OBJECTIVES: To develop, validate, and implement algorithms to identify diabetic retinopathy (DR) cases and controls from electronic health care records (EHRs). MATERIALS AND METHODS: We developed and validated electronic health record (EHR)-based algorithms to identify DR cases and individuals with type I or II diabetes without DR (controls) in 3 independent EHR systems: Vanderbilt University Medical Center Synthetic Derivative (VUMC), the VA Northeast Ohio Healthcare System (VANEOHS), and Massachusetts General Brigham (MGB). Cases were required to meet 1 of the following 3 criteria: (1) 2 or more dates with any DR ICD-9/10 code documented in the EHR, (2) at least one affirmative health-factor or EPIC code for DR along with an ICD9/10 code for DR on a different day, or (3) at least one ICD-9/10 code for any DR occurring within 24 hours of an ophthalmology examination. Criteria for controls included affirmative evidence for diabetes as well as an ophthalmology examination. RESULTS: The algorithms, developed and evaluated in VUMC through manual chart review, resulted in a positive predictive value (PPV) of 0.93 for cases and negative predictive value (NPV) of 0.91 for controls. Implementation of algorithms yielded similar metrics in VANEOHS (PPV = 0.94; NPV = 0.86) and lower in MGB (PPV = 0.84; NPV = 0.76). In comparison, the algorithm for DR implemented in Phenome-wide association study (PheWAS) in VUMC yielded similar PPV (0.92) but substantially reduced NPV (0.48). Implementation of the algorithms to the Million Veteran Program identified over 62 000 DR cases with genetic data including 14 549 African Americans and 6209 Hispanics with DR. CONCLUSIONS/DISCUSSION: We demonstrate the robustness of the algorithms at 3 separate healthcare centers, with a minimum PPV of 0.84 and substantially improved NPV than existing automated methods. We strongly encourage independent validation and incorporation of features unique to each EHR to enhance algorithm performance for DR cases and controls.

MOTIVATION: Making sense of networked multivariate association patterns is vitally important to many areas of high-dimensional analysis. Unfortunately, as the data-space dimensions grow, the number of association pairs increases in O(n2); this means that traditional visualizations such as heatmaps quickly become too complicated to parse effectively. RESULTS: Here, we present associationSubgraphs: a new interactive visualization method to quickly and intuitively explore high-dimensional association datasets using network percolation and clustering. The goal is to provide an efficient investigation of association subgraphs, each containing a subset of variables with stronger and more frequent associations among themselves than the remaining variables outside the subset, by showing the entire clustering dynamics and providing subgraphs under all possible cutoff values at once. Particularly, we apply associationSubgraphs to a phenome-wide multimorbidity association matrix generated from an electronic health record and provide an online, interactive demonstration for exploring multimorbidity subgraphs. AVAILABILITY AND IMPLEMENTATION: An R package implementing both the algorithm and visualization components of associationSubgraphs is available at https://github.com/tbilab/associationsubgraphs. Online documentation is available at https://prod.tbilab.org/associationsubgraphs_info/. A demo using a multimorbidity association matrix is available at https://prod.tbilab.org/associationsubgraphs-example/.

OBJECTIVE: Developing algorithms to extract phenotypes from electronic health records (EHRs) can be challenging and time-consuming. We developed PheMap, a high-throughput phenotyping approach that leverages multiple independent, online resources to streamline the phenotyping process within EHRs. MATERIALS AND METHODS: PheMap is a knowledge base of medical concepts with quantified relationships to phenotypes that have been extracted by natural language processing from publicly available resources. PheMap searches EHRs for each phenotype's quantified concepts and uses them to calculate an individual's probability of having this phenotype. We compared PheMap to clinician-validated phenotyping algorithms from the Electronic Medical Records and Genomics (eMERGE) network for type 2 diabetes mellitus (T2DM), dementia, and hypothyroidism using 84 821 individuals from Vanderbilt Univeresity Medical Center's BioVU DNA Biobank. We implemented PheMap-based phenotypes for genome-wide association studies (GWAS) for T2DM, dementia, and hypothyroidism, and phenome-wide association studies (PheWAS) for variants in FTO, HLA-DRB1, and TCF7L2. RESULTS: In this initial iteration, the PheMap knowledge base contains quantified concepts for 841 disease phenotypes. For T2DM, dementia, and hypothyroidism, the accuracy of the PheMap phenotypes were >97% using a 50% threshold and eMERGE case-control status as a reference standard. In the GWAS analyses, PheMap-derived phenotype probabilities replicated 43 of 51 previously reported disease-associated variants for the 3 phenotypes. For 9 of the 11 top associations, PheMap provided an equivalent or more significant P value than eMERGE-based phenotypes. The PheMap-based PheWAS showed comparable or better performance to a traditional phecode-based PheWAS. PheMap is publicly available online. CONCLUSIONS: PheMap significantly streamlines the process of extracting research-quality phenotype information from EHRs, with comparable or better performance to current phenotyping approaches.

