Clinical Electronic Health Record Research Articles

Individual Pharmacotherapy Management (IPM-II) for Patient and Drug Safety in Polypharmacy via Clinical Electronic Health Record Is Associated with Significant Fall Prevention.

Falls and fractures are emerging as a near-pandemic and major global health concern, placing an enormous burden on ageing patients and public health economies. Despite the high risk of polypharmacy in the elderly patients, falls are usually attributed to age-related changes. For the "Individual Pharmacotherapy Management (IPM)" established at the University Hospital Halle, the IPM medication adjustments and their association with in-hospital fall prevention were analysed. On the basis of the most updated digital overall patient view via his inpatient electronic health record (EHR), IPM adapts each drug's Summary of Product Characteristics to the patient's condition. A retrospective pre-post intervention study in geriatric traumatology on ≥70 years old patients compared 200 patients before IPM implementation (CG) with 204 patients from the IPM intervention period (IG) for the entire medication list, organ, cardiovascular and vital functions and fall risk parameters. Statistically similar baseline data allowed a comparison of the average 80-year-old patient with a mean of 11.1 ± 4.9 (CG) versus 10.4 ± 3.6 (IG) medications. The IPM adjusted for drug-drug interactions, drug-disease interactions, overdoses, anticholinergic burden, adverse drug reactions, esp. from opioids inducing increased intrasynaptic serotonin, psychotropic drugs, benzodiazepines, contraindications and missing prescriptions. IPM was associated with a significant reduction in in-hospital falls from 18 (9%) in CG to 3 (1.5%) in IG, a number needed to treat of 14, relative risk reduction 83%, OR 0.17 [95% CI 0.04; 0.76], p = 0.021 in multivariable regression analysis. Factors associated with falls were antipsychotics, digitoxin, corticosteroids, Würzburg pain drip (combination of tramadol, metamizole, metoclopramide), head injury, cognitive impairment and aspects of the Huhn Fall Risk Scale including urinary catheter. The results indicate medication risks constitute a major iatrogenic cause of falls in this population and support the use of EHR-based IPM in standard care for the prevention of falls in the elderly and for patient and drug safety. In terms of global efforts, IPM contributes to the running WHO and United Nations Decade of Healthy Ageing (2021-2030).

Predicting involuntary admission following inpatient psychiatric treatment using machine learning trained on electronic health record data

Abstract Background Involuntary admissions to psychiatric hospitals are on the rise. If patients at elevated risk of involuntary admission could be identified, prevention may be possible. Our aim was to develop and validate a prediction model for involuntary admission of patients receiving care within a psychiatric service system using machine learning trained on routine clinical data from electronic health records (EHRs). Methods EHR data from all adult patients who had been in contact with the Psychiatric Services of the Central Denmark Region between 2013 and 2021 were retrieved. We derived 694 patient predictors (covering e.g. diagnoses, medication, and coercive measures) and 1134 predictors from free text using term frequency-inverse document frequency and sentence transformers. At every voluntary inpatient discharge (prediction time), without an involuntary admission in the 2 years prior, we predicted involuntary admission 180 days ahead. XGBoost and elastic net models were trained on 85% of the dataset. The models with the highest area under the receiver operating characteristic curve (AUROC) were tested on the remaining 15% of the data. Results The model was trained on 50 634 voluntary inpatient discharges among 17 968 patients. The cohort comprised of 1672 voluntary inpatient discharges followed by an involuntary admission. The best XGBoost and elastic net model from the training phase obtained an AUROC of 0.84 and 0.83, respectively, in the test phase. Conclusion A machine learning model using routine clinical EHR data can accurately predict involuntary admission. If implemented as a clinical decision support tool, this model may guide interventions aimed at reducing the risk of involuntary admission.

Generating Biomedical Knowledge Graphs from Knowledge Bases, Registries, and Multiomic Data.

As large clinical and multiomics datasets and knowledge resources accumulate, they need to be transformed into computable and actionable information to support automated reasoning. These datasets range from laboratory experiment results to electronic health records (EHRs). Barriers to accessibility and sharing of such datasets include diversity of content, size and privacy. Effective transformation of data into information requires harmonization of stakeholder goals, implementation, enforcement of standards regarding quality and completeness, and availability of resources for maintenance and updates. Systems such as the Biomedical Data Translator leverage knowledge graphs (KGs), structured and machine learning readable knowledge representation, to encode knowledge extracted through inference. We focus here on the transformation of data from multiomics datasets and EHRs into compact knowledge, represented in a KG data structure. We demonstrate this data transformation in the context of the Translator ecosystem, including clinical trials, drug approvals, cancer, wellness, and EHR data. These transformations preserve individual privacy. We provide access to the five resulting KGs through the Translator framework. We show examples of biomedical research questions supported by our KGs, and discuss issues arising from extracting biomedical knowledge from multiomics data.

