DLKN-MLC: A Disease Prediction Model via Multi-Label Learning

We examine the information acquisition process regarding a patient’s status under emergency department (ED) congestion conditions. We focus on two key information channels: 1) Electronic Health Record (EHR) that provide the patient’s medical history and 2) Medical tests conducted in real-time. Whereas the EHR provides the physician with easily accessible information with little delay, real-time medical tests can provide more current information, but are time-consuming. We examine physicians’ decisions in cases of ED congestion, using a dataset that includes more than 1.4 million visits. When congestion is low, the information channels are complementary – acquiring information from the EHR is positively correlated with information acquisition from the medical tests channel, representing an incentive for the physician to acquire all possible information before providing diagnosis. However, as the congestion increases, there is less reliance on medical tests; this effect is amplified when EHR information is used. To avoid excessive congestion, physicians apparently refrain from sending patients for medical tests, and compensate for loss of information using EHR information. The impact of high system workload on the quality of medical service is an essential concern for managers; we show the indirect benefit of investment in EHRs through reduced blood-tests without increasing revisit rates.

Efficient querying for medication information in Electronic Health Record (EHR) datasets is crucial for effective patient care and clinical research. To address the complexity and data volume challenges involved in efficient medication information retrieval, we propose an ontology-driven medication query (ODMQ) optimization approach, leveraging the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). Integrating semantic ontology structures from the OMOP CDM can help enhance query accuracy and efficiency by broadening the scope of relevant medication terms like drug names, National Drug Codes, and generics, resulting in more comprehensive query outcomes than traditional methods. ODMQ significantly reduces manual search time and enhances query capabilities. We validate ODMQ's efficacy using real-world COVID-19 EHR data, demonstrating improved query performance. Through a comprehensive manual review, ODMQ ensures that expanded search terms are relevant to user inputs. It also includes an intuitive query interface and visualizes patient history for result validation and exploration.

Electronic health records (EHR) capture “real-world” disease and care processes and hence offer richer and more generalizable data for comparative effectiveness research than traditional randomized clinical trial studies. With the increasingly broadening adoption of EHR worldwide, there is a growing need to widen the use of EHR data to support clinical research. A big barrier to this goal is that much of the information in EHR is still narrative. This chapter describes the foundation of biomedical language processing and explains how traditional machine learning and the state-of-the-art deep learning techniques can be employed in the context of extracting and transforming narrative information in EHR to support clinical research.

BackgroundWith the population aging, multiple chronic diseases, depressive symptoms, and stroke are increasingly common among middle-aged and elderly adults worldwide. This study aimed to explore the independent associations of multiple chronic diseases and depressive symptoms as well as their combination with incident stroke in a prospective cohort of Chinese middle-aged and elderly adults, and to sensitively estimate the association between each type of chronic disease and incident stroke.MethodsThis study used data from the China Health and Retirement Longitudinal Study (CHARLS). A total of 8389 participants meeting the inclusion criteria at baseline (between 2011 and 2012) survey were included, and 7108 eligible participants completed the follow-up survey over 8 years (Wave 4, in 2018). Questionnaire information, physical examination, and clinical and biochemical measurements were collected.ResultsThe mean (SD) age at baseline was 58.5 (± 9.1) years. Multiple chronic disease and depressive symptoms were independently associated with incident stroke. After adjusting for control variables, patients having 1 type of chronic disease and depressive symptoms were at 1.943 (95% CI = 1.166–3.238) times higher risk of incident stroke than those without chronic disease and depressive symptoms, and patients having at least 2 types of chronic diseases and depressive symptoms were at 3.000 (95% CI = 1.846–4.877) times higher risk of incident stroke; the magnitudes of the associations increased by the numbers of having chronic diseases and depressive symptoms. Sensitivity analyses incorporating all five types of chronic disease (i.e., hypertension, dyslipidemia, heart disease, diabetes, and chronic kidney disease) showed that the magnitude of the associations between hypertension and incident stroke was most significant.ConclusionsWe identified significant independent and combined longitudinal associations of multiple chronic diseases and depressive symptoms with incident stroke, and the combined associations reflected a dose–response relationship. The association between hypertension and incident stroke was strongest among the five chronic diseases.

