Abstract
Purpose of Review
To illustrate the utility of quantile regression in epidemiology for outcomes that are continuous and when exposure effects may differ across the distribution of the outcome. Linear regression methods estimate only the effects at the mean level which may be an incomplete and biased summary of the effect of exposures for some continuous health outcomes.
Recent Findings
There are several variations of the quantile regression method including classical linear quantile regression, nonparametric quantile regression for growth trajectories, and the modified quantile regression for case–control designs. Such methods offer several applications including (1) the use of quantile regression to test whether the effects of exposure are similar across quantiles, (2) the use of quantile regression for risk prediction, and (3) the use of quantile regression to examine the effects of growth trajectories over time.
Summary
Quantile regression is an important tool for understanding continuous health outcomes, especially outcomes that are not normally distributed, as it offers insight into the relation of exposures with respect to the distribution of the outcome. Quantile regression methods have the potential to deepen and expand the existing quantitative evidence from more common mean-based analyses.
Similar content being viewed by others
References
Papers of particular interest, published recently, have been highlighted as: • Of importance
Yang J, Loos RJ, Powell JE, Medland SE, Speliotes EK, Chasman DI, et al. FTO genotype is associated with phenotypic variability of body mass index. Nature. 2012;490(7419):267–72. https://doi.org/10.1038/nature11401.
Frayling TM, Timpson NJ, Weedon MN, Zeggini E, Freathy RM, Lindgren CM, et al. A common variant in the FTO gene is associated with body mass index and predisposes to childhood and adult obesity. Science. 2007;316(5826):889–94. https://doi.org/10.1126/science.1141634.
Koenker R, Bassett G Jr. Regression quantiles. Econometrica. 1978;46:33–50.
Zhao Z, Xiao Z. Efficient regressions via optimally combining quantile information. Economic Theory. 2014;30(6):1272–314. https://doi.org/10.1017/s0266466614000176.
Gutenbrunner C, Jurečková J, Koenker R, Portnoy S. Tests of linear hypotheses based on regression rank scores. J Nonparametr Stat. 1993;2(4):307–31.
He X, Hu F. Markov chain marginal bootstrap. J Am Stat Assoc. 2002;97(459):783–95.
Feng X, He X, Hu J. Wild bootstrap for quantile regression. Biometrika. 2011;98(4):995–9.
Kocherginsky M, He X, Mu Y. Practical confidence intervals for regression quantiles. J Comput Graph Stat. 2005;14(1):41–55.
Hjartåker A, Langseth H, Weiderpass E. Obesity and diabetes epidemics. In: Innovative Endocrinology of Cancer: Springer; 2008. p. 72–93.
Terry MB, Wei Y, Esserman D. Maternal, birth, and early-life influences on adult body size in women. Am J Epidemiol. 2007;166(1):5–13. https://doi.org/10.1093/aje/kwm094.
Koenker RW, D’Orey V, Algorithm AS. 229: computing regression quantiles. J R Stat Soc: Ser C: Appl Stat. 1987;36(3):383–93. https://doi.org/10.2307/2347802.
Koenker R, d’Orey V, Remark AS. R92: a remark on algorithm AS 229: computing dual regression quantiles and regression rank scores. J R Stat Soc: Ser C: Appl Stat. 1994;43(2):410–4.
Wei Y, Pere A, Koenker R, He X. Quantile regression methods for reference growth charts. Stat Med. 2006;25(8):1369–82.
Wei Y, He X. Conditional growth charts. Ann Stat. 2006;34(5):2069–97.
• Wei Y, Ma X, Liu X, Terry MB. Using time-varying quantile regression approaches to model the influence of prenatal and infant exposures on childhood growth. Biostat Epidemiol. 2017;1(1):133–47. https://doi.org/10.1080/24709360.2017.1358137. This is a paper that shows how to do repeated measures analysis with quantile regression .
Terry MB, Wei Y, Esserman D, McKeague IW, Susser E. Pre- and postnatal determinants of childhood body size: cohort and sibling analyses. J Dev Orig Health Dis. 2011;2(2):99–111. https://doi.org/10.1017/s2040174411000067.
