EHR-based Studies Research Articles

Statistical inference for association studies using electronic health records: handling both selection bias and outcome misclassification.

Health research using electronic health records (EHR) has gained popularity, but misclassification of EHR-derived disease status and lack of representativeness of the study sample can result in substantial bias in effect estimates and can impact power and type I error. In this paper, we develop new strategies for handling disease status misclassification and selection bias in EHR-based association studies. We first focus on each type of bias separately. For misclassification, we propose three novel likelihood-based bias correction strategies. A distinguishing feature of the EHR setting is that misclassification may be related to patient-varying factors, and the proposed methods leverage data in the EHR to estimate misclassification rates without gold standard labels. For addressing selection bias, we describe how calibration and inverse probability weighting methods from the survey sampling literature can be extended and applied to the EHRsetting. Addressing misclassification and selection biases simultaneously is a more challenging problem than dealing with each on its own, and we propose several new strategies. For all methods proposed, we derive valid standard error estimators and provide software for implementation. We provide a new suite of statistical estimation and inference strategies for addressing misclassification and selection bias simultaneously that is tailored to problems arising in EHR data analysis. We apply these methods to data from The Michigan Genomics Initiative, a longitudinal EHR-linkedbiorepository.

OP520 Evaluating Long-Term Survival From Clinical Trials: Does Real-World Evidence Change the Paradigm?

IntroductionBoth patient composition and medical care received in clinical trials may not be representative of clinical practice, yet health technology assessments (HTAs) commonly use extrapolation results from trials to estimate incremental benefit. Due to data limitations, external validation of trial extrapolations are uncommon. With the goal of better estimating the benefit of new therapies in practice, we compared long-term survival estimated from real-world patients who received therapy similar to the comparator arm of the OAK trial, a phase III study of patients with advanced non-small cell lung cancer (aNSCLC) who progressed following initial chemotherapy, to standard estimation approaches.MethodsWe estimated long-term survival from: (i) direct extrapolation of trial survival curves; and (ii) aNSCLC patients from the United States Flatiron Health Electronic Health Record ()-derived de-identified database diagnosed between January 2011 and August 2019 who received docetaxel monotherapy after platinum-doublet and had adequate organ function as well as functional status. Patients with unknown organ function and functional status were also included. Standard parametric extrapolations were applied and selected based on visual inspection and goodness-of-fit tests for each cohort.ResultsUsing a log-logistic model to extrapolate the trial comparator arm (N = 425), estimated lifetime mean overall survival was 19.2 months (95% confidence interval [95% CI]: 16.5–22.6), and 14.4 months (95% CI: 12.4–17.0) for the real-world cohort (N = 415). Estimated 5-year overall survival rates were 5.4 percent (95% CI: 3.9–7.3) for the trial patients, compared to 3.7 percent (95% CI: 2.6–5.0) among real-world cohort patients.ConclusionsOur results suggest that directly extrapolating observed survival for trial patients may overestimate the long-term survival compared to the experience of patients treated in routine practice. Our findings have implications for those wishing to estimate the incremental benefit for novel versus established treatments. We plan to compare our results to a generic patient cohort from national cancer registry. Further EHR-based studies utilizing real world data are needed to confirm our findings and to extend beyond this use case for other cancer types and anti-neoplastic therapies.

Research and Reporting Considerations for Observational Studies Using Electronic Health Record Data.

Electronic health records (EHRs) are an increasingly important source of real-world health care data for observational research. Analyses of data collected for purposes other than research require careful consideration of data quality as well as the general research and reporting principles relevant to observational studies. The core principles for observational research in general also apply to observational research using EHR data, and these are well addressed in prior literature and guidelines. This article provides additional recommendations for EHR-based research. Considerations unique to EHR-based studies include assessment of the accuracy of computer-executable cohort definitions that can incorporate unstructured data from clinical notes and management of data challenges, such as irregular sampling, missingness, and variation across time and place. Principled application of existing research and reporting guidelines alongside these additional considerations will improve the quality of EHR-based observational studies.

