In vivo alkaline comet assay: Statistical considerations on historical negative and positive control data.

The alkaline comet assay is frequently used as in vivo follow-up test within different regulatory environments to characterize the DNA-damaging potential of different test items. The corresponding OECD Test guideline 489 highlights the importance of statistical analyses and historical control data (HCD) but does not provide detailed procedures. Therefore, the working group "Statistics" of the German-speaking Society for Environmental Mutation Research (GUM) collected HCD from five laboratories and >200 comet assay studies and performed several statistical analyses. Key results included that (I) observed large inter-laboratory effects argue against the use of absolute quality thresholds, (II) > 50% zero values on a slide are considered problematic, due to their influence on slide or animal summary statistics, (III) the type of summarizing measure for single-cell data (e.g., median, arithmetic and geometric mean) may lead to extreme differences in resulting animal tail intensities and study outcome in the HCD. These summarizing values increase the reliability of analysis results by better meeting statistical model assumptions, but at the cost of information loss. Furthermore, the relation between negative and positive control groups in the data set was always satisfactorily (or sufficiently) based on ratio, difference and quantile analyses.

Rapid detection and online analysis of microbial changes through flow cytometry.

Short-read 16 S rRNA gene sequencing is the dominating technology to profile microbial communities in different habitats. Its uncontested taxonomic resolution paved the way for major contributions to the field. Sample measurement and analysis, that is, sequencing, is rather slow-in order of days. Alternatively, flow cytometry can be used to profile the microbiota of various sources within a few minutes per sample. To keep up with high measurement speed, we developed the open source-analyzing tool FlowSoFine. To validate the ability to distinguish microbial profiles, we examined human skin samples of three body sites (N = 3 × 54) with flow cytometry and 16 S rRNA gene amplicon sequencing. Confirmed by sequencing of the very same samples, body site was found to be significantly different by flow cytometry. For a proof-of-principle multidimensional approach, using stool samples of patients (N = 40) with/without inflammatory bowel diseases, we could discriminate the health status by their bacterial patterns. In conclusion, FlowSoFine enables the generation and comparison of cytometric fingerprints of microbial communities from different sources. The implemented interface supports the user through all analytical steps to work out the biological relevant signals from raw measurements to publication ready figures. Furthermore, we present flow cytometry as a valid method for skin microbiota analysis.

Polymorphisms in genes of melatonin biosynthesis and signaling support the light-at-night hypothesis for breast cancer.

Light-at-night triggers the decline of pineal gland melatonin biosynthesis and secretion and is an IARC-classified probable breast-cancer risk factor. We applied a large-scale molecular epidemiology approach to shed light on the putative role of melatonin in breast cancer. We investigated associations between breast-cancer risk and polymorphisms at genes of melatonin biosynthesis/signaling using a study population of 44,405 women from the Breast Cancer Association Consortium (22,992 cases, 21,413 population-based controls). Genotype data of 97 candidate single nucleotide polymorphisms (SNPs) at 18 defined gene regions were investigated for breast-cancer risk effects. We calculated adjusted odds ratios (ORs) and 95% confidence intervals (CI) by logistic regression for the main-effect analysis as well as stratified analyses by estrogen- and progesterone-receptor (ER, PR) status. SNP-SNP interactions were analyzed via a two-step procedure based on logic regression. The Bayesian false-discovery probability (BFDP) was used for all analyses to account for multiple testing. Noteworthy associations (BFDP < 0.8) included 10 linked SNPs in tryptophan hydroxylase 2 (TPH2) (e.g. rs1386492: OR = 1.07, 95% CI 1.02-1.12), and a SNP in the mitogen-activated protein kinase 8 (MAPK8) (rs10857561: OR = 1.11, 95% CI 1.04-1.18). The SNP-SNP interaction analysis revealed noteworthy interaction terms with TPH2- and MAPK-related SNPs (e.g. rs1386483R ∧ rs1473473D ∧ rs3729931D: OR = 1.20, 95% CI 1.09-1.32). In line with the light-at-night hypothesis that links shift work with elevated breast-cancer risks our results point to SNPs in TPH2 and MAPK-genes that may impact the intricate network of circadian regulation.

