Study population

The HPFS is an ongoing prospective cohort of 51,529 male health professionals who were ages 40–75 at enrollment in 1986. Participants responded to a baseline questionnaire concerning medical history, medications, lifestyle, and diet, and have since completed follow-up questionnaires biennially. Average follow-up rates exceed 90%.

Primary analyses were restricted to individuals diagnosed with PCa after return of the 1994 follow-up questionnaire (n = 5695) so that we were able to address reverse causation by lagging exposure 4–6 years (beginning in 1990) and address confounding by adjusting for behaviours 8–10 years prior to diagnosis (beginning in 1986) (Fig. 1). (In the first post-diagnostic follow-up windows, exposure was measured in the years leading up to diagnosis.) We excluded 333 individuals diagnosed with any cancer (other than non-melanoma skin cancer) prior to 1994 and 17 individuals known to have died but whose date of death was unavailable. Among the remaining participants with PCa, we excluded 166 diagnosed with T3b or higher disease and 597 with unknown stage at diagnosis, such that the remaining 4582 men were diagnosed with stage T3a or lower disease. We further restricted analyses of each score to individuals with values of the score 4–6 years and 8–10 years prior to diagnosis. The number excluded due to missing data was specific to each score based on participants having responded to questionnaire items corresponding to score components (2021 PCa Behaviour Score: 220; 2015 PCa Behaviour Score: 262; WCRF/AICR Score: 64; ACS Score: 77).

The analyses are depicted for a hypothetical individual diagnosed in 1997, who died from prostate cancer in 2013. Dietary information in years lacking a food frequency questionnaire was carried forward from prior years. The exception was the transition from pre- to post-diagnosis; if the first questionnaire administered after diagnosis did not include a food frequency questionnaire, then dietary data were considered missing. The analyses corresponding to Table 3 and Supplementary Table 2 started follow-up at the time of diagnosis and lagged exposure by 4–6 years. The analyses corresponding to Supplementary Table 3 started follow-up at the time of diagnosis and did not lag exposure. The analyses corresponding to Supplementary Table 4 started follow-up at the time of first post-diagnosis questionnaire with all score components and anchored exposure at first post-diagnostic health behaviour scores without updating. The analyses corresponding to Supplementary Table 5 started follow-up 4–6 years after the first post-diagnosis questionnaire with all score components and lagged exposure 4–6 years.