Electronic Health Record Dataset Research Articles

Enabling scalable clinical interpretation of machine learning (ML)-based phenotypes using real world data

Background: Large and deep electronic health record (EHR) datasets have the potential to increase understanding of real-world patient journeys, and to identify subgroups of patients currently grouped with a common disease label but differing in outcomes and medical need. However, working with EHRs is still relatively new and challenging due to the heterogeneous nature of data. Increasing interest in machine learning (ML)-based EHR aggregation is mostly method-driven, i.e., building on available or newly developed methods. These methods, input requirements, and output are frequently difficult to interpret, especially without data science or statistical training. This endangers the ultimate aim of such analyses: generating actionable and clinically meaningful interpretation.

Associations of remote mental healthcare with clinical outcomes: a natural language processing enriched electronic health record data study protocol

IntroductionPeople often experience significant difficulties in receiving mental healthcare due to insufficient resources, stigma and lack of access to care. Remote care technology has the potential to overcome these barriers...

Risks of mortality and severe coronavirus disease 19 (COVID-19) outcomes in patients with or without systemic lupus erythematosus

ObjectivesWe compared the outcomes of patients with or without systemic lupus erythematosus (SLE) who were diagnosed with coronavirus disease 19 (COVID-19) and evaluated factors within patients with SLE associated with...

The incidence of new mental health disorders after acute pancreatitis: A large, propensity-matched, observational study

IntroductionThe prevalence of acute pancreatitis (AP) and mental health disorders (MHDs) are rising. While the association between chronic pancreatitis (CP) and MHDs is established, it is unknown whether there is a risk of MHDs after an index episode of AP. The aim of this study was to evaluate the incidence of MHDs and pharmacotherapy use after an episode of AP. MethodsThis was a large observational study using the TriNetX research network, an electronic health record dataset containing inpatient and outpatient data from more than 50 healthcare organizations. Patients with AP from 2015–2020 were identified. Four cohorts were created: acute necrotizing pancreatitis (ANP), acute pancreatitis without necrosis (AP-WON), acute appendicitis, and healthy controls without pancreatitis. The cohorts were matched by age, sex, race, ethnicity, and nicotine and alcohol use. The primary outcome was new composite MHDs at one-year. Secondary outcomes included stratified MHDs, psychiatric medication use, opioid analgesic use, and all-cause mortality. ResultsThe ANP, AP-WON, appendicitis, and healthy control cohorts contained 11,806, 177,266, 27,187, and 561,833 patients, respectively. Patients with AP-WON had significantly higher rates of composite MHDs compared with those hospitalized for appendicitis (9.7% vs 4.7%, HR 1.9, 95% CI 1.7–1.9). This association was augmented when comparing ANP to appendicitis (12.8% vs 5.2%, HR 2.4, 95% CI 2.1–2.7). All secondary outcomes were observed at significantly higher rates in the AP-WON cohort when compared to appendicitis. Again, these associations were augmented comparing ANP to appendicitis. ConclusionCompared with controls, patients with AP had significantly higher rates of new MHDs and their associated pharmacotherapies at one-year, suggesting that a single episode of AP may independently place patients at risk for developing MHDs irrespective of whether they go on to develop CP.

Paxlovid associated with decreased hospitalization rate among adults with COVID-19 — United States, April–September 2022

Nirmatrelvir-ritonavir (Paxlovid), an oral antiviral treatment, is authorized for adults with mild-to-moderate COVID-19 who are at increased risk for progression to severe illness. However, real-world evidence on the benefit of Paxlovid, according to vaccination status, age group, and underlying health conditions, is limited. To examine the benefit of Paxlovid in adults aged ≥18 years in the United States, a large electronic health record (EHR) data set (Cosmos†) was analyzed to assess the association between receiving a prescription for Paxlovid and hospitalization with a COVID-19 diagnosis in the ensuing 30 days.