Causal inference aims to estimate the causal relationships and effect sizes among treatments and outcomes. Electronic health record (EHR) data is a valuable healthcare data source that can support causal inference. However, a large percentage of the data is missing in EHRs and they are missing not at random (MNAR). Ignoring MNAR can lead to severe biases, to the extent where the causal structure underlying the data gets distorted. We proposed a new causal inference methodology that addresses the MNAR problem and thus helps preserve the causal structure. We compared the performance of our proposed method with the traditional causal inference method, structural equation modeling (SEM). We evaluated these methods for their accuracy in estimating the causal effect sizes and their ability to converge at all. We employed both simulation studies and real-world EHR data sets. We demonstrated that imputation under the improper missingness mechanism distorted the causal structure to a degree where SEM found it incompatible with the data and failed to. converge. Even when 20 to 30 % of the values were missing, SEM failed to converge in as many as 50% of the runs. The proposed causal inference method achieved a higher convergence rate and more accurate estimation of latent treatment effects both on the synthetic data and on a real EHR data set. We proposed a new methodology that incorporates the knowledge of missing data mechanisms. It significantly mitigated the biases associated with MNAR in the EHR dataset and substantially outperformed SEM that uses the improper missing data mechanism.

AB1206 SELF-REPORT OF FRACTURE HISTORY COMPARED TO FRACTURE CODES FROM AN ELECTRONIC HEALTH RECORD DATASET

Mental health disorders represent a significant global health burden, often characterized by late diagnoses and inconsistent treatment outcomes. The integration of artificial intelligence (AI) with electronic health records (EHRs) offers a transformative approach to early diagnosis and intervention. This study presents the development of an AI-powered predictive model designed to enhance the diagnosis of mental health disorders by leveraging structured and unstructured data from EHRs. The model integrates natural language processing (NLP), machine learning (ML), and deep learning techniques to analyze clinical notes, patient histories, demographic data, and behavioral indicators. Our methodology involves preprocessing EHR datasets, extracting relevant features, and applying advanced algorithms such as random forests, support vector machines (SVM), and neural networks to build robust predictive models. Additionally, NLP techniques are used to process narrative text data, identifying critical indicators of mental health conditions such as depression, anxiety, bipolar disorder, and schizophrenia. The model is trained and validated using a large, anonymized EHR dataset, ensuring a high level of accuracy, precision, and recall in identifying at-risk individuals. Preliminary results demonstrate the model’s ability to outperform traditional diagnostic methods by identifying subtle patterns and risk factors often overlooked in standard clinical evaluations. Moreover, the integration of interpretability tools such as SHAP (SHapley Additive exPlanations) enables clinicians to understand the rationale behind each prediction, promoting trust and clinical applicability. This research underscores the potential of AI-driven tools in revolutionizing mental healthcare by enabling timely diagnosis, personalized treatment planning, and improved patient outcomes. However, the study also highlights challenges such as data quality, privacy concerns, algorithmic bias, and the need for interdisciplinary collaboration in deploying these systems responsibly. In conclusion, the proposed AI-powered predictive model demonstrates promise in augmenting mental health diagnostics using EHR data, paving the way for scalable and proactive mental healthcare delivery.

