Computational Phenotyping Research Articles

Long-COVID incidence proportion in adults and children between 2020 and 2024.

Incidence estimates of post-acute sequelae of SARS-CoV-2 infection, also known as long-COVID, have varied across studies and changed over time. We estimated long-COVID incidence among adult and pediatric populations in three nationwide research networks of electronic health records (EHR) participating in the RECOVER Initiative using different classification algorithms (computable phenotypes). This EHR-based retrospective cohort study included adult and pediatric patients with documented acute SARS-CoV-2 infection and two control groups-- contemporary COVID-19 negative and historical patients (2019). We examined the proportion of individuals identified as having symptoms or conditions consistent with probable long-COVID within 30-180 days after COVID-19 infection (incidence proportion). Each network (the National COVID Cohort Collaborative (N3C), National Patient-Centered Clinical Research Network (PCORnet), and PEDSnet) implemented its own long-COVID definition. We introduced a harmonized definition for adults in a supplementary analysis. Overall, 4% of children and 10-26% of adults developed long-COVID, depending on computable phenotype used. Excess incidence among SARS-CoV-2 patients was 1.5% in children and ranged from 5-6% among adults, representing a lower-bound incidence estimation based on our control groups. Temporal patterns were consistent across networks, with peaks associated with introduction of new viral variants. Our findings indicate that preventing and mitigating long-COVID remains a public health priority. Examining temporal patterns and risk factors of long-COVID incidence informs our understanding of etiology and can improve prevention and management.

The Marriage of Computable Phenotypes With Machine Learning—A Pathway to Evidence-Based Care for Critically Ill Children

The Marriage of Computable Phenotypes With Machine Learning—A Pathway to Evidence-Based Care for Critically Ill Children

578. Metformin reduces the risk of Long COVID or Death over 6 months in an Emulated Target Trial of Primarily Omicron-infected Adults without Diabetes or Prediabetes: a New-User, Active-Comparator Analysis Using the National COVID Cohort Collaborative (N3C) Electronic Health Record Database. This research was supported in part by the Intramural/Extramural research program of the National Center for Advancing Translational Science,

