Dear Statalist community,
I was wondering how to interpret, from a cox regression, statistically significant coefficients from degree 1 and degree 2 fractional polynomials.
Particulary, I am interested in how to interpret:
a) two coefficients of the same variable (the deviance test indicated that a fractional polynomial exponentiated to powers -2 and -2 was better than a reduced model with a linear tern) signed negatively
b) One Fractional Polynomial degree 1 covariate exponentiated as ^-0.5
c) Another Fractional Polynomial degree 1 covariate exponentiated as ^-2
About my data set:
Panel data, with subject-year (multiple-record-per-subject) observations measured in discrete units (years), I am modelling the time until first failure with a vector of, mostly, time dependent continuous covariates.
I am interested in testing whether covariates should be included in the model as linear or whether a functional transformation might be advisable, as one of many diagnostics tools for cox models.
I found in the literature (Hosmer et. al. 2010: 145 and Cleves et al. 2010: 182) that, appart from the visual test of plotting martin gale residuals, it was highly advisable to compare deviances from different models through fractional polynomial analysis.
Thanks a lot for your support!
References:
Hosmer et. al. 2010. Applied Survival Analysis. Regression Modeling of Time-to-Event Data. Second Edition.
Cleves et al. 2010. An Introduction to Survival Analysis Using Stata. Third Edition
I was wondering how to interpret, from a cox regression, statistically significant coefficients from degree 1 and degree 2 fractional polynomials.
Particulary, I am interested in how to interpret:
a) two coefficients of the same variable (the deviance test indicated that a fractional polynomial exponentiated to powers -2 and -2 was better than a reduced model with a linear tern) signed negatively
b) One Fractional Polynomial degree 1 covariate exponentiated as ^-0.5
c) Another Fractional Polynomial degree 1 covariate exponentiated as ^-2
About my data set:
Panel data, with subject-year (multiple-record-per-subject) observations measured in discrete units (years), I am modelling the time until first failure with a vector of, mostly, time dependent continuous covariates.
I am interested in testing whether covariates should be included in the model as linear or whether a functional transformation might be advisable, as one of many diagnostics tools for cox models.
I found in the literature (Hosmer et. al. 2010: 145 and Cleves et al. 2010: 182) that, appart from the visual test of plotting martin gale residuals, it was highly advisable to compare deviances from different models through fractional polynomial analysis.
Thanks a lot for your support!
References:
Hosmer et. al. 2010. Applied Survival Analysis. Regression Modeling of Time-to-Event Data. Second Edition.
Cleves et al. 2010. An Introduction to Survival Analysis Using Stata. Third Edition
