Large Electronic Health Record Research Articles

Treatment Utilization Pattern of Preschool Children With Attention-Deficit/Hyperactivity Disorder

Objective: The aim of this study was to identify patterns of ADHD care, including factors that guide selection and sequencing of treatments in a large nationwide sample of preschool-aged youth over the past 6 years. Method: A retrospective cohort study utilizing a large electronic health record (TriNetX) of nearly 24,000 children ages 3 to 6 diagnosed with ADHD. Results: One in three preschoolers with ADHD were prescribed psychotropic medication, most commonly methylphenidate and guanfacine. One in 10 had at least one psychotherapy billing code during the entire assessment with most youth starting medication before psychotherapy. Rates of most treatments, including polypharmacy, increased with comorbid psychiatric disorders or sleep problems and over the course of the coronavirus pandemic. Conclusion: Rates of treatment have increased over time but are still largely inconsistent with published care guidelines that advise therapy before medication. Clinicians appear to prioritize psychiatric comorbidity and sleep problems when selecting treatments.

Machine‐learning based classification of Frontotemporal dementia in electronic health records for genetic discovery

AbstractBackgroundFrontotemporal dementia (FTD) is a neurodegenerative syndrome characterized by distinct subtypes, each with unique clinical presentations and varying neuropathologic features. Although FTD is highly heritable, it is frequently undiagnosed and genetic causes of FTD are rarely determined, limiting what is known about the medical phenome associated with FTD. Here, we perform a combination of innovative strategies to improve detection of FTD in a large electronic health record (EHR) using machine‐learning (ML) and uncover genetic factors for FTD.Method123 FTD cases and 615 matched controls were identified via diagnostic codes and confirmed by manual expert review of medical notes in Vanderbilt University Medical Center’s (VUMC) EHR. We then applied a feature selection algorithm which calculated the frequency for each phecode within a randomly generated and demographically matched control set. A p‐value denoting the enrichment of each phecode was calculated based on the observed phecode frequency relative to the permutation‐derived null distribution. 28 phecodes were enriched in FTD cases compared to controls. Top predictors included “Neurological disorders”, “Memory loss”, Aphasia”, and “Parkinson’s disorder”. We then developed a classification and regression tree model that leveraged enriched phecodes to label an undiagnosed case set in VUMC’s EHR‐linked DNA databank, BioVU. The model was trained on 96 cases and 481 controls, with 28 cases and 139 controls reserved for testing.ResultOf the testing set, 19 of the FTD cases were classified as cases and 5 were misclassified as controls. 135 controls were classified as such and 4 were misclassified as cases, indicating 82% PPV. Application of this model in BioVU resulted in an expanded case set of inferred FTD, which is currently being leveraged in genomic analyses to identify causal variants. We have identified a genomic segment on chromosome 1 (146.9MB‐147.3MB) spanning ACP6 and GJA5 that is shared in significant excess among FTD cases. These genes have been associated with intellectual disability, aphasia, and autism.ConclusionOur results demonstrate both the potential of machine learning‐based approaches for case acquisition in dementia subtypes that are often underdiagnosed in medical record databases and illustrate how inferred cases can be leveraged for novel gene discovery.

Occurrence of sleep disorders, anxiety, and depression among patients with mild cognitive impairment and Alzheimer’s disease in diverse populations: A large nationwide electronic health record cohort study

AbstractBackgroundPatients affected with mild cognitive impairment (MCI) and Alzheimer’s disease (AD) often experience anxiety, depression, and sleep disturbances. Some of the published studies reported a higher incidence of these disorders in minority populations. The objectives of this study were to characterize the occurrence of sleep disorders, anxiety, and depression among patients with MCI and/or AD, stratified by gender, age and race/ethnicity in the United States clinical practice setting.MethodA retrospective cohort study was conducted using electronic health records and insurance claims from the Optum® Market Clarity database containing data from more than 63 million people in the US insured through Medicare, Medicaid, and other insurance providers. The study included patients diagnosed, utilizing ICD diagnosis codes, with MCI and/or AD between January 2017 and September 2021. Prevalence of depression, anxiety, and sleep disorders, as well as associated medication prescriptions, were identified in the 12 months before and after MCI and/or AD diagnoses. The occurrence of these diagnoses and prescriptions was evaluated stratified by gender, age, and race/ethnicity.ResultA total of 197,346 patients were diagnosed with MCI (56.9% female, mean age of 70.0 [SD 14.8]) and 144,321 with AD (63.9% female, mean age of 79.3 [SD 8.2]) at the baseline. Overall, a large proportion of MCI and AD cases had depression (pre‐MCI diagnosis: 40.3%; post‐MCI diagnosis: 44.8%; pre‐AD diagnosis: 32.2%; post‐AD diagnosis: 38.3%), anxiety (pre‐MCI diagnosis: 36.2%; post‐MCI diagnosis: 40.2%; pre‐AD diagnosis: 26.4%; post‐AD diagnosis: 31.5%), and sleep disorders (pre‐MCI diagnosis: 36.2%; post‐MCI diagnosis: 39.3%; pre‐AD diagnosis: 23.2%; post‐AD diagnosis: 25.4%). Occurrences of depression and anxiety were highest among females. Among individuals with MCI and/or AD, the gap between the proportion experiencing depression and the proportion prescribed an antidepressant was greatest for African American males and females. This was not observed for anxiety and anti‐anxiety medications.ConclusionA large proportion of individuals with MCI and/or AD also had depression, anxiety, and sleep issues. Further understanding of the populations with these comorbidities and their associated prescriptions by different ages, sex, and races/ethnicities is important when supporting patients with MCI and AD.