Abstract Background Metformin has decreased SARS-CoV-2 RNA in 4 cell lines. In a RCT of > 1,000 majority vaccinated outpatients, COVID-19 related ED visits/Hospitalization/Death occurred in 4.7% of the metformin vs 9.0% of the placebo group by Day 28; clinician-diagnosed Long Covid (LC) occurred in 6.2% of metformin vs 10.3% of placebo by Day 300; and 14% of metformin vs 23% of the placebo group had detectable nasal viral load on Day 10. In a trial of 20 adults, 60% of metformin vs 100% of placebo had detectable viral load by Day 4. Observational analyses report associations between prevalent metformin and less severe acute COVID. Given these data, we assessed whether metformin currently prevents LC. Cumulative Incidence Curve Cumulative incidence curves for the outcome of Long Covid/Death in the 180 Days after a metformin prescription (blue) or active comparator prescription (red). The increase in outcomes at Day 90 reflects the computable phenotype, which cannot be computed until 90 days after infection. Per data use requirements in the National Covid Cohort Collaborative (N3C), the starting number at risk is not shown due to small cell sizes or the ability to calculate small cell sizes. Methods We emulated a randomized trial of metformin vs. control in SARS-CoV-2-infected outpatients. Intervention: prescription for metformin within 6 days of infection. Control: prescription for fluvoxamine, fluticasone, ivermectin, or montelukast (drugs used off-label for Covid but clinical trials have shown no effect on acute COVID outcomes). Exclusions: age < 18; metformin or comparator prescribed within 365 days; indication for chronic metformin use; contraindication for metformin or control. Outcome: a composite of LC or Death (LC/D), LC defined by U09.9 or a symptom/condition based computable phenotype. We used entropy balancing to estimate the average treatment effect with a weighted log linear model. This is a forest plot showing metformin use versus active comparator use and the development of Long Covid/Death by Day 180. The black squares represent risk ratios (RR), and the lines represent 95% confidence intervals. The number of outcomes, denominators, and exact percentages are sometimes omitted so that cell sizes <10 are not able to be calculated, in accordance with the data use requirements in the National Covid Cohort Collaborative (N3C). The Pre-Omicron Era is larger than the Omicron era in both the Metformin and Comparator Cohorts, but the denominator for the Day 0-6 sample is not shown because of cell sizes <10. The subgroup of those with new Paxlovid use was too small to analyze, so we present only the subgroup who did not also receive a prescription for Paxlovid. Results After balancing, the standardized mean differences were < 0.01. Overall, most (5,787/9,660, 59.9%) were infected during Omicron. In the metformin and control groups, 16% and 17% were Black; 16% and 13% were Hispanic, respectively. In the metformin group, 10/248 (4.0%) developed LC/D vs. 21/248 (8.5%) in the control group, RR 0.47 (95% CI 0.25 to 0.89). For prescriptions on Days 0-1 relative to infection the RR was 0.39 (95% CI 0.12-1.24); for prescriptions on Days 0-14 the RR was 0.75 (95% CI 0.52-1.08). Cohort Balance Cohort balance of the standardized mean differences (SMD) before weighting (in pink), and after weighting (in teal), for the primary analysis. Conclusion Metformin prescribed within a week of SARS-CoV-2 infection was associated with a 53% reduction in LC/D. Anything between an 11% to 75% reduction in risk is highly compatible with our data. Important questions remain: are results consistent in randomized clinical trial data from adults with a normal body mass index and those previously infected with SARS-CoV-2.Table 1:Individuals who experienced death before the prescription could be started were excluded from the trial emulation to mimic both target trials (COVID-OUT and ACTIV-6), as they were both decentralized trials that entailed medication delivery to the home. Those who experienced hospitalization between -1 to 3 Days were also excluded to mimic the target trial: those who were hospitalized on Day -1 to 0 relative to infection could not have received and taken an outpatient prescription of metformin or comparator for their SARS-CoV-2; those hospitalized on Days 1 to 2 likely spent at least one day in the emergency department before hospitalization, so it is also unlikely that they received an outpatient prescription, obtained the prescription from a pharmacy, and took a dose before being hospitalized. Disclosures All Authors: No reported disclosures

Helping authors produce FAIR taxonomic data: evaluation of an author-driven phenotype data production prototype.

It is well-known that the use of vocabulary in phenotype treatments is often inconsistent. An earlier survey of biologists who create or use phenotypic characters revealed that this lack of standardization leads to ambiguities, frustrating both the consumers and producers of phenotypic data. Such ambiguities are challenging for biologists, and more so for Artificial Intelligence, to resolve. That survey also indicated a strong interest in a new authoring workflow supported by ontologies to ensure published phenotype data are FAIR (Findable, Accessible, Interoperable, and Reusable) and suitable for large-scale computational analyses. In this article, we introduce a prototype software system designed for authors to produce computational phenotype data. This platform includes a web-based, ontology-enhanced editor for taxonomic characters (Character Recorder), an Ontology Backend holding standardized vocabulary (the Cared Ontology), and a mobile application for resolving ontological conflicts (Conflict Resolver). We present two formal user evaluations of Character Recorder, the main interface authors would interact with to produce FAIR data. The evaluations were conducted with undergraduate biology students and Carex experts. We evaluated Character Recorder against Microsoft Excel on their effectiveness, efficiency, and the cognitive demands of the users in producing computable taxon-by-character matrices. The evaluations showed that Character Recorder is quickly learnable for both student and professional participants, with its cognitive demand comparable to Excel's. Participants agreed that the quality of the data Character Recorder yielded was superior. Students praised Character Recorder's educational value, while Carex experts were keen to recommend it and help evolve it from a prototype into a comprehensive tool. Feature improvements recommended by expert participants have been implemented after the evaluation.

