John Tran

A Potential Causal Association Mining Algorithm for Screening Adverse Drug Reactions in Postmarketing Surveillance

Early detection of unknown adverse drug reactions (ADRs) in postmarketing surveillance saves lives and prevents harmful consequences. We propose a novel data mining approach to signaling potential ADRs from electronic health databases. More specifically, we introduce potential causal association rules (PCARs) to represent the potential causal relationship between a drug and ICD-9 (CDC. (2010). International Classification of Diseases, Ninth Revision (ICD-9). [Online]. Available: http://www.cdc.gov/nchs/icd/icd9.html) coded signs or symptoms representing potential ADRs. Due to the infrequent nature of ADRs, the existing frequency-based data mining methods cannot effectively discover PCARs. We introduce a new interestingness measure, potential causal leverage, to quantify the degree of association of a PCAR. This measure is based on the computational, experience-based fuzzy recognition-primed decision (RPD) model that we developed previously (Y. Ji, R. M. Massanari, J. Ager, J. Yen, R. E. Miller, and H. Ying, “A fuzzy logic-based computational recognition-primed decision model,” Inf. Sci., vol. 177, pp. 4338-4353, 2007) on the basis of the well-known, psychology-originated qualitative RPD model (G. A. Klein, “A recognition-primed decision making model of rapid decision making,” in Decision Making in Action: Models and Methods, 1993, pp. 138-147). The potential causal leverage assesses the strength of the association of a drug-symptom pair given a collection of patient cases. To test our data mining approach, we retrieved electronic medical data for 16 206 patients treated by one or more than eight drugs of our interest at the Veterans Affairs Medical Center in Detroit between 2007 and 2009. We selected enalapril as the target drug for this ADR signal generation study. We used our algorithm to preliminarily evaluate the associations between enalapril and all the ICD-9 codes associated with it. The experimental results indicate that our approach has a potential to better signal potential ADRs than risk ratio and leverage, two traditional frequency-based measures. Among the top 50 signal pairs (i.e., enalapril versus symptoms) ranked by the potential causal-leverage measure, the physicians on the project determined that eight of them probably represent true causal associations.

A Method for Mining Infrequent Causal Associations and Its Application in Finding Adverse Drug Reaction Signal Pairs

In many real-world applications, it is important to mine causal relationships where an event or event pattern causes certain outcomes with low probability. Discovering this kind of causal relationships can help us prevent or correct negative outcomes caused by their antecedents. In this paper, we propose an innovative data mining framework and apply it to mine potential causal associations in electronic patient data sets where the drug-related events of interest occur infrequently. Specifically, we created a novel interestingness measure, exclusive causal-leverage, based on a computational, fuzzy recognition-primed decision (RPD) model that we previously developed. On the basis of this new measure, a data mining algorithm was developed to mine the causal relationship between drugs and their associated adverse drug reactions (ADRs). The algorithm was tested on real patient data retrieved from the Veterans Affairs Medical Center in Detroit, Michigan. The retrieved data included 16,206 patients (15,605 male, 601 female). The exclusive causal-leverage was employed to rank the potential causal associations between each of the three selected drugs (i.e., enalapril, pravastatin, and rosuvastatin) and 3,954 recorded symptoms, each of which corresponded to a potential ADR. The top 10 drug-symptom pairs for each drug were evaluated by the physicians on our project team. The numbers of symptoms considered as likely real ADRs for enalapril, pravastatin, and rosuvastatin were 8, 7, and 6, respectively. These preliminary results indicate the usefulness of our method in finding potential ADR signal pairs for further analysis (e.g., epidemiology study) and investigation (e.g., case review) by drug safety professionals.

