Electronic Health Record Database Research Articles

Question Answering for Electronic Health Records: Scoping Review of Datasets and Models.

Question answering (QA) systems for patient-related data can assist both clinicians and patients. They can, for example, assist clinicians in decision-making and enable patients to have a better understanding of their medical history. Substantial amounts of patient data are stored in electronic health records (EHRs), making EHR QA an important research area. Because of the differences in data format and modality, this differs greatly from other medical QA tasks that use medical websites or scientific papers to retrieve answers, making it critical to research EHR QA. This study aims to provide a methodological review of existing works on QA for EHRs. The objectives of this study were to identify the existing EHR QA datasets and analyze them, study the state-of-the-art methodologies used in this task, compare the different evaluation metrics used by these state-of-the-art models, and finally elicit the various challenges and the ongoing issues in EHR QA. We searched for articles from January 1, 2005, to September 30, 2023, in 4 digital sources, including Google Scholar, ACL Anthology, ACM Digital Library, and PubMed, to collect relevant publications on EHR QA. Our systematic screening process followed PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. A total of 4111 papers were identified for our study, and after screening based on our inclusion criteria, we obtained 47 papers for further study. The selected studies were then classified into 2 non-mutually exclusive categories depending on their scope: "EHR QA datasets" and "EHR QA models." A systematic screening process obtained 47 papers on EHR QA for final review. Out of the 47 papers, 53% (n=25) were about EHR QA datasets, and 79% (n=37) papers were about EHR QA models. It was observed that QA on EHRs is relatively new and unexplored. Most of the works are fairly recent. In addition, it was observed that emrQA is by far the most popular EHR QA dataset, both in terms of citations and usage in other papers. We have classified the EHR QA datasets based on their modality, and we have inferred that Medical Information Mart for Intensive Care (MIMIC-III) and the National Natural Language Processing Clinical Challenges datasets (ie, n2c2 datasets) are the most popular EHR databases and corpuses used in EHR QA. Furthermore, we identified the different models used in EHR QA along with the evaluation metrics used for these models. EHR QA research faces multiple challenges, such as the limited availability of clinical annotations, concept normalization in EHR QA, and challenges faced in generating realistic EHR QA datasets. There are still many gaps in research that motivate further work. This study will assist future researchers in focusing on areas of EHR QA that have possible future research directions.

Dynamic QTc interval changes observed in real-world patients with implanted insertable cardiac monitors taking class III antiarrhythmic drugs

Abstract Background QT prolongation can lead to sustained ventricular arrhythmias . Class III Antiarrhythmic drugs (AAD) are known to cause QT prolongation. Insertable cardiac monitors (ICM) have the capability to continuously monitor and detect dynamic changes in QT. Purpose To investigate the characteristics of QTc interval changes before and after antiarrhythmic drug loading in real-world ICM patients administered with Sotalol, Amiodarone, and Dofetilide in hospital. Methods The QT detection algorithm detects the T-waves and determines the QTc intervals for every beat, thus providing continuous long-term QTc trends from ICM ECG. Real-world patients from de-identified Electronic Health Record (EHR) database linked with ICM device data warehouse with implanted ICM being administered sotalol, amiodarone or dofetilide AAD for the first time in the hospital were analyzed. The patients were broken down into groups based on the AAD taken. Patients not taking AAD were selected as control. QTc intervals 7-days pre and post administered AAD are reported, and a random 14-day period was used for control patients. The paired difference in median QTc 7 days pre and post administered AAD were analyzed using the Wilcoxon signed rank test. The Wilcoxon rank-sum test was used to analyze QTc between the different groups. Results A total of 487 patients with ICM were included in this analysis: 56 sotalol patients (avg. age: 67±10 years, 50% male), 100 amiodarone patients (avg. age: 72±10 years, 50% male), 52 dofetilide patients (avg. age: 66±12 years, 50% male), and 248 control patients (avg. age: 70±13 years, 50% male). Sotalol, amiodarone, and dofetilide patients showed significant increase in QTc 1-7 days post AAD compared to 1-7 days pre AAD (figure 1). For sotalol (post AAD median[IQR] = 426.5[408.8,446.6] msec, pre AAD median[IQR] = 415.9[391,437.3] msec, p=0.004); Dofetilide (post median[IQR] = 432.4[413,458.7] msec; pre median[IQR] = 409.3[397.5,435.2] msec, p<0.001); Amiodarone (post median[IQR] = 426.3[400.9,449] msec, pre median[IQR] = 415.5[397.5,435.2] msec, p = 0.037). The QTc 1-7 days pre AAD compared to 1-7 days post AAD for control patients was insignificant p>0.05. The median change in QTc 1-7 days post AAD from 1-7 days pre AAD was 13.0, 7.5, 21.25 and -1.0 msec for sotalol, amiodarone, dofetilide, and control patients respectively (figure 2). Dofetilide patients had a significantly higher change in QTc compared to the amiodarone patients, p=0.018. QTc for all three AAD groups when compared to control group post AAD were significantly higher, p<0.001. QTc was also higher pre AAD compared to control, p=0.003, p<0.001, p=0.009 for sotalol, amiodarone, and dofetilide respectively. Conclusions There is a significant increase in QTc detected by ICM in patients receiving index drug loading of Sotalol, amiodarone and dofetilide. QTc monitoring via ICM can detect these dynamic changes in QTc allowing for early intervention.Figure 1Figure 2

