Multi-ancestry genome- and phenome-wide association studies of diverticular disease in electronic health records with natural language processing enriched phenotyping algorithm.

OBJECTIVE: Diverticular disease (DD) is one of the most prevalent conditions encountered by gastroenterologists, affecting ~50% of Americans before the age of 60. Our aim was to identify genetic risk variants and clinical phenotypes associated with DD, leveraging multiple electronic health record (EHR) data sources of 91,166 multi-ancestry participants with a Natural Language Processing (NLP) technique. MATERIALS AND METHODS: We developed a NLP-enriched phenotyping algorithm that incorporated colonoscopy or abdominal imaging reports to identify patients with diverticulosis and diverticulitis from multicenter EHRs. We performed genome-wide association studies (GWAS) of DD in European, African and multi-ancestry participants, followed by phenome-wide association studies (PheWAS) of the risk variants to identify their potential comorbid/pleiotropic effects in clinical phenotypes. RESULTS: Our developed algorithm showed a significant improvement in patient classification performance for DD analysis (algorithm PPVs ≥ 0.94), with up to a 3.5 fold increase in terms of the number of identified patients than the traditional method. Ancestry-stratified analyses of diverticulosis and diverticulitis of the identified subjects replicated the well-established associations between ARHGAP15 loci with DD, showing overall intensified GWAS signals in diverticulitis patients compared to diverticulosis patients. Our PheWAS analyses identified significant associations between the DD GWAS variants and circulatory system, genitourinary, and neoplastic EHR phenotypes. DISCUSSION: As the first multi-ancestry GWAS-PheWAS study, we showcased that heterogenous EHR data can be mapped through an integrative analytical pipeline and reveal significant genotype-phenotype associations with clinical interpretation. CONCLUSION: A systematic framework to process unstructured EHR data with NLP could advance a deep and scalable phenotyping for better patient identification and facilitate etiological investigation of a disease with multilayered data.

Ambiguous and Incomplete: Natural Language Processing Reveals Problematic Reporting Styles in Thyroid Ultrasound Reports.

OBJECTIVE: Natural language processing (NLP) systems convert unstructured text into analyzable data. Here, we describe the performance measures of NLP to capture granular details on nodules from thyroid ultrasound (US) reports and reveal critical issues with reporting language. METHODS: We iteratively developed NLP tools using clinical Text Analysis and Knowledge Extraction System (cTAKES) and thyroid US reports from 2007 to 2013. We incorporated nine nodule features for NLP extraction. Next, we evaluated the precision, recall, and accuracy of our NLP tools using a separate set of US reports from an academic medical center (A) and a regional health care system (B) during the same period. Two physicians manually annotated each test-set report. A third physician then adjudicated discrepancies. The adjudicated "gold standard" was then used to evaluate NLP performance on the test-set. RESULTS: A total of 243 thyroid US reports contained 6,405 data elements. Inter-annotator agreement for all elements was 91.3%. Compared with the gold standard, overall recall of the NLP tool was 90%. NLP recall for thyroid lobe or isthmus characteristics was: laterality 96% and size 95%. NLP accuracy for nodule characteristics was: laterality 92%, size 92%, calcifications 76%, vascularity 65%, echogenicity 62%, contents 76%, and borders 40%. NLP recall for presence or absence of lymphadenopathy was 61%. Reporting style accounted for 18% errors. For example, the word "heterogeneous" interchangeably referred to nodule contents or echogenicity. While nodule dimensions and laterality were often described, US reports only described contents, echogenicity, vascularity, calcifications, borders, and lymphadenopathy, 46, 41, 17, 15, 9, and 41% of the time, respectively. Most nodule characteristics were equally likely to be described at hospital A compared with hospital B. CONCLUSIONS: NLP can automate extraction of critical information from thyroid US reports. However, ambiguous and incomplete reporting language hinders performance of NLP systems regardless of institutional setting. Standardized or synoptic thyroid US reports could improve NLP performance.

Under-specification as the source of ambiguity and vagueness in narrative phenotype algorithm definitions.

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.

Novel EDGE encoding method enhances ability to identify genetic interactions.