Methods to identify dementia in the electronic health record: Comparing cognitive test scores with dementia algorithms

BackgroundEpidemiologic studies often use diagnosis codes to identify dementia outcomes. It remains unknown to what extent cognitive screening test results add value in identifying dementia cases in big data studies leveraging electronic health record (EHR) data. We examined test scores from EHR data and compared results with dementia algorithms. MethodsThis retrospective cohort study included patients 60+ years of age from Kaiser Permanente Washington (KPWA) during 2013–2018 and the Veterans Health Affairs (VHA) during 2012–2015. Results from the Mini Mental State Examination (MMSE) and the Saint Louis University Mental Status Examination (SLUMS) cognitive screening exams, were classified as showing dementia or not. Multiple dementia algorithms were created using combinations of diagnosis codes, pharmacy records, and specialty care visits. Correlations between test scores and algorithms were assessed. Results3,690 of 112,917 KPWA patients and 2,981 of 102,981 VHA patients had cognitive test results in the EHR. In KPWA, dementia prevalence ranged from 6.4%-8.1% depending on the algorithm used and in the VHA, 8.9%–12.1%. The algorithm which best agreed with test scores required ≥2 dementia diagnosis codes in 12 months; at KPWA, 14.8% of people meeting this algorithm had an MMSE score, of whom 65% had a score indicating dementia. Within VHA, those figures were 6.2% and 77% respectively. ConclusionsAlthough cognitive test results were rarely available, agreement was good with algorithms requiring ≥2 dementia diagnosis codes, supporting the accuracy of this algorithm. ImplicationsThese scores may add value in identifying dementia cases for EHR-based research studies.

Case contamination in electronic health records based case-control studies.

Clinically relevant information from electronic health records (EHRs) permits derivation of a rich collection of phenotypes. Unlike traditionally designed studies where scientific hypotheses are specified a priori before data collection, the true phenotype status of any given individual in EHR-based studies is not directly available. Structured and unstructured data elements need to be queried through preconstructed rules to identify case and control groups. A sufficient number of controls can usually be identified with high accuracy by making the selection criteria stringent. But more relaxed criteria are often necessary for more thorough identification of cases to ensure achievable statistical power. The resulting pool of candidate cases consists of genuine cases contaminated with noncase patients who do not satisfy the control definition. The presence of patients who are neither true cases nor controls among the identified cases is a unique challenge in EHR-based case-control studies. Ignoring case contamination would lead to biased estimation of odds ratio association parameters. We propose an estimating equation approach to bias correction, study its large sample property, and evaluate its performance through extensive simulation studies and an application to a pilot study of aortic stenosis in the Penn medicine EHR. Our method holds the promise of facilitating more efficient EHR studies by accommodating enlarged albeit contaminated casepools.

RW4 THE ROLE OF ELECTRONIC HEALTH RECORDS IN THE ADVANCEMENT OF REAL WORLD EVIDENCE BASED RESEARCH IN ONCOLOGY

In the recent years, real-world evidence (RWE) has evolved as a complementary source of evidence for traditional clinical trials. Electronic health records (EHRs) generated from clinical practices is one of the vital components RWE. This systematic review assessed the range of analytical questions answered by EHR-based studies. Research articles reporting any oncology EHR, published in Pubmed between Jan 1, 2015 and Aug 5, 2019 were included. We evaluated the objective(s), study endpoints, and analytical approaches adopted in each publication. A total of 3906 abstracts were screened, of which 35 were included, with 27 (77%) being published in the past 2 years. Studies ranged from basic descriptive to advanced analytics including correlations studies and predictive models. Treatment pattern was most frequently studied (37%), followed by comparative effectiveness (14%), data extraction validations (14%), clinical effectiveness (9%), disease prediction models (9%), safety (6%), burden of illness (3%), cost-effectiveness (3%), treatment drivers (3%), and patient journey (3%). Treatment pattern was frequently accompanied by treatment outcome (10/13), often assessing how change impacts survival. Regarding EHR data, we observed use of both structured and unstructured information. Seven publications (20%) accessed unstructured data (e.g., clinician notes, lab/radiology reports) to understand underlying drivers of observed trends. However, many of these studies had indicated that such analyses are limited by small sample sizes or limitations of available manual data extraction techniques. There has been a notable increase in EHR publications in the past 2 years, with a majority of research focused on practice trends and clinical outcomes. Although advanced analytical approaches have enabled the use of structured and unstructured data from EHR notes to understand drivers of treatment decisions, there is more opportunity to enhance data completeness and improve extraction techniques.

