Search

Rationale: Birth cohort studies have identified several temporal patterns of wheezing, only some of which are associated with asthma. Whether 17q12-21 genetic variants, which are closely associated...

Background Machine learning is increasingly used for risk stratification in health care. Achieving accurate predictive models do not improve outcomes if they cannot be translated into efficacious intervention. Here we examine the potential utility of automated risk stratification and referral intervention to screen older adults for fall risk after emergency department (ED) visits. Objective This study evaluated several machine learning methodologies for the creation of a risk stratification algorithm using electronic health record data and estimated the effects of a resultant intervention based on algorithm performance in test data. Methods Data available at the time of ED discharge were retrospectively collected and separated into training and test datasets. Algorithms were developed to predict the outcome of a return visit for fall within 6 months of an ED index visit. Models included random forests, AdaBoost, and regression-based methods. We evaluated models both by the area under the receiver operating characteristic (ROC) curve, also referred to as area under the curve (AUC), and by projected clinical impact, estimating number needed to treat (NNT) and referrals per week for a fall risk intervention. Results The random forest model achieved an AUC of 0.78, with slightly lower performance in regression-based models. Algorithms with similar performance, when evaluated by AUC, differed when placed into a clinical context with the defined task of estimated NNT in a real-world scenario. Conclusion The ability to translate the results of our analysis to the potential tradeoff between referral numbers and NNT offers decisionmakers the ability to envision the effects of a proposed intervention before implementation.

Just under half of the 85.7 million US adults with hypertension have uncontrolled blood pressure using a hypertension threshold of systolic pressure ≥ 140 or diastolic pressure ≥ 90. Uncontrolled hypertension increases risks of death, stroke, heart failure, and myocardial infarction. Guidelines on hypertension management include lifestyle modification such as diet and exercise. In order to improve hypertension control, it is important to identify predictors of lifestyle modification assessment or advice to tailor future interventions using these effective, low-risk interventions. Electronic health record data from 14,360 adult hypertension patients at an academic medical center were analyzed using statistical and machine learning methods to identify predictors and timing of lifestyle modification. Multiple variables were statistically significant in analysis of lifestyle modification documentation at multiple time points. Random Forest was the best machine learning method to classify lifestyle modification documentation at any time with Area Under the Receiver Operator Curve (AUROC) 0.831. Logistic regression was the best machine learning method for classifying lifestyle modification documentation at ≤3 months with an AUROC of 0.685. Analyzing narrative and coded data from electronic health records can improve understanding of timing of lifestyle modification and patient, clinic and provider characteristics that are correlated with or predictive of documentation of lifestyle modification for hypertension. This information can inform improvement efforts in hypertension care processes, treatment implementation, and ultimately hypertension control.

Objectives To evaluate the utility of routinely collected Hendrich II fall scores in predicting returns to the emergency department (ED) for falls within 6 months. Design Retrospective electronic record review. Setting Academic medical center ED. Participants Individuals aged 65 and older seen in the ED from January 1, 2013, through September 30, 2015. Measurements We evaluated the utility of routinely collected Hendrich II fall risk scores in predicting ED visits for a fall within 6 months of an all‐cause index ED visit. Results For in‐network patient visits resulting in discharge with a completed Hendrich II score (N = 4,366), the return rate for a fall within 6 months was 8.3%. When applying the score alone to predict revisit for falls among the study population the resultant receiver operating characteristic (ROC) plot had an area under the curve (AUC) of 0.64. In a univariate model, the odds of returning to the ED for a fall in 6 months were 1.23 times as high for every 1‐point increase in Hendrich II score (odds ratio (OR)=1.23 (95% confidence interval (CI)=1.19–1.28). When included in a model with other potential confounders or predictors of falls, the Hendrich II score is a significant predictor of a return ED visit for fall (adjusted OR=1.15, 95% CI=1.10–1.20, AUC=0.75). Conclusion Routinely collected Hendrich II scores were correlated with outpatient falls, but it is likely that they would have little utility as a stand‐alone fall risk screen. When combined with easily extractable covariates, the screen performs much better. These results highlight the potential for secondary use of electronic health record data for risk stratification of individuals in the ED. Using data already routinely collected, individuals at high risk of falls after discharge could be identified for referral without requiring additional screening resources.

