I am trying to run a rather complex linear mixed model on a longitudinal dataset with ~approx. 2,043 subjects. To do this, I used the "mixed command" and wrote the model:
This is longitudinal data, in which every person has at least 2 time points and up to 6 time points (but some folks have missing data points).
Here:
-fullTandImperfect = the dependent variable (a continuous measure of balance time)
-height = a covariate of no interest
-ev1-ev10 = 10 covariates controlling for population structure (as is common in genetic association studies)
-ib0.rs1611115_A = a SNP (3 genotype groups, coded 0, 1, and 2)
-timesince = the number of years since baseline visit at each measurement (values = 0-6 years)
-agei_group = the age group at the first measurement (coded as 0 = 65-74 year olds; 1 = 75-84 year olds; 2 = 85+ year olds)
-GENDER = where 0 is male & 1 is female
I attached example output to this post (PDF).
I used the allbaselevels command to try and figure out how to interpret, in particular, the interactions.
My primary research questions are:
1) Does genotype group (rs1611115_A) predict change over time in fullTandImperfect score?
2) Do certain genotype (rs1611115_A) / age groups (agei_group) change differently in fullTandImperfect score over time? (e.g., do 85+ year old AA folks decline faster over time than other groups?)
3) Do certain genotype (rs1611115_A) / gender groups (GENDER) change differently in fullTandImperfect score over time? (e.g., do AA males decline faster over time than other groups?)
4) Do certain genotype (rs1611115_A) / age groups (agei_group) / gender groups (GENDER) change differently in fullTandImperfect score over time? (e.g., do 85+ year old AA males decline faster over time than other groups?)
Right now, I am stuck on a few points:
1) Am I correct in my interpretation that I should be looking at the results of the interactions with timesince for all of these questions (which are related to how groups might change differently over time)?
2) Are the interaction results comparing each group to ALL of the other groups? e.g., if in the attached output, the following interaction is "significant" at p = 0.032, (if for ib0.rs1611115_A, 2 = GG genotype and for GENDER, 2 = female), would this translate to being able to say:
"GG genotype females are declining significantly faster in balance time across years since baseline testing (p < 0.032) than ALL of the other genotype/gender combination groups?"
3) Additionally, since the coefficient here is -5.697933, could we say that GG females are declining by -5.697933 per year? (i.e., that the slope for GG females is -5.697933?)
I want to run this model with a number of different SNPs, so right now my primary concern is how to interpret the model and figure out if this is an appropriate way to answer my research questions.
Code:
mixed fullTandImperfect c.height ev1 ev2 ev3 ev4 ev5 ev6 ev7 ev8 ev9 ev10 ib0.rs1611115_A##c.timesince##ib0.agei_group##GENDER, allbaselevels || hhidpn: timesince, variance mle
Here:
-fullTandImperfect = the dependent variable (a continuous measure of balance time)
-height = a covariate of no interest
-ev1-ev10 = 10 covariates controlling for population structure (as is common in genetic association studies)
-ib0.rs1611115_A = a SNP (3 genotype groups, coded 0, 1, and 2)
-timesince = the number of years since baseline visit at each measurement (values = 0-6 years)
-agei_group = the age group at the first measurement (coded as 0 = 65-74 year olds; 1 = 75-84 year olds; 2 = 85+ year olds)
-GENDER = where 0 is male & 1 is female
I attached example output to this post (PDF).
I used the allbaselevels command to try and figure out how to interpret, in particular, the interactions.
My primary research questions are:
1) Does genotype group (rs1611115_A) predict change over time in fullTandImperfect score?
2) Do certain genotype (rs1611115_A) / age groups (agei_group) change differently in fullTandImperfect score over time? (e.g., do 85+ year old AA folks decline faster over time than other groups?)
3) Do certain genotype (rs1611115_A) / gender groups (GENDER) change differently in fullTandImperfect score over time? (e.g., do AA males decline faster over time than other groups?)
4) Do certain genotype (rs1611115_A) / age groups (agei_group) / gender groups (GENDER) change differently in fullTandImperfect score over time? (e.g., do 85+ year old AA males decline faster over time than other groups?)
Right now, I am stuck on a few points:
1) Am I correct in my interpretation that I should be looking at the results of the interactions with timesince for all of these questions (which are related to how groups might change differently over time)?
2) Are the interaction results comparing each group to ALL of the other groups? e.g., if in the attached output, the following interaction is "significant" at p = 0.032, (if for ib0.rs1611115_A, 2 = GG genotype and for GENDER, 2 = female), would this translate to being able to say:
"GG genotype females are declining significantly faster in balance time across years since baseline testing (p < 0.032) than ALL of the other genotype/gender combination groups?"
3) Additionally, since the coefficient here is -5.697933, could we say that GG females are declining by -5.697933 per year? (i.e., that the slope for GG females is -5.697933?)
Code:
rs1611115_A#GENDER#c.timesince 0 1 0 (base) 0 2 0 (base) 1 1 0 (base) 1 2 .3524937 1.224969 0.29 0.774 -2.048402 2.75339 2 1 0 (base) 2 2 -5.697933 2.660539 -2.14 0.032 -10.91249 -.4833723
I want to run this model with a number of different SNPs, so right now my primary concern is how to interpret the model and figure out if this is an appropriate way to answer my research questions.
Comment