The formula should read f[GHz]<=75/(h[mm]Sqrt(er-1)) for single grounded dielectric layer. And this condition doesn't imply that there is no PROPAGATING surface wave. Instead, the dominant TM surface wave has zero-cutoff, and the above condition implies there is only this dominant TM surface wave propagating in a grounded dielectirc slab, all higher order modes are evanscent. This is not related to the metal trace. In orther words, metal traces (like the microstrip) are removed when doing this eigenvalue analysis.
The reference should be: J R Mosig and F E Gardiol, "A dynamical radiation model for microstrip structures," Adv. Electron. Electron Phys. Vol 59, pp.139-237, 1982.