Electronic Health Record Information Research Articles

Natural language processing to identify suicidal ideation and anhedonia in major depressive disorder.

Anhedonia and suicidal ideation are symptoms of major depressive disorder (MDD) that are not regularly captured in structured scales but may be captured in unstructured clinical notes. Natural language processing (NLP) techniques may be used to extract longitudinal data on suicidal behaviors and anhedonia within unstructured clinical notes. This study assessed the accuracy of using NLP techniques on electronic health records (EHRs) to identify these symptoms among patients with MDD. EHR-derived, de-identified data were used from the NeuroBlu Database (version 23R1), a longitudinal behavioral health real-world database. Mental health clinicians annotated instances of anhedonia and suicidal symptoms in clinical notes creating a ground truth. Interrater reliability (IRR) was calculated using Krippendorff's alpha. A novel transformer architecture-based NLP model was trained on clinical notes to recognize linguistic patterns and contextual cues. Each sentence was categorized into one of four labels: (1) anhedonia; (2) suicidal ideation without intent or plan; (3) suicidal ideation with intent or plan; (4) absence of suicidal ideation or anhedonia. The model was assessed using positive predictive values (PPV), negative predictive values, sensitivity, specificity, F1-score, and AUROC. The model was trained, tested, and validated on 2,198, 1,247, and 1,016 distinct clinical notes, respectively. IRR was 0.80. For anhedonia, suicidal ideation with intent or plan, and suicidal ideation without intent or plan the model achieved a PPV of 0.98, 0.93, and 0.87, an F1-score of 0.98, 0.91, and 0.89 during training and a PPV of 0.99, 0.95, and 0.87 and F1-score of 0.99, 0.95, and 0.89 during validation. NLP techniques can leverage contextual information in EHRs to identify anhedonia and suicidal symptoms in patients with MDD. Integrating structured and unstructured data offers a comprehensive view of MDD's trajectory, helping healthcare providers deliver timely, effective interventions. Addressing current limitations will further enhance NLP models, enabling more accurate extraction of critical clinical features and supporting personalized, proactive mental health care.

Toward a Computable Phenotype for Determining Eligibility of Lung Cancer Screening Using Electronic Health Records.

Lung cancer screening (LCS) has the potential to reduce mortality and detect lung cancer at its early stages, but the high false-positive rate associated with low-dose computed tomography (LDCT) for LCS acts as a barrier to its widespread adoption. This study aims to develop computable phenotype (CP) algorithms on the basis of electronic health records (EHRs) to identify individual's eligibility for LCS, thereby enhancing LCS utilization in real-world settings. The study cohort included 5,778 individuals who underwent LDCT for LCS from 2012 to 2022, as recorded in the University of Florida Health Integrated Data Repository. CP rules derived from LCS guidelines were used to identify potential candidates, incorporating both structured EHR and clinical notes analyzed via natural language processing. We then conducted manual reviews of 453 randomly selected charts to refine and validate these rules, assessing CP performance using metrics, for example, F1 score, specificity, and sensitivity. We developed an optimal CP rule that integrates both structured and unstructured data, adhering to the US Preventive Services Task Force 2013 and 2020 guidelines. This rule focuses on age (55-80 years for 2013 and 50-80 years for 2020), smoking status (current, former, and others), and pack-years (≥30 for 2013 and ≥20 for 2020), achieving F1 scores of 0.75 and 0.84 for the respective guidelines. Including unstructured data improved the F1 score performance by up to 9.2% for 2013 and 12.9% for 2020, compared with using structured data alone. Our findings underscore the critical need for improved documentation of smoking information in EHRs, demonstrate the value of artificial intelligence techniques in enhancing CP performance, and confirm the effectiveness of EHR-based CP in identifying LCS-eligible individuals. This supports its potential to aid clinical decision making and optimize patient care.

The association of distress and depression screening measures and other electronic health record information with adjuvant endocrine therapy persistence.