PS-C08-7: TIME IN THERAPEUTIC RANGE IS NOT A RISK FACTOR FOR CARDIOVASCULAR DISEASE WHEN ADJUSTED FOR MEAN SYSTOLIC BLOOD PRESSURE

Background: Time in therapeutic range (TTR) has been suggested as a novel metric to assess blood pressure (BP) control, the quality of antihypertensive therapy, and the risk of cardiovascular events in people with hypertension. Methods: We used a validated electronic health record dataset, including all BP recordings in primary care in two Swedish counties from 1993 to 2011. People were eligible if they were 18 years or older, had a diagnosis of hypertension, no previous cardiovascular events, and at least five BP recordings during the first five years from the date of hypertension diagnosis. TTR was defined as the percentage of BP readings with a systolic value less than 140 mmHg. Participants were followed until death, migration or the 31st of December 2019. The risk of myocardial infarction and stroke was assessed using Cox proportional hazards model with TTR as independent variable, with and without mean systolic BP as a covariate. All analyses were adjusted for age, sex, diabetes, county, income and educational level. Results: 72 724 participants (56% female, mean age 62 years) were included in the analyses. The mean BP level during the first five years after diagnosis was 151/85 mmHg, with a median TTR of 17.7% (interquartile range 0–40%). During an average 7.7 years of follow-up, 6551 participants experienced a first-ever stroke and 5513 a first-ever myocardial infarction. In the unadjusted analyses, the risk of stroke decreased by 7% (HR 0.93, 95%CI 0.92–0.94) and the risk of myocardial infarction decreased by 6% (0.94, 0.93–0.95) for each 10 percentage points increase in TTR. When including mean systolic BP as a covariate, however, TTR was not associated with either stroke (0.99, 0.97–1.01) or myocardial infarction (1.00, 0.99–1.02). Conclusion: TTR for systolic BP is not an independent risk factor for stroke and myocardial infarction in clinical practice and does not add importantly to mean systolic BP as a measure of BP control.

Health record hiccups-5,526 real-world time series with change points labelled by crowdsourced visual inspection.

Large routinely collected data such as electronic health records (EHRs) are increasingly used in research, but the statistical methods and processes used to check such data for temporal data quality issues have not moved beyond manual, ad hoc production and visual inspection of graphs. With the prospect of EHR data being used for disease surveillance via automated pipelines and public-facing dashboards, automation of data quality checks will become increasingly valuable. We generated 5,526 time series from 8 different EHR datasets and engaged >2,000 citizen-science volunteers to label the locations of all suspicious-looking change points in the resulting graphs. Consensus labels were produced using density-based clustering with noise, with validation conducted using 956 images containing labels produced by an experienced data scientist. Parameter tuning was done against 670 images and performance calculated against 286 images, resulting in a final sensitivity of 80.4% (95% CI, 77.1%-83.3%), specificity of 99.8% (99.7%-99.8%), positive predictive value of 84.5% (81.4%-87.2%), and negative predictive value of 99.7% (99.6%-99.7%). In total, 12,745 change points were found within 3,687 of the time series. This large collection of labelled EHR time series can be used to validate automated methods for change point detection in real-world settings, encouraging the development of methods that can successfully be applied in practice. It is particularly valuable since change point detection methods are typically validated using synthetic data, so their performance in real-world settings cannot be assumed to be comparable. While the dataset focusses on EHRs and data quality, it should also be applicable in other fields.

Machine Learning and Real-World Data to Predict Lung Cancer Risk in Routine Care.

