Association Between Breast Cancer Risk and Levonorgestrel‐releasing Intrauterine System

Sir, It is our privilege to respond to a few comments recently published about our paper “Levonorgestrelreleasing intrauterine system and breast cancer risk: A systematic review and metaanalysis.”1 In a Letter to the Editor,2 Kiyumi and coworkers stated that “ ... the pooled analysis in Figure 2 combined adjusted and crude odds ratios. For example, Soini et al. did not adjust for any covariates and crude odd ratio (OR) was reported. Moreover, this study contributed to the highest percentage of weight (18.5%), which therefore might have influenced the overall effect size.” As described in our paper, the studies included in the metaanalysis obtained their association estimates for the association between levonorgestrelreleasing intrauterine system (LNGIUS) use and breast cancer using several different analytical approaches. These technical choices and the reasons put forward by the authors, when available, were meticulously described in the systematic review section of our manuscript, and resulted in effect descriptors such as risk ratios, OR and SIR. In our manuscript, as detailed in the statistics section, the effect sizes for LNGIUS use retrieved from each study were derived using these different estimates (eg hazard ratios, OR, risk ratios, SIR) and their standard errors, and these effects were ultimately used in the metaanalyses calculations. This is a perfectly valid approach in mathematical terms, and the essence of our metaanalysis was thus to compute a more robust effect size measure for the association of LNGIUS with breast cancer than those sparsely reported throughout the available literature. Of course, we had no reasons to doubt that the analytical approach used by each of the contributing papers was technically correct. Reader discretion is always advised, and all technical dissimilarities across studies were rigorously described in our manuscript. Despite randomized clinical trials being considered the gold standard for evidencebased medicine, due to their lower chances of selection bias and “confounding” effects, in some circumstances they are not feasible. The available observational studies did not whittle down the results of our metaanalysis; the studies included a high number of participants and, importantly, they most likely reflect the daily routine in clinical practice, as observational studies have a greater external validity than studies with narrow inclusion criteria, such as the randomized controlled trials. This notwithstanding, we are satisfied to learn that Kiyumi et al. mirrored exactly our own conclusions1 when they stated that“ ... causal inference in this study should be cautiously interpreted and more prospective welldesigned studies that mitigate the potential effects of confounding bias, such as randomized controlled trials, are necessary.”