Deep Learning-based Propensity Scores for Confounding Control in Comparative Effectiveness Research: A Large-scale, Real-world Data Study.

BACKGROUND: Due to the non-randomized nature of real-world data, prognostic factors need to be balanced, which is often done by propensity scores (PSs). This study aimed to investigate whether autoencoders, which are unsupervised deep learning architectures, might be leveraged to compute PS. METHODS: We selected patient-level data of 128,368 first-line treated cancer patients from the Flatiron Health EHR-derived de-identified database. We trained an autoencoder architecture to learn a lower-dimensional patient representation, which we used to compute PS. To compare the performance of an autoencoder-based PS with established methods, we performed a simulation study. We assessed the balancing and adjustment performance using standardized mean differences, root mean square errors (RMSE), percent bias, and confidence interval coverage. To illustrate the application of the autoencoder-based PS, we emulated the PRONOUNCE trial by applying the trial's protocol elements within an observational database setting, comparing two chemotherapy regimens. RESULTS: All methods but the manual variable selection approach led to well-balanced cohorts with average standardized mean differences <0.1. LASSO yielded on average the lowest deviation of resulting estimates (RMSE 0.0205) followed by the autoencoder approach (RMSE 0.0248). Altering the hyperparameter setup in sensitivity analysis, the autoencoder approach led to similar results as LASSO (RMSE 0.0203 and 0.0205, respectively). In the case study, all methods provided a similar conclusion with point estimates clustered around the null (e.g., HRautoencoder 1.01 [95% confidence interval = 0.80, 1.27] vs. HRPRONOUNCE 1.07 [0.83, 1.36]). CONCLUSIONS: Autoencoder-based PS computation was a feasible approach to control for confounding but did not perform better than some established approaches like LASSO.

Matching by OS Prognostic Score to Construct External Controls in Lung Cancer Clinical Trials.

External controls (eControls) leverage historical data to create non-randomized control arms. The lack of randomization can result in confounding between the experimental and eControl cohorts. To balance potentially confounding variables between the cohorts, one of the proposed methods is to match on prognostic scores. Still, the performance of prognostic scores to construct eControls in oncology has not been analyzed yet. Using an electronic health record-derived de-identified database, we constructed eControls using one of three methods: ROPRO, a state-of-the-art prognostic score, or either a propensity score composed of five (5Vars) or 27 covariates (ROPROvars). We compared the performance of these methods in estimating the overall survival (OS) hazard ratio (HR) of 11 recent advanced non-small cell lung cancer. The ROPRO eControls had a lower OS HR error (median absolute deviation (MAD), 0.072, confidence interval (CI): 0.036-0.185), than the 5Vars (MAD 0.081, CI: 0.025-0.283) and ROPROvars eControls (MAD 0.087, CI: 0.054-0.383). Notably, the OS HR errors for all methods were even lower in the phase III studies. Moreover, the ROPRO eControl cohorts included, on average, more patients than the 5Vars (6.54%) and ROPROvars cohorts (11.7%). The eControls matched with the prognostic score reproduced the controls more reliably than propensity scores composed of the underlying variables. Additionally, prognostic scores could allow eControls to be built on many prognostic variables without a significant increase in the variability of the propensity score, which would decrease the number of matched patients.

Automatic filtering and substantiation of drug safety signals.

Drug safety issues pose serious health threats to the population and constitute a major cause of mortality worldwide. Due to the prominent implications to both public health and the pharmaceutical industry, it is of great importance to unravel the molecular mechanisms by which an adverse drug reaction can be potentially elicited. These mechanisms can be investigated by placing the pharmaco-epidemiologically detected adverse drug reaction in an information-rich context and by exploiting all currently available biomedical knowledge to substantiate it. We present a computational framework for the biological annotation of potential adverse drug reactions. First, the proposed framework investigates previous evidences on the drug-event association in the context of biomedical literature (signal filtering). Then, it seeks to provide a biological explanation (signal substantiation) by exploring mechanistic connections that might explain why a drug produces a specific adverse reaction. The mechanistic connections include the activity of the drug, related compounds and drug metabolites on protein targets, the association of protein targets to clinical events, and the annotation of proteins (both protein targets and proteins associated with clinical events) to biological pathways. Hence, the workflows for signal filtering and substantiation integrate modules for literature and database mining, in silico drug-target profiling, and analyses based on gene-disease networks and biological pathways. Application examples of these workflows carried out on selected cases of drug safety signals are discussed. The methodology and workflows presented offer a novel approach to explore the molecular mechanisms underlying adverse drug reactions.

