Prospective and retrospective studies of vitamin D levels and breast cancer have produced discrepant results. This may be due to variations in serum 25-hydroxyvitamin D (25(OH)D) concentrations over time, including systematic changes after breast cancer diagnosis. We measured total serum 25(OH)D levels in participants from the Sister Study, a US cohort study of sisters of breast cancer patients, who provided samples at baseline (2003-2009) and 4-10 years later (2013-2015). This included 827 women with an intervening breast cancer and 771 women without one. Although 25(OH)D levels were modestly correlated over time (R = 0.42), 25(OH)D concentrations increased in both groups, with larger increases among cases (averaging 31.6 ng/mL at baseline and 43.5 ng/mL at follow-up) than among controls (32.3 ng/mL at baseline, 40.4 ng/mL at follow-up). Consequently, the estimated association between 25(OH)D and breast cancer depended on whether baseline measurements (per 10-ng/mL increase, odds ratio = 0.87, 95% confidence interval: 0.78, 0.98) or measurements from the second blood draw (per 10-ng/mL increase, odds ratio = 1.17, 95% confidence interval: 1.08, 1.26) were used. Concentrations were related to regular use (≥4 times/week) of vitamin D supplements, which became more common over time; increases in regular use were greater in cases (from 56% to 84%) than in controls (from 56% to 77%). Our results do not explain previously observed differences between retrospective and prospective studies, but they do demonstrate how reverse causation and temporal trends in exposure can distort inference.
Keywords: 25-hydroxyvitamin D; breast cancer; reliability; reverse causation bias; vitamin D.
Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health 2019.