Health Records Research Articles

Predicting adverse drug event using machine learning based on electronic health records: a systematic review and meta-analysis

IntroductionAdverse drug events (ADEs) pose a significant challenge in current clinical practice. Machine learning (ML) has been increasingly used to predict specific ADEs using electronic health record (EHR) data. This systematic review provides a comprehensive overview of the application of ML in predicting specific ADEs based on EHR data.MethodsA systematic search of PubMed, Web of Science, Embase, and IEEE Xplore was conducted to identify relevant articles published from the inception to 20 May 2024. Studies that developed ML models for predicting specific ADEs or ADEs associated with particular drugs were included using EHR data.ResultsA total of 59 studies met the inclusion criteria, covering 15 drugs and 15 ADEs. In total, 38 machine learning algorithms were reported, with random forest (RF) being the most frequently used, followed by support vector machine (SVM), eXtreme gradient boosting (XGBoost), decision tree (DT), and light gradient boosting machine (LightGBM). The performance of the ML models was generally strong, with an average area under the curve (AUC) of 76.68% ± 10.73, accuracy of 76.00% ± 11.26, precision of 60.13% ± 24.81, sensitivity of 62.35% ± 20.19, specificity of 75.13% ± 16.60, and an F1 score of 52.60% ± 21.10. The combined sensitivity, specificity, diagnostic odds ratio (DOR), and AUC from the summary receiver operating characteristic (SROC) curve using a random effects model were 0.65 (95% CI: 0.65–0.66), 0.89 (95% CI: 0.89–0.90), 12.11 (95% CI: 8.17–17.95), and 0.8069, respectively. The risk factors associated with different drugs and ADEs varied.DiscussionFuture research should focus on improving standardization, conducting multicenter studies that incorporate diverse data types, and evaluating the impact of artificial intelligence predictive models in real-world clinical settings.Systematic Review Registrationhttps://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42024565842, identifier CRD42024565842.

Laparoscopic transabdominal vs. totally extraperitoneal inguinal hernia repair: outcomes, healthcare utilization, and cost differences.

Laparoscopic inguinal hernia repair utilizes either a transabdominal preperitoneal (TAPP) or totally extraperitoneal (TEP) approach. The literature remains mixed on outcomes comparing TAPP versus TEP. The objective of our study was to assess outcomes, healthcare utilization, and cost differences between TAPP and TEP inguinal hernia repair. Adult patients who underwent elective inguinal hernia repair between 2013 and 2021 were retrospectively identified from our institution's electronic health record. Baseline characteristics and postoperative complications were compared using chi-squared test. Multivariable logistic regression was used to model the odds of experiencing a postoperative emergency department visit, readmission, and/or reoperation within 6months were surgery. Generalized linear models were used to investigate differences in cost between TAPP and TEP groups. 1086 patients underwent TAPP repair, while 1277 patients underwent TEP repair. TAPP patients had more than double the rates of readmissions (3.1% vs. 1.3%, p = 0.002) and reoperations (1.5% vs. 0.2%, p = 0.001) within 6months of surgery. On multivariable analysis, undergoing TAPP inguinal hernia repair was associated with higher odds of reoperations and/or readmissions within 6months of surgery (OR 2.8, CI [1.5, 5.1], p = 0.001). TAPP repair had a higher index surgery cost and higher costs associated with reoperations and readmissions compared to TEP repair. Although both approaches are very safe, TAPP inguinal hernia repair had higher rates of postoperative complications, increased healthcare utilization, and higher associated costs. Future studies should be directed toward decreasing postoperative healthcare utilization in order to decrease costs in inguinal hernia repair.

Learning health system linchpins: information exchange and a common data model.

To demonstrate the potential for a centrally managed health information exchange standardized to a common data model (HIE-CDM) to facilitate semantic data flow needed to support a learning health system (LHS). The Rhode Island Quality Institute operates the Rhode Island (RI) statewide HIE, which aggregates RI health data for more than half of the state's population from 47 data partners. We standardized HIE data to the Observational Medical Outcomes Partnership (OMOP) CDM. Atherosclerotic cardiovascular disease (ASCVD) risk and primary prevention practices were selected to demonstrate LHS semantic data flow from 2013 to 2023. We calculated longitudinal 10-year ASCVD risk on 62,999 individuals. Nearly two-thirds had ASCVD risk factors from more than one data partner. This enabled granular tracking of individual ASCVD risk, primary prevention (ie, statin therapy), and incident disease. The population was on statins for fewer than half of the guideline-recommended days. We also found that individuals receiving care at Federally Qualified Health Centers were more likely to have unfavorable ASCVD risk profiles and more likely to be on statins. CDM transformation reduced data heterogeneity through a unified health record that adheres to defined terminologies per OMOP domain. We demonstrated the potential for an HIE-CDM to enable observational population health research. We also showed how to leverage existing health information technology infrastructure and health data best practices to break down LHS barriers. HIE-CDM facilitates knowledge curation and health system intervention development at the individual, health system, and population levels.