Diagnosis patterns among diverse populations with mild cognitive impairment and Alzheimer’s disease: A large nationwide electronic health record cohort study

AbstractBackgroundWith recent advances and emerging research on diagnostic biomarkers and evolving diagnostic criteria, practitioners managing patients with cognitive complaints operate within a complex, changing environment. Although some evidence is available on the importance of prompt diagnosis, little information exists on the diagnostic process used in real‐world settings among diverse populations. This study’s objective was to characterize diagnosis patterns among adults with mild cognitive impairment (MCI), Alzheimer’s disease (AD), and/or dementia in the United States, and stratify by gender, age and race/ethnicity.MethodThis retrospective observational cohort study utilized the Optum® Market Clarity database from January 1, 2017–September 30, 2021. Eligible individuals had ≥1 diagnosis of MCI, AD, and/or dementia in their electronic health records or ≥2 diagnosis codes separated by ≥30 days in claims data, had ≥12 months of continuous enrollment before the cohort entry date, and had ≥1 day of continuous healthcare plan enrollment after the index date. Demographic information and the occurrence of diagnostic tests were also obtained from electronic medical records and insurance claims. Descriptive statistics were calculated and stratified by demographic factors.ResultA total of 179,419 newly diagnosed MCI cases, 81,267 AD cases, and 338,739 dementia cases were identified. The mean age at MCI, AD, or dementia diagnosis was similar overall for males and females. Among Caucasians, the mean age of MCI diagnosis was 70.5±14.8 for females and 69.8±14.5 for males. Among African Americans, the mean age at MCI diagnosis was 66.3±15.4 for females and 63.6±16.1 for males. PET scans and CSF biomarker screening occurred less often than MRI and CT (MCI: PET 0.6%, CSF 1.6%, MRI 16.6%, CT 17.5%; AD: PET 0.5%; CSF 0.9%; MRI 9.0%; CT 18.4%; dementia: PET 0.2%; CSF 1.6%; MRI 9.5%; CT 25.9%).ConclusionThe mean age at MCI, AD, and dementia diagnosis varied by demographic groups. The diagnosis of MCI, AD, and dementia remains largely clinical, with low use of brain imaging or CSF biomarkers. This knowledge of the patient populations and diagnostic experiences in a real‐world setting allows for greater understanding and growth to improve diagnosis across diverse groups and lead to faster support.

Drug utilization among patients with mild cognitive impairment and Alzheimer’s disease in diverse populations: A large nationwide electronic health record cohort study

AbstractBackgroundCholinesterase inhibitors (ChE‐Is), such as donepezil, rivastigmine, and galantamine, and N‐methyl‐d‐aspartate receptor antagonists, such as memantine, are prescribed to decrease symptoms associated with cognitive impairment. Few studies have examined the initiation and utilization of these medications among diverse populations. This study’s objective was to characterize drug utilization among patients with mild cognitive impairment (MCI) and/or Alzheimer’s disease (AD) in the United States and stratify by gender, age and race/ethnicity.MethodThis retrospective cohort study was conducted using the Optum® Market Clarity database, which contains electronic health records and insurance claims. Cases of MCI and AD were identified utilizing the International Classification of Diseases, Ninth and Tenth Revisions diagnosis codes. Drug prescriptions were determined to occur before, on, or after the date of first diagnosis. All cases were required to have ≥12 months of enrollment in the database prior to first diagnosis. The study period was from January 2017 through September 2021. Descriptive statistics were calculated stratified by demographic groups.ResultA large number of cases were identified (197,346 MCI and 144,321 AD cases). Approximately 6% of individuals with MCI had prescriptions for ChE‐Is prior to diagnosis and 14% after diagnosis. For memantine, the proportions with prescriptions before and after MCI diagnosis were 2% and 5%, respectively. Among individuals with AD, prescriptions for ChE‐Is occurred in 22% prior to diagnosis and 32% after diagnosis. Memantine prescriptions increased from 10% prior to AD diagnosis to 18% after. The results did not differ by gender. The highest proportions with ChE‐Is and memantine prescriptions were older age groups <85 years. When examining race/ethnic groups, the proportion of MCI and AD cases with these prescriptions was generally lowest among African Americans.ConclusionAs expected, prescriptions for ChE‐Is and memantine increased after diagnosis. The proportion of MCI and/or AD cases with prescriptions were similar across genders. The proportion with prescriptions was generally lowest among African Americans. Future research into the discrepancy in the occurrence of prescriptions by race/ethnicity as well as the timing of certain prescriptions will be valuable in further understanding the patient population and their access to appropriate medications.

