Skip to main content

Advertisement

Table 2 Event times generated according to a Cox-exponential model with a confounder X and an exposure status Z with time-dependent effect

From: Model-based estimation of measures of association for time-to-event outcomes

  BIAS MSE  
  PV Cox Cox PV Cox Cox
  r c s(t) log(t) r c s(t) r c s(t) log(t) r c s(t)
200 -0.0290 0.0036 0.0151 0.0792 0.0689 0.1403
400 -0.0248 -0.0108 -0.0188 0.0730 0.0657 0.1119
600 -0.0185 -0.0134 -0.0242 0.0730 0.0657 0.1119
  Width Coverage  
  PV Cox Cox PV Cox Cox
  r c s(t) log(t) r c s(t) r c s(t) log(t) r c s(t)
200 0.2899 0.2700 0.5162 0.9270 0.9530 0.9120
400 0.2747 0.2528 0.4211 0.9470 0.9500 0.9170
600 0.2158 0.1860 0.3256 0.9410 0.9410 0.9300
  1. The effect of Z is multiplied by log(t). CGPM was adopted using the Cox proportional hazard regression model with the time dependent effect of Z specified in two ways: (1) with an interaction between Z and log(t), corresponding to the generating model; (2) with and interaction between Z and a restricted cubic spline of time (rcs(t)). PV model was adopted with identity link. Pseudo times were fixed at quantiles 5%, 25%, 50%, 75%, 95% of the failure time distribution. The Z covariate was inserted with an interaction with the baseline risk function modelled with a restricted cubic spline with 3 knots, giving an estimated risk difference varying through follow-up time, RD(t).