How do patients and healthcare professionals experience foot examinations in diabetes care? - A randomised controlled study of digital foot examinations versus traditional foot examinations.

Digital solutions in healthcare can facilitate and improve care. However, the experiences and the usefulness of using either digital foot examinations or traditional foot examinations need to be evaluated. The aims of the study were to evaluate: 1) The differences in patient experiences, having their foot examined supported by the Clinical Decision Support System as compared with having their foot examined in traditional practice, 2) How healthcare professionals, by using the digital tool, experienced the routine compared with performing the foot examination as in traditional practice. Of a total of 141 patients, 100 patients with diabetes were single-blind digitally randomised to one of two parallel arms: having their foot examined by a healthcare professional using a digital tool (n = 47) or having their foot examined as in traditional practice (n = 53) at the Department of Prosthetics and Orthotics at Sahlgrenska University Hospital, Gothenburg, Sweden. Patients filled in a modified version of the National Patient Survey and the Orthotics and Prosthetics Users' Survey at study end. Two healthcare professionals, working at a Department of Prosthetics and Orthotics, answered surveys regarding the interaction between the patient and the certified prosthetist and orthotist. Patients, aged 66 ± 13 years, perceived a high level of satisfaction with the service at the department, regardless of the method used. No significant differences between groups were found when evaluated by 27 questions in the National Patient Survey or by the Orthotics and Prosthetics Users' Survey, with scores of 67.17 ± 12.18 vs. 66.35 ± 16.52 (p = 0.78) for the intervention and control group respectively. For the same patient that healthcare professionals foot examined, the risk class was fully obtained when the risk to develop foot ulcers was assessed by using the digital tool, whereas only 2% of the patients were classified when foot assessed in traditional practice. Regardless of the method used for the foot examination, patients perceived a high level of satisfaction with the services at the Department of Prosthetics and Orthotics. All the patients were risk classified in the intervention group. The healthcare professionals found that, by using the Clinical Decision Support System, the foot examination was structured and followed clinical guidelines. Furthermore, the documentation in the electronic health record was thorough, even though further improvements, such as integration with co-existing health record systems, were requested. Clinical Trials NCT03088566 , Registered 23 March 2017.

Abstract 4135624: A Hypertrophic Cardiomyopathy Polygenic Score Modifies Penetrance of Pathogenic Hypertrophic and Dilated Cardiomyopathy Variants in Opposite Directions

Background: Hypertrophic (HCM) and dilated (DCM) cardiomyopathy have been postulated to represent morphologically opposing cardiac disease traits, with genetic pathways implicated in HCM demonstrating inverse relationships with DCM. The degree to which polygenic risk for HCM protects from DCM in individuals at genetic risk of DCM remains unexplored. Research Question: Is polygenic risk for HCM protective against developing DCM in carriers of titin truncating variants? Methods: Associations between polygenic risk for HCM, monogenic variants, and cardiomyopathy outcomes were modeled among Penn Medicine BioBank (PMBB) participants genetically similar to European reference populations using logistic regression. Polygenic risk for HCM was determined using a previously published HCM polygenic score (PGS; PGS Catalog PGS000739; Harper et al. 2021); scores were PCA adjusted, and the mean and variance were normalized using pgsc_calc. Pathogenic variants in definitive HCM genes were identified using ACMG criteria. Titin truncating variants in high percentage spliced in exons (TTNtv) were considered pathogenic. HCM and DCM outcomes were assigned based on diagnosis codes +/- clinical notes in the electronic health record. Results: Of 26,728 PMBB participants, 205 (0.77%) had HCM and 589 (2.2%) had DCM. 289 had a TTNtv, and 144 had a pathogenic HCM variant. Pathogenic HCM variant status (n=144) conferred a 76-fold increase in HCM risk (p=1.8 e-87). A one standard deviation change in HCM PGS had a significant positive association with HCM (OR 1.6; p=1.5e-9) and negative association with DCM (OR 0.72; p=1.e-12). TTNtv status (n=289) was associated with an increased risk of DCM (OR 5.0; p=1.4e-12). An interaction term between TTNtv status and HCM PGS had a negative association with DCM (OR 0.61; p=4.5e-2). Among the top HCM PGS quintile, there was no statistically significant difference in DCM risk between those with and without TTNtv (Figure 1). Conclusions: These results provide additional evidence that HCM and DCM are influenced by polygenic background. We observe that polygenic risk of HCM additively increases risk of HCM among pathogenic HCM variant carriers and may synergistically protect from DCM among TTNtv carriers.