This study used machine learning to develop a 3-year lung cancer risk prediction model with large real-world data in a mostly younger population. Over 4.7 million individuals, aged 45 to 65 years with no history of any cancer or lung cancer screening, diagnostic, or treatment procedures, with an outpatient visit in 2013 were identified in Optum's de-identified Electronic Health Record (EHR) dataset. A least absolute shrinkage and selection operator model was fit using all available data in the 365 days prior. Temporal validation was assessed with recent data. External validation was assessed with data from Mercy Health Systems EHR and Optum's de-identified Clinformatics Data Mart Database. Racial inequities in model discrimination were assessed with xAUCs. The model AUC was 0.76. Top predictors included age, smoking, race, ethnicity, and diagnosis of chronic obstructive pulmonary disease. The model identified a high-risk group with lung cancer incidence 9 times the average cohort incidence, representing 10% of patients with lung cancer. Model performed well temporally and externally, while performance was reduced for Asians and Hispanics. A high-dimensional model trained using big data identified a subset of patients with high lung cancer risk. The model demonstrated transportability to EHR and claims data, while underscoring the need to assess racial disparities when using machine learning methods. This internally and externally validated real-world data-based lung cancer prediction model is available on an open-source platform for broad sharing and application. Model integration into an EHR system could minimize physician burden by automating identification of high-risk patients.

Assessment of the Frequency, Phenotypes, and Outcomes of Acute Liver Injury Associated with Amoxicillin/Clavulanate in 1.4 Million Patients in the Veterans Health Administration

Drug-induced liver injury is a significant health issue, yet the exposure-based incidence remains to be characterized. We aimed to assess the frequency, phenotypes, and outcomes of acute liver injury associated with amoxicillin/clavulanate using a large electronic health record system. Using the Veterans Health Administration electronic health record system, we developed the framework to identify unexplained acute liver injury, defined by alanine aminotransferase and/or alkaline phosphatase elevation temporally linked to prescription records of amoxicillin/clavulanate, a major culprit of clinically significant drug-induced liver injury, excluding other competing causes. The population was subcategorized by pre-existing liver conditions and inpatient status at the time of exposure for the analysis. Among 1,445,171 amoxicillin/clavulanate first exposures in unique individuals [92% men; mean age (standard deviation): 59 (15) years], 6476 (incidence: 0.448%) acute liver injuries were identified. Of these, 4427 (65%) had alternative causes, yielding 2249 (incidence: 0.156%) with unexplained acute liver injuries. The incidence of unexplained acute liver injury was lowest in outpatients without underlying liver disease (0.067%) and highest in inpatients with pre-existing liver conditions (0.719%). Older age, male sex, and American Indian or Alaska Native (vs White) were associated with a higher incidence of unexplained acute liver injury. Cholestatic injury affected 74%, exhibiting a higher frequency with advanced age, inpatient exposure, and pre-existing liver conditions. Hepatocellular injury with bilirubin elevation affected 0.003%, with a higher risk at age >45 years. During a 12-month follow-up, patients with unexplained acute liver injury had a higher adjusted overall mortality risk than those without evident acute liver injury. This framework identifies unexplained acute liver injury following drug exposure in large electronic health record datasets. After validating in other systems, this framework can aid in deducing drug-induced liver injury in the general patient population and regulatory decision making to promote drug safety and public health.