A fuzzy recognition-primed decision model-based causal association mining algorithm for detecting adverse drug reactions in postmarketing surveillance

The current approach to postmarketing surveillance primarily relies on spontaneous reporting. It is a passive surveillance system and limited by gross underreporting (<10% reporting rate), latency, and inconsistent reporting. We propose a new interestingness measure, causal-leverage, to signal potential adverse drug reactions (ADRs) from electronic health databases which are readily available in most modern hospitals. This measure is based on an experience-based fuzzy recognition-primed decision (RPD) model that we developed previously [1] which assesses the strength of association of a drug-ADR pair within each individual patient case. Using the causal-leverage measure, we develop a data mining algorithm to evaluate the associations between a given drug enalapril and all potential ADRs in a real-world electronic health database. The experimental results have shown that our approach can effectively shortlist some known ADRs. For example, the known ADR hyperkalemia caused by enalapril was ranked as top 1% among all the 3954 potential ADRs in our database.

Mining Infrequent Causal Associations in Electronic Health Databases

Discovering infrequent causal relationships can help us prevent or correct negative outcomes caused by their antecedents. In this paper, we propose an innovative data mining framework and apply it to mine potential causal associations in electronic patient datasets where the drug-related events of interest occur infrequently. Specifically, we created a novel interestingness measure, exclusive causal-leverage, based on a computational, fuzzy recognition-primed decision (RPD) model that we previously developed. On the basis of this new measure, a data mining algorithm was developed to mine the causal relationship between drugs and their associated adverse drug reactions (ADRs). The algorithm was tested on real patient data retrieved from the Veterans Affairs Medical Center in Detroit, Michigan. The exclusive causal-leverage was employed to rank the potential causal associations between each of the two selected drugs (i.e., enalapril and pravastatin) and 3,954 recorded symptoms, each of which corresponds to a potential ADR. The top 10 drug-symptom pairs for each drug were evaluated by our physicians on the project team. The results showed that the number of symptoms considered as real ADRs for enalapril and pravastatin was 8 and 7 out of 10, respectively.

Leveraging MapReduce to efficiently extract associations between biomedical concepts from large text data

Large biomedical text data represents an important source of information that not only enables researchers to discover in-depth knowledge about biological systems, but also helps healthcare professionals do evidence-based medicine in clinical settings. However, investigating and analyzing these data is often both data-intensive and computation-intensive. In this paper, we investigate how to use MapReduce, a parallel and distributed programming paradigm, to efficiently mine the associations between biomedical concepts extracted from a large set of biomedical articles. First, biomedical concepts were obtained by matching text to Unified Medical Language System (UMLS) Metathesaurus, a biomedical vocabulary and standard database. Then we developed a MapReduce algorithm that could be used to calculate a category of interestingness measures defined on the basis of a 22 contingency table. This algorithm consists of two MapReduce jobs and takes a stripes approach to reduce the number of intermediate results. Experiments were conducted using Amazon Elastic MapReduce (EMR) with an input of 33,960 articles from TREC (Text REtrieval Conference) 2006 Genomics Track. Performance test indicated that our algorithm had approximately linear scalability and was more efficient than a pairs approach in the literature. The physician in our project team evaluated a subset of the association mining results related to drug-disease treatment and found that meaningful association rules ranked high.

A functional temporal association mining approach for screening potential drug-drug interactions from electronic patient databases.

ABSTRACT Aims: Drug–drug interactions (DDIs) can result in serious consequences, including death. Existing methods for identifying potential DDIs in post-marketing surveillance primarily rely on spontaneous reports. These methods suffer from severe underreporting, incompleteness, and various bias. The aim of this study was to more effectively screen potential DDIs using patient electronic data and temporal association mining techniques. Methods: We focus on discovery of potential DDIs by analyzing the temporal relationships between the concurrent use of two drugs of interest and the occurrences of various symptoms. We introduced innovative functional temporal association rules where the degree of temporal association between two events within a patient case was defined by a function. Results: Preliminary test results on two drug pairs (i.e., and ) were classified into 260 clinically meaningful categories. These categories were evaluated by physicians and the results exhibited that all the potential DDIs were confined to top 20 of the 260 outcomes. Conclusions: Our methodology can be used to dramatically reduce a long list of association rules to a manageable list for further analysis and investigation by drug safety professionals.