Improving screening, brief intervention and referral to treatment for unhealthy alcohol use in diverse, low-resourced primary care clinics

BackgroundImplementation of screening brief intervention and referral to treatment (SBIRT) and medication-assisted treatment for alcohol use disorder (MAUD) remains low in primary care. ANTECEDENT (Partnerships to Enhance Alcohol Screening, Treatment, and Intervention) was a practice-facilitator led implementation study to increase SBIRT and MAUD use in diverse primary care clinics.MethodsFrom November 2019 – April 2023, we conducted a convergent parallel mixed methods evaluation. Participants were small and medium-sized primary care clinics in the Northwestern U.S. Clinics received foundational support (i.e., baseline/exit assessment, access to SBIRT Oregon website) and the option for supplemental implementation support (e.g., practice facilitation, expert consultation) over the 15-month intervention to improve SBIRT and MAUD. Qualitative and quantitative data regarding clinic characteristics, implementation strategies, and SBIRT/MAUD outcomes were collected through practice facilitator interviews, periodic reflections and clinic contact logs, interviews, pre-post surveys, and electronic health record (EHR) queries. Quantitative analyses included descriptive statistics, logistic regression, and the Wilcoxon sign-ranked test. Qualitative analysis took an immersion crystallization approach. Data were reviewed in a matrix to evaluate intervention adoption, implementation, and effectiveness.ResultsWe recruited 75 unique clinics; 66 participated and 48 (73%) completed the study. Eight participating clinics chose to receive foundational support only (12%) while 58 chose to engage in supplemental support (88%) activities. Clinics that received supplemental support and completed the intervention (n = 42) engaged in practice facilitation (Mean: 293 min, range: 75–550 min); data review (38%), HIT support (31%), expert consultation (19%), and peer-to-peer learning (5%). Pre- and post-intervention assessments showed significant improvement in self-reported SBIRT process outcomes. Performance data improved among the subset of completing clinics able to produce data (n = 17). Clinics described positive experiences with ANTECEDENT support and highlighted remaining barriers to SBIRT and MAUD implementation.ConclusionsParticipating clinics engaged in a diverse range of supportive activities. A flexible approach using practice facilitation and implementation support could be helpful for low-resourced primary care clinics in improving SBIRT and MAUD for unhealthy alcohol use.

CriteriaMapper: establishing the automatic identification of clinical trial cohorts from electronic health records by matching normalized eligibility criteria and patient clinical characteristics

The use of electronic health records (EHRs) holds the potential to enhance clinical trial activities. However, the identification of eligible patients within EHRs presents considerable challenges. We aimed to develop a CriteriaMapper system for phenotyping eligibility criteria, enabling the identification of patients from EHRs with clinical characteristics that match those criteria. We utilized clinical trial eligibility criteria and patient EHRs from the Mount Sinai Database. The CriteriaMapper system was developed to normalize the criteria using national standard terminologies and in-house databases, facilitating computability and queryability to bridge clinical trial criteria and EHRs. The system employed rule-based pattern recognition and manual annotation. Our system normalized 367 out of 640 unique eligibility criteria attributes, covering various medical conditions including non-small cell lung cancer, small cell lung cancer, prostate cancer, breast cancer, multiple myeloma, ulcerative colitis, Crohn’s disease, non-alcoholic steatohepatitis, and sickle cell anemia. About 174 criteria were encoded with standard terminologies and 193 were normalized using the in-house reference tables. The agreement between automated and manual normalization was high (Cohen’s Kappa = 0.82), and patient matching demonstrated a 0.94 F1 score. Our system has proven effective on EHRs from multiple institutions, showing broad applicability and promising improved clinical trial processes, leading to better patient selection, and enhanced clinical research outcomes.

The Trial Enrollment Diversity Dashboard for Acute Leukemia Clinical Research: Intervention Development and Cohort Analysis.

Participation in acute leukemia clinical trials is inequitable across multiple sociodemographic categories. Tools that provide researchers with performance feedback on the representativeness of the patients they enroll are limited. We aimed to develop an electronic health record (EHR)-based dashboard to provide such feedback and to describe any enrollment inequities uncovered. We created a visual dashboard linking leukemia clinical trial registration and EHR data at the Dana-Farber Cancer Institute. Accuracy of a patient inclusion and assignment algorithm was tested with a target area under the receiver-operator curve (AUROC) of >0.90 against manual review. Demographic metric identification, visualization construction, and dashboard refinement were performed through stakeholder cognitive testing. Analysis of a recent 5-year cohort generated by the final algorithm assessed bivariate associations between enrollment and demographic metrics. Multivariable logistic regression included significant bivariate results. The final algorithm assignment AUROC was 0.98. Metrics were identified and visualizations successfully constructed. Fourteen individuals participated in testing and identified areas for revision: category mergers, denominator filters, and data delivery preferences. In the initial cohort of 1,315 patients, 1,020 (77.6%) had enrolled in any study protocol: 553 (42.1%) in a treatment trial and 936 (71.2%) in a biobanking study. In a multivariable model, older age (odds ratio [OR], 0.83 [95% CI, 0.73 to 0.94]) and Non-Hispanic Black race-ethnicity (OR, 0.38 [95% CI, 0.18 to 0.82]) were associated with lower enrollment, and English primary language with higher enrollment (OR, 2.50 [95% CI, 1.30 to 4.79]). We developed a research participation equity performance feedback dashboard for clinical researchers, and we identified actionable inequities. Next steps include feasibility and efficacy testing as well as implementation.

