Summary of performance measures reported in the 41 development studies.

Overview of performance measures reported by ≥2 studies. At least one performance measure was required for inclusion; most studies reported more than one. A summary of measures per study is given in Table S2, and the measures reported for each model are in Table S3. The heterogeneity of the studies precludes direct comparison of model performance.
AUC  =  area under the curve; ideal  = 1.0 and chance  = 0.5.
R2  =  coefficient of determination; the proportion of variability that is accounted for by the model. ideal  =  1.0.
H-L p  =  Hosmer-Lemeshow p value; any non-significant value indicates acceptable calibration.
% correct; ideal  = 1.0 and chance ≈ %survival2.
PPV  =  positive predictive value; ideal  = 1.0 and chance related to prevalence and cut-off.
NPV  =  negative predictive value; ideal  = 1.0 and chance related to prevalence and cut-off.
sensitivity; ideal  = 1.0 and chance related to prevalence.
specificity; ideal  = 1.0 and chance related to prevalence.
range  =  the range of probabilities generated by the model.
Note that performance measures of unvalidated models may be overestimated.

≥2项研究报告的性能指标概述：
纳入本分析的研究需至少报告一项性能指标，多数研究同时报告了多项指标。各项研究的性能指标汇总详见补充表S2，各模型所报告的性能指标详见补充表S3。由于各项研究间存在异质性，无法直接比较模型性能。

各性能指标定义如下：
1. AUC（area under the curve，曲线下面积）：理想取值为1.0，随机猜测的基准值为0.5。
2. R²（coefficient of determination，决定系数）：表示模型所能解释的变异占总变异的比例，理想取值为1.0。
3. H-L p（Hosmer-Lemeshow p value，霍斯默-莱梅肖p值）：若该值无统计学显著性，则说明模型的校准性能可接受。
4. 正确率（% correct）：理想取值为1.0，随机猜测的取值近似于%survival2。
5. PPV（positive predictive value，阳性预测值）：理想取值为1.0，随机猜测的取值与患病率及截断值相关。
6. NPV（negative predictive value，阴性预测值）：理想取值为1.0，随机猜测的取值与患病率及截断值相关。
7. 灵敏度（sensitivity）：理想取值为1.0，随机猜测的取值与患病率相关。
8. 特异度（specificity）：理想取值为1.0，随机猜测的取值与患病率相关。
9. range（概率范围）：指模型生成的概率值的取值区间。

请注意：未经过验证的模型，其性能指标可能存在高估情况。