Article

Averaged or Stratified Measures of Risk Profile Discrimination Horses for Courses

Program in Molecular and Genetic Epidemiology, Harvard School of Public Health, Boston, MA 02115, USA.
Epidemiology (Cambridge, Mass.) (Impact Factor: 6.18). 11/2011; 22(6):813-4. DOI: 10.1097/EDE.0b013e3182320edc
Source: PubMed
0 Followers
 · 
38 Views
  • [Show abstract] [Hide abstract]
    ABSTRACT: In case-control studies evaluating the classification accuracy of a marker, controls are often matched to cases with respect to factors associated with the marker and disease status. In contrast with matching in epidemiologic etiology studies, matching in the classification setting has not been rigorously studied. In this article, we consider the implications of matching in terms of the choice of statistical analysis, efficiency, and assessment of the incremental value of the marker over the matching covariates. We find that adjustment for the matching covariates is essential, as unadjusted summaries of classification accuracy can be biased. In many settings, matching is the most efficient covariate-dependent sampling scheme, and we provide an expression for the optimal matching ratio. However, we also show that matching greatly complicates estimation of the incremental value of the marker. We recommend that matching be carefully considered in the context of these findings.
    Biometrics 04/2008; 64(1):1-9. DOI:10.1111/j.1541-0420.2007.00823.x · 1.52 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Recently, several genome-wide association studies have identified various genetic susceptibility loci for breast cancer. Relatively little is known about the possible interactions between these loci and the established risk factors for breast cancer. To assess interactions between single-nucleotide polymorphisms (SNPs) and established risk factors, we prospectively collected DNA samples and questionnaire data from 8576 breast cancer case subjects and 11 892 control subjects nested within the National Cancer Institute's Breast and Prostate Cancer Cohort Consortium (BPC3). We genotyped 17 germline SNPs (FGFR2-rs2981582, FGFR2-rs3750817, TNRC9-rs3803662, 2q35-rs13387042, MAP3K1-rs889312, 8q24-rs13281615, CASP8-rs1045485, LSP1-rs3817198, COL1A1-rs2075555, COX11-rs6504950, RNF146-rs2180341, 6q25-rs2046210, SLC4A7-rs4973768, NOTCH2-rs11249433, 5p12-rs4415084, 5p12-rs10941679, RAD51L1-rs999737), and odds ratios were estimated by logistic regression to confirm previously reported associations with breast cancer risk. We performed likelihood ratio test to assess interactions between 17 SNPs and nine established risk factors (age at menarche, parity, age at menopause, use of hormone replacement therapy, family history, height, body mass index, smoking status, and alcohol consumption), and a correction for multiple testing of 153 tests (adjusted P value threshold = .05/153 = 3 × 10(-4)) was done. Case-case comparisons were performed for possible differential associations of polymorphisms by subgroups of tumor stage, estrogen and progesterone receptor status, and age at diagnosis. All statistical tests were two-sided. We confirmed the association of 14 SNPs with breast cancer risk (P(trend) = 2.57 × 10(-3) -3.96 × 10(-19)). Three SNPs (LSP1-rs3817198, COL1A1-rs2075555, and RNF146-rs2180341) did not show association with breast cancer risk. After accounting for multiple testing, no statistically significant interactions were detected between the 17 SNPs and the nine risk factors. We also confirmed that SNPs in FGFR2 and TNRC9 were associated with greater risk of estrogen receptor-positive than estrogen receptor-negative breast cancer (P(heterogeneity) = .0016 for FGFR2-rs2981582 and P(heterogeneity) = .0053 for TNRC9-rs3803662). SNP 5p12-rs10941679 was statistically significantly associated with greater risk of progesterone receptor-positive than progesterone receptor-negative breast cancer (P(heterogeneity) = .0028). This study does not support the hypothesis that known common breast cancer susceptibility loci strongly modify the associations between established risk factors and breast cancer.
    CancerSpectrum Knowledge Environment 08/2011; 103(16):1252-63. DOI:10.1093/jnci/djr265 · 15.16 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Epidemiologic methods are well established for investigating the association of a predictor of interest and disease status in the presence of covariates also associated with disease. There is less consensus on how to handle covariates when the goal is to evaluate the increment in prediction performance gained by a new marker when a set of predictors already exists. We distinguish between adjusting for covariates and joint modeling of disease risk in this context. We show that adjustment and joint modeling are distinct concepts, and we describe the specific conditions where they are the same. We also discuss the concept of interaction among variables and describe a notion of interaction that is relevant to prediction performance. We conclude with a discussion of the most appropriate methods for evaluating new biomarkers in the presence of existing predictors.
    Epidemiology (Cambridge, Mass.) 11/2011; 22(6):805-12. DOI:10.1097/EDE.0b013e31823035fb · 6.18 Impact Factor