Computable Phenotypes for Respiratory Viral Infections in the All of Us Research Program.

Electronic health records (EHRs) contain rich temporal data about infectious diseases, but an optimal approach to identify infections remains undefined. Using the All of Us Research Program, we developed computable phenotypes for respiratory viruses by integrating billing codes, prescriptions, and laboratory results within 90-day episodes. Phenotypes computed from 265,222 participants yielded cohorts ranging from 238 (adenovirus) to 28,729 (SARS-CoV-2) cases. Virus-specific billing codes showed varied sensitivity (8-67%) and high positive predictive value (90-97%), except for influenza virus and SARS-CoV-2 where lower PPV (69-70%) improved with increasing billing codes. Identified infections exhibited expected seasonal patterns and virus proportions when compared with CDC data. This integrated approach identified episodic disease more effectively than individual components alone and demonstrated utility in identifying severe infections. The method enables large-scale studies of host genetics, health disparities, and clinical outcomes across episodic diseases.

Large-scale audiometric phenotyping identifies distinct genes and pathways involved in hearing loss subtypes.

Age-related hearing loss affects one-third of the population over 65 years. However, the diverse pathologies underlying these heterogenous phenotypes complicate genetic studies. To overcome challenges associated with accurate phenotyping for older adults with hearing loss, we applied computational phenotyping approaches based on audiometrically measured hearing loss. This novel phenotyping strategy uncovered distinct genetic variants associated with sensory and metabolic hearing loss. Sex-stratified analyses of these sexually dimorphic hearing loss phenotypes revealed a novel locus of relevance to sensory hearing loss in males, but not females. Enrichment analyses revealed that genes involved in frontotemporal dementia were implicated in metabolic hearing loss, while genes relating to sensory processing of sound by hair cells were implicated in sensory hearing loss. Our study has enhanced our understanding of these two distinct hearing loss phenotypes, representing the first step in the development of more precise treatments for these pathologically distinct hearing loss phenotypes.

Toward a Computable Phenotype for Determining Eligibility of Lung Cancer Screening Using Electronic Health Records.

Lung cancer screening (LCS) has the potential to reduce mortality and detect lung cancer at its early stages, but the high false-positive rate associated with low-dose computed tomography (LDCT) for LCS acts as a barrier to its widespread adoption. This study aims to develop computable phenotype (CP) algorithms on the basis of electronic health records (EHRs) to identify individual's eligibility for LCS, thereby enhancing LCS utilization in real-world settings. The study cohort included 5,778 individuals who underwent LDCT for LCS from 2012 to 2022, as recorded in the University of Florida Health Integrated Data Repository. CP rules derived from LCS guidelines were used to identify potential candidates, incorporating both structured EHR and clinical notes analyzed via natural language processing. We then conducted manual reviews of 453 randomly selected charts to refine and validate these rules, assessing CP performance using metrics, for example, F1 score, specificity, and sensitivity. We developed an optimal CP rule that integrates both structured and unstructured data, adhering to the US Preventive Services Task Force 2013 and 2020 guidelines. This rule focuses on age (55-80 years for 2013 and 50-80 years for 2020), smoking status (current, former, and others), and pack-years (≥30 for 2013 and ≥20 for 2020), achieving F1 scores of 0.75 and 0.84 for the respective guidelines. Including unstructured data improved the F1 score performance by up to 9.2% for 2013 and 12.9% for 2020, compared with using structured data alone. Our findings underscore the critical need for improved documentation of smoking information in EHRs, demonstrate the value of artificial intelligence techniques in enhancing CP performance, and confirm the effectiveness of EHR-based CP in identifying LCS-eligible individuals. This supports its potential to aid clinical decision making and optimize patient care.

Development and validation of computable social phenotypes for health-related social needs.

