Skip to main content

Table 4 Comparison of the mean absolute error for ln (HR) and var. ln (HR) using the Parmar, Guyot and nlopt method (Exact and Non-exact P value) described in this study

From: A non-linear optimisation method to extract summary statistics from Kaplan-Meier survival plots using the published P value

Method Mean Absolute Error (95% CIs) Validation
ln (HR) Var ln (HR)
1. Nlopt method: exact P value 0.014 (0.007–0.022) 0.0039 (0.0022–0.0056) 13 KM plots (this study)
2. Nlopt method: non-exact P value 0.087 (−0.036–0.210) 0.0033 (0.0010–0.0055)
3. Parmar: this study 0.077 (0.039–0.115) 0.0038 (0.0016–0.0060)
4. Parmar 0.079 (0.048–0.110) 0.0104 (0.0070–0.0139) 48 KM plots [8]
5. Guyot: All information 0.017 (0.002–0.122) 0.0026 (2e-5–0.1332) 6 KM plots [16]
6. Guyot: No numbers at risk 0.036 (0.003–0.242) 0.0015 (6e-6–0.0541)
7. Guyot: No total events 0.028 (0.002–0.149) 0.0065 (2e-5–0.2497)
8. Guyot: No additional information 0.198 (0.021–1.556) 0.1227 (6e-4–3.2501)
  1. Ln (HR) and var. ln (HR) reported as mean absolute error with 95% confidence intervals. The var ln (HR) for Guyot were calculated from the published standard errors by using var. = SE^2*n