Incidence of atrial fibrillation detected by an insertable cardiac monitor in a large real-world cohort of heart failure patients with reduced and preserved ejection fraction

Abstract Background Subcutaneous implantable cardiac monitors (ICM) have the capability to detect atrial fibrillation (AF) episodes with high degree of accuracy [1]. Objective We investigated the incidence of AF in patients implanted with an ICM and a clinical history of heart failure (HF) with reduced or preserved left ventricular ejection fraction (LVEF). Methods Patients with history of HF admissions who were implanted with an ICM were identified from the aggregated and de-identified electronic health record (EHR) database during 2007-2021. The device collected data were merged with the EHR data to create a de-identified database of real-world patients. Patients were included if they had ≥180 days of device follow-up. ICMs detect AF based on incoherence of RR intervals and absence of single p-wave between two R-waves [1]. All ICM detected AF episodes stored with at least 30 seconds of ECG at onset were adjudicated using an artificial intelligence (AI) model. The AI model was previously trained and validated using over 60K manually adjudicated episodes and was shown to have an accuracy of over 97% [2]. ICM detected AF episodes with AI model based true probability of AF ≥ 0.9 was analyzed. The Kaplan-Meier incidence curves for AF incidence as a function of episode duration, clinical history of AF, and LVEF are reported. Results A total of 1020 LINQ ICM patients with history of HF admission prior to implant were identified from the deidentified real-world dataset. LVEF was available in 889 (87%) patients, of whom 394 (44%) had EF <50%, and 495 (56%) had EF≥50%. NYHA class was available in 296 (29%) patients, of whom 39 (13%) were class I, 126 (43%) were class II, 112 (38%) were class III, and 19 (6%) were class IV. Patients had an average age of 68±13 years with 52% being males and had clinical history of hypertension in 95%, diabetes in 56%, coronary artery disease in 75%, AF in 57%, stroke/TIA in 51%, and renal dysfunction in 53%. A total of 911 patients with ≥180 days of follow-up (average follow-up of 25.8 months) were included. A total of 8407 episodes from 358 patients were classified by the AI model as true AF. The incidence of AF as a function of episode duration, clinical history of AF and reduced vs preserved LVEF are shown in Figures A, B, and C respectively. Incidence of AF, as detected by an ICM over 42 months of follow-up, was estimated to be 38% to 46% depending on AF episode duration. Estimated 42-month AF incidence was 23% in patients with no clinical history of AF. The 42-month AF incidence was 44% vs 47% in HF patients with reduced versus preserved LVEF. Conclusion Incidence of AF, as detected by an ICM over 3 years of follow-up, was estimated to be in close to half of ICM patients with history of HF events. Close to one fourth of the patients with no prior clinical history of AF had newly detected AF. AF incidence was similar in HF patients with preserved versus reduced LVEF in this real-world cohort.

Characterization of individuals with atherosclerotic cardiovascular disease, with and without chronic kidney disease: results from a UK electronic health record database

