Blue-202110-2383OC simulation
Anon reviewer
11/7/2021
Introduction
The purpose of this simulation is to show that the patterns of EBA reported in the MS can be reproduced with random variation within a fixed population effect of time on MPN (i.e. no true subgroups of PD response, only “noise”), making reasonable assumptions about the data generating mechanisms.
Definition of response patterns
From suppl table 5 have extracted these pattern definitions. Note these are not mutually exclusive and do not cover the entire set of possible responses. Have applied the definitions hierarchically to account for the former problem.
- classic_biphasic = (t_0 - t_3)>0.5 AND t_7<t_0 AND t_14<t_0
- early_nonrespond = [(t_0 - t_3)<=0.5 OR (t_0 - t_7)<=0.5] AND t_14 < t_7
- paradoxical = ((t_3 > t_0) OR (t_7 > t_0) OR (t_14 > t_0))
- negligible = |(t_0 - t_3)|<0.5 AND |(t_0 - t_7)|<0.5 AND |(t_0 - t_14)|<0.5
(Here t_x refers to MPN count at time x)
Simulation model
The model assumes:
- a fixed population intercept for MPN at time 0 of 7 log (average baseline value), and fixed slope for average rate of change in MPN per day -0.25 (0.25 log decrease in MPN per day) i.e. a monoexponential time-kill.
- an individual participant has random variation in intercept and slope around this population average intercept and monoexponential decline in MPN; these random effects are normally distributed with mean 0 and s.d. 1.5 and 0.05 respectively.
- there is residual error (e.g. from technical variation related to sampling and laboratory noise); this residual error is normally distributed with mean 0 and s.d. set to 1.
Simulation based on above model
100 participants simulated from above parameters, classified by the above pattern definitions: all the patterns are possible assuming only random variance in MPN as specified above.
