Let the alarming difference between those figures be a warning to you that FR4 is not really an RF capable material. You use it because it's cheap, and you compromise on the RF properties.
Many circuits can be built successfully even with a 10% spread in dielectric constant, transmission lines for high speed digital, RF amplifiers etc. If you want to build distributed filters, or achieve very low loss (have you looked at the tanδ column?) then you need to use a different material.
At the start of my RF career, we researched FR4 materials for use at 1 to 2 GHz, both consistency of \$\epsilon_r\$ from the same manufacturer, and across manufacturers, and determined that it wasn't good enough for signal generator manufacture (to make oscillators and filters), whereas G200 would just manage (it had a higher glass transistion temperature than FR4).
In the last couple of decades, low cost materials (R4350 for instance) have become available that can be processed like FR4 but with much tighter control of \$\epsilon_r\$ and much better loss tangent (shown as hydrocarbon/ceramic) in your linked table. Our more recent RF products use FR4 for the inner power/signal layers with R4350 on the outer layers for RF microstrip. This ability to mix materials keeps the cost down.
Note that the impedance figures for FR4 seem to have dropped any significant decimal places, and are given without a tolerance. Compare that to the way the \$\epsilon_r\$ figures are presented for hydrocarbon/ceramic.
You will have to turn the RF specifications for your particular application into an er tolerance, buy some FR4 samples, and see whether it delivers the consistency you need. If it doesn't, then you'll have to pay more for a better material.
You may be able to make the difference between FR4 working for you and not by choosing the layup of the board correctly. Most PCB fab houses will try to sell you core in the middle, with pre-preg and foil for the outer layers. Instead, insist on core for the outer layers. It's slightly more expensive, but factory-made core has far better repeatability than fab-house assembled pre-preg.
Avoid thin dielectrics in microstrip, as tolerances will get you both on the thickness of the dielectric, and on etching tolerances for the narrower track. If you can improve these two tolerances, you may get away with a larger \$\epsilon_r\$ tolerance in your error budget. Of course a wider track will increase the size of the circuit, another compromise to consider.