From my point of view, No you can't predict earthquake using total electronic content (TEC) derived using GPS. Earthquake phenomenon is more complicated to be predicted in such way. You may learn more about fracture and active areas but not to predict an earthquake especially when talking about occurrence of time.
Please, you may check the work by relevant experts in the field like Sergey Pulinets (I attach 3 screenshots with some of his references), Dimitar Ouzounov and Manuel Hernandez-Pajares (I attach links to their ResearchGate webpages). I think they would say it is possible.
Regretably, NO! Major roblem is that viable cause/effect mechanism does not exist - only untested (or untestable) hypotheses. If you do not know the cause (Lithospere - Ionosphere Coupling) it is impossible to constrain the effect (TEC changes). Also, other factors affecting TEC, some of them significantly, are not all understood and some may even still be unknown. Minor problem is that TEC computation with current methods is, least to say, simplistic - e.g. do not account for Ionosphere's gyrotropy although this can be fixed. In consequece, published "results" are largely anecdotal of nature, based on poorly contrained experiments that do not yield repeatable results. I'm sorry to have to be so blunt, but this line of research is still at its early infancy and there's a good chance that it'll never mature (like VAN which a lot of people prefer to forget).