Abstract Introduction Elevated C-reactive protein (CRP) levels are often present in individuals with atherosclerotic cardiovascular disease (ASCVD), including in those who also have chronic kidney disease (CKD), and are associated with an increased risk of major adverse cardiovascular events (MACE). To understand unmet needs in the identification and management of at-risk individuals, we characterized patients with ASCVD in a UK database, according to CKD stage and CRP levels. Methods This study comprised separate retrospective cross-sectional and longitudinal cohort analyses of patients in the Clinical Practice Research Datalink (CPRD) Aurum database, a representative UK database of anonymized primary care records. In the cross-sectional analysis, CPRD Aurum was linked with Hospital Episode Statistics (HES) and Index of Multiple Deprivation data to assess patient characteristics on 1 Jan 2021 (defined as index date). In the cohort analysis, CPRD Aurum was linked with HES and Office for National Statistics mortality data to determine the incidence of 3-point MACE (3P-MACE) over approximately 5 years of follow-up (index date: 1 Jan 2016). Both analyses included individuals ≥ 18 years old with an ASCVD diagnosis in both primary and secondary care in the 5 years pre-index. Data were analysed using descriptive statistics. Patient characteristics were assessed by CKD stage; for individuals with CKD stage 3–4, characteristics were also assessed by CRP levels. Results In total, 225,767 and 230,055 individuals were included in the cross-sectional and cohort analyses, respectively. The prevalence of CKD stage 3–4 among individuals with ASCVD was 24.3% (Table 1). The proportion of individuals with elevated CRP (CRP ≥ 2 mg/L) and anaemia increased for successively higher CKD severity groups; in addition, glycated haemoglobin (HbA1c) ≥ 7% and elevated blood pressure were generally more frequently observed in individuals with CKD than in those without (Table 1). Individuals with ASCVD, CKD stage 3–4 and elevated CRP had higher HbA1c and total cholesterol levels and a greater proportion had obesity, anaemia and elevated blood pressure, compared with individuals with CRP < 2 mg/L (Table 2). Over 5 years’ follow-up, the incidence of 3P-MACE increased with CKD severity (Table 1) and with increasing CRP levels in individuals with ASCVD and CKD stage 3–4 (Table 2). However, only 28% of individuals with ASCVD and CKD stage 3–4 had CRP measurements. Conclusion Adults with ASCVD, CKD and elevated CRP identified from primary and secondary UK care records had a higher disease burden than individuals with CRP levels < 2 mg/L. In the UK, testing CRP levels is not routinely recommended, resulting in a high proportion of missing CRP measurements in this study. Routine testing of CRP may help in identifying patients with residual cardiovascular risk and has the potential to support risk assessment and management of patients with ASCVD and CKD.

Use of non-vitamin K antagonist oral anticoagulants in patients with atrial fibrillation across five stages of chronic kidney diseases

Abstract Background Although non-vitamin K antagonist oral anticoagulants (NOACs) were more and more commonly prescribed in the daily practice, data about the use of NOACs in atrial fibrillation (AF) patients with stage 4 and 5 chronic kidney diseases (CKD) are limited. In the present study, we compared the clinical outcomes of AF patients receiving NOACs and those receiving warfarin or non-OACs in different stages of CKD. Methods A total of 39217 AF patients aged >20 years whose data about creatinine clearance (CCr) were available were identified from the electronic health record database of a multi-center healthcare system in Taiwan. Patients were categorized into stage 1 to stage 5 CKDs, and the composite risks of ischemic stroke/systemic embolism, major bleeding and mortality were compared between non-OACs, warfarin and NOACs. Results Around 12.7% and 9.4% of study patients had a stage 4 and stage 5 CKD, respectively. Compared to non-OACs, the use of OACs (either warfarin or NOACs) were consistently associated with a lower risk of composite events from stage 1 to stage 4 CKD, but not for stage 5 CKD (warfarin: aHR 0.912, 95%CI 0.742-1.121; NOACs: aHR 0.993, 95%CI 0.653-1.537)(Figure 1). Compared to warfarin, NOACs were associated with a lower risk of adverse events for patients with stage 2 (aHR 0.630), stage 3 (aHR 0.712) and stage 4 CKD (aHR 0.749). However, the risk did not differ between warfarin and NOACs for stage 1 and stage 5 CKD (Figure 2). Conclusions Although data of the clinical trials for AF patients with stage 4 CKD were relatively lacking, NOACs should still be considered for this AF population. More data are necessary to guide optimal stroke prevention strategy for patients with stage 5 CKD.

Application of a digital quality measure for cancer diagnosis in Epic Cosmos.

Missed and delayed cancer diagnoses are common, harmful, and often preventable. We previously validated a digital quality measure (dQM) of emergency presentation (EP) of lung cancer in 2 US health systems. This study aimed to apply the dQM to a new national electronic health record (EHR) database and examine demographic associations. We applied the dQM (emergency encounter followed by new lung cancer diagnosis within 30 days) to Epic Cosmos, a deidentified database covering 184 million US patients. We examined dQM associations with sociodemographic factors. The overall EP rate was 19.6%. EP rate was higher in Black vs White patients (24% vs 19%, P < .001) and patients with younger age, higher social vulnerability, lower-income ZIP code, and self-reported transport difficulties. We successfully applied a dQM based on cancer EP to the largest US EHR database. This dQM could be a marker for sociodemographic vulnerabilities in cancer diagnosis.

