0

What I would like to do

I would like to reconstruct a logistic regression model with splines (Lymph Node Involvement (Cores)) using published coefficients and spline knots. All sources that I posted here are from: https://www.mskcc.org/nomograms/prostate/pre_op/coefficients

Why I would like to do that

My aim is to reconstruct the model "Lymph node involment (Cores)" in R, so that I can apply it on a large number of patients of a clinical study (without having to type all the data in an online calculator for every subject) and predict their probability of Lymph node involvement.

The published information about the model

The model definition is as follows:

enter image description here

The restricted cubic spline terms are as follows:

enter image description here enter image description here

The intercept and coefficients are as follow: enter image description here

Question

  1. Is it possible to reconstruct this model with the published data
  2. If yes, how can this be achieved using R

What I have already done

I found following source, however, it is slightly different (I do not want to change anything of the model) Reconstructing a logistic regression model from literature using published coefficients

I understand that I somehow have to reconstruct the model with something like:

#reconstruct model
copylogit <- ...

And then apply it to my data:

#make test data
newdata <-data.frame(age=as.numeric(80),psa=as.numeric(10),gleason_grade=as.factor(4),clinical_stage=as.character("2A"), no_of_positive_cores=as.numeric(2),no_of_negativ_cores=as.numeric(10))

#apply model to test data
predict(copylogit, newdata = newdata, type = "response")

And the expected result for this example would be: 17%

Update: Since I am not sure if it is possible to reconstruct the model with the published data I thought about to generate a prediction equation so that I can calculate the probabilites of lymph node involvement in a larger dataset. Therefore, I opened a new question: Calculate spline terms of a logistic regression using published knots and formula

captcoma
  • 155
  • 13
  • I see the formulation of a task, 'reconstruct a prediction curve based on published coefficients', but I believe that you missed stating your question related to it. What is the question? – Sextus Empiricus Oct 12 '20 at 07:26
  • Pardon me! I added the questions – captcoma Oct 12 '20 at 07:46
  • 1
    What do you mean by "reconstruct"? Is it different from "to generate a prediction equation" and what you did in the other question? – Sextus Empiricus Oct 12 '20 at 15:39
  • Thx for your patience. It is actually my goal to compare the model's performance with another model's performance in a validation cohort. To do so I wanted to calculate the probabilities for both models and compare them using the c-index. Since I learned that this is a power procedure, I thought about to compare the models using a likelihood test. With my very limited understanding I thought that I could re-fit both models to my data (this is what I meant by reconstruct) and then compare both models (i.e. glm objects) in R. If this makes any sense (I doubt that), I can add it to the question. – captcoma Oct 14 '20 at 22:40

0 Answers0