我正在为具有几个二进制预测变量以及一个时间点的二进制结果建立一个混合效应模型。每个对象都有至少一个或多达三个时间点的数据。数据看起来像这样,但是更长:
df <- read.table(text="subject,drug1,drug2,drug3,timepoint,readmit
subj1,1,0,0,time1,1
subj2,0,1,1,time1,0
subj3,1,1,0,time1,1
subj4,0,0,0,time1,0
subj5,0,0,1,time1,0
subj1,1,1,0,time2,1
subj3,1,0,1,time2,0
subj1,0,1,0,time3,0", sep=",", header=TRUE)
## > df
## subject drug1 drug2 drug3 timepoint readmit
## 1 subj1 1 0 0 time1 1
## 2 subj2 0 1 1 time1 0
## 3 subj3 1 1 0 time1 1
## 4 subj4 0 0 0 time1 0
## 5 subj5 0 0 1 time1 0
## 6 subj1 1 1 0 time2 1
## 7 subj3 1 0 1 time2 0
## 8 subj1 0 1 0 time3 0
因此受试者可以获得这3种药物的任意组合。在90天内跟踪它们,以查看是否重新进入(readmit = 1
)。如果是这样,他们将在下一个时间点有另一个条目。我的问题是关于如何在混合模型中包含timepoint
。我计划建立的模型将3种药物作为固定效应随时间推移,并对受试者进行随机拦截:
glmer(readmit ~ drug1 + drug2 + drug3 + timepoint + (1|subject), data=df,
family=binomial, control=glmerControl(optimizer = "bobyqa")
我的问题:在处理重复测量的混合效应模型中如何指定时间点?是否应将其作为随机斜率添加?任何指导将不胜感激-我是这种混合模型的新手。