Exploiting nonsystematic covariate monitoring to broaden the scope of evidence about the causal effects of adaptive treatment strategies.

In studies based on electronic health records (EHR), the frequency of covariate monitoring can vary by covariate type, across patients, and over time, which can limit the generalizability of inferences about the effects of adaptive treatment strategies. In addition, monitoring is a health intervention in itself with costs and benefits, and stakeholders may be interested in the effect of monitoring when adopting adaptive treatment strategies. This paper demonstrates how to exploit nonsystematic covariate monitoring in EHR-based studies to both improve the generalizability of causal inferences and to evaluate the health impact of monitoring when evaluating adaptive treatment strategies. Using a real world, EHR-based, comparative effectiveness research (CER) study of patients with type II diabetes mellitus, we illustrate how the evaluation of joint dynamic treatment and static monitoring interventions can improve CER evidence and describe two alternate estimation approaches based on inverse probability weighting (IPW). First, we demonstrate the poor performance of the standard estimator of the effects of joint treatment-monitoring interventions, due to a large decrease in data support and concerns over finite-sample bias from near-violations of the positivity assumption (PA) for the monitoring process. Second, we detail an alternate IPW estimator using a no direct effect assumption. We demonstrate that this estimator can improve efficiency but at the potential cost of increase in bias from violations of the PA for the treatmentprocess.

Population-Based Registry Linkages to Improve Validity of Electronic Health Record-Based Cancer Research.

There is tremendous potential to leverage the value gained from integrating electronic health records (EHR) and population-based cancer registry data for research. Registries provide diagnosis details, tumor characteristics, and treatment summaries, while EHRs contain rich clinical detail. A carefully conducted cancer registry linkage may also be used to improve the internal and external validity of inferences made from EHR-based studies. We linked the EHRs of a large, multispecialty, mixed-payer health care system with the statewide cancer registry and assessed the validity of our linked population. For internal validity, we identify patients that might be "missed" in a linkage, threatening the internal validity of an EHR study population. For generalizability, we compared linked cases with all other cancer patients in the 22-county EHR catchment region. From an EHR population of 4.5 million, we identified 306,554 patients with cancer, 26% of the catchment region patients with cancer; 22.7% of linked patients were diagnosed with cancer after they migrated away from our health care system highlighting an advantage of system-wide linkage. We observed demographic differences between EHR patients and non-EHR patients in the surrounding region and demonstrated use of selection probabilities with model-based standardization to improve generalizability. Our experiences set the foundation to encourage and inform researchers interested in working with EHRs for cancer research as well as provide context for leveraging linkages to assess and improve validity and generalizability. Researchers conducting linkages may benefit from considering one or more of these approaches to establish and evaluate the validity of their EHR-based populations.See all articles in this CEBP Focus section, "Modernizing Population Science."

Assessment of the impact of EHR heterogeneity for clinical research through a case study of silent brain infarction

