TY - JOUR
T1 - Statistical Methods for Unusual Count Data
T2 - Examples From Studies of Microchimerism
AU - Guthrie, Katherine A.
AU - Gammill, Hilary S.
AU - Kamper-Jørgensen, Mads
AU - Tjønneland, Anne
AU - Gadi, Vijayakrishna K.
AU - Nelson, J. Lee
AU - Leisenring, Wendy
N1 - © The Author 2016. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: [email protected].
PY - 2016/11/15
Y1 - 2016/11/15
N2 - Natural acquisition of small amounts of foreign cells or DNA, referred to as microchimerism, occurs primarily through maternal-fetal exchange during pregnancy. Microchimerism can persist long-term and has been associated with both beneficial and adverse human health outcomes. Quantitative microchimerism data present challenges for statistical analysis, including a skewed distribution, excess zero values, and occasional large values. Methods for comparing microchimerism levels across groups while controlling for covariates are not well established. We compared statistical models for quantitative microchimerism values, applied to simulated data sets and 2 observed data sets, to make recommendations for analytic practice. Modeling the level of quantitative microchimerism as a rate via Poisson or negative binomial model with the rate of detection defined as a count of microchimerism genome equivalents per total cell equivalents tested utilizes all available data and facilitates a comparison of rates between groups. We found that both the marginalized zero-inflated Poisson model and the negative binomial model can provide unbiased and consistent estimates of the overall association of exposure or study group with microchimerism detection rates. The negative binomial model remains the more accessible of these 2 approaches; thus, we conclude that the negative binomial model may be most appropriate for analyzing quantitative microchimerism data.
AB - Natural acquisition of small amounts of foreign cells or DNA, referred to as microchimerism, occurs primarily through maternal-fetal exchange during pregnancy. Microchimerism can persist long-term and has been associated with both beneficial and adverse human health outcomes. Quantitative microchimerism data present challenges for statistical analysis, including a skewed distribution, excess zero values, and occasional large values. Methods for comparing microchimerism levels across groups while controlling for covariates are not well established. We compared statistical models for quantitative microchimerism values, applied to simulated data sets and 2 observed data sets, to make recommendations for analytic practice. Modeling the level of quantitative microchimerism as a rate via Poisson or negative binomial model with the rate of detection defined as a count of microchimerism genome equivalents per total cell equivalents tested utilizes all available data and facilitates a comparison of rates between groups. We found that both the marginalized zero-inflated Poisson model and the negative binomial model can provide unbiased and consistent estimates of the overall association of exposure or study group with microchimerism detection rates. The negative binomial model remains the more accessible of these 2 approaches; thus, we conclude that the negative binomial model may be most appropriate for analyzing quantitative microchimerism data.
U2 - 10.1093/aje/kww093
DO - 10.1093/aje/kww093
M3 - Journal article
C2 - 27769989
SN - 0002-9262
VL - 184
SP - 779
EP - 786
JO - American Journal of Epidemiology
JF - American Journal of Epidemiology
IS - 10
ER -
