Regional cerebral blood flow changes in patients with internet addiction: Authors' reply

To the comments of Prof. Andreas Otte to our work we reply as follows: As was mentioned by Prof. Andreas Otte, our study was the first study of regional cerebral blood flow changes in patients with internet addiction. Therefore, there was not much previous, established experience to refer to. As an exploration study, it was inevitable that there were some insufficiencies. We feel great appreciation to Prof. Andreas Otte for his comments on our work, which will improve our studying quality in this field in the future. Prof. Andreas Otte inquired in his letter how the regional cerebral blood flow (rCBF) was calculated, and was scaled, relatively to the whole brain mean value or to the cerebellar mean value. All rCBF data were scaled relatively to the whole brain, in our study. As for the question in relation to the test level, the P-value of 0.05 was only used when comparing intergroup differences of baseline or clinical information of patients using SPSS, while the P-value of 0.01 was used for the statistical parametric mapping (SPM) t-test. We had tried using the test level of P<0.001 to indicate the significance when performing the SPM t-test, but there were not very meaningful findings identified at this test level. This might be explained by the relatively old generation and low sensitivity of the SPET scanner used in our study. For the purpose to identify more potential cerebral regions with abnormal rCBF, we lowered down the test level to 0.01, although this might had resulted in some false-positive results. However, in order to control false-positive errors, we performed intragroup comparisons from rest to adenosine-stressed status firstly by paired t test to identify cerebral regions with obvious rCBF changes because of administration of adenosine. On the basis of these cerebral regions, we subsequently performed a two-sample t test to compare intergroup differences to identify cerebral regions with rCBF that could have attributed to internet addiction. We believed that this "twostep" statistical mode might reduce the probability of falsepositive results to some extent. As for the cluster question in relation to SPM analysis, it seems more problematic and more prone for clusterwise inference to produce false-positive results than voxelwise inference, as mentioned in the paper by Eklund et al (2016). We did not take the two-way ANOVA analysis, instead, we performed the more suitable "twostep" statistical mode mentioned above. However, when facing multiple comparisons, we did adjust the P values with the relatively conservative Bonferroni method. Thanks again for the interesting and the meaningful comments of Prof. Andreas Otte to our study. If there are any more queries, please do not hesitate to contact us.