BackgroundThe rapid adoption of electronic health records (EHRs) holds great promise for advancing medicine through practice-based knowledge discovery. However, the validity of EHR-based clinical research is questionable due to poor research reproducibility caused by the heterogeneity and complexity of healthcare institutions and EHR systems, the cross-disciplinary nature of the research team, and the lack of standard processes and best practices for conducting EHR-based clinical research.MethodWe developed a data abstraction framework to standardize the process for multi-site EHR-based clinical studies aiming to enhance research reproducibility. The framework was implemented for a multi-site EHR-based research project, the ESPRESSO project, with the goal to identify individuals with silent brain infarctions (SBI) at Tufts Medical Center (TMC) and Mayo Clinic. The heterogeneity of healthcare institutions, EHR systems, documentation, and process variation in case identification was assessed quantitatively and qualitatively.ResultWe discovered a significant variation in the patient populations, neuroimaging reporting, EHR systems, and abstraction processes across the two sites. The prevalence of SBI for patients over age 50 for TMC and Mayo is 7.4 and 12.5% respectively. There is a variation regarding neuroimaging reporting where TMC are lengthy, standardized and descriptive while Mayo’s reports are short and definitive with more textual variations. Furthermore, differences in the EHR system, technology infrastructure, and data collection process were identified.ConclusionThe implementation of the framework identified the institutional and process variations and the heterogeneity of EHRs across the sites participating in the case study. The experiment demonstrates the necessity to have a standardized process for data abstraction when conducting EHR-based clinical studies.

Dynamic ElecTronic hEalth reCord deTection (DETECT) of individuals at risk of a first episode of psychosis: a case-control development and validation study

Many individuals who will experience a first episode of psychosis (FEP) are not detected before occurrence, limiting the effect of preventive interventions. The combination of machine-learning methods and electronic health records (EHRs) could help address this gap. This case-control development and validation study is based on EHR data from IBM Explorys. The IBM Explorys Platform holds standardised, longitudinal, de-identified, patient-level EHR data pooled from different health-care systems with distinct EHRs. The present EHR-based studies were retrospective, matched (1:1), case-control studies compliant with RECORD, STROBE, and TRIPOD statements. The study included individuals in the IBM Explorys database who at some point between 1990 and 2018 had a diagnosis of FEP followed by schizophrenia, and psychosis-free matched control individuals from a random subsample of the full cohort. For every individual in the FEP cohort, the individual in the control cohort was matched to have a similar date for inclusion in the database and a similar total observation time. Individuals in the FEP cohort had their index date defined as the first diagnosis of psychosis or the first prescription of antipsychotic medication. Individuals in the control cohort had their index date defined to occur the same number of days after inclusion in the database as their matching FEP individual. The FEP and control cohorts were both randomly split into development and validation datasets in a ratio of 7:3. The subset of individuals in the validation dataset who had all their health-care encounters at providers that were not seen in the development dataset made up the external validation subset. A novel recurrent neural network model was developed to predict the risk of FEP 1 year before the index date by employing demographics and medical events (in the categories diagnoses, prescriptions, procedures, encounters and admissions, observations, and laboratory test results) dynamically collected in the EHR as part of clinical routine. We named the recurrent neural network Dynamic ElecTronic hEalth reCord deTection (DETECT). The main outcomes were accuracy and area under receiver operating characteristic curve (AUROC). Decision-curve analyses and dynamic patient journey plots were used to evaluate clinical usefulness. The FEP and control cohorts each comprised 72 860 individuals. 102 030 individuals (51 015 matching pairs) were randomly allocated to the development dataset and the remaining 43 690 to the validation dataset. In the validation dataset, 4770 individuals had all their encounters outside of the 118 790 health-care providers that were encountered in the development dataset. The data from these individuals made up the external validation subset. The median follow-up (observation time before index date) was 6·0 years (IQR 3·0-10·4). In the development dataset, DETECT's prognostic accuracy was 0·787 and AUROC was 0·868. In the validation dataset, DETECT's prognostic accuracy was 0·774 and AUROC was 0·856. In the external test subset, DETECT's balanced prognostic accuracy was 0·724 and AUROC was 0·799. Prevalence-adjusted decision-curve analyses suggested that DETECT was associated with a positive net benefit in two different scenarios for FEP detection. DETECT showed adequate prognostic accuracy to detect individuals at risk of developing a FEP in primary and secondary care. Replication and refinement in a population-based setting are needed to consolidate these findings. Lundbeck.

Data science and machine learning in anesthesiology.

