About helix parameterization used in STAR (this what I basically copied to StPicoHelix from StHelix) can be found in the appendix A (page 109):

Now let's ask whether the primary vertex sits inside or outside that circle. If it is inside, then the inclusion of the primary vertex in the momentum fit to get the
primary track pT will decrease the radius of the circle, i.e. bias towards pTprimary < pTglobal. However, if the  primary vertex sits outside this circle then the
opposite is the case, pTprimary > pTglobal. There is no upper limit on pTprimary from this mechanism. A very large shift from this mechanism may be rare,
but then so are true high pT tracks, and low pT weak decays are much more common. Eventually, at very high pT this is the single largest background to high pT tracking. 

I hope it is clear that there are two equivalent ways to discriminate this background: either cut on a large mis-match between pTprimary and pTglobal,
or cut on the signed gDCAs like you just generated, with the cut having different sign depending on the charge of the track (since pos and neg curve oppositely).
A cut on large mis-match between pTprimary and pTglobal would be fuzzy and vague in meaning, while a cut on gDCAs addresses the mechanism directly and is strongly preferred, I think.
 

To show this correlation, I suggest that you plot pTprimary vs pTglobal separately for  pos and neg gDCAs and pos and neg tracks (four 2-d plots in total).
If the above picture is correct we should see a large off-diagonal contribution in pTprimary vs pTglobal for gDCAs<0 for pos tracks and gDACs>0 for neg tracks,
but not for the other two combinations.