Specialty Tobacco Treatment Implementation in Oncology: A Qualitative Study.

In 2017, the National Cancer Institute (NCI) funded the Cancer Center Cessation Initiative (C3I) to implement and expand tobacco treatment programs in routine oncology care. Many C3I programs developed specialty care programs staffed by tobacco treatment specialists (TTSs) to deliver evidence-based treatment to adult patients who smoke. People involved in specialty tobacco treatment programs can help to identify implementation strategies and adaptations that may enhance tobacco treatment reach and effectiveness in cancer care and help more patients with cancer quit using tobacco. We conducted semistructured interviews with TTSs from 21 C3I-funded cancer centers and applied content analysis to interview transcripts from 37 TTSs and 17 respondents in other program roles. We used the Consolidated Framework for Implementation Research to code interview data. We identified final themes and implementation strategies and adaptations recommended by respondents on the basis of these codes. Respondents shared that implementation of specialty tobacco treatment programs in cancer care settings could be facilitated by training staff to provide patient connection to services, incorporating prescription of no- or low-cost cessation medications, hiring additional staff to deliver tobacco treatment, allocating space to the program, and automating electronic health record workflows. TTSs and others involved in specialty tobacco treatment in NCI-designated cancer centers identified ways to improve tobacco treatment access, use, and effectiveness by (1) adapting specialty tobacco treatment delivery to meet patient needs; (2) facilitating referrals and pharmacotherapy coordination; and (3) committing staffing, space, and support resources to tobacco treatment programs. Key program participants suggested that these approaches would help more oncology patients connect with evidence-based tobacco treatment and quit smoking.

Abstract 4138564: Fitbit-Measured Physical Activity Is Inversely Associated With Incident Atrial Fibrillation Among All Of Us Participants

Background: Studies show that moderate to vigorous physical activity (MVPA) reduces atrial fibrillation (AF) risk. However, these studies used subjective questionnaires or short-term (<1 week) accelerometry assessments. Objective: This study investigated the link between MVPA levels and AF incidence using long-term (1-year) accelerometry data from the NIH All of Us Fitbit database. Methods: The study included All of Us participants who agreed to share electronic health record (EHR) and Fitbit data between January 1, 2016 to July 1, 2022. Among participants without an existing diagnosis of AF, the average weekly minutes of MVPA over the first year of Fitbit use was computed. Incident AF events were identified using validated ICD-10 codes in a five-year follow-up period. The risk of incident AF was evaluated using Cox proportional hazards regression models, for both MVPA as a continuous and a categorical variable (<30, 30-150,151-300, and >300 minutes MVPA per week; aligned with WHO guidelines). Covariates included age, sex, race/ethnicity, BMI, smoking status, and diagnoses of hypertension and diabetes. In a subset of participants with whole genome data, we evaluated if the association between MVPA and AF incidence varied by a participant’s AF genetic risk score (GRS) Results: 6086 participants (mean age 50 years, 71% female, 83% White) had EHR and at least 1 year of Fitbit data. Over the 1-year MVPA assessment period, participants had 41±12 weeks of complete data and a median 202 [241] minutes of MVPA per week. 55 participants (0.9%) experienced incident AF in the five-year follow-up period. In adjusted analysis, every additional hour of MVPA was associated with an 11% reduced risk of AF (HR = 0.89 [0.81,0.98], p = 0.02). There was a step-wise reduction in risk of AF by MVPA category with <30 min/week as the reference: 30-150 (HR = 0.62 [0.24,1.6], p = 0.31), 151-300 (HR = 0.40 [0.15,1.07], p = 0.07), and >300 (HR = 0.35 [0.13,0.95], p = 0.04) ( Figure 1 ). These associations also held after adjusting for GRS. Conclusion: Higher amounts of objectively measured MVPA were inversely associated with risk of incident AF, independent of clinical and genetic risk factors.

Supporting long-term condition management: a workflow framework for the co-development and operationalization of machine learning models using electronic health record data insights

The prevalence of long-term conditions such as cardiovascular disease, chronic obstructive pulmonary disease (COPD), asthma, and diabetes mellitus is rising. These conditions are leading sources of premature mortality, hospital admission, and healthcare expenditure. Machine learning approaches to improve the management of these conditions have been widely explored, with data-driven insights demonstrating the potential to support earlier diagnosis, triage, and treatment selection. The translation of this research into tools used in live clinical practice has however been limited, with many projects lacking clinical involvement and planning beyond the initial model development stage. To support the move toward a more coordinated and collaborative working process from concept to investigative use in a live clinical environment, we present a multistage workflow framework for the co-development and operationalization of machine learning models which use routine clinical data derived from electronic health records. The approach outlined in this framework has been informed by our multidisciplinary team’s experience of co-developing and operationalizing risk prediction models for COPD within NHS Greater Glasgow & Clyde. In this paper, we provide a detailed overview of this framework, alongside a description of the development and operationalization of two of these risk-prediction models as case studies of this approach.