Association of LDL cholesterol levels and all cause mortality in a real world analysis using data from an international network

Abstract Background Current cholesterol guidelines have recommended low low-density lipoprotein cholesterol (LDL-C) treatment targets for secondary prevention of cardiovascular disease. However, recent studies have suggested that very low or very high levels of LDL-C may be associated with increased mortality and other adverse outcomes. The association between LDL-C levels and mortality remains to be determined. Methods and Results This cohort study included data from a large multinational electronic health record network (TriNetX) of patients aged 18 years or older with coronary artery disease. LDL-C levels were divided into 4 categories: <30 (cohort 1), 30-50 (cohort 2), 50-70 (cohort 3), and 70-100 (cohort 4). Deaths were ascertained by linkage to death records in the last 10 years before the analysis. Five-year Kaplan-Meier curves and hazard ratios were calculated. We identified 1,136,600 patients, of whom 36,060 had LDL-C <30 mg/dL, 282,500 had LDL-C <50 mg/dL, 255,964 had LDL-C 50-70 mg/dL and 596,136 had LDL-C 70-100 mg/dL. Comparative analyzes were performed on the cohorts (cohort 1 vs. cohort 2, cohort 2 vs. cohort 3, and cohort 3 vs. cohort 4) after propensity score matching. After adjustment, individuals of cohort 2 (n = 36,004), compared with those of cohort 1 (n = 36,004), had higher all-cause mortality (HR 1.086; CI 95% 1.064-1.109, p <0.0001), individuals of cohort 3 (n = 244,908), compared to those of cohort 2 (n = 270,359), had higher risk of all-cause mortality (HR 1.063; 95% CI 1.055-1.1071, p <0.0001), and induviduals of cohort 4 (n = 254,465), compared with those of cohort 3 (n = 254,465), had lower risk of all-cause mortality (HR 0.937; CI 95% 0.929-0.944, p <0.0001). Conclusions Lower LDL-C levels are associated with a reduced risk of all-cause mortality in patients with coronary artery disease.

A Transformer Natural Language Processing Algorithm for Cancer Associated Thrombosis Phenotype