2298. Clinical and Demographic Characteristics of COVID-19 in Pediatric Patients in the United States

Abstract Background The percentage of children infected with COVID-19 has outpaced that of adults. As children >5 years are now eligible to receive vaccines, it is necessary to understand the effect of vaccination in the context of demographic characteristics, clinical factors, and variants on pediatric COVID-19 illness severity. Methods We conducted a descriptive study of patients ≤18 years from the Optum® COVID-19 electronic health record dataset. Patients were included if positive for COVID-19 by polymerase chain reaction or antigen testing for the first time from 3/12/2020 to 1/20/2022. We compare race and ethnicity, age, gender, US region of residence, vaccination status, body mass index (BMI), pediatric comorbidity index (PCI) (Sun, Am. J. Epidemiol. 2021), and predominant variant (by time and region) with 2-tailed t-test, multi-category chi-square test, and odds ratios (R version 4.1.2; α = 0.05). PCI is a validated comorbidity index predicting hospitalization in pediatric patients. Results Of all pediatric patients in our dataset, 165,468 (13.2%) tested positive for COVID-19. 3,087 (1.9%) were hospitalized, 1,417 (0.9%) were admitted to the ICU, 1545 (0.9%) received respiratory support, and 31 (0.02%) died, comparable to AAP-reported hospitalization and mortality rates in US children. Patients with severe outcomes were more likely to be younger, non-Caucasian, from the US South, unvaccinated, and have a higher PCI (Figure 1). Excluding non-severe outcomes, rates of death and ICU admission were higher in 0–4-year-olds compared to 5–11 or 12–18-year-olds (Figure 2). All patients receiving at least one dose of the vaccine survived. The odds ratio of a severe outcome is 0.11 (95% CI 0.07–0.16) in fully vaccinated patients compared to unvaccinated patients. The odds ratio of a severe outcome is 0.55 (95% CI 0.49–0.63) in partially vaccinated patients compared to unvaccinated patients. Demographic and clinical characteristics of pediatric patients with COVID-19 Relative proportion of clinically severe outcomes within age groups, excluding non-severe outcomes Conclusion In this large population, incidence rate of severe outcomes from COVID-19 in pediatric patients was higher among non-Caucasian patients, living in the South, with underlying comorbid illness, and those not yet eligible for vaccination. These findings reinforce the need for a vaccine for younger patients and targeted vaccine outreach to racial and ethnic minorities and children with chronic conditions. Disclosures Christoph U. Lehmann, MD, Celanese: Stocks/Bonds|Markel: Stocks/Bonds|Springer: Honoraria|UTSW: Employee.

1161. Vaccination During Acute COVID-19 Infection Protects Against 30 Day Adverse Outcomes

Abstract Background Therapeutic vaccination following SARS-CoV-2 infection might stimulate anti-viral immunity and improve patient outcomes. We investigated, amongst previously unvaccinated patients, whether vaccination with the Pfizer, Moderna, or Johnson & Johnson vaccines within 14 days of a positive SARS-CoV-2 test affected 30-day patient outcomes. Methods Using a deidentified national electronic health record dataset (Optum, Inc.), we identified previously unvaccinated patients who tested positive for COVID-19 between 12/11/2020 and 12/19/2021. Among this cohort, 1,909 patients received a first vaccine dose within 14 days (vaccinated) while 446,309 did not receive a first dose of vaccine within 30 days of their first positive test (unvaccinated). We performed 1:1 propensity score matching based on age, gender, race, ethnicity, region, BMI, insurance, and comorbidities from the Charlson Comorbidity Index. Next, we compared odds of severe outcomes within 30 days between vaccinated and unvaccinated groups using a partial proportional odds model with the following ordinal severity outcome: no hospitalization, hospitalization, ICU stay, or death (Stata version 17.0, α = 0.05). Results 1,909 vaccinated patients were propensity score-matched to 1,909 unvaccinated patients. The final matched cohort was statistically indistinguishable (p > 0.05) for all metrics used in propensity score calculation. This matched cohort (n = 3,818) was 58.6% female, 67.7% white, 12.6% Hispanic, and 56.4% commercially insured, with a mean age of 50.6 years and a similar comorbidity profile. A partial proportional odds model showed that unvaccinated patients were at increased risk for hospitalization and higher ordered outcomes (OR = 1.19, 95% CI: 1.02-1.39), ICU stay and higher ordered outcomes (OR 1.63, 95% CI: 1.21-2.20), and death (OR 4.57, 95% CI: 2.50-8.37). Conclusion Among previously unvaccinated patients, those who received a first dose vaccine within 14 days of a positive COVID-19 test were less likely to experience hospitalization, ICU stay, or death compared to matched peers who did not receive a first dose in the acute phase of infection. The sample size of patients vaccinated during the acute phase is limited, so further studies are indicated to evaluate the safety and efficacy of this practice. Disclosures Christoph U. Lehmann, MD, FAAP, FACMI, FIAHSI, Celanese: Stocks/Bonds|Markel: Stocks/Bonds|Springer: Honoraria.