The EU-ADR Web Platform: delivering advanced pharmacovigilance tools.

PURPOSE: Pharmacovigilance methods have advanced greatly during the last decades, making post-market drug assessment an essential drug evaluation component. These methods mainly rely on the use of spontaneous reporting systems and health information databases to collect expertise from huge amounts of real-world reports. The EU-ADR Web Platform was built to further facilitate accessing, monitoring and exploring these data, enabling an in-depth analysis of adverse drug reactions risks. METHODS: The EU-ADR Web Platform exploits the wealth of data collected within a large-scale European initiative, the EU-ADR project. Millions of electronic health records, provided by national health agencies, are mined for specific drug events, which are correlated with literature, protein and pathway data, resulting in a rich drug-event dataset. Next, advanced distributed computing methods are tailored to coordinate the execution of data-mining and statistical analysis tasks. This permits obtaining a ranked drug-event list, removing spurious entries and highlighting relationships with high risk potential. RESULTS: The EU-ADR Web Platform is an open workspace for the integrated analysis of pharmacovigilance datasets. Using this software, researchers can access a variety of tools provided by distinct partners in a single centralized environment. Besides performing standalone drug-event assessments, they can also control the pipeline for an improved batch analysis of custom datasets. Drug-event pairs can be substantiated and statistically analysed within the platform's innovative working environment. CONCLUSIONS: A pioneering workspace that helps in explaining the biological path of adverse drug reactions was developed within the EU-ADR project consortium. This tool, targeted at the pharmacovigilance community, is available online at https://bioinformatics.ua.pt/euadr/.

Correlation Between Early Trends of a Prognostic Biomarker and Overall Survival in Non-Small-Cell Lung Cancer Clinical Trials.

PURPOSE: Overall survival (OS) is the primary end point in phase III oncology trials. Given low success rates, surrogate end points, such as progression-free survival or objective response rate, are used in early go/no-go decision making. Here, we investigate whether early trends of OS prognostic biomarkers, such as the ROPRO and DeepROPRO, can also be used for this purpose. METHODS: Using real-world data, we emulated a series of 12 advanced non-small-cell lung cancer (aNSCLC) clinical trials, originally conducted by six different sponsors and evaluated four different mechanisms, in a total of 19,920 individuals. We evaluated early trends (until 6 months) of the OS biomarker alongside early OS within the joint model (JM) framework. Study-level estimates of early OS and ROPRO trends were correlated against the actual final OS hazard ratios (HRs). RESULTS: We observed a strong correlation between the JM estimates and final OS HR at 3 months (adjusted R2 = 0.88) and at 6 months (adjusted R2 = 0.85). In the leave-one-out analysis, there was a low overall prediction error of the OS HR at both 3 months (root-mean-square error [RMSE] = 0.11) and 6 months (RMSE = 0.12). In addition, at 3 months, the absolute prediction error of the OS HR was lower than 0.05 for three trials. CONCLUSION: We describe a pipeline to predict trial OS HRs using emulated aNSCLC studies and their early OS and OS biomarker trends. The method has the potential to accelerate and improve decision making in drug development.

DisGeNET: a Cytoscape plugin to visualize, integrate, search and analyze gene-disease networks.