Few risk factors for early adjuvant endocrine discontinuation have been identified, but clinical trials suggest pre-AET symptom burden might be important. We sought to assess this in an academic practice. We examined baseline and up to five years of follow-up information for postmenopausal women with stage I-III hormone-receptor positive breast cancer 2014-2019 receiving oncologist prescriptions for AET. The Distress Thermometer (DT) and its problem list, the Patient Health Questionnaire 2/9 (PHQ 2/9), cancer extent of disease and treatments, comorbidities, sociodemographics, and pharmacy prescription fill dates were abstracted from the cancer registry and electronic health record (EHR). The association of these variables with early AET prescription fill discontinuation (prior to 5years) was examined using survival analysis and Bayesian machine learning, with censoring for recurrence, death, or provider change. Among the cohort of 961 women (mean 68.8years, SD 2.88), 91.6% were white, 74.6% had Stage I disease, and 45.0% a pre-AET DT showing high distress (> 3). The median follow-up time was 820 (25, 75% 448,1282) days, and 29.6% discontinued early. Neither the DT score nor the PHQ 2/9 was associated with nonadherence, although three physical problems were modestly associated. Over 25% of women who stopped filling prescriptions did not have their prescriptions discontinued in the EHR. Several commonly available baseline EHR variables were not associated with early discontinuation, although some symptoms may have modest effects. Many women who discontinued still had EHR prescriptions, suggesting that physicians could use prescription fill information to intervene earlier.

Comparing the performance of screening surveys versus predictive models in identifying patients in need of health-related social need services in the emergency department.

Health-related social needs (HRSNs), such as housing instability, food insecurity, and financial strain, are increasingly prevalent among patients. Healthcare organizations must first correctly identify patients with HRSNs to refer them to appropriate services or offer resources to address their HRSNs. Yet, current identification methods are suboptimal, inconsistently applied, and cost prohibitive. Machine learning (ML) predictive modeling applied to existing data sources may be a solution to systematically and effectively identify patients with HRSNs. The performance of ML predictive models using data from electronic health records (EHRs) and other sources has not been compared to other methods of identifying patients needing HRSN services. A screening questionnaire that included housing instability, food insecurity, transportation barriers, legal issues, and financial strain was administered to adult ED patients at a large safety-net hospital in the mid-Western United States (n = 1,101). We identified those patients likely in need of HRSN-related services within the next 30 days using positive indications from referrals, encounters, scheduling data, orders, or clinical notes. We built an XGBoost classification algorithm using responses from the screening questionnaire to predict HRSN needs (screening questionnaire model). Additionally, we extracted features from the past 12 months of existing EHR, administrative, and health information exchange data for the survey respondents. We built ML predictive models with these EHR data using XGBoost (ML EHR model). Out of concerns of potential bias, we built both the screening question model and the ML EHR model with and without demographic features. Models were assessed on the validation set using sensitivity, specificity, and Area Under the Curve (AUC) values. Models were compared using the Delong test. Almost half (41%) of the patients had a positive indicator for a likely HRSN service need within the next 30 days, as identified through referrals, encounters, scheduling data, orders, or clinical notes. The screening question model had suboptimal performance, with an AUC = 0.580 (95%CI = 0.546, 0.611). Including gender and age resulted in higher performance in the screening question model (AUC = 0.640; 95%CI = 0.609, 0.672). The ML EHR models had higher performance. Without including age and gender, the ML EHR model had an AUC = 0.765 (95%CI = 0.737, 0.792). Adding age and gender did not improve the model (AUC = 0.722; 95%CI = 0.744, 0.800). The screening questionnaire models indicated bias with the highest performance for White non-Hispanic patients. The performance of the ML EHR-based model also differed by race and ethnicity. ML predictive models leveraging several robust EHR data sources outperformed models using screening questions only. Nevertheless, all models indicated biases. Additional work is needed to design predictive models for effectively identifying all patients with HRSNs.

Optimizing Electronic Health Records for Enhanced Clinical Decision Support and Improved Patient Care through Advanced Data Integration and Machine Learning Algorithms

