Search

Abstract The electronic Medical Records and Genomics (eMERGE) Network assessed the feasibility of deploying portable phenotype rule-based algorithms with natural language processing (NLP) components added to improve performance of existing algorithms using electronic health records (EHRs). Based on scientific merit and predicted difficulty, eMERGE selected six existing phenotypes to enhance with NLP. We assessed performance, portability, and ease of use. We summarized lessons learned by: (1) challenges; (2) best practices to address challenges based on existing evidence and/or eMERGE experience; and (3) opportunities for future research. Adding NLP resulted in improved, or the same, precision and/or recall for all but one algorithm. Portability, phenotyping workflow/process, and technology were major themes. With NLP, development and validation took longer. Besides portability of NLP technology and algorithm replicability, factors to ensure success include privacy protection, technical infrastructure setup, intellectual property agreement, and efficient communication. Workflow improvements can improve communication and reduce implementation time. NLP performance varied mainly due to clinical document heterogeneity; therefore, we suggest using semi-structured notes, comprehensive documentation, and customization options. NLP portability is possible with improved phenotype algorithm performance, but careful planning and architecture of the algorithms is essential to support local customizations.

Abstract Background Currently available medications for chronic osteoarthritis pain are only moderately effective, and their use is limited in many patients because of serious adverse effects and contraindications. The primary surgical option for osteoarthritis is total joint replacement (TJR). The objectives of this study were to describe the treatment history of patients with osteoarthritis receiving prescription pain medications and/or intra-articular corticosteroid injections, and to estimate the incidence of TJR in these patients. Methods This retrospective, multicenter, cohort study utilized health plan administrative claims data (January 1, 2013, through December 31, 2019) of adult patients with osteoarthritis in the Innovation in Medical Evidence Development and Surveillance Distributed Database, a subset of the US FDA Sentinel Distributed Database. Patients were analyzed in two cohorts: those with prevalent use of “any pain medication” (prescription non-steroidal anti-inflammatory drugs [NSAIDs], opioids, and/or intra-articular corticosteroid injections) using only the first qualifying dispensing (index date); and those with prevalent use of “each specific pain medication class” with all qualifying treatment episodes identified. Results Among 1 992 670 prevalent users of “any pain medication”, pain medications prescribed on the index date were NSAIDs (596 624 [29.9%] patients), opioids (1 161 806 [58.3%]), and intra-articular corticosteroids (323 459 [16.2%]). Further, 92 026 patients received multiple pain medications on the index date, including 71 632 (3.6%) receiving both NSAIDs and opioids. Altogether, 20.6% of patients used an NSAID at any time following an opioid index dispensing and 17.2% used an opioid following an NSAID index dispensing. The TJR incidence rates per 100 person-years (95% confidence interval [CI]) were 3.21 (95% CI: 3.20–3.23) in the “any pain medication” user cohort, and among those receiving “each specific pain medication class” were NSAIDs, 4.63 (95% CI: 4.58–4.67); opioids, 7.45 (95% CI: 7.40–7.49); and intra-articular corticosteroids, 8.05 (95% CI: 7.97–8.13). Conclusions In patients treated with prescription medications for osteoarthritis pain, opioids were more commonly prescribed at index than NSAIDs and intra-articular corticosteroid injections. Of the pain medication classes examined, the incidence of TJR was highest in patients receiving intra-articular corticosteroids and lowest in patients receiving NSAIDs.

Abstract Background Patients and their loved ones often report symptoms or complaints of cognitive decline that clinicians note in free clinical text, but no structured screening or diagnostic data are recorded. These symptoms/complaints may be signals that predict who will go on to be diagnosed with mild cognitive impairment (MCI) and ultimately develop Alzheimer’s Disease or related dementias. Our objective was to develop a natural language processing system and prediction model for identification of MCI from clinical text in the absence of screening or other structured diagnostic information. Methods There were two populations of patients: 1794 participants in the Adult Changes in Thought (ACT) study and 2391 patients in the general population of Kaiser Permanente Washington. All individuals had standardized cognitive assessment scores. We excluded patients with a diagnosis of Alzheimer’s Disease, Dementia or use of donepezil. We manually annotated 10,391 clinic notes to train the NLP model. Standard Python code was used to extract phrases from notes and map each phrase to a cognitive functioning concept. Concepts derived from the NLP system were used to predict future MCI. The prediction model was trained on the ACT cohort and 60% of the general population cohort with 40% withheld for validation. We used a least absolute shrinkage and selection operator logistic regression approach (LASSO) to fit a prediction model with MCI as the prediction target. Using the predicted case status from the LASSO model and known MCI from standardized scores, we constructed receiver operating curves to measure model performance. Results Chart abstraction identified 42 MCI concepts. Prediction model performance in the validation data set was modest with an area under the curve of 0.67. Setting the cutoff for correct classification at 0.60, the classifier yielded sensitivity of 1.7%, specificity of 99.7%, PPV of 70% and NPV of 70.5% in the validation cohort. Discussion and conclusion Although the sensitivity of the machine learning model was poor, negative predictive value was high, an important characteristic of models used for population-based screening. While an AUC of 0.67 is generally considered moderate performance, it is also comparable to several tests that are widely used in clinical practice.