Influence of breast cancer risk factors on proliferation and DNA damage in human breast glandular tissues: role of intracellular estrogen levels, oxidative stress and estrogen biotransformation.

Breast cancer etiology is associated with both proliferation and DNA damage induced by estrogens. Breast cancer risk factors (BCRF) such as body mass index (BMI), smoking, and intake of estrogen-active drugs were recently shown to influence intratissue estrogen levels. Thus, the aim of the present study was to investigate the influence of BCRF on estrogen-induced proliferation and DNA damage in 41 well-characterized breast glandular tissues derived from women without breast cancer. Influence of intramammary estrogen levels and BCRF on estrogen receptor (ESR) activation, ESR-related proliferation (indicated by levels of marker transcripts), oxidative stress (indicated by levels of GCLC transcript and oxidative derivatives of cholesterol), and levels of transcripts encoding enzymes involved in estrogen biotransformation was identified by multiple linear regression models. Metabolic fluxes to adducts of estrogens with DNA (E-DNA) were assessed by a metabolic network model (MNM) which was validated by comparison of calculated fluxes with data on methoxylated and glucuronidated estrogens determined by GC- and UHPLC-MS/MS. Intratissue estrogen levels significantly influenced ESR activation and fluxes to E-DNA within the MNM. Likewise, all BCRF directly and/or indirectly influenced ESR activation, proliferation, and key flux constraints influencing E-DNA (i.e., levels of estrogens, CYP1B1, SULT1A1, SULT1A2, and GSTP1). However, no unambiguous total effect of BCRF on proliferation became apparent. Furthermore, BMI was the only BCRF to indeed influence fluxes to E-DNA (via congruent adverse influence on levels of estrogens, CYP1B1 and SULT1A2).

Identifying treatment effect heterogeneity in dose-finding trials using Bayesian hierarchical models: Identifying treatment effect heterogeneity in dose-finding trials.

An important task in drug development is to identify patients, which respond better or worse to an experimental treatment. Identifying predictive covariates, which influence the treatment effect and can be used to define subgroups of patients, is a key aspect of this task. Analyses of treatment effect heterogeneity are however known to be challenging, since the number of possible covariates or subgroups is often large, while samples sizes in earlier phases of drug development are often small. In addition, distinguishing predictive covariates from prognostic covariates, which influence the response independent of the given treatment, can often be difficult. While many approaches for these types of problems have been proposed, most of them focus on the two-arm clinical trial setting, where patients are given either the treatment or a control. In this article we consider parallel groups dose-finding trials, in which patients are administered different doses of the same treatment. To investigate treatment effect heterogeneity in this setting we propose a Bayesian hierarchical dose-response model with covariate effects on dose-response parameters. We make use of shrinkage priors to prevent overfitting, which can easily occur, when the number of considered covariates is large and sample sizes are small. We compare several such priors in simulations and also investigate dependent modeling of prognostic and predictive effects to better distinguish these two types of effects. We illustrate the use of our proposed approach using a Phase II dose-finding trial and show how it can be used to identify predictive covariates and subgroups of patients with increased treatment effects.

Combining heterogeneous subgroups with graph-structured variable selection priors for Cox regression.

