Recognition of medication information from discharge summaries using ensembles of classifiers

ABSTRACT: BACKGROUND: Extraction of clinical information such as medications or problems from clinical text is an important task of clinical natural language processing (NLP). Rule-based methods are often used in clinical NLP systems because they are easy to adapt and customize. Recently, supervised machine learning methods have proven to be effective in clinical NLP as well. However, combining different classifiers to further improve the performance of clinical entity recognition systems has not been investigated extensively. Combining classifiers into an ensemble classifier presents both challenges and opportunities to improve performance in such NLP tasks. METHODS: We investigated ensemble classifiers that used different voting strategies to combine outputs from three individual classifiers: a rule-based system, a support vector machine (SVM) based system, and a conditional random field (CRF) based system. Three voting methods were proposed and evaluated using the annotated data sets from the 2009 i2b2 NLP challenge: simple majority, local SVM-based voting, and local CRF-based voting. RESULTS: Evaluation on 268 manually annotated discharge summaries from the i2b2 challenge showed that the local CRF-based voting method achieved the best F-score of 90.84% (94.11% Precision, 87.81% Recall) for 10-fold cross-validation. We then compared our systems with the first-ranked system in the challenge by using the same training and test sets. Our system based on majority voting achieved a better F-score of 89.65% (93.91% Precision, 85.76% Recall) than the previously reported F-score of 89.19% (93.78% Precision, 85.03% Recall) by the first-ranked system in the challenge. CONCLUSIONS: Our experimental results using the 2009 i2b2 challenge datasets showed that ensemble classifiers that combine individual classifiers into a voting system could achieve better performance than a single classifier in recognizing medication information from clinical text. It suggests that simple strategies that can be easily implemented such as majority voting could have the potential to significantly improve clinical entity recognition.     

Zyflamend Reduces the Expression of Androgen Receptor in a Model of Castrate-Resistant Prostate Cancer

Prostate cancer is the most commonly diagnosed solid malignancy, and tumor cells eventually transform to castrate resistance through multiple pathways including activation of the androgen receptor via insulin-like growth factor receptor (IGF-1R) signaling involving phospho-AKT (pAKT). In this study, a mixture of herbal extracts, Zyflamend®, was used as a treatment in a model of castrate-resistant prostate cancer using CWR22Rv1 cells. Zyflamend reduced androgen receptor and IGF-1R expression along with a reduction of IGF-1-mediated proliferation of CWR22Rv1 cells. IGF-1 induced downstream AKT phosphorylation; however, the induction of pAKT was not associated with androgen receptor expression. Further, constitutively active form of AKT had no effect on nuclear expression of androgen receptor, indicating that upregulation of pAKT did not promote androgen receptor expression or nuclear translocation in castrate-resistant CWR22Rv1 cells. Conversely, Zyflamend reduced androgen receptor expression following IGF-1 stimulation and in cells overexpressing pAKT. These results demonstrated that Zyflamend inhibited IGF-1-stimulated cell growth, IGF-1R expression, and androgen receptor expression and its nuclear localization, but these effects were not dependent upon phosphatidylinositol 3-kinase/pAKT signaling. In conclusion, Zyflamend decreased cell proliferation and inhibited IGF-1R and androgen receptor expression in a phosphatidylinositol 3-kinase/pAKT independent manner.     

Heterogeneous phenotypes of platelet and plasma von Willebrand factor in obligatory heterozygotes for severe von Willebrand disease

To characterize the heterogeneity of severe (type III) von Willebrand disease (vWD), plasma and platelet von Willebrand factor antigen (vWF:Ag) and ristocetin cofactor activity (Ricof) were measured in 28 obligatory heterozygotes (ie, parents or children of probands from 15 different kindreds with severe vWD). On the average, heterozygotes had low levels of vWF in both platelets and plasma. There was, however, considerable heterogeneity, with four distinct patterns. Eleven heterozygotes had concordant reduction of vWF:Ag and Ricof in both plasma and platelets; five had low levels of vWF:Ag and Ricof in plasma contrasting with normal levels in platelets; eight had a peculiar pattern, the reverse of the above (ie, low levels in platelets and normal levels in plasma); and in one, both vWF measurements were normal in plasma and platelets. These patterns were genetically determined: they were consistent in four couples of consanguineous heterozygotes and in two couples carrying the same gene deletion. Only the remaining three heterozygotes had no clearly identifiable pattern. Other findings of this study were that although most of the heterozygotes had normal bleeding times, the 7 of 28 who had prolonged bleeding times had concordantly low levels of vWF measurements in both plasma and platelets. In conclusion, this large series of obligatory heterozygotes provides evidence for phenotypic and genotypic heterogeneity of severe vWD.     

