Comment on: ‘Vitamin D supplements and cancer incidence and mortality: a meta-analysis'

Sir,

We read with deep interest the article by Keum and Giovannucci (2014) entitled ‘Vitamin D supplements and cancer incidence and mortality: a meta-analysis' published in August 2014 in British Journal of Cancer, and believe this is a well-conducted meta-analysis of randomised controlled trials (RCTs) which attempted to elucidate the effects of vitamin D supplementation on total cancer incidence and mortality. The findings of study will have a profound influence on future cancer treatment. Nevertheless, we also found some worthwhile issues worth being discussed.

First, two electronic databases (PubMed and Embase) were systematically searched to identify potential RCTs, and the small number of included articles is likely to have negative effect on credibility of the meta-analysis. To make the study more credible, we hope more electronic databases should be thoroughly retrieved by the authors.

Second, in accordance with recommendations in the Cochrane Handbook (Higgins and Green, 2011), the methodological quality of trials should be independently evaluated by two investigators. However, the authors did not describe this in their meta-analysis. In our opinion, the methodological quality of all eligible trials should be evaluated. The items of the methodological quality assessment should include randomisation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, withdrawals and dropouts. Meanwhile, the specific grades or scores for each included article should be provided according to the Cochrane Collaboration guidelines (Higgins et al, 2011).

Third, the authors reported that the Q-test and I² statistic were used to assess heterogeneity across studies. However, the P and I² values considered significant were not stated in ‘statistical analyses' part. From our own perspective, a P-value <0.10 is considered significant in Q-test (Cochran, 1954) and a I² value of at least 50% is taken as indicator of substantial heterogeneity of outcomes in I² statistic (Higgins and Thompson, 2002). We wish to know the researchers' viewpoint on this issue.

Fourth, the heterogeneity (I²) across studies in Figures 2A and B were both equal to 0.0% (Keum and Giovannucci, 2014). This showed no variations between studies, which could not potentially bias the results of this study. Therefore, it was inappropriate that a random effects model was used to calculate the summary relative risk with 95% confidence interval. As far as we know, the Mantel–Haenszel fixed effects model should be used to calculate the data of total cancer incidence and mortality by authors.

Fifth, we suggest that the type of mortality endpoint (the number of days, hospitalisation or intensive care unit) and gender ratio should be listed in Table 1 of main characteristics of the RCTs included (Keum and Giovannucci, 2014), which may provide specific and important guidance for clinical treatment.

Last but not least, the investigators conducted the publication bias assessment using Egger's test and Begg's test, we advise that the visual funnel plots should be provided. The search was limited to English articles, in our opinion, the language bias should be clarified as a limitation of this meta-analysis in ‘Discussion' part. The ‘Avenell, 2011' in Table 1 and Figure 2 should be replaced by ‘Avenell, 2012', which is consistent with the ‘Discussion' and ‘References' part (Keum and Giovannucci, 2014), please check this point.

All in all, we respect the great contributions of the researchers, they reached an important conclusion that over 2–7 years of duration, the benefit of vitamin D supplementation may be limited to cancer mortality. We agree on the above conclusion of the authors, which has an important value to instruct clinical therapy for cancer. The conclusion will probably not be changed because of our comments. However, we believe that our remarks will further improve the above conclusion and contribute to more accurate elaboration of the results presented by Keum and Giovannucci (2014).