Assumptions are made about the genetic model of single nucleotide polymorphisms (SNPs) when choosing a traditional genetic encoding: additive, dominant, and recessive. Furthermore, SNPs across the genome are unlikely to demonstrate identical genetic models. However, running SNP-SNP interaction analyses with every combination of encodings raises the multiple testing burden. Here, we present a novel and flexible encoding for genetic interactions, the elastic data-driven genetic encoding (EDGE), in which SNPs are assigned a heterozygous value based on the genetic model they demonstrate in a dataset prior to interaction testing. We assessed the power of EDGE to detect genetic interactions using 29 combinations of simulated genetic models and found it outperformed the traditional encoding methods across 10%, 30%, and 50% minor allele frequencies (MAFs). Further, EDGE maintained a low false-positive rate, while additive and dominant encodings demonstrated inflation. We evaluated EDGE and the traditional encodings with genetic data from the Electronic Medical Records and Genomics (eMERGE) Network for five phenotypes: age-related macular degeneration (AMD), age-related cataract, glaucoma, type 2 diabetes (T2D), and resistant hypertension. A multi-encoding genome-wide association study (GWAS) for each phenotype was performed using the traditional encodings, and the top results of the multi-encoding GWAS were considered for SNP-SNP interaction using the traditional encodings and EDGE. EDGE identified a novel SNP-SNP interaction for age-related cataract that no other method identified: rs7787286 (MAF: 0.041; intergenic region of chromosome 7)-rs4695885 (MAF: 0.34; intergenic region of chromosome 4) with a Bonferroni LRT p of 0.018. A SNP-SNP interaction was found in data from the UK Biobank within 25 kb of these SNPs using the recessive encoding: rs60374751 (MAF: 0.030) and rs6843594 (MAF: 0.34) (Bonferroni LRT p: 0.026). We recommend using EDGE to flexibly detect interactions between SNPs exhibiting diverse action.

Medical records-based chronic kidney disease phenotype for clinical care and "big data" observational and genetic studies.

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.

Adverse Drug Reaction Discovery from Electronic Health Records with Deep Neural Networks.

Adverse drug reactions (ADRs) are detrimental and unexpected clinical incidents caused by drug intake. The increasing availability of massive quantities of longitudinal event data such as electronic health records (EHRs) has redefined ADR discovery as a big data analytics problem, where data-hungry deep neural networks are especially suitable because of the abundance of the data. To this end, we introduce neural self-controlled case series (NSCCS), a deep learning framework for ADR discovery from EHRs. NSCCS rigorously follows a self-controlled case series design to adjust implicitly and efficiently for individual heterogeneity. In this way, NSCCS is robust to time-invariant confounding issues and thus more capable of identifying associations that reflect the underlying mechanism between various types of drugs and adverse conditions. We apply NSCCS to a large-scale, real-world EHR dataset and empirically demonstrate its superior performance with comprehensive experiments on a benchmark ADR discovery task.

Preliminary outcomes of preemptive warfarin pharmacogenetic testing at a large rural healthcare center.

PURPOSE: As a preliminary evaluation of the outcomes of implementing pharmacogenetic testing within a large rural healthcare system, patients who received pre-emptive pharmacogenetic testing and warfarin dosing were monitored until June 2017. SUMMARY: Over a 20-month period, 749 patients were genotyped for VKORC1 and CYP2C9 as part of the electronic Medical Records and Genomics Pharmacogenetics (eMERGE PGx) study. Of these, 27 were prescribed warfarin and received an alert for pharmacogenetic testing pertinent to warfarin; 20 patients achieved their target international normalized ratio (INR) of 2.0-3.0, and 65% of these patients achieved target dosing within the recommended pharmacogenetic alert dose (± 0.5 mg/day). Of these, 10 patients had never been on warfarin prior to the alert and were further evaluated with regard to time to first stable target INR, bleeds and thromboembolic events, hospitalizations, and mortality. There was a general trend of faster time to first stable target INR when the patient was initiated at a warfarin dose within the alert recommendation versus a dose outside of the alert recommendation with a mean (± SD) of 34 (± 28) days versus 129 (± 117) days, respectively. No trends regarding bleeds, thromboembolic events, hospitalization, or mortality were identified with respect to the pharmacogenetic alert. The pharmacogenetic alert provided pharmacogenetic dosing information to prescribing clinicians and appeared to deploy appropriately with the correct recommendation based upon patient genotype. CONCLUSION: Implementing pharmacogenetic testing as a standard of care service in anticoagulation monitoring programs may improve dosage regimens for patients on anticoagulation therapy.

Comorbidity Characterization Among eMERGE Institutions: A Pilot Evaluation with the Johns Hopkins Adjusted Clinical Groups® System.