Introduction: Cancer patients have high risk of venous thromboembolism (VTE). The ability to accurately capture VTE outcome in large electronic health record (EHR) is critical for both surveillance and risk stratification. We previously published a rule-based natural language processing (NLP) algorithm that relied on radiology reports alone with a precision (positive predictive value) of 89% and recall (sensitivity) of 69-78% (sTable 5 PMID36626707). The low recall metric was expected as VTE diagnosis that occur at outside hospital would not be captured by radiology reports. In the current study, we aimed to assess the performance of a machine learning NLP algorithm using both radiology and medical notes to detect cancer associated VTE longitudinally in a large cancer cohort. Methods: VTEwasdefined as symptomatic or incidental pulmonary embolism, lower or upper extremity deep vein thrombosis. Unusual thrombosis such as splanchnic vein thrombosis (mostly tumor thrombi) was classified as negative. Figure 1 details the study design: i) patient selection, ii) pre-annotation filtering, iii) gold standard annotation, iv) transformer model development. From a cohort of 9,769 patients with active cancer receiving first-line systemic therapy, we first pre-filtered all clinic progress notes, discharge summaries, and radiology reports to exclude sections with little clinical relevance (e.g. review of systems, physical exam), while maintaining sections with high clinical value (e.g. history of present illness, assessment/plan, hospital problems/course). Then, we kept notes with VTE keyword stems (e.g. VTE, PE, DVT, thromb, embol, clot, filling defect, phlebitis). We selected 808 patients with at least 1 keyword to determine the gold standard VTE date and location through independent chart review process. This cohort was randomly split into a training and testing cohort at an 7:1 ratio. In the training cohort (n=708), we further annotated all keyword-containing phrases within 30 days (d) from VTE date and from index date until 180d if no VTE (n=~2,000). The ~800 positive and ~1,200 negative keyword-containing phrases were used to train ClinicalBERT (bidirectional encoder representation transformer) large language model through split-sample cross validation to finetune the loss function and to determine the prediction threshold. Finally, we applied the entire trained model package to every available notes/reports (n=~2300) in the test cohort (n=100) to assess if the first positively predicted date would correlate with the gold standard VTE date. Results: The finetuned ClinicalBERT modelachieved 98% precision and 98% recall on the patient level in the training cohort (n=708). When applied on the 100-patient test cohort, the model accurately predicted first VTE events in 50/54 patients (93% precision) and captured 50/54 of patients with VTE (93% recall sensitivity) (Table 1). Most of the tumor thrombi and superficial venous thrombi were correctly classified as negative. The false positives were related to unusual sentences with both positive and negative findings such as “Thrombosis surrounding the left brachial catheter in basilic vein,” “CT chest with PE was done,” “Lovenox was initiated for UE DVT, but this was port thrombosis in the basilic vein.” The false negatives were related to isolated VTE diagnosis close to death or loss to follow-up. Notably, the precision and recall tradeoff can be further finetuned by changing the minimum number of positively predicted notes from the NLP. Conclusion: In conclusion, we successfully developed and tested a transformer NLP model to detect cancer associated VTE longitudinally using a combination of clinical notes and radiology reports that achieved a precision of 93% and recall of 93%. This represented an improvement over our original algorithm with radiology report alone. Further validation of the model in external cohorts and finetuning through federated or transferred learning are ongoing to ensure model generalizability and usability.

The Use of a Large Electronic Health Records System to Define and Characterize a Thrombotic Thrombocytopenic Purpura Population

Background: Thrombotic thrombocytopenic purpura (TTP) is a rare life-threatening microangiopathy due to a deficiency of the von Willebrand factor (VWF)-cleaving protease ADAMTS13. Microangiopathic hemolytic anemia, severe thrombocytopenia and ischemic end organ injury due to microvascular thrombi are the defining features of TTP. TTP can be caused by auto-antibodies against ADAMTS13 enzyme (immune TTP [ITTP]) or by inherited mutations of the ADAMTS13 gene (congenital TTP [CTTP]). Therapeutic plasma exchange (TPE) to replenish ADAMTS13, with or without steroids, has been the mainstay of treatment. Due to the auto-immune nature of ITTP the International Society on Thrombosis and Hemostasis (ISTH) made a strong recommendation in 2020 to add corticosteroids to TPE for ITTP and a conditional recommendation to add the CD20 inhibitor rituximab as well the VWF monoclonal antibody caplacizumab. Objective: To define and characterize TTP in a living population from a large electronic health record (EHR) database in terms of disease prevalence, demographic characteristics and therapeutic interventions and procedures in light of recent ISTH guidelines. Methods: We conducted a study of data from TriNetX Analytics Network, which contains EHRs from >140 million individuals from 79 Health Care Organizations in the US. We used ICD-10-CM diagnosis codes, ATC and CPT codes applied in the last 10 years to evaluate TTP disease prevalence, therapeutic interventions and procedural interventions respectively. Eligibility criteria applied to define the TTP cohort were as follows: Inclusion criteria: ICD-10 M31.1 diagnosis of thrombotic microangiopathy (TMA)History of thrombocytopenia (platelet volume ≤150,000 per microliter)ICD-10-PCS code for pheresis of plasma or a CPT code for therapeutic apheresis, for plasma pheresis Exclusion criteria: ICD-10-CM codes linked to etiologies of TMA other than TTP (hemopoietic stem cell-associated TMA [M31.11], hemolytic uremic syndrome [D59.3], disseminated intravascular coagulation [D65], antiphospholipid syndrome [D68.61], malignant hypertension [I10, I15])Deceased Results: As of July 31, 2023, there were a total of 142,046,930 living individuals registered in the database in the last 10 years. Of these, 560 individuals met eligibility criteria to define the TTP population (prevalence 0.0004% or 4 cases per million). Demographic characteristics included mean age at diagnosis (39.3 years [SD=14.3]), gender (73% female, 27% male), race (White = 44%, Black or African American = 33%, Asian = 3%, Other or unknown = 20%) and ethnicity (Not Hispanic or Latino = 73%, Unknown = 16%, Hispanic or Latino = 12%). Therapeutic interventions included systemic glucocorticoids (93% [prednisone 84%, methylprednisolone 68%, hydrocortisone 46%, dexamethasone 38%, triamcinolone 14%, betamethasone 9%, prednisolone 5%]), antithrombotics (86% [heparin group 79%, platelet aggregation inhibitors {excluding heparin} 29%, enzymes 27%, other antithrombotic agents {including caplacizumab} 18%, vitamin K antagonists 7%, direct factor Xa inhibitors 7%, direct thrombin inhibitors 2%]) and CD20 inhibitors (including rituximab) (40%). Conclusion: This this is the first report to define and characterize a TTP population from a large US EHR database using a combination administrative codes and laboratory data. The study found a very low prevalence rate; that closely aligns with previously published estimates from registries of 4-5 cases per million for ITTP and 1-2 cases per million for CTTP. It was not possible to differentiate between CTTP and ITTP cases in this study due to a lack of ADAMTS13 genomic data. Demographic findings were consistent with previous registry data (median age of acute TTP presentation in the third to fourth decade of life, typically affecting females more than males). We found a high level of systemic glucocorticoid use as expected, but also high percentage of patients treated with rituximab in line with recent ISTH guidance. The authors are aware of other potential treatments in early development including gene therapy candidates for CTTP and cell therapy candidates for ITTP. This represents a tantalizing prospect for this population; however, to demonstrate efficacy and safety, these new therapies will require scientifically rigorous and well planned clinical studies including robust long-term follow-up.

