NHANES - prediction

Dear all,
We are using the ols function from RMS R package for predicting/modelling a biomarker (NHANES database). For the independent variables we are using some clinical relevant variables (with rcs). The trouble is: the calibration curve is ok (almost a perfect line), but the R2 is too low (~ 0.04).

Can anyone, please, suggest any approach, or and explanation?


I assume your model was pre-specified and you did not screen any variables using p-values or other variable selection procedures. In this case the R^2 is probably telling you the right story and the outcome is just difficult to predict with your input variables.


Dear Professor,
Your assumptions were correct. Thank you for your insight.


I am a PhD student from Austria. My group has worked with NHANES data before. I would be interested in the background of your analysis (but not sure if I can be of much help tbh). Would you like to share some more information?