Model development for bespoke large language models for digital triage assistance in mental health care

Contemporary large language models (LLMs) may have utility for processing unstructured, narrative free-text clinical data contained in electronic health records (EHRs) — a particularly important use-case for mental health where a majority of routinely-collected patient data lacks structured, machine-readable content.A significant problem for the United Kingdom’s National Health Service (NHS) are the long waiting lists for specialist mental healthcare. According to NHS data (NHS Digital, 2024), in each month of 2023, there were between 370,000 and 470,000 individual new referrals into secondary mental healthcare services. Referrals must be triaged by clinicians, using clinical information contained in the patient’s EHR to arrive at a decision about the most appropriate mental healthcare team to assess and potentially treat these patients.The ability to efficiently recommend a relevant team by ingesting potentially voluminous clinical notes could help services both reduce referral waiting times and with the right technology, improve the evidence available to justify triage decisions.We present and evaluate three different approaches for LLM-based, end-to-end ingestion of variable-length clinical EHR data to assist clinicians when triaging referrals. Our model is able to deliver triage recommendations consistent with existing clinical practices and its architecture was implemented on a single GPU, making it practical for implementation in resource-limited NHS environments where private implementations of LLM technology will be necessary to ensure confidential clinical data are appropriately controlled and governed. Code available at: https://github.com/NtaylorOX/BespokeLLM_Triage.

Inferring Population HIV Viral Load From a Single HIV Clinic's Electronic Health Record: Simulation Study With a Real-World Example.

Population viral load (VL), the most comprehensive measure of the HIV transmission potential, cannot be directly measured due to lack of complete sampling of all people with HIV. A given HIV clinic's electronic health record (EHR), a biased sample of this population, may be used to attempt to impute this measure. We simulated a population of 10,000 individuals with VL calibrated to surveillance data with a geometric mean of 4449 copies/mL. We sampled 3 hypothetical EHRs from (A) the source population, (B) those diagnosed, and (C) those retained in care. Our analysis imputed population VL from each EHR using sampling weights followed by Bayesian adjustment. These methods were then tested using EHR data from an HIV clinic in Delaware. Following weighting, the estimates moved in the direction of the population value with correspondingly wider 95% intervals as follows: clinic A: 4364 (95% interval 1963-11,132) copies/mL; clinic B: 4420 (95% interval 1913-10,199) copies/mL; and clinic C: 242 (95% interval 113-563) copies/mL. Bayesian-adjusted weighting further improved the estimate. These findings suggest that methodological adjustments are ineffective for estimating population VL from a single clinic's EHR without the resource-intensive elucidation of an informative prior.

#1559 Using nephrology clinic data to emulate cohorts: a comparison between an electronic health records service and Chronic Renal Insufficiency Cohort