Electronic Health Records (EHRs) have changed the way healthcare is provided by recording patient information, making it easier to handle data, and letting healthcare workers talk to each other without any problems. But EHRs haven't completely come to their full guarantee however when it comes to making strides quiet care and making proficient choices superior. This theoretical talks almost a other way to make strides EHR frameworks by utilizing progressed information integration and machine learning methods. The objective is to create clinical choice back way better and, within the conclusion, make understanding comes about way better. The recommended approach makes the foremost of the colossal sum of information in electronic wellbeing records (EHRs) by combining distinctive sorts of information, such as organized clinical information, unorganized composed notes, restorative images, genetic information, and real-time real signs. Diverse information sources are brought together utilizing progressed information integration strategies to form a full understanding profile that appears the patient's wellbeing state and therapeutic foundation as a entirety. This bound together data environment is the premise for building progressed machine learning models that can draw valuable conclusions, figure how patients will do, and recommend the most excellent care for each individual. Machine learning methods are exceptionally vital for getting valuable data from EHR information. These calculations can discover mystery associations and complex designs by utilizing strategies like profound learning, normal dialect preparing, and expectation modeling. They can also accurately predict clinical events. These models are always learning and getting better, so they can change to new patient data and clinical situations.

Perceptions of hospital electronic health record (EHR) training, support, and patient safety by staff position and tenure

BackgroundHospitals rely on their electronic health record (EHR) systems to assist with the provision of safe, high quality, and efficient health care. However, EHR systems have been found to disrupt clinical workflows and may lead to unintended consequences associated with patient safety and health care professionals’ perceptions of and burden with EHR usability and interoperability. This study sought to explore the differences in staff perceptions of the usability and safety of their hospital EHR system by staff position and tenure.MethodsWe used data from the AHRQ Surveys on Patient Safety Culture® (SOPS®) Hospital Survey Version 1.0 Database and the SOPS Health Information Technology Patient Safety Supplemental Items (“Health IT item set”) collected from 44 hospitals and 8,880 staff in 2017. We used regression modeling to examine perceptions of EHR system training, EHR support & communication, EHR-related workflow, satisfaction with the EHR system, and the frequency of EHR-related patient safety and quality issues by staff position and tenure, while controlling for hospital ownership type and bed-size.ResultsIn comparison to RNs, pharmacists had significantly lower (unfavorable) scores for EHR system training (regression coefficient = -0.07; p = 0.047), and physicians, hospital management, and the IT staff were significantly more likely to report high frequency of inaccurate EHR information (ORs = 2.03, 1.34, 1.72, respectively). Compared to staff with 11 or more years of hospital tenure, new staff (less than 1 year at the hospital) had significantly lower scores for EHR system training, but higher scores for EHR support & communication (p < 0.0001). Dissatisfaction of the EHR system was highest among physicians and among staff with 11 or more years tenure at the hospital.ConclusionsThere were significant differences in the Health IT item set’s results across staff positions and hospital tenure. Hospitals can implement the SOPS Health IT Patient Safety Supplemental Items as a valuable tool for identifying incongruity in the perceptions of EHR usability and satisfaction across staff groups to inform targeted investment in EHR system training and support.

The reuse of electronic health records information models in the oncology domain: Studies with the bioframe framework

Objective:The reuse of Electronic Health Records (EHR) information models (e.g., templates and archetypes) may bring various benefits, including higher standardization, integration, interoperability, increased productivity in developing EHR systems, and unlock potential Artificial Intelligence applications built on top of medical records. The literature presents recent advances in standards for modeling EHR, in Knowledge Organization Systems (KOS) and EHR data reuse. However, methods, development processes, and frameworks to improve the reuse of EHR information models are still scarce. This study proposes a software engineering framework, named BioFrame, and analyzes how the reuse of EHR information models can be improved during the development of EHR systems. Methods:EHR standards and KOS, including ontologies, identified from systematic reviews were considered in developing the BioFrame framework. We used the structure of the OpenEHR to model templates and archetypes, as well as its relationship to international KOS used in the oncology domain. Our framework was applied in the context of pediatric oncology. Three data entry forms concerning nutrition and one utilized during the first pediatric oncology consultations were analyzed to measure the reuse of information models. Results:There was an increase in the adherence rate to international KOS of 18% to the original forms. There was an increase in the concepts reused in all 12 scenarios analyzed, with an average reuse of 6.55% in the original forms compared to 17.1% using BioFrame, resulting in significant differences. Conclusions:Our results point to higher reuse rates achieved due to an engineering process that provided greater adherence to EHR standards combined with semantic artifacts. This reveals the potential to develop new methods and frameworks aimed at EHR information model reuse. Additional research is needed to evaluate the impacts of the reuse of the EHR information model on interoperability, EHR data reuse, and data quality and assess the proposed framework in other health domains.

