Toward better public health reporting using existing off the shelf approaches: The value of medical dictionaries in automated cancer detection using plaintext medical data.

OBJECTIVES: Existing approaches to derive decision models from plaintext clinical data frequently depend on medical dictionaries as the sources of potential features. Prior research suggests that decision models developed using non-dictionary based feature sourcing approaches and "off the shelf" tools could predict cancer with performance metrics between 80% and 90%. We sought to compare non-dictionary based models to models built using features derived from medical dictionaries. MATERIALS AND METHODS: We evaluated the detection of cancer cases from free text pathology reports using decision models built with combinations of dictionary or non-dictionary based feature sourcing approaches, 4 feature subset sizes, and 5 classification algorithms. Each decision model was evaluated using the following performance metrics: sensitivity, specificity, accuracy, positive predictive value, and area under the receiver operating characteristics (ROC) curve. RESULTS: Decision models parameterized using dictionary and non-dictionary feature sourcing approaches produced performance metrics between 70 and 90%. The source of features and feature subset size had no impact on the performance of a decision model. CONCLUSION: Our study suggests there is little value in leveraging medical dictionaries for extracting features for decision model building. Decision models built using features extracted from the plaintext reports themselves achieve comparable results to those built using medical dictionaries. Overall, this suggests that existing "off the shelf" approaches can be leveraged to perform accurate cancer detection using less complex Named Entity Recognition (NER) based feature extraction, automated feature selection and modeling approaches.

Toward better public health reporting using existing off the shelf approaches: A comparison of alternative cancer detection approaches using plaintext medical data and non-dictionary based feature selection.

OBJECTIVES: Increased adoption of electronic health records has resulted in increased availability of free text clinical data for secondary use. A variety of approaches to obtain actionable information from unstructured free text data exist. These approaches are resource intensive, inherently complex and rely on structured clinical data and dictionary-based approaches. We sought to evaluate the potential to obtain actionable information from free text pathology reports using routinely available tools and approaches that do not depend on dictionary-based approaches. MATERIALS AND METHODS: We obtained pathology reports from a large health information exchange and evaluated the capacity to detect cancer cases from these reports using 3 non-dictionary feature selection approaches, 4 feature subset sizes, and 5 clinical decision models: simple logistic regression, naïve bayes, k-nearest neighbor, random forest, and J48 decision tree. The performance of each decision model was evaluated using sensitivity, specificity, accuracy, positive predictive value, and area under the receiver operating characteristics (ROC) curve. RESULTS: Decision models parameterized using automated, informed, and manual feature selection approaches yielded similar results. Furthermore, non-dictionary classification approaches identified cancer cases present in free text reports with evaluation measures approaching and exceeding 80-90% for most metrics. CONCLUSION: Our methods are feasible and practical approaches for extracting substantial information value from free text medical data, and the results suggest that these methods can perform on par, if not better, than existing dictionary-based approaches. Given that public health agencies are often under-resourced and lack the technical capacity for more complex methodologies, these results represent potentially significant value to the public health field.

DCMS: A data analytics and management system for molecular simulation.

Molecular Simulation (MS) is a powerful tool for studying physical/chemical features of large systems and has seen applications in many scientific and engineering domains. During the simulation process, the experiments generate a very large number of atoms and intend to observe their spatial and temporal relationships for scientific analysis. The sheer data volumes and their intensive interactions impose significant challenges for data accessing, managing, and analysis. To date, existing MS software systems fall short on storage and handling of MS data, mainly because of the missing of a platform to support applications that involve intensive data access and analytical process. In this paper, we present the database-centric molecular simulation (DCMS) system our team developed in the past few years. The main idea behind DCMS is to store MS data in a relational database management system (DBMS) to take advantage of the declarative query interface (i.e., SQL), data access methods, query processing, and optimization mechanisms of modern DBMSs. A unique challenge is to handle the analytical queries that are often compute-intensive. For that, we developed novel indexing and query processing strategies (including algorithms running on modern co-processors) as integrated components of the DBMS. As a result, researchers can upload and analyze their data using efficient functions implemented inside the DBMS. Index structures are generated to store analysis results that may be interesting to other users, so that the results are readily available without duplicating the analysis. We have developed a prototype of DCMS based on the PostgreSQL system and experiments using real MS data and workload show that DCMS significantly outperforms existing MS software systems. We also used it as a platform to test other data management issues such as security and compression.

A vision for the systematic monitoring and improvement of the quality of electronic health data.

In parallel with the implementation of information and communications systems, health care organizations are beginning to amass large-scale repositories of clinical and administrative data. Many nations seek to leverage so-called Big Data repositories to support improvements in health outcomes, drug safety, health surveillance, and care delivery processes. An unsupported assumption is that electronic health care data are of sufficient quality to enable the varied use cases envisioned by health ministries. The reality is that many electronic health data sources are of suboptimal quality and unfit for particular uses. To more systematically define, characterize and improve electronic health data quality, we propose a novel framework for health data stewardship. The framework is adapted from prior data quality research outside of health, but it has been reshaped to apply a systems approach to data quality with an emphasis on health outcomes. The proposed framework is a beginning, not an end. We invite the biomedical informatics community to use and adapt the framework to improve health data quality and outcomes for populations in nations around the world.

