文献上看到mixed logistic model，查了相关
Conclusions - mixed-effects logistic regression models useful
for incomplete longitudinal dichotomous data
can handle subjects measured incompletely or at different
timepoints (missing data assumed MAR)
degree of within-subjects variation on dichotomous outcome is
important to consider (might have 3-timepoint study where
90% of subjects have same response across timepoints)
subject-specific (or conditional) interpretation of regression
coefficients
generalizations to other categorical outcomes
– ordinal outcomes - mixed-effects ordinal logistic regression
 proportional odds model
 partial or non-proportional odds model
– nominal outcomes - mixed-effects nominal logistic
regression
不知道何解，英文看的晕晕的,不知道怎么通过sas 实现