B-110 Descriptive Analysis of Multiple Myeloma Participants in the NIH All of Us Research Program Cohort: Insights into Demographics Disparities

Abstract Background Multiple Myeloma (MM) is a complex hematological malignancy characterized by significant heterogeneity in participant characteristics and treatment responses. The All of Us Research Program seeks to enroll 1 million diverse participants who share multiple sources of data, providing unique opportunities for research. A comprehensive descriptive analysis of MM participants within Electronic Health Records (EHR) databases can provide valuable insights into the disease's epidemiology, participant demographics, comorbidities, and treatment outcomes. Objective Using AI/ML predictive analytics, this study conducted a detailed descriptive analysis of MM participants’ EHR data in the All of Us Research Program cohort, which is enrolling 1 million diverse participants who share multiple sources of data. Methods We extracted data for participants diagnosed with MM from an extensive EHR database. The analysis included participant demographics (age, gender, race, ethnicity), diagnostic laboratory tests (SPEP, UPEP, CBC, renal function), comorbidities (renal impairment, bone disease, anemia, cardiovascular diseases), treatment details (chemotherapy, stem cell transplantation), and genetic markers (cytogenetic abnormalities). Results The analysis revealed diverse demographic patterns and a wide range of comorbidities associated with MM (Table 1). Conclusions The analysis demonstrated marked disparities in MM prevalence among African Americans and “More than one race,” revealing social and medical disparities. Implications: By understanding the diverse profiles of MM participants, stakeholders can more effectively address disparities in healthcare outcomes (1). Reference: 1. Zeng, C., Schlueter, D. J., Tran, T. C., Babbar, A., Cassini, T., Bastarache, L. A., &amp; Denny, J. C. (2024). Comparison of phenomic profiles in the All of Us Research Program against the US general population and the UK Biobank. Journal of the American Medical Informatics Association. Retrieved from, https://doi.org/10.1093/jamia/ocad260

A genome-wide Association study of the Count of Codeine prescriptions

Opioid prescription records in existing electronic health record (EHR) databases are a potentially useful, high-fidelity data source for opioid use-related risk phenotyping in genetic analyses. Prescriptions for codeine derived from EHR records were used as targeting traits by screening 16 million patient-level medication records. Genome-wide association analyses were then conducted to identify genomic loci and candidate genes associated with different count patterns of codeine prescriptions. Both low- and high-prescription counts were captured by developing 8 types of phenotypes with selected ranges of prescription numbers to reflect potentially different levels of opioid risk severity. We identified one significant locus associated with low-count codeine prescriptions (1, 2 or 3 prescriptions), while up to 7 loci were identified for higher counts (≥ 4, ≥ 5, ≥6, or ≥ 7 prescriptions), with a strong overlap across different thresholds. We identified 9 significant genomic loci with all-count phenotype. Further, using the polygenic risk approach, we identified a significant correlation (Tau = 0.67, p = 0.01) between an externally derived polygenic risk score for opioid use disorder and numbers of codeine prescriptions. As a proof-of-concept study, our research provides a novel and generalizable phenotyping pipeline for the genomic study of opioid-related risk traits.

COVID-19 Breakthrough Infections Among People With HIV: A Statewide Cohort Analysis.

This study aims to identify COVID-19 breakthrough infections among people with HIV (PWH) across different phases of the pandemic and explore whether differential immune dysfunctions are associated with breakthrough infections. This retrospective population-based cohort study used data from an integrated electronic health record (EHR) database in South Carolina (SC). Breakthrough infection was defined as the first COVID-19 diagnosis documented in the state agency after the date an individual was fully vaccinated (ie, 2 doses of Pfizer/BNT162b2 or Moderna/mRNA-1273, or 1 dose of Janssen/Ad26.COV2.S) through June 14, 2022. We analyzed the risk and associated factors of the outcome using Cox proportional hazards models. Among 7596 fully vaccinated PWH, the overall rate of breakthrough infections was 118.95 cases per 1000 person-years. When compared with the alpha-dominant period, the breakthrough infection rate was higher during both delta-dominant (HR: 1.50; 95% CI: 1.25 to 1.81) and omicron-dominant (HR: 2.86; 95% CI: 1.73 to 4.73) periods. Individuals who received a booster dose had a lower likelihood of breakthrough infections (HR: 0.19; 95% CI: 0.15 to 0.24). There was no association of breakthrough infections with degree of HIV viral suppression, but a higher CD4 count was significantly associated with fewer breakthroughs among PWH (>500 vs <200 cells/mm3: HR: 0.68; 95% CI: 0.49 to 0.94). In our PWH population, the incidence of breakthrough infections was high (during both delta-dominant and omicron-dominant periods) and mainly associated with the absence of a booster dose in patients older than 50 years, with comorbidities and low CD4 count.