About 99,000 people are waiting for a kidney in the US, and many will die waiting. The concept of “imminent death” donation, a type of living donation, has been gaining attention among physicians, patients, and ethicists. We estimated the number of potential imminent death kidney donors at the University of Wisconsin Hospital and Clinics by estimating the number of annual deaths in individuals with normal kidney function. Based on a previous survey suggesting that 1/3 of patients might be willing to donate at imminent death, we estimate that between 76 and 396 people in the state of Wisconsin would be medically eligible and willing to donate each year at the time of imminent death. We extrapolated these numbers to all transplant centers in the US, estimating that between 5925 and 31,097 people might be eligible and willing to donate. Our results suggest that allowing donation at imminent death and including discussions about organ donation in end-of-life planning could substantially reduce the nation's kidney waiting list while providing many more donors the opportunity to give this gift. This article is protected by copyright. All rights reserved.

ObjectiveClinical research involving humans is critically important, but it is a lengthy and expensive process. Most studies require institutional review board (IRB) approval. Our objective is to identify predictors of delays or accelerations in the IRB review process and apply this knowledge to inform process change in an effort to improve IRB efficiency, transparency, consistency and communication.MethodsWe analyzed timelines of protocol submissions to determine protocol or IRB characteristics associated with different processing times. Our evaluation included single variable analysis to identify significant predictors of IRB processing time and machine learning methods to predict processing times through the IRB review system. Based on initial identified predictors, changes to IRB workflow and staffing procedures were instituted and we repeated our analysis.ResultsOur analysis identified several predictors of delays in the IRB review process including type of IRB review to be conducted, whether a protocol falls under Veteran’s Administration purview and specific staff in charge of a protocol's review.ConclusionsWe have identified several predictors of delays in IRB protocol review processing times using statistical and machine learning methods. Application of this knowledge to process improvement efforts in two IRBs has led to increased efficiency in protocol review. The workflow and system enhancements that are being made support our four-part goal of improving IRB efficiency, consistency, transparency, and communication.

The purpose of this article is to describe the association between initial CT for atraumatic headache and repeat emergency department (ED) visitation within 30 days of ED discharge.A retrospective observational study was performed at an academic urban ED with more than 85,000 annual visits. All adult patients with a chief complaint of headache from January through December 2010 who were discharged after ED evaluation were included in the analysis. Patients were excluded if they were transferred, died in the ED, or had a diagnosis indicating a traumatic mechanism. A propensity score-matched logistic regression model was used to determine whether the use of brain CT was associated with the primary outcome of ED revisitation within 30 days, controlling for potential confounding variables.Of 80,619 total patient visits to the ED during the study period, 922 ED discharges with a chief complaint of headache were included. A total of 139 (15.1%) patients revisited within 30 days. The return rate was 11.2% among patients who underwent CT at their initial visit and 21.1% among those who did not. In the adjusted analysis, controlling for age, race, sex, insurance status, triage vital signs, laboratory values, and triage pain level, the odds ratio for revisitation given CT performance was 0.49 (95% CI, 0.27-0.86).After adjustment for clinical factors, we found that patients who underwent a brain CT examination for atraumatic headache at an initial ED visit were less likely to return to the ED within 30 days. Future appropriate use quality metrics regarding ED imaging use may need to incorporate downstream health care use.

Background Falls among older adults are both a common reason for presentation to the emergency department, and a major source of morbidity and mortality. It is critical to identify fall patients quickly and reliably during, and immediately after, emergency department encounters in order to deliver appropriate care and referrals. Unfortunately, falls are difficult to identify without manual chart review, a time intensive process infeasible for many applications including surveillance and quality reporting. Here we describe a pragmatic NLP approach to automating fall identification. Methods In this single center retrospective review, 500 emergency department provider notes from older adult patients (age 65 and older) were randomly selected for analysis. A simple, rules-based NLP algorithm for fall identification was developed and evaluated on a development set of 1084 notes, then compared with identification by consensus of trained abstractors blinded to NLP results. Results The NLP pipeline demonstrated a recall (sensitivity) of 95.8%, specificity of 97.4%, precision of 92.0%, and F1 score of 0.939 for identifying fall events within emergency physician visit notes, as compared to gold standard manual abstraction by human coders. Conclusions Our pragmatic NLP algorithm was able to identify falls in ED notes with excellent precision and recall, comparable to that of more labor-intensive manual abstraction. This finding offers promise not just for improving research methods, but as a potential for identifying patients for targeted interventions, quality measure development and epidemiologic surveillance. Electronic supplementary material The online version of this article (10.1186/s12911-019-0843-7) contains supplementary material, which is available to authorized users.