BACKGROUND: Important objectives in cancer research are the prediction of a patient's risk based on molecular measurements such as gene expression data and the identification of new prognostic biomarkers (e.g. genes). In clinical practice, this is often challenging because patient cohorts are typically small and can be heterogeneous. In classical subgroup analysis, a separate prediction model is fitted using only the data of one specific cohort. However, this can lead to a loss of power when the sample size is small. Simple pooling of all cohorts, on the other hand, can lead to biased results, especially when the cohorts are heterogeneous. RESULTS: We propose a new Bayesian approach suitable for continuous molecular measurements and survival outcome that identifies the important predictors and provides a separate risk prediction model for each cohort. It allows sharing information between cohorts to increase power by assuming a graph linking predictors within and across different cohorts. The graph helps to identify pathways of functionally related genes and genes that are simultaneously prognostic in different cohorts. CONCLUSIONS: Results demonstrate that our proposed approach is superior to the standard approaches in terms of prediction performance and increased power in variable selection when the sample size is small.

Influence of breast cancer risk factors and intramammary biotransformation on estrogen homeostasis in the human breast.

Understanding intramammary estrogen homeostasis constitutes the basis of understanding the role of lifestyle factors in breast cancer etiology. Thus, the aim of the present study was to identify variables influencing levels of the estrogens present in normal breast glandular and adipose tissues (GLT and ADT, i.e., 17ß-estradiol, estrone, estrone-3-sulfate, and 2-methoxy-estrone) by multiple linear regression models. Explanatory variables (exVARs) considered were (a) levels of metabolic precursors as well as levels of transcripts encoding proteins involved in estrogen (biotrans)formation, (b) data on breast cancer risk factors (i.e., body mass index, BMI, intake of estrogen-active drugs, and smoking) collected by questionnaire, and (c) tissue characteristics (i.e., mass percentage of oil, oil%, and lobule type of the GLT). Levels of estrogens in GLT and ADT were influenced by both extramammary production (menopausal status, intake of estrogen-active drugs, and BMI) thus showing that variables known to affect levels of circulating estrogens influence estrogen levels in breast tissues as well for the first time. Moreover, intratissue (biotrans)formation (by aromatase, hydroxysteroid-17beta-dehydrogenase 2, and beta-glucuronidase) influenced intratissue estrogen levels, as well. Distinct differences were observed between the exVARs exhibiting significant influence on (a) levels of specific estrogens and (b) the same dependent variables in GLT and ADT. Since oil% and lobule type of GLT influenced levels of some estrogens, these variables may be included in tissue characterization to prevent sample bias. In conclusion, evidence for the intracrine activity of the human breast supports biotransformation-based strategies for breast cancer prevention. The susceptibility of estrogen homeostasis to systemic and tissue-specific modulation renders both beneficial and adverse effects of further variables associated with lifestyle and the environment possible.

Statistical analysis of in vivo alkaline comet assay data - Comparison of median and geometric mean as centrality measures.

The comet assay is one of the standard tests for evaluating the genotoxic potential of a test item able to detect DNA strand breaks in cells or isolated nuclei from various tissues. The in vivo alkaline comet assay is part of the standard test battery, given in option 2 of the ICH guidance S2 (R1) and a follow-up test in the EFSA framework on genotoxicity testing. The current OECD guideline for the testing of chemicals No. 489 directly affects the statistical analysis of comet data as it suggests using the median per slide and the mean of all medians per animal. However, alternative approaches can be used if scientifically justified. In this work, we demonstrated that the selection of different centrality measures to describe an average value per slide may lead to fundamentally different statistical test results and contradicting interpretations. Our focus was on geometric means and medians per slide for the primary endpoint "tail intensity". We compared both strategies using original and simulated data in different experimental settings incl. a varying number of animals, slides and cells per slide. In general, it turned out that the chosen centrality measure has an immense impact on the final statistical test result.

Benefits of improved air quality on ageing lungs: impacts of genetics and obesity.

