Spline transformation (predictor) in the logistic regression

[Rlearner123]

I am working on a regression model and hoping to incorporate spline transformation for my continuous predictor (pre-score). However, I am struggling with incorporating the spline transformation into the logistic regression (using disease as an outcome). Any advice on how I should fix my code below is greatly appreciated.

Here is how I attempted my spline transformed logistic regression:

knots <- quantile(data_B$Pre.score, p = c(0.25, 0.5, 0.75))
# Build the model
library(splines)
model <- glm (disease ~ bs(Pre.score, knots = knots), data = data_B, family=binomial)
# Make predictions
pred.val.2 <- predict(model, type ="response")
# Model performance
library(caret)
data.frame(
  RMSE = RMSE(pred.val.2, data_B$disease),
  R2 = R2(pred.val.2, data_B$disease)
)

Here is my data:

structure(list(Pre.score = c(15.87301587, 5.310939628, 
5.707491082, 3.089700997, 41.27569847, 4.200567644, 14.30503889, 
6.699928724, 4.148471616, 3.70212766, 19.41605839, 11.99368753, 
5.991232343, 4.42804428, 20.77562327, 3.432367595, 1.1574886, 
4.186655037, 29.27974948, 0.874453467, 13.66459627, 0.564652739, 
13.06160259, 19.66829268, 0, 8.623969562, 13.18289786, 3.505734689, 
4.769475358, 0.208768267, 1.23683005, 11.11934766, 0.567000567, 
4.077429984, 1.179835538, 2.582781457, 18.62888102, 0.540151242, 
9.014084507, 2.714285714, 15.17327505, 17.41935484, 11.01306036, 
13.53987378, 7.38645816, 18.69688385, 14.6287403), 
    disease = structure(c(1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 1L, 
    1L, 1L, 2L, 1L, 2L, 2L, 2L, 1L, 1L, 2L, 1L, 2L, 1L, 2L, 2L, 
    1L, 2L, 2L, 2L, 2L, 1L, 2L, 2L, 1L, 2L, 1L, 1L, 2L, 1L, 1L, 
    1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L), .Label = c("none", "disease"
    ), class = "factor")), row.names = c(NA, 47L), class = "data.frame")

[Max]

Can you please provide a minimal reprex (reproducible example)? The goal of a reprex is to make it as easy as possible for me to recreate your problem so that I can fix it: please help me help you!

If you've never heard of a reprex before, start by reading "What is a reprex", and follow the advice further down that page.

For example, there are multiple packages with RMSE functions etc.

Also, you appear to be fitting a classification model but evaluating the model using regression performance metrics.

[Rlearner123]

Thanks, Max! I have updated (e.g., specify which package I used) my original post. Does it help you better understand my question?

Also, can you pls elaborate on your last comment re: "you appear to be fitting a classification model but evaluating the model using regression performance metrics"? What do you mean?

[Max]

@Max

you appear to be fitting a classification model but evaluating the model using regression performance metrics.

So you are fitting a logistic regression model for classifification.

RMSE and R² are performance metrics for cases when the outcome data is quantitative. There are a lot of metrics that are more appropriate, such as the area under the ROC curve, the binomial log likelihood, and (to a lesser extent) classification accuracy.

[Rlearner123]

Yes, I realized that as well! I will fix it. However, how about the other parts of my R code? For example, are thess parts correctly entered?

# Build the model
library(splines)
model <- glm (disease ~ bs(Pre.score, knots = knots), data = data_B, family=binomial)
# Make predictions
pred.val.2 <- predict(model, type ="response")

[Rlearner123]

Hi Max - I am following up re: the spline transformation.
Do you think there is anything incorrect about the following code:

model <- glm (disease ~ bs(percentage, knots = knots), data = data_B, family=binomial)

Below is the output for another predictor (percentage) from the above code. Can you pls help me intepret it?

Coefficients:
                                 Estimate Std. Error z value Pr(>|z|)
(Intercept)                        0.4845     1.1982   0.404    0.686
bs(percentage, knots = knots)1     2.7183     3.3578   0.810    0.418
bs(percentage, knots = knots)2    -1.7655     2.3966  -0.737    0.461
bs(percentage, knots = knots)3     2.1134     3.1838   0.664    0.507
bs(percentage, knots = knots)4    -6.1535    10.9851  -0.560    0.575
bs(percentage, knots = knots)5    -8.1928    56.9180  -0.144    0.886
bs(percentage, knots = knots)6 10636.9163  8119.2865   1.310    0.190

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 61.513  on 46  degrees of freedom
Residual deviance: 51.817  on 40  degrees of freedom
AIC: 65.817

Number of Fisher Scoring iterations: 14

[Max]

I do not know. The code seems syntactically correct but we have no idea about the purpose of this code. We can't tell if you are trying to unscrew something with a butterknife (as an analogy).

Even if the code is correct, the size of the estimate and standard error for the sixth spline parameter is excessive.