Relationships Among Physician Vendor-Derived Proficiency Score, Gender, and Time in the Electronic Health Record

Background and Objectives: Electronic health record (EHR) customization is proposed to mitigate EHR-related burnout. Gender disparities in EHR usage are established, though less is known regarding differences in customization and its impact on EHR time. This study examined gender differences in vendor-derived proficiency score (PS) and its relationship to EHR time. Methods: This was a retrospective observational study of ambulatory EHR use for adult primary care and medical subspecialty physicians at an academic safety-net health care system. The EHR vendor provided a physician PS (0–10), derived from customization and efficiency tool utilization. Primary outcomes were PS, time in system per day, and time in system per appointment stratified by gender. We used multiple variable linear regression to determine whether gender differences persisted with the inclusion of other factors. Results: A total of 228 physicians were included in the study; 122 were women, and 106 were men. Women had higher median PS (7.6 vs 6.6, P=.021) and EHR time per day (150.5 vs 119.9 minutes, P=.013), but no difference in time per appointment (24.7 vs 26.1 minutes, P=.665). After adjusting for potential confounders, gender remained a significant predictor of PS, but not time in EHR. Higher PS was significantly associated with greater time in the system per appointment, but not per day. Conclusions: While women had higher PS than men, gender was not significantly associated with measures of EHR time after adjusting for potential confounders. Higher PS was associated with greater time in the EHR per appointment, suggesting factors that influence EHR time are complex and varied.

Abstract 4136136: Left Ventricular Global Longitudinal Strain: An Imaging Marker Associated with Improved Survival in Paradoxical Low-Flow, Low-Gradient Severe Aortic Stenosis

Background: The optimal clinical management and timing of intervention are less well defined in paradoxical low-flow, low-gradient severe aortic stenosis (PLFLG AS). Left ventricular global longitudinal strain (LV-GLS) has been shown to predict outcomes in high flow severe AS, but there is lack of data in patients with PLFLG AS. Given the exaggerated LV hypertrophy and remodeling pattern in PLFLG AS, LV-GLS may be a mechanistic imaging marker for worse outcomes. Hypothesis: In patients with PLFLG AS, LV-GLS is associated with adverse clinical outcomes by detecting subclinical myocardial fibrosis resulting from myocardial remodeling due to LV pressure overload. Methods: We examined patients with PLFLG AS defined as AVA <1cm 2 , mean gradient <40mmHg and preserved left ventricular ejection fraction (LVEF) ≥50%, with low-flow state defined as transvalvular flow rate (Q) ≤210 ml/sec. Exclusion criteria included moderate or greater mitral or aortic regurgitation and presence of atrial fibrillation at the time of echocardiogram. LV strain analysis was performed using 2D-strain imaging software (TomTec, Chicago, IL). Clinical outcome data were obtained via the electronic health record. The primary outcome was defined as all-cause mortality. Patients were stratified by LV-GLS above and below the median value. Results: A total of 198 patients were included in the analysis, with a median LV-GLS of -12.3% [IQR: (-9.8,-16.3)]. Over a median follow-up time of 1.2 years, 130 deaths (65.6%) were identified. Patients with less negative LV-GLS (n=101) had worse survival than those with more negative LV-GLS (n=97; 22 vs 38%; p=0.02) (Fig 1). After multivariable adjustment for potential confounders (stroke volume index, aortic valve mean gradient, relative wall thickness, age, sex, heart failure, hypertension, coronary artery disease, and diabetes), less negative LV-GLS was independently associated with adverse outcomes (HR 1.56; 95% CI (1.03-2.38); p=0.04). Conclusions: In patients with PLFLG AS as defined by a transvalvular flow rate (Q) ≤210 ml/sec, reduced LV-GLS is associated with increased all-cause mortality. LV-GLS may serve as an imaging marker to help guide clinical decision making in PLFLG AS.

