The health effects of omega-3 fatty acids have been controversial. Here we report the results of a de novo pooled analysis conducted with data from 17 prospective cohort studies examining the associations between blood omega-3 fatty acid levels and risk for all-cause mortality. Over a median of 16 years of follow-up, 15,720 deaths occurred among 42,466 individuals. We found that, after multivariable adjustment for relevant risk factors, risk for death from all causes was significantly lower (by 15–18%, at least p < 0.003) in the highest vs the lowest quintile for circulating long chain (20–22 carbon) omega-3 fatty acids (eicosapentaenoic, docosapentaenoic, and docosahexaenoic acids). Similar relationships were seen for death from cardiovascular disease, cancer and other causes. No associations were seen with the 18-carbon omega-3, alpha-linolenic acid. These findings suggest that higher circulating levels of marine n-3 PUFA are associated with a lower risk of premature death.

In the analysis of current genomic data, application of machine learning and data mining techniques has become more attractive given the rising complexity of the projects. As part of the Genetic Analysis Workshop 19, approaches from this domain were explored, mostly motivated from two starting points. First, assuming an underlying structure in the genomic data, data mining might identify this and thus improve downstream association analyses. Second, computational methods for machine learning need to be developed further to efficiently deal with the current wealth of data. In the course of discussing results and experiences from the machine learning and data mining approaches, six common messages were extracted. These depict the current state of these approaches in the application to complex genomic data. Although some challenges remain for future studies, important forward steps were taken in the integration of different data types and the evaluation of the evidence. Mining the data for underlying genetic or phenotypic structure and using this information in subsequent analyses proved to be extremely helpful and is likely to become of even greater use with more complex data sets.

De novo lipogenesis (DNL) is the primary metabolic pathway synthesizing fatty acids from carbohydrates, protein, or alcohol. Our aim was to examine associations of in vivo levels of selected fatty acids (16:0, 16:1n7, 18:0, 18:1n9) in DNL with incidence of type 2 diabetes (T2D). Seventeen cohorts from 12 countries (7 from Europe, 7 from the United States, 1 from Australia, 1 from Taiwan; baseline years = 1970–1973 to 2006–2010) conducted harmonized individual-level analyses of associations of DNL-related fatty acids with incident T2D. In total, we evaluated 65,225 participants (mean ages = 52.3–75.5 years; % women = 20.4%–62.3% in 12 cohorts recruiting both sexes) and 15,383 incident cases of T2D over the 9-year follow-up on average. Cohort-specific association of each of 16:0, 16:1n7, 18:0, and 18:1n9 with incident T2D was estimated, adjusted for demographic factors, socioeconomic characteristics, alcohol, smoking, physical activity, dyslipidemia, hypertension, menopausal status, and adiposity. Cohort-specific associations were meta-analyzed with an inverse-variance-weighted approach. Each of the 4 fatty acids positively related to incident T2D. Relative risks (RRs) per cohort-specific range between midpoints of the top and bottom quintiles of fatty acid concentrations were 1.53 (1.41–1.66; p < 0.001) for 16:0, 1.40 (1.33–1.48; p < 0.001) for 16:1n-7, 1.14 (1.05–1.22; p = 0.001) for 18:0, and 1.16 (1.07–1.25; p < 0.001) for 18:1n9. Heterogeneity was seen across cohorts (I² = 51.1%–73.1% for each fatty acid) but not explained by lipid fractions and global geographical regions. Further adjusted for triglycerides (and 16:0 when appropriate) to evaluate associations independent of overall DNL, the associations remained significant for 16:0, 16:1n7, and 18:0 but were attenuated for 18:1n9 (RR = 1.03, 95% confidence interval (CI) = 0.94–1.13). These findings had limitations in potential reverse causation and residual confounding by imprecisely measured or unmeasured factors. Concentrations of fatty acids in the DNL were positively associated with T2D incidence. Our findings support further work to investigate a possible role of DNL and individual fatty acids in the development of T2D.

The use of simulation-based methods for introducing inferen-ce is growing in popularity for the Stat 101 course, due in part to increasing evidence of the methods ability to improve studen-ts’ statistical thinking. This impact comes from simulation-based methods (a) clearly presenting the overarching logic of inference, (b) strengthening ties between statistics and probability/mathematical concepts, (c) encouraging a focus on the entire research process, (d) facilitating student thinking about advanced statistical concepts, (e) allowing more time to explore, do, and talk about real research and messy data, and (f) acting as a firm-er foundation on which to build statistical intuition. Thus, we argue that simulation-based inference should be an entry point to an undergraduate statistics program for all students, and that simulation-based inference should be used throughout all under-graduate statistics courses. To achieve this goal and fully recognize the benefits of simulation-based inference on the undergraduate statistics program, we will need to break free of historical forces tying undergraduate statistics curricula to mathematics, consider radical and innovative new pedagogical approaches in our courses, fully implement assessment-driven content innovations, and embrace computation throughout the curriculum. [Received December 2014. Revised July 2015]

Machine learning methods continue to show promise in the analysis of data from genetic association studies because of the high number of variables relative to the number of observations. However, few best practices exist for the application of these methods. We extend a recently proposed supervised machine learning approach for predicting disease risk by genotypes to be able to incorporate gene expression data and rare variants. We then apply 2 different versions of the approach (radial and linear support vector machines) to simulated data from Genetic Analysis Workshop 19 and compare performance to logistic regression. Method performance was not radically different across the 3 methods, although the linear support vector machine tended to show small gains in predictive ability relative to a radial support vector machine and logistic regression. Importantly, as the number of genes in the models was increased, even when those genes contained causal rare variants, model predictive ability showed a statistically significant decrease in performance for both the radial support vector machine and logistic regression. The linear support vector machine showed more robust performance to the inclusion of additional genes. Further work is needed to evaluate machine learning approaches on larger samples and to evaluate the relative improvement in model prediction from the incorporation of gene expression data.