Abstract Background and Aims The amount of data contributed to evidence-synthesis is inequitably distributed among various countries in a way that most of the data is emerged from developed and/or upper middle and high income countries. One efficient and low-cost solution can be the implementation of scattered electronic health records (EHR) sources in those countries into emulated cohort studies. We aim to compare the characteristics of a custom-built nephrology clinic electronic health records (CBNC-EHR) service that is in practice for more than 13 years with Chronic Renal Insufficiency Cohort (CRIC) study which has approximately the same span and enrollment to find out about the feasibility, strengths, and challenges of the proposition. Method We compared measured variables of baseline characteristics and follow-up strategy of CRIC study with the data provided by CBNC-EHR to estimate if the care-centered database is capable of emulating a cohort study. Results Among variables of socioeconomic information at CRIC study, CBNC-EHR provides 33% at full, 33% are “partially available or easily accessible” (PAEA), and 33% are not available. 71% of variables of medical history are fully available for all patients, while only 14% are PAEA, and 14% are not available. For kidney function measures, CBNC-EHR provides 55% of variables completely, while the remaining 45% are PAEA. Obviously, no blood or urine sample were collected during ambulatory care, and no questionnaire on different aspects of health were used to gather supplementary data, contrary to CRIC study. Compared to follow-up strategy of CRIC study, which indicated Annual visits and laboratory tests and 6-month telephone follow-ups, CBNC-EHR yields data only upon personally-decided attention of patients for follow-up care, which varies widely from one patient to another. Nevertheless, the majority of patients have at least one office visit in a year, accompanied by relevant laboratory tests. CBNC-EHR provides a complete list of risk factors and comorbidities of each patient, which makes it suitable for subgrouping, as was designed in CRIC study regard the diabetes status of patients. CBNC-EHR currently has data of 6481 individuals with renal failure, compared to 3000 enrollment goal of CRIC study. However, much higher rates of dropout (estimated to be approximately 20%) and utilization of inclusion and exclusion criteria will contribute to the actual available cases to emulate a cohort study. Conclusion Implementing a care-centered database into emulation of a cohort study, is challenging and complicated. Lack of consistency, high dropout rates, imperfect baseline and complementary information, and need for extensive data cleaning are all difficulties. However, contribution to equitability of data share in evidence-synthesis, extremely lower cost of conduction, and less heterogeneity are strengths of this procedure which can have a remarkable effect on transforming medical data from less-developed countries (which understandably prioritize healthcare to research due to scarcity of resources) to effective research studies. In this paper, we compared a renowned nephrology cohort with a private nephrology clinic EHR in a developing country and showed virtual availability of data to emulate a cohort study based on the care-centered database.

Abstract 2551: Harness the power of real-world-data for oncology target and biomarker research: A case study by applying clinic-genomics data for 2,659 lung cancer patients and beyond

Abstract Randomized controlled trials (RCTs) are the gold standard for evaluating drug safety and efficacy in controlled settings, but they have limited generalizability due to their focus on selective and homogeneous populations. Real-world-data (RWD), collected from larger and more varied patient populations, can address inherent limitations of RCTs and aid in clinical trial design and drug approval. However, with the challenges of RWD, unlocking its full potential for tumor target and biomarker identification and validation demands innovative analytical approaches. Here we present RWD analyses of ConcertAI® electronic health records (EHR) linked with Caris Lab data for 2,659 lung cancer patients. Comprehensive genomics analyses incorporate WES and RNA-seq data, coupled with rich clinical EHR data. By integrating clinical and genomics data, we showcase the potential of harnessing RWD as follows: (1) we investigated genomic alterations across five histology groups (adenocarcinoma, squamous, large cell, small cell lung cancer [SCLC], neuroendocrine [NE] lung cancer). Notably, TP53 and LRP1B mutations were prevalent across all groups. LRP1B mutations correlated with higher tumor mutation burden, indicative of a favorable response to immunotherapy. Biallelic TP53 and RB1 inactivation was predominant in SCLC and NE lung tumors. The five histological groups exhibit distinct expression profiles, reflecting tumor intrinsic characteristics. Most immune gene signature scores were significantly lower in SCLC and NE tumors, aligned with that they are cold tumors with immune resistance, except CD56-centered NK and neuroendocrine gene signatures; (2) we analyzed patient treatment patterns and persistence by line of therapy. Among the 206 NSCLC patients with clinically actionable EGFR mutations, 143 received targeted therapy during their treatment courses (69.4%). Time to treatment discontinuation (TTD) was used as a surrogate clinical outcome endpoint. TTD was significantly longer when TKI was the first line treatment compared to later lines (p = 0.01), while the trend is opposite for ICI where TTD is significantly shorter in the 1st line setting (p = 0.025); (3) we highlight the unique potential of leveraging H&E imaging modality in such RWD. In summary, we successfully identified unique molecular signatures among various lung cancer histological groups, pinpointed distinct subpopulations within lung adenocarcinoma and squamous tumors linked to their immune phenotypes, and investigated the different patient treatment patterns and outcomes, facilitating personalized medicine. In the broader context of our work, we have developed a framework for harnessing various data modalities (clinical, multi-omics, imaging) within AbbVie's oncology RWD realm, enhancing the discovery and validation of new oncology targets and biomarkers. Citation Format: Si Wu, Weilong Zhao, Christy Choi, Mona Cai, Peter Ansell, Alexander Liede, Rong Chen, Xi Zhao, Josue Samayoa. Harness the power of real-world-data for oncology target and biomarker research: A case study by applying clinic-genomics data for 2,659 lung cancer patients and beyond [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 2551.

An age-friendly approach to primary care in an academic health system.