Electronic health records (EHR) are valuable to define phenotype selection algorithms used to identify cohorts ofpatients for sequencing or genome wide association studies (GWAS). To date, the electronic medical records and genomics (eMERGE) network institutions have developed and applied such algorithms to identify cohorts with associated DNA samples used to discover new genetic associations. For complex diseases, there are benefits to stratifying cohorts using comorbidities in order to identify their genetic determinants. The objective of this study was to: (a) characterize comorbidities in a range of phenotype-selected cohorts using the Johns Hopkins Adjusted Clinical Groups® (ACG®) System, (b) assess the frequency of important comorbidities in three commonly studied GWAS phenotypes, and (c) compare the comorbidity characterization of cases and controls. Our analysis demonstrates a framework to characterize comorbidities using the ACG system and identified differences in mean chronic condition count among GWAS cases and controls. Thus, we believe there is great potential to use the ACG system to characterize comorbidities among genetic cohorts selected based on EHR phenotypes.

Machine Learning Assisted Discovery of Novel Predictive Lab Tests Using Electronic Health Record Data.

Epidemiological studies identifying biological markers of disease state are valuable, but can be time-consuming, expensive, and require extensive intuition and expertise. Furthermore, not all hypothesized markers will be borne out in a study, suggesting that higher quality initial hypotheses are crucial. In this work, we propose a high-throughput pipeline to produce a ranked list of high-quality hypothesized marker laboratory tests for diagnoses. Our pipeline generates a large number of candidate lab-diagnosis hypotheses derived from machine learning models, filters and ranks them according to their potential novelty using text mining, and corroborate final hypotheses with logistic regression analysis. We test our approach on a large electronic health record dataset and the PubMed corpus, and find several promising candidate hypotheses.

Facilitating phenotype transfer using a common data model.

BACKGROUND: Implementing clinical phenotypes across a network is labor intensive and potentially error prone. Use of a common data model may facilitate the process. METHODS: Electronic Medical Records and Genomics (eMERGE) sites implemented the Observational Health Data Sciences and Informatics (OHDSI) Observational Medical Outcomes Partnership (OMOP) Common Data Model across their electronic health record (EHR)-linked DNA biobanks. Two previously implemented eMERGE phenotypes were converted to OMOP and implemented across the network. RESULTS: It was feasible to implement the common data model across sites, with laboratory data producing the greatest challenge due to local encoding. Sites were then able to execute the OMOP phenotype in less than one day, as opposed to weeks of effort to manually implement an eMERGE phenotype in their bespoke research EHR databases. Of the sites that could compare the current OMOP phenotype implementation with the original eMERGE phenotype implementation, specific agreement ranged from 100% to 43%, with disagreements due to the original phenotype, the OMOP phenotype, changes in data, and issues in the databases. Using the OMOP query as a standard comparison revealed differences in the original implementations despite starting from the same definitions, code lists, flowcharts, and pseudocode. CONCLUSION: Using a common data model can dramatically speed phenotype implementation at the cost of having to populate that data model, though this will produce a net benefit as the number of phenotype implementations increases. Inconsistencies among the implementations of the original queries point to a potential benefit of using a common data model so that actual phenotype code and logic can be shared, mitigating human error in reinterpretation of a narrative phenotype definition.

A Trans-Ethnic Genome-Wide Association Study of Uterine Fibroids.

Uterine fibroids affect up to 77% of women by menopause and account for up to $34 billion in healthcare costs each year. Although fibroid risk is heritable, genetic risk for fibroids is not well understood. We conducted a two-stage case-control meta-analysis of genetic variants in European and African ancestry women with and without fibroids classified by a previously published algorithm requiring pelvic imaging or confirmed diagnosis. Women from seven electronic Medical Records and Genomics (eMERGE) network sites (3,704 imaging-confirmed cases and 5,591 imaging-confirmed controls) and women of African and European ancestry from UK Biobank (UKB, 5,772 cases and 61,457 controls) were included in the discovery genome-wide association study (GWAS) meta-analysis. Variants showing evidence of association in Stage I GWAS (P < 1 × 10-5) were targeted in an independent replication sample of African and European ancestry individuals from the UKB (Stage II) (12,358 cases and 138,477 controls). Logistic regression models were fit with genetic markers imputed to a 1000 Genomes reference and adjusted for principal components for each race- and site-specific dataset, followed by fixed-effects meta-analysis. Final analysis with 21,804 cases and 205,525 controls identified 326 genome-wide significant variants in 11 loci, with three novel loci at chromosome 1q24 (sentinel-SNP rs14361789; P = 4.7 × 10-8), chromosome 16q12.1 (sentinel-SNP rs4785384; P = 1.5 × 10-9) and chromosome 20q13.1 (sentinel-SNP rs6094982; P = 2.6 × 10-8). Our statistically significant findings further support previously reported loci including SNPs near WT1, TNRC6B, SYNE1, BET1L, and CDC42/WNT4. We report evidence of ancestry-specific findings for sentinel-SNP rs10917151 in the CDC42/WNT4 locus (P = 1.76 × 10-24). Ancestry-specific effect-estimates for rs10917151 were in opposite directions (P-Het-between-groups = 0.04) for predominantly African (OR = 0.84) and predominantly European women (OR = 1.16). Genetically-predicted gene expression of several genes including LUZP1 in vagina (P = 4.6 × 10-8), OBFC1 in esophageal mucosa (P = 8.7 × 10-8), NUDT13 in multiple tissues including subcutaneous adipose tissue (P = 3.3 × 10-6), and HEATR3 in skeletal muscle tissue (P = 5.8 × 10-6) were associated with fibroids. The finding for HEATR3 was supported by SNP-based summary Mendelian randomization analysis. Our study suggests that fibroid risk variants act through regulatory mechanisms affecting gene expression and are comprised of alleles that are both ancestry-specific and shared across continental ancestries.