UNLABELLED: DisGeNET is a plugin for Cytoscape to query and analyze human gene-disease networks. DisGeNET allows user-friendly access to a new gene-disease database that we have developed by integrating data from several public sources. DisGeNET permits queries restricted to (i) the original data source, (ii) the association type, (iii) the disease class or (iv) specific gene(s)/disease(s). It represents gene-disease associations in terms of bipartite graphs and provides gene centric and disease centric views of the data. It assists the user in the interpretation and exploration of the genetic basis of human diseases by a variety of built-in functions. Moreover, DisGeNET permits multicolouring of nodes (genes/diseases) according to standard disease classification for expedient visualization. AVAILABILITY: DisGeNET is compatible with Cytoscape 2.6.3 and 2.7.0, please visit http://ibi.imim.es/DisGeNET/DisGeNETweb.html for installation guide, user tutorial and download.

Artificial Intelligence for Prognostic Scores in Oncology: a Benchmarking Study.

Introduction: Prognostic scores are important tools in oncology to facilitate clinical decision-making based on patient characteristics. To date, classic survival analysis using Cox proportional hazards regression has been employed in the development of these prognostic scores. With the advance of analytical models, this study aimed to determine if more complex machine-learning algorithms could outperform classical survival analysis methods. Methods: In this benchmarking study, two datasets were used to develop and compare different prognostic models for overall survival in pan-cancer populations: a nationwide EHR-derived de-identified database for training and in-sample testing and the OAK (phase III clinical trial) dataset for out-of-sample testing. A real-world database comprised 136K first-line treated cancer patients across multiple cancer types and was split into a 90% training and 10% testing dataset, respectively. The OAK dataset comprised 1,187 patients diagnosed with non-small cell lung cancer. To assess the effect of the covariate number on prognostic performance, we formed three feature sets with 27, 44 and 88 covariates. In terms of methods, we benchmarked ROPRO, a prognostic score based on the Cox model, against eight complex machine-learning models: regularized Cox, Random Survival Forests (RSF), Gradient Boosting (GB), DeepSurv (DS), Autoencoder (AE) and Super Learner (SL). The C-index was used as the performance metric to compare different models. Results: For in-sample testing on the real-world database the resulting C-index [95% CI] values for RSF 0.720 [0.716, 0.725], GB 0.722 [0.718, 0.727], DS 0.721 [0.717, 0.726] and lastly, SL 0.723 [0.718, 0.728] showed significantly better performance as compared to ROPRO 0.701 [0.696, 0.706]. Similar results were derived across all feature sets. However, for the out-of-sample validation on OAK, the stronger performance of the more complex models was not apparent anymore. Consistently, the increase in the number of prognostic covariates did not lead to an increase in model performance. Discussion: The stronger performance of the more complex models did not generalize when applied to an out-of-sample dataset. We hypothesize that future research may benefit by adding multimodal data to exploit advantages of more complex models.

Prior fluid and electrolyte imbalance is associated with COVID-19 mortality.

Background: The COVID-19 pandemic represents a major public health threat. Risk of death from the infection is associated with age and pre-existing comorbidities such as diabetes, dementia, cancer, and impairment of immunological, hepatic or renal function. It remains incompletely understood why some patients survive the disease, while others do not. As such, we sought to identify novel prognostic factors for COVID-19 mortality. Methods: We performed an unbiased, observational retrospective analysis of real world data. Our multivariable and univariable analyses make use of U.S. electronic health records from 122,250 COVID-19 patients in the early stages of the pandemic. Results: Here we show that a priori diagnoses of fluid, pH and electrolyte imbalance during the year preceding the infection are associated with an increased risk of death independently of age and prior renal comorbidities. Conclusions: We propose that future interventional studies should investigate whether the risk of death can be alleviated by diligent and personalized management of the fluid and electrolyte balance of at-risk individuals during and before COVID-19.

A novel multilevel statistical method for the study of the relationships between multireceptorial binding affinity profiles and in vivo endpoints.

