Hi everyone,
Thanks in advance for your help. I am running a cox regression model and ran fractional polynomials on age in the dataset (among the rest of the variables that I want in my regression) and found that FP weight (0.5, 0.5) is best. However, if I drop (age_1 and age_2) and run fp generate ^age (0.5, 0.5), essentially recreating this, about 60K more individuals are in my model (5,507,332 vs. . Why is this? I sometimes need to recreate the FP because my dataset is large and I'm returning to my analysis.
Thank you.
___________________________________________
Model after the original FP :
fp <age>: stcox <age> ib2.bmi_clincat i.fin_gender i.aud_ard_ever i.aud_cat i.smknum hiv_ever tot_panc i.fin_raceeth diabetes_ever hcv_chron cyst currentcancer if age>18 & age<90 & bmi> 14 & bmi <70
(fitting 44 models)
(....10%....20%....30%....40%....50%....60%....70% ....80%....90%....100%)
Fractional polynomial comparisons:
-------------------------------------------------------------------
| Test Deviance
age | df Deviance diff. P Powers
-------------+-----------------------------------------------------
omitted | 4 709982.18 14108.24 0.000
linear | 3 697341.73 1467.79 0.000 1
m = 1 | 2 695957.52 83.58 0.000 -2
m = 2 | 0 695873.94 0.00 -- .5 .5
-------------------------------------------------------------------
Note: Test df is degrees of freedom, and P = P > chi2 is sig. level
for tests comparing models vs. model with m = 2 based on
deviance difference, chi2.
Cox regression with Breslow method for ties
No. of subjects = 5,507,332 Number of obs = 5,507,332
No. of failures = 23,922
Time at risk = 59,940,509
LR chi2(21) = 18687.85
Log likelihood = -347936.97 Prob > chi2 = 0.0000
Model after fp generate:
fp generate age^(0.5, 0.5)
. stcox age_1 age_2 ib2.bmi_clincat i.fin_gender i.aud_ard_ever i.aud_cat i.smknum hiv_ever tot_panc i.fin_raceeth diabetes_ever hcv_chron cyst currentcancer
Failure _d: totpdac_10yr_dp
Analysis time _t: time_10yr
ID variable: scrssn
Cox regression with Breslow method for ties
No. of subjects = 5,565,414 Number of obs = 5,565,414
No. of failures = 14,169
Time at risk = 1.91032e+10
LR chi2(21) = 14107.52
Log likelihood = -212012.11 Prob > chi2 = 0.0000
Thanks in advance for your help. I am running a cox regression model and ran fractional polynomials on age in the dataset (among the rest of the variables that I want in my regression) and found that FP weight (0.5, 0.5) is best. However, if I drop (age_1 and age_2) and run fp generate ^age (0.5, 0.5), essentially recreating this, about 60K more individuals are in my model (5,507,332 vs. . Why is this? I sometimes need to recreate the FP because my dataset is large and I'm returning to my analysis.
Thank you.
___________________________________________
Model after the original FP :
fp <age>: stcox <age> ib2.bmi_clincat i.fin_gender i.aud_ard_ever i.aud_cat i.smknum hiv_ever tot_panc i.fin_raceeth diabetes_ever hcv_chron cyst currentcancer if age>18 & age<90 & bmi> 14 & bmi <70
(fitting 44 models)
(....10%....20%....30%....40%....50%....60%....70% ....80%....90%....100%)
Fractional polynomial comparisons:
-------------------------------------------------------------------
| Test Deviance
age | df Deviance diff. P Powers
-------------+-----------------------------------------------------
omitted | 4 709982.18 14108.24 0.000
linear | 3 697341.73 1467.79 0.000 1
m = 1 | 2 695957.52 83.58 0.000 -2
m = 2 | 0 695873.94 0.00 -- .5 .5
-------------------------------------------------------------------
Note: Test df is degrees of freedom, and P = P > chi2 is sig. level
for tests comparing models vs. model with m = 2 based on
deviance difference, chi2.
Cox regression with Breslow method for ties
No. of subjects = 5,507,332 Number of obs = 5,507,332
No. of failures = 23,922
Time at risk = 59,940,509
LR chi2(21) = 18687.85
Log likelihood = -347936.97 Prob > chi2 = 0.0000
Model after fp generate:
fp generate age^(0.5, 0.5)
. stcox age_1 age_2 ib2.bmi_clincat i.fin_gender i.aud_ard_ever i.aud_cat i.smknum hiv_ever tot_panc i.fin_raceeth diabetes_ever hcv_chron cyst currentcancer
Failure _d: totpdac_10yr_dp
Analysis time _t: time_10yr
ID variable: scrssn
Cox regression with Breslow method for ties
No. of subjects = 5,565,414 Number of obs = 5,565,414
No. of failures = 14,169
Time at risk = 1.91032e+10
LR chi2(21) = 14107.52
Log likelihood = -212012.11 Prob > chi2 = 0.0000
