Background: Gene-set analysis evaluates the expression of biological pathways, or a priori defined gene sets, rather than that of individual genes, in association with a binary phenotype, and is of great biologic interest in many DNA microarray studies. Gene Set Enrichment Analysis (GSEA) has been applied widely as a tool for gene-set analyses. We describe here some critical problems with GSEA and propose an alternative method by extending the individual-gene analysis method, Significance Analysis of Microarray (SAM), to gene-set analyses (SAM-GS). Results: Using a mouse microarray dataset with simulated gene sets, we illustrate that GSEA gives statistical significance to gene sets that have no gene associated with the phenotype (null gene sets), and has very low power to detect gene sets in which half the genes are moderately or strongly associated with the phenotype (truly-associated gene sets). SAM-GS, on the other hand, performs very well. The two methods are also compared in the analyses of three real microarray datasets and relevant pathways, the diverging results of which clearly show advantages of SAM-GS over GSEA, both statistically and biologically. In a microarray study for identifying biological pathways whose gene expressions are associated with p53 mutation in cancer cell lines, we found biologically relevant performance differences between the two methods. Specifically, there are 31 additional pathways identified as significant by SAM-GS over GSEA, that are associated with the presence vs. absence of p53. Of the 31 gene sets, 11 actually involve p53 directly as a member. A further 6 gene sets directly involve the extrinsic and intrinsic apoptosis pathways, 3 involve the cell-cycle machinery, and 3 involve cytokines and/or JAK/STAT signaling. Each of these 12 gene sets, then, is in a direct, well-established relationship with aspects of p53 signaling. Of the remaining 8 gene sets, 6 have plausible, if less well established, links with p53. Conclusion: We conclude that GSEA has important limitations as a gene-set analysis approach for microarray experiments for identifying biological pathways associated with a binary phenotype. As an alternative statistically-sound method, we propose SAM-GS. A free Excel Add-In for performing SAM-GS is available for public use.

Multiple data-analytic methods have been proposed for evaluating gene-expression levels in specific biological pathways, assessing differential expression associated with a binary phenotype. Following Goeman and Bühlmann's recent review, we compared statistical performance of three methods, namely Global Test, ANCOVA Global Test, and SAM-GS, that test "self-contained null hypotheses" Via. subject sampling. The three methods were compared based on a simulation experiment and analyses of three real-world microarray datasets. In the simulation experiment, we found that the use of the asymptotic distribution in the two Global Tests leads to a statistical test with an incorrect size. Specifically, p-values calculated by the scaled χ2 distribution of Global Test and the asymptotic distribution of ANCOVA Global Test are too liberal, while the asymptotic distribution with a quadratic form of the Global Test results in p-values that are too conservative. The two Global Tests with permutation-based inference, however, gave a correct size. While the three methods showed similar power using permutation inference after a proper standardization of gene expression data, SAM-GS showed slightly higher power than the Global Tests. In the analysis of a real-world microarray dataset, the two Global Tests gave markedly different results, compared to SAM-GS, in identifying pathways whose gene expressions are associated with p53 mutation in cancer cell lines. A proper standardization of gene expression variances is necessary for the two Global Tests in order to produce biologically sensible results. After the standardization, the three methods gave very similar biologically-sensible results, with slightly higher statistical significance given by SAM-GS. The three methods gave similar patterns of results in the analysis of the other two microarray datasets. An appropriate standardization makes the performance of all three methods similar, given the use of permutation-based inference. SAM-GS tends to have slightly higher power in the lower α-level region (i.e. gene sets that are of the greatest interest). Global Test and ANCOVA Global Test have the important advantage of being able to analyze continuous and survival phenotypes and to adjust for covariates. A free Microsoft Excel Add-In to perform SAM-GS is available from http://www.ualberta.ca/~yyasui/homepage.html .

Gene-set analysis aims to identify differentially expressed gene sets (pathways) by a phenotype in DNA microarray studies. We review here important methodological aspects of gene-set analysis and illustrate them with varying performance of several methods proposed in the literature. We emphasize the importance of distinguishing between ‘self-contained’ versus ‘competitive’ methods, following Goeman and Bühlmann. We also discuss reducing a gene set to its subset, consisting of ‘core members’ that chiefly contribute to the statistical significance of the differential expression of the initial gene set by phenotype. Significance analysis of microarray for gene-set reduction (SAM-GSR) can be used for an analytical reduction of gene sets to their core subsets. We apply SAM-GSR on a microarray dataset for identifying biological gene sets (pathways) whose gene expressions are associated with p53 mutation in cancer cell lines. Codes to implement SAM-GSR in the statistical package R can be downloaded from http://www.ualberta.ca/~yyasui/homepage.html.