Age-friendly care, addressing what matters most, medications, mentation, and mobility, is a successful model for improving older adult care. We describe the initial outcomes of age-friendly care implementation in five primary care clinics in an academic health system. In partnership with a regional quality improvement (QI) organization, we used practice facilitation to implement age-friendly care from July 2020 to June 2023. Clinic workflows and electronic health record (EHR) templates were modified to capture six QI measures for patients ≥65 years: Documenting what matters most to patients Advance care planning (ACP) Annual cognitive screening Caregiver referral to dementia community resources Fall-risk screening Co-prescription of opioid and sedative-hypnotic drugs Providers were alerted if patients had positive screens and given support tools for clinical decision-making. QI measures from January-June 2023 were compared to the year prior to implementation. Providers and staff were interviewed about implementation barriers and facilitators. All six measures improved in Geriatrics and and other clinics showed improvement in ACP and cognitive screening. All clinics had high fall-risk screening rates (≥85%). The least improved measure was co-prescription of opioids and sedative-hypnotics with co-prescription rates ranging from 7% to 39%. Implementation hinged on leadership prioritization, practice facilitator guidance, clinical team buy-in, EHR functionality, and clinical performance review. Three clinics received Age-Friendly Health System recognition. A QI approach using practice facilitation and EHR templates improved some but not all age-friendly care measures. Future interventions will focus on training in high-risk medication tapering and elicitation of health goals.

Strategies to Increase Cervical Cancer Screening With Mailed Human Papillomavirus Self-Sampling Kits: A Randomized Clinical Trial

ABSTRACT In the United States, adherence for cervical cancer screening decreased from 86% to 77% between 2005 and 2019. Over 50% of cervical cancers are diagnosed in individuals overdue for screening, and promoting access and adherence is essential. Unlike traditional Papanicolaou testing, human papillomavirus (HPV) screening can use self-collected samples, which have comparable sensitivity and specificity to clinician-collected samples. Self-collecting has been shown to increase screening adherence, possibly by circumventing important barriers such as scheduling clinic appointments and negativity about pelvic examinations. Despite this, most individuals in prior studies remained unscreened with low follow-up of HPV-positive results, and kit uptake among populations that are screening adherent or have unknown screening history is unknown. The STEP trial (self-testing options in the era for primary HPV screening for cervical cancer) was a parallel, single-blind, randomized clinical trial comparing cervical cancer screening completion across groups of individuals with due (screening-adherent), overdue, or unknown screening history. Participants were randomized to traditional care (patient reminders and clinician electronic health record [HER] alerts), education (usual care plus educational materials about screening), direct mail (traditional care plus educational materials plus a mailed HPV self-collection kit), or to opt in (traditional care plus educational materials plus the option to request a self-collection kit). Patients were identified using the Kaiser Permanente Washington HER and administrative claims. Eligible participants were females aged 30 to 64 years with current Kaiser Permanente Washington insurance and primary care, intact cervix, who were either due or overdue for screening. The primary study outcome was screening completion within 6 months after randomization, which could be done either in clinic or using the self-collection kit. Modified Poisson regression was used to estimate relative risk (RR) of screening completion for the direct-mail and opt-in groups, relative to the education group. A total of 32,771 participants were randomized between November 2020 and January 2022, of which 13,356 were due for screening, 8682 were overdue, and 10,733 had unknown screening history. Of those due for screening, the RR for screening completion was 1.30 (95% confidence interval [CI], 1.23–1.36; absolute difference, 14.1% [95% CI, 11.2%–16.9%]) for direct mail and 1.07 (95% CI, 1.02–1.12; absolute difference, 3.5% [95% CI, 1.2%–5.7%]) for the opt-in group compared with education. Of those overdue for screening, the RR for screening completion was 1.90 (95% CI, 1.68–2.16; absolute difference 16.9% [95% CI, 13.8%–20.0%]) for direct mail compared with education. Of those with unknown screening history, the RR for screening completion was 1.14 (95% CI, 1.03–1.25; absolute difference 2.2% [95% CI, 0.5%–3.9%]) for the opt-in group compared with education. The results of this randomized controlled trial showed that directly mailing HPV self-collection kits increased cervical cancer screening by over 14% compared with education alone. The effect was similar between participants that were due or overdue for screening, and an opt-in approach was found to minimally increase screening.

Automatic quantitative stroke severity assessment based on Chinese clinical named entity recognition with domain-adaptive pre-trained large language model