Background The ECOG-ACRIN Cancer Research Group evaluated rituximab treatment schedules for patients with newly-diagnosed low-tumor-burden follicular-lymphoma (FL). All patients received 4-weekly rituximab treatments as induction therapy. Clinically-responding patients were randomized to receive rituximab every 13 weeks (“maintenance”) vs. no additional rituximab until progression (“non-maintenance”). Based on “time-to-rituximab-failure (TTRF)”, the study-committee reported there was no overall-benefit for maintenance rituximab in this setting. Tumor-reactive mAbs, like rituximab, trigger natural killer (NK) cells. NK-cell responses are regulated, in part, by interactions between killer immunoglobulin-like receptors (KIRs) on NK cells and their interactions with KIR-ligands. In a separate study of children with neuroblastoma treated with a different mAb, we found certain KIR/KIR-ligand genotypes associated with improved outcome. Here, we assessed whether a subset of FL patients show improved outcome from the maintenance rituximab based on these same KIR/KIR-ligand genotypes. Methods Genotypes for KIR/KIR-ligand were determined and assessed for associations with outcome [duration of response, TTRF and % tumor shrinkage] as a post-hoc analysis of this phase III trial. Our primary objective was to assess specific KIR/KIR-ligand genotype associations, followed by separate prespecified KIR/KIR-ligand genotype associations in follow-up analyses. Statistical analyses for association of genotype with clinical outcome included: Log-rank tests and Cox proportional hazards regression models to assess duration of response and TTRF; analysis of variance (ANOVA) was used for assessment of % tumor shrinkage. Results We found that patients inheriting KIR2DL2 and its ligand (HLA-C1) along with KIR3DL1 and its ligand (HLA-Bw4) had improved outcome over patients without this genotype. In addition, patients with KIR2DL2 and HLA-C1 along with KIR3DL1 and HLA-Bw4 also showed improved duration of response and tumor shrinkage if they received maintenance, while patients without this genotype showed no such improvement when receiving maintenance. Conclusions The data presented here indicate that a subset of FL patients, identified by certain KIRs/KIR-ligands, have improved outcome and may benefit from additional rituximab treatment. Taken together, this suggests that the efficacy of tumor-reactive mAb treatment for some patients is influenced by KIRs on NK cells. However, prior to considering these genotypes in a clinically-actionable manner, these findings need independent validation in other studies. Electronic supplementary material The online version of this article (10.1186/s40425-019-0538-8) contains supplementary material, which is available to authorized users.

Lifestyle modification, including diet, exercise, and tobacco cessation, is the first-line treatment of many disorders including hypertension, obesity, and diabetes. Lifestyle modification data are not easily retrieved or used in research due to their textual nature. This study addresses this knowledge gap using natural language processing to automatically identify lifestyle modification documentation from electronic health records. Electronic health record notes from hypertension patients were analyzed using an open-source natural language processing tool to retrieve assessment and advice regarding lifestyle modification. These data were classified as lifestyle modification assessment or advice and mapped to a coded standard ontology. Combined lifestyle modification (advice and assessment) recall was 99.27 percent, precision 94.44 percent, and correct classification 88.15 percent. Through extraction and transformation of narrative lifestyle modification data to coded data, this critical information can be used in research, metric development, and quality improvement efforts regarding care delivery for multiple medical conditions that benefit from lifestyle modification.

Objectives The Patient, Intervention, Control/Comparison, and Outcome (PICO) framework is an effective technique for framing a clinical question. We aim to develop the counterpart of PICO to structure clinical research data needs. Methods We use a data-driven approach to abstracting key concepts representing clinical research data needs by adapting and extending an expert-derived framework originally developed for defining cancer research data needs. We annotated clinical trial eligibility criteria, EHR data request logs, and data queries to electronic health records (EHR), to extract and harmonize concept classes representing clinical research data needs. We evaluated the class coverage, class preservation from the original framework, schema generalizability, schema understandability, and schema structural correctness through a semi-structured interview with eight multidisciplinary domain experts. We iteratively refined the schema based on the evaluations. Results Our data-driven schema preserved 68% of the 63 classes from the original framework and covered 88% (73/82) of the classes proposed by evaluators. Class coverage for participants of different backgrounds ranged from 60% to 100% with a median value of 95% agreement among the individual evaluators. The schema was found understandable and structurally sound. Conclusions Our proposed schema may serve as the counterpart to PICO for improving the research data needs communication between researchers and informaticians.