INTRODUCTION: The beneficial effect of improving air quality on lung function in the elderly remains unclear. We examined associations between decline in air pollutants and lung function, and effect modifications by genetics and body mass index (BMI), in elderly German women. METHODS: Data were analysed from the prospective SALIA (Study on the influence of Air pollution on Lung function, Inflammation and Aging) study (n=601). Spirometry was conducted at baseline (1985-1994; age 55 years), in 2007-2010 and in 2012-2013. Air pollution concentrations at home addresses were determined for each time-point using land-use regression models. Global Lung Initiative 2012 z-scores were calculated. Weighted genetic risk scores (GRSs) were determined from lung function-related risk alleles and used to investigate interactions with improved air quality. Multiple linear mixed models were fitted. RESULTS: Air pollution levels decreased substantially during the study period. Reduction of air pollution was associated with an increase in z-scores for forced expiratory volume in 1 s (FEV1) and the FEV1/forced vital capacity ratio. For a decrease of 10 µg·m-3 in nitrogen dioxide (NO2), the z-score for FEV1 increased by 0.14 (95% CI 0.01-0.26). However, with an increasing number of lung function-related risk alleles, the benefit from improved air quality decreased (GRS×NO2 interaction: p=0.029). Interactions with BMI were not significant. CONCLUSIONS: Reduction of air pollution is associated with a relative improvement of lung function in elderly women, but also depends on their genetic make-up.

Identification of interactions of binary variables associated with survival time using survivalFS.

Many medical studies aim to identify factors associated with a time to an event such as survival time or time to relapse. Often, in particular, when binary variables are considered in such studies, interactions of these variables might be the actual relevant factors for predicting, e.g., the time to recurrence of a disease. Testing all possible interactions is often not possible, so that procedures such as logic regression are required that avoid such an exhaustive search. In this article, we present an ensemble method based on logic regression that can cope with the instability of the regression models generated by logic regression. This procedure called survivalFS also provides measures for quantifying the importance of the interactions forming the logic regression models on the time to an event and for the assessment of the individual variables that take the multivariate data structure into account. In this context, we introduce a new performance measure, which is an adaptation of Harrel's concordance index. The performance of survivalFS and the proposed importance measures is evaluated in a simulation study as well as in an application to genotype data from a urinary bladder cancer study. Furthermore, we compare the performance of survivalFS and its importance measures for the individual variables with the variable importance measure used in random survival forests, a popular procedure for the analysis of survival data. These applications show that survivalFS is able to identify interactions associated with time to an event and to outperform random survival forests.

Qualitative and quantitative differences in estrogen biotransformation in human breast glandular and adipose tissues: implications for studies using mammary biospecimens.

Because of its assumed role in breast cancer etiology, estrogen biotransformation (and interaction of compounds therewith) has been investigated in human biospecimens for decades. However, little attention has been paid to the well-known fact that large inter-individual variations exist in the proportion of breast glandular (GLT) and adipose (ADT) tissues and less to adequate tissue characterization. To assess the relevance of this, the present study compares estrogen biotransformation in GLT and ADT. GLT and ADT were isolated from 47 reduction mammoplasty specimens derived from women without breast cancer and were characterized histologically and by their percentages of oil. Levels of 12 unconjugated and five conjugated estrogens were analyzed by GC- and UHPLC-MS/MS, respectively, and levels of 27 transcripts encoding proteins involved in estrogen biotransformation by Taqman® probe-based PCR. Unexpectedly, one-third of specimens provided neat GLT only after cryosection. Whereas 17ß-estradiol, estrone, and estrone-3-sulfate were detected in both tissues, estrone-3-glucuronide and 2-methoxy-estrone were detected predominately in GLT and ADT, respectively. Estrogen levels as well as ratios 17ß-estradiol/estrone and estrone-3-sulfate/estrone differed significantly between GLT and ADT, yet less than between individuals. Furthermore, estrogen levels in GLT and ADT correlated significantly with each other. In contrast, levels of most transcripts encoding enzymes involved in biotransformation differed more than between individuals and did not correlate between ADT and GLT. Thus, mixed breast tissues (and plasma) will not provide meaningful information on local estrogen biotransformation (and interaction of compounds therewith) whereas relative changes in 17ß-estradiol levels may be investigated in the more abundant ADT.

Boron-exposed male workers in Turkey: no change in sperm Y:X chromosome ratio and in offspring's sex ratio.

