Department of Biostatistics
Statistical Methods in Epidemiology Working Group
2014 - 2015
ABSTRACT: Serum albumin is a leading index of protein-energy malnutrition (PEM) that has been associated with mortality among hemodialysis patients. Studies have found that albumin levels at the start of dialysis and the slope of albumin over time are independent risk factors for mortality. It is also natural to hypothesize that high within-subject variability in albumin measured over time may also be indicative of increased mortality. That is, high instability around a patient's first-order trend is likely an indication of nutritional instability and hence may be a risk factor for morbidity and mortality. We develop a Gaussian process model for estimating a summary measure of within-subject volatility in serum albumin measured over time. The proposed model includes a parameter to allow for subject-to-subject variability and places a Dirichlet process prior on the unknown distribution from which these subject-specific parameters are drawn in order to cluster subjects with similar longitudinal patterns without specifying the number of clusters. Simulation studies that assess the proposed model are presented and an illustrative example is provided where the induced summary measure of within-subject volatility is associated with mortality using patients from the United States Renal Data System.
ABSTRACT: In nutritional epidemiology studies of diet and health often focus separately on single dietary nutrients or foods as they pertain to disease risk. As the bioactive component of foods, nutrients are believed to be the causal compounds related to disease. However, the entirety of health effects of foods is rarely captured by considering a single nutrient. Additionally, considering the effects of a food is difficult as they aren't consumed in isolation. Here we present an alternative approach allowing for a more comprehensive analysis of multiple foods on the risk of disease using a two-stage regression approach. The proposed hierarchical model uses nutrient-level information to "shrink" or pull ordinary estimates of food effects towards or away from one another when a food has a similar nutrient composition. We present results using these models to assess the effects of many different foods on the risk of diabetes or cardiovascular disease using data from the Nurses' Health Study. Research by Kathryn Fitzgerald, Bernard Rosner and Karin Michels.
ABSTRACT: Diagnostic biomarkers are used frequently in epidemiologic and clinical work. The ability of a diagnostic biomarker to discriminate between subjects who develop disease (cases) and subjects who do not (controls) is often measured by the area under the receiver operating characteristic curve (AUC). The diagnostic biomarkers are usually measured with error. Ignoring measurement error can cause biased estimation of AUC, which results in misleading interpretation of the efficacy of a diagnostic biomarker. In this article, we propose a method to correct the AUC for measurement error and derive confidence limits for the corrected AUC. The proposed method does not require the normality assumption for the distributions of diagnostic biomarkers. Both real data analyses and simulation studies show good performance of the proposed measurement error correction method.
ABSTRACT: To access the calibration of a predictive model in a survival analysis setting, several authors have extended the Hosmer and Lemeshow goodness of fit test to survival data. Gronnesby and Borgan developed a test under the assumption of proportional hazards, and D'Agostino and Nam developed a nonparametric test that is applicable in a more general survival setting for data with limited censoring. We analyze the performance of the two tests and show that the Gronnesby-Borgan test attains appropriate size in a variety of settings, whereas the D'Agostino-Nam method has a high Type 1 error when there is more than trivial censoring. Both tests are sensitive to small cell sizes. We develop a modification of the D'Agostino-Nam test to allow for higher censoring rates. We show that this modified D'Agostino-Nam test has appropriate Type 1 error and comparable power to the Gronnesby and Borgan test, and is applicable to settings other than proportional hazards. We prove that in the absence of censoring the proposed method is closely related to Hosmer-Lemeshow test statistic. We also discuss the application to small cell sizes.
|Back to SPH Biostatistics||
Maintained by the
Last Update: March 18, 2015