Respiratory Syncytial Virus and US Pediatric Intensive Care Utilization

Bronchiolitis is the most common diagnosis necessitating respiratory support and pediatric intensive care, and respiratory syncytial virus (RSV) is the most common cause of bronchiolitis. In 2023, the monoclonal antibody nirsevimab and the maternal RSVpreF vaccine were implemented to prevent RSV in infants. To determine the potential association of novel RSV prevention strategies with pediatric intensive care unit (ICU) utilization. This retrospective cross-sectional study examined US pediatric ICU encounters with and without RSV from January 1, 2017, to June 1, 2023. Data were from Oracle Cerner RealWorld Data, a national electronic health record database. All children (aged >1 day and <18 years) admitted to an ICU during the study period were included in the analysis. Statistical analysis was performed from February to May 2024. The primary outcomes were ICU encounters with RSV and ICU encounters with RSV eligible for RSV prevention. There were 119 782 ICU encounters from 53 hospitals; 65 757 encounters (54.9%) were male; median (IQR) age was 4.5 (1.1-12.5) years, median (IQR) ICU length of stay was 1.8 (1.0-3.9) days; 13 702 ICU encounters (11.4%) had RSV, of which 38.6% (5217) were eligible for prevention (4.4% of total ICU encounters). Encounters with RSV accounted for 21.1% (109 334) of ICU days, of which 43.8% (47 888) were eligible for prevention (9.2% of total ICU days). Encounters with RSV were associated with a greater use of positive pressure ventilation (4074 [29.7%] vs 18 821 [17.7%]; P < .001) and vasoactive medications (3057 [22.3%] vs 18 570 [17.5%]; P < .001) compared with those without RSV. The findings of this retrospective cross-sectional study of RSV and US pediatric intensive care utilization suggest that if 65% to 85% of eligible children received RSV prevention, an estimated 2.1% to 2.8% reduction in pediatric ICU encounters and an estimated 4.5% to 5.9% reduction in pediatric ICU days could be achieved.

Calibrating Observational Health Record Data Against a Randomized Trial

The conditions required for health record data sources to accurately assess treatment effectiveness remain unclear. Emulation of randomized clinical trials (RCTs) with health record data and subsequent calibration of the results can help elucidate this. To pilot an emulation of the KEYNOTE-189 RCT using a commercially available electronic health record (EHR) data source. This retrospective cohort study used an EHR database spanning from April 2007 to February 2023. Follow-up began on treatment initiation and proceeded until an outcome event, loss to follow-up, end of data, or end of study period (640 days). The population-based cohort was ascertained from EHRs provided by 52 health systems across the US. Eligibility criteria were defined as closely as possible to the benchmark RCT. Patients with non-small cell lung cancer initiating first-line treatment for metastatic disease were included. Patients with evidence of squamous non-small cell lung cancer, primary nonlung malignant neoplasms, or identified EGFR/ALK variations were excluded. Data were analyzed from June to October 2023. Initiation of first-line pembrolizumab and chemotherapy and chemotherapy alone. Chemotherapy in both groups was defined as a combination of pemetrexed and platinum-based (carboplatin or cisplatin) therapy. Outcomes of interest were 12-month survival probability and mortality hazard ratio (HR). A total of 1854 patients (mean [SD] age, 63.7 [9.6] years; 971 [52.4%] men) were eligible, including 589 patients who initiated pembrolizumab and chemotherapy and 1265 patients who initiated chemotherapy only. The cohort included 364 Black patients (19.6%) and 1445 White patients (77.9%). The 12-month survival probabilities were 0.60 (95% CI, 0.54-0.65) in the pembrolizumab group and 0.58 (95% CI, 0.55-0.62) in the chemotherapy-only group, compared with 0.69 (95% CI, 0.64-0.74) in the KEYNOTE-189 pembrolizumab group and 0.49 (95% CI, 0.42-0.56) in the KEYNOTE-189 chemotherapy-only group. The mortality HR was 0.95 (95% CI, 0.78-1.16), compared with 0.49 (95% CI, 0.38-0.64) in the KEYNOTE-189 RCT. In this cohort study piloting an RCT emulation, results were incongruous with the benchmark trial. Differences in patient treatment and data capture between the RCT and EHR populations, confounding by indication, treatment crossover, and accuracy of captured diagnoses may explain these findings. Future feasibility assessments will require data sources to have important oncology-specific measures curated.

Prevalent Metformin Use in Adults With Diabetes and the Incidence of Long COVID: An EHR-Based Cohort Study From the RECOVER Program.

Studies show metformin use before and during SARS-CoV-2 infection reduces severe COVID-19 and postacute sequelae of SARS-CoV-2 (PASC) in adults. Our objective was to describe the incidence of PASC and possible associations with prevalent metformin use in adults with type 2 diabetes mellitus (T2DM). This is a retrospective cohort analysis using the National COVID Cohort Collaborative (N3C) and Patient-Centered Clinical Research Network (PCORnet) electronic health record (EHR) databases with an active comparator design that examined metformin-exposed individuals versus nonmetformin-exposed individuals who were taking other diabetes medications. T2DM was defined by HbA1c ≥6.5 or T2DM EHR diagnosis code. The outcome was death or PASC within 6 months, defined by EHR code or computable phenotype. In the N3C, the hazard ratio (HR) for death or PASC with a U09.9 diagnosis code (PASC-U09.0) was 0.79 (95% CI 0.71-0.88; P < 0.001), and for death or N3C computable phenotype PASC (PASC-N3C) was 0.85 (95% CI 0.78-0.92; P < 0.001). In PCORnet, the HR for death or PASC-U09.9 was 0.87 (95% CI 0.66-1.14; P = 0.08), and for death or PCORnet computable phenotype PASC (PASC-PCORnet) was 1.04 (95% CI 0.97-1.11; P = 0.58). Incident PASC by diagnosis code was 1.6% metformin vs. 2.0% comparator in the N3C, and 2.1% metformin vs. 2.5% comparator in PCORnet. By computable phenotype, incidence was 4.8% metformin and 5.2% comparator in the N3C and 24.7% metformin vs. 26.1% comparator in PCORnet. Prevalent metformin use is associated with a slightly lower incidence of death or PASC after SARS-CoV-2 infection. PASC incidence by computable phenotype is higher than by EHR code, especially in PCORnet. These data are consistent with other observational analyses showing prevalent metformin is associated with favorable outcomes after SARS-CoV-2 infection in adults with T2DM.