Boron-associated shifts in sex ratios at birth were suggested earlier and attributed to a decrease in Y- vs. X-bearing sperm cells. As the matter is pivotal in the discussion of reproductive toxicity of boron/borates, re-investigation in a highly borate-exposed population was required. In the present study, 304 male workers in Bandirma and Bigadic (Turkey) with different degrees of occupational and environmental exposure to boron were investigated. Boron was quantified in blood, urine and semen, and the persons were allocated to exposure groups along B blood levels. In the highest ("extreme") exposure group (n = 69), calculated mean daily boron exposures, semen boron and blood boron concentrations were 44.91 ± 18.32 mg B/day, 1643.23 ± 965.44 ng B/g semen and 553.83 ± 149.52 ng B/g blood, respectively. Overall, an association between boron exposure and Y:X sperm ratios in semen was not statistically significant (p > 0.05). Also, the mean Y:X sperm ratios in semen samples of workers allocated to the different exposure groups were statistically not different in pairwise comparisons (p > 0.05). Additionally, a boron-associated shift in sex ratio at birth towards female offspring was not visible. In essence, the present results do not support an association between boron exposure and decreased Y:X sperm ratio in males, even under extreme boron exposure conditions.

Birth weights of newborns and pregnancy outcomes of environmentally boron-exposed females in Turkey.

Boric acid and sodium borates are currently classified as being toxic to reproduction under "Category 1B" with the hazard statement of "H360 FD" in the European CLP regulation. This has prompted studies on boron-mediated reprotoxic effects in male workers in boron mining areas and boric acid production plants. By contrast, studies on boron-mediated developmental effects in females are scarce. The present study was designed to fill this gap. Hundred and ninety nine females residing in Bandirma and Bigadic participated in this study investigating pregnancy outcomes. The participants constituted a study group covering blood boron from low (< 100 ng B/g blood, n = 143) to high (> 150 ng B/g blood, n = 27) concentrations. The mean blood boron concentration and the mean estimated daily boron exposure of the high exposure group was 274.58 (151.81-975.66) ng B/g blood and 24.67 (10.47-57.86) mg B/day, respectively. In spite of the high level of daily boron exposure, boron-mediated adverse effects on induced abortion, spontaneous abortion (miscarriage), stillbirth, infant death, neonatal death, early neonatal death, preterm birth, congenital anomalies, sex ratio and birth weight of newborns were not observed.

Evaluation of FSH, LH, testosterone levels and semen parameters in male boron workers under extreme exposure conditions.

Boric acid and sodium borates are currently classified in the EU-CLP regulation as "toxic to reproduction" under "Category 1B", with hazard statement of H360FD. However, so far field studies on male reproduction in China and in Turkey could not confirm such boron-associated toxic effects. As validation by another independent study is still required, the present study has investigated possible boron-associated effects on male reproduction in workers (n = 212) under different boron exposure conditions. The mean daily boron exposure (DBE) and blood boron concentration of workers in the extreme exposure group (n = 98) were 47.17 ± 17.47 (7.95-106.8) mg B/day and 570.6 ± 160.1 (402.6-1100) ng B/g blood, respectively. Nevertheless, boron-associated adverse effects on semen parameters, as well as on FSH, LH and total testosterone levels were not seen, even within the extreme exposure group. With this study, a total body of evidence has accumulated that allows to conclude that male reproductive effects are not relevant to humans, under any feasible and realistic conditions of exposure to inorganic boron compounds.

Identification and replication of the interplay of four genetic high-risk variants for urinary bladder cancer.

