Skip to main content

Table 2 Mean Square Error (MSE) comparison for B-spline, P-spline and fractional polynomial (FP) network meta-analysis models. Displayed are the mean and standard deviation (SD) of the respective MSEs averaged over 50 simulated data sets for each scenario. Unless stated otherwise, considered outcomes are continuous. Scenarios that contain non-monotonic temporal behaviors appear to be challenging for the FP method

From: Bayesian splines versus fractional polynomials in network meta-analysis

Scenario B-spline model   P-spline model   FP model  
  Mean SD Mean SD Mean SD
Linear 0.0019 0.0213 0.0021 0.0243 0.3505 0.0388
Logarithmic 0.0014 0.0111 0.0015 0.0135 0.3505 0.0118
Piecewise linear monotonic 0.0011 0.0923 0.0052 0.1437 1.3069 0.1823
Mixed 0.0037 0.0313 0.0040 0.0931 1.5747 0.1338
Non-monotonic 0.0019 0.0389 0.0027 0.1039 41.7533 1.8354
MTC 0.0374 0.0483 0.0402 0.0984 0.9198 0.2864
BEST-ITC 0.0397 0.0503 0.0402 0.1003 0.1701 0.2898
Piecewise linear monotonic (binary) 0.0034 0.0118 0.0053 0.0978 0.0897 0.3698
Non-monotonic (binary) 0.0004 0.0019 0.0004 0.0178 0.9748 0.2976
Piecewise linear (non-closed network) 0.0172 0.0173 0.0235 0.0774 0.0903 0.3854