Karolina Carlsson  presents her master's thesis

Time-dependent Effects of Prognostic Factors in Breast Cancer


Abstract:

Cox regression analysis was used to investigate distant disease-free
survival in 237 breast premenopausal patients with lymph node-negative
breast cancer. The purpose was to examine prognostic factors and
especially to study possible time-dependent effects. Of interest was
also to investigate the effect of age at diagnosis on breast cancer
survival.  The results suggest strongly that factors exhibit
time-dependent effects where the hazard ratio is stronger the first
years after diagnosis but decreases with analysis time. This pattern
can be captured by including an interaction between a fixed effect and
a logarithmic function of analysis time in the Cox model. Furthermore
the analysis showed that young age at diagnosis is an adverse
prognostic factor.  A final model with age, S-phase fraction, uPA and
size, the latter three modelled as time-dependent, were evaluated and
was shown to be able to predict risk groups later in survival time
better than a model without time dependency.