Real world analysis of treatment change and response in adults with attention-deficit/hyperactivity disorder (ADHD) alone and with concomitant psychiatric comorbidities: results from an electronic health record database study is the United States

BackgroundThe objectives of this study were to examine the association of psychiatric comorbidities and patient characteristics with treatment change and response as well as to assess the association between treatment change and healthcare resource utilization (HCRU) among adult patients with attention-deficit/hyperactivity disorder (ADHD) and psychiatric comorbidities.MethodsDe-identified electronic health records from the NeuroBlu Database (2002–2021) were used to select patients ≥ 18 years with ADHD who were prescribed ADHD-specific medication. The index date was set as the first prescription of ADHD medication. The outcomes were treatment change (discontinuation, switch, add-on, or drop) and HCRU (inpatient, outpatient, composite) within 12 months of follow-up. Cox proportional-hazard model was used to assess the association between clinical and demographic patient characteristics and treatment change, while generalized linear model with negative binomial distribution and log link function was used to assess the association between key risk factors linked to treatment change and HCRU rates.ResultsA total of 3,387 patients with ADHD were included (ADHD only: 1,261; ADHD + major depressive disorder (MDD): 755; ADHD + anxiety disorder: 467; ADHD + mood disorder: 164). Nearly half (44.8%) of the study cohort experienced a treatment change within the 12-month follow-up period. Treatment switch and add-on were more common in patients with ADHD and comorbid MDD and anxiety disorder (switch: 18.9%; add-on: 20.5%) compared to other cohorts (range for switch: 8.5–13.6%; range for add-on: 8.9–12.1%) Survival analysis demonstrated that the probability of treatment change within 12 months from treatment initiation in the study cohort was estimated to be 42.4%. Outpatient visit rates statistically significantly increased from baseline (mean [SD] 1.03 [1.84] visits/month) to 3 months post-index (mean [SD] 1.62 [1.91] visits/month; p < 0.001), followed by a gradual decline up to 12 months post-index. Being prescribed both a stimulant and a non-stimulant at index date was statistically significantly associated with increased risk of treatment change (adjusted hazard ratio: 1.64; 95% CI: 1.13, 2.38; p = 0.01).ConclusionsThis real-world study found that treatment change was common among patients with ADHD and psychiatric comorbidities. These findings support the need for future studies to examine the unmet medical and treatment needs of this complex patient population.

Cross-EHR validation of antidepressant response algorithm and links with genetics of psychiatric traits.

Antidepressants are commonly prescribed medications in the United States, however, factors underlying response are poorly understood. Electronic health records (EHRs) provide a cost-effective way to create and test response algorithms on large, longitudinal cohorts. We describe a new antidepressant response algorithm, validation in two independent EHR databases, and genetic associations with antidepressant response. We deployed the algorithm in EHRs at Vanderbilt University Medical Center (VUMC), the All of Us Research Program, and the Mass General Brigham Healthcare System (MGB) and validated response outcomes with patient health questionnaire (PHQ) scores. In a meta-analysis across all sites, worse antidepressant response associated with higher PHQ-8 scores (beta = 0.20, p-value = 1.09 x 10 -18 ). We used polygenic scores to investigate the relationship between genetic liability of psychiatric disorders and response to first antidepressant trial across VUMC and MGB. After controlling for depression diagnosis, higher polygenic scores for depression, schizophrenia, bipolar, and cross-disorders associated with poorer response to the first antidepressant trial (depression: p-value = 2.84 x 10 -8 , OR = 1.07; schizophrenia: p-value = 5.93 x 10 -4 , OR = 1.05; bipolar: p-value = 1.99 x 10 -3 , OR = 1.04; cross-disorders: p-value = 1.03 x 10 -3 , OR = 1.05). Overall, we demonstrate our antidepressant response algorithm can be deployed across multiple EHR systems to increase sample size of genetic and epidemiologic studies of antidepressant response.