Measurement of health-related social needs (HRSNs) is complex. We sought to develop and validate computable phenotypes (CPs) using structured electronic health record (EHR) data for food insecurity, housing instability, financial insecurity, transportation barriers, and a composite-type measure of these, using human-defined rule-based and machine learning (ML) classifier approaches. We collected HRSN surveys as the reference standard and obtained EHR data from 1550 patients in 3 health systems from 2 states. We followed a Delphi-like approach to develop the human-defined rule-based CP. For the ML classifier approach, we trained supervised ML (XGBoost) models using 78 features. Using surveys as the reference standard, we calculated sensitivity, specificity, positive predictive values, and area under the curve (AUC). We compared AUCs using the Delong test and other performance measures using McNemar's test, and checked for differential performance. Most patients (63%) reported at least one HRSN on the reference standard survey. Human-defined rule-based CPs exhibited poor performance (AUCs=.52 to .68). ML classifier CPs performed significantly better, but still poor-to-fair (AUCs = .68 to .75). Significant differences for race/ethnicity were found for ML classifier CPs (higher AUCs for White non-Hispanic patients). Important features included number of encounters and Medicaid insurance. Using a supervised ML classifier approach, HRSN CPs approached thresholds of fair performance, but exhibited differential performance by race/ethnicity. CPs may help to identify patients who may benefit from additional social needs screening. Future work should explore the use of area-level features via geospatial data and natural language processing to improve model performance.

Developing a Computational Phenotype of the Fourth Universal Definition of Myocardial Infarction for Inpatients.

Background: The fourth universal definition of myocardial infarction (MI) introduced the differentiation of acute myocardial injury from MI. In this study, we developed a computational phenotype for distinct identification of acute myocardial injury and MI within electronic medical records (EMRs). Methods: Two cohorts were used from a Calgary-wide EMR system: a chart review of 3042 randomly selected inpatients from Dec 2014 to Jun 2015; and 11,685 episodes of care that included cardiac catheterization from Jan 2013 to Apr 2017. Electrocardiogram (ECG) reports were processed using natural language processing and combined with high-sensitivity troponin lab results to classify patients as having an acute myocardial injury, MI, or neither. Results: For patients with an MI diagnosis, only 64.0% (65.7%) in the catheterized cohorts (chart review cohort) had two troponin measurements within 6 h of each other. For patients with two troponin measurements within 6 h; of those with an MI diagnosis, our phenotype classified 25.2% (31.3%) with an acute myocardial injury and 62.2% (55.2%) with an MI in the catheterized cohort (chart review cohort); and of those without an MI diagnosis, our phenotype classified 12.9% (12.4%) with an acute myocardial injury and 10.0% (13.1%) with an MI in the catheterized cohort (chart review cohort). Conclusions: Patients with two troponin measurements within 6 h, identified by our phenotype as having either an acute myocardial injury or MI, will at least meet the diagnostic criteria for an acute myocardial injury (barring lab errors) and indicate many previously uncaptured cases. Myocardial infarctions are harder to be certain of because ECG report findings might be superseded by evidence not included in our phenotype, or due to errors with the natural language processing.

Active Choice Nudge to Increase Screening for Primary Aldosteronism in At-Risk Patients.

Primary aldosteronism (PA) is the most common cause of secondary hypertension, yet screening remains startlingly infrequent. We describe (1) PA screening practices in a large, diverse health system, (2) the development of a computable phenotype for PA screening, and (3) the design and pilot deployment of an electronic health record (EHR)-based active choice nudge to recommend PA screening. A multidisciplinary team developed a multipronged intervention to improve PA screening informed by guidelines, expertise, and multivariable analyses of factors associated with screening. The intervention included EHR-based tools to automatically identify screen-eligible patients, an active choice nudge recommending screening for these patients, and screening result interpretation. The intervention was piloted across 2 primary care practices for 7 months. Screening frequencies were compared with clinics not receiving the intervention. The baseline frequency of screening of eligible patients within 1 year was 1.4%. Higher mean systolic blood pressure (odds ratio [OR] 1.4; p < 0.001), more antihypertensive medications (OR 1.3; p = 0.002), lower minimum serum potassium (OR 2.0; p = 0.001), specialist care (OR 3.0; p < 0.001), and Black race (OR 1.5; p = 0.001) were associated with a higher likelihood of screening. The refined computable phenotype identified a subcohort with a higher frequency of positive screening (8.6% vs 4.1%; p = 0.03). In a pilot study of an active choice nudge, a greater proportion of eligible patients were screened in the intervention clinics (16.4%) than in the nonintervention clinics (1.8%; p < 0.001). PA screening rates are low. This pilot study suggests an EHR-based nudge leveraging a precise computable phenotype can dramatically increase appropriate PA screening.

