You will have to do things that reduce the attenuation of the lines. This includes lower loss dielectric material. Lower dielectric constant (so the conductor traces are wider.)
in addition to the method explained by Flatulent, depending on a type of your filter, you might use shapes different from ractangular coupled lines. For example, microstrip disk resonators can have Q factors 4 times (!!) greater than half wavelength line resonator.
Unfortunately there is few to do.
I don't agree flatulet when he wrote reduce εr.
The way is reduce the losses!
So you may reduce tanδ and ρ (metal resistivity).
But practically, for many frequency the losses of the copper dominate the losses of the PFTE dielectric.
The only way for a large Q improvement is reduce the metal losses, ...so Superconductors. Usually HTS structures has 1/10 of the losses of a Roger 5880
Reducing er has the effect of widening the traces to maintain the impedance which reduce conductor loss. Increasing the dielectric thickness also widens the traces.
The calculator by Rogers Corp is very good and sorts out all of the loss components.