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.

Age-dependent topic modeling of comorbidities in UK Biobank identifies disease subtypes with differential genetic risk

The analysis of longitudinal data from electronic health records (EHRs) has the potential to improve clinical diagnoses and enable personalized medicine, motivating efforts to identify disease subtypes from patient comorbidity information. Here we introduce an age-dependent topic modeling (ATM) method that provides a low-rank representation of longitudinal records of hundreds of distinct diseases in large EHR datasets. We applied ATM to 282,957 UK Biobank samples, identifying 52 diseases with heterogeneous comorbidity profiles; analyses of 211,908 All of Us samples produced concordant results. We defined subtypes of the 52 heterogeneous diseases based on their comorbidity profiles and compared genetic risk across disease subtypes using polygenic risk scores (PRSs), identifying 18 disease subtypes whose PRS differed significantly from other subtypes of the same disease. We further identified specific genetic variants with subtype-dependent effects on disease risk. In conclusion, ATM identifies disease subtypes with differential genome-wide and locus-specific genetic risk profiles.

Characteristics and Predictors of Apparent Treatment-Resistant Hypertension in Real-World Populations Using Electronic Health Record-Based Data.

Apparent treatment-resistant hypertension (aTRH) is defined as uncontrolled blood pressure (BP) despite using ≥3 antihypertensive classes or controlled BP while using ≥4 antihypertensive classes. Patients with aTRH have a higher risk for adverse cardiovascular outcomes compared with patients with controlled hypertension (HTN). Although there have been prior reports on the prevalence, characteristics, and predictors of aTRH, these have been broadly derived from smaller datasets, randomized controlled trials, or closed healthcare systems. We extracted patients with HTN defined by ICD-9 and ICD-10 codes during 1/1/2015-12/31/2018, from 2 large electronic health record databases: the OneFlorida Data Trust (n = 223,384) and Research Action for Health Network (REACHnet) (n = 175,229). We applied our previously validated aTRH and stable controlled HTN computable phenotype algorithms and performed univariate and multivariate analyses to identify the prevalence, characteristics, and predictors of aTRH in these populations. The prevalence of aTRH among patients with HTN in OneFlorida (16.7%) and REACHnet (11.3%) was similar to prior reports. Both populations had a significantly higher proportion of Black patients with aTRH compared with those with stable controlled HTN. aTRH in both populations shared similar significant predictors, including Black race, diabetes, heart failure, chronic kidney disease, cardiomegaly, and higher body mass index. In both populations, aTRH was significantly associated with similar comorbidities, when compared with stable controlled HTN. In 2 large, diverse real-world populations, we observed similar comorbidities and predictors of aTRH as prior studies. In the future, these results may be used to improve healthcare professionals' understanding of aTRH predictors and associated comorbidities.