Background:Stroke is a prevalent disease with a significant global impact. Effective assessment of stroke severity is vital for an accurate diagnosis, appropriate treatment, and optimal clinical outcomes. The National Institutes of Health Stroke Scale (NIHSS) is a widely used scale for quantitatively assessing stroke severity. However, the current manual scoring of NIHSS is labor-intensive, time-consuming, and sometimes unreliable. Applying artificial intelligence (AI) techniques to automate the quantitative assessment of stroke on vast amounts of electronic health records (EHRs) has attracted much interest. Objective:This study aims to develop an automatic, quantitative stroke severity assessment framework through automating the entire NIHSS scoring process on Chinese clinical EHRs. Methods:Our approach consists of two major parts: Chinese clinical named entity recognition (CNER) with a domain-adaptive pre-trained large language model (LLM) and automated NIHSS scoring. To build a high-performing CNER model, we first construct a stroke-specific, densely annotated dataset “Chinese Stroke Clinical Records” (CSCR) from EHRs provided by our partner hospital, based on a stroke ontology that defines semantically related entities for stroke assessment. We then pre-train a Chinese clinical LLM coined “CliRoberta” through domain-adaptive transfer learning and construct a deep learning-based CNER model that can accurately extract entities directly from Chinese EHRs. Finally, an automated, end-to-end NIHSS scoring pipeline is proposed by mapping the extracted entities to relevant NIHSS items and values, to quantitatively assess the stroke severity. Results:Results obtained on a benchmark dataset CCKS2019 and our newly created CSCR dataset demonstrate the superior performance of our domain-adaptive pre-trained LLM and the CNER model, compared with the existing benchmark LLMs and CNER models. The high F1 score of 0.990 ensures the reliability of our model in accurately extracting the entities for the subsequent automatic NIHSS scoring. Subsequently, our automated, end-to-end NIHSS scoring approach achieved excellent inter-rater agreement (0.823) and intraclass consistency (0.986) with the ground truth and significantly reduced the processing time from minutes to a few seconds. Conclusion:Our proposed automatic and quantitative framework for assessing stroke severity demonstrates exceptional performance and reliability through directly scoring the NIHSS from diagnostic notes in Chinese clinical EHRs. Moreover, this study also contributes a new clinical dataset, a pre-trained clinical LLM, and an effective deep learning-based CNER model. The deployment of these advanced algorithms can improve the accuracy and efficiency of clinical assessment, and help improve the quality, affordability and productivity of healthcare services.

The relationship between perceptions of electronic health record usability and clinical importance of social and environmental determinants of health on provider documentation.

Social and environmental determinants of health (SEDH) data in the electronic health record (EHR) can be inaccurate and incomplete. Providers are in a unique position to impact this issue as they both obtain and enter this data, however, the variability in screening and documentation practices currently limits the ability to mobilize SEDH data for secondary uses. This study explores whether providers' perceptions of clinical importance of SEDH or EHR usability influenced data entry by analyzing two relationships: (1) provider charting behavior and clinical consideration of SEDH and (2) provider charting behavior and ease of EHR use in charting. We performed a cross-sectional study using an 11-question electronic survey to assess self-reported practices related to clinical consideration of SEDH elements, EHR usability and SEDH documentation of all staff physicians, identified using administrative listserves, at Penn State Health Hershey Medical Center during September to October 2021. A total of 201 physicians responded to and completed the survey out of a possible 2,478 identified staff physicians (8.1% response rate). A five-point Likert scale from "never" to "always" assessed charting behavior and clinical consideration. Responses were dichotomized as consistent/inconsistent and vital/not vital respectively. EHR usability was assessed as "yes" or "no" responses. Fisher's exact tests assessed the relationship between charting behavior and clinical consideration and to compare charting practices between different SEDHs. Cumulative measures were constructed for consistent charting and ease of charting. A generalized linear mixed model (GLMM) compared SDH and EDH with respect to each cumulative measure and was quantified using odds ratios (OR) and 95% confidence intervals (CI). Our results show that provider documentation frequency of an SEDH is associated with perceived clinical utility as well as ease of charting and that providers were more likely to consistently chart on SDH versus EDH. Nuances in these relationships did exist with one notable example comparing the results of smoking (SDH) to infectious disease outbreaks (EDH). Despite similar percentages of physicians reporting that both smoking and infectious disease outbreaks are vital to care, differences in charting consistency and ease of charting between these two were seen. Taken as a whole, our results suggest that SEDH quality optimization efforts cannot consider physician perceptions and EHR usability as siloed entities and that EHR design should not be the only target for intervention. The associations found in this study provide a starting point to understand the complexity in how clinical utility and EHR usability influence charting consistency of each SEDH element, however, further research is needed to understand how these relationships intersect at various levels in the SEDH data optimization process.

Artificial intelligence and machine learning technologies in ulcerative colitis.

Interest in artificial intelligence (AI) applications for ulcerative colitis (UC) has grown tremendously in recent years. In the past 5 years, there have been over 80 studies focused on machine learning (ML) tools to address a wide range of clinical problems in UC, including diagnosis, prognosis, identification of new UC biomarkers, monitoring of disease activity, and prediction of complications. AI classifiers such as random forest, support vector machines, neural networks, and logistic regression models have been used to model UC clinical outcomes using molecular (transcriptomic) and clinical (electronic health record and laboratory) datasets with relatively high performance (accuracy, sensitivity, and specificity). Application of ML algorithms such as computer vision, guided image filtering, and convolutional neural networks have also been utilized to analyze large and high-dimensional imaging datasets such as endoscopic, histologic, and radiological images for UC diagnosis and prediction of complications (post-surgical complications, colorectal cancer). Incorporation of these ML tools to guide and optimize UC clinical practice is promising but will require large, high-quality validation studies that overcome the risk of bias as well as consider cost-effectiveness compared to standard of care.

