Using Median Regression to Obtain Adjusted Estimates of Central Tendency for Skewed Laboratory and Epidemiologic Data
McGreevy, K. M.
Lipsitz, S. R.
Linder, J. A.
Rimm, Eric Bruce::0ab2926c8242f35e5a982e3cf59f4987::600
Hoel, D. G.
MetadataShow full item record
CitationMcGreevy, K. M., S. R. Lipsitz, J. A. Linder, E. Rimm, and D. G. Hoel. 2008. “Using Median Regression to Obtain Adjusted Estimates of Central Tendency for Skewed Laboratory and Epidemiologic Data.” Clinical Chemistry 55 (1): 165–69. https://doi.org/10.1373/clinchem.2008.106260.
AbstractBACKGROUND: Laboratory studies often involve analyses of highly skewed data for which means are not an adequate measure of central tendency because they are sensitive to outliers. Attempts to transform skewed data to symmetry are not always successful, and medians are better measures of central tendency for such skewed distributions. When medians are compared across groups, confounding can be an issue, so there is a need for adjusted medians. Methods: We illustrate the use of quantile regression to obtain adjusted medians. The method is illustrated by use of skewed nutrient data obtained from black and white men attending a prostate cancer screening. For 3 nutrients, saturated fats, caffeine, and vitamin K, we obtained medians adjusted by age, body mass index, and calories for men in each race group. Results: Quantile regression, linear regression, and log-normal regression produced substantially different adjusted estimates of central tendency for saturated fats, caffeine, and vitamin K. ConclUSIONS: Our method was useful for analysis of skewed and other nonnmormally distributed continuous outcome data and for calculation of adjusted medians.
Citable link to this pagehttp://nrs.harvard.edu/urn-3:HUL.InstRepos:41254598
- SPH Scholarly Articles