Prevalence and risk factors for long COVID among adults in Scotland using electronic health records: a national, retrospective, observational cohort study

Long COVID is a debilitating multisystem condition. The objective of this study was to estimate the prevalence of long COVID in the adult population of Scotland, and to identify risk factors associated with its development. In this national, retrospective, observational cohort study, we analysed electronic health records (EHRs) for all adults (≥18 years) registered with a general medical practice and resident in Scotland between March 1, 2020, and October 26, 2022 (98-99% of the population). We linked data from primary care, secondary care, laboratory testing and prescribing. Four outcome measures were used to identify long COVID: clinical codes, free text in primary care records, free text on sick notes, and a novel operational definition. The operational definition was developed using Poisson regression to identify clinical encounters indicative of long COVID from a sample of negative and positive COVID-19 cases matched on time-varying propensity to test positive for SARS-CoV-2. Possible risk factors for long COVID were identified by stratifying descriptive statistics by long COVID status. Of 4,676,390 participants, 81,219 (1.7%) were identified as having long COVID. Clinical codes identified the fewest cases (n=1,092, 0.02%), followed by free text (n=8,368, 0.2%), sick notes (n=14,469, 0.3%), and the operational definition (n=64,193, 1.4%). There was limited overlap in cases identified by the measures; however, temporal trends and patient characteristics were consistent across measures. Compared with the general population, a higher proportion of people with long COVID were female (65.1% versus 50.4%), aged 38-67 (63.7% versus 48.9%), overweight or obese (45.7% versus 29.4%), had one or more comorbidities (52.7% versus 36.0%), were immunosuppressed (6.9% versus 3.2%), shielding (7.9% versus 3.4%), or hospitalised within 28 days of testing positive (8.8% versus 3.3%%), and had tested positive before Omicron became the dominant variant (44.9% versus 35.9%). The operational definition identified long COVID cases with combinations of clinical encounters (from four symptoms, six investigation types, and seven management strategies) recorded in EHRs within 4-26 weeks of a positive SARS-CoV-2 test. These combinations were significantly (p<0.0001) more prevalent in positive COVID-19 patients than in matched negative controls. In a case-crossover analysis, 16.4% of those identified by the operational definition had similar healthcare patterns recorded before testing positive. The prevalence of long COVID presenting in general practice was estimated to be 0.02-1.7%, depending on the measure used. Due to challenges in diagnosing long COVID and inconsistent recording of information in EHRs, the true prevalence of long COVID is likely to be higher. The operational definition provided a novel approach but relied on a restricted set of symptoms and may misclassify individuals with pre-existing health conditions. Further research is needed to refine and validate this approach. Chief Scientist Office (Scotland), Medical Research Council, and BREATHE.

Mapping the Landscape of Electronic Health Records and Health Information Exchange Through Bibliometric Analysis and Visualization.

The adoption of Electronic Health Records (EHRs) and the establishment of Health Information Exchange (HIE) systems have significantly transformed healthcare delivery and management. This study presents a comprehensive bibliometric analysis and visualization of the landscape surrounding EHRs and HIE to provide insights into the current state and emerging trends in this field. Leveraging advanced bibliometric methodologies, including co-citation analysis, keyword co-occurrence analysis, and network visualization techniques, we systematically map the scholarly literature spanning several decades. Our analysis reveals key thematic clusters, influential publications, prolific authors, and collaborative networks within the domain of EHRs and HIE. Furthermore, we identify significant research gaps and areas for future exploration, including interoperability challenges, privacy concerns, and the integration of emerging technologies such as artificial intelligence and blockchain. The findings of this study offer valuable insights for researchers, policymakers, and healthcare practitioners seeking to navigate and contribute to the ongoing evolution of EHRs and HIE systems, ultimately enhancing the quality, efficiency, and accessibility of healthcare services.

