||January 28th, 2019
||Kelley Engineering Center (KEC) Room 1001
||This seminar is free and open to the public.
Sample size estimation for calibration studies for variables with heteroscedastic measurement error.
Vitamin D levels, as measured by 25-Hydroxyvitamin D (25(OH)D) concentration in blood, have been associated with clinical and public health outcomes, including bone health, autoimmune diseases, cancer, cardiovascular disease morbidity and mortality, cancer, all-cause mortality. Comparisons across populations and the estimation of adequate levels of circulating levels of vitamin D, have been difficult to establish due in part to the lack of standardized methods of measurement of 25(OH)D. In 2010 the Office of Dietary Supplements/ National Health Institutes of Health (NIH) launched the Vitamin D Standardization Program (VDSP) in an effort to promote accurate and time/location-comparable measurement of 25(OH)D.
One of the objectives of the VDSP is to standardize measured 25(OH)D concentrations in national health surveys and other large epidemiological studies to the National Institute of Standards and Technology (NIST) reference measurement procedure (RMP). The main goal is to develop calibrating equations that transform the existing 25(OH)D levels (test values) into an LC/MS-based approach (reference values) comparable to the NIST RFP. Calibration methods for 25(OH)D and other analytes have been used by the National Center for Health Statistics (NCHS) in the US to update measurements in the National Health and Nutrition Examination Survey (NHANES). Serum samples of test values will be selected from each participating study and re-measured using the reference method.
The number and distribution of the selected serum samples needs to be determined in advance so that the reference values, obtained through the calibration equations, could be estimated with enough precision and statistical power.
Both test and reference measurement procedures yield 25(OH)D values that are subjected to error, making the estimation of a calibration equation more convoluted. In particular, sample size estimation techniques need to be updated to accommodate this error. Linnet (1999) suggests the application of weighted least squares or Deming regression for situations where the error is constant across test values and weighted Deming regression when the error is proportional to the level. Lu, et al (2012) have developed methodology for the estimation, inference and power/sample size calculation that allows for more general assumptions. This paper describes the sample size estimation for each of the studies using the approach proposed by Lu, et al(2012). Further developments to accommodate non-linearity are briefly discussed.
For more information about Ramon Durazo-Aruizo. Click here.