Method | Advantages | Limits |
---|---|---|
Raw coverage | Based on the reported and non pre-processed data A graphical check can easily be performed to assess the quality of the coverage | Does not take into account the OTU Strongly dependent on the width of the predicted interval Does not rely on a statistical test |
Juncture | Takes into consideration the OTU A graphical check can easily be performed to see how well the observed and predicted intervals overlap | Strongly dependent on the width of both observed and predicted intervals A minimal overlap between the two intervals is enough to consider the predictions as validated for a given time point Does not rely on a statistical test |
Bootstrapped log-rank (without OTU) | Based on a statistical test frequently used in a TTE context Combined with a bootstrap approach to avoid an excess of statistical power | Does not take into account the OTU Credibility of the result if the proportional hazards assumption is not met |
Bootstrapped log-rank (with OTU) | Based on a statistical test frequently used in a TTE context Combined with a bootstrap approach to avoid an excess of statistical power Takes into consideration the OTU | Credibility of the result if the proportional hazards assumption is not met |
Bootstrapped combination of weighted log-ranks (without OTU) | Based on an improved version of the log-rank test, more robust in case of non-proportional hazards Combined with a bootstrap approach to avoid an excess of statistical power | Does not take into account the OTU Can be overly sensitive to minor differences because of its design |
Bootstrapped combination of weighted log-ranks (with OTU) | Based on an improved version of the log-rank test, more robust in case of non-proportional hazards Combined with a bootstrap approach to avoid an excess of statistical power Same as above Takes into consideration the OTU | Can be overly sensitive to minor differences because of its design |