view of the patient's record along with contextual statements needed for care decisions with minimum effort. We propose several other applications of a PSKB to improve multiple EHR functions to guide future research.

Venous Thromboembolism (VTE) is a serious, preventable public health problem that requires timely treatment. Because signs and symptoms are non-specific, patients often present to primary care providers with VTE symptoms prior to diagnosis. Today there are no federal measurement tools in place to track delayed diagnosis of VTE. We developed and tested an electronic clinical quality measure (eCQM) to quantify Diagnostic Delay of Venous Thromboembolism (DOVE); the rate of avoidable delayed VTE events occurring in patients with a VTE who had reported VTE symptoms in primary care within 30 days of diagnosis. DOVE uses routinely collected EHR data without contributing to documentation burden. DOVE was tested in two geographically distant healthcare systems. Overall DOVE rates were 72.60% (site 1) and 77.14% (site 2). This novel, data-driven eCQM could inform healthcare providers and facilities about opportunities to improve care, strengthen incentives for quality improvement, and ultimately improve patient safety.