In addition to the above comments
(1) Since you use NPN, when temperature goes higher the NW-->Sub leaskage will go up. You need to check the NW-->SUB current through tne NWell node. You can also compensate it, but it won't be easy to get it accurate. This is very depending on modeling or PDK.
(2) You use big ration between 2 BJT. You need to check the IC vs VCE curve to make sure the the bias point is not close to the "corner or border" between active and saturation. Usually that area is not modeled very well and may have bigger variation.