LAUSR

Dear Editor, It was with great interest I read the recently published study by Seliger et al. evaluating the prognostic effect of diabetic metformin in high-grade glioma (HGG) patients. The authors should be commended on producing such a thorough analysis and providing insight into how this may translate clinically, as any effective medical adjunct is a welcome finding in managing this deadly disease. However, I believe a few points require clarification in the interest of validity that may temper the implication of metformin-related survival improvement in Grade III glioma only, and not Grade IV. Statistically, the authors analyze the survival metrics of a cohort of 1,093 HGG patients, with only 55 receiving metformin (5.0%), and only 11 of them being Grade III. A large concern of mine is that there appears to be no attempt made case–control or propensity match overall or within Grades, which would have reduced statistical noise in the analyses. Without this, we run the risk of overexaggerating the alternative hypothesis, and depreciating the null hypothesis, especially given the small cohort size of the experimental arms in question. In addition, the authors note that due to the data generating from pre-existing population-based and multicenter databases, the results are not likely prone to selection bias. This may not be completely accurate, for the prospective nature data was collected from 1998 to 2013 without clearly defined variables for our study in mind. For example, IDH testing only became standard of practice in 2009, meaning a large proportion of glioma patients prior were subject to availability and suitability criteria for testing—perhaps patients with a perceived better prognosis or greater tolerance of adjuvant therapy were offered testing only, indicating selection bias. Also, given the Stupp Protocol for Grade IV glioma standard of care was established in 2005, it is possible many included Grade IV patients received heterogeneous care prior to that. These selection bias concerns, in the absence of controlling, are amplified further by the vast intrinsic clinical heterogeneity involved in recruitment of patients from 53 regional hospitals managed by over 1,500 practitioners. These concerns carry on into the reported multivariate regression analyses, as a number of included covariates such as extent of resection (EOR), isocitrate dehydrogenase (IDH) and O-alkylguanine DNA alkyltransferase (MGMT) statuses, and body mass index (BMI) have >50% of their data listed as “Unknown,” in both the metformin group and overall. This is concerning for how the regression handled this data—was it included as another category in the data? If so, this would have diluted the statistical power of the analysis, and weakened the fairness toward the null hypothesis. Biologically, it is difficult to rationalize any potential prognostic benefit by metformin would be exclusive to Grade III glioma only. At a cellular level, evidence of anticancer properties have been demonstrated in vitro in Grade IV glioma, as well as in patients where metformin showed statistically significant improvement in progression free survival (PFS) in diabetic Grade IV glioma. Given the likelihood that all HGG share similarities in both origin and microenvironment, why and how exactly Grade III glioma cells interact differently with Grade IV glioma cells is uncertain. The authors have proposed a very interesting explanation, in that an association with survival metrics may be related to IDH status. If the advantage conferred to Grade III glioma is indeed via a positive IDH mutation modulating ketoglutarate and other oxidative products, it would suggest that variations in the biochemical activity of IDH mutations themselves could also modulate survival success. This is not necessarily the case as observed with The Cancer Genome Atlas (TCGA) survival data with respect to glioma. The authors note in their discussion that when HGG patients with known IDH mutation status were analyzed separately, there was a “trend” toward improved survival it may be precarious to use the term “trend” when describing these dichotomous data results without reporting statistical significance, as a summary statistic of perfect nullity is largely theoretical, and thus most analyses would otherwise show a “trend” as well. Subgroup analysis of Grade III glioma only with known IDH status was not reported, ultimately leaving the significance of metformin use in this specific cohort questionable. Clinically, there are a number of important considerations that may have confounded the ability to associate metformin and HGG survival outcomes. Firstly, patients receiving metformin have a clinical indication, most probably Diabetes Mellitus (DM) Type II. DM is a chronic condition, requiring consistent monitoring and medical follow-up. DM patients with glioma are more likely to be vigilant of symptoms, and present earlier to medical attention, as opposed to glioma patient without any active medical monitoring. Lead-time bias due to earlier diagnosis may create what appears to be superior OS. Secondly, after diagnosis, the improved PFS metric may be a result of greater symptomatic control in DM patients rather than relief of cancer burden. Patients with DM are already at risk for symptom development, and thus better control of DM would reduce this risk, and possibility then the tendency to present for progression evaluation of the tumor due to similar symptoms, e.g. headache due to International Journal of Cancer IJC