Machine learning for phenotyping opioid overdose events.

OBJECTIVE: To develop machine learning models for classifying the severity of opioid overdose events from clinical data. MATERIALS AND METHODS: Opioid overdoses were identified by diagnoses codes from the Marshfield Clinic population and assigned a severity score via chart review to form a gold standard set of labels. Three primary feature sets were constructed from disparate data sources surrounding each event and used to train machine learning models for phenotyping. RESULTS: Random forest and penalized logistic regression models gave the best performance with cross-validated mean areas under the ROC curves (AUCs) for all severity classes of 0.893 and 0.882 respectively. Features derived from a common data model outperformed features collected from disparate data sources for the same cohort of patients (AUCs 0.893 versus 0.837, p value = 0.002). The addition of features extracted from free text to machine learning models also increased AUCs from 0.827 to 0.893 (p value < 0.0001). Key word features extracted using natural language processing (NLP) such as 'Narcan' and 'Endotracheal Tube' are important for classifying overdose event severity. CONCLUSION: Random forest models using features derived from a common data model and free text can be effective for classifying opioid overdose events.

Heritability and genome-wide association study of benign prostatic hyperplasia (BPH) in the eMERGE network.

Benign prostatic hyperplasia (BPH) results in a significant public health burden due to the morbidity caused by the disease and many of the available remedies. As much as 70% of men over 70 will develop BPH. Few studies have been conducted to discover the genetic determinants of BPH risk. Understanding the biological basis for this condition may provide necessary insight for development of novel pharmaceutical therapies or risk prediction. We have evaluated SNP-based heritability of BPH in two cohorts and conducted a genome-wide association study (GWAS) of BPH risk using 2,656 cases and 7,763 controls identified from the Electronic Medical Records and Genomics (eMERGE) network. SNP-based heritability estimates suggest that roughly 60% of the phenotypic variation in BPH is accounted for by genetic factors. We used logistic regression to model BPH risk as a function of principal components of ancestry, age, and imputed genotype data, with meta-analysis performed using METAL. The top result was on chromosome 22 in SYN3 at rs2710383 (p-value = 4.6 × 10-7; Odds Ratio = 0.69, 95% confidence interval = 0.55-0.83). Other suggestive signals were near genes GLGC, UNCA13, SORCS1 and between BTBD3 and SPTLC3. We also evaluated genetically-predicted gene expression in prostate tissue. The most significant result was with increasing predicted expression of ETV4 (chr17; p-value = 0.0015). Overexpression of this gene has been associated with poor prognosis in prostate cancer. In conclusion, although there were no genome-wide significant variants identified for BPH susceptibility, we present evidence supporting the heritability of this phenotype, have identified suggestive signals, and evaluated the association between BPH and genetically-predicted gene expression in prostate.

Unfolding of hidden white blood cell count phenotypes for gene discovery using latent class mixed modeling.

Resting-state white blood cell (WBC) count is a marker of inflammation and immune system health. There is evidence that WBC count is not fixed over time and there is heterogeneity in WBC trajectory that is associated with morbidity and mortality. Latent class mixed modeling (LCMM) is a method that can identify unobserved heterogeneity in longitudinal data and attempts to classify individuals into groups based on a linear model of repeated measurements. We applied LCMM to repeated WBC count measures derived from electronic medical records of participants of the National Human Genetics Research Institute (NHRGI) electronic MEdical Record and GEnomics (eMERGE) network study, revealing two WBC count trajectory phenotypes. Advancing these phenotypes to GWAS, we found genetic associations between trajectory class membership and regions on chromosome 1p34.3 and chromosome 11q13.4. The chromosome 1 region contains CSF3R, which encodes the granulocyte colony-stimulating factor receptor. This protein is a major factor in neutrophil stimulation and proliferation. The association on chromosome 11 contain genes RNF169 and XRRA1; both involved in the regulation of double-strand break DNA repair.