Enhancing healthcare decision support through explainable AI models for risk prediction

Electronic health records (EHRs) are a valuable source of information that can aid in understanding a patient’s health condition and making informed healthcare decisions. However, modelling longitudinal EHRs with heterogeneous information is a challenging task. Although recurrent neural networks (RNNs) are frequently utilized in artificial intelligence (AI) models for capturing longitudinal data, their explanatory capabilities are limited. Predictive clustering stands as the most recent advancement within this domain, offering interpretable indications at the cluster level for predicting disease risk. Nonetheless, the challenge of determining the optimal number of clusters has put a brake on the widespread application of predictive clustering for disease risk prediction. In this paper, we introduce a novel non-parametric predictive clustering-based risk prediction model that integrates the Dirichlet Process Mixture Model (DPMM) with predictive clustering via neural networks. To enhance the model’s interpretability, we integrate attention mechanisms that enable the capture of local-level evidence in addition to the cluster-level evidence provided by predictive clustering. The outcome of this research is the development of a multi-level explainable artificial intelligence (AI) model. We evaluated the proposed model on two real-world datasets and demonstrated its effectiveness in capturing longitudinal EHR information for disease risk prediction. Moreover, the model successfully produced interpretable evidence to bolster its predictions.

Securing the Patient’s Breast Cancer Data using Blockchain-based IBE with Deep Learning Model in IoT

Computational intelligence (CI) and artificial intelligence (AI) have a portion to offer in the way of making healthcare systems smart and long-lasting. These technological advances have the potential to lessen our environmental impact while simultaneously increasing expectations for performance. The use of electronic health monitoring systems in healthcare management has been crucial. E-health has the potential to offer patients useful and efficient tracking tools. However, the existing E-Health system has protection disputes. However, there are safety concerns with the present online healthcare system. Doctors with bad intentions might collude with CSPs to compromise their patients' electronic health records (EHRs) or quickly leak EHR information to other opponents for financial gain. Patients might be manipulated by clinicians who work in collusion with a Patient Healthcare Monitoring Service Provider (PHMSP). EHRs for financial gain or disclose the HER content of EHRs directly to other adversaries. Recently, block-chain has emerged as one of the most potent strategies for maintaining privacy and security. The current security issues in e-health monitoring systems are expected to be replaced by this promising security method. To prevent unauthorised parties from accessing encrypted data, blockchain uses encryption. Using a computational intelligence approach, this study presented a blockchain-based encryption framework based on identity-based encryption (IBE) to deliver secure solutions. Furthermore, the study effort utilises optimised deep learning (ODL) to forecast the patient's condition from breast cancer input data. The suggested method improves performance, with an accuracy of 0.93 during training and 0.91 during validation.

420 Computable Phenotyping with “Big Data” as a Foundation for Artificial Intelligence Algorithm Construction: Puberty as a Transdisciplinary Case Example

OBJECTIVES/GOALS: Artificial intelligence (AI) depends on quality machine learning (ML) algorithms constructed with high-quality training data. This TL1 trainee project develops a disease-agnostic computable phenotype framework for ML algorithm construction, modeling male puberty as a case example. METHODS/STUDY POPULATION: A computable phenotype of male puberty was constructed to answer the question: “Does early pubertal timing increase the risk of developing type II diabetes (T2D) in males?” A computable phenotype of males &lt; 18 years old was created in the TriNetX© Diamond Network utilizing Boolean operator data queries. TriNetX© contains patient electronic health record information (ICD-10 diagnoses, anthropometric measures). An exploratory analysis of patient counts reflecting various computable phenotypes allowed for outcome (T2D) comparison of males diagnosed with precocious puberty (E30.1, ICD code for early pubertal timing) to those without, controlling for body mass index (BMI). RESULTS/ANTICIPATED RESULTS: Subjects (n=12,996,132) displayed the following computable phenotype: Male, &lt; 18 years old, without ever having a BMI documented &gt;85th percentile. Males diagnosed with precocious puberty (E30.1) were 6.89 times more likely to develop T2D when aged 14-18 years old than those without (OR 6.89, 95% CI: 5.17-9.19, p&lt;0.0001). Next steps involve training a ML model on each computable phenotype groupings’ health data, with anticipated results identifying underlying salient pathophysiologic variables. A generalized computable phenotype approach is further developed to: 1) explore clinical questions in large databases like TriNetX©, and 2) model disease development with AI/ML algorithm construction. DISCUSSION/SIGNIFICANCE: Computed phenotypes reveal males with precocious puberty may have increased T2D risk. Next steps utilize subject data to train an AI/ML algorithm, model development to identify salient pathophysiologic variables, and synthesize a generalized AI/ML developmental research framework for dissemination.