• Ester WA, Houghton LC, Lumey LH, Michels KB, Hoek HW, Wei Y, et al. Maternal and early childhood determinants of women’s body size in midlife: overall cohort and sibling analyses. Am J Epidemiol. 2017;185(5):385–94. https://doi.org/10.1093/aje/kww222. This analysis updated quantile-specific results from 2007 showing the association between maternal BMI and gestational weight gain and offspring BMI persists through midlife.
• Briollais L, Durrieu G. Quantile regression for genetic and genomic applications. In: Handbook of quantile regression: Chapman and Hall/CRC; 2017. p. 409–27. This paper is an example of applying quantile regression to genetic data.
Lin D, Zeng D. Proper analysis of secondary phenotype data in case-control association studies. Genet Epidemiol. 2009;33(3):256–65.
Wei Y, Song X, Liu M, Ionita-Laza I, Reibman J. Quantile regression in the secondary analysis of case–control data. J Am Stat Asoc. 2016;111(513):344–54.
Liu M, Rogers L, Cheng Q, Shao Y, Fernandez-Beros ME, Hirschhorn JN, et al. Genetic variants of TSLP and asthma in an admixed urban population. PLoS One. 2011;6(9):e25099.
Song X, Ionita-Laza I, Liu M, Reibman J, We Y. A general and robust framework for secondary traits analysis. Genetics. 2016;202:1329–43.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of Interest
The authors declare that they have no potential conflicts of interest.
Human and Animal Rights and Informed Consent
This article does not contain any studies with human or animal subjects performed by any of the authors.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This article is part of the Topical Collection on Epidemiologic Methods
Appendix
Appendix
Basic Quantile Regression Syntax in R
> install.packages(“quantreg”).
> library(quantreg).
> fit = rq(y~×1 + ×2, tau = .5, data = data).
Note: tau is the quantile level(s) of interest. It could a single value for a fixed quantile level, or a vector of quantile levels, tau = c(0.25, 0.5, 0.75). The function rq() will return regression quantiles from multiple quantiles. If tau is smaller than 0 or larger than 1, the function will return the entire quantile process.
Basic Quantile Regression Syntax in SAS
PROC QUANTREG.
DATA = sas-data-set;
CLASS X1;
MODEL Y = X1 X2 / QUANTILE = 0.25 0.5 0.75;
RUN;
Note: if the option QUANTILE = ALL, it returns the entire quantile process. Same as in R, the default value is 0.5, corresponding to the median.
Statistical Inference of Quantile Regression in R and SAS
To obtain statistical inference of quantile regression in R, we need to use the function summary.rq(object, se = “nid”, ...), where object is the returned object from the function rq(), and the parameter se specify the inference methods. In SAS, the inference options are specified at the PROC QUANTREG Statement following the syntax “PROC QUANTREG CI= <NONE|RANK|...> ALPHA = value ;” where ALPHA is the significance level, and CI specifies the choice of inference. The table below lists the available methods in R and SAS.
Options | |||
---|---|---|---|
Inference method | Subcategories | R | SAS |
Direct | i.i.d. model n.i.d. model | se =” iid” se = “nid” | CI = SPARCITY/IID CI = SPARCITY |
Rank Score | se =” rank” | CI = RANK | |
resampling | Pairwise | se =” boot”, bsmethod = “xy” | Not available |
Parzen, Wei and Ying | se =” boot”, bsmethod = “pxy” | Not available | |
MCMB | se =” boot”, bsmethod = “mcmb” | CI = RESAMPLING | |
Wild | se =” boot”, bsmethod = “wild” | Not available |
R script for the nonparametric quantile regression for growth trajectories with B-spline approximation
Rights and permissions
About this article
Cite this article
Wei, Y., Kehm, R.D., Goldberg, M. et al. Applications for Quantile Regression in Epidemiology. Curr Epidemiol Rep 6, 191–199 (2019). https://doi.org/10.1007/s40471-019-00204-6
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40471-019-00204-6