From the IN828 datasheet we have this (in your case G=1 so the black diamond):
You need to stay within the 6-sided diamond figure.
If the highest possible output voltage is +/-100mV then you need to keep GND2 and (Vd-I*R/2) within about +/-3.5V of each other. If it is +/-3V then more like within +/- 2.5V of each other. It's up to you to maintain that relationship, and if you don't the amplifier will not work properly (it will saturate in one way or another).
Floating the entire front and circuit and using optical or other means to transmit the signal, as Andy suggests, is a possibility (maybe the only practical one, depending on your situation) but it comes with many possible complications. You still have to maintain the common-mode voltage as shown in the figure (so maybe you need a direct connection to Vd or a buffer amplifier to maintain the CM input voltage as with bio signals) and if you use something more pleasant than batteries such as a DC-DC converter it could disrupt the measurement due to AC noise and capacitive coupling. The means of transmitting the signal could also disrupt the measurement. A nA signal, while not inherently difficult to measure, could be challenging to galvanically isolate.
One off-the-shelf possibility might be a battery-powered DMM with serial interface, but there will still be some coupling through the interface which could disrupt the measurement. COM for the meter would best be connected to Vd, most likely, depending on your setup.