Most likely the substrate Er is NOT what you thought it was.
You need to get some material, measure its Er (some use a simple circuit, like a 1 pole resonator), and when you are happy with the Er then make the full circuit with the SAME LOT NUMBER material.
It is possible that you did not properly model the open circuit end and stepped width discontinuity effects.
If you are desperate to ship some product, you might be able to paint thin layers of varnish or conformal coating or lay a thin dielectric sheet over the entire filter to move the frequency down. The varnish acts to raise the effective Er of the substrate, thereby lowering the frequency of the measured passband.