Common Errors in Statistics (and How to Avoid Them), 4th Edition

SURVIVAL ANALYSIS

Survival analysis is used to assess time-to-event data including time to recovery and time to revision.

Most contemporary survival analysis is built around the Cox model for which the hazard function takes the form , where for each observation is a 1 × p row vector of covariate values and is a p × 1 column vector of to-be-estimated coefficients. Possible sources of error in the application of this model include all of the following:

• Neglecting the possible dependence of the baseline function λ₀ on the predictors.
• Overmatching, that is, using highly correlated predictors that may well mask each other’s effects.
• Using the parametric Breslow or Kaplan–Meier estimators of the survival function rather than the nonparametric Nelson–Aalen estimator.
• Excluding patients based on post-hoc criteria. Pathology workups on patients who died during the study may reveal that some of them were wrongly diagnosed. Regardless, patients cannot be eliminated from the study as we lack the information needed to exclude those who might have been similarly diagnosed but who are still alive at the conclusion of the study.
• Failure to account for differential susceptibility (frailty) of the ...