Background: Data generated using 'omics' technologies are characterized by high dimensionality, where the number of features measured per subject vastly exceeds the number of subjects in the study. In this paper, we consider issues relevant in the design of biomedical studies in which the goal is the discovery of a subset of features and an associated algorithm that can predict a binary outcome, such as disease status. We compare the performance of four commonly used classifiers (K-Nearest Neighbors, Prediction Analysis for Microarrays, Random Forests and Support Vector Machines) in high-dimensionality data settings. We evaluate the effects of varying levels of signal-to-noise ratio in the dataset, imbalance in class distribution and choice of metric for quantifying performance of the classifier. To guide study design, we present a summary of the key characteristics of 'omics' data profiled in several human or animal model experiments utilizing high-content mass spectrometry and multiplexed immunoassay based techniques. Results: The analysis of data from seven 'omics' studies revealed that the average magnitude of effect size observed in human studies was markedly lower when compared to that in animal studies. The data measured in human studies were characterized by higher biological variation and the presence of outliers. The results from simulation studies indicated that the classifier Prediction Analysis for Microarrays (PAM) had the highest power when the class conditional feature distributions were Gaussian and outcome distributions were balanced. Random Forests was optimal when feature distributions were skewed and when class distributions were unbalanced. We provide a free open-source R statistical software library (MVpower) that implements the simulation strategy proposed in this paper. Conclusion: No single classifier had optimal performance under all settings. Simulation studies provide useful guidance for the design of biomedical studies involving high-dimensionality data.

Tumors are capable of shedding DNA into the blood stream. This shed DNA may be recovered from serum or plasma. The objective of this study was to evaluate whether pyrosequencing promoter DNA in a panel of 12 breast cancer-related genes (APC, BRCA1, CCND2, CDH1, ESR1, GSTP1, HIN1, P16, RARβ, RASSF1, SFRP1 and TWIST) to measure the degree of methylation would lead to a useful serum-based marker of breast cancer. Serum was obtained from women who were about to undergo a breast biopsy or mastectomy at three hospitals from 1977 to 1987 in Grand Rapids, MI USA. We compared the methylation status of 12 genes in serum DNA obtained from three groups of postmenopausal women (mean age at blood collection: 63.0 y; SD 9.9; range 35–91): breast cancer cases with lymph node-positive disease (n = 241); breast cancer cases with lymph node-negative disease (n = 63); and benign breast disease control subjects (n = 234). Overall, median levels of promoter methylation were low, typically below 5%, for all genes in all study groups. For all genes, median levels of methylation were higher (by 3.3 to 47.6%) in lymph node-positive breast cancer cases than in the controls. Comparing mean methylation level between lymph-node positive cases and controls, the most statistically significant findings, after adjustment of the false-positive rate (q-value), were for TWIST (p = 0.04), SFRP1 (p = 0.16), ESR1 (p = 0.17), P16 (p = 0.19) and APC (p = 0.19). For two of these four genes (TWIST, P16), the median methylation level was also highest in lymph-node positive cases, intermediate in lymph node-negative cases and lowest in the controls. The percent of study subjects with mean methylation scores ≥ 5% was higher among lymph node-positive cases than controls for ten genes, and significantly higher for HIN1 and TWIST (22.0 vs. 12.2%, p = 0.04 and 37.9 vs. 24.5%, p = 0.004, respectively). Despite relatively consistent variation in methylation patterns among groups, these modest differences did not provide sufficient ability to distinguish between cases and controls in a clinical setting.

Metabolomics profiling has shown promise in elucidating the biological pathways underpinning mortality, but there are limited data in female populations. We applied a liquid chromatography-tandem mass spectrometry metabolomics platform to EDTA-plasma to measure 470 metabolites at baseline in a discovery set of 943 postmenopausal women (including 417 incident deaths, median time to death of 10.6 years) with validation in an independent set of 1355 postmenopausal women (including 685 deaths, median time to death of 9.1 years) in the Women’s Health Initiative. Eight new metabolites were discovered to be associated with all-cause mortality. Findings included protective effects of increased levels of three amino acids (asparagine, homoarginine and tryptophan) and docosatrienoic acid; and detrimental effects of increased levels of C4-OH-carnitine, hexadecanedioate and two purine/pyrimidines (N2, N2-dimethylguanosine and N4-acetylcytidine). In addition, a set of nine previously published metabolite associations were replicated. A metabolite score comprising 17 metabolites was associated with mortality (P < 10^–8) after adjustment for risk factors, with a hazard ratio of 1.95 (95% CI: 1.46–2.62) for women in the highest quartile compared with the lowest quartile of metabolite score. The score was robust among younger women and older women, for both cardiovascular and non-cardiovascular mortality, and associated with both early deaths (within the first 10 years of baseline) and later deaths. Our study fills a gap in the literature by identifying eight novel metabolite associations with all-cause mortality in women, using a robust study design involving independent discovery and validation datasets.

