Multi-database mining.

Biomedical data useful for data mining are often distributed across multiple databases. These databases may be aggregated using several techniques to create single data sets that may be mined using standard approaches; however, separate databases may, in their design or data representation, capture information that is analytically useful and that is lost on integration. Recent techniques for mining multiple databases simultaneously but separately may preserve and leverage the unique perspectives within each database. This article presents an example, "dual mining," in which concurrent analysis of a target database with a related knowledge base can improve the identification of association patterns in the target most likely to be of interest for further analysis.

Air travel and venous thromboembolism: a systematic review.

CONTEXT: Despite multiple attempts to document and quantify the danger of venous thromboembolism (VTE) following prolonged travel, there is still uncertainty about the magnitude of risk and what can be done to lower it. OBJECTIVES: To review the methodologic strength of the literature, estimate the risk of travel-related VTE, evaluate the efficacy of preventive treatments, and develop evidence-based recommendations for practice. DATA SOURCES: Studies identified from MEDLINE from 1966 through December 2005, supplemented by a review of the Cochrane Central Registry of Controlled Trials, the Database of Abstracts of Reviews of Effects, and relevant bibliographies. STUDY SELECTION: We included all clinical studies that either reported primary data concerning travel as a risk factor for VTE or tested preventive measures for travel-related VTE. DATA EXTRACTION AND ANALYSIS: Two reviewers reviewed each study independently to assess inclusion criteria, classify research design, and rate methodologic features. The effect of methodologic differences, VTE risk, and travel duration on VTE rate was evaluated using a logistic regression model. DATA SYNTHESIS: Twenty-four published reports, totaling 25 studies, met inclusion criteria (6 case-control studies, 10 cohort studies, and 9 randomized controlled trials). Method of screening for VTE [screening ultrasound compared to usual clinical care, odds ratio (OR) 390], outcome measure [all VTE compared to pulmonary embolism (PE) only, OR 21], duration of travel (<6 hours compared to 6-8 hours, OR 0.011), and clinical risk ("higher" risk travelers compared to "lower," OR 3.6) were significantly related to VTE rate. Clinical VTE after prolonged travel is rare [27 PE per million flights diagnosed through usual clinical care, 0.05% symptomatic deep venous thrombosis (DVT) diagnosed through screening ultrasounds], but asymptomatic thrombi of uncertain clinical significance are more common. Graduated compression stockings prevented travel-related VTE (P < 0.05 in 4 of 6 studies), aspirin did not, and low-molecular-weight heparin (LMWH) showed a trend toward efficacy in one study. CONCLUSIONS: All travelers, regardless of VTE risk, should avoid dehydration and frequently exercise leg muscles. Travelers on a flight of less than 6 hours and those with no known risk factors for VTE, regardless of the duration of the flight, do not need DVT prophylaxis. Travelers with 1 or more risk factors for VTE should consider graduated compression stockings and/or LMWH for flights longer than 6 hours.

Relemed: sentence-level search engine with relevance score for the MEDLINE database of biomedical articles.

BACKGROUND: Receiving extraneous articles in response to a query submitted to MEDLINE/PubMed is common. When submitting a multi-word query (which is the majority of queries submitted), the presence of all query words within each article may be a necessary condition for retrieving relevant articles, but not sufficient. Ideally a relationship between the query words in the article is also required. We propose that if two words occur within an article, the probability that a relation between them is explained is higher when the words occur within adjacent sentences versus remote sentences. Therefore, sentence-level concurrence can be used as a surrogate for existence of the relationship between the words. In order to avoid the irrelevant articles, one solution would be to increase the search specificity. Another solution is to estimate a relevance score to sort the retrieved articles. However among the >30 retrieval services available for MEDLINE, only a few estimate a relevance score, and none detects and incorporates the relation between the query words as part of the relevance score. RESULTS: We have developed "Relemed", a search engine for MEDLINE. Relemed increases specificity and precision of retrieval by searching for query words within sentences rather than the whole article. It uses sentence-level concurrence as a statistical surrogate for the existence of relationship between the words. It also estimates a relevance score and sorts the results on this basis, thus shifting irrelevant articles lower down the list. In two case studies, we demonstrate that the most relevant articles appear at the top of the Relemed results, while this is not necessarily the case with a PubMed search. We have also shown that a Relemed search includes not only all the articles retrieved by PubMed, but potentially additional relevant articles, due to the extended 'automatic term mapping' and text-word searching features implemented in Relemed. CONCLUSION: By using sentence-level matching, Relemed can deliver higher specificity, thus eliminating more false-positive articles. By introducing an appropriate relevance metric, the most relevant articles on which the user wishes to focus are listed first. Relemed also shrinks the displayed text, and hence the time spent scanning the articles.