The present work introduces a novel method for drug research based on the sequential building of linked multivariate statistical models, each one introducing a different level of drug description. The use of multivariate methods allows us to overcome the traditional one-target assumption and to link in vivo endpoints with drug binding profiles, involving multiple receptors. The method starts with a set of drugs, for which in vivo or clinical observations and binding affinities for potentially relevant receptors are known, and allows obtaining predictions of the in vivo endpoints highlighting the most influential receptors. Moreover, provided that the structure of the receptor binding sites is known (experimentally or by homology modeling), the proposed method also highlights receptor regions and ligand-receptor interactions that are more likely to be linked to the in vivo endpoints, which is information of high interest for the design of novel compounds. The method is illustrated by a practical application dealing with the study of the metabolic side effects of antipsychotic drugs. Herein, the method detects related receptors confirmed by experimental results. Moreover, the use of structural models of the receptor binding sites allows identifying regions and ligand-receptor interactions that are involved in the discrimination between antipsychotic drugs that show metabolic side effects and those that do not. The structural results suggest that the topology of a hydrophobic sandwich involving residues in transmembrane helices (TM) 3, 5, and 6, as well as the assembling of polar residues in TM5, are important discriminators between target/antitarget receptors. Ultimately, this will provide useful information for the design of safer compounds inducing fewer side effects.

Pathway databases and tools for their exploitation: benefits, current limitations and challenges.

In past years, comprehensive representations of cell signalling pathways have been developed by manual curation from literature, which requires huge effort and would benefit from information stored in databases and from automatic retrieval and integration methods. Once a reconstruction of the network of interactions is achieved, analysis of its structural features and its dynamic behaviour can take place. Mathematical modelling techniques are used to simulate the complex behaviour of cell signalling networks, which ultimately sheds light on the mechanisms leading to complex diseases or helps in the identification of drug targets. A variety of databases containing information on cell signalling pathways have been developed in conjunction with methodologies to access and analyse the data. In principle, the scenario is prepared to make the most of this information for the analysis of the dynamics of signalling pathways. However, are the knowledge repositories of signalling pathways ready to realize the systems biology promise? In this article we aim to initiate this discussion and to provide some insights on this issue.

From SNPs to pathways: integration of functional effect of sequence variations on models of cell signalling pathways.

BACKGROUND: Single nucleotide polymorphisms (SNPs) are the most frequent type of sequence variation between individuals, and represent a promising tool for finding genetic determinants of complex diseases and understanding the differences in drug response. In this regard, it is of particular interest to study the effect of non-synonymous SNPs in the context of biological networks such as cell signalling pathways. UniProt provides curated information about the functional and phenotypic effects of sequence variation, including SNPs, as well as on mutations of protein sequences. However, no strategy has been developed to integrate this information with biological networks, with the ultimate goal of studying the impact of the functional effect of SNPs in the structure and dynamics of biological networks. RESULTS: First, we identified the different challenges posed by the integration of the phenotypic effect of sequence variants and mutations with biological networks. Second, we developed a strategy for the combination of data extracted from public resources, such as UniProt, NCBI dbSNP, Reactome and BioModels. We generated attribute files containing phenotypic and genotypic annotations to the nodes of biological networks, which can be imported into network visualization tools such as Cytoscape. These resources allow the mapping and visualization of mutations and natural variations of human proteins and their phenotypic effect on biological networks (e.g. signalling pathways, protein-protein interaction networks, dynamic models). Finally, an example on the use of the sequence variation data in the dynamics of a network model is presented. CONCLUSION: In this paper we present a general strategy for the integration of pathway and sequence variation data for visualization, analysis and modelling purposes, including the study of the functional impact of protein sequence variations on the dynamics of signalling pathways. This is of particular interest when the SNP or mutation is known to be associated to disease. We expect that this approach will help in the study of the functional impact of disease-associated SNPs on the behaviour of cell signalling pathways, which ultimately will lead to a better understanding of the mechanisms underlying complex diseases.

