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

Cancer caregivers are often informal family members who may not be prepared to adequately meet the needs of patients and often experience high stress along with significant physical, emotional, and financial burdens. Accurate prediction of caregiver's burden level is highly valuable for early intervention and support. In this study, we used several machine learning approaches to build prediction models from the National Alliance for Caregiving/AARP dataset. We performed data cleansing and imputation on the raw data to give us a working dataset of cancer caregivers. Then a series of feature selection methods were used to identify predictive risk factors for burden level. Using supervised machine learning classifiers, we achieved reasonably good prediction performance (Accuracy ∼ 0.94; AUC ∼ 0.97; F1∼ 0.93). We identify a small set of 15 features that are strong predictors of burden and can be used to build Clinical Decision Support Systems.

With the recent advances in natural language processing and deep learning, the development of tools that can assist medical coders in ICD-10 diagnosis coding and increase their efficiency in coding discharge summaries is significantly more viable than before. To that end, one important component in the development of these models is the datasets used to train them. In this study, such datasets are presented, and it is shown that one of them can be used to develop a BERT-based language model that can consistently perform well in assigning ICD-10 codes to discharge summaries written in Swedish. Most importantly, it can be used in a coding support setup where a tool can recommend potential codes to the coders. This reduces the range of potential codes to consider and, in turn, reduces the workload of the coder. Moreover, the de-identified and pseudonymised dataset is open to use for academic users.

The Pediatric Cancer Data Commons (PCDC) comprises an international community whose ironclad commitment to data sharing is combatting pediatric cancer in an unprecedented way. The byproduct of their data sharing efforts is a gold-standard consensus data model covering many types of pediatric cancer. This article describes an effort to utilize SSSOM, an emerging specification for semantically-rich data mappings, to provide a "hub and spoke" model of mappings from several common data models (CDMs) to the PCDC data model. This provides important contributions to the research community, including: 1) a clear view of the current coverage of these CDMs in the domain of pediatric oncology, and 2) a demonstration of creating standardized mappings. These mappings can allow downstream crosswalk for data transformation and enhance data sharing. This can guide those who currently create and maintain brittle ad hoc data mappings in order to utilize the growing volume of viable research data.

Translating prediction models into practice and supporting clinicians' decision-making demand demonstration of clinical value. Existing approaches to evaluating machine learning models emphasize discriminatory power, which is only a part of the medical decision problem. We propose the Applicability Area (ApAr), a decision-analytic utility-based approach to evaluating predictive models that communicate the range of prior probability and test cutoffs for which the model has positive utility; larger ApArs suggest a broader potential use of the model. We assess ApAr with simulated datasets and with three published medical datasets. ApAr adds value beyond the typical area under the receiver operating characteristic curve (AUROC) metric analysis. As an example, in the diabetes dataset, the top model by ApAr was ranked as the 23^rd best model by AUROC. Decision makers looking to adopt and implement models can leverage ApArs to assess if the local range of priors and utilities is within the respective ApArs.

Biomedical ontologies are a key component in many systems for the analysis of textual clinical data. They are employed to organize information about a certain domain relying on a hierarchy of different classes. Each class maps a concept to items in a terminology developed by domain experts. These mappings are then leveraged to organize the information extracted by Natural Language Processing (NLP) models to build knowledge graphs for inferences. The creation of these associations, however, requires extensive manual review. In this paper, we present an automated approach and repeatable framework to learn a mapping between ontology classes and terminology terms derived from vocabularies in the Unified Medical Language System (UMLS) metathesaurus. According to our evaluation, the proposed system achieves a performance close to humans and provides a substantial improvement over existing systems developed by the National Library of Medicine to assist researchers through this process.

Despite the high prevalence and burden of mental health conditions, there is a global shortage of mental health providers. Artificial Intelligence (AI) methods have been proposed as a way to address this shortage, by supporting providers with less extensive training as they deliver care. To this end, we developed the AI-Assisted Provider Platform (A2P2), a text-based virtual therapy interface that includes a response suggestion feature, which supports providers in delivering protocolized therapies empathetically. We studied providers with and without expertise in mental health treatment delivering a therapy session using the platform with (intervention) and without (control) AI-assistance features. Upon evaluation, the AI-assisted system significantly decreased response times by 29.34% (p=0.002), tripled empathic response accuracy (p=0.0001), and increased goal recommendation accuracy by 66.67% (p=0.001) across both user groups compared to the control. Both groups rated the system as having excellent usability.

The effectiveness of digital treatments can be measured by requiring patients to self-report their state through applications, however, it can be overwhelming and causes disengagement. We conduct a study to explore the impact of gamification on self-reporting. Our approach involves the creation of a system to assess cognitive load (CL) through the analysis of photoplethysmography (PPG) signals. The data from 11 participants is utilized to train a machine learning model to detect CL. Subsequently, we create two versions of surveys: a gamified and a traditional one. We estimate the CL experienced by other participants (13) while completing surveys. We find that CL detector performance can be enhanced via pre-training on stress detection tasks. For 10 out of 13 participants, a personalized CL detector can achieve an F1 score above 0.7. We find no difference between the gamified and non-gamified surveys in terms of CL but participants prefer the gamified version.

The diversity of patient information recorded on electronic medical records generally, presents a challenge for converting it into fixed-length vectors that align with clinical characteristics. To address this issue, this study aimed to utilize an unsupervised graph representation learning method to transform the unstructured inpatient information from electronic medical records into a fixed-length vector. Infograph, one of the unsupervised graph representation learning algorithms was applied to the graphed inpatient information, resulting in embedded vectors of fixed length. The embedded vectors were then evaluated for whether the clinical information was preserved in it. The results indicated that the embedded representation contained information that could predict readmission within 30 days, demonstrating the feasibility of using unsupervised graph representation learning to transform patient information into fixed-length vectors that retain clinical characteristics.

Post-acute sequelae of SARS-CoV-2 (PASC) is an increasingly recognized yet incompletely understood public health concern. Several studies have examined various ways to phenotype PASC to better characterize this heterogeneous condition. However, many gaps in PASC phenotyping research exist, including a lack of the following: 1) standardized definitions for PASC based on symptomatology; 2) generalizable and reproducible phenotyping heuristics and meta-heuristics; and 3) phenotypes based on both COVID-19 severity and symptom duration. In this study, we defined computable phenotypes (or heuristics) and meta-heuristics for PASC phenotypes based on COVID-19 severity and symptom duration. We also developed a symptom profile for PASC based on a common data standard. We identified four phenotypes based on COVID-19 severity (mild vs. moderate/severe) and duration of PASC symptoms (subacute vs. chronic). The symptoms groups with the highest frequency among phenotypes were cardiovascular and neuropsychiatric with each phenotype characterized by a different set of symptoms.

Communicating health-related probabilities to patients and the public presents challenges, although multiple studies have demonstrated that we can promote comprehension and appropriate application of numbers by matching presentation formats (e.g., percentage, bar charts, icon arrays) to communication goal (e.g., improving recall, decreasing worry, taking action). We used this literature to create goal-driven, evidence-based guidance to support health communicators in conveying probabilities. We then conducted semi-structured interviews with 39 health communicators to understand: communicators' goals for expressing probabilities, formats they choose to convey probabilities, and perceptions of prototypes of our "communicating numbers clearly" guidance. We found that communicators struggled to articulate granular goals for their communication, impeding their ability to select appropriate guidance. Future work should consider how best to support health communicators in selecting granular, differentiable goals to support broadly comprehensible information design.