Machine learning (ML) is revolutionizing anesthesiology research. Unlike classical research methods that are largely inference-based, ML is geared more towards making accurate predictions. ML is a field of artificial intelligence concerned with developing algorithms and models to perform prediction tasks in the absence of explicit instructions. Most ML applications, despite being highly variable in the topics that they deal with, generally follow a common workflow. For classification tasks, a researcher typically tests various ML models and compares the predictive performance with the reference logistic regression model. The main advantage of ML lies in its ability to deal with many features with complex interactions and its specific focus on maximizing predictive performance. However, emphasis on data-driven prediction can sometimes neglect mechanistic understanding. This article mainly focuses on the application of supervised ML to electronic health record (EHR) data. The main limitation of EHR-based studies is in the difficulty of establishing causal relationships. However, the associated low cost and rich information content provide great potential to uncover hitherto unknown correlations. In this review, the basic concepts of ML are introduced along with important terms that any ML researcher should know. Practical tips regarding the choice of software and computing devices are also provided. Towards the end, several examples of successful ML applications in anesthesiology are discussed. The goal of this article is to provide a basic roadmap to novice ML researchers working in the field of anesthesiology.

Variation in methods, results and reporting in electronic health record-based studies evaluating routine care in gout: A systematic review.

To perform a systematic review examining the variation in methods, results, reporting and risk of bias in electronic health record (EHR)-based studies evaluating management of a common musculoskeletal disease, gout. Two reviewers systematically searched MEDLINE, Scopus, Web of Science, CINAHL, PubMed, EMBASE and Google Scholar for all EHR-based studies published by February 2019 investigating gout pharmacological treatment. Information was extracted on study design, eligibility criteria, definitions, medication usage, effectiveness and safety data, comprehensiveness of reporting (RECORD), and Cochrane risk of bias (registered PROSPERO CRD42017065195). We screened 5,603 titles/abstracts, 613 full-texts and selected 75 studies including 1.9M gout patients. Gout diagnosis was defined in 26 ways across the studies, most commonly using a single diagnostic code (n = 31, 41.3%). 48.4% did not specify a disease-free period before 'incident' diagnosis. Medication use was suboptimal and varied with disease definition while results regarding effectiveness and safety were broadly similar across studies despite variability in inclusion criteria. Comprehensiveness of reporting was variable, ranging from 73% (55/75) appropriately discussing the limitations of EHR data use, to 5% (4/75) reporting on key data cleaning steps. Risk of bias was generally low. The wide variation in case definitions and medication-related analysis among EHR-based studies has implications for reported medication use. This is amplified by variable reporting comprehensiveness and the limited consideration of EHR-relevant biases (e.g. data adequacy) in study assessment tools. We recommend accounting for these biases and performing a sensitivity analysis on case definitions, and suggest changes to assessment tools to foster this.

An augmented estimation procedure for EHR-based association studies accounting for differential misclassification.

The ability to identify novel risk factors for health outcomes is a key strength of electronic health record (EHR)-based research. However, the validity of such studies is limited by error in EHR-derived phenotypes. The objective of this study was to develop a novel procedure for reducing bias in estimated associations between risk factors and phenotypes in EHR data. The proposed method combines the strengths of a gold-standard phenotype obtained through manual chart review for a small validation set of patients and an automatically-derived phenotype that is available for all patients but is potentially error-prone (hereafter referred to as the algorithm-derived phenotype). An augmented estimator of associations is obtained by optimally combining these 2 phenotypes. We conducted simulation studies to evaluate the performance of the augmented estimator and conducted an analysis of risk factors for second breast cancer events using data on a cohort from Kaiser Permanente Washington. The proposed method was shown to reduce bias relative to an estimator using only the algorithm-derived phenotype and reduce variance compared to an estimator using only the validation data. Our simulation studies and real data application demonstrate that, compared to the estimator using validation data only, the augmented estimator has lower variance (ie, higher statistical efficiency). Compared to the estimator using error-prone EHR-derived phenotypes, the augmented estimator has smaller bias. The proposed estimator can effectively combine an error-prone phenotype with gold-standard data from a limited chart review in order to improve analyses of risk factors using EHR data.

A Comprehensive Review of Methods to Measure Oral Oncolytic Dose Intensity Using Retrospective Data.