Estimating medication adherence from Electronic Health Records: comparing methods for mining and processing asthma treatment prescriptions

BackgroundMedication adherence is usually defined as the extent of the agreement between the medication regimen agreed to by patients with their healthcare provider and the real-world implementation. Proactive identification of those with poor adherence may be useful to identify those with poor disease control and offers the opportunity for ameliorative action. Adherence can be estimated from Electronic Health Records (EHRs) by comparing medication dispensing records to the prescribed regimen. Several methods have been developed in the literature to infer adherence from EHRs, however there is no clear consensus on what should be considered the gold standard in each use case.Our objectives were to critically evaluate different measures of medication adherence in a large longitudinal Scottish EHR dataset. We used asthma, a chronic condition with high prevalence and high rates of non-adherence, as a case study.MethodsOver 1.6 million asthma controllers were prescribed for our cohort of 91,334 individuals, between January 2009 and March 2017. Eight adherence measures were calculated, and different approaches to estimating the amount of medication supply available at any time were compared.ResultsEstimates from different measures of adherence varied substantially. Three of the main drivers of the differences between adherence measures were the expected duration (if taken as in accordance with the dose directions), whether there was overlapping supply between prescriptions, and whether treatment had been discontinued. However, there are also wider, study-related, factors which are crucial to consider when comparing the adherence measures.ConclusionsWe evaluated the limitations of various medication adherence measures, and highlight key considerations about the underlying data, condition, and population to guide researchers choose appropriate adherence measures. This guidance will enable researchers to make more informed decisions about the methodology they employ, ensuring that adherence is captured in the most meaningful way for their particular application needs.

Association Between Fluctuations in Blood Lipid Levels Over Time With Incident Alzheimer Disease and Alzheimer Disease-Related Dementias.

Prevention strategies for Alzheimer disease and Alzheimer disease-related dementias (AD/ADRDs) are urgently needed. Lipid variability, or fluctuations in blood lipid levels at different points in time, has not been examined extensively and may contribute to the risk of AD/ADRD. Lipid panels are a part of routine screening in clinical practice and routinely available in electronic health records (EHR). Thus, in a large geographically defined population-based cohort, we investigated the variation of multiple lipid types and their association to the development of AD/ADRD. All residents living in Olmsted County, Minnesota on the index date January 1, 2006, aged 60 years or older without an AD/ADRD diagnosis were identified. Persons with ≥3 lipid measurements including total cholesterol, triglycerides, low-density lipoprotein cholesterol (LDL-C), or high-density lipoprotein cholesterol (HDL-C) in the 5 years before index date were included. Lipid variation was defined as any change in individual's lipid levels over time regardless of direction and was measured using variability independent of the mean (VIM). Associations between lipid variation quintiles and incident AD/ADRD were assessed using Cox proportional hazards regression. Participants were followed through 2018 for incident AD/ADRD. The final analysis included 11,571 participants (mean age 71 years; 54% female). Median follow-up was 12.9 years with 2,473 incident AD/ADRD cases. After adjustment for confounding variables including sex, race, baseline lipid measurements, education, BMI, and lipid-lowering treatment, participants in the highest quintile of total cholesterol variability had a 19% increased risk of incident AD/ADRD, and those in highest quintile of triglycerides, variability had a 23% increased risk. In a large EHR derived cohort, those in the highest quintile of variability for total cholesterol and triglyceride levels had an increased risk of incident AD/ADRD. Further studies to identify the mechanisms behind this association are needed.

726-P: Medication Adherence among American Indians with Type 2 Diabetes—Assessing the Quality of Electronic Health Record Data within a Large Tribal Health Care System

Electronic health record (EHR) systems within tribal healthcare facilities offer a rich but underutilized data source about medication adherence and its potential to predict HbA1c. The purpose of this analysis was to describe data quality issues for medication events in a dataset extracted from a large tribal healthcare EHR system. This analysis was a critical step prior to conducting a study of medication adherence among American Indians with type 2 diabetes (T2D). The EHR dataset from 2017-2021 included 10,506 patients who were enrolled adult tribal citizens, had T2D, and had at least one health system encounter. Tribal pharmacy refill data were linked to the EHR data and included information such as prescription regimen (e.g., drug, dose, route of administration, quantity), date new medication prescribed, date picked up, and refill dates. Data quality indicators were: duplicate data, missing data (e.g., dispense and fill dates, RX number), and data inconsistent with eligibility criteria. The mean age of the sample was 60.2 ±13.4 years and 52% were female. Out of the 1,400,593 patient medication fill/refill events we excluded 64,015 medication events due to a) duplicate prescription fill data (n=112); b) no dispense date (n=1,081); c) no order date (n=0); d) no fill date (n=0); e) no RX number (n=0); f) missing number of days dispensed (n=0); g) missing drug (n=0); h) missing dose (n=0); and i) data extracted not meeting the eligibility criteria (n=62,822). The date the medication was prescribed was not extracted from the EHR. Discovering these results was labor intensive taking more than 6 months. This analysis revealed few data-related concerns. With the exception of one missing variable, the dataset includes high quality data from which we will calculate medication adherence among American Indians with T2D. Disclosure L.Scarton: None. T.N.Nelson: None. A.Devaughan circles: None. A.Legaspi: None. D.J.Wilkie: Other Relationship; eNursing llc. Funding National Institutes of Health (1R01NR020386-01)