Abstract 4144549: Proof of Concept of a Novel Culinary Medicine Service Line to Address Diet-Related Cardiovascular Disease Risk Factors

Background: Poor diet has emerged as a top contributor to cardiovascular disease (CVD) in the United States. Evidence-based dietary strategies have been shown to decrease CVD risk by addressing comorbid conditions such as hypertension and diabetes, and there is a growing demand for practical nutrition guidance from health providers. However, implementation of nutrition support in healthcare settings remains limited due to lack of time, expertise, and reimbursement, necessitating creative interprofessional collaboration. Culinary Medicine (CM) combines medical nutrition education with culinary literacy and holds promise to address these challenges. Our study reports on the cascade implementation of a 3-prong CM service line which includes a CM eConsult, 1:1 CM specialty clinic, and shared medical appointment (group visit) in a teaching kitchen at a large, academic medical center from 2021 to 2024. Methods: A CM-specific eConsult was designed and launched on August 1 st 2021 utilizing existing eConsult workflows in the electronic health record. Next, an in-person, interprofessional CM specialty clinic launched at a community-based satellite clinic December 4 th 2022 open to referrals from physicians and patients. In April 2024, established patients in the CM clinic were invited to participate in six, biweekly cooking classes billed as shared medical appointments. A Microsoft PowerBI dashboard was developed to track patient demographics, referrals and participation, associated diagnoses, and billing data for the service line for continuous quality improvement. Results: In the first year of eConsults, 25 eConsults were completed with 19 (76%) reimbursed by insurance. In the first year from launch of the CM specialty clinic, 96 patients completed initial visits with a 100% insurance collections rate. In the recent launch of shared medical appointments, the initial cohort included 17 unique patients with 58% completing at least 4 of 6 classes. Referrals to the CM clinic included hypertension, hyperlipidemia, coronary artery disease, and type 2 diabetes as the top 4 associated diagnoses. Conclusions: We demonstrated proof-of-concept of a novel CM service line as an interprofessional, financially sustainable preventive strategy to target patients with diet-sensitive CVD risk factors such as diabetes, hypertension, and obesity. Future study will investigate the impact on biomarkers of CVD health for patients served by the CM service line.

Abstract Su706: New risk markers of sudden cardiac death in the young : insights from machine learning on EHR data

Background: Sudden Cardiac Death (SCD) in young adults is a significant public health issue, with survival rates often less than 15%. Effective prevention is critical as many SCD cases go undetected due to nonspecific symptoms and ineffective risk stratification methods, especially among the youth. Objective: To explore risk markers specific to SCD in younger populations using Artificial Intelligence (AI) trained on systematically collected Electronic Health Records (EHR) data, and to compare these markers with those in older populations. Methods: We conducted a retrospective cohort study using data from 46,060 participants in Paris and its inner suburbs, spanning 2011 to 2020. The cohort included 1,100 SCD cases and 1,100 controls aged 40 years or younger, and 21,930 SCD cases and 21,930 controls aged over 40. A machine learning model was developed, trained, and validated on these cohorts. We used the Catboost algorithm and evaluated model performance using AUC, sensitivity, and specificity. The SHAP algorithm was employed to elucidate the relationship between variables and predicted risk. Results: The model achieved on validation sets an AUC of 0.75 (95% CI 0.67-0.83) for the younger cohort and AUC of 0.82 (95% CI: 0.70-0.83), for the older cohort. Neuropsychiatric drugs were the most influential predictors for the younger cohort, whereas age and cardiovascular drugs were most predictive for the older cohort. The results indicate that traditional cardiovascular risk factors play a secondary role in predicting SCD among younger individuals. Conclusion: In analyzing predictors of sudden death in individuals under 40, we developed a personalized SCD prediction model using EHR data tailored to this demographic. Psychiatric medications, and possibly underlying psychiatric conditions, emerged as significant predictors. Our findings emphasize the need for a multidisciplinary approach to SCD prevention, integrating neuropsychiatric and cardiovascular considerations.

Abstract 4136273: A Machine Learning Approach to Simplify Risk Stratification of Patients with Atherosclerotic Cardiovascular Disease

