Clinically, the dependent variable and the clinical outcome are sometimes not linearly related, and the regression model has an important assumption that the independent variable and the dependent variable are linearly related, so the nonlinear relationship model is limited by regression analysis. Therefore, a better solution is to fit the nonlinear relationship between the independent variable and the dependent variable. Restricted cubic spline (RCS) is one of the most common methods to analyze the nonlinear relationship.
Recently, a fan asked whether ordinal multi-categorical variables can be applied to restricted cube bar graphs? is allowed. We can show the difference of data by grouping visualization.
Need to use the three packages of splines, rms, ggplot2, you must download them first,
first load the package and import our data:
library(ggplot2)
library(splines)
library(rms)
be<-read.csv("E:/r/test/qztp2.csv",sep=',',header=TRUE)
names(be)
Build spline regression
model.spline <- lm(be$Gestational.week ~ rcs(be$HB))#建立样条回归
summary(model.spline)
Pay attention to this P value, less than 0.05 indicates a non-linear relationship.
Use anova(model.spline) to view the same,
and then use ggplot2 to draw the graph.
ggplot(be, aes(HB, Gestational.week)) +
geom_point()+
stat_smooth(method = lm, formula = y ~ rcs(x,4)) ##绘制样条回归拟合效果图
We can also set the cut-off point and draw the picture by ourselves. It feels that there is no software that can set it by itself.
model.spline1 <- lm(be$Gestational.week ~ rcs(be$HB, c(20,100,126,150)))#建立样条回归，设置4个节点
ggplot(be, aes(HB, Gestational.week)) +
geom_point()+
stat_smooth(method = lm, formula = y ~ rcs(x, c(20,100,126,150))) ##绘制样条回归拟合效果图
You can also further beautify the graphics
ggplot(be, aes(HB, Gestational.week,fill=Prenatal.hemorrhage,size=Prenatal.hemorrhage1)) +
geom_point(shape=21,size=4,col="black")+
stat_smooth(method = lm, formula = y ~ rcs(x,4)) ##美化一下图形
Adding that we want to know the impact of the two classification indicators of bleeding and no bleeding on the result Y, we can add a grouping variable to perceive the amount of bleeding from the shade of the color. At present, the confidence intervals of the two variables almost overlap. Whether bleeding should be the result No effect, the aggregate depends on the model index.
ggplot(be, aes(HB, Gestational.week,group=Prenatal.hemorrhage,fill=Prenatal.hemorrhage1)) +
geom_point(shape=21,size=4,col="black")+
stat_smooth(method = lm, formula = y ~ rcs(x,4)) ##分组表示
The colors of the two groups are a bit similar, it is not easy to distinguish, we can also modify it.
More exciting articles are available in the public account: zero-based research