Abstract Introduction Currently, one of the commonly used methods for disseminating electronic health record (EHR)-based phenotype algorithms is providing a narrative description of the algorithm logic, often accompanied by flowcharts. A challenge with this mode of dissemination is the potential for under-specification in the algorithm definition, which leads to ambiguity and vagueness. Methods This study examines incidents of under-specification that occurred during the implementation of 34 narrative phenotyping algorithms in the electronic Medical Record and Genomics (eMERGE) network. We reviewed the online communication history between algorithm developers and implementers within the Phenotype Knowledge Base (PheKB) platform, where questions could be raised and answered regarding the intended implementation of a phenotype algorithm. Results We developed a taxonomy of under-specification categories via an iterative review process between two groups of annotators. Under-specifications that lead to ambiguity and vagueness were consistently found across narrative phenotype algorithms developed by all involved eMERGE sites. Discussion and conclusion Our findings highlight that under-specification is an impediment to the accuracy and efficiency of the implementation of current narrative phenotyping algorithms, and we propose approaches for mitigating these issues and improved methods for disseminating EHR phenotyping algorithms.

Abstract Chronic Kidney Disease (CKD) represents a slowly progressive disorder that is typically silent until late stages, but early intervention can significantly delay its progression. We designed a portable and scalable electronic CKD phenotype to facilitate early disease recognition and empower large-scale observational and genetic studies of kidney traits. The algorithm uses a combination of rule-based and machine-learning methods to automatically place patients on the staging grid of albuminuria by glomerular filtration rate (“A-by-G” grid). We manually validated the algorithm by 451 chart reviews across three medical systems, demonstrating overall positive predictive value of 95% for CKD cases and 97% for healthy controls. Independent case-control validation using 2350 patient records demonstrated diagnostic specificity of 97% and sensitivity of 87%. Application of the phenotype to 1.3 million patients demonstrated that over 80% of CKD cases are undetected using ICD codes alone. We also demonstrated several large-scale applications of the phenotype, including identifying stage-specific kidney disease comorbidities, in silico estimation of kidney trait heritability in thousands of pedigrees reconstructed from medical records, and biobank-based multicenter genome-wide and phenome-wide association studies.

Abstract Background Elevated triglycerides (TG) are associated with, and may be causal for, cardiovascular disease (CVD), and co-morbidities such as type II diabetes and metabolic syndrome. Pathogenic variants in APOA5 and APOC3 as well as risk SNVs in other genes [APOE (rs429358, rs7412), APOA1/C3/A4/A5 gene cluster (rs964184), INSR (rs7248104), CETP (rs7205804), GCKR (rs1260326)] have been shown to affect TG levels. Knowledge of genetic causes for elevated TG may lead to early intervention and targeted treatment for CVD. We previously identified linkage and association of a rare, highly conserved missense variant in SLC25A40, rs762174003, with hypertriglyceridemia (HTG) in a single large family, and replicated this association with rare, highly conserved missense variants in a European American and African American sample. Methods Here, we analyzed a longitudinal mixed-ancestry cohort (European, African and Asian ancestry, N = 8966) from the Electronic Medical Record and Genomics (eMERGE) Network. We tested associations between median TG and the genes of interest, using linear regression, adjusting for sex, median age, median BMI, and the first two principal components of ancestry. Results We replicated the association between TG and APOC3, APOA5, and risk variation at APOE, APOA1/C3/A4/A5 gene cluster, and GCKR. We failed to replicate the association between rare, highly conserved variation at SLC25A40 and TG, as well as for risk variation at INSR and CETP. Conclusions Analysis using data from electronic health records presents challenges that need to be overcome. Although large amounts of genotype data is becoming increasingly accessible, usable phenotype data can be challenging to obtain. We were able to replicate known, strong associations, but were unable to replicate moderate associations due to the limited sample size and missing drug information.

Thus far, pleiotropy analysis using individual-level Electronic Health Records data has been limited to data from one site. Here, the authors introduce Sum-Share, a method designed to efficiently and losslessly integrate EHR and genetic data from multiple sites to perform pleiotropy analysis.

