Lara Lusa

Class prediction for high-dimensional class-imbalanced data

BackgroundThe goal of class prediction studies is to develop rules to accurately predict the class membership of new samples. The rules are derived using the values of the variables available for each subject: the main characteristic of high-dimensional data is that the number of variables greatly exceeds the number of samples. Frequently the classifiers are developed using class-imbalanced data, i.e., data sets where the number of samples in each class is not equal. Standard classification methods used on class-imbalanced data often produce classifiers that do not accurately predict the minority class; the prediction is biased towards the majority class. In this paper we investigate if the high-dimensionality poses additional challenges when dealing with class-imbalanced prediction. We evaluate the performance of six types of classifiers on class-imbalanced data, using simulated data and a publicly available data set from a breast cancer gene-expression microarray study. We also investigate the effectiveness of some strategies that are available to overcome the effect of class imbalance.ResultsOur results show that the evaluated classifiers are highly sensitive to class imbalance and that variable selection introduces an additional bias towards classification into the majority class. Most new samples are assigned to the majority class from the training set, unless the difference between the classes is very large. As a consequence, the class-specific predictive accuracies differ considerably. When the class imbalance is not too severe, down-sizing and asymmetric bagging embedding variable selection work well, while over-sampling does not. Variable normalization can further worsen the performance of the classifiers.ConclusionsOur results show that matching the prevalence of the classes in training and test set does not guarantee good performance of classifiers and that the problems related to classification with class-imbalanced data are exacerbated when dealing with high-dimensional data. Researchers using class-imbalanced data should be careful in assessing the predictive accuracy of the classifiers and, unless the class imbalance is mild, they should always use an appropriate method for dealing with the class imbalance problem.

Subcellular Localization of Activated Leukocyte Cell Adhesion Molecule Is a Molecular Predictor of Survival in Ovarian Carcinoma Patients

Purpose: Currently available clinicopathologic prognostic factors are imperfect predictors of clinical course in advanced-stage epithelial ovarian cancer patients. New molecular predictors are needed to identify patients with higher risk of relapse or death from disease. In a retrospective study, we investigated the prognostic impact of activated leukocyte cell adhesion molecule (ALCAM) expression in epithelial ovarian cancer. Experimental Design: We analyzed the effect of cell-anchorage loss on ALCAM cellular localization in vitro and assessed ALCAM expression by immunohistochemistry in a series of 109 well-characterized epithelial ovarian cancer patient samples. Chi-square test, Kaplan-Meier method, and Cox proportional hazard analyses were used to relate ALCAM cellular localization to clinical-pathologic parameters and to overall survival (OS) rate. Results: Loss of epithelial ovarian cancer cell anchorage was associated both in vitro and in vivo with decreased ALCAM membrane expression. In vivo, ALCAM was localized to cell membrane in normal surface ovarian epithelium, whereas in 67% of the epithelial ovarian cancer samples, membrane localization was decreased or even lost, and the molecule was mainly expressed in cytoplasm. Median OS in this group of patients was 58 months, whereas a median OS was not yet reached in patients with ALCAM membrane localization (P = 0.036, hazard ratio [HR] = 2.0, 95% confidence interval [CI] 1.1 to 3.5). In a multivariate Cox regression model including all the available clinicopathologic variables, loss of ALCAM membrane expression was an independent factor of unfavorable prognosis (P = 0.042, HR = 2.15, 95% CI: 1.0 to 4.5). Conclusions: Decreased/lost ALCAM membrane expression is a marker of poorer outcome in epithelial ovarian cancer patients and might help to identify patients who could benefit from more frequent follow-up or alternative therapeutic modalities.