Locating previously unknown patterns in data-mining results: a dual data- and knowledge-mining method.

BACKGROUND: Data mining can be utilized to automate analysis of substantial amounts of data produced in many organizations. However, data mining produces large numbers of rules and patterns, many of which are not useful. Existing methods for pruning uninteresting patterns have only begun to automate the knowledge acquisition step (which is required for subjective measures of interestingness), hence leaving a serious bottleneck. In this paper we propose a method for automatically acquiring knowledge to shorten the pattern list by locating the novel and interesting ones. METHODS: The dual-mining method is based on automatically comparing the strength of patterns mined from a database with the strength of equivalent patterns mined from a relevant knowledgebase. When these two estimates of pattern strength do not match, a high "surprise score" is assigned to the pattern, identifying the pattern as potentially interesting. The surprise score captures the degree of novelty or interestingness of the mined pattern. In addition, we show how to compute p values for each surprise score, thus filtering out noise and attaching statistical significance. RESULTS: We have implemented the dual-mining method using scripts written in Perl and R. We applied the method to a large patient database and a biomedical literature citation knowledgebase. The system estimated association scores for 50,000 patterns, composed of disease entities and lab results, by querying the database and the knowledgebase. It then computed the surprise scores by comparing the pairs of association scores. Finally, the system estimated statistical significance of the scores. CONCLUSION: The dual-mining method eliminates more than 90% of patterns with strong associations, thus identifying them as uninteresting. We found that the pruning of patterns using the surprise score matched the biomedical evidence in the 100 cases that were examined by hand. The method automates the acquisition of knowledge, thus reducing dependence on the knowledge elicited from human expert, which is usually a rate-limiting step.

Data mining and clinical data repositories: Insights from a 667,000 patient data set.

Clinical repositories containing large amounts of biological, clinical, and administrative data are increasingly becoming available as health care systems integrate patient information for research and utilization objectives. To investigate the potential value of searching these databases for novel insights, we applied a new data mining approach, HealthMiner, to a large cohort of 667,000 inpatient and outpatient digital records from an academic medical system. HealthMiner approaches knowledge discovery using three unsupervised methods: CliniMiner, Predictive Analysis, and Pattern Discovery. The initial results from this study suggest that these approaches have the potential to expand research capabilities through identification of potentially novel clinical disease associations.

Incremental value of combined perfusion and function over perfusion alone by gated SPECT myocardial perfusion imaging for detection of severe three-vessel coronary artery disease.

