Table 3. Suggested plots for residual-based diagnostics.

Y-axis	X-axis	Potential implication	Suggested remedy
Martingale residual	Any omitted covariate	Covariate excluded wrongly	Refit model with covariate included
Martingale residual	Any included covariate	Covariate modelled incorrectly (e.g. nonlinear effect)	Fit nonlinear term (e.g. a squared term)
Martingale residual	Date of enrolment in study	Evidence of temporal effect	Incorporate time of entry as a covariate
Deviance residual	Survival time, log(survival time) or ranks of survival time	Model fails to predict consistently for all survival times	Fit time-dependent PH model or consider using a different (i.e. non-PH) model
Deviance residual	Subject identifier	Individual is an outlier	(1) Check if the data are correct
			(2) Refit model with individual removed. If effect sizes alter substantially, consider removing individual altogether
Scaled Schoenfeld residual	Survival time, log(survival time) or ranks of survival time	Non-PH	Fit time-dependent PH model or consider using a different (i.e. non-PH) model

PH=proportional hazards. All of the above X–Y plots should give rise to a plot evenly scattered along a horizontal line that displays no trend. The possible implications where this does not occur and suggested remedies are presented.