MUC5B Genotype and Other Common Variants are Associated with Computational Imaging Features of UIP.

Idiopathic pulmonary fibrosis (IPF) is a complex and heterogeneous disease. Given this, we reasoned that differences in genetic profiles may be associated with unique clinical and radiologic features. Computational image analysis, sometimes referred to as radiomics, provides objective, quantitative assessments of radiologic features in subjects with pulmonary fibrosis. To determine if the genetic risk profile of patients with IPF identifies unique computational imaging phenotypes. Participants with IPF were included in this study if they had genotype data and CT scans of the chest available for computational image analysis. Extent of lung fibrosis and likelihood of a usual interstitial pneumonia (UIP) pattern were scored automatically by using two separate, previously validated deep learning techniques for CT analysis. UIP pattern was also classified visually by radiologists according to established criteria. Among 334 participants with IPF, MUC5B, FAM13A and ZKSCAN1 were independently associated with the deep learning-based UIP score. None of the common variants were associated with fibrosis extent by computational imaging. We did not find an association between MUC5B, FAM13A or ZKSCAN1 and visually assessed UIP pattern. Select genetic variants are associated with computer-based classification of UIP on CT among patients with IPF. Analysis of radiologic features using deep learning may enhance our ability to identify important genotype-phenotype associations in fibrotic lung diseases.

A Computable Phenotype Algorithm for Postvaccination Myocarditis/Pericarditis Detection Using Real-World Data: Validation Study.

Adverse events (AEs) associated with vaccination have traditionally been evaluated by epidemiological studies. More recently, they have gained attention due to the emergency use authorization of several COVID-19 vaccines. As part of its responsibility to conduct postmarket surveillance, the US Food and Drug Administration continues to monitor several AEs of interest to ensure the safety of vaccines, including those for COVID-19. This study is part of the Biologics Effectiveness and Safety Initiative, which aims to improve the US Food and Drug Administration's postmarket surveillance capabilities while minimizing the burden of collecting clinical data on suspected postvaccination AEs. The objective of this study was to enhance active surveillance efforts through a pilot platform that can receive automatically reported AE cases through a health care data exchange. We detected cases by sharing and applying computable phenotype algorithms to real-world data in health care providers' electronic health records databases. Using the fast healthcare interoperability resources standard for secure data transmission, we implemented a computable phenotype algorithm on a new health care system. The study focused on the algorithm's positive predictive value, validated through clinical records, assessing both the time required for implementation and the accuracy of AE detection. The algorithm required 200-250 hours to implement and optimize. Of the 6,574,420 clinical encounters across 694,151 patients, 30 cases were identified as potential myocarditis/pericarditis. Of these, 26 cases were retrievable, and 24 underwent clinical validation. In total, 14 cases were confirmed as definite or probable myocarditis/pericarditis, yielding a positive predictive value of 58.3% (95% CI 37.3%-76.9%). These findings underscore the algorithm's capability for real-time detection of AEs, though they also highlight variability in performance across different health care systems. The study advocates for the ongoing refinement and application of distributed computable phenotype algorithms to enhance AE detection capabilities. These tools are crucial for comprehensive postmarket surveillance and improved vaccine safety monitoring. The outcomes suggest the need for further optimization to achieve more consistent results across diverse health care settings.