OBJECTIVES: We hypothesized that combining functional assessment to perfusion enhances the ability of electrocardiographic gating Tc-99m sestamibi single photon emission computed tomography (gated SPECT) myocardial perfusion imaging (MPI) to detect defects in multiple vascular territories in patients with severe three-vessel coronary artery disease (3VD). BACKGROUND: In patients with 3VD, perfusion defects in multiple vascular territories may not always be evident due to globally reduced perfusion. METHODS: Gated SPECT MPIs were interpreted sequentially with perfusion first, followed by combined perfusion/function, in 143 patients with angiographic 3VD and a control group of 112 non-3VD patients. All patients underwent coronary arteriography within one month of MPI. RESULTS: In 3VD patients, combined perfusion/function analysis yielded significantly greater numbers of abnormal segments/patient (6.2 +/- 4.7 vs. 4.1 +/- 2.8, p < 0.001) and more defects in multiple vascular territories (60% vs. 46%, p < 0.05) than perfusion alone. In the control group, there were no differences between the combined perfusion/function and perfusion alone interpretations. Multivariate analysis of 15 different clinical, stress, and scintigraphic variables in all patients revealed age (p < 0.0001) and number of abnormal vascular territories by combined perfusion/function (p < 0.0001) to be the most powerful predictors of 3VD. Addition of functional data to clinical, stress, and perfusion yielded a significant increase in the predictive value of 3VD (global chi-square: 131.7 vs. 89.8, p < 0.00001). Specificity of combined perfusion/function analysis was not lower than perfusion alone (72% vs. 69%, p = NS). CONCLUSIONS; Adjunctive assessment of function with perfusion by gated SPECT MPI enhances the detection of defects in multiple vascular territories in patients with severe 3VD, without adversely affecting its specificity.

GeneX Va: VBC open source microarray database and analysis software.

Developed by the Virginia Bioinformatics Consortium (VBC), GeneX Va is an open source, freeware database and bioinformatics analysis software for archiving and analyzing Affymetrix GeneChip data. It provides an integrated framework for management, documentation, and analysis of microarray experiments and data to support a range of users, from individual research laboratories to institutional microarray facilities. GeneX Va also provides web-based access to a PostgreSQL relational database system with a comprehensive security system. Data can be extracted from the database and delivered to interactive or scriptable statistical analysis protocols. The security system allows each investigator to manage their own array data and analysis output files and also provides custom access privileges for other users, groups, and internal/external collaborators. The analysis interface uses "Analysis Trees," an innovative user interface that allows researchers to interactively create a tree-structured flow chart of analysis routines. The latest GeneX Va software is available from and can be freely downloaded at the Sourceforge web site http://va-genex.sourceforge.net. To allow researchers to access the database and analysis capabilities of the GeneX Va system, microarray data from many VBC GeneChip experiments have been deposited into a public section of the GeneX Va system at the University of Virginia. The VBC GeneX Va sites, which include documentation, are at http://genes.med.virginia.edu/ of the University of Virginia and at http://genex.csbc.vcu.edu/ of the Virginia Commonwealth University.

The effect of physician feedback and an action checklist on diabetes care measures.

The objective was to evaluate whether physician feedback accompanied by an action checklist improved diabetes care process measures. Eighty-three physicians in an academic general medicine clinic were provided a single feedback report on the most recent date and result of diabetes care measures (glycosylated hemoglobin [A1c], urine microalbumin, serum creatinine, lipid levels, retinal examination) as well as recent diabetes medication refills with calculated dosing and adherence on 789 patients. An educational session regarding the feedback and adherence information was provided. The physicians were asked to complete a checklist accompanying the feedback on each of their patients, indicating requested actions with respect to follow-up, testing, and counseling. The physicians completed 82% of patient checklists, requesting actions consistent with patient needs on the basis of the feedback. Of the physicians, 93% felt the patient information and intervention format to be useful. The odds of urine microalbumin testing, serum creatinine, lipid profile, A1c, and retinal examination increased in the 6 months after the feedback. The increase was sustained at 1 year only for microalbumin and retinal exams. There was no significant change in refill adherence for the group overall after the feedback, although adherence did improve among patients of physicians attending the educational session. No significant change was noted in lipid or A1c levels during the study period. In conclusion, a simple physician feedback tool with action checklist can be both helpful and popular for improving rates of diabetes care guideline adherence. More complex interventions are likely required to improve diabetes outcomes.

Pacifiers and breastfeeding: a systematic review.