Regulatory T-cell Genes Drive Altered Immune Microenvironment in Adult Solid Cancers and Allow for Immune Contextual Patient Subtyping.

Background: The tumor microenvironment is an important factor in cancer immunotherapy response. To further understand how a tumor affects the local immune system, we analyzed immune gene expression differences between matching normal and tumor tissue.Methods: We analyzed public and new gene expression data from solid cancers and isolated immune cell populations. We also determined the correlation between CD8, FoxP3 IHC, and our gene signatures.Results: We observed that regulatory T cells (Tregs) were one of the main drivers of immune gene expression differences between normal and tumor tissue. A tumor-specific CD8 signature was slightly lower in tumor tissue compared with normal of most (12 of 16) cancers, whereas a Treg signature was higher in tumor tissue of all cancers except liver. Clustering by Treg signature found two groups in colorectal cancer datasets. The high Treg cluster had more samples that were consensus molecular subtype 1/4, right-sided, and microsatellite-instable, compared with the low Treg cluster. Finally, we found that the correlation between signature and IHC was low in our small dataset, but samples in the high Treg cluster had significantly more CD8+ and FoxP3+ cells compared with the low Treg cluster.Conclusions: Treg gene expression is highly indicative of the overall tumor immune environment.Impact: In comparison with the consensus molecular subtype and microsatellite status, the Treg signature identifies more colorectal tumors with high immune activation that may benefit from cancer immunotherapy. Cancer Epidemiol Biomarkers Prev; 27(1); 103-12. ©2017 AACR.

Functional evaluation of out-of-the-box text-mining tools for data-mining tasks.

OBJECTIVE: The trade-off between the speed and simplicity of dictionary-based term recognition and the richer linguistic information provided by more advanced natural language processing (NLP) is an area of active discussion in clinical informatics. In this paper, we quantify this trade-off among text processing systems that make different trade-offs between speed and linguistic understanding. We tested both types of systems in three clinical research tasks: phase IV safety profiling of a drug, learning adverse drug-drug interactions, and learning used-to-treat relationships between drugs and indications. MATERIALS: We first benchmarked the accuracy of the NCBO Annotator and REVEAL in a manually annotated, publically available dataset from the 2008 i2b2 Obesity Challenge. We then applied the NCBO Annotator and REVEAL to 9 million clinical notes from the Stanford Translational Research Integrated Database Environment (STRIDE) and used the resulting data for three research tasks. RESULTS: There is no significant difference between using the NCBO Annotator and REVEAL in the results of the three research tasks when using large datasets. In one subtask, REVEAL achieved higher sensitivity with smaller datasets. CONCLUSIONS: For a variety of tasks, employing simple term recognition methods instead of advanced NLP methods results in little or no impact on accuracy when using large datasets. Simpler dictionary-based methods have the advantage of scaling well to very large datasets. Promoting the use of simple, dictionary-based methods for population level analyses can advance adoption of NLP in practice.

DisGeNET: a discovery platform for the dynamical exploration of human diseases and their genes.

DisGeNET is a comprehensive discovery platform designed to address a variety of questions concerning the genetic underpinning of human diseases. DisGeNET contains over 380,000 associations between >16,000 genes and 13,000 diseases, which makes it one of the largest repositories currently available of its kind. DisGeNET integrates expert-curated databases with text-mined data, covers information on Mendelian and complex diseases, and includes data from animal disease models. It features a score based on the supporting evidence to prioritize gene-disease associations. It is an open access resource available through a web interface, a Cytoscape plugin and as a Semantic Web resource. The web interface supports user-friendly data exploration and navigation. DisGeNET data can also be analysed via the DisGeNET Cytoscape plugin, and enriched with the annotations of other plugins of this popular network analysis software suite. Finally, the information contained in DisGeNET can be expanded and complemented using Semantic Web technologies and linked to a variety of resources already present in the Linked Data cloud. Hence, DisGeNET offers one of the most comprehensive collections of human gene-disease associations and a valuable set of tools for investigating the molecular mechanisms underlying diseases of genetic origin, designed to fulfill the needs of different user profiles, including bioinformaticians, biologists and health-care practitioners. Database URL: http://www.disgenet.org/

Proton Pump Inhibitor Usage and the Risk of Myocardial Infarction in the General Population.