Abstract Background We characterised the phenotypic consequence of genetic variation at the PCSK9 locus and compared findings with recent trials of pharmacological inhibitors of PCSK9. Methods Published and individual participant level data (300,000+ participants) were combined to construct a weighted PCSK9 gene-centric score (GS). Seventeen randomized placebo controlled PCSK9 inhibitor trials were included, providing data on 79,578 participants. Results were scaled to a one mmol/L lower LDL-C concentration. Results The PCSK9 GS (comprising 4 SNPs) associations with plasma lipid and apolipoprotein levels were consistent in direction with treatment effects. The GS odds ratio (OR) for myocardial infarction (MI) was 0.53 (95% CI 0.42; 0.68), compared to a PCSK9 inhibitor effect of 0.90 (95% CI 0.86; 0.93). For ischemic stroke ORs were 0.84 (95% CI 0.57; 1.22) for the GS, compared to 0.85 (95% CI 0.78; 0.93) in the drug trials. ORs with type 2 diabetes mellitus (T2DM) were 1.29 (95% CI 1.11; 1.50) for the GS, as compared to 1.00 (95% CI 0.96; 1.04) for incident T2DM in PCSK9 inhibitor trials. No genetic associations were observed for cancer, heart failure, atrial fibrillation, chronic obstructive pulmonary disease, or Alzheimer’s disease – outcomes for which large-scale trial data were unavailable. Conclusions Genetic variation at the PCSK9 locus recapitulates the effects of therapeutic inhibition of PCSK9 on major blood lipid fractions and MI. While indicating an increased risk of T2DM, no other possible safety concerns were shown; although precision was moderate.

Abstract Background Non-alcoholic fatty liver disease (NAFLD) is a common chronic liver illness with a genetically heterogeneous background that can be accompanied by considerable morbidity and attendant health care costs. The pathogenesis and progression of NAFLD is complex with many unanswered questions. We conducted genome-wide association studies (GWASs) using both adult and pediatric participants from the Electronic Medical Records and Genomics (eMERGE) Network to identify novel genetic contributors to this condition. Methods First, a natural language processing (NLP) algorithm was developed, tested, and deployed at each site to identify 1106 NAFLD cases and 8571 controls and histological data from liver tissue in 235 available participants. These include 1242 pediatric participants (396 cases, 846 controls). The algorithm included billing codes, text queries, laboratory values, and medication records. Next, GWASs were performed on NAFLD cases and controls and case-only analyses using histologic scores and liver function tests adjusting for age, sex, site, ancestry, PC, and body mass index (BMI). Results Consistent with previous results, a robust association was detected for the PNPLA3 gene cluster in participants with European ancestry. At the PNPLA3-SAMM50 region, three SNPs, rs738409, rs738408, and rs3747207, showed strongest association (best SNP rs738409 p = 1.70 × 10− 20). This effect was consistent in both pediatric (p = 9.92 × 10− 6) and adult (p = 9.73 × 10− 15) cohorts. Additionally, this variant was also associated with disease severity and NAFLD Activity Score (NAS) (p = 3.94 × 10− 8, beta = 0.85). PheWAS analysis link this locus to a spectrum of liver diseases beyond NAFLD with a novel negative correlation with gout (p = 1.09 × 10− 4). We also identified novel loci for NAFLD disease severity, including one novel locus for NAS score near IL17RA (rs5748926, p = 3.80 × 10− 8), and another near ZFP90-CDH1 for fibrosis (rs698718, p = 2.74 × 10− 11). Post-GWAS and gene-based analyses identified more than 300 genes that were used for functional and pathway enrichment analyses. Conclusions In summary, this study demonstrates clear confirmation of a previously described NAFLD risk locus and several novel associations. Further collaborative studies including an ethnically diverse population with well-characterized liver histologic features of NAFLD are needed to further validate the novel findings.

Biomarker identification requires prohibitively large cohorts with gene expression and phenotype data. The approach introduced here learns polygenic predictors of expression from genetic and expression data, used to infer biomarker levels in patients with genetic and disease information.

Abstract Background The genetic etiology of human lipid quantitative traits is not fully elucidated, and interactions between variants may play a role. We performed a gene-centric interaction study for four different lipid traits: low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), and triglycerides (TG). Results Our analysis consisted of a discovery phase using a merged dataset of five different cohorts (n = 12,853 to n = 16,849 depending on lipid phenotype) and a replication phase with ten independent cohorts totaling up to 36,938 additional samples. Filters are often applied before interaction testing to correct for the burden of testing all pairwise interactions. We used two different filters: 1. A filter that tested only single nucleotide polymorphisms (SNPs) with a main effect of p