Introduction: The 2018 AHA/ACC/Multisociety Blood Cholesterol Guideline recommends that patients with atherosclerotic cardiovascular disease (ASCVD) be classified as very high-risk (VHR) or not very high-risk (NVHR). This designation is based on the presence of > 2 major ASCVD events (recent acute coronary syndrome [ACS], history of myocardial infarction [MI], history of ischemic stroke, symptomatic peripheral artery disease [PAD]) or 1 major ASCVD event and > 2 high-risk conditions (age ≥65 years, heterozygous familial hypercholesterolemia, history of coronary revascularization, diabetes, hypertension, chronic kidney disease, current smoking, persistently elevated LDL-cholesterol [LDL-C], history of heart failure). While this approach was initially instituted to better predict future ASCVD events alone, more intensive LDL-C lowering is now recommended for those at VHR (LDL-C treatment threshold of 55 mg/dL for VHR patients vs 70 mg/dL for NVHR patients). We aimed to simplify ASCVD risk stratification in patients with clinical ASCVD using machine learning models. Methods: We used electronic health record data (Providence Health system) from 2022 to identify patients with ASCVD based on ICD-10 codes. Patient demographics, comorbidities, and laboratory data stratified patients as VHR or NVHR. Then, a classification and regression tree analysis was performed, with the outcome being VHR classification. Patients were randomly assigned to either the testing or training dataset in equal proportion. Variables in the model included age, sex, race, ethnicity, and the Guideline criteria. Model performance was assessed using misclassification rate and area under the receiver operative curve (AUC). Results: A total of 491,514 patients with ASCVD were included, of which 56% were male, 77% were white, and the mean (SD) age was 72.5 (13.0) years. Over half of patients (59%) were VHR. Those with recent ACS, history of MI, history of ischemic stroke, or PAD were classified as VHR 93%, 92%, 84%, and 91%, respectively. Presence of hypertension in these patients increased the likelihood of being VHR to 97%, 97%, 94%, and 98%, respectively. The misclassification rate and AUC for the testing set were 6.8% and 96.2%, respectively. Conclusions: Most patients with a major ASCVD event and hypertension were found to be VHR. Relying on these two factors alone can help simplify risk assessment. This approach may also help increase use of guideline-recommended LDL-C lowering therapy.

Abstract 4136203: Risk of Mortality and Heart Failure with Myosin Inhibition Among Hypertrophic Cardiomyopathy Patients

Background: While prior studies have established that mavacamten therapy in symptomatic obstructive hypertrophic cardiomyopathy (oHCM) leads to significant clinical and functional improvement, long-term data on the effect of mavacamten on mortality and heart failure with reduced ejection fraction (HFrEF) in these patients is lacking. Methods: In this retrospective cohort study utilizing electronic health record data from the TriNetX Network, we aimed to assess the association of mavacamten therapy with mortality and HFrEF compared to medical management among patients with oHCM between 04/2022 and 04/2024. The hazard ratio (HR) and 95% confidence interval (CI) for the study outcomes were estimated by comparing propensity-score matched groups. Results: Among 47,601 adults (45.4% females; mean age: 63.5±16.6 years), 701 and 46,900 patients were included in the mavacamten and medical management groups, respectively. ( Table ) In the propensity-matched analysis, mavacamten was associated with a lower risk of mortality (HR=0.40;95%CI:0.21-0.73,p=0.002), but a similar risk of incident HFrEF (HR=1.03;95%CI:0.56-1.88,p=0.93) compared with medical management. ( Figure ) Conclusions: The current study is among the first to report improved survival with mavacamten therapy among patients with oHCM utilizing real-world data. The similar risk of HFrEF between the mavacamten and medical management cohorts highlighted in this study indicates the efficacy of the Risk Evaluation and Mitigation Strategy (REMS) program in maintaining patient safety and offsetting any potential excess risk of HFrEF.

Machine learning-based infection diagnostic and prognostic models in post-acute care settings: a systematic review.

This study aims to (1) review machine learning (ML)-based models for early infection diagnostic and prognosis prediction in post-acute care (PAC) settings, (2) identify key risk predictors influencing infection-related outcomes, and (3) examine the quality and limitations of these models. PubMed, Web of Science, Scopus, IEEE Xplore, CINAHL, and ACM digital library were searched in February 2024. Eligible studies leveraged PAC data to develop and evaluate ML models for infection-related risks. Data extraction followed the CHARMS checklist. Quality appraisal followed the PROBAST tool. Data synthesis was guided by the socio-ecological conceptual framework. Thirteen studies were included, mainly focusing on respiratory infections and nursing homes. Most used regression models with structured electronic health record data. Since 2020, there has been a shift toward advanced ML algorithms and multimodal data, biosensors, and clinical notes being significant sources of unstructured data. Despite these advances, there is insufficient evidence to support performance improvements over traditional models. Individual-level risk predictors, like impaired cognition, declined function, and tachycardia, were commonly used, while contextual-level predictors were barely utilized, consequently limiting model fairness. Major sources of bias included lack of external validation, inadequate model calibration, and insufficient consideration of data complexity. Despite the growth of advanced modeling approaches in infection-related models in PAC settings, evidence supporting their superiority remains limited. Future research should leverage a socio-ecological lens for predictor selection and model construction, exploring optimal data modalities and ML model usage in PAC, while ensuring rigorous methodologies and fairness considerations.