Automated Medical Records Review for Mild Cognitive Impairment and Dementia.

Unstructured and structured data in electronic health records (EHR) are a rich source of information for research and quality improvement studies. However, extracting accurate information from EHR is labor-intensive. Here we introduce an automated EHR phenotyping model to identify patients with Alzheimer's Disease, related dementias (ADRD), or mild cognitive impairment (MCI). We assembled medical notes and associated International Classification of Diseases (ICD) codes and medication prescriptions from 3,626 outpatient adults from two hospitals seen between February 2015 and June 2022. Ground truth annotations regarding the presence vs. absence of a diagnosis of MCI or ADRD were determined through manual chart review. Indicators extracted from notes included the presence of keywords and phrases in unstructured clinical notes, prescriptions of medications associated with MCI/ADRD, and ICD codes associated with MCI/ADRD. We trained a regularized logistic regression model to predict the ground truth annotations. Model performance was evaluated using area under the receiver operating curve (AUROC), area under the precision-recall curve (AUPRC), accuracy, specificity, precision/positive predictive value, recall/sensitivity, and F1 score (harmonic mean of precision and recall). Thirty percent of patients in the cohort carried diagnoses of MCI/ADRD based on manual review. When evaluated on a held-out test set, the best model using clinical notes, ICDs, and medications, achieved an AUROC of 0.98, an AUPRC of 0.98, an accuracy of 0.93, a sensitivity (recall) of 0.91, a specificity of 0.96, a precision of 0.96, and an F1 score of 0.93 The estimated overall accuracy for patients randomly selected from EHRs was 99.88%. Automated EHR phenotyping accurately identifies patients with MCI/ADRD based on clinical notes, ICD codes, and medication records. This approach holds potential for large-scale MCI/ADRD research utilizing EHR databases.

Use of Palliative Care Among Adults With Newly Diagnosed Heart Failure: Insights From a US National Insured Patient Sample.

Despite the known benefits for individuals with heart failure (HF), incomplete data suggest a low use of palliative care (PC) for HF in the United States. We aimed to investigate the national PC use for adults with HF by determining when they received their first PC consultation (PCC) and the associations with clinical factors following diagnosis of HF. We conducted a retrospective cohort study in a national all-payer electronic health record database to identify adults (aged ≥18 years) with newly diagnosed HF between 2011 and 2018. The proportion of those who received PCC within 5 years following a diagnosis of HF, and associations of time to first PCC with patient characteristics and HF-specific clinical markers were determined. We followed 127 712 patients for a median of 792 days, of whom 18.3% received PCC in 5 years. Shorter time to receive PCC was associated with diagnoses of HF in 2016 to 2018 (compared with 2010-2015: adjusted hazard ratio [aHR], 1.421 [95% CI, 1.370-1.475]), advanced HF (aHR, 2.065 [95% CI, 1.940-2.198]), cardiogenic shock (aHR, 2.587 [95% CI, 2.414-2.773]), implantable cardioverter-defibrillator (aHR, 5.718 [95% CI, 5.327-6.138]), and visits at academic medical centers (aHR, 1.439 [95% CI, 1.381-1.500]). Despite an expanded definition of PC and recommendations by professional societies, PC for HF remains low in the United States. Racial and geographic variations in access and use of PC exist for patients with HF. Future studies should interrogate the mechanisms of PC underusage, especially before advanced stages, and address barriers to PC services across the health care system.

Reliable generation of privacy-preserving synthetic electronic health record time series via diffusion models.

Electronic health records (EHRs) are rich sources of patient-level data, offering valuable resources for medical data analysis. However, privacy concerns often restrict access to EHRs, hindering downstream analysis. Current EHR deidentification methods are flawed and can lead to potential privacy leakage. Additionally, existing publicly available EHR databases are limited, preventing the advancement of medical research using EHR. This study aims to overcome these challenges by generating realistic and privacy-preserving synthetic EHRs time series efficiently. We introduce a new method for generating diverse and realistic synthetic EHR time series data using denoizing diffusion probabilistic models. We conducted experiments on 6 databases: Medical Information Mart for Intensive Care III and IV, the eICU Collaborative Research Database (eICU), and non-EHR datasets on Stocks and Energy. We compared our proposed method with 8 existing methods. Our results demonstrate that our approach significantly outperforms all existing methods in terms of data fidelity while requiring less training effort. Additionally, data generated by our method yield a lower discriminative accuracy compared to other baseline methods, indicating the proposed method can generate data with less privacy risk. The proposed model utilizes a mixed diffusion process to generate realistic synthetic EHR samples that protect patient privacy. This method could be useful in tackling data availability issues in the field of healthcare by reducing barrier to EHR access and supporting research in machine learning for health. The proposed diffusion model-based method can reliably and efficiently generate synthetic EHR time series, which facilitates the downstream medical data analysis. Our numerical results show the superiority of the proposed method over all other existing methods.