Associations between the leptin receptor gene and adiposity in middle-aged Caucasian males from the HERITAGE family study

Linkage and association studies between three exonic polymorphisms in the leptin receptor gene and body composition variables in the HERITAGE Family Study were undertaken. Polymorphisms K109R, Q223R, and K656N have been analyzed with body mass index (BMI), sum of height skinfolds (SF8), fat mass (FM), percent body fat (%FAT), fat free mass, and plasma leptin level. Single-point linkage analysis and covariance analysis across genotypes were performed, by race, on phenotypes adjusted for age and sex. Blacks (88 parents; 231 adult offspring) from 115 nuclear families (72-119 sibpairs) and Caucasians (192 parents; 330 adult offspring) from 99 nuclear families (319-364 sibpairs) were used for these analyses. In Caucasians, BMI and FM showed suggestive linkages with K109R (P = 0.02 and P = 0.05, respectively) and associations with Q223R (P = 0.005 and P = 0.03, respectively). In blacks, no statistically significant linkage or association was observed. In Caucasians, associations with Q223R were observed in parents, but not in offspring, for BMI, FM, and %FAT (0.04< or =P< or =0.0001). Males, not females, showed differences across genotypes for the same phenotypes plus SF8 and leptin (0.03< or = P< or =0.0002). Carriers of the R223 allele showed higher values than noncarriers for BMI (+4 U, P = 0.0001), SF8 (+30 mm, P = 0.01), FM (+7 kg, P = 0.0004), %FAT (+5%, P = 0.0002), and leptin (+4 ng/mL, P = 0.0006). These results indicate a significant effect of leptin receptor on adiposity in middle-aged Caucasian males.     

Plasma human immunodeficiency virus (HIV) type 1 RNA load in men and women with advanced HIV-1 disease

Several studies of patients infected with human immunodeficiency virus (HIV) type 1 have suggested that women have lower plasma HIV-1 RNA levels than men, even when controlling for CD4 T cell levels. A cross-sectional analysis was performed in 494 patients (21% of whom were women) who enrolled in a prospective study of anemic HIV-1-infected patients requiring transfusion. The median CD4 T cell count and plasma HIV-1 RNA levels were 15 cells/microL and 4.83 log(10) copies/mL (67,350 copies/mL), respectively. In unadjusted analyses, women had slightly higher mean log HIV-1 RNA titers than men (0.19 log(10) higher copies/mL; 95% confidence interval, -0.05 to 0.44; P=.11). Adjustment for CD4 T cell count, race or ethnicity, injection drug use, and age yielded a smaller sex difference (0.13 log(10) copies/mL higher in women; P=.28). In this population of patients with very advanced HIV disease, there is no evidence that women have lower HIV-1 RNA levels than men.     

Mapping the lethal factor and edema factor binding sites on oligomeric anthrax protective antigen

Assembly of anthrax toxin complexes at the mammalian cell surface involves competitive binding of the edema factor (EF) and lethal factor (LF) to heptameric oligomers and lower order intermediates of PA(63), the activated carboxyl-terminal 63-kDa fragment of protective antigen (PA). We used sequence differences between PA(63) and homologous PA-like proteins to delineate a region within domain 1' of PA that may represent the binding site for these ligands. Substitution of alanine for any of seven residues in or near this region (R178, K197, R200, P205, I207, I210, and K214) strongly inhibited ligand binding. Selected mutations from this set were introduced into two oligomerization-deficient PA mutants, and the mutants were used in various combinations to map the single ligand site within dimeric PA(63). The site was found to span the interface between two adjacent subunits, explaining the dependence of ligand binding on PA oligomerization. The locations of residues comprising the site suggest that a single ligand molecule sterically occludes two adjacent sites, consistent with the finding that the PA(63) heptamer binds a maximum of three ligand molecules. These results elucidate the process by which the components of anthrax toxin, and perhaps other binary bacterial toxins, assemble into toxic complexes.     

A Snapshot of Lipid-Reporting Practices in Canadian Clinical Laboratories: An Urgent Need for Harmonisation

To effectively implement the Canadian Cardiovascular Society (CCS) guidelines for dyslipidemia management into clinical laboratories, clear recommendations for lipid reporting are essential. In this study, the Canadian Society of Clinical Chemists Working Group on Reference Interval Harmonisation surveyed Canadian laboratories on adult lipid reporting practices to set a foundation for the development and implementation of harmonised lipid reporting across Canada. Key aspects of the survey asked laboratories: what reporting parameters were in place to assess lipid results; what interpretative comments were provided; whether nonfasting lipids were permitted and, if so, what strategy was used to document fasting status; and whether there was interest in implementing a harmonised lipid report. A total of 101 laboratories were represented by 24 respondents, as many responses were submitted by laboratory networks that included more than 1 laboratory. There was at least 1 response from 9 Canadian provinces and representation across 5 testing platforms. Upper and lower limits for lipid parameters and referenced source of limits varied substantially across laboratories, with only 56% of laboratories (9 respondents) referencing the 2016 CCS guidelines. Eighty-six percent of laboratories (19 respondents) report nonfasting lipids, although the method of documenting nonfasting status varied. Overall, 36% of laboratories (8 respondents) reported interest in implementing a harmonised lipid report. Assessment of current lipid-reporting practices supports the need for harmonised lipid reporting across Canada. Development of a harmonised lipid report for the adult population, consistent with up-to-date Canadian guidelines, will improve continuity of lipid test interpretation across Canada and improve clinical decision making.