Statistical analysis
We calculated sample-size by two independent proportions.17, 18 Using estimated 6-month improvement rate of 83% with LNG-IUD insertion and 71% with hysteroscopic niche resection based on our pilot study, and assuming an alpha level of 0.025 for a one-sided test, we calculated that a sample of 68 cases who underwent therapy would result in 80% power (1:1 ratio). The sample size was increased to 76 to allow for a 10% rate of loss to follow-up over the 12-month period.
All statistical tests were conducted using IBM SPSS 23(SPSS Inc., Chicago, IL, USA). We used both ITT and per protocol analyses. For the primary outcome (effective rate)was based on the ITT analyses. The mean ± standard deviation was used to expressed the continuous data that were normally distributed, the differences between groups were tested using the student t test. Otherwise median and interquartile range (IQR) were used and tested with Mann-Whitney U test. Categorical variables were compared by the χ 2 test; if more than 25% of cells were found to have an expected value of between 1 and 5, Continuity Correction Chi-Square test was used instead; if more than 25% of cells were found to have an expected value of less than 1, Fisher’s exact test was used. The mix linear model was used to compare the effective rate of spotting symptoms repeatedly measured within 1 year after surgery between two groups and the spotting days between each follow-up time points and baseline. Linear-by-linear association test was used to calculated the P for trend to expressed the efficacy change over time. Logistic regression analysis was used to analyze the effect of the TRM and also of age on spotting and total bleeding days at each follow-up point. The Statistical significance was determined by a p value of <0.05.