SMOTE for high-dimensional class-imbalanced data

BackgroundClassification using class-imbalanced data is biased in favor of the majority class. The bias is even larger for high-dimensional data, where the number of variables greatly exceeds the number of samples. The problem can be attenuated by undersampling or oversampling, which produce class-balanced data. Generally undersampling is helpful, while random oversampling is not. Synthetic Minority Oversampling TEchnique (SMOTE) is a very popular oversampling method that was proposed to improve random oversampling but its behavior on high-dimensional data has not been thoroughly investigated. In this paper we investigate the properties of SMOTE from a theoretical and empirical point of view, using simulated and real high-dimensional data.ResultsWhile in most cases SMOTE seems beneficial with low-dimensional data, it does not attenuate the bias towards the classification in the majority class for most classifiers when data are high-dimensional, and it is less effective than random undersampling. SMOTE is beneficial for k-NN classifiers for high-dimensional data if the number of variables is reduced performing some type of variable selection; we explain why, otherwise, the k-NN classification is biased towards the minority class. Furthermore, we show that on high-dimensional data SMOTE does not change the class-specific mean values while it decreases the data variability and it introduces correlation between samples. We explain how our findings impact the class-prediction for high-dimensional data.ConclusionsIn practice, in the high-dimensional setting only k-NN classifiers based on the Euclidean distance seem to benefit substantially from the use of SMOTE, provided that variable selection is performed before using SMOTE; the benefit is larger if more neighbors are used. SMOTE for k-NN without variable selection should not be used, because it strongly biases the classification towards the minority class.

Improved Metabolic Control in Pediatric Patients with Type 1 Diabetes: A Nationwide Prospective 12-Year Time Trends Analysis

BACKGROUND This study estimated temporal trends of metabolic control over 12 years in a national cohort of childhood-onset type 1 diabetes. SUBJECTS AND METHODS Data from the prospective childhood-onset diabetes register, which included 886 case subjects from 0 to 17.99 years of age at diagnosis and at least 1 year of follow-up until the age of 22.99 years, were analyzed using multivariable linear and logistic regression models in the observational period between 2000 and 2011. RESULTS Hemoglobin A1c (HbA1c) significantly decreased over 12 years, from 78 mmol/mol (interquartile range [IQR], 68-88 mmol/mol) (9.26% [IQR, 8.41-10.24%]) in the year 2000 to 61 mmol/mol (IQR, 55-67 mmol/mol) (7.75% [IQR, 7.20-8.30%]) in the year 2011 (P<0.001). HbA1c was significantly associated with age, treatment modality, and duration of diabetes (P<0.001), with females having on average 1.02% higher HbA1c (P=0.01; 95% confidence interval [CI] 1.005-1.035). The overall use of insulin pumps was 74%. The incidence rate of severe acute complications was low: 1.07 per 100 patient-years for severe diabetic ketoacidosis (95% CI 0.81-1.40) and 1.21 per 100 patient-years for severe (requiring intravenous or intramuscular therapy) hypoglycemia (95% CI 0.81-1.40). CONCLUSIONS The metabolic control of the entire nationwide pediatric type 1 diabetes population significantly improved during the 12-year observational period with a low rate of severe acute complications events. The improvement was associated with the treatment modality. Additional efforts and solutions are necessary to further improve metabolic control and the quality of life of young people with type 1 diabetes.

Treatment of Erythema Migrans With Doxycycline for 10 Days Versus 15 Days

BACKGROUND The efficacy of 10-day doxycycline treatment in patients with erythema migrans has been assessed in the United States but not in Europe. Experts disagree on the significance of post-Lyme borreliosis symptoms. METHODS In a noninferiority trial, the efficacies of 10 days and 15 days of oral doxycycline therapy were evaluated in adult European patients with erythema migrans. The prevalence of nonspecific symptoms was compared between patients with erythema migrans and 81 control subjects without a history of Lyme borreliosis. The efficacy of treatment, determined on the basis of clinical observations and microbiologic tests, was assessed at 14 days and at 2, 6, and 12 months. Nonspecific symptoms in patients and controls were compared at 6 months after enrollment. RESULTS A total of 117 patients (52%) were treated with doxycycline for 15 days, and 108 (48%) received doxycycline for 10 days. Twelve months after enrollment, 85 of 91 patients (93.4%) in the 15-day group and 79 of 86 (91.9%) in the 10-day group had complete response (difference, 1.6 percentage points; upper limit of the 95% confidence interval, 9.1 percentage points). At 6 months, the frequency of nonspecific symptoms in the patients was similar to that among controls. CONCLUSIONS The 10-day regimen of oral doxycycline was not inferior to the 15-day regimen among adult European patients with solitary erythema migrans. Six months after treatment, the frequency of nonspecific symptoms among erythema migrans patients was similar to that among control subjects. CLINICAL TRIALS REGISTRATION NCT00910715.

