304 chapter 4 Statistical analysis The trial was designed to have 80% power to detect a reduction in unbeneficial management from ~75% to ~40% at a significance level of 0.05. At least 90 evaluable patients with nodules >10 mm were required (2:1 allocation). After correction for 82.7% expected nodule size >10mm and 15% estimated data-attrition, the sample size was set at 132 patients [29, 38]. After half of the anticipated patients of the diagnostic surgery group were recruited, the study safety committee conducted a predetermined interim analysis and reported no objections to safe continuation of the trial. The applied descriptive statistics were mean ± standard deviation or median and interquartile range for continuous variables, and absolute numbers and relative frequencies (%) for categorical variables. Intention-to-treat analysis was performed. Categorical primary and secondary outcomes were compared between allocated groups using Pearson’s chi-squared or Fisher’s exact tests, where appropriate. We adjusted for the stratifying variables using binary logistic regression; the corrected p-values are reported together with an adjusted odds ratio and their 95% confidence intervals (CI) [479]. Sensitivity, specificity, negative and positive predictive value (NPV, PPV) were calculated using the traditional formulas. 95% CIs were calculated using the β-distribution (Clopper-Pearson interval). For EQ-5D-5L, iMCQ and iPCQ questionnaires, we used multiple imputation to account for possible selectively missing values. To estimate HRQoL, we calculated Dutch utility scores from the EQ-5D5L and estimated the mean one-year QALYs as the area under the utility curves. One-year societal costs were estimated as the mean sum of [volume x costs] for all components. QALYs and costs are presented as mean and 95% CI, and compared using independent samples T-tests with unequal variances. In the analysis of QALYs and costs, we adjusted for the stratifying variables and additionally adjusted for possible influences of the unevenly distributed malignancy/borderline rate (see Results), using a generalized linear model (GLM). The local benign/malignant histopathological diagnosis was included in the GLM, as this diagnosis determined the patients’ postoperative course of treatment and thus contributed to the costs and perceived HRQoL. The adjusted means, p-values, and mean differences are presented. Two prespecified subgroup analyses for the primary outcome were performed: one for nodules >10 mm (ultrasonographic largest diameter), and one for Hürthle cell nodules (defined as Bethesda IV HCN/SHCN cytology) and non-Hürthle cell nodules (defined as Bethesda III AUS/FLUS and Bethesda IV FN/SFN cytology). Data collection was performed using Castor EDC (Castor EDC, Amsterdam, the Netherlands). Statistical analysis was performed using SPSS Statistics version 26 (IBM Corp, Armonk, NY, USA). This trial is registered with ClinicalTrials.gov NCT02208544 (5 August 2014).
RkJQdWJsaXNoZXIy MTk4NDMw