Little is known whether genetic variants identified in genome-wide association studies interact to increase bladder cancer risk. Recently, we identified two- and three-variant combinations associated with a particular increase of bladder cancer risk in a urinary bladder cancer case-control series (Leibniz Research Centre for Working Environment and Human Factors at TU Dortmund (IfADo), 1501 cases, 1565 controls). In an independent case-control series (Nijmegen Bladder Cancer Study, NBCS, 1468 cases, 1720 controls) we confirmed these two- and three-variant combinations. Pooled analysis of the two studies as discovery group (IfADo-NBCS) resulted in sufficient statistical power to test up to four-variant combinations by a logistic regression approach. The New England and Spanish Bladder Cancer Studies (2080 cases and 2167 controls) were used as a replication series. Twelve previously identified risk variants were considered. The strongest four-variant combination was obtained in never smokers. The combination of rs1014971[AA] near apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like 3A (APOBEC3A) and chromobox homolog 6 (CBX6), solute carrier family 1s4 (urea transporter), member 1 (Kidd blood group) (SLC14A1) exon single nucleotide polymorphism (SNP) rs1058396[AG, GG], UDP glucuronosyltransferase 1 family, polypeptide A complex locus (UGT1A) intron SNP rs11892031[AA] and rs8102137[CC, CT] near cyclin E1 (CCNE1) resulted in an unadjusted odds ratio (OR) of 2.59 (95% CI = 1.93-3.47; P = 1.87 × 10-10), while the individual variant ORs ranged only between 1.11 and 1.30. The combination replicated in the New England and Spanish Bladder Cancer Studies (ORunadjusted = 1.60, 95% CI = 1.10-2.33; P = 0.013). The four-variant combination is relatively frequent, with 25% in never smoking cases and 11% in never smoking controls (total study group: 19% cases, 14% controls). In conclusion, we show that four high-risk variants can statistically interact to confer increased bladder cancer risk particularly in never smokers.

Detection of gene-environment interactions in the presence of linkage disequilibrium and noise by using genetic risk scores with internal weights from elastic net regression.

BACKGROUND: For the analysis of gene-environment (GxE) interactions commonly single nucleotide polymorphisms (SNPs) are used to characterize genetic susceptibility, an approach that mostly lacks power and has poor reproducibility. One promising approach to overcome this problem might be the use of weighted genetic risk scores (GRS), which are defined as weighted sums of risk alleles of gene variants. The gold-standard is to use external weights from published meta-analyses. METHODS: In this study, we used internal weights from the marginal genetic effects of the SNPs estimated by a multivariate elastic net regression and thereby provided a method that can be used if there are no external weights available. We conducted a simulation study for the detection of GxE interactions and compared power and type I error of single SNPs analyses with Bonferroni correction and corresponding analysis with unweighted and our weighted GRS approach in scenarios with six risk SNPs and an increasing number of highly correlated (up to 210) and noise SNPs (up to 840). RESULTS: Applying weighted GRS increased the power enormously in comparison to the common single SNPs approach (e.g. 94.2% vs. 35.4%, respectively, to detect a weak interaction with an OR ≈ 1.04 for six uncorrelated risk SNPs and n = 700 with a well-controlled type I error). Furthermore, weighted GRS outperformed the unweighted GRS, in particular in the presence of SNPs without any effect on the phenotype (e.g. 90.1% vs. 43.9%, respectively, when 20 noise SNPs were added to the six risk SNPs). This outperforming of the weighted GRS was confirmed in a real data application on lung inflammation in the SALIA cohort (n = 402). However, in scenarios with a high number of noise SNPs (>200 vs. 6 risk SNPs), larger sample sizes are needed to avoid an increased type I error, whereas a high number of correlated SNPs can be handled even in small samples (e.g. n = 400). CONCLUSION: In conclusion, weighted GRS with weights from the marginal genetic effects of the SNPs estimated by a multivariate elastic net regression were shown to be a powerful tool to detect gene-environment interactions in scenarios of high Linkage disequilibrium and noise.

Comparison of weighting approaches for genetic risk scores in gene-environment interaction studies.