BACKGROUND AND AIMS: Proton pump inhibitors (PPIs) have been associated with adverse clinical outcomes amongst clopidogrel users after an acute coronary syndrome. Recent pre-clinical results suggest that this risk might extend to subjects without any prior history of cardiovascular disease. We explore this potential risk in the general population via data-mining approaches. METHODS: Using a novel approach for mining clinical data for pharmacovigilance, we queried over 16 million clinical documents on 2.9 million individuals to examine whether PPI usage was associated with cardiovascular risk in the general population. RESULTS: In multiple data sources, we found gastroesophageal reflux disease (GERD) patients exposed to PPIs to have a 1.16 fold increased association (95% CI 1.09-1.24) with myocardial infarction (MI). Survival analysis in a prospective cohort found a two-fold (HR = 2.00; 95% CI 1.07-3.78; P = 0.031) increase in association with cardiovascular mortality. We found that this association exists regardless of clopidogrel use. We also found that H2 blockers, an alternate treatment for GERD, were not associated with increased cardiovascular risk; had they been in place, such pharmacovigilance algorithms could have flagged this risk as early as the year 2000. CONCLUSIONS: Consistent with our pre-clinical findings that PPIs may adversely impact vascular function, our data-mining study supports the association of PPI exposure with risk for MI in the general population. These data provide an example of how a combination of experimental studies and data-mining approaches can be applied to prioritize drug safety signals for further investigation.

Mining clinical text for signals of adverse drug-drug interactions.

BACKGROUND AND OBJECTIVE: Electronic health records (EHRs) are increasingly being used to complement the FDA Adverse Event Reporting System (FAERS) and to enable active pharmacovigilance. Over 30% of all adverse drug reactions are caused by drug-drug interactions (DDIs) and result in significant morbidity every year, making their early identification vital. We present an approach for identifying DDI signals directly from the textual portion of EHRs. METHODS: We recognize mentions of drug and event concepts from over 50 million clinical notes from two sites to create a timeline of concept mentions for each patient. We then use adjusted disproportionality ratios to identify significant drug-drug-event associations among 1165 drugs and 14 adverse events. To validate our results, we evaluate our performance on a gold standard of 1698 DDIs curated from existing knowledge bases, as well as with signaling DDI associations directly from FAERS using established methods. RESULTS: Our method achieves good performance, as measured by our gold standard (area under the receiver operator characteristic (ROC) curve >80%), on two independent EHR datasets and the performance is comparable to that of signaling DDIs from FAERS. We demonstrate the utility of our method for early detection of DDIs and for identifying alternatives for risky drug combinations. Finally, we publish a first of its kind database of population event rates among patients on drug combinations based on an EHR corpus. CONCLUSIONS: It is feasible to identify DDI signals and estimate the rate of adverse events among patients on drug combinations, directly from clinical text; this could have utility in prioritizing drug interaction surveillance as well as in clinical decision support.

Integration of genomic information with biological networks using Cytoscape.

Cytoscape is an open-source software for visualizing, analyzing, and modeling biological networks. This chapter explains how to use Cytoscape to analyze the functional effect of sequence variations in the context of biological networks such as protein-protein interaction networks and signaling pathways. The chapter is divided into five parts: (1) obtaining information about the functional effect of sequence variation in a Cytoscape readable format, (2) loading and displaying different types of biological networks in Cytoscape, (3) integrating the genomic information (SNPs and mutations) with the biological networks, and (4) analyzing the effect of the genomic perturbation onto the network structure using Cytoscape built-in functions. Finally, we briefly outline how the integrated data can help in building mathematical network models for analyzing the effect of the sequence variation onto the dynamics of the biological system. Each part is illustrated by step-by-step instructions on an example use case and visualized by many screenshots and figures.

