Performance on the complete dataset.

Overview of the performance of the different algorithms using the univariate and multivariate approach and on 8, 24 and 48 weeks of therapy. The multivariate approach includes additional variables in the model: start year of therapy, information on start of a new drug class, number of previous therapy switches, previous drug class experience, baseline viral load, baseline CD4, gender, age, risk group. Reported are the odds ratio (OR), 95% confidence interval (CI) and P-value (P) of the logistic model and the median and standard deviation (SD) of the 10-fold cross-validation area under the ROC curve (AUC).*The performance of the algorithms is compared with that of Rega 8 using a Wilcoxon signed-rank test and the P-value corrected for multiple testing is reported.

本研究呈现了不同算法在单变量、多变量分析框架下，以及分别针对8周、24周、48周治疗周期时的性能表现。其中多变量分析模型纳入了以下额外变量：治疗起始年份、新药物类别启动信息、既往治疗转换次数、既往药物类别暴露情况、基线病毒载量、基线CD4计数、性别、年龄以及风险组别。本次报告涵盖逻辑回归模型的优势比（odds ratio, OR）、95%置信区间（confidence interval, CI）与P值（P），以及10折交叉验证下受试者工作特征曲线下面积（area under the ROC curve, AUC）的中位数与标准差（standard deviation, SD）。*本研究采用Wilcoxon符号秩检验，将各算法的性能与Rega 8进行对比，并报告校正多重检验后的P值。