The risk assessment matrix is a widely accepted, semi-quantitative tool for assessing risks, and setting priorities in risk management. Although the method can be useful to promote discussion to distinguish high risks from low risks, a published critique described a problem when the frequency and severity of risks are negatively correlated. A theoretical analysis showed that risk predictions could be misleading. We evaluated a practical public health example because it provided experiential risk data that allowed us to assess the practical implications of the published concern that risk matrices would make predictions that are worse than random. We explored this predicted problem by constructing a risk assessment matrix using a public health risk scenario—Tainted blood transfusion infection risk—That provides negative correlation between harm frequency and severity. We estimated the risk from the experiential data and compared these estimates with those provided by the risk assessment matrix. Although we validated the theoretical concern, for these authentic experiential data, the practical scope of the problem was limited. The risk matrix has been widely used in risk assessment. This method should not be abandoned wholesale, but users must address the source of the problem, apply the risk matrix with a full understanding of this problem and use matrix predictions to inform, but not drive decision-making.

We aimed to determine predictors of, and outcomes after, veno-arterial extracorporeal membrane oxygenation instituted within 48 h after cardiac surgery (early ECMO) in young infants. Patients ≤ 6 weeks old having cardiac surgery from 2003 to 2012 were enrolled prospectively. Patients cannulated pre-operatively, intra-operatively, or ≥ 48 h post-operatively were excluded. Variables at p ≤ 0.1 on univariate regression were entered into multiple logistic regression to predict early ECMO. Early-ECMO cases were matched 1:2 for six demographic variables, and death by age 2 years old (determined using conditional logistic regression; presented as odds ratio (OR), 95% confidence interval (CI)) and General Adaptive Composite scores at age 2 years (determined using Wilcoxon rank sum) were compared; p ≤ 0.05 was considered statistically significant. Of 565 eligible patients over the 10-year period, 20 had early ECMO instituted at a mean (standard deviation) of 12.4 (11.4) h post-operatively, 10 of whom had extracorporeal cardiopulmonary resuscitation. Of early-ECMO patients, 8 (40%) were found to have residual anatomic defects requiring intervention with catheterization (n = 1) and/or surgery (n = 7). On multiple regression, the post-operative day 1 highest vasoactive-inotrope score (OR 1.02; 95%CI 1.06,1.08; p < 0.001), highest lactate (OR 1.2; 95%CI 1.06,1.35; p = 0.003), and lowest base deficit (OR 0.82; 95%CI 0.71,0.94; p = 0.004), CPB time (OR 1.01; 95%CI 1.00,1.02; p = 0.002), and single-ventricle anatomy (OR 5.35; 95%CI 1.66,17.31; p = 0.005) were associated with early ECMO. Outcomes at 2 years old compared between early-ECMO and matched patients were mortality 11/20 (55%) vs 11/40 (28%) (OR 3.22, 95%CI 0.98,10.63; p = 0.054) and General Adaptive Composite median 65 [interquartile range (IQR) 58, 81.5] in 9 survivors vs 93 [IQR 86.5, 102.5] in 29 survivors (p = 0.02). The identified risk factors for, and outcomes after, having early ECMO may aid decision making in the acute period and confirm that neurodevelopmental follow-up for these children is necessary. The hypothesis that earlier institution of ECMO may improve long-term outcomes requires further study.

Perfluoroalkyl acids (PFAAs) within the broader class of per- and polyfluoroalkyl substances (PFAS) are present in human serum as isomer mixtures, but epidemiological studies have yet to address isomer-specific associations with child development and behavior. To examine associations between prenatal exposure to 25 PFAAs, including perfluorooctane sulfonate (PFOS) and perfluorooctanoate (PFOA) isomers, and child neurodevelopment among 490 mother-child pairs in a prospective Canadian birth cohort, the Alberta Pregnancy Outcomes and Nutrition (APrON) study. To consider the influence of a classic neurotoxicant, total mercury (THg), based on its likelihood of co-exposure with PFAAs from common dietary sources. Maternal blood samples were collected in the second trimester and child neurodevelopment was assessed at 2 years of age using the Bayley Scales of Infant and Toddler Development, 3rd Edition (Bayley-III). Linear or curvilinear multiple regression models were used to examine associations between exposures and neurodevelopment outcomes. Select PFAAs were associated with lower Cognitive composite scores, including perfluoroheptanoate (PFHpA) (β = -0.88, 95% confidence interval (CI): -1.7, -0.06) and perfluorododecanoate (PFDoA) (β = -2.0, 95% CI: -3.9, -0.01). Non-linear relationships revealed associations of total PFOS (β = -4.4, 95% CI: -8.3, -0.43), and linear-PFOS (β = -4.0, 95% CI: -7.5, -0.57) and 1m-PFOS (β = -1.8, 95% CI: -3.3, -0.24) isomers with lower Language composite scores. Although there was no effect modification, including THg interaction terms in PFAA models revealed negative associations between perfluorononanoate (PFNA) and Motor (β = -3.3, 95% CI: -6.2, -0.33) and Social-Emotional (β = -3.0, 95% CI: -5.6, -0.40) composite scores. These findings reinforce previous reports of adverse effects of maternal PFAA exposure during pregnancy on child neurodevelopment. The unique hazards posed from isomers of PFOS justify isomer-specific analysis in future studies. To control for possible confounding, mercury co-exposure may be considered in studies of PFAAs.