Abstract 3569: Using AI to automatically process data from unstructured health records of patients with lung cancer

Abstract Introduction: Structured databases created from electronic health records (EHR) are crucial for cancer research. Manual data entry into databases is both labor-intensive and error-prone. We aimed to develop an artificial intelligence (AI)-driven approach for automatically inputting patient information from EHRs. Methods: A REDCap database for 53 patients with advanced lung cancer treated with first line immuno+/-chemotherapy at Gustave Roussy between 02/2021-06/2023 was manually populated by physicians with demographics, risk factors, cancer history, and treatment data, with 137 variables/patient. Given unstructured medical letters and a schematic description of each variable, generative AI was used to find, quote and process variables into a structured form. We directed large language model actions with prompt engineering and tailored few-shot examples. Mortality data were auto-extracted from the French public registry, INSEE. We assessed consistency between manual (MDE) and automated data entry (ADE), with a secondary manual review for mismatches performed by a senior physician. Results: In total, 7,261 data points were assessed. ADE averaged 10 minutes per patient inclusive of quality checks vs 1 h for MDA. The concordance rate between ADE and MDE was 90.2% (6,550/7,261) with a discordance of 6.8% (496/7,261). Data was missing in 0.8% (59/7,261) of cases for both methods, 1% (73/7,261) for ADE and 1,1% (83/7,261) for MDA. After checking discordances, ADE correctness was 95.3% (6,922/7,261) (Table), MDE was 93.8% (6,809/7,261). Errors in ADE (207/7,261) were due to algorithm refinement needs (70%) and missing EHR information (30%). Conclusion: Generative AI has a strong potential for identifying and structuring data from EHRs, yielding consistency superior to manual entry by physicians and 85% reduced amount of time. This may enhance the efficiency, accuracy, and scalability of EHR-to-database conversions. ADE for &amp;gt;1000 patients will be presented at the meeting. ADE correctness 100% 95-99% 90-94% 80-89% Sex; Age; Life status; Last follow-up; Metastatic from diagnosis Molecular alterations; Metastatic sites; Progression sites; Treatment type Tobacco &amp; cannabis consumption; Histology; PDL1 score; Progression event Pack years; Date of diagnosis &amp; first metastasis; Stage; Date of start treatment Citation Format: Mihaela Aldea, Pierre Rolland, Solenne Simon, Aliette Poplu, Muriel Wartelle, Benjamin Vignal, Jean-Charles Louis, Francois Lion, Arnaud Borie, David Planchard, Caroline Robert, Stefan Michiels, Fabrice Andre, Fabrice Barlesi, Franck Le Ouay, Benjamin Besse. Using AI to automatically process data from unstructured health records of patients with lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3569.

Characterisation and validation of lactation information from structured electronic health records for use in pharmacoepidemiological studies.

Breastfeeding information stored within electronic health records (EHR) has recently been used for pharmacoepidemiological research, however the data are primarily collected for clinical care. To characterise breastfeeding information recorded in structured fields in EHR during infant and postpartum health care visits, and to assess the validity of lactation status based on EHR data versus maternal report at research study visits. We assessed breastfeeding information recorded in structured fields in EHR from one health system for a subset of 211 patients who were also enrolled in a study on breast milk composition between 2014 and 2017 that required participants to exclusively breastfeed their infants until at least 1 month of age. We assessed the frequency of breastfeeding information in EHR during the first 12 months of age and compared lactation status based on EHR with maternal report at 1 and 6-month study visits (reference standard). The median number of breastfeeding records in the EHR per infant was six (interquartile range 3) with most observations clustering in the first few weeks of life and around well-infant visits. At the 6-month study visit, 93.8% of participants were breastfeeding and 80.1% were exclusively breastfeeding according to maternal report. Sensitivity of EHR data for identifying ever breastfeeding was at or near 100%, and sensitivity for identifying ever exclusive breastfeeding was 98.0% (95% CI: 95.0%, 99.2%). Sensitivities were 97.3% (95% CI: 93.9%, 98.9%) for identifying any breastfeeding and 94.4% (95% CI: 89.7%, 97.0%) for exclusive breastfeeding, and positive predictive values were 99.5% (95% CI: 97.0%, 99.9%) for any breastfeeding and 95.0% (95% CI: 90.4%, 97.4%) for exclusive breastfeeding. Breastfeeding information in structured EHR fields have the potential to accurately classify lactation status. The validity of these data should be assessed in populations with a lower breastfeeding prevalence.