A machine learning approach to identifying patients with pulmonary hypertension using real-world electronic health records

BackgroundThis study aimed to develop a machine learning (ML) model to identify patients who are likely to have pulmonary hypertension (PH), using a large patient-level US-based electronic health record (EHR) database. MethodsA gradient boosting model, XGBoost, was developed using data from Optum's US-based de-identified EHR dataset (2007–2019). PH and disease control adult patients were identified using diagnostic, treatment and procedure codes and were randomly split into the training (90%) or test set (10%). Model features included patient demographics, physician visits, diagnoses, procedures, prescriptions, and laboratory test results. SHapley Additive exPlanations values were used to determine feature importance. ResultsWe identified 11,279,478 control and 115,822 PH patients (mean age, respectively: 62 and 68 years, both 53% female). The final model used 165 features, with the most important predictive features including diagnosis of heart failure, shortness of breath and atrial fibrillation. The model predicted PH with an area under the receiver operating characteristic curve (AUROC) of 0.92. AUROC remained above 0.80 for the prediction of PH up to and beyond 18 months before diagnosis. Among the PH patients, we also identified 955 pulmonary arterial hypertension (PAH) and 1432 chronic thromboembolic pulmonary hypertension (CTEPH) patients, and the range of AUROCs obtained for these cohorts was 0.79–0.90 and 0.87–0.96, respectively. ConclusionsThis model to detect PH based on patients’ EHR records is viable and performs well in subgroups of PAH and CTEPH patients. This approach has the potential to improve patient outcomes by reducing diagnostic delay in PH.

Paxlovid Associated with Decreased Hospitalization Rate Among Adults with COVID-19 - United States, April-September 2022.

Nirmatrelvir-ritonavir (Paxlovid), an oral antiviral treatment, is authorized for adults with mild-to-moderate COVID-19 who are at increased risk for progression to severe illness. However, real-world evidence on the benefit of Paxlovid, according to vaccination status, age group, and underlying health conditions, is limited. To examine the benefit of Paxlovid in adults aged ≥18 years in the United States, a large electronic health record (EHR) data set (Cosmos†) was analyzed to assess the association between receiving a prescription for Paxlovid and hospitalization with a COVID-19 diagnosis in the ensuing 30 days. A Cox proportional hazards model was used to estimate this association, adjusted for demographic characteristics, geographic location, vaccination, previous infection, and number of underlying health conditions. Among 699,848 adults aged ≥18 years eligible for Paxlovid during April-August 2022, 28.4% received a Paxlovid prescription within 5 days of COVID-19 diagnosis. Being prescribed Paxlovid was associated with a lower hospitalization rate among the overall study population (adjusted hazard ratio [aHR]=0.49), among those who had received ≥3 mRNA COVID-19 vaccines (aHR=0.50), and across age groups (18-49 years: aHR=0.59; 50-64 years: aHR=0.40; and ≥65 years: aHR=0.53). Paxlovid should be prescribed to eligible adults to reduce the risk of COVID-19-associated hospitalization.

External validation of the Johns Hopkins Fall Risk Assessment Tool in older Dutch hospitalized patients.

Fall prevention is a safety goal in many hospitals. The performance of the Johns Hopkins Fall Risk Assessment Tool (JHFRAT) in older inpatients is largely unknown. We aimed to assess the JHFRAT performance in a large sample of Dutch older inpatients, including its trend over time. We used an Electronic Health Records (EHR) dataset with hospitalized patients (≥ 70), admitted for ≥ 24h between 2016 and 2021. Inpatient falls were extracted from structured and free-text data. We assessed the association between JHFRAT and falls using logistic regression. For test accuracy, we calculated sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Discrimination was measured by the AUC. For calibration, we plotted the predicted fall probability with the actual probability of falls. For time-related effects, we calculated the AUC per 6months (using data of patients admitted during the 6months' time interval) and plotted these different AUC values over time. Furthermore, we compared the model (JHFRAT and falls) with and without adjusting for seasonal influenza, COVID-19, spring, summer, fall or winter periods. Data included 17,263 admissions with at least 1 JHFRAT measurement, a median age of 76 and a percentage female of 47%. The in-hospital fall prevalence was 2.5%. JHFRAT [OR = 1.11 (1.03-1.20)] and its subcategories were significantly associated with falls. For medium/high risk of falls (JHFRAT > 5), sensitivity was 73%, specificity 51%, PPV 4% and NPV 99%. The overall AUC was 0.67, varying over time between 0.62 and 0.71 (for 6months' time intervals). Seasonal influenza did affect the association between JHFRAT and falls. COVID-19, spring, summer, fall or winter did not affect the association. Our results show an association between JHFRAT and falls, a low discrimination by JHFRAT for older inpatients and over-prediction in the calibration. Improvements in the fall-risk assessment are warranted to improve efficiency.