Learning signals of adverse drug-drug interactions from the unstructured text of electronic health records.

Drug-drug interactions (DDI) account for 30% of all adverse drug reactions, which are the fourth leading cause of death in the US. Current methods for post marketing surveillance primarily use spontaneous reporting systems for learning DDI signals and validate their signals using the structured portions of Electronic Health Records (EHRs). We demonstrate a fast, annotation-based approach, which uses standard odds ratios for identifying signals of DDIs from the textual portion of EHRs directly and which, to our knowledge, is the first effort of its kind. We developed a gold standard of 1,120 DDIs spanning 14 adverse events and 1,164 drugs. Our evaluations on this gold standard using millions of clinical notes from the Stanford Hospital confirm that identifying DDI signals from clinical text is feasible (AUROC=81.5%). We conclude that the text in EHRs contain valuable information for learning DDI signals and has enormous utility in drug surveillance and clinical decision support.

Learning signals of adverse drug-drug interactions from the unstructured text of electronic health records.

Drug-drug interactions (DDI) account for 30% of all adverse drug reactions, which are the fourth leading cause of death in the US. Current methods for post marketing surveillance primarily use spontaneous reporting systems for learning DDI signals and validate their signals using the structured portions of Electronic Health Records (EHRs). We demonstrate a fast, annotation-based approach, which uses standard odds ratios for identifying signals of DDIs from the textual portion of EHRs directly and which, to our knowledge, is the first effort of its kind. We developed a gold standard of 1,120 DDIs spanning 14 adverse events and 1,164 drugs. Our evaluations on this gold standard using millions of clinical notes from the Stanford Hospital confirm that identifying DDI signals from clinical text is feasible (AUROC=81.5%). We conclude that the text in EHRs contain valuable information for learning DDI signals and has enormous utility in drug surveillance and clinical decision support.

Profiling risk factors for chronic uveitis in juvenile idiopathic arthritis: a new model for EHR-based research.

BACKGROUND: Juvenile idiopathic arthritis is the most common rheumatic disease in children. Chronic uveitis is a common and serious comorbid condition of juvenile idiopathic arthritis, with insidious presentation and potential to cause blindness. Knowledge of clinical associations will improve risk stratification. Based on clinical observation, we hypothesized that allergic conditions are associated with chronic uveitis in juvenile idiopathic arthritis patients. METHODS: This study is a retrospective cohort study using Stanford's clinical data warehouse containing data from Lucile Packard Children's Hospital from 2000-2011 to analyze patient characteristics associated with chronic uveitis in a large juvenile idiopathic arthritis cohort. Clinical notes in patients under 16 years of age were processed via a validated text analytics pipeline. Bivariate-associated variables were used in a multivariate logistic regression adjusted for age, gender, and race. Previously reported associations were evaluated to validate our methods. The main outcome measure was presence of terms indicating allergy or allergy medications use overrepresented in juvenile idiopathic arthritis patients with chronic uveitis. Residual text features were then used in unsupervised hierarchical clustering to compare clinical text similarity between patients with and without uveitis. RESULTS: Previously reported associations with uveitis in juvenile idiopathic arthritis patients (earlier age at arthritis diagnosis, oligoarticular-onset disease, antinuclear antibody status, history of psoriasis) were reproduced in our study. Use of allergy medications and terms describing allergic conditions were independently associated with chronic uveitis. The association with allergy drugs when adjusted for known associations remained significant (OR 2.54, 95% CI 1.22-5.4). CONCLUSIONS: This study shows the potential of using a validated text analytics pipeline on clinical data warehouses to examine practice-based evidence for evaluating hypotheses formed during patient care. Our study reproduces four known associations with uveitis development in juvenile idiopathic arthritis patients, and reports a new association between allergic conditions and chronic uveitis in juvenile idiopathic arthritis patients.