Age and sex-specific stroke epidemiology in COVID-19

BackgroundCOVID-19 has emerged as an independent risk factor for stroke. We aimed to determine age and sex-specific stroke incidence and risk factors with COVID-19 in the US using a large electronic health record (EHR) that included both inpatients and outpatients.MethodsA retrospective cohort study was conducted using individual-level data from Optum® de-identified COVID-19 EHR. A total of 387,330 individuals aged ≥ 18 with laboratory-confirmed COVID-19 between March 1, 2020 and December 31, 2020 were included. The primary outcome was cumulative incidence of stroke after COVID-19 confirmation within 180 days of follow-up or until death. Kaplan–Meier cumulative incidence curves for acute ischemic stroke (AIS), intracerebral hemorrhage (ICH), and a composite outcome of all strokes were stratified by sex and age, and the differences in curves were assessed using a log-rank test. The relative risk of stroke by demographics and risk factors was estimated using multivariable Cox-proportional hazards regressions and adjusted hazard ratios (aHRs).ResultsOf 387,330 COVID-19 patients, 2,752 patients (0.71%, 95% CI 0.68–0.74) developed stroke during the 180-day follow-up, AIS in 0.65% (95% CI 0.62–0.67), and ICH in 0.11% (95% CI 0.10–0.12). Of strokes among COVID-19 patients, 57% occurred within 3 days. Advanced age was associated with a substantially higher stroke risk, with aHR 6.92 (5.72–8.38) for ages 65–74, 9.42 (7.74–11.47) for ages 75–84, and 11.35 (9.20–14.00) for ages 85 and older compared to ages 18–44 years. Men had a 32% higher risk of stroke compared to women. African-American [aHR 1.78 (1.61–1.97)] and Hispanic patients [aHR 1.48 (1.30–1.69)] with COVID-19 had an increased risk of stroke compared to white patients.ConclusionThis study has several important findings. AIS and ICH risk in patients with COVID-19 is highest in the first 3 days of COVID-19 positivity; this risk decreases with time. The incidence of stroke in patients with COVID-19 (both inpatient and outpatient) is 0.65% for AIS and 0.11% for ICH during the 180-day follow-up. Traditional stroke risk factors increase the risk of stroke in patients with COVID-19. Male sex is an independent risk factor for stroke in COVID-19 patients across all age groups. African-American and Hispanic patients have a higher risk of stroke from COVID-19.

Advancing heart failure research using machine learning

Advancing heart failure research using machine learning

Generalizability of an EHR-network dataset to the United States for cardiovascular disease conditions: Comparison of Cerner real world data with the national inpatient sample

Electronic Health Record (EHR) data from health systems are increasingly being combined for clinical research purposes. Yet, it remains unclear whether these large EHR data sources provide a representative assessment of national disease prevalence and treatment. To evaluate this, we compared Cerner RealWorldData (CRWD), a large EHR data source, to those seen in the National Inpatient Sample (NIS) for three cardiovascular conditions (myocardial infarction (MI), congestive heart failure (CHF), and stroke. Adult patients (age ≥18 years) hospitalized with MI, CHF, and stroke were identified in both CRWD (86 health systems) and the NIS (4,782 hospitals). Patient demographics, comorbidities, procedures, outcomes (length of stay and in-hospital mortality) and hospital type (teaching or non-teaching) were compared between NIS and CRWD patients. Of 86 health systems participating in CRWD, 33 were excluded for potential data quality issues which accounted for about 11% of hospitalizations in the dataset, leaving 53 for inclusion in analysis which accounted for about 89% of hospitalizations in the dataset. Between January 1st 2017 and December 31st 2018, 116,956 MI, 188,107 CHF, and 93,968 stroke hospitalizations were identified in CRWD versus 2,245,300 MI, 4,310,745 CHF, and 1,333,480 stroke hospitalizations in the NIS. Patient demographics were similar among patients in CWRD and the NIS for all three cardiovascular groups except for ethnicity, with underrepresentation of Hispanic individuals in CRWD vs the NIS. Patients hospitalized in CRWD had a slightly higher proportion of coded co-morbidities compared with NIS hospitalizations due to a longer potential look-back period. For patients with MI, hospital mortality, length of stay, coronary artery bypass graft (CABG) rates, and percutaneous coronary intervention (PCI) rates were similar between CRWD and NIS. Additionally, there was similar in hospital mortality and length of stay for those with CHF and stroke hospitalizations between CRWD and NIS. On aggregate, characteristics of hospitalizations for MI, CHF, and stroke using EHR data from one nationwide EHR-derived database, CRWD, appears similar to characteristics of hospitalizations in the nationally representative NIS. Important limitations of CRWD include lack of geographic representativeness, under-representation of Hispanic adults, and the need to exclude health systems for missing data.