Patient and public involvement and engagement (PPIE) is integral to health research. Reporting of PPIE methods and impact is becoming increasingly common in health research. However, reporting on PPIE in studies using large, routinely collected electronic health record data sets is less common. Anecdotal evidence suggests that involvement in this research context is more challenging and offers fewer opportunities for meaningful influence on the research process. This paper reports the involvement approach for a Clinical Practice Research Datalink (CPRD) study and critically reflects on the process and impact of involving young people, parents and carers in research using this UK primary care electronic health record data set. The CHOOSE study investigated mental health diagnoses of children and young people (1-24 years) during the COVID-19 pandemic using the CPRD. The study was informed by a Lived Experience Advisory Panel (LEAP) which consisted of 13 members including 8 young people (13-25 years) with lived experience of mental health difficulties and 5 parents/carers, with involvement activities facilitated by project partners, mental health research charity, The McPinR Foundation. We reflect on this process in this manuscript. Key benefits of involving people with lived experience in this research included making sense of and contextualising findings and ensuring that they were focused on making a difference to young people's lives. Challenges included the fixed nature of the CPRD data, which did not capture all the information people with lived experience perceived to be important. Researchers expressed limited time for PPIE activities although that was compensated by McPin colleagues who organised and facilitated online meetings, and supported the young people, parents and carers during and between meetings. This paper describes an approach to patient and public involvement in an electronic health record database study. Working collaboratively with young people, carers and other stakeholders requires sufficient time and adequate resources. We also highlight the importance of appropriate training and support and being transparent about the limitations of PPIE involvement. Three members of the CHOOSE LEAP have been involved in conceptualising and writing this paper.

The automated processing of Electronic Health Records (EHRs) poses a significant challenge due to their unstructured nature, rich in valuable, yet disorganized information. Natural Language Processing (NLP), particularly Named Entity Recognition (NER), has been instrumental in extracting structured information from EHR data. However, existing literature primarly focuses on extracting handcrafted clinical features through NLP and NER methods without delving into their learned representations. In this work, we explore the untapped potential of these representations by considering their contextual richness and entity-specific information. Our proposed methodology extracts representations generated by a transformer-based NER model on EHRs data, combines them using a hierarchical attention mechanism, and employs the obtained enriched representation as input for a clinical prediction model. Specifically, this study addresses Overall Survival (OS) in Non-Small Cell Lung Cancer (NSCLC) using unstructured EHRs data collected from an Italian clinical centre encompassing 838 records from 231 lung cancer patients. Whilst our study is applied on EHRs written in Italian, it serves as use case to prove the effectiveness of extracting and employing high level textual representations that capture relevant information as named entities. Our methodology is interpretable because the hierarchical attention mechanism highlights the information in EHRs that the model considers the most crucial during the decision-making process. We validated this interpretability by measuring the agreement of domain experts on the importance assigned by the hierarchical attention mechanism to EHRs information through a questionnaire. Results demonstrate the effectiveness of our method, showcasing statistically significant improvements over traditional manually extracted clinical features.

We integrate literature- and data-driven task analysis methods to derive an initial task taxonomy for electronic health record (EHR) and electronic medical record (EMR) data analysis. An EHR (EMR) is a digital and longitudinal version of a patients health(medical) information and may include all key clinical events relevant to that persons health (medical) history, such as provider, demographics, progress notes, medicine, diagnosis, etc. Our goal is to arrive a task taxonomy for analyzing EHR (EMR) datasets because tasks play an important role in the design and evaluation of visualization techniques. Our method has three stages: data collection, task modelling, and task taxonomy summary. In data collection, we first survey related literature from the past two decades and extract typical tasks and corresponding data by extracting goals and scenarios of the particular work. We introduce multiple continuous relations to describe specific binary or multiple continuous relation-seeking tasks. Finally, we arrive an initial set of task types for EHR/EMR analysis that guide the design and evaluation of visualization techniques.

Deep Phenotyping on Electronic Health Records Facilitates Genetic Diagnosis by Clinical Exomes

Many efforts have proved that deep learning models are effective for disease prediction using electronic health records (EHRs). However, these models are not yet precise enough to predict diseases. Additionally, ethical concerns and the use of clustering and classification algorithms on small datasets limit their effectiveness. The complexity of data processing further complicates the interpretation of patient representation learning models, even though data augmentation strategies may help. Incomplete patient data also hinder model accuracy. This study aims to develop and evaluate a deep learning model that addresses these challenges. Our proposed approach is to design a disease prediction model based on deep Q-learning (DQL), which replaces the traditional Q-learning reinforcement learning algorithm with a neural network deep learning model, and the mapping capabilities of the Q-network are utilized. We conclude that the proposed model achieves the best accuracy (98%) compared with other models.