miR-342 Regulates BRCA1 Expression through Modulation of ID4 in Breast Cancer

A miRNAs profiling on a group of familial and sporadic breast cancers showed that miRNA-342 was significantly associated with estrogen receptor (ER) levels. To investigate at functional level the role of miR-342 in the pathogenesis of breast cancer, we focused our attention on its “in silico” predicted putative target gene ID4, a transcription factor of the helix-loop-helix protein family whose expression is inversely correlated with that of ER. ID4 is expressed in breast cancer and can negatively regulate BRCA1 expression. Our results showed an inverse correlation between ID4 and miR-342 as well as between ID4 and BRCA1 expression. We functionally validated the interaction between ID4 and miR-342 in a reporter Luciferase system. Based on these findings, we hypothesized that regulation of ID4 mediated by miR-342 could be involved in the pathogenesis of breast cancer by downregulating BRCA1 expression. We functionally demonstrated the interactions between miR-342, ID4 and BRCA1 in a model provided by ER-negative MDA-MB-231 breast cancer cell line that presented high levels of ID4. Overexpression of miR-342 in these cells reduced ID4 and increased BRCA1 expression, supporting a possible role of this mechanism in breast cancer. In the ER-positive MCF7 and in the BRCA1-mutant HCC1937 cell lines miR-342 over-expression only reduced ID4. In the cohort of patients we studied, a correlation between miR-342 and BRCA1 expression was found in the ER-negative cases. As ER-negative cases were mainly BRCA1-mutant, we speculate that the mechanism we demonstrated could be involved in the decreased expression of BRCA1 frequently observed in non BRCA1-mutant breast cancers and could be implicated as a causal factor in part of the familial cases grouped in the heterogeneous class of non BRCA1 or BRCA2-mutant cases (BRCAx). To validate this hypothesis, the study should be extended to a larger cohort of ER-negative cases, including those belonging to the BRCAx class.

Quantitative detection of Borrelia burgdorferi sensu lato in erythema migrans skin lesions using internally controlled duplex real time PCR.

B. burgdorferi sensu stricto, B. afzelii, B. garinii and B. bavariensis are the principal species which account for Lyme borreliosis (LB) globally. We have developed an internally controlled duplex quantitative real time PCR assay targeting the Borrelia 16S rRNA and the human RNAseP genes. This assay is well-suited for laboratory confirmation of suspected cases of LB and will be used to assess the efficacy of a vaccine against LB in clinical trials. The assay is highly specific, successfully detecting DNA extracted from 83 diverse B. burgdorferi sensu lato strains representing all major species causing LB, while 21 unrelated microbial species and human genomic DNA tested negative. The assay was highly reproducible and sensitive, with a lower limit of detection of 6 copies per PCR reaction. Together with culture, the assay was used to evaluate paired 3 mm skin biopsy samples taken from 121 patients presenting with solitary erythema migrans (EM) lesion. PCR testing identified more positive biopsy samples than culture (77.7% PCR positive versus 55.1% culture positive) and correctly identified all specimens scored as culture positive. OspA-based typing identified the majority of isolates as B. afzelii (96.8%) and the bacterial load was significantly higher in culture positive biopsies than in culture negative biopsies (P<0.001). The quantitative data also enabled relationships between Borrelia burden and patient symptoms to be evaluated. The bacterial load was significantly higher among patients with systemic symptoms than without (P = 0.02) and was significantly higher for biopsies retrieved from patients with EM lesions with central clearing (P<0.001). 16S copy numbers were moderately lower in samples from patients reporting a history of LB (P = 0.10). This is the first quantitative PCR study of human skin biopsies predominantly infected with B. afzelii and the first study to demonstrate a clear relationship between clinical symptoms in B. afzelii-infected patients and Borrelia burden.

