Purpose: Diet has been postulated to influence breast cancer prognosis; however, existing evidence is weak and inconsistent. Previous studies have sought evidence of a linear relationship between diet and breast cancer outcomes. Because of a U-shaped association of body mass index (BMI) with survival in breast cancer, we hypothesized that a nonlinear association also existed for dietary variables.
Patients and Methods: Four hundred seventy-seven women with surgically resected T1 to T3, N0/1, M0 breast cancer completed the Block Food Frequency Questionnaire 9.3 ± 4.6 weeks (mean ± standard deviation) after diagnosis, reporting intake over the preceding 12 months. Data on tumor-related factors, treatment, and outcomes were obtained prospectively from medical records. A series of Cox models was performed, modeling the association of dietary factors with breast cancer survival linearly and quadratically, adjusting for total energy intake, tumor- and treatment-related variables, and BMI.
Results: Significant nonlinear survival associations were found for protein, oleic acid, cholesterol, polyunsaturated-saturated fat ratio, and for percentage of calories from fat and percentage of calories from carbohydrates in multivariate models. The shape of the survival associations varied across nutrients. Hazard ratios for highest risk quintiles ranged from 2.1 to 6.5. For total fat, adjustment for BMI reduced the multivariate P value obtained from nonlinear Cox models from .05 to .10. No significant linear associations were identified.
Conclusion: The association of key dietary variables with breast cancer survival may be U-shaped rather than linear. Our data suggest that midrange intake of most major energy sources is associated with the most favorable outcomes, and extremes are associated with less favorable outcomes.