Patients with allergic asthma have lower risk of severe COVID-19 outcomes than patients with nonallergic asthma

BackgroundAlthough asthma does not appear to be a risk factor for severe coronavirus disease 2019 (COVID-19), outcomes could vary for patients with different asthma subtypes. The objective of this analysis was to compare COVID-19 outcomes in real-world cohorts in the United States among patients with asthma, with or without evidence of allergy.MethodsIn a retrospective analysis of the COVID-19 Optum electronic health record dataset (February 20, 2020–January 28, 2021), patients diagnosed with COVID-19 with a history of moderate-to-severe asthma were divided into 2 cohorts: those with evidence of allergic asthma and those without (nonallergic asthma). After 1:1 propensity score matching, in which covariates were balanced and potential bias was removed, COVID-19 outcomes were compared between cohorts.ResultsFrom a COVID-19 population of 591,198 patients, 1595 patients with allergic asthma and 8204 patients with nonallergic asthma were identified. After propensity score matching (n = 1578 per cohort), risk of death from any cause after COVID-19 diagnosis was significantly lower for patients with allergic vs nonallergic asthma (hazard ratio, 0.48; 95% CI 0.28–0.83; P = 0.0087), and a smaller proportion of patients with allergic vs nonallergic asthma was hospitalized within − 7 to + 30 days of COVID-19 diagnosis (13.8% [n = 217] vs 18.3% [n = 289]; P = 0.0005). Among hospitalized patients, there were no significant differences between patients with allergic or nonallergic asthma in need for intensive care unit admission, respiratory support, or COVID-19 treatment.ConclusionsAsthma subtype may influence outcomes after COVID-19; patients with allergic asthma are at lower risk for hospitalization/death than those with nonallergic asthma.

Immunolab: Combining targeted real-world data with advanced analytics to generate evidence at scale in immunology

Real-world evidence (RWE) has traditionally been used by regulatory or payer authorities to inform disease burden, background risk, or conduct post-launch pharmacovigilance, but in recent years RWE has been increasingly used to inform regulatory decision-making. However, RWE data sources remain fragmented, and datasets are disparate and often collected inconsistently. To this end, we have constructed an RWE-generation platform, Immunolab, to facilitate data-driven insights, hypothesis generation and research in immunological diseases driven by type 2 inflammation. Immunolab leverages a large, anonymized patient cohort from the Optum electronic health record and claims dataset containing over 17 million patient lives. Immunolab is an interactive platform that hosts three analytical modules: the Patient Journey Mapper, to describe the drug treatment patterns over time in patient cohorts; the Switch Modeler, to model treatment switching patterns and identify its drivers; and the Head-to-Head Simulator, to model the comparative effectiveness of treatments based on relevant clinical outcomes. The Immunolab modules utilize various analytic methodologies including machine learning algorithms for result generation which can then be presented in various formats for further analysis and interpretation.

Using set visualisation to find and explain patterns of missing values: a case study with NHS hospital episode statistics data

ObjectivesMissing data is the most common data quality issue in electronic health records (EHRs). Missing data checks implemented in common analytical software are typically limited to counting the number of...