BACKGROUND: Weighted genetic risk scores (GRS), defined as weighted sums of risk alleles of single nucleotide polymorphisms (SNPs), are statistically powerful for detection gene-environment (GxE) interactions. To assign weights, the gold standard is to use external weights from an independent study. However, appropriate external weights are not always available. In such situations and in the presence of predominant marginal genetic effects, we have shown in a previous study that GRS with internal weights from marginal genetic effects ("GRS-marginal-internal") are a powerful and reliable alternative to single SNP approaches or the use of unweighted GRS. However, this approach might not be appropriate for detecting predominant interactions, i.e. interactions showing an effect stronger than the marginal genetic effect. METHODS: In this paper, we present a weighting approach for such predominant interactions ("GRS-interaction-training") in which parts of the data are used to estimate the weights from the interaction terms and the remaining data are used to determine the GRS. We conducted a simulation study for the detection of GxE interactions in which we evaluated power, type I error and sign-misspecification. We compared this new weighting approach to the GRS-marginal-internal approach and to GRS with external weights. RESULTS: Our simulation study showed that in the absence of external weights and with predominant interaction effects, the highest power was reached with the GRS-interaction-training approach. If marginal genetic effects were predominant, the GRS-marginal-internal approach was more appropriate. Furthermore, the power to detect interactions reached by the GRS-interaction-training approach was only slightly lower than the power achieved by GRS with external weights. The power of the GRS-interaction-training approach was confirmed in a real data application to the Traffic, Asthma and Genetics (TAG) Study (N = 4465 observations). CONCLUSION: When appropriate external weights are unavailable, we recommend to use internal weights from the study population itself to construct weighted GRS for GxE interaction studies. If the SNPs were chosen because a strong marginal genetic effect was hypothesized, GRS-marginal-internal should be used. If the SNPs were chosen because of their collective impact on the biological mechanisms mediating the environmental effect (hypothesis of predominant interactions) GRS-interaction-training should be applied.

Phase II dose-response trials: A simulation study to compare analysis method performance under design considerations.

Phase II trials are intended to provide information about the dose-response relationship and to support the choice of doses for a pivotal phase III trial. Recently, new analysis methods have been proposed to address these objectives, and guidance is needed to select the most appropriate analysis method in specific situations. We set up a simulation study to evaluate multiple performance measures of one traditional and three more recent dose-finding approaches under four design options and illustrate the investigated analysis methods with an example from clinical practice. Our results reveal no general recommendation for a particular analysis method across all design options and performance measures. However, we also demonstrate that the new analysis methods are worth the effort compared to the traditional ANOVA-based approach.

Multivariate binary classification of imbalanced datasets-A case study based on high-dimensional multiplex autoimmune assay data.

The classification of a population by a specific trait is a major task in medicine, for example when in a diagnostic setting groups of patients with specific diseases are identified, but also when in predictive medicine a group of patients is classified into specific disease severity classes that might profit from different treatments. When the sizes of those subgroups become small, for example in rare diseases, imbalances between the classes are more the rule than the exception and make statistical classification problematic when the error rate of the minority class is high. Many observations are classified as belonging to the majority class, while the error rate of the majority class is low. This case study aims to investigate class imbalance for Random Forests and Powered Partial Least Squares Discriminant Analysis (PPLS-DA) and to evaluate the performance of these classifiers when they are combined with methods to compensate imbalance (sampling methods, cost-sensitive learning approaches). We evaluate all approaches with a scoring system taking the classification results into consideration. This case study is based on one high-dimensional multiplex autoimmune assay dataset describing immune response to antigens and consisting of two classes of patients: Rheumatoid Arthritis (RA) and Systemic Lupus Erythemathodes (SLE). Datasets with varying degrees of imbalance are created by successively reducing the class of RA patients. Our results indicate possible benefit of cost-sensitive learning approaches for Random Forests. Although further research is needed to verify our findings by investigating other datasets or large-scale simulation studies, we claim that this work has the potential to increase awareness of practitioners to this problem of class imbalance and stresses the importance of considering methods to compensate class imbalance.