This study, designed to address the issue confounding by indication  and reverse causality, found that pre-diagnostic vitamin D supplementation has little effect on survival in women with one of four major cancers. Furthermore, our results highlight the need for caution in interpreting observational data of vitamin D supplementation and cancer survival, given the marked protective effect on mortality seen in women with breast cancer who are prescribed supplements, compared with those never prescribed supplements.
A high validity of using cancer diagnoses as recorded in the GPRD has previously been reported . Even if some cases of cancers were omitted erroneously from the dataset, it is unlikely that this would introduce any selection bias into the study, since the association between vitamin D supplementation and mortality is unlikely to differ between those included and those excluded. In the validation study, the median time between diagnosis in GPRD and in the cancer registry data was 11 days , suggesting that our sensitivity analysis of excluding a full year prior to the date of diagnosis would be sufficiently sensitive.
A systematic review of the validity of reporting in GPRD found just one validation study, on sudden death, which was well reported . Potential under-ascertainment of outcome may have diluted our effect, but is unlikely to have been an important source of bias in this study.
We were limited by our measure of supplementation; in particular, we had no information on vitamin D bought over the counter. Some women who had not received any vitamin D prescription may have bought vitamin D. However, it is unlikely that women who had received a prescription would instead buy vitamin D, since prescriptions for women are free after age 60 years. Therefore, this should have had minimal impact on our a priori results. In CPRD, dosage is reported, but instructions for use are not complete, precluding a calculation of average daily dose, or equivalent. We have previously reported a moderate degree of correlation between number of prescriptions and duration of intake (r2 = 0.66, p < 0.01) . Any exposure misclassification is likely to be non-differential, and therefore have diluted our results towards the null effect, rather than to have caused measurement bias.
We also do not have any information on adherence to vitamin D supplementation in the UK. Previous studies have shown that among elderly female hip fracture patients, compliance to recommended supplements was low (28.9 %), but that it can be increased through written recommendations in the hospital discharge letter . In the UK, current guidelines include recommendations to improve vitamin D access for women over the age of 65, including free prescriptions to women over 60. These guidelines may keep non-adherece to a lower level.
We were unable to adjust for key clinical determinants of survival, such as stage of disease. There is some evidence that low levels of vitamin D may be associated with faster progression of cancer . For example, in the Health, Eating, Activity, and Lifestyle study, stage of disease predicted vitamin D levels, independent of other potential confounders .
Other studies have demonstrated that vitamin D levels may be related to adverse prognostic indicators, such as tumour size (but not grade)  and hormone receptor profiles with poorer prognosis (but not tumour size or invasiveness) . Although we could not test this, it seems plausible that the lack of adjustment of key prognostic markers will not have affected our inferences of an effect of vitamin D on survival to a strong degree. Given that our exposure and reference groups differed by discontinuation rather than by initiation of supplementation, it seems unlikely that the two groups would differ by key determinants of survival. Indeed, this has been empirically shown in a study of ovarian cancer .
It is worth noting that most studies looking at cancer progression used blood levels of vitamin D as its measure, and might not be directly comparable to our study using vitamin D prescriptions. Vitamin D supplementation might not correlate with serum levels of vitamin D, since sun exposure and intake of vitamin D food sources affects vitamin D serum levels. Nevertheless, vitamin D from sun exposure is limited in the UK, as much of the country is situated above the latitude that permits optimal vitamin D synthesis, particularly during fall and winter. The elderly, such as the women in our study population, also spend relatively large amounts of time indoors, have reduced dermal capacity to synthesize vitamin D and were more likely to use sun protection when outdoors.
More comparable clinical trials have shown inconsistent results and it remains unclear whether the post-diagnostic supplementation of patients with cancer can improve survival. Three trials (summarised in ) of vitamin D supplementation in men with prostate cancer provided conflicting results; after the promsing ASCENT trial, the ASCENT-II trial was stopped early, due to a higher rate of death in the supplemented group. Ongoing trials are evaluating the role that vitamin D may play on survival in patients with metastatic breast cancer, chronic lymphoid leukaemia and melanoma . Moreover, further research is required for other outcomes. For example, initial results show a possible role vitamin D supplementation may play in reducing aromatase inhibitor-induced joint symptoms  and loss of bone density  in women with breast cancer.