Dear Sir,Earlier this year, we reported the results of a systematic review and meta-regression analysis determining the relationships between circulating total IGF-I and IGF binding protein-3 (IGFBP-3) concentrations and common cancer risk, including premenopausal and postmenopausal breast cancers.
1Two subsequent meta-analyses have been published specifically examining relationships with breast cancer risk.2,3 At first glance, there are some differences in the results and the derived conclusions. This correspondence addresses the reasons.Our study determined associations between circulating IGF peptides and cancer risk at 4 sites: prostate, colorectum, breast (considered pre-and postmenopausal) and lung. Strict inclusion criteria were used, namely, that findings were reported as odds ratios (ORs) with 95% confidence intervals (CIs). Statistical analyses were performed at 2 levels: (i) a baseline metaanalysis of ORs comparing the highest and lowest (referent) category for each cancer site using conventional randomeffects methods 4 and (ii) a meta-regression analysis of the study-specific dose-response slopes generated using the method described by Greenland and Longnecker. 5 For breast cancer, the results from the 2 analyses were similar and demonstrated that higher concentrations of both IGF-I and IGFBP-3 were significantly positively associated with increased risk of premenopausal, but not postmenopausal, breast cancer. Furthermore, through sensitivity analyses, we demonstrated that the associations with IGF-I and IGFBP-3 were likely to be independent of each other (webappendix8 www.christie.man. ac.uk/profinfo/departments/surgery/default.htm). The positive association between higher circulating IGF-I and premenopausal breast cancer is not unexpected as IGF-I is mitogenic, antiapoptotic and proangiogenic, attributes that favour tumour development. However, the observation that circulating IGFBP-3 levels are positively associated with premenopausal breast cancer risk is contrary to perceived convention, which suggests that inverse associations should prevail as IGFBP-3 is predominantly antitumour in laboratory experiments and consequently the subject of debate (www.christie.man.ac.uk/ profinfo/departments/surgery/default.htm).The review of Shi et al. 2 used less restrictive inclusion criteria and included 16 publications. The meta-analysis was performed using a method known as Hedges' standardised mean differences, involving calculation of weighted mean effect sizes and their 95% CIs from the reported concentrations of each peptide for each individual study. From this and knowing the number of cases and controls, a correlation coefficient was calculated, with ORs and their 95% CIs estimated. While this approach allowed the meta-analysis to capture more studies for associations with IGF-I, the estimated ORs and 95% CIs differed for most studies from those reported originally. Moreover, this approach failed to calculate estimates for associations with IGFBP-3 in several studies despite ORs and 95% CIs being repor...