Coronavirus infections in hospitalized pediatric patients with acute respiratory tract disease

BackgroundAcute viral respiratory infections are an important cause of morbidity and mortality in humans worldwide. The etiological backgrounds of these infections remain unconfirmed in most clinical cases. The aim of this study was to estimate the prevalence of human coronavirus infections in a series of children hospitalized with symptoms of acute respiratory tract disease in a one-year period in Slovenia.MethodsThe 664 specimens from 592 children under six years of age hospitalized at the University Children’s Hospital in Ljubljana were sent for the routine laboratory detection of respiratory viruses. Respiratory viruses were detected with a direct immunofluorescence assay and human coronaviruses were detected with a modified real-time RT–PCR.ResultsHCoV RNA was detected in 40 (6%, 95% CI: 4.3%–8.1%) of 664 samples. Of these specimens, 21/40 (52.5%) were identified as species HKU1, 7/40 (17.5%) as OC43, 6/40 (15%) as 229E, and 6/40 (15%) as NL63. Infection with HCoV occurred as a coinfection with one or more other viruses in most samples (70%). Of the HCoV-positive children, 70.3% had lower respiratory tract infections.ConclusionThe results of our study show that HCoV are frequently detected human pathogens, often associated with other respiratory viruses and acute respiratory tract infections in hospitalized children. An association between age and the viral load was found. The highest viral load was detected in children approximately 10 months of age.

Suspected early Lyme neuroborreliosis in patients with erythema migrans.

BACKGROUND Our objective was to obtain data on patients with erythema migrans (EM) who have symptoms/signs suggesting nervous system involvement and to compare epidemiologic, clinical, and microbiologic findings in patients with and without cerebrospinal fluid (CSF) pleocytosis. METHODS Adult patients with EM and suspected early Lyme neuroborreliosis were included in this study. RESULTS Of 161 patients, 31 (19%) had elevated and 130 (81%) had normal CSF cell counts. In contrast to patients with normal CSF cell counts, those with pleocytosis (1) more often reported radicular pain and more often presented with meningeal signs but less frequently complained of malaise; (2) had larger EM skin lesions despite similar duration; (3) more commonly had Borrelia garinii isolated from EM skin lesions (odds ratio for pleocytosis was 31 times higher in patients with established B. garinii skin infection compared to patients with other Borrelia species isolated from their EM skin lesion) and from CSF; and (4) more frequently fulfilled microbiologic criteria for established borrelial infection of the central nervous system. The positive predictive value of pleocytosis for microbiologically proven borrelial infection of the central nervous system (defined by isolation of Borrelia from CSF and/or demonstration of intrathecal synthesis of borrelial antibodies) was 67.9%, whereas normal CSF white cell counts ruled out Lyme neuroborreliosis with a predictive value of 91.9%. CONCLUSIONS Comparison of European patients with EM who had symptoms/signs suggesting early Lyme neuroborreliosis revealed several differences in the clinical presentation and in microbiologic test results according to CSF findings.

Evaluation of SMOTE for High-Dimensional Class-Imbalanced Microarray Data

Synthetic Minority Oversampling TEchnique (SMOTE) is a popular oversampling method that was proposed to improve random oversampling but its behavior on high-dimensional data has not been thoroughly investigated. In this paper we evaluate the performance of SMOTE on high-dimensional data, using gene expression microarray data. We observe that SMOTE does not attenuate the bias towards the classification in the majority class for most classifiers, and it is less effective than random undersampling. SMOTE is beneficial for k-NN classifiers based on the Euclidean distance if the number of variables is reduced performing some type of variable selection and the benefit is larger if more neighbors are used. If the variable selection is not performed than the k-NN classification is counter intuitively biased towards the minority class, so SMOTE for k-NN without variable selection should not be used in practice.