LO feedthru can be due to dc offset in your mixer. check your layout for systematic offsets and run some monte-carlo simulations. Another major source will be crosstalk, be it in the silicon or in the package. Good layout planning (i.e having tx output bodnwires perpendicular to the lo carrying (the LO itself or LO supply) bondwires should help

If you have access to dc bias points, triming dc offset can reduce the leakage to crosstalk induced leakage.

Also when you say 5 degree mismatch, what do you mean ? duty cycle ? this can be improved by improving oscillator design and layout. it usually higher level harmonics or crosstalk non-symmetry which impacts fundamental duty cycle.

cheers
aw

u can reject this LO feedthrough by high pass filter at the output of the Mixer ,  other way  u can use subharmonic mixer , so the LO will be half the 4.5 Ghz  and u use the 2nd harmonic

khouly