Towards Reducing Diagnostic Errors with Interpretable Risk Prediction.

Many diagnostic errors occur because clinicians cannot easily access relevant information in patient Electronic Health Records (EHRs). In this work we propose a method to use LLMs to identify pieces of evidence in patient EHR data that indicate increased or decreased risk of specific diagnoses; our ultimate aim is to increase access to evidence and reduce diagnostic errors. In particular, we propose a Neural Additive Model to make predictions backed by evidence with individualized risk estimates at time-points where clinicians are still uncertain, aiming to specifically mitigate delays in diagnosis and errors stemming from an incomplete differential. To train such a model, it is necessary to infer temporally fine-grained retrospective labels of eventual "true" diagnoses. We do so with LLMs, to ensure that the input text is from before a confident diagnosis can be made. We use an LLM to retrieve an initial pool of evidence, but then refine this set of evidence according to correlations learned by the model. We conduct an in-depth evaluation of the usefulness of our approach by simulating how it might be used by a clinician to decide between a pre-defined list of differential diagnoses.

Factors Associated with Online Patient Portal Utilization Experience in an Arkansas Phone Survey.

Introduction: Accessing electronic health record information through a patient portal is associated with numerous benefits to both health care providers and patients. However, patient portal utilization remains low. Little is known about the factors associated with patient portal utilization following the onset of the COVID-19 pandemic. Methods: In March 2022, we conducted a random digit dial phone survey of both cell phones and landlines of adults living in Arkansas that asked numerous demographic and health-related measures, including patient portal utilization in the past 12 months. A total of 2,201 adult Arkansans completed the survey between March 1 and March 28, 2022. Weighted estimates were generated using rank ratio estimation to approximate the 2019 American Community Survey 1-year Arkansas estimates for race/ethnicity (72% White, 15% Black/African American, 7.8% Hispanic, 4.9% other race/ethnicity), age (73% 18-39, 32% 40-59, and 31% 60+), and gender (49% male, 51% female). We fit the data to a logistic regression model. Results: We found that education, employment, prior telehealth experience, having a check-up in the past 2 years, and having a primary care provider were all positively associated with patient portal utilization. We also found that non-Hispanic Black/African-American respondents were less likely to access a patient portal relative to non-Hispanic White respondents. Discussion: Patient portal utilization is related to several demographic and health-related factors among an adult population in Arkansas. Given that the documented benefits of patient portal utilization are broad, under-utilization by groups that already experience relatively worse health outcomes could reproduce or even exacerbate existing health disparities. Additional research is needed to further investigate what barriers to patient portal utilization remain for these populations.

Trustworthy Data and AI Environments for Clinical Prediction: Application to Crisis-Risk in People with Depression.

Depression is a common mental health condition that often occurs in association with other chronic illnesses, and varies considerably in severity. Electronic Health Records (EHRs) contain rich information about a patient's medical history and can be used to train, test and maintain predictive models to support and improve patient care. This work evaluated the feasibility of implementing an environment for predicting mental health crisis among people living with depression based on both structured and unstructured EHRs. A large EHR from a mental health provider, Mersey Care, was pseudonymised and ingested into the Natural Language Processing (NLP) platform CogStack, allowing text content in binary clinical notes to be extracted. All unstructured clinical notes and summaries were semantically annotated by MedCAT and BioYODIE NLP services. Cases of crisis in patients with depression were then identified. Random forest models, gradient boosting trees, and Long Short-Term Memory (LSTM) networks, with varying feature arrangement, were trained to predict the occurrence of crisis. The results showed that all the prediction models can use a combination of structured and unstructured EHR information to predict crisis in patients with depression with good and useful accuracy. The LSTM network that was trained on a modified dataset with only 1000 most-important features from the random forest model with temporality showed the best performance with a mean AUC of 0.901 and a standard deviation of 0.006 using a training dataset and a mean AUC of 0.810 and 0.01 using a hold-out test dataset. Comparing the results from the technical evaluation with the views of psychiatrists shows that there are now opportunities to refine and integrate such prediction models into pragmatic point-of-care clinical decision support tools for supporting mental healthcare delivery.