Electronic health record modification and dashboard development to improve clinical care in pediatric rheumatology.

This report describes our experience in electronic health record (EHR) note modification and creation of an external dashboard to create a local learning health system that contributes to quality improvement and patient care within our pediatric rheumatology clinic. We applied quality improvement methodology to develop a more reliable and accurate system to identify patients with juvenile idiopathic arthritis and track important measures that aide in improving patient care and performance outcomes. From 2019 to 2021, we iteratively modified our outpatient clinic EHR note to include structured data elements to improve longitudinal monitoring. We then validated data transferred to an electronic dashboard external to the EHR and demonstrated utility for identifying an accurate patient population and tracking quality improvement initiatives. Creation of the structured data elements improved the identification of patients with JIA with >99% accuracy and without requiring manual review of the chart. Using the dashboard to monitor performance, we improved documentation of critical disease activity measures that resulted in improvement in those scores across the local population of patients with JIA. The structured data elements also enabled us to automate electronic data transfer to a multicenter learning network registry. The structured data element modifications made to our outpatient EHR note populate a local dashboard that allows real time access to critical information for patient care, population management, and improvement in quality metrics. The collection and monitoring of structured data can be scaled to other quality improvement initiatives in our clinic and shared with other centers.

Identifying incarceration status in the electronic health record using large language models in emergency department settings.

Incarceration is a significant social determinant of health, contributing to high morbidity, mortality, and racialized health inequities. However, incarceration status is largely invisible to health services research due to inadequate clinical electronic health record (EHR) capture. This study aims to develop, train, and validate natural language processing (NLP) techniques to more effectively identify incarceration status in the EHR. The study population consisted of adult patients (≥ 18 y.o.) who presented to the emergency department between June 2013 and August 2021. The EHR database was filtered for notes for specific incarceration-related terms, and then a random selection of 1,000 notes was annotated for incarceration and further stratified into specific statuses of prior history, recent, and current incarceration. For NLP model development, 80% of the notes were used to train the Longformer-based and RoBERTa algorithms. The remaining 20% of the notes underwent analysis with GPT-4. There were 849 unique patients across 989 visits in the 1000 annotated notes. Manual annotation revealed that 559 of 1000 notes (55.9%) contained evidence of incarceration history. ICD-10 code (sensitivity: 4.8%, specificity: 99.1%, F1-score: 0.09) demonstrated inferior performance to RoBERTa NLP (sensitivity: 78.6%, specificity: 73.3%, F1-score: 0.79), Longformer NLP (sensitivity: 94.6%, specificity: 87.5%, F1-score: 0.93), and GPT-4 (sensitivity: 100%, specificity: 61.1%, F1-score: 0.86). Our advanced NLP models demonstrate a high degree of accuracy in identifying incarceration status from clinical notes. Further research is needed to explore their scaled implementation in population health initiatives and assess their potential to mitigate health disparities through tailored system interventions.

Analysis of the Larissa Heart Failure Risk Score: Predictive Value in 9207 Patients Hospitalized for Heart Failure from a Single Center.

Early risk stratification is of outmost clinical importance in hospitalized patients with heart failure (HHF). We examined the predictive value of the Larissa Heart Failure Risk Score (LHFRS) in a large population of HHF patients from the Cleveland Clinic. A total of 13,309 admissions for heart failure (HF) from 9207 unique patients were extracted from the Cleveland Clinic's electronic health record system. For each admission, components of the 3-variable simple LHFRS were obtained, including hypertension history, myocardial infarction history, and red blood cell distribution width (RDW) ≥ 15%. The primary outcome was a HF readmission and/or all-cause mortality at one year, and the secondary outcome was all-cause mortality at one year of discharge. For both outcomes, all variables were statistically significant, and the Kaplan-Meier curves were well-separated and in a consistent order (Log-rank test p-value < 0.001). Higher LHFRS values were found to be strongly related to patients experiencing an event, showing a clear association of LHFRS with this study outcomes. The bootstrapped-validated area under the curve (AUC) for the logistic regression model for each outcome revealed a C-index of 0.64 both for the primary and secondary outcomes, respectively. LHFRS is a simple risk model and can be utilized as a basis for risk stratification in patients hospitalized for HF.