BackgroundIntracranial aneurysms (IAs), abdominal aortic aneurysms (AAAs), and thoracic aortic aneurysms (TAAs) all have a familial predisposition. Given that aneurysm types are known to co‐occur, we hypothesized that there may be shared genetic risk factors for IAs, AAAs, and TAAs. Methods and ResultsWe performed a mega‐analysis of 1000 Genomes Project‐imputed genome‐wide association study (GWAS) data of 4 previously published aneurysm cohorts: 2 IA cohorts (in total 1516 cases, 4305 controls), 1 AAA cohort (818 cases, 3004 controls), and 1 TAA cohort (760 cases, 2212 controls), and observed associations of 4 known IA, AAA, and/or TAA risk loci (9p21, 18q11, 15q21, and 2q33) with consistent effect directions in all 4 cohorts. We calculated polygenic scores based on IA‐, AAA‐, and TAA‐associated SNPs and tested these scores for association to case‐control status in the other aneurysm cohorts; this revealed no shared polygenic effects. Similarly, linkage disequilibrium–score regression analyses did not show significant correlations between any pair of aneurysm subtypes. Last, we evaluated the evidence for 14 previously published aneurysm risk single‐nucleotide polymorphisms through collaboration in extended aneurysm cohorts, with a total of 6548 cases and 16 843 controls (IA) and 4391 cases and 37 904 controls (AAA), and found nominally significant associations for IA risk locus 18q11 near RBBP8 to AAA (odds ratio [OR]=1.11; P=4.1×10−5) and for TAA risk locus 15q21 near FBN1 to AAA (OR=1.07; P=1.1×10−3). ConclusionsAlthough there was no evidence for polygenic overlap between IAs, AAAs, and TAAs, we found nominally significant effects of two established risk loci for IAs and TAAs in AAAs. These two loci will require further replication.

Phenome-wide association studies (PheWAS) have demonstrated utility in validating genetic associations derived from traditional genetic studies as well as identifying novel genetic associations. Here we used an electronic health record (EHR)-based PheWAS to explore pleiotropy of genetic variants in the fat mass and obesity associated gene (FTO), some of which have been previously associated with obesity and type 2 diabetes (T2D). We used a population of 10,487 individuals of European ancestry with genome-wide genotyping from the Electronic Medical Records and Genomics (eMERGE) Network and another population of 13,711 individuals of European ancestry from the BioVU DNA biobank at Vanderbilt genotyped using Illumina HumanExome BeadChip. A meta-analysis of the two study populations replicated the well-described associations between FTO variants and obesity (odds ratio [OR]=1.25, 95% Confidence Interval=1.11-1.24, p=2.10 x 10 9) and FTO variants and T2D (OR=1.14, 95% CI=1.08-1.21, p=2.34 x 10 6). The meta-analysis also demonstrated that FTO variant rs8050136 was significantly associated with sleep apnea (OR=1.14, 95% CI=1.07-1.22, p=3.33 x 10 5); however, the association was attenuated after adjustment for body mass index (BMI). Novel phenotype associations with obesity-associated FTO variants included fibrocystic breast disease (rs9941349, OR=0.81, 95% CI=0.74-0.91, p=5.41x10 5) and trends toward associations with nonalcoholic liver disease and gram-positive bacterial infections. FTO variants not associated with obesity demonstrated other potential disease associations including noninflammatory disorders of the cervix and chronic periodontitis. These results suggest that genetic variants in FTO may have pleiotropic associations, some of which are not mediated by obesity.

A single mutation can alter cellular and global homeostatic mechanisms and give rise to multiple clinical diseases. We hypothesized that these disease mechanisms could be identified using low minor allele frequency (MAF

Background: Clinical text contains valuable information but must be de-identified before it can be used for secondary purposes. Accurate annotation of personally identifiable information (PII) is essential to the development of automated de-identification systems and to manual redaction of PII. Yet the accuracy of annotations may vary considerably across individual annotators and annotation is costly. As such, the marginal benefit of incorporating additional annotators has not been well characterized., Objectives: This study models the costs and benefits of incorporating increasing numbers of independent human annotators to identify the instances of PII in a corpus. We used a corpus with gold standard annotations to evaluate the performance of teams of annotators of increasing size., Methods: Four annotators independently identified PII in a 100-document corpus consisting of randomly selected clinical notes from Family Practice clinics in a large integrated health care system. These annotations were pooled and validated to generate a gold standard corpus for evaluation., Results: Recall rates for all PII types ranged from 0.90 to 0.98 for individual annotators to 0.998 to 1.0 for teams of three, when meas-ured against the gold standard. Median cost per PII instance discovered during corpus annotation ranged from $ 0.71 for an individual annotator to $ 377 for annotations discovered only by a fourth annotator., Conclusions: Incorporating a second annotator into a PII annotation process reduces unredacted PII and improves the quality of annotations to 0.99 recall, yielding clear benefit at reasonable cost; the cost advantages of annotation teams larger than two diminish rapidly.