Hi,
In the description of the output prediction format (reproduced below) we have a hint on what should be the output of the would-be submitted model, however, it is unclear if the prediction score should be computed according to any (standard?) metric. Or, we should come up with our own prediction score metric? (Presumably, based on some analysis of the survival/disease progression factors of the patients?) Likewise, more clarification on the highriskflag column would be very much appreciated.
Which interval scale should the prediction adhere to, are we expected to output probabilities, likelihoods, any number?
Many thanks in advance,
Asan
study patient predictionscore highriskflag
MMRF MMRF_1021 7.95518535355131 FALSE
MMRF MMRF_1024 7.64328116526648 FALSE
MMRF MMRF_1029 7.59964358015746 FALSE
MMRF MMRF_1030 7.42126240254235 FALSE
Created by Asan Agibetov asan.agibetov Hi Brian,
Thank you for your clarifications regarding the prediction scales. Actually, by re-reading the description of the challenge and parallel threads in the discussion forums (in particular, "Determination of progression or death"), I think I have understood what is it that we need to predict:
"Identify high risk patients defined by disease progression (or death) within 18 months from time of diagnosis"
Could you please consider answering questions raised in the "Determination of progression or death" thread? Hi Asan,
I have updated the Wiki to attempt to clarify this here:
https://www.synapse.org/#!Synapse:syn6187098/wiki/449445
This is what it now says:
A continuous prediction score should be output in column predictionscore. The prediction scores are not assumed to be on any particular scale. Rather, model confidence in assigning high risk should increase monotonically with prediction score--i.e., for two patients, the one with the numerically higher prediction score is at higher risk.
A dichotomized prediction score should be output in column highriskflag, where TRUE indicates that the associated patient is classified as high risk.
Please let me know if that is not clear.
Thank you,
Brian