CancelRx case study: implications for clinic and community pharmacy work systems

BackgroundMedication prescribing and discontinuation processes are complex and involve the patient, numerous health care professionals, organizations, health information technology (IT). CancelRx is a health IT that automatically communicates medication discontinuations from the clinic electronic health record to the community pharmacy dispensing platform, theoretically improving communication. CancelRx was implemented across a Midwest academic health system in October 2017. The health system also operates 15 outpatient community pharmacies.ObjectiveThe goal of this qualitative study was to describe how both the clinic and community pharmacy work systems change and interact over time regarding medication discontinuations, before and after CancelRx implantation.ApproachMedical Assistants (n = 9), Community Pharmacists (n = 12), and Pharmacy Administrators (n = 3), employed by the health system were interviewed across 3-time periods between 2017 and 2018— 3-months prior to CancelRx implementation, 3-months after CancelRx implementation, and 9-months after CancelRx implementation. Interviews were audio recorded, transcribed, and conducted a hybrid analysis with deductive content analysis following the Systems Engineering Initiative for Patient Safety (SEIPS) framework and inductive analysis to capture additional codes and themes.Key resultsCancelRx changed the medication discontinuation process at both clinics and community pharmacies. In the clinics, the workflows and medication discontinuation tasks changed over time while MA roles and clinic staff communication practices remained variable. In the pharmacy, CancelRx automated and streamlined how medication discontinuation messages were received and processed, but also increased workload for the pharmacists and introduced new errors.ConclusionsThis study utilizes a systems approach to assess disparate systems within a patient network. Future studies may consider health IT implications for systems that are not in the same health system as well as assessing the role of implementation decisions on health IT use and dissemination.

Prevalence, Incidence, and Characterization of LIGHT Chain Amyloidosis in the USA: A Real-World Analysis Utilizing Electronic Health Records (EHR)

Background: Light chain (AL) amyloidosis is a rare disease characterized by the deposition of misfolded immunoglobulin light chains, resulting in organ dysfunction and significant morbidity and mortality. However, population-based studies on the epidemiology of AL are limited. This study aimed to provide an up-to-date assessment of the prevalence and incidence of AL, examine temporal changes in prevalence over a 5-year period, and characterize the AL population in the USA. Methods: A retrospective analysis was conducted using de-identified patient records from Optum's Clinical Electronic Health Record (EHR) database (&amp;gt;140,000 providers and &amp;gt;60 million unique patients). Adults with AL were identified based on ≥2 diagnosis codes (ICD-10-CM code E85.81) or positive mentions about the disease (including clinical features indicative of AL) found in the physician's notes (≥30 days apart). Patients with clinical data at least 12 months pre- index event (defined as the date of the first AL amyloidosis diagnosis or positive mention during the identification period: January 01, 2017 - December 31, 2021) were included in the analysis. Prevalence estimates were calculated for each year between 2017 and 2021, and annual percentage changes were assessed to determine temporal trends. Annual incidence was calculated as the number of newly diagnosed AL cases occurring each year between 2019 and 2021. Additionally, patient demographics, comorbidities, and Mayo 2004 staging with European modification (based on availability of the cardiac biomarkers: NT-proBNP and troponin) were characterized at index date or in the 12-month post index period. Results: A total of 1471 AL patients were included in the study population. The mean age of the prevalent patients was 67.2 years, with 54.5% being male. Most patients were White (78.5%), followed by Black (14.5%), Hispanic (3.3%), Asian (1.2%), and other/unknown (2.6%). The diagnosed prevalence of AL substantially increased from 22.7 to 69.1 cases per million between 2017 and 2021, with an annual growth rate of 32.1%. In 2021, the annual incidence of AL was estimated to be 16.7 patients per million in the general population, with an annual growth rate of 7.7% since 2019, at which time incidence was 14.4 patients per million. The distribution of patients by Mayo stage at diagnosis among those with available biomarker information (n=613, 42% of total sample) was: Stage I, 27%; Stage II, 42%; Stage IIIa, 16%; and Stage IIIb, 15%. The most prevalent comorbidities reported in the year after diagnosis were hypertension (56.6%), dyslipidaemia (51.7%), chronic kidney disease (41.3%), multiple myeloma and malignant plasma cell neoplasms (37.5%), and heart failure (34.5%). Cardiomyopathy was observed in 40.2% of the AL patients. Conclusions: This study provides an updated report of the prevalence and incidence of AL using a large-scale EHR database. The observed rise in AL prevalence and incidence is likely due to a combination of increased disease awareness and improvements in detection and management. The number of comorbidities, especially those related to cardiac function, suggest a continuing unmet need for more therapeutic options. Furthermore, the Mayo stage distribution of patients highlights the need for enhanced efforts to diagnose AL earlier due to the high mortality associated with advanced Mayo stages. These epidemiological findings reinforce the importance of continuous monitoring of prevalence, incidence trends, and the burden of AL in the US.