Pharmacogenomic clinical decision support design and multi-site process outcomes analysis in the eMERGE Network.

To better understand the real-world effects of pharmacogenomic (PGx) alerts, this study aimed to characterize alert design within the eMERGE Network, and to establish a method for sharing PGx alert response data for aggregate analysis. Seven eMERGE sites submitted design details and established an alert logging data dictionary. Six sites participated in a pilot study, sharing alert response data from their electronic health record systems. PGx alert design varied, with some consensus around the use of active, post-test alerts to convey Clinical Pharmacogenetics Implementation Consortium recommendations. Sites successfully shared response data, with wide variation in acceptance and follow rates. Results reflect the lack of standardization in PGx alert design. Standards and/or larger studies will be necessary to fully understand PGx impact. This study demonstrated a method for sharing PGx alert response data and established that variation in system design is a significant barrier for multi-site analyses.

A Machine-Learning-Based Drug Repurposing Approach Using Baseline Regularization.

We present the baseline regularization model for computational drug repurposing using electronic health records (EHRs). In EHRs, drug prescriptions of various drugs are recorded throughout time for various patients. In the same time, numeric physical measurements (e.g., fasting blood glucose level) are also recorded. Baseline regularization uses statistical relationships between the occurrences of prescriptions of some particular drugs and the increase or the decrease in the values of some particular numeric physical measurements to identify potential repurposing opportunities.

Probing the Virtual Proteome to Identify Novel Disease Biomarkers.

BACKGROUND: Proteomic approaches allow measurement of thousands of proteins in a single specimen, which can accelerate biomarker discovery. However, applying these technologies to massive biobanks is not currently feasible because of the practical barriers and costs of implementing such assays at scale. To overcome these challenges, we used a "virtual proteomic" approach, linking genetically predicted protein levels to clinical diagnoses in >40 000 individuals. METHODS: We used genome-wide association data from the Framingham Heart Study (n=759) to construct genetic predictors for 1129 plasma protein levels. We validated the genetic predictors for 268 proteins and used them to compute predicted protein levels in 41 288 genotyped individuals in the Electronic Medical Records and Genomics (eMERGE) cohort. We tested associations for each predicted protein with 1128 clinical phenotypes. Lead associations were validated with directly measured protein levels and either low-density lipoprotein cholesterol or subclinical atherosclerosis in the MDCS (Malmö Diet and Cancer Study; n=651). RESULTS: In the virtual proteomic analysis in eMERGE, 55 proteins were associated with 89 distinct diagnoses at a false discovery rate q<0.1. Among these, 13 associations involved lipid (n=7) or atherosclerosis (n=6) phenotypes. We tested each association for validation in MDCS using directly measured protein levels. At Bonferroni-adjusted significance thresholds, levels of apolipoprotein E isoforms were associated with hyperlipidemia, and circulating C-type lectin domain family 1 member B and platelet-derived growth factor receptor-ß predicted subclinical atherosclerosis. Odds ratios for carotid atherosclerosis were 1.31 (95% CI, 1.08-1.58; P=0.006) per 1-SD increment in C-type lectin domain family 1 member B and 0.79 (0.66-0.94; P=0.008) per 1-SD increment in platelet-derived growth factor receptor-ß. CONCLUSIONS: We demonstrate a biomarker discovery paradigm to identify candidate biomarkers of cardiovascular and other diseases.

A study paradigm integrating prospective epidemiologic cohorts and electronic health records to identify disease biomarkers.

Defining the full spectrum of human disease associated with a biomarker is necessary to advance the biomarker into clinical practice. We hypothesize that associating biomarker measurements with electronic health record (EHR) populations based on shared genetic architectures would establish the clinical epidemiology of the biomarker. We use Bayesian sparse linear mixed modeling to calculate SNP weightings for 53 biomarkers from the Atherosclerosis Risk in Communities study. We use the SNP weightings to computed predicted biomarker values in an EHR population and test associations with 1139 diagnoses. Here we report 116 associations meeting a Bonferroni level of significance. A false discovery rate (FDR)-based significance threshold reveals more known and undescribed associations across a broad range of biomarkers, including biometric measures, plasma proteins and metabolites, functional assays, and behaviors. We confirm an inverse association between LDL-cholesterol level and septicemia risk in an independent epidemiological cohort. This approach efficiently discovers biomarker-disease associations.