The LT1881 has an
input offset voltage of ~30uV typical, so -19uV is well within spec.
Similarly, the input bias current is ~100pA, so that through a 1M resistor would give an additional 100uV offset, which is pretty much what you're seeing.
So all in all what you're seeing is about what one would expect to see for a LT1881. It's maybe not the greatest device, but the design is only meant for 10V to 5 decades, which means a basic uncertainty of +/-50uV so it'll certainly do the job.