Abstract 4137570: Atherosclerotic Cardiovascular Disease Risks in Children: An electronic health record data analytics approach

Background and Aims: There is little data on atherosclerotic cardiovascular disease (ASCVD) risk conditions, factors, and tiers in the pediatric population. This study aimed to develop and validate electronic health record (EHR) data analytic algorithms to attain the prevalence of 24 risk conditions, 7 factors, and 4 tiers in children in a rural community. Methods: This study included all the 495,902 subjects from 0 to 18 years at the Geisinger Medical Center from 2004 to 2022. The demographic and clinical data were collected using ICD-10 codes, clinical and laboratory data, and a dual-coded computational algorithm was developed based on the AHA guideline to derive the ASCVD risk conditions, factors, and tiers. To validate the EHR algorithms, the algorithm (rater 1) and a clinician (rater 2) extracted and analyzed the data from 40 randomly selected subjects from this cohort. Kendall’s coefficient of concordance (W) and Gwet’s agreement coefficient were used to evaluate algorithm and clinician agreement within the ASCVD tiers, conditions and factors, respectively. Results: High concordance was observed between algorithmic and clinician raters for risk tiers (W=80%), conditions (80-97%), and factors (49% for blood pressure, and 92% to 97% for the rest of the risk factors). In this cohort, the median age was 3.9 years, 51.0% were male, 80.6% were Non-Hispanic/Latino, and 88.1% were White. Overall, 15.5% (N = 76,945) were at-risk, 7.0% (N = 34,530) at moderate risk, and 4.2% (N = 20,761) at high risk for ASCVD, respectively. The most prevalent risk conditions were obesity (9.1%, N = 45,074), chronic inflammatory conditions (6.8%, N = 33,685), severe obesity (2.7%, N = 13,442), insulin resistance with comorbidities (2.7%, N = 13,397), and hypertension (2.1%, N = 10,436). The most prevalent risk factors were elevated BMI (27.4%, N = 135,904), blood pressure (12.5%, N = 62,015), and fasting lipid profile (8.7%, N = 43,100). Conclusion: Our EHR data analytics algorithm is in high agreement with the clinician rater. The high, moderate, and at-risk ASCVD risk tiers are prevalent in a large contemporary pediatric cohort at Geisinger. Both ASCVD risk conditions and factors are also prevalent, especially obesity, diabetes, hypertension, and hyperlipidemia. These findings suggest the further application and refinement of these algorithms to design innovative approaches for ASCVD risk reduction and prevention in children.

Evaluation of a clinical decision support alert to identify hepatic dysfunction and need for medication therapy adjustment in hospitalized patients.

In an effort to expedite the publication of articles, AJHP is posting manuscripts online as soon as possible after acceptance. Accepted manuscripts have been peer-reviewed and copyedited, but are posted online before technical formatting and author proofing. These manuscripts are not the final version of record and will be replaced with the final article (formatted per AJHP style and proofed by the authors) at a later time. To optimize the hepatic dysfunction alert tool at our institution to identify appropriate patients and minimize irrelevant alerts. This single-center, retrospective review included adults hospitalized over a 1-month period for whom a hepatic dysfunction alert fired for a medication order placed in the electronic health record. The existing alert determines hepatic dysfunction based on laboratory tests. The primary objective was to determine the proportion of patients with an alert that was deemed to be clinically relevant. Alerts were considered relevant if the patient had a Child-Pugh score in class B or C and were ordered a medication with a hepatic warning from FDA or LiverTox. The performance of 14 alternative models was evaluated. A total of 1,541 alerts fired for 309 patients. Of these patients, 155 were randomly selected for the analysis, and the alert was deemed relevant in 86 patients (55%). Patients with relevant alerts were more likely to have documented liver disease and worsening measures on liver function tests. Of the alternative models evaluated, a model that excluded INR and albumin resulted in a 27% decrease in the number of alerts fired, of which 73% were relevant; however, it failed to identify 30% of patients with relevant hepatic dysfunction. None of the other models performed better. The existing hepatic dysfunction clinical decision support tool correctly identifies patients with relevant hepatic dysfunction only 55% of the time. Alternative models were able to improve the rate of relevant results, but not without missing patients with relevant hepatic dysfunction.