Quantitative Measurement of the Direct Public Health Impact of Medicines Withdrawals in Europe: Development of a Modelling Method and Proof-of-Concept Study to Estimate the Morbidity and Mortality Prevented by Regulatory Action.

Removing medicines from market may benefit public health by preventing adverse drug reactions (ADRs), which should be quantified. This study's aim was to identify a model to quantify the impact of medicines' marketing authorisation (MA) withdrawal and revocation in terms of preventing morbidity and mortality. MA withdrawals and revocations for safety reasons in France, Germany and/or the United Kingdom between July 2012 and December 2016 were identified for prescription medicines. Annual exposure was estimated for each medicine, using IQVIA Medical Research Data (IMRD)-France, IMRD-Germany and IMRD-UK primary care electronic health record databases. European Medicines Agency records provided reasons for regulatory action for each medicine. Absolute risks of ADRs which led to MA withdrawal were estimated for patients exposed to each medicine by systematic review of quantitative research. Public health impact, expressed as annual number of ADRs avoided, was estimated by modelling exposure and ADR risk. Four MA withdrawals and two revocations met study inclusion criteria. Each product's usage decreased following MA withdrawal or revocation. Absolute risk for ADRs was 0.1%-41.25%. To estimate impact of each withdrawal or revocation, its average annual exposure within each IMRD population was multiplied by the absolute risk to give the crude number of ADRs prevented annually due to regulatory action. This model quantifies the public health impact of MA withdrawal and revocation in terms of serious morbidity, resulting from eliminated or reduced usage of medicines. This method can be applied to products in other settings to quantify the impact of other pharmacovigilance actions.

Validation of diagnostic coding for chronic obstructive pulmonary disease in an electronic health record system in Hong Kong.

Electronic health record databases can facilitate epidemiology research regarding diseases such as chronic obstructive pulmonary disease (COPD), a common medical condition worldwide. We aimed to assess the validity of International Classification of Diseases, 9th Revision (ICD-9) code algorithms for identifying COPD in Hong Kong's territory-wide electronic health record system, the Clinical Data Analysis and Reporting System (CDARS). Adult patients diagnosed with COPD at all public hospitals in Hong Kong and specifically at Queen Mary Hospital from 2011 to 2020 were identified using the ICD-9 code 496 (Chronic airway obstruction, not elsewhere classified) within the CDARS. Two respiratory specialists reviewed clinical records and spirometry results to confirm the presence of COPD in a randomly selected group of cases. During the study period, 93 971 and 2479 patients had the diagnostic code for COPD at all public hospitals in Hong Kong and specifically at Queen Mary Hospital, respectively. Two hundred cases were randomly selected from Queen Mary Hospital for validation using medical records and spirometry results. The overall positive predictive value was 81.5% (95% confidence interval=76.1%-86.9%). We also developed an algorithm to identify COPD cases in our cohort. This study represents the first validation of ICD-9 coding for COPD in the CDARS. Our findings demonstrated that the ICD-9 code 496 is a reliable indicator for identifying COPD cases, supporting the use of the CDARS database for further clinical research concerning COPD.

Prescribing Inertia or Not? Quantitative Investigation of Loop Diuretics Prescribing after Palliative Care Consultation among Patients with Heart Failure

Purpose: Loop Diuretics (LD) are the first-line pharmacotherapy to address Heart Failure (HF)-associated edema and dyspnea. However, LD causes frequent urinary, resulting in inconvenience and possibly undermining the quality of life. While prescription adjustment is an essential part of Palliative Care Consultation (PCC), it remains unclear how PCC affects the deprescribing of diuretics for adults with HF. Methods: We conducted a pre-post analysis of the percentage of HF patients who were prescribed LD in a national Electronic Health Record (EHR) database 12 months before and after the first PCC. The difference in prescription rates between the periods was determined. Adjusted associations of post-PCC LD prescription with pre-PCC LD prescription and patient’s characteristics, insurance, provider type, and clinical factors were quantified. Results: From 2010 to 2018, 5,969 patients with newly diagnosed HF received at least one PCC, among whom 2,539 (42.5%) were prescribed LD before and 1,552 (26.0%) after their first PCC. Despite a decrease in LD prescription rate encompassing the date of PCC, post-PCC LD prescribing was strongly associated with pre-PCC prescribing (aOR[95%CI] 3.2[2.8,3.7]) and varied by age at first PCC, year of HF diagnosis (aOR[95%CI] 2.1[1.9,2.4]) and months from HF diagnosis to first PCC. While our finding demonstrates reduced polypharmacy associated with PCC, the strong association between pre- and post-PCC indicates reverse therapeutic inertia. Future research should investigate the benefits and costs of polypharmacy among specific patient groups to help develop personalized treatment for HF.