Hi,

I aim to perform an independent external validation of a published cox prediction model. From the published cox prediction model, I have access to Level 2 information (i.e. Cox model yielding PI, i.e. a set of regression coefficients plus risk groups plus Kaplan-Meier curves, Ref(1))


According to Ref(2), I shall technically be performing a partial validation ("7 Partial validation", page 286). Ref(2) clearly demonstrates, in Stata code, how to perform that if one has access to both derivation and validation data, which is usually not the case for an independent external validation study. However it is not clear to me how I can code Stata to perform Steps 1 and 2* with published regression coefficients and cutpoints on the PI (prognostic index) without using the individual derivation data.The steps in Ref(2) apply if one has access to both development and validation data.

Could anyone kindly point me to Stata code that can be used to do that, or kindly show me how one can "manually" perform steps 1 & 2 with published regression coefficients please?




=================
*THE STEPS on Ref(2):
=================
Step 1:
Fit the model on the derivation data, and predict the PI on both datasets. Keep only the validation data together with the predicted PI.

use pbc, clear
stcox x1 x2 x3 if val==0
predict xb, xb

Step 2:
Define 3 prognostic groups from the 25th and 75th centiles of xb in the derivation dataset. (This is done on events because the number of events gets very small in the Good prognostic group.)

centile xb if val==0 & _d==1, centile(25 75)
generate byte group = cond(xb <= r(c_1), 1, cond(xb >= r(c_2), 3, 2))


===========
REFERENCES
===========
Ref(1): External validation of a Cox prognostic model: principles and methods, Royston P, Altman, D. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3667097/

Ref(2) : Tools for checking calibration of a Cox model in external validation: Prediction of population-averaged survival curves based on risk groups, Royston P. Stata Journal