Understanding the real-world use of oral oncolytics is essential to assess drug effectiveness. Retrospective analyses using medical and pharmacy claims data allow observation of drug use patterns and health outcomes. However, studies of medication adherence to oral oncolytics may not be sufficient in characterizing exposure because they typically measure refill frequency, not the administered dose or dose changes. Patients who appear fully adherent by traditional measures may be receiving different doses and experiencing differing effectiveness. Relative dose intensity (RDI) is a measure that has been used for intravenous drugs to capture the amount of a particular chemotherapeutic agent administered per unit of time (dose intensity), expressed as the fraction of the amount recommended in evidence-based guidelines. Such a measure would be useful for real-world studies of comparative effectiveness to characterize patient exposure to oral oncolytics. To identify studies that used administrative claims data to measure real-world oral oncolytic dose intensity, RDI, or similar constructs. Two health sciences librarians conducted a literature search (PubMed, January 1, 1809-February 6, 2018) including terms in each of the following concept areas: oncology drugs, dosage, and retrospective data sources. At least 2 reviewers scanned each title and abstract of publications retrieved from PubMed. Abstracts that indicated the study reported dose or related concepts and oral oncolytics using retrospective data sources were marked for full-text review. During full-text review, papers were excluded if they did not study oral oncolytics (i.e., only described intravenous chemotherapy); if they did not report drug dosage; or if the study was not retrospective. Resulting studies were included for full-text data extraction. Of the 1,640 publications returned from the search, 41 were marked for full-text review. Full-text review established that 17 studies addressed a concept related to dose of oral oncolytics using retrospective data. Twenty-four studies were excluded: 11 did not measure dose; 9 did not study oral oncolytics; and 4 were not retrospective studies. Among the 17 articles marked for extraction, 5 articles reported dose intensity or RDI using medical records or electronic health record (EHR) data. This study reveals not only the need for a claims-based measure of dose intensity for oral oncolytics, but also provides a basis for the development of such a measure based on previous EHR-based studies. While several claims data studies have characterized oral oncolytic dosing and duration, we found that no studies combined these dimensions into a single measure such as dose intensity. Methods using EHR data may be translatable to a claims data study. Future research is needed to develop and validate such measures. Novartis Pharmaceuticals provided funding for this study and is a manufacturer of oral onalytics, which is under study in this article. Arcona and Zacker are employees of Novartis. Slejko reports grants from PhRMA, PhRMA Foundation, and Takeda Pharmaceuticals and consulting fees from Pfizer, outside the submitted work. Stuart reports consulting fees from the University of Maryland during the study. The other authors have nothing to disclose. The preliminary findings of this study were presented in a poster at AMCP Nexus 2018, October 22-25, 2018, in Orlando, FL.

Barriers and facilitators to data quality of electronic health records used for clinical research in China: a qualitative study

ObjectivesThere is an increasing trend in the use of electronic health records (EHRs) for clinical research. However, more knowledge is needed on how to assure and improve data quality. This...

Using big data to improve cardiovascular care and outcomes in China: a protocol for the CHinese Electronic health Records Research in Yinzhou (CHERRY) Study

IntroductionData based on electronic health records (EHRs) are rich with individual-level longitudinal measurement information and are becoming an increasingly common data source for clinical risk prediction worldwide. However, few EHR-based...

Addressing Bias in Electronic Health Record-Based Surveillance of Cardiovascular Disease Risk: Finding the Signal Through the Noise.

Use of the electronic health record (EHR) for CVD surveillance is increasingly common. However, these data can introduce systematic error that influences the internal and external validity of study findings. We reviewed recent literature on EHR-based studies of CVD risk to summarize the most common types of bias that arise. Subsequently, we recommend strategies informed by work from others as well as our own to reduce the impact of these biases in future research. Systematic error, or bias, is a concern in all observational research including EHR-based studies of CVD risk surveillance. Patients captured in an EHR system may not be representative of the general population, due to issues such as informed presence bias, perceptions about the healthcare system that influence entry, and access to health services. Further, the EHR may contain inaccurate information or be missing key data points of interest due to loss to follow-up or over-diagnosis bias. Several strategies, including implementation of unique patient identifiers, adoption of standardized rules for inclusion/exclusion criteria, statistical procedures for data harmonization and analysis, and incorporation of patient-reported data have been used to reduce the impact of these biases. EHR data provide an opportunity to monitor and characterize CVD risk in populations. However, understanding the biases that arise from EHR datasets is instrumental in planning epidemiological studies and interpreting study findings. Strategies to reduce the impact of bias in the context of EHR data can increase the quality and utility of these data.