New vocabularies are rapidly evolving in the literature relative to the practice of clinical medicine and translational research. To provide integrated access to new terms, we developed a mobile and desktop online reference—Marshfield Dictionary of Clinical and Translational Science (MD-CTS). It is the first public resource that comprehensively integrates Wiktionary (word definition), BioPortal (ontology), Wiki (image reference), and Medline abstract (word usage) information. MD-CTS is accessible at http://spellchecker.mfldclin.edu/. The website provides a broadened capacity for the wider clinical and translational science community to keep pace with newly emerging scientific vocabulary. An initial evaluation using 63 randomly selected biomedical words suggests that online references generally provided better coverage (73%-95%) than paper-based dictionaries (57–71%).

Display Omitted We compared ILP to propositional machine learning approaches for EHR phenotyping.Training subject selection for machine learning was automated using ICD-9 codes.ILP out-performed propositional machine learning approaches in AUROC.Relational learning using ILP offers a viable approach to EHR-driven phenotyping. ObjectiveElectronic health records (EHR) offer medical and pharmacogenomics research unprecedented opportunities to identify and classify patients at risk. EHRs are collections of highly inter-dependent records that include biological, anatomical, physiological, and behavioral observations. They comprise a patient's clinical phenome, where each patient has thousands of date-stamped records distributed across many relational tables. Development of EHR computer-based phenotyping algorithms require time and medical insight from clinical experts, who most often can only review a small patient subset representative of the total EHR records, to identify phenotype features. In this research we evaluate whether relational machine learning (ML) using inductive logic programming (ILP) can contribute to addressing these issues as a viable approach for EHR-based phenotyping. MethodsTwo relational learning ILP approaches and three well-known WEKA (Waikato Environment for Knowledge Analysis) implementations of non-relational approaches (PART, J48, and JRIP) were used to develop models for nine phenotypes. International Classification of Diseases, Ninth Revision (ICD-9) coded EHR data were used to select training cohorts for the development of each phenotypic model. Accuracy, precision, recall, F-Measure, and Area Under the Receiver Operating Characteristic (AUROC) curve statistics were measured for each phenotypic model based on independent manually verified test cohorts. A two-sided binomial distribution test (sign test) compared the five ML approaches across phenotypes for statistical significance. ResultsWe developed an approach to automatically label training examples using ICD-9 diagnosis codes for the ML approaches being evaluated. Nine phenotypic models for each ML approach were evaluated, resulting in better overall model performance in AUROC using ILP when compared to PART (p=0.039), J48 (p=0.003) and JRIP (p=0.003). DiscussionILP has the potential to improve phenotyping by independently delivering clinically expert interpretable rules for phenotype definitions, or intuitive phenotypes to assist experts. ConclusionRelational learning using ILP offers a viable approach to EHR-driven phenotyping.

Purpose: In 2010, a Children's Oncology Group (COG) phase III randomized trial for patients with high-risk neuroblastoma (ANBL0032) demonstrated improved event-free survival (EFS) and overall survival (OS) following treatment with an immunotherapy regimen of dinutuximab, GM-CSF, IL2, and isotretinoin compared with treatment with isotretinoin alone. Dinutuximab, a chimeric anti-GD2 monoclonal antibody, acts in part via natural killer (NK) cells. Killer immunoglobulin-like receptors (KIR) on NK cells and their interactions with KIR-ligands can influence NK cell function. We investigated whether KIR/KIR-ligand genotypes were associated with EFS or OS in this trial. Experimental Design: We genotyped patients from COG study ANBL0032 and evaluated the effect of KIR/KIR-ligand genotypes on clinical outcomes. Cox regression models and log-rank tests were used to evaluate associations of EFS and OS with KIR/KIR-ligand genotypes. Results: In this trial, patients with the “all KIR-ligands present” genotype as well as patients with inhibitory KIR2DL2 with its ligand (HLA-C1) together with inhibitory KIR3DL1 with its ligand (HLA-Bw4) were associated with improved outcome if they received immunotherapy. In contrast, for patients with the complementary KIR/KIR-ligand genotypes, clinical outcome was not significantly different for patients who received immunotherapy versus those receiving isotretinoin alone. Conclusions: These data show that administration of immunotherapy is associated with improved outcome for neuroblastoma patients with certain KIR/KIR-ligand genotypes, although this was not seen for patients with other KIR/KIR-ligand genotypes. Further investigation of KIR/KIR-ligand genotypes may clarify their role in cancer immunotherapy and may enable KIR/KIR-ligand genotyping to be used prospectively for identifying patients likely to benefit from certain cancer immunotherapy regimens. Clin Cancer Res; 24(1); 189–96. ©2017 AACR. See related commentary by Cheung and Hsu, p. 3

