Statistical analyses

Patients with MC were compared with non-cases within the total cohort. The associations between dietary factors (divided into quartiles with the first quartile used as the reference) and lifestyle factors and MC risk were estimated through the Cox proportional-hazards regression models with time from baseline until diagnosis, death or end of follow-up as the underlying time variable. The analyses were adjusted for potential confounding factors (age, gender and smoking habits). Analyses with dietary factors were also adjusted for total energy intake, season for inclusion in the study and version of the dietary method. We also tested P for trend, that is, the linear association between lifestyle and dietary factors and MC risk by using these variables as continuous variables (for example, 1–4 for the dietary factors). This analysis tested whether increasing amounts of exposure had a positive or a negative association with MC risk. All dietary parameters were also tested in a multiple logistic regression model but with the same non-significant outcome (data not shown). In a sensitivity analysis, we excluded individuals with a reported dietary change in the past.^{31, 32} In the calculation of the association between smoking, alcohol and MC risk, both smoking and alcohol intake were set as reference. Values are given as hazard ratios (HRs) and 95% confidence interval (CI) or mean and range. A P-value below 0.05 was considered statistically significant. For all statistical analyses, SPSS Version 22 (IBM Corp., Armonk, NY, USA) was used.