Evaluating common data models for use with a longitudinal community registry

ObjectiveTo evaluate common data models (CDMs) to determine which is best suited for sharing data from a large, longitudinal, electronic health record (EHR)-based community registry. Materials and methodsFour CDMs were chosen from models in use for clinical research data: Sentinel v5.0 (referred to as the Mini-Sentinel CDM in previous versions), PCORnet v3.0 (an extension of the Mini-Sentinel CDM), OMOP v5.0, and CDISC SDTM v1.4. Each model was evaluated against 11 criteria adapted from previous research. The criteria fell into six categories: content coverage, integrity, flexibility, ease of querying, standards compatibility, and ease and extent of implementation. ResultsThe OMOP CDM accommodated the highest percentage of our data elements (76%), fared well on other requirements, and had broader terminology coverage than the other models. Sentinel and PCORnet fell short in content coverage with 37% and 48% matches respectively. Although SDTM accommodated a significant percentage of data elements (55% true matches), 45% of the data elements mapped to SDTM’s extension mechanism, known as Supplemental Qualifiers, increasing the number of joins required to query the data. ConclusionThe OMOP CDM best met the criteria for supporting data sharing from longitudinal EHR-based studies. Conclusions may differ for other uses and associated data element sets, but the methodology reported here is easily adaptable to common data model evaluation for other uses.

A General Framework for Considering Selection Bias in EHR-Based Studies: What Data Are Observed and Why?

Electronic health records (EHR) data are increasingly seen as a resource for cost-effective comparative effectiveness research (CER). Since EHR data are collected primarily for clinical and/or billing purposes, their use for CER requires consideration of numerous methodologic challenges including the potential for confounding bias, due to a lack of randomization, and for selection bias, due to missing data. In contrast to the recent literature on confounding bias in EHR-based CER, virtually no attention has been paid to selection bias possibly due to the belief that standard methods for missing data can be readily-applied. Such methods, however, hinge on an overly simplistic view of the available/missing EHR data, so that their application in the EHR setting will often fail to completely control selection bias. Motivated by challenges we face in an on-going EHR-based comparative effectiveness study of choice of antidepressant treatment and long-term weight change, we propose a new general framework for selection bias in EHR-based CER. Crucially, the framework provides structure within which researchers can consider the complex interplay between numerous decisions, made by patients and health care providers, which give rise to health-related information being recorded in the EHR system, as well as the wide variability across EHR systems themselves. This, in turn, provides structure within which: (i) the transparency of assumptions regarding missing data can be enhanced, (ii) factors relevant to each decision can be elicited, and (iii) statistical methods can be better aligned with the complexity of the data.

Using Electronic Health Records for Population Health Research: A Review of Methods and Applications.

The use and functionality of electronic health records (EHRs) have increased rapidly in the past decade. Although the primary purpose of EHRs is clinical, researchers have used them to conduct epidemiologic investigations, ranging from cross-sectional studies within a given hospital to longitudinal studies on geographically distributed patients. Herein, we describe EHRs, examine their use in population health research, and compare them with traditional epidemiologic methods. We describe diverse research applications that benefit from the large sample sizes and generalizable patient populations afforded by EHRs. These have included reevaluation of prior findings, a range of diseases and subgroups, environmental and social epidemiology, stigmatized conditions, predictive modeling, and evaluation of natural experiments. Although studies using primary data collection methods may have more reliable data and better population retention, EHR-based studies are less expensive and require less time to complete. Future EHR epidemiology with enhanced collection of social/behavior measures, linkage with vital records, and integration of emerging technologies such as personal sensing could improve clinical care and population health.