Figure 2 : Assessment of the Cox PH regression model adequacy.

The martingale (A), deviance (B) and score (C) residuals in the scatterplots are plotted against “patient observation numbers”. The martingale and deviance residuals are graphed to identify possible outliers. The score residuals are plotted to detect influential observations. Observations 30, 38 and 40 in (A) and (B) stand out from the rest. A large number of positive martingale as well as deviance residuals form a horizontal band of points because of tied events. Possible influential observations on the estimated coefficients in (C) are 38, 41 and 80.