Abstract 4124585: Identification of Low-Value Carotid Ultrasound Tests: A Comparison of Administrative Claims Versus Chart Review

Introduction/Background: Low-value care is defined as care where harms or costs outweigh patient benefits. Prior studies have suggested that approximately 1 of every 10 carotid ultrasound tests are low-value because they were performed among patients without neurologic symptoms or a history of stroke. However, these studies have used administrative claims data that may not account for all patient symptoms or include all guideline-recommended indications. We assessed the validity of a claims-based measure of low-value carotid ultrasound testing compared to a reference standard of chart review. Hypothesis: Claims-based measures overestimate the proportion of carotid ultrasound tests classified as low-value care. Methods: We conducted a retrospective cohort study of Medicare fee-for-service beneficiaries attributed to an accountable care organization with administrative claims data linked to electronic health record data. Using Current Procedural Technology codes, the first 150 patients that underwent carotid ultrasound in 2021 were identified. We applied a claims-based algorithm used by researchers and policymakers to identify if testing was low-value or appropriate. We used linked electronic health record data to assess appropriateness of testing using the 2012 American Heart Association clinical practice guidelines. We assessed agreement between claims and chart review using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and Cohen’s Kappa. Results: Among 150 patients (mean age 76, 55% male, 86% white) who underwent carotid ultrasound, 76 tests (51%) were considered low value by the claims measure and 63 (42%) by chart review. Claims based measures were 67% sensitive, 61% specific, with a PPV of 55% and NPV of 62%. Cohen’s Kappa coefficient was 0.27 (CI 0.11 - 0.42), indicating only fair agreement between claims and chart review. Conclusion: We found that an often-used claims-based measure to identify low-value carotid ultrasound testing misclassifies a substantial proportion of tests as low-value. Using claims-based measures alone for assessing provider quality and making payment decisions could harm provider reputations and erode trust in the quality reporting system.

Abstract 4121624: Ambulatory Inotropic Support in Patients with Stage D Heart Failure Does Not Increase ICD Therapies

Background: Ambulatory inotropic infusions can be used as a palliative therapy (PT) or bridge to advanced therapies in end-stage heart failure (ESHF) patients, many of whom have implantable cardioverter defibrillators (ICDs). Research Questions: How often do patients receive ICD therapies after initiation of outpatient inotropic support and how often is defibrillation inactivated? Methods: We conducted a retrospective review of all patients discharged from an academic tertiary medical center with a new prescription for inotropic therapy between November 2017 and March 2023. Clinical data, Palliative Care consultation, and ICD interrogations before and after inotrope initiation were extracted from the electronic health record. Results: The cohort comprised 45 patients discharged on new inotropic support, including 22 (49%) with intent of PT and 23 (51%) as a bridge to advanced therapies (Figure). Among PT patients, 15 (68%) had an ICD in place with only 4 having ICD therapies deactivated at discharge. Only 2 additional PT patients had a documented discussion regarding ongoing ICD therapies. By 1 year, 4 (18%) PT patients were transitioned to hospice and 13 (59%) had died, with only 1 additional patient having ICD deactivation. Of 28 patients discharged on inotropic infusions with ICD therapies on (both PT and bridge patients), only 4 patients had ICD therapies turned off later and no ICD shocks were reported for the year following initiation of inotropic support. Conclusion: In a series of 45 patients initiated on inotropic support for PT or bridge to advanced therapies, the majority (77%) had an active ICD, but therapies were rarely delivered. Notably, there were very few documented discussions regarding ICD therapies in PT patients even with specialty Palliative Care consultation. In patients with ESHF initiated on inotropic support as PT, most patients chose to leave ICD therapies in place. For those patients, there was a low rate of ICD therapies delivered.

Abstract Su903: Prognostic Value of Early Grey-White Ratio in the presence of Hypercapnia after Cardiac Arrest

Background: Cerebral edema following cardiac arrest can be quantified by measuring the ratio of grey matter to white matter radiodensity (GWR). Decreased GWR may reflect edema from irrecoverable neuronal cell death or reversible hyperemia driven by hypercapnia. We tested for an interaction between GWR and partial pressure of carbon dioxide (PCO2) predicting outcomes. Methods: We identified comatose post-arrest patients treated at our single center between 2010 and 2023 who underwent brain computed tomography (CT) imaging within 24 hours and PCO2 measured within one hour of CT. We extracted demographics, GWR, and PCO2 from our prospective registry and electronic health record. We scaled and centered data, then fit a multivariate logistic model to quantify the interaction between GWR and PCO2 in predicting survival. Results: Of 901 included patients, 356 (40%) were female, median [IQR] age was 61 [51-70], and 221 (25%) survived to discharge. CTs were obtained 4.18 [2.8-5.7] hours post-arrest. PCO2 was sampled within 36 median [24-47] minutes of CT. The median PCO2 was 47 [39-59] mmHg. GWR predicted survival (coefficient estimate 1.16, 95% confidence interval [CI], 0.91-1.44, p < 0.001, but PCO2 (-0.11, 95% CI -0.32-0.09, p = 0.29) did not. There was no significant interaction between GWR and PCO2 (0.009, 95% CI -0.24-0.29, p = 0.95). Conclusion: There is no evidence of PCO2 modifying the association of early GWR with survival after resuscitation from cardiac arrest.