OBJECTIVE: To summarize current evidence on the association between infant pacifier use and breastfeeding. DATA SOURCES: MEDLINE, CINAHL, the Cochrane Library, EMBASE, POPLINE, and bibliographies of identified articles. STUDY SELECTION: A search for English-language records (from January 1950 through August 2006) containing the Medical Subject Heading terms pacifiers and breastfeeding was conducted, resulting in 1098 reports. Duplicate and irrelevant studies were excluded, yielding 29 studies that fit inclusion criteria for the review (4 randomized controlled trials, 20 cohort studies, and 5 cross-sectional studies). Two independent reviewers abstracted data and scored these studies for quality; disagreements were settled through consensus with a third investigator. MAIN EXPOSURE: Pacifier use. MAIN OUTCOME MEASURES: Breastfeeding duration or exclusivity. RESULTS: Results from 4 randomized controlled trials revealed no difference in breastfeeding outcomes with different pacifier interventions (pacifier use during tube feeds, pacifier use at any time after delivery, an educational program for mothers emphasizing avoidance of pacifiers, and a UNICEF [United Nations Children's Fund]/World Health Organization Baby Friendly Hospital environment). Most observational studies reported an association between pacifier use and shortened duration of breastfeeding. CONCLUSIONS: The highest level of evidence does not support an adverse relationship between pacifier use and breastfeeding duration or exclusivity. The association between shortened duration of breastfeeding and pacifier use in observational studies likely reflects a number of other complex factors, such as breastfeeding difficulties or intent to wean. Ongoing quantitative and qualitative research is needed to better understand the relationship between pacifier use and breastfeeding.

Impact of lifestyle intervention on lost productivity and disability: improving control with activity and nutrition.

OBJECTIVE: To evaluate the effectiveness of a lifestyle intervention (LI) in reducing work loss and disability days. METHODS: One year randomized controlled trial of health plan members (n = 147) with type 2 diabetes and obesity. Members were randomized to modest-cost LI or usual care (UC). Outcomes were group differences in cumulative days either missed at work or with disability using Mann-Whitney U-tests and Poisson regression models. RESULTS: LI reduced the risk of workdays lost by 64.3% (P

Papaverine prolongs patency of peripheral arterial catheters in neonates.

OBJECTIVE: To test the hypothesis that the continuous infusion of papaverine-containing solutions in peripheral arterial catheters would decrease the catheter failure rate and increase the functional duration of the catheter in neonates. STUDY DESIGN: In a prospective, randomized, placebo-controlled, masked trial, 82 catheters were placed in 70 neonates in the papaverine group and 98 catheters were placed in 71 neonates in the placebo group. RESULTS: The catheters in the papaverine group remained functional for a significantly longer duration than those in the placebo group. The median (25th%, 75th%) time before catheter failure was 16.6 (9.5, 24.3) days in the papaverine group and 12 days (6.1, 18.2) in the placebo group ( P = .023; Cox proportional hazards model). There was no significant difference in the incidence of intraventricular hemorrhage (IVH) between groups, and there was no evidence of hepatic toxicity. CONCLUSIONS: The continuous infusion of papaverine-containing fluids prolongs the patency of peripheral arterial catheters in neonates. In this small number of infants, we found no difference in the incidence of IVH or hepatic toxicity.

Do pacifiers reduce the risk of sudden infant death syndrome? A meta-analysis.