Pay-for-performance for appropriate prescribing using routine healthcare data from general practices

Abstract Background A national pay-for-performance (P4P) scheme was introduced in the Netherlands to incentivize appropriate prescribing among general practitioners (GPs) using routine care data of electronic health records (EHRs). The goal of this program was to encourage GPs to choose the most clinically appropriate and cost-effective medication for their patients to promote an affordable and sustainable health system. Nivel facilitated and evaluated this program. Facilitators and barriers of this P4P program were assessed as part of this endeavour. Methods Semi-structured interviews were conducted with all stakeholder representatives (n = 15) to evaluate their experiences with the program. The interviews were analysed thematically. Results Our research was structured around the three ‘flows’ necessary for an effective learning health system. The ‘data to knowledge’ flow was impacted by: 1) effect of different EHR information systems on data quality, and 2) advantages of using EHR data for measuring quality of care, minimising GPs’ administrative burden. The ‘knowledge to practice’ flow was impacted by: 1) meaningfulness and robustness of indicators, and 2) measure for appropriate prescribing. The ‘practice to data’ flow was impacted by: 1) feedback to GPs, with most stakeholders emphasizing the need for GPs to reflect on their results with other GPs to stimulate learning, 2) impact of the P4P model, such as improvement in quality of GP registration in EHRs and better stakeholder collaboration, and 3) methods, communication and GP perspectives on the performance payment. Conclusions The study showed the factors that can support or hinder P4P programs in promoting appropriate care with EHR data for a sustainable and affordable health system. Enhancing uniformity of EHR information systems is crucial for more consistent quality of care assessment. To encourage GPs to deliver better care, it is recommended to complement monetary incentives with other forms of incentives. Key messages • This study showed that routine healthcare data are feasible for the use in P4P schemes promoting a sustainable and affordable health system. • P4P as a standalone approach may not be optimal for promoting a sustainable and affordable health system.

Vulnerable Patients with Inflammatory Bowel Disease

Crohn’s Disease and ulcerative colitis, collectively known as Inflammatory Bowel Disease (IBD), are conditions characterized by inflammation and ulceration of the intestines. This work aims to highlight disparities among vulnerable IBD populations using the OneFlorida Clinical Research Consortium, a data network housing claims and electronic health record information for more than 15 million patients in Florida. A retrospective analysis using data collected between 01/01/2012 and 06/30/2020 was performed using OneFlorida, and patients were identified using diagnosis and pharmacy codes for IBD. 10,578 patients met the inclusion criteria. Healthcare utilization (emergency department visits, inpatient stay, medications, procedures), disease severity (c-reactive protein, white blood cell count, hemoglobin, albumin), and outcomes (i.e., abdominal surgery) were assessed for patients of different ages, gender, ethnicities, races, and locations. Notably, adolescent and young adult (AYA) patients (ages 17-25) use significantly more corticosteroids than both the pediatric and older adult groups (p-values 0.005 and 0.0007). Contrary to prior literature, Hispanic IBD patients used more biologic medications compared to non-Hispanic White patients (p&lt;0.0001), whereas non-Hispanic Black patients used more steroids (p=0.0004). These findings illustrate a need for targeted educational interventions to support these vulnerable populations.

EHR-KnowGen: Knowledge-enhanced multimodal learning for disease diagnosis generation

Electronic health records (EHRs) contain diverse patient information, including medical notes, clinical events, and laboratory test results. Integrating this multimodal data can improve disease diagnoses using deep learning models. However, effectively combining different modalities for diagnosis remains challenging. Previous approaches, such as attention mechanisms and contrastive learning, have attempted to address this but do not fully integrate the modalities into a unified feature space. This paper presents EHR-KnowGen, a multimodal learning model enhanced with external domain knowledge, for improved disease diagnosis generation from diverse patient information in EHRs. Unlike previous approaches, our model integrates different modalities into a unified feature space with soft prompts learning and leverages large language models (LLMs) to generate disease diagnoses. By incorporating external domain knowledge from different levels of granularity, we enhance the extraction and fusion of multimodal information, resulting in more accurate diagnosis generation. Experimental results on real-world EHR datasets demonstrate the superiority of our generative model over comparative methods, providing explainable evidence to enhance the understanding of diagnosis results.