Evaluating a Computable Phenotype for CKD Detection in Adult Patients Treated in Primary Care

Evaluating a Computable Phenotype for CKD Detection in Adult Patients Treated in Primary Care

Developing an automated algorithm for identification of children and adolescents with diabetes using electronic health records from the OneFlorida+ clinical research network.

To develop an automated computable phenotype (CP) algorithm for identifying diabetes cases in children and adolescents usingelectronic health records (EHRs) from the UF Health System. The CP algorithm was iteratively derived based on structured data from EHRs (UF Health System 2012-2020). We randomly selected 536 presumed cases among individuals aged <18 years who had (1) glycated haemoglobin levels ≥ 6.5%; or (2) fasting glucose levels ≥126 mg/dL; or (3) random plasma glucose levels ≥200 mg/dL; or (4) a diabetes-related diagnosis code from an inpatient or outpatient encounter; or (5) prescribed, administered, or dispensed diabetes-related medication. Four reviewers independently reviewed the patient charts to determine diabetes status and type. Presumed cases without type 1 (T1D) or type 2 diabetes (T2D) diagnosis codes were categorized as non-diabetes/other types of diabetes. The rest were categorized as T1D if the most recent diagnosis was T1D, or otherwise categorized as T2D if the most recent diagnosis was T2D. Next, we applied a list of diagnoses and procedures that can determine diabetes type (e.g., steroid use suggests induced diabetes) to correct misclassifications from Step 1. Among the 536 reviewed cases, 159 and 64 had T1D and T2D, respectively. The sensitivity, specificity, and positive predictive values of the CP algorithm were 94%, 98% and 96%, respectively, for T1D and 95%, 95% and 73% for T2D. We developed a highly accurate EHR-based CP for diabetes in youth based on EHR data from UF Health. Consistent with prior studies, T2D was more difficult to identify using these methods.

The impact of the COVID-19 pandemic on mental health care utilization among people living with HIV: A real-world data study.

The COVID-19 pandemic has profoundly impacted mental health worldwide, particularly among vulnerable populations such as people living with HIV (PLWH). However, large-scale, real-world data on mental health care utilization and associated factors among PLWH remain limited. This study leveraged electronic health records (EHR) and Basics survey data from the All of Us program to explore mental health care utilization and associated factors among PLWH during the COVID-19 pandemic. Using a retrospective cohort design, we identified and included 4,575 PLWH through computational phenotyping based on relevant Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) concept sets from the All of Us program between March 2018 and March 2022. Mental health care utilization was measured using the yearly count of mental health care visits over this period. The pattern of mental health care utilization was compared between pre-pandemic (2018-2020) and during the pandemic (2020-2022). Incidence rate ratios (IRR) from the Poisson generalized linear mixed models (GLMM) were used to examine associations between mental health care utilization, history of COVID-19 infection, demographic factors, pre-existing chronic conditions (e.g., hypertension, diabetes), and socioeconomic status. Among 4,575 PLWH, the annual number of mental health care visits decreased significantly during the pandemic period (March 2020 - March 2022) compared to the pre-pandemic period (March 2018 - February 2020) (IRR = 0.89, p < 0.001). The Poisson regression analysis found that a history of COVID-19 infection was associated with a higher number of mental health care visits (IRR = 1.35, p < 0.001). Middle-aged groups with participants aged 30-39 (IRR= 2.35, p = 0.002), 40-49 (IRR= 3.49, p < 0.001), and 50-64 (IRR= 2.07, p = 0.004) had significantly higher visit numbers compared to the youngest group (18-29 years). Black or African American participants were less likely to have visits compared to White participants (IRR = 0.71, p = 0.002). Medicaid health insurance was associated with an increase (IRR = 1.29, p = 0.007), while employer- or union-sponsored insurance was associated with a decrease in mental health care utilization (IRR = 0.54, p < 0.001, p < 0.001). Pre-existing comorbidities greatly increased the number of mental health care visits (one comorbidity: IRR = 5.49, two or more: IRR = 10.4, p < 0.001). Our study reveals a significant decrease in mental health care utilization and the diverse experiences of mental healthcare among PLWH during the COVID-19 pandemic. These findings underscore the importance of addressing disparities in mental health care access, particularly during public health emergencies, and suggest the need for tailored interventions to meet the mental health care needs of PLWH.