OBJECTIVE: Phenotyping algorithms applied to electronic health record (EHR) data enable investigators to identify large cohorts for clinical and genomic research. Algorithm development is often iterative, depends on fallible investigator intuition, and is time- and labor-intensive. We developed and evaluated 4 types of phenotyping algorithms and categories of EHR information to identify hypertensive individuals and controls and provide a portable module for implementation at other sites. MATERIALS AND METHODS: We reviewed the EHRs of 631 individuals followed at Vanderbilt for hypertension status. We developed features and phenotyping algorithms of increasing complexity. Input categories included International Classification of Diseases, Ninth Revision (ICD9) codes, medications, vital signs, narrative-text search results, and Unified Medical Language System (UMLS) concepts extracted using natural language processing (NLP). We developed a module and tested portability by replicating 10 of the best-performing algorithms at the Marshfield Clinic. RESULTS: Random forests using billing codes, medications, vitals, and concepts had the best performance with a median area under the receiver operator characteristic curve (AUC) of 0.976. Normalized sums of all 4 categories also performed well (0.959 AUC). The best non-NLP algorithm combined normalized ICD9 codes, medications, and blood pressure readings with a median AUC of 0.948. Blood pressure cutoffs or ICD9 code counts alone had AUCs of 0.854 and 0.908, respectively. Marshfield Clinic results were similar. CONCLUSION: This work shows that billing codes or blood pressure readings alone yield good hypertension classification performance. However, even simple combinations of input categories improve performance. The most complex algorithms classified hypertension with excellent recall and precision.

With the recent success of genome-wide association studies (GWAS), a wealth of association data has been accomplished for more than 200 complex diseases/traits, proposing a strong demand for data integration and interpretation. A combinatory analysis of multiple GWAS datasets, or an integrative analysis of GWAS data and other high-throughput data, has been particularly promising. In this study, we proposed an integrative analysis framework of multiple GWAS datasets by overlaying association signals onto the protein-protein interaction network, and demonstrated it using schizophrenia datasets. Building on a dense module search algorithm, we first searched for significantly enriched subnetworks for schizophrenia in each single GWAS dataset and then implemented a discovery-evaluation strategy to identify module genes with consistent association signals. We validated the module genes in an independent dataset, and also examined them through meta-analysis of the related SNPs using multiple GWAS datasets. As a result, we identified 205 module genes with a joint effect significantly associated with schizophrenia; these module genes included a number of well-studied candidate genes such as DISC1, GNA12, GNA13, GNAI1, GPR17, and GRIN2B. Further functional analysis suggested these genes are involved in neuronal related processes. Additionally, meta-analysis found that 18 SNPs in 9 module genes had P(meta)<1 × 10⁻⁴, including the gene HLA-DQA1 located in the MHC region on chromosome 6, which was reported in previous studies using the largest cohort of schizophrenia patients to date. These results demonstrated our bi-directional network-based strategy is efficient for identifying disease-associated genes with modest signals in GWAS datasets. This approach can be applied to any other complex diseases/traits where multiple GWAS datasets are available.

MOTIVATION: Next-generation targeted resequencing of genome-wide association study (GWAS)-associated genomic regions is a common approach for follow-up of indirect association of common alleles. However, it is prohibitively expensive to sequence all the samples from a well-powered GWAS study with sufficient depth of coverage to accurately call rare genotypes. As a result, many studies may use next-generation sequencing for single nucleotide polymorphism (SNP) discovery in a smaller number of samples, with the intent to genotype candidate SNPs with rare alleles captured by resequencing. This approach is reasonable, but may be inefficient for rare alleles if samples are not carefully selected for the resequencing experiment. RESULTS: We have developed a probability-based approach, SampleSeq, to select samples for a targeted resequencing experiment that increases the yield of rare disease alleles substantially over random sampling of cases or controls or sampling based on genotypes at associated SNPs from GWAS data. This technique allows for smaller sample sizes for resequencing experiments, or allows the capture of rarer risk alleles. When following up multiple regions, SampleSeq selects subjects with an even representation of all the regions. SampleSeq also can be used to calculate the sample size needed for the resequencing to increase the chance of successful capture of rare alleles of desired frequencies. SOFTWARE: http://biostat.mc.vanderbilt.edu/SampleSeq

BACKGROUND: Multifactor Dimensionality Reduction (MDR) has been introduced previously as a non-parametric statistical method for detecting gene-gene interactions. MDR performs a dimensional reduction by assigning multi-locus genotypes to either high- or low-risk groups and measuring the percentage of cases and controls incorrectly labelled by this classification - the classification error. The combination of variables that produces the lowest classification error is selected as the best or most fit model. The correctly and incorrectly labelled cases and controls can be expressed as a two-way contingency table. We sought to improve the ability of MDR to detect gene-gene interactions by replacing classification error with a different measure to score model quality. RESULTS: In this study, we compare the detection and power of MDR using a variety of measures for two-way contingency table analysis. We simulated 40 genetic models, varying the number of disease loci in the model (2 - 5), allele frequencies of the disease loci (.2/.8 or .4/.6) and the broad-sense heritability of the model (.05 - .3). Overall, detection using NMI was 65.36% across all models, and specific detection was 59.4% versus detection using classification error at 62% and specific detection was 52.2%. CONCLUSION: Of the 10 measures evaluated, the likelihood ratio and normalized mutual information (NMI) are measures that consistently improve the detection and power of MDR in simulated data over using classification error. These measures also reduce the inclusion of spurious variables in a multi-locus model. Thus, MDR, which has already been demonstrated as a powerful tool for detecting gene-gene interactions, can be improved with the use of alternative fitness functions.