Skip to main content

Table 1 Bias (bold font), coverage (italics font) and MSE (normal font) for simulation scenario 1 comparing the non parametric Pohar Perme estimator, a conditional model (without covariates), regression standardization under proportional hazards (PH), regression standardization under non-proportional hazards (Non PH) and a marginal model. Bias is expressed as a difference in probabilities

From: Direct modelling of age standardized marginal relative survival through incorporation of time-dependent weights

  Years from diagnosis)
Method 1 5 10
Pohar Perme 0.0005 0.0006 0.0006
  95.4 95.1 95.4
  198.676 302.290 430.220
Conditional Model 0.0261 0.0486 0.0457
  53.6 19.5 29.3
  873.184 2653.730 2450.234
Regression standardization (PH) -0.0044 -0.0093 -0.0202
  92.5 93.5 80.7
  204.942 353.266 737.678
Regression standardization (Non PH) ∗ 0.0025 0.0016 -0.0006
  93.6 97.2 96.5
  188.909 283.351 381.325
Marginal model 0.0032 0.0020 0.0018
  95.7 95.7 95.5
  192.043 288.790 420.932
Relative % increase in precision +
Regression standardization (PH) 7.2 12.9 30.0
Regression standardization (Non PH) ∗ 8.6 7.6 12.9
Marginal model 9.0 6.0 3.0
  1. bias, Coverage, MSE
  2. ∗ 0.5% of models did not converge
  3. + compared to Pohar Perme
  4. PH - proportional hazards
  5. Non PH - non proportional hazards
Back to article page

BMC Medical Research Methodology

ISSN: 1471-2288

Contact us