Objective. To examine the association of dietary quality and risk of incident diabetes overall and by race/ethnicity among postmenopausal women enrolled in the Women's Health Initiative (WHI). Research methods and procedures. The WHI recruited 161,808 postmenopausal women between 1993 and 1998, and followed them until 2005. Incident diabetes was determined annually over an average of 7.6 years from enrollment. At baseline, all participants completed a Food Frequency Questionnaire (FFQ). Dietary quality was assessed by the Alternate Healthy Eating Index (AHEI), calculated from the baseline FFQ responses. Results. There were 10,307 incident cases of self-reported treated diabetes over 1,172,761 person-years of follow-up. Most participants did not meet the AHEI dietary goals; that is, only 0.1% of women met or exceeded the recommended consumption of vegetables, and few (17.3%) met or exceeded the recommended level for total fiber. After adjusting for potential confounders, women in the highest quintile of the AHEI score were 24% less likely to develop diabetes relative to women in the lowest quintile of AHEI [hazard ratio (HR) = 0.76 (95% CI: 0.70–0.82)]. This association was observed in Whites [HR = 0.74 (95% CI: 0.68–0.82)] and Hispanics [HR = 0.68 (95% CI: 0.46–0.99)], but not in Blacks [HR = 0.85 (95% CI: 0.69–1.05)] or Asians [HR = 0.88 (95% CI: 0.57–1.38)]. Conclusion. These findings support a protective role of healthful eating choices in reducing the risk of developing diabetes, after adjusting for other lifestyle factors, in White and Hispanic postmenopausal women. Future studies are needed to investigate the relationship between dietary quality and risk of diabetes among Blacks and Asians in relationship to other lifestyle factors.

Background: Racial differences in metabolomic profiles may reflect underlying differences in social determinants of health by self-reported race and may be related to racial disparities in coronary heart disease (CHD) among women in the United States. However, the magnitude of differences in metabolomic profiles between Black and White women in the United States has not been well-described. It also remains unknown whether such differences are related to differences in CHD risk. Methods: Plasma metabolomic profiles were analyzed using liquid chromatography-tandem mass spectrometry in the WHI-OS (Women’s Health Initiative-Observational Study; 138 Black and 696 White women), WHI-HT trials (WHI-Hormone Therapy; 156 Black and 1138 White women), MESA (Multi-Ethnic Study of Atherosclerosis; 114 Black and 219 White women), JHS (Jackson Heart Study; 1465 Black women with 107 incident CHD cases), and NHS (Nurses’ Health Study; 2506 White women with 136 incident CHD cases). First, linear regression models were used to estimate associations between self-reported race and 472 metabolites in WHI-OS (discovery); findings were replicated in WHI-HT and validated in MESA. Second, we used elastic net regression to construct a racial difference metabolomic pattern (RDMP) representing differences in the metabolomic patterns between Black and White women in the WHI-OS; the RDMP was validated in the WHI-HT and MESA. Third, using conditional logistic regressions in the WHI (717 CHD cases and 719 matched controls), we examined associations of metabolites with large differences in levels by race and the RDMP with risk of CHD, and the results were replicated in Black women from the JHS and White women from the NHS. Results: Of the 472 tested metabolites, levels of 259 (54.9%) metabolites, mostly lipid metabolites and amino acids, significantly differed between Black and White women in both WHI-OS and WHI-HT after adjusting for baseline characteristics, socioeconomic status, lifestyle factors, baseline health conditions, and medication use (false discovery rate <0.05); similar trends were observed in MESA. The RDMP, composed of 152 metabolites, was identified in the WHI-OS and showed significantly different distributions between Black and White women in the WHI-HT and MESA. Higher RDMP quartiles were associated with an increased risk of incident CHD (odds ratio=1.51 [0.97–2.37] for the highest quartile comparing to the lowest; P _trend =0.02), independent of self-reported race and known CHD risk factors. In race-stratified analyses, the RDMP-CHD associations were more pronounced in White women. Similar patterns were observed in Black women from the JHS and White women from the NHS. Conclusions: Metabolomic profiles significantly and substantially differ between Black and White women and may be associated with CHD risk and racial disparities in US women.