我正在对使用MICE多次估算的数据运行二元结果变量的逻辑回归 . 汇总glm模型的系数似乎很简单:imp = mice(nhanes2,print = F)

imp$meth

 fit0=with(data=imp, glm(hyp~age, family = binomial))

 fit1=with(data=imp, glm(hyp~age+chl, family = binomial))

 summary(pool(fit1))

但是,我无法想出一种方法来汇集glm生成的其他输出 . 例如,glm函数产生可用于模型测试的AIC,Null偏差和残差偏差 . 池(汇总(fit1))##估算汇总1:

Call:
     glm(formula = hyp ~ age + chl, family = binomial)

    Deviance Residuals: 
     Min       1Q   Median       3Q      Max  
      -1.0117  -0.7095  -0.4862  -0.2169   2.2267  

    Coefficients:
   Estimate Std. Error z value Pr(>|z|)
    (Intercept) -5.69937    3.78119  -1.507    0.132
    age2         1.34014    1.35545   0.989    0.323
    age3         1.55824    1.39266   1.119    0.263
    chl          0.01662    0.01749   0.950    0.342

  (Dispersion parameter for binomial family taken to be 1)

  **Null deviance: 25.020  on 24  degrees of freedom
  Residual deviance: 21.898  on 21  degrees of freedom
  AIC: 29.898**

   Number of Fisher Scoring iterations: 5

我尝试了pool.compare函数,但也无法使用二进制结果变量

pool.compare(fit1,fit0,data = imp,method =“likelihood”)

Error in Summary.factor(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L,  : 
  ‘min’ not meaningful for factors

有没有办法使用MICE使用多重插补数据来完成这些事情(或获得对数似然测试输出),或者有没有办法使用像rms这样的另一个包来处理由MICE生成的MI数据?