R语言Logistic回归

病例对照研究
队列研究

单因素logistic回归分析

单个分类变量的logistic

点击下载

处理数据

y <- ifelse(expr$YTHDC1>median(expr$YTHDC1),1,0)
df <- data.frame(y=y,x=expr$stage,YTHDC1 = expr$YTHDC1)

建立模型

logistic <- glm(y~x,family = binomial(link="logit"),data = df)
summ <- summary(logistic)

Call:
glm(formula = y ~ x, family = binomial(link = "logit"), data = df)

Deviance Residuals: 
     Min        1Q    Median        3Q       Max  
-1.23169  -1.16396  -0.00977   1.19092   1.21159  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)
(Intercept)   0.1268     0.2255   0.562    0.574
xStage II    -0.2068     0.2715  -0.762    0.446
xStage III   -0.0795     0.2870  -0.277    0.782
xStage IV    -0.1585     0.3382  -0.469    0.639

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 615.51  on 443  degrees of freedom
Residual deviance: 614.83  on 440  degrees of freedom
AIC: 622.83

Number of Fisher Scoring iterations: 3

查看系数

coefficients(logistic)

使用模型预测

predict_=predict.glm(logistic,type="response",newdata=df)
table(predict_)
my_predict <- function(x2=0,x3=0,x4=0){
  a <- 0.12675171+ (x2* -0.20679441)+ (x3* -0.07949882) +(x4* -0.15850040)
  res <- exp(a)/(exp(a)+1)
  return(res)
}
my_predict()
my_predict(x2=1)
my_predict(x3=1)
my_predict(x4=1)

图片alt

如果观测到某个分类变量，该分类变量的取值为1，其余虚拟变量的取值为0

计算OR值

conf <- confint(logistic,level = 0.95)
cbind(OR=exp(summ$coefficients[,1]),
      OR.95L=exp(conf[,1]),
      OR.95H=exp(conf[,2]),
      p=summ$coefficients[,4])

多因素logistic回归分析

参考

• How to Handle R Warning: glm.fit: algorithm did not converge