Table 2 Royston-Parmar model specifications
Author (year) | Country | Reason for flexible parametric survival model | Relative survival model | Number of interior knots (i.e., df-1) | Expressed as knots or degrees of freedom? | Placement of knots | Sensitivity analysis for knots | Scale used | Software (command) | Candidate variables | Variable selection strategy | Variables in final model |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
Andersson et al. (2014) [34] | Sweden | Prognostic model | Yes, cure model | 5 | Degrees of freedom | Evenly spaced | Not stated | PH | Stata (stpm2) | 4 | Not stated | 4 |
Baade et al. (2015) [33] | Australia | Prognostic model | Yes | 6 | Degrees of freedom | Not stated | Not stated | PH | Stata (stpm2) | 3 | Not stated | 3 |
Baade et al. (2015) [41] | Australia | Prognostic model | No | 2 | Degrees of freedom | Not stated | Compared models with varying number of knots on the PH, PO, and probit scales using BIC | Probit | Stata (stpm2) | 9 | Backwards selection | 6 |
Castillo et al. (2013) [38] | United States of America | Prognostic model | Yes, cure model | Not stated | N/A | N/A | Not stated | PH | Stata (not stated) | 5 | Not stated | 5 |
Csordas et al. (2016) [37] | Switzerland | Prognostic model | No | 1 | Degrees of freedom | Not stated | Not stated | PH | Stata (not stated) | 4 | Not stated | 2 |
Eyre et al. (2012) [40] | United Kingdom | Complement Cox PH prognostic model | No | Not stated | N/A | N/A | Models compared with AIC; the types of models compared are not described | PH | Stata (not stated) | Royston-Parmar model not used for model | Royston-Parmar model not used for model | Royston-Parmar model not used for model |
Fox et al. (2014) [32] | United Kingdom | Prognostic model | No | Not stated | N/A | N/A | Not stated | PH | Stata (stpm2) | 10 | Backwards selection | 10 |
Li et al. (2016) [39] | United Kingdom | Prognostic model | No | 2 | Knots | Not stated | Compared models with 0 to 4 knots on the PH, PO, and probit scales using AIC | PH | Stata (stpm2) | 17 | Backwards selection | 6 |
Miladinovic et al. (2012) [30] | United States of America | Prognostic model | No | 1 | Knots | Evenly spaced | Compared models with 0 to 5 knots on the PH, PO, and probit scales using AIC, BIC, and R2 | Probit | Stata (not stated) | 1 | Not stated | 1 |
Myklebust et al.(2016) [31] | Norway | Prognostic model | Yes | 4 | Degrees of freedom | Evenly spaced | Not stated | PH | Not stated | 3 | Not stated | 3 |
Ramezani Tehrani et al. (2016) [35] | Iran | Prognostic model | No | 1 | Degrees of freedom | Not stated | Compared model with 2 knots on PH scale versus model 1 knot on PO scale using AIC | PO | Stata (not stated) | 4 | Forward selection | 2 |
Sanchis et al. (2014) [36] | Spain | Complement Cox PH prognostic model | No | Not stated | N/A | N/A | Not stated | PH | Stata (not stated) | Royston-Parmar model not used for model | Royston-Parmar model not used for model | Royston-Parmar model not used for model |