Purpose: Fc-gamma receptors (FCGRs) are expressed on immune cells, bind to antibodies, and trigger antibody-induced cell-mediated antitumor responses when tumor-reactive antibodies are present. The affinity of the FCGR/antibody interaction is variable and dependent upon FCGR polymorphisms. Prior studies of patients with cancer treated with immunotherapy indicate that FCGR polymorphisms can influence antitumor response for certain immunotherapies that act via therapeutically administered mAbs or via endogenous tumor-reactive antibodies induced from tumor antigen vaccines. The previously published “SELECT” trial of high-dose aldesleukin (HD-IL2) for metastatic renal cell carcinoma resulted in an objective response rate of 25%. We evaluated the patients in this SELECT trial to determine whether higher-affinity FCGR polymorphisms are associated with outcome. Experimental Design: SNPs in FCGR2A, FCGR3A, and FCGR2C were analyzed, individually and in combination, for associations between genotype and clinical outcome. Results: When higher-affinity genotypes for FCGR2A, FCGR3A, and FCGR2C were considered together, they were associated with significantly increased tumor shrinkage and prolonged survival in response to HD-IL2. Conclusions: Although associations of higher-affinity FCGR genotype with clinical outcome have been demonstrated with mAb therapy and with idiotype vaccines, to our knowledge, this is the first study to show associations of FCGR genotypes with outcome following HD-IL2 treatment. We hypothesize that endogenous antitumor antibodies may engage immune cells through their FCGRs, and HD-IL2 may enhance antibody-induced tumor destruction, or antibody-enhanced tumor antigen presentation, via augmented activation of innate or adaptive immune responses; this FCGR-mediated immune activity would be augmented through immunologically favorable FCGRs. Clin Cancer Res; 23(9); 2159–68. ©2016 AACR.

NK cells play a role in many cancer immunotherapies. NK cell activity is tightly regulated by killer immunoglobulin-like receptor (KIR) and KIR–ligand interactions. Inhibitory KIR–ligands have been identified as HLA molecules, while activating KIR–ligands are largely unknown. Individuals that have not inherited the corresponding KIR–ligand for at least one inhibitory KIR gene are termed the “KIR–ligand missing” genotype, and they are thought to have a subset of NK cells that express inhibitory KIRs for which the corresponding KIR–ligand is missing on autologous tissue, and thus will not be inhibited through KIR–ligand recognition. In some settings where an anticancer immunotherapeutic effect is likely mediated by NK cells, individuals with a KIR–ligand missing genotype have shown improved clinical outcome compared to individuals with an “all KIR–ligands present” genotype. In addition, patients receiving hematopoietic stem cell transplants for leukemia may do better if their donor has more activating KIR genes (i.e., KIR haplotype-B). In a recent multi-institution clinical trial of patients with metastatic renal cell carcinoma receiving high-dose IL2 (HD-IL2), 25 % of patients showed a complete or partial tumor response to this therapy. We genotyped KIR and KIR–ligand genes for these patients (n = 107) and tested whether KIR/KIR–ligand genotypes correlated with patient clinical outcomes. In these analyses, we did not find any significant association of KIR/KIR–ligand genotype (either KIR–ligand missing or the presence of KIR haplotype-B) with patient outcome in response to the HD-IL2 therapy.