OBJECTIVE: Pacifier use has been reported to be associated with a reduced risk of sudden infant death syndrome (SIDS), but most countries around the world, including the United States, have been reluctant to recommend the use of pacifiers because of concerns about possible adverse effects. This meta-analysis was undertaken to quantify and evaluate the protective effect of pacifiers against SIDS and to make a recommendation on the use of pacifiers to prevent SIDS. METHODS: We searched the Medline database (January 1966 to May 2004) to collect data on pacifier use and its association with SIDS, morbidity, or other adverse effects. The search strategy included published articles in English with the Medical Subject Headings terms "sudden infant death syndrome" and "pacifier" and the keywords "dummy" and "soother." Combining searches resulted in 384 abstracts, which were all read and evaluated for inclusion. For the meta-analysis, articles with data on the relationship between pacifier use and SIDS risk were limited to published original case-control studies, because no prospective observational reports were found; 9 articles met these criteria. Two independent reviewers evaluated each study on the basis of the 6 criteria developed by the American Academy of Pediatrics Task Force on Infant Positioning and SIDS; in cases of disagreement, a third reviewer evaluated the study, and a consensus opinion was reached. We developed a script to calculate the summary odds ratio (SOR) by using the reported ORs and respective confidence intervals (CI) to weight the ORs. We then pooled them together to compute the SOR. We performed the Breslow-Day test for homogeneity of ORs, Cochran-Mantel-Haenszel test for the null hypothesis of no effect (OR = 1), and the Mantel-Haenszel common OR estimate. The consistency of findings was evaluated and the overall potential benefits of pacifier use were weighed against the potential risks. Our recommendation is based on the taxonomy of the 5-point (A-E) scale adopted by the US Preventive Services Task Force. RESULTS: Seven studies were included in the meta-analysis. The SOR calculated for usual pacifier use (with univariate ORs) is 0.90 (95% confidence interval [CI]: 0.79-1.03) and 0.71 (95% CI: 0.59-0.85) with multivariate ORs. For pacifier use during last sleep, the SORs calculated using univariate and multivariate ORs are 0.47 (95% CI: 0.40-0.55) and 0.39 (95% CI: 0.31-0.50), respectively. CONCLUSIONS: Published case-control studies demonstrate a significant reduced risk of SIDS with pacifier use, particularly when placed for sleep. Encouraging pacifier use is likely to be beneficial on a population-wide basis: 1 SIDS death could be prevented for every 2733 (95% CI: 2416-3334) infants who use a pacifier when placed for sleep (number needed to treat), based on the US SIDS rate and the last-sleep multivariate SOR resulting from this analysis. Therefore, we recommend that pacifiers be offered to infants as a potential method to reduce the risk of SIDS. The pacifier should be offered to the infant when being placed for all sleep episodes, including daytime naps and nighttime sleeps. This is a US Preventive Services Task Force level B strength of recommendation based on the consistency of findings and the likelihood that the beneficial effects will outweigh any potential negative effects. In consideration of potential adverse effects, we recommend pacifier use for infants up to 1 year of age, which includes the peak ages for SIDS risk and the period in which the infant's need for sucking is highest. For breastfed infants, pacifiers should be introduced after breastfeeding has been well established.

D-dimer testing for deep venous thrombosis: a metaanalysis.

BACKGROUND: The use of D-dimer assays as a rule-out test for deep venous thrombosis (DVT) is controversial. To clarify this issue we performed a systematic review of the relevant literature. METHODS: We identified eligible studies, using MEDLINE entries from February 1995 through October 2003, supplemented by a review of bibliographies of relevant articles. Studies reporting accuracy evaluations comparing D-dimer test results with lower extremity ultrasound or venography in symptomatic patients with suspected acute DVT were selected for review. Two reviewers critically appraised each study independently according to previously established methodologic standards for diagnostic test research. Those studies judged to be of highest quality were designated Level 1. RESULTS: The 23 Level 1 studies reported data on 21 different D-dimer assays. There was wide variation in assay sensitivity, specificity, and negative predictive values, and major differences in methodology of reviewed studies. A multivariate analysis of assay performance, controlling for sample size, DVT prevalence, reference standard, and patient mix, found few differences among the assays in effect on test performance as measured by diagnostic odds ratio. Increasing prevalence of DVT was associated with poorer test performance (P = 0.01), whereas the choice of venography as the reference standard was associated with better test performance (P <0.005). CONCLUSIONS: Explanations for the wide variation in assay performance include differences in biochemical and technical characteristics of the assays, heterogeneity and small size of patient groups, and bias introduced by choice of reference standards. Assay sensitivity and negative predictive value were frequently <90%, uncharacteristic of a good rule-out test. General use of D-dimer assays as a stand-alone test for the diagnosis of DVT is not supported by the literature.