Letter to the Editor, “Effectiveness of Implant Therapy Analyzed in a Swedish Population: Early and Late Implant Loss”

We have concerns with a recent article in the journal by Derks et al. (2015) that we want to bring to the attention of the readership. The authors identify factors affecting the probability of implant loss using a multiple logistic multilevel model without adjusting for or discussing the risk for false significances due to multiple significance testing. All factors in the full model were tested for importance. A factor was considered related to implant loss if the P value for “importance” was statistically significant (P < 0.05). The final model included only “significant” factors. The methodology could be criticized as no adjustment for multiple testing was done—or even discussed. We believe that it is most important to better interpret the results not by solely performing mathematical adjustments on the significance levels (or on the P values). For many readers, it is tempting to interpret P values <5% as statistically proven. However, this could be done only if the P values are properly adjusted for multiplicity. The discussion above strongly suggests that no formal conclusions should be drawn—or indicated—on the basis of the results in the article. Potentially important factors have been identified but need to be confirmed in further research—preferably in a randomized clinical trial setting or with use of data from another historical database. This should have been emphasized clearly in the article. It is our opinion that the results of Derks et al. should not be considered confirmative but as hypothesis generating.