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.

Linking Maternal and Child Health Data to Enhance Public Health Surveillance in Indiana

Indiana faces significant challenges in maternal and infant health, ranking 45th worst in infant mortality with a rate of 6.5 deaths per 1,000 live-births, and 21st in stillbirths. Indiana further has the 3rd highest rate of maternal mortality with 44 deaths per 100,000 live-births. Despite these alarming statistics, comprehensive surveillance of maternal and infant health is absent, and data often remains fragmented across disparate information systems. In response, the Indiana University Fairbanks School of Public Health and Regenstrief Institute secured a cooperative agreement through the CDC's Pregnant People-Infant Linked Longitudinal Surveillance (PILLARS) program, collaborating with the Indiana Health Information Exchange, Indiana Department of Health, and Marion County Public Health Department. The initiative aims to enhance Indiana's infrastructure for surveillance of maternal and child health (MCH) using electronic health records (EHRs), public health, and administrative data. Key efforts have included development, validation, and application of linkage algorithms across records for mothers and children; integration of case data across data sources; design of routine surveillance reports; and design of longitudinal studies to examine determinants and outcomes in MCH populations. Using deterministic linkage algorithms, we have created over 800,000 mother-infant dyads; this number will expand with the implementation of probabilistic approaches. We also have developed a computational phenotype for identifying cases of pregnancy in the EHR even when not explicitly flagged in the EHR. Future efforts will build on this infrastructure to draw from additional public health data sources and/or expand surveillance efforts to include prioritized MCH outcomes.

Evaluation of a Multi-Component Phenotype Algorithm for Systemic Lupus Erythematosus across the PCORnet, OMOP, and i2b2 Common Data Models

The aim of this study was to evaluate the performance of a computable phenotype for systemic lupus erythematosus (SLE) patients when it is ported from a local data warehouse to the i2b2, OMOP, and PCORnet CDMs. We adapted the SLE phenotype to the Northwestern Medicine (NM) Enterprise Data Warehouse (EDW) and NU i2b2, OMOP, and PCORnet instances. Each of the phenotype’s 17 components were determined by a rules-based algorithm built on diagnosis, medication, lab, and procedure codes. We assessed the phenotype over 168 clinician-confirmed SLE patients and 100 healthy controls and calculated agreement between the EDW and CDMs for overall SLE classification, individual SLICC criteria membership, and individual code occurrence level aggregates - count, first and last date, lab values using Cohen’s kappa (Cκ) and intraclass correlation coefficients (ICC). For overall SLE classification, agreement of the OMOP and PCORnet datamarts with the EDW was high (Cκ 0.928, 0.802, respectively) while that of i2b2 was low (Cκ 0.328). For the panel of SLICC criteria, OMOP had high agreement for 14/17 SLICC criteria (Cκ 0.792-1.000). PCORnet had high agreement for 13/17 criteria (Cκ 0.708-1.000). i2b2 had high agreement for 10/17 criteria (Cκ 0.813-1.000) with the exception of lab-based criteria. While agreement between the EDW and CDMs at the level of individual codes was relatively poor, agreement at the levels of overall SLE classification and individual SLICC criteria were reasonably high for OMOP and PCORnet, suggesting that inconsistencies at a micro, code level become less apparent at the macro, phenotype level.

Assessing the Effect of Electronic Health Record Data Quality on Identifying Patients With Type 2 Diabetes: Cross-Sectional Study.