ABSTRACTPeople’s privacy concerns about electronic health records (EHRs) have been widely discussed. Given that in principle, patients have the right to control their information in EHRs, the system will not reach its full potential without their support. Although human beings are generally inclined toward privacy, contextual differences play a role in individual decisions to disclose personal information. Likewise, patients exhibit different responses in terms of sharing their health information in diverse scenarios. Empirical work on patients’ attitudes towards the secondary use of their health information is scarce. This study aims to investigate individuals’ willingness to share their health information based on anonymity, information type (partial vs. whole), and the type of information user (health professionals, health researchers, health-related governmental agencies, and other governmental agencies). Furthermore, this study attempts to examine the effects of interaction between the three factors. A survey was conducted in South Korea, and the data obtained were analyzed by ANOVA. Despite the recent rapid diffusion of EHRs in South Korea, there is little discussion of patients’ privacy in society. The results show that, although anonymity and the user type have a significant effect on willingness to share health information, the information type has no significant effect. The results also indicate that the willingness to share was higher for health-related governmental agencies and health professionals than for other governmental agencies. The findings reveal that, although willingness to share anonymous information does not vary significantly, the willingness differs for identified information. The study contributes to research on patients’ privacy behavior by analyzing their comprehensive responses to health information-sharing investigations.

When using electronic health records (EHRs) for clinical and translational research, additional data is often available from external sources to enrich the information extracted from EHRs. For example, academic biobanks have more granular data available, and patient reported data is often collected through small-scale surveys. It is common that the external data is available only for a small subset of patients who have EHR information. We propose efficient and robust methods for building and evaluating models for predicting the risk of binary outcomes using such integrated EHR data. Our method is built upon an idea derived from the two-phase design literature that modeling the availability of a patient's external data as a function of an EHR-based preliminary predictive score leads to effective utilization of the EHR data. Through both theoretical and simulation studies, we show that our method has high efficiency for estimating log-odds ratio parameters, the area under the ROC curve, as well as other measures for quantifying predictive accuracy. We apply our method to develop a model for predicting the short-term mortality risk of oncology patients, where the data was extracted from the University of Pennsylvania hospital system EHR and combined with survey-based patient reported outcome data.

Introduction: Electronic health records (EHRs) represent an opportunity for real-time early risk prediction. EHR records were used to determine the extent to which machine learning tools and a purely data driven approach to modeling (DDM) could detect heart failure subtypes, i.e., preserved ejection fraction (HFpEF) and reduced ejection fraction (HFrEF), 12 months before a clinical diagnosis. Methods: Incident HF cases were identified from Geisinger Clinic primary care patients age 50-85, diagnosed between 2001 and 2010 and further defined as HFrEF if LVEF≤40 and HFpEF if LVEF>50. Controls were chosen to match HF cases by age, gender, location and primary care physician. EHR data were extracted on demographics, ICD-9 codes, medication orders, clinical and behavioral measures. Modeling was completed to detect HF using data from a 24-month observation window 12 months before the HF diagnosis. Patient feature vectors were generated from the data and summarized by one or more aggregation functions (e.g., counts, means). For the HF endpoint, modeling was done with and without patients who had an acute coronary syndrome (ACS) event within 12 months of the diagnosis. Regularized logistic regression was applied using information gain feature selection and 10-fold cross validation. Model performance was assessed by the area under the ROC curve (AUC) and complexity by the number of selected features. Results: Performance for HFpEF is better than for HFrEF. The HFpEF model is more complex than the HFrEF model as indicated by more EHR information that was needed to discriminate the HFpEF cases from controls. Performance with and without ACS cases is similar though models including ACS cases are more complex than models excluding them (Table). Conclusions: Purely data driven approaches to modeling can be used to detect HF 12 months before clinical diagnosis. Model performance and complexity varies with the HF subtypes indicating differences in the complexity of modeling the HF subtypes.