Demographic factors, psychiatric and physical comorbidities associated with starting preexposure prophylaxis in a nationally distributed cohort

Due to a large number of small studies and limited control for confounding, existing evidence regarding patient characteristics associated with PrEP initiation is inconsistent. We used a large electronic health record cohort to determine which demographic, physical morbidity and psychiatric conditions are associated with PrEP initiation. Eligible adult (≥18 years) patients were selected from the Optum® de-identified Electronic Health Record dataset (2010–2018). Non-HIV sexually transmitted diseases and high risk sexual behavior was used to identify patients eligible for PrEP. A fully adjusted Poisson regression model estimated the association between age, gender, race, insurance status, comorbidity index, depression, anxiety, dysthymia, severe mental illness, substance use disorder and nicotine dependence/smoking and rate of PrEP initiation. The cohort (n = 30,909) was mostly under 40 years of age (64.3%), 67.6% were female and 58.2% were White. The cumulative incidence of PrEP initiation was 1.3% (n = 408). Patients ≥60 years of age, compared to 18–29 year olds and Black compared to White patients had significantly lower rates of PrEP initiation. Anxiety disorder was significantly associated with higher rate of PrEP initiation (RR = 1.67; 95%CI:1.20–2.33) and nicotine dependence/smoking with a lower rate (RR = 0.73; 95%CI:0.54–0.97). PrEP is underutilized, and a race disparity exists in PrEP initiation. In the context of existing research, nicotine dependence/smoking is the patient characteristic most consistently associated lower rates of starting PrEP. Given the high prevalence of smoking in PrEP eligible patients, physicians may want to integrate discussions of smoking cessation in patient-provider decisions to start PrEP.

Predicting Clinical Remission of Chronic Urticaria Using Random Survival Forests: Machine Learning Applied to Real-World Data.

The time required to reach clinical remission varies in patients with chronic urticaria (CU). The objective of this study is to develop a predictive model using a machine learning methodology to predict time to clinical remission for patients with CU. Adults with ≥ 2 ICD-9/10 relevant CU diagnosis codes/CU-related treatment > 6weeks apart were identified in the Optum deidentified electronic health record dataset (January 2007 to June 2019). Clinical remission was defined as ≥ 12months without CU diagnosis/CU-related treatment. A random survival forest was used to predict time from diagnosis to clinical remission for each patient based on clinical and demographic features available at diagnosis. Model performance was assessed using concordance, which indicates the degree of agreement between observed and predicted time to remission. To characterize clinically relevant groups, features were summarized among cohorts that were defined based on quartiles of predicted time to remission. Among 112,443 patients, 73.5% reached clinical remission, with a median of 336days from diagnosis. From 1876 initial features, 176 were retained in the final model, which predicted a median of 318days to remission. The model showed good performance with a concordance of 0.62. Patients with predicted longer time to remission tended to be older with delayed CU diagnosis, and have more comorbidities, more laboratory tests, higher body mass index, and polypharmacy during the 12-month period before the first CU diagnosis. Applying machine learning to real-world data enabled accurate prediction of time to clinical remission and identified multiple relevant demographic and clinical variables with predictive value. Ongoing work aims to further validate and integrate these findings into clinical applications for CU management.

Modeling electronic health record data using an end-to-end knowledge-graph-informed topic model

The rapid growth of electronic health record (EHR) datasets opens up promising opportunities to understand human diseases in a systematic way. However, effective extraction of clinical knowledge from EHR data has been hindered by the sparse and noisy information. We present Graph ATtention-Embedded Topic Model (GAT-ETM), an end-to-end taxonomy-knowledge-graph-based multimodal embedded topic model. GAT-ETM distills latent disease topics from EHR data by learning the embedding from a constructed medical knowledge graph. We applied GAT-ETM to a large-scale EHR dataset consisting of over 1 million patients. We evaluated its performance based on topic quality, drug imputation, and disease diagnosis prediction. GAT-ETM demonstrated superior performance over the alternative methods on all tasks. Moreover, GAT-ETM learned clinically meaningful graph-informed embedding of the EHR codes and discovered interpretable and accurate patient representations for patient stratification and drug recommendations. GAT-ETM code is available at https://github.com/li-lab-mcgill/GAT-ETM.