Integrating unified medical language system and association mining techniques into relevance feedback for biomedical literature search

BackgroundFinding highly relevant articles from biomedical databases is challenging not only because it is often difficult to accurately express a user’s underlying intention through keywords but also because a keyword-based query normally returns a long list of hits with many citations being unwanted by the user. This paper proposes a novel biomedical literature search system, called BiomedSearch, which supports complex queries and relevance feedback.MethodsThe system employed association mining techniques to build a k-profile representing a user’s relevance feedback. More specifically, we developed a weighted interest measure and an association mining algorithm to find the strength of association between a query and each concept in the article(s) selected by the user as feedback. The top concepts were utilized to form a k-profile used for the next-round search. BiomedSearch relies on Unified Medical Language System (UMLS) knowledge sources to map text files to standard biomedical concepts. It was designed to support queries with any levels of complexity.ResultsA prototype of BiomedSearch software was made and it was preliminarily evaluated using the Genomics data from TREC (Text Retrieval Conference) 2006 Genomics Track. Initial experiment results indicated that BiomedSearch increased the mean average precision (MAP) for a set of queries.ConclusionsWith UMLS and association mining techniques, BiomedSearch can effectively utilize users’ relevance feedback to improve the performance of biomedical literature search.

High-Performance Biomedical Association Mining with MapReduce

MapReduce has been applied to data-intensive applications in different domains because of its simplicity, scalability and fault-tolerance. However, its uses in biomedical association mining are still very limited. In this paper, we investigate using MapReduce to efficiently mine the associations between biomedical terms extracted from a set of biomedical articles. First, biomedical terms were obtained by matching text to Unified Medical Language System (UMLS) Metathesaurus, a biomedical vocabulary and standard database. Then we developed a MapReduce algorithm that could be used to calculate a category of interestingness measures defined on the basis of a 2×2 contingency table. This algorithm consists of two MapReduce jobs and takes a stripes approach to reduce the number of intermediate results. Experiments were conducted using Amazon Elastic MapReduce (EMR) with an input of 3610 articles retrieved from two biomedical journals. Test results indicate that our algorithm has linear scalability.

An exclusive causal-leverage measure for detecting adverse drug reactions from electronic medical records

Early detection of causal relationships between drugs and their associated adverse drug reactions (ADRs) can prevent harmful consequences or even deaths. Rare ADRs cannot be detected by pre-marketing clinical trials due to limitations in their size and duration. Existing postmarketing surveillance methods mainly rely on spontaneous reporting which is limited by severe underreporting (<10 percentage reporting rate), latency and inconsistency. In this paper, we propose to identify potential ADRs from electronic medical records which are accessible now in many hospitals. Specifically, we created a new interestingness measure, exclusive causal-leverage, based on a computational, fuzzy recognition-primed decision (RPD) model[1]. This measure extends our previous measure, called causal-leverage, and can more effectively reduce the effects of background noises in the data. On the basis of this new measure, a data mining algorithm was developed and tested on real patient data retrieved from the Veterans Affairs Medical Center in Detroit, Michigan. The retrieved data included 16,206 patients (15,605 male, 601 female). Experimental results showed that two known ADRs (i.e. hyperpotassemia and cough) associated with drug enalapril were ranked as 3 and 21, respectively, among all the 3,954 potential ADRs (ICD-9 codes) in our database.

A Multi-relational Association Mining Algorithm for Screening Suspected Adverse Drug Reactions

Existing association mining algorithms generally assume that the data is in a single table (relation). One approach to mining multi-relational data tables is to convert the data into a single table and then apply the existing algorithms. However, the converted table may be too large to fit into memory. Moreover, these algorithms often need structures to store large intermediate data, which further restricts them by available memory. In this study, we developed an efficient SQL-based algorithm that directly dealt with multi-relational data tables that need less allocated memory. We also investigated how database indexes and the number of connections affect the performance of such an algorithm. The proposed algorithm was tested using data from the FDAs (Food and Drug Administration) spontaneous reporting system. The data collected was used for detecting potential adverse drug reactions (ADRs) which represent a serious worldwide problem. Our experiment results indicate that the algorithm performs well and is scalable in terms of the number of association rules that are evaluated and the size of the data.