Objective: Clinicians often have difficulty translating information needs into effective search strategies to find appropriate answers. Information retrieval systems employing an intelligent search agent that generates adaptive search strategies based on human search expertise could be helpful in meeting clinician information needs. A prerequisite for creating such systems is an information seeking model that facilitates the representation of human search expertise. The purpose of developing such a model is to provide guidance to information seeking system development and to shape an empirical research program.Design: The information seeking process was modeled as a complex problem-solving activity. After considering how similarly complex activities had been modeled in other domains, we determined that modeling context-initiated information seeking across multiple problem spaces allows the abstraction of search knowledge into functionally consistent layers. The knowledge layers were identified in the information science literature and validated through our observations of searches performed by health science librarians.Results: A hierarchical multi-level model of context-initiated information seeking is proposed. Each level represents (1) a problem space that is traversed during the online search process, and (2) a distinct layer of knowledge that is required to execute a successful search. Grand strategy determines what information resources will be searched, for what purpose, and in what order. The strategy level represents an overall approach for searching a single resource. Tactics are individual moves made to further a strategy. Operations are mappings of abstract intentions to information resource-specific concrete input. Assessment is the basis of interaction within the strategic hierarchy, influencing the direction of the search.Conclusion: The described multi-level model provides a framework for future research and the foundation for development of an automated information retrieval system that uses an intelligent search agent to bridge clinician information needs and human search expertise.

e14014Background: High-risk neuroblastoma (NBL) patients (pts) enrolled in a COG phase III clinical trial (ANBL0032) were randomized to isotretinoin (RA) alone or RA + immunotherapy (IM): dinutuxim...

Computational models in biomedicine rely on biological and clinical assumptions. The selection of these assumptions contributes substantially to modeling success or failure. Assumptions used by experts at the cutting edge of research, however, are rarely explicitly described in scientific publications. One can directly collect and assess some of these assumptions through interviews and surveys. Here we investigate diversity in expert views about a complex biological phenomenon, the process of cancer metastasis. We harvested individual viewpoints from 28 experts in clinical and molecular aspects of cancer metastasis and summarized them computationally. While experts predominantly agreed on the definition of individual steps involved in metastasis, no two expert scenarios for metastasis were identical. We computed the probability that any two experts would disagree on k or fewer metastatic stages and found that any two randomly selected experts are likely to disagree about several assumptions. Considering the probability that two or more of these experts review an article or a proposal about metastatic cascades, the probability that they will disagree with elements of a proposed model approaches 1. This diversity of conceptions has clear consequences for advance and deadlock in the field. We suggest that strong, incompatible views are common in biomedicine but largely invisible to biomedical experts themselves. We built a formal Markov model of metastasis to encapsulate expert convergence and divergence regarding the entire sequence of metastatic stages. This model revealed stages of greatest disagreement, including the points at which cancer enters and leaves the bloodstream. The model provides a formal probabilistic hypothesis against which researchers can evaluate data on the process of metastasis. This would enable subsequent improvement of the model through Bayesian probabilistic update. Practically, we propose that model assumptions and hunches be harvested systematically and made available for modelers and scientists., Author Summary Mathematical models and scientific theories fail not only from internal inconsistency, but also from the poor selection of basic assumptions. Assumptions in computational models of biomedicine are typically provided by scientists who interact directly with empirical data. If we seek to model the dynamics of cancer metastasis and ask experts regarding valid assumptions, how widely will they agree and on which assumptions? To answer this question, we queried 28 faculty-level experts about the progression of metastasis. We demonstrate an unexpected diversity of assumptions across experts leading to a striking lack of agreement over the basic stages and sequence of metastasis. We suggest a formal model and framework that builds on this diversity and enables researchers to evaluate divergent hypotheses about metastasis with experimental data. We conclude that modeling biomedical processes could be substantially improved by harvesting scientific assumptions and exposing them for formalization and experiment.

The evolving complexity of genome-scale experiments has increasingly centralized the role of a highly computable, accurate, and comprehensive resource spanning multiple biological scales and viewpoints. To provide a resource to meet this need, we have significantly extended the PhenoGO database with gene-disease specific annotations and included an additional ten species. This a computationally-derived resource is primarily intended to provide phenotypic context (cell type, tissue, organ, and disease) for mining existing associations between gene products and GO terms specified in the Gene Ontology Databases Automated natural language processing (BioMedLEE) and computational ontology (PhenOS) methods were used to derive these relationships from the literature, expanding the database with information from ten additional species to include over 600,000 phenotypic contexts spanning eleven species from five GO annotation databases. A comprehensive evaluation evaluating the mappings (n = 300) found precision (positive predictive value) at 85%, and recall (sensitivity) at 76%. Phenotypes are encoded in general purpose ontologies such as Cell Ontology, the Unified Medical Language System, and in specialized ontologies such as the Mouse Anatomy and the Mammalian Phenotype Ontology. A web portal has also been developed, allowing for advanced filtering and querying of the database as well as download of the entire dataset http://www.phenogo.org.