Logistic 回归的假设

• 假设1：因变量唯一，且为有序多分类变量
• 假设2：存在一个或多个自变量，可为连续、有序多分类或无序分类变量。
• 假设3：自变量之间无多重共线性。
• 假设4：模型满足“比例优势”假设。

选择自变量

paste0('"',paste(colnames(data$training_set), collapse = '", "'),'"')
str(data$training_set$INFARCT_RELATED_VESSEL) # 确定标记为了无序多分类  factor(ordered = F)
independent_variable = c(
    "SEX", "AGE", "HEIGHT", "WEIGHT", "BMI", "BSA", "HR", "SBP", "DBP", 
)

单因素回归

数据归一化

• 方式1

normalize <- function(x, mean_, sd_) {
    return ((x - mean_) / sd_)
}
res = data.frame()
for (iv in independent_variable){
    res[iv, 'VarName'] = iv
    if(is.factor(data$training_set[[iv]])){
        res[iv, 'mean'] = 0
        res[iv, 'sd'] = 0
    }else{
        res[iv, 'mean'] = mean(data$training_set[[iv]], na.rm = T)
        res[iv, 'sd'] = sd(data$training_set[[iv]], na.rm = T)
        data$training_set[[iv]] = normalize(data$training_set[[iv]], res[iv, 'mean'], res[iv, 'sd'])
    }
}
res
readr::write_csv(x = res, file = '../../data/normalize_train.csv')

• 方式2

normalize <- function(x, min_, max_) {
    return ((x - min_) / (max_ - min_))
}
res = data.frame()
for (iv in independent_variable){
    res[iv, 'VarName'] = iv
    if(is.factor(data$training_set[[iv]])){
        res[iv, 'min'] = 0
        res[iv, 'max'] = 0
    }else{
        res[iv, 'min'] = min(data$training_set[iv], na.rm = T)
        res[iv, 'max'] = max(data$training_set[iv], na.rm = T)
        data$training_set[[iv]] = normalize(data$training_set[[iv]], res[iv, 'min'], res[iv, 'max'])
    }
}
res
readr::write_csv(x = res, file = '../../data/normalize_train.csv')

Logistic 回归

res = data.frame()
for (iv in independent_variable){
    lmf <- formula(paste0("IMH~", iv))
    m <- glm(lmf, family=binomial(link="logit"), data=data$training_set)
    vns = names(coef(m))
    vns = vns[2:length(vns)] # 删去常数项
    for (vn in vns){
        res[vn, 'VarName'] = iv
        res[vn, 'VarValue'] = vn
        res[vn, 'mean'] = exp(coef(m)[vn]) # to OR
        res[vn, 'lower'] = exp(confint(m)[vn,1])
        res[vn, 'upper'] = exp(confint(m)[vn,2])
        res[vn, 'Pvalue'] = summary(m)$coefficients[vn, "Pr(>|z|)"]
    }
}
res
readr::write_csv(x = res, file = '../../data/test_glm_train.csv')

附加 线性回归

res = data.frame()
for (iv in independent_variable){
    lmf <- formula(paste0("HEMO_VOLUME~", iv))
    m <- lm(lmf, data$training_set)
    vns = names(coef(m))
    vns = vns[2:length(vns)] # 删去常数项
    for (vn in vns){
        res[vn, 'VarName'] = iv
        res[vn, 'VarValue'] = vn
        res[vn, 'mean'] = coef(m)[vn]
        res[vn, 'lower'] = confint(m)[vn,1]
        res[vn, 'upper'] = confint(m)[vn,2]
        res[vn, 'Pvalue'] = summary(m)$coefficients[vn, "Pr(>|t|)"]
    }
}
res
readr::write_csv(x = res, file = '../../data/test_lm_train.csv')

多重共线性检测

qr(as.matrix(data$training_set[independent_variable]))$rank == length(independent_variable)
kappa(cor(as.matrix(data$training_set[independent_variable])), exact= TRUE) < 100

缺失值插补

• 更多信息见 mice 包

多因素回归

• 用到再写

绘制森林图

• f_forestplot

df <- readRDS('Logistic.rds')
options(repr.plot.width=8, repr.plot.height=6)
f_forestplot(df, zero = 1)