Temporal dynamics of prognostic factors in breast cancer survival

Abstract

Objectives: We aimed to explore how relevant factors could have shifting prognostic impacts on mortality among BC patients over time, applying alternative survival model specifications.

Methods: Given data from the Cancer Registry of Norway we followed 36 412 women aged 40 + , breast cancer (BC) diagnosed 2006-2020. We analyzed survival comparing patient's molecular subtype group and BMI, adjusting for age, tumor stage and treatment, considering incidence rates for BC specific death (BCSD) and other causes of death (OCOD). We explored cause-specific survival models with and without time invariant coefficients.

Results: Altogether 8.6% (2 542 patients) and 7.0% (3 125 patients) died from other causes and BC throughout the study period, respectively.Molecular subtype HR + /HER2-patients had similar BCSD and OCOD incidence rates. For the other patients the BCSD incidence was higher than the OCOD incidence, especially for HR-/HER2+ and HR-/HER2- patients. Allowing for the association between time to BCSD and molecular subtype groups and radiation therapy to vary over time we found that HR-/HER2+ and HR-/HER2- patients had hazard ratios of 3.38 and 3.53 compared to HR + /HER2- patients within 3 years following BC diagnosis, and hazard ratios of respectively 0.93 and 1.62 in the subsequent years. The positive effect of radiation therapy on BCSD was highest in the first years following diagnosis and still positive, albeit much smaller, in the later years. We found no difference in BC mortality risk between underweight, normal weight, overweight and obese patients.

Conclusions: Our findings contribute to understanding long-term BC survival. Cause-specific Cox models that assume proportional hazards do not allow for time-varying coefficients. This may lead to overlooked time-varying effects of relevant prognostic factors. Especially with longer time horizons one might draw incorrect conclusions.