Increasing and substantial reliance on electronic health records (EHRs) and data types (ie, diagnosis, medication, and laboratory data) demands assessment of their data quality as a fundamental approach, especially since there is a need to identify appropriate denominator populations with chronic conditions, such as type 2 diabetes (T2D), using commonly available computable phenotype definitions (ie, phenotypes). To bridge this gap, our study aims to assess how issues of EHR data quality and variations and robustness (or lack thereof) in phenotypes may have potential impacts in identifying denominator populations. Approximately 208,000 patients with T2D were included in our study, which used retrospective EHR data from the Johns Hopkins Medical Institution (JHMI) during 2017-2019. Our assessment included 4 published phenotypes and 1 definition from a panel of experts at Hopkins. We conducted descriptive analyses of demographics (ie, age, sex, race, and ethnicity), use of health care (inpatient and emergency room visits), and the average Charlson Comorbidity Index score of each phenotype. We then used different methods to induce or simulate data quality issues of completeness, accuracy, and timeliness separately across each phenotype. For induced data incompleteness, our model randomly dropped diagnosis, medication, and laboratory codes independently at increments of 10%; for induced data inaccuracy, our model randomly replaced a diagnosis or medication code with another code of the same data type and induced 2% incremental change from -100% to +10% in laboratory result values; and lastly, for timeliness, data were modeled for induced incremental shift of date records by 30 days to 365 days. Less than a quarter (n=47,326, 23%) of the population overlapped across all phenotypes using EHRs. The population identified by each phenotype varied across all combinations of data types. Induced incompleteness identified fewer patients with each increment; for example, at 100% diagnostic incompleteness, the Chronic Conditions Data Warehouse phenotype identified zero patients, as its phenotypic characteristics included only diagnosis codes. Induced inaccuracy and timeliness similarly demonstrated variations in performance of each phenotype, therefore resulting in fewer patients being identified with each incremental change. We used EHR data with diagnosis, medication, and laboratory data types from a large tertiary hospital system to understand T2D phenotypic differences and performance. We used induced data quality methods to learn how data quality issues may impact identification of the denominator populations upon which clinical (eg, clinical research and trials, population health evaluations) and financial or operational decisions are made. The novel results from our study may inform future approaches to shaping a common T2D computable phenotype definition that can be applied to clinical informatics, managing chronic conditions, and additional industry-wide efforts in health care.

EHR-based Case Identification of Pediatric Long COVID: A Report from the RECOVER EHR Cohort.

Long COVID, marked by persistent, recurring, or new symptoms post-COVID-19 infection, impacts children's well-being yet lacks a unified clinical definition. This study evaluates the performance of an empirically derived Long COVID case identification algorithm, or computable phenotype, with manual chart review in a pediatric sample. This approach aims to facilitate large-scale research efforts to understand this condition better. The algorithm, composed of diagnostic codes empirically associated with Long COVID, was applied to a cohort of pediatric patients with SARS-CoV-2 infection in the RECOVER PCORnet EHR database. The algorithm classified 31,781 patients with conclusive, probable, or possible Long COVID and 307,686 patients without evidence of Long COVID. A chart review was performed on a subset of patients (n=651) to determine the overlap between the two methods. Instances of discordance were reviewed to understand the reasons for differences. The sample comprised 651 pediatric patients (339 females, M age = 10.10 years) across 16 hospital systems. Results showed moderate overlap between phenotype and chart review Long COVID identification (accuracy = 0.62, PPV = 0.49, NPV = 0.75); however, there were also numerous cases of disagreement. No notable differences were found when the analyses were stratified by age at infection or era of infection. Further examination of the discordant cases revealed that the most common cause of disagreement was the clinician reviewers' tendency to attribute Long COVID-like symptoms to prior medical conditions. The performance of the phenotype improved when prior medical conditions were considered (accuracy = 0.71, PPV = 0.65, NPV = 0.74). Although there was moderate overlap between the two methods, the discrepancies between the two sources are likely attributed to the lack of consensus on a Long COVID clinical definition. It is essential to consider the strengths and limitations of each method when developing Long COVID classification algorithms.