Reply to the letter to the editor ‘Body mass index and 20-specific cancers—re-analyses of dose–response meta-analyses of observational studies’ by Markozannes et al.

We would like to thank Markozannes et al. for their insightful comments [1.Markozannes G. Kalliala I. Kyrgiou M. Tsilidis KK. Letter to the editor on ‘Body mass index and 20-specific cancers—re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 1490-1491Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar] and respond to these accordingly. First, they described that their umbrella review [2.Kyrgiou M. Kalliala I. Markozannes G. et al.Adiposity and cancer at major anatomical sites: umbrella review of the literature.BMJ. 2017; 356: j477.Crossref PubMed Scopus (420) Google Scholar] used a grading scheme that has been extensively applied and justified in previous studies. However, the methods of an umbrella review or of determining evidence of grading have not been established yet. The article referred to Li et al. [3.Li X. Meng X. Timofeeva M. et al.Serum uric acid levels and multiple health outcomes: umbrella review of evidence from observational studies, randomised controlled trials, and Mendelian randomisation studies.BMJ. 2017; 357: j2376Crossref PubMed Scopus (193) Google Scholar] is based on the opinion of the collaborative group of which Markozannes et al. are part of rather than a grading scheme that has been extensively applied and justified by other independent groups. Second, our grading scheme [4.Choi E.K. Park H.B. Lee K.H. et al.Body mass index and 20-specific cancers: re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 749-757Abstract Full Text Full Text PDF PubMed Scopus (43) Google Scholar] would fail to correctly classify the evidence, because the results of our study were inconsistent with previous assessments by the World Cancer Research Fund and the International Agency for Research on Cancer and theirs in some cancers [1.Markozannes G. Kalliala I. Kyrgiou M. Tsilidis KK. Letter to the editor on ‘Body mass index and 20-specific cancers—re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 1490-1491Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar]. In order to gain an understanding of the discrepancy of results of cancer–obesity association studies, the authors need to carefully analyse previous assessments by their own groups members including Kyrgiou et al.’s study [2.Kyrgiou M. Kalliala I. Markozannes G. et al.Adiposity and cancer at major anatomical sites: umbrella review of the literature.BMJ. 2017; 356: j477.Crossref PubMed Scopus (420) Google Scholar] which analysed the associations of many kinds of obesity parameters with cancers, while ours focussed on dose–response meta-analyses [body mass index (BMI) per 5 kg/m2 increase] and therefore results should be interpreted with caution as results from our study can for example not be directly compared with those of a study reporting on different obesity parameters versus normal. As shown in Kyrgiou et al.'s study [2.Kyrgiou M. Kalliala I. Markozannes G. et al.Adiposity and cancer at major anatomical sites: umbrella review of the literature.BMJ. 2017; 356: j477.Crossref PubMed Scopus (420) Google Scholar], if many kinds of obesity parameters are used, the results of evidence of grading in one cancer type will be different according to the parameter used. Therefore, if some discrepancies exist among the results of evidence grading, one should consider the possible reasons for the discrepancy before judging the results themselves. Third, they also pointed out that we pooled studies from several meta-analyses on the same topic (e.g. for prostate and pancreatic cancer) without acknowledging/correcting for study overlap [1.Markozannes G. Kalliala I. Kyrgiou M. Tsilidis KK. Letter to the editor on ‘Body mass index and 20-specific cancers—re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 1490-1491Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar]. Importantly, however, as was also indicated in our article, we went through the correction of individual study overlap. In the Kyrgiou et al.'s study [2.Kyrgiou M. Kalliala I. Markozannes G. et al.Adiposity and cancer at major anatomical sites: umbrella review of the literature.BMJ. 2017; 356: j477.Crossref PubMed Scopus (420) Google Scholar], an evidence grading of one cancer–obesity association was classified not only as convincing but also as a weak evidence according to the kinds of obesity parameters. In this situation, someone might wonder what is the conclusion for the association between BMI and one cancer type. To interpret the results correctly, two things should be considered. One thing is that only one obesity parameter should be selected for one comparison and it cannot be applied to the results by other obesity parameters. If one performs an update of a meta-analysis, individual studies included in the first meta-analysis should at least be included in the second updating meta-analysis. However, we found many missing individual studies in the second update or overlapping meta-analysis, although the authors executed an extensive search strategy. In this situation, therefore, to get to a final conclusion for the association between BMI and one cancer type, we think that all individual studies should be used in the final meta-analysis, because the results of a meta-analysis can change completely by adding or omitting several individual studies. Therefore, we recommend that an evidence of grading should be done on this final result of a meta-analysis rather than separate re-analysis of each meta-analysis alone. Fourth, Markozannes et al. criticize that we updated existing meta-analyses by selectively adding only one newly published study without conducting a systematic literature search [1.Markozannes G. Kalliala I. Kyrgiou M. Tsilidis KK. Letter to the editor on ‘Body mass index and 20-specific cancers—re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 1490-1491Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar]. Clearly, the results of meta-analysis can change by adding data of individual studies and one should consider what data are most convincing: the results of a meta-analysis or the results of the single largest study. One set of data included in the current analysis was derived from a recently published large study in the UK and it we attempted to assess whether it would change the results of meta-analysis. Therefore, if one performs a new systematic literature search and adds data from additional individual studies, it could lead to a change in the conclusion. However, this conclusion cannot be drawn by simple re-analysis of each published meta-analysis. Fifth, Markozannes et al. pointed out that we did not report or differentiate the analysis by study design (e.g. cohort versus case–control) [1.Markozannes G. Kalliala I. Kyrgiou M. Tsilidis KK. Letter to the editor on ‘Body mass index and 20-specific cancers—re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 1490-1491Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar]. We believe this is beyond the scope of our umbrella review and already indicated several limitations of our study. Sixth, it was pointed out that several associations by cancer subsite or according to modifying factors were omitted from the current article [1.Markozannes G. Kalliala I. Kyrgiou M. Tsilidis KK. Letter to the editor on ‘Body mass index and 20-specific cancers—re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 1490-1491Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar]. However, these limitations of our study were already mentioned in the discussion where we described that we could not perform re-meta-analyses according to gender, smoking status or ethnicity, which should therefore be interpreted based on the summary of the umbrella review summarized in the supplementary Table S1, available atAnnals of Oncology online [4.Choi E.K. Park H.B. Lee K.H. et al.Body mass index and 20-specific cancers: re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 749-757Abstract Full Text Full Text PDF PubMed Scopus (43) Google Scholar]. Finally, Markozannes et al. pointed out that we used aP-value threshold of 0.05 for Egger’s regression asymmetry test as evidence for small-study effects, but this test is known to be underpowered and 0.10 is the widely accepted threshold [1.Markozannes G. Kalliala I. Kyrgiou M. Tsilidis KK. Letter to the editor on ‘Body mass index and 20-specific cancers—re-analyses of dose-response meta-analyses of observational studies.Ann Oncol. 2018; 29: 1490-1491Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar]. We are sympathetic to this opinion, but already described that the application of prediction interval (PI), heterogeneity and publication bias may not be definitive criteria in limitation section. When we applied aP-value threshold of 0.1 for Egger’s regression asymmetry test to all our results, the results almost did not change. In conclusion, the evidence for the associations of BMI with each cancer should be interpreted with caution and the evidence grading for each comparison should be done according to the obesity parameter used, which could change the evidence grading for the same cancer type. The methods for the evidence grading deserve further attention and discussion. None declared.