Hyper-structure mining of frequent patterns in uncertain data streams.

Data uncertainty is inherent in many real-world applications such as sensor monitoring systems, location-based services, and medical diagnostic systems. Moreover, many real-world applications are now capable of producing continuous, unbounded data streams. During the recent years, new methods have been developed to find frequent patterns in uncertain databases; nevertheless, very limited work has been done in discovering frequent patterns in uncertain data streams. The current solutions for frequent pattern mining in uncertain streams take a FP-tree-based approach; however, recent studies have shown that FP-tree-based algorithms do not perform well in the presence of data uncertainty. In this paper, we propose two hyper-structure-based false-positive-oriented algorithms to efficiently mine frequent itemsets from streams of uncertain data. The first algorithm, UHS-Stream, is designed to find all frequent itemsets up to the current moment. The second algorithm, TFUHS-Stream, is designed to find frequent itemsets in an uncertain data stream in a time-fading manner. Experimental results show that the proposed hyper-structure-based algorithms outperform the existing tree-based algorithms in terms of accuracy, runtime, and memory usage.

Performance analysis of a dual-tree algorithm for computing spatial distance histograms.

Many scientific and engineering fields produce large volume of spatiotemporal data. The storage, retrieval, and analysis of such data impose great challenges to database systems design. Analysis of scientific spatiotemporal data often involves computing functions of all point-to-point interactions. One such analytics, the Spatial Distance Histogram (SDH), is of vital importance to scientific discovery. Recently, algorithms for efficient SDH processing in large-scale scientific databases have been proposed. These algorithms adopt a recursive tree-traversing strategy to process point-to-point distances in the visited tree nodes in batches, thus require less time when compared to the brute-force approach where all pairwise distances have to be computed. Despite the promising experimental results, the complexity of such algorithms has not been thoroughly studied. In this paper, we present an analysis of such algorithms based on a geometric modeling approach. The main technique is to transform the analysis of point counts into a problem of quantifying the area of regions where pairwise distances can be processed in batches by the algorithm. From the analysis, we conclude that the number of pairwise distances that are left to be processed decreases exponentially with more levels of the tree visited. This leads to the proof of a time complexity lower than the quadratic time needed for a brute-force algorithm and builds the foundation for a constant-time approximate algorithm. Our model is also general in that it works for a wide range of point spatial distributions, histogram types, and space-partitioning options in building the tree.

Identification of a gene signature in cell cycle pathway for breast cancer prognosis using gene expression profiling data.

BACKGROUND: Numerous studies have used microarrays to identify gene signatures for predicting cancer patient clinical outcome and responses to chemotherapy. However, the potential impact of gene expression profiling in cancer diagnosis, prognosis and development of personalized treatment may not be fully exploited due to the lack of consensus gene signatures and poor understanding of the underlying molecular mechanisms. METHODS: We developed a novel approach to derive gene signatures for breast cancer prognosis in the context of known biological pathways. Using unsupervised methods, cancer patients were separated into distinct groups based on gene expression patterns in one of the following pathways: apoptosis, cell cycle, angiogenesis, metastasis, p53, DNA repair, and several receptor-mediated signaling pathways including chemokines, EGF, FGF, HIF, MAP kinase, JAK and NF-kappaB. The survival probabilities were then compared between the patient groups to determine if differential gene expression in a specific pathway is correlated with differential survival. RESULTS: Our results revealed expression of cell cycle genes is strongly predictive of breast cancer outcomes. We further confirmed this observation by building a cell cycle gene signature model using supervised methods. Validated in multiple independent datasets, the cell cycle gene signature is a more accurate predictor for breast cancer clinical outcome than the previously identified Amsterdam 70-gene signature that has been developed into a FDA approved clinical test MammaPrint. CONCLUSION: Taken together, the gene expression signature model we developed from well defined pathways is not only a consistently powerful prognosticator but also mechanistically linked to cancer biology. Our approach provides an alternative to the current methodology of identifying gene expression markers for cancer prognosis and drug responses using the whole genome gene expression data.

DGEM--a microarray gene expression database for primary human disease tissues.

Gene expression patterns can reflect gene regulations in human tissues under normal or pathologic conditions. Gene expression profiling data from studies of primary human disease samples are particularly valuable since these studies often span many years in order to collect patient clinical information and achieve a large sample size. Disease-to-Gene Expression Mapper (DGEM) provides a beneficial community resource to access and analyze these data; it currently includes Affymetrix oligonucleotide array datasets for more than 40 human diseases and 1400 samples. The data are normalized to the same scale and stored in a relational database. A statistical-analysis pipeline was implemented to identify genes abnormally expressed in disease tissues or genes whose expressions are associated with clinical parameters such as cancer patient survival. Data-mining results can be queried through a web-based interface at http://dgem.dhcp.iupui.edu/. The query tool enables dynamic generation of graphs and tables that are further linked to major gene and pathway resources that connect the data to relevant biology, including Entrez Gene and Kyoto Encyclopedia of Genes and Genomes (KEGG). In summary, DGEM provides scientists and physicians a valuable tool to study disease mechanisms, to discover potential disease biomarkers for diagnosis and prognosis, and to identify novel gene targets for drug discovery. The source code is freely available for non-profit use, on request to the authors.