An efficient landmark model for prediction of suicide attempts in multiple clinical settings

Growing evidence has shown that applying machine learning models to large clinical data sources may exceed clinician performance in suicide risk stratification. However, many existing prediction models either suffer from “temporal bias” (a bias that stems from using case-control sampling) or require training on all available patient visit data. Here, we adopt a “landmark model” framework that aligns with clinical practice for prediction of suicide-related behaviors (SRBs) using a large electronic health record database. Using the landmark approach, we developed models for SRB prediction (regularized Cox regression and random survival forest) that establish a time-point (e.g., clinical visit) from which predictions are made over user-specified prediction windows using historical information up to that point. We applied this approach to cohorts from three clinical settings: general outpatient, psychiatric emergency department, and psychiatric inpatients, for varying prediction windows and lengths of historical data. Models achieved high discriminative performance (area under the Receiver Operating Characteristic curve 0.74–0.93 for the Cox model) across different prediction windows and settings, even with relatively short periods of historical data. In short, we developed accurate, dynamic SRB risk prediction models with the landmark approach that reduce bias and enhance the reliability and portability of suicide risk prediction models.

A systems approach points to a therapeutic role for retinoids in asparaginase-associated pancreatitis.

Among drug-induced adverse events, pancreatitis is life-threatening and results in substantial morbidity. A prototype example is the pancreatitis caused by asparaginase, a crucial drug used to treat acute lymphoblastic leukemia (ALL). Here, we used a systems approach to identify the factors affecting asparaginase-associated pancreatitis (AAP). Connectivity Map analysis of the transcriptomic data showed that asparaginase-induced gene signatures were potentially reversed by retinoids (vitamin A and its analogs). Analysis of a large electronic health record database (TriNetX) and the U.S. Federal Drug Administration Adverse Events Reporting System demonstrated a reduction in AAP risk with concomitant exposure to vitamin A.Furthermore, we performed a global metabolomic screening of plasma samples from 24 individuals with ALL who developed pancreatitis (cases) and 26 individuals with ALL who did not develop pancreatitis (controls), before and after a single exposure to asparaginase. Screening from this discovery cohort revealed that plasma carotenoids were lower in the cases than in controls. This finding was validated in a larger external cohort. A 30-day dietary recall showed that the cases received less dietary vitamin A than the controls did. In mice, asparaginase administration alone was sufficient to reduce circulating and hepatic retinol. Based on these data, we propose that circulating retinoids protect against pancreatic inflammation and that asparaginase reduces circulating retinoids. Moreover, we show that AAP is more likely to develop with reduced dietary vitamin A intake. The systems approach taken for AAP provides an impetus to examine the role of dietary vitamin A supplementation in preventing or treating AAP.

Variability in Lipid Levels and Risk for Cardiovascular Disease: An Electronic Health Record-Based Population Cohort Study.

Background Larger within-patient variability of lipid levels has been associated with increased risk of cardiovascular disease (CVD); however, measures of lipid variability require ≥3 measurements and are not currently used clinically. We investigated the feasibility of calculating lipid variability within a large electronic health record-based population cohort and assessed associations with incident CVD. Methods and Results We identified all individuals ≥40 years of age who resided in Olmsted County, MN, on January 1, 2006 (index date), without prior CVD, defined as myocardial infarction, coronary artery bypass graft surgery, percutaneous coronary intervention, or CVD death. Patients with ≥3 measurements of total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, or triglycerides during the 5 years before the index date were retained. Lipid variability was calculated using variability independent of the mean. Patients were followed through December 31, 2020 for incident CVD. We identified 19 652 individuals (mean age 61 years; 55% female), who were CVD-free and had variability independent of the mean calculated for at least 1 lipid type. After adjustment, those with highest total cholesterol variability had a 20% increased risk of CVD (Q5 versus Q1 hazard ratio, 1.20 [95% CI, 1.06-1.37]). Results were similar for low-density lipoprotein cholesterol and high-density lipoprotein cholesterol. Conclusions In a large electronic health record-based population cohort, high variability in total cholesterol, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol was associated with an increased risk of CVD, independent of traditional risk factors, suggesting it may be a possible risk marker and target for intervention. Lipid variability can be calculated in the electronic health record environment, but more research is needed to determine its clinical utility.