Thanks for the tip, I had found an application note very similar to that one but targeting the INA138. Additionally, I did read Section 7.4.4 of the INA21x datasheet, but it made it seem more like a protection circuit than for sustained use. I'm also looking into the LM5066 at @Bharbour suggestion, though price/complexity might rule that one out.

You could float the INA21x monitoring a current shunt at the top rail, using a regulator to create the "ground" at maybe (Vrail - 5V).  In fact, you could float the entire current sensor including the display circuitry, or use an opto-isolator or RF module (even ESP8266) to transmit the data from the sensor to wherever you need.

Here's an example of the technique, with different parts: https://www.analog.com/en/analog-dialogue/articles/wireless-current-sense-circuit-floats-with-sense-resistor.html

In their case, they used 2x AA batteries for power, but you could come up with a regulator.

For a different approach, see also:

http://www.ti.com/lit/an/sboa295/sboa295.pdf

Thanks for the tip, I had found an application note very similar to that one but targeting the INA138. Additionally, I did read Section 7.4.4 of the INA21x datasheet, but it made it seem more like a protection circuit than for sustained use. I'm also looking into the LM5066 at @Bharbour suggestion, though price/complexity might rule that one out.

The part is designed as a hot swap power controller. I have used it as an eFuse in a couple of applications. We used to use the low voltage version in some crazy high current systems , like 250A. We tested them by slamming a pair of pliers across the output. It would shut down without trashing the FETs or the power supply. It was tough on the pliers though. The LM5066 should run without the FET in place. It will squawk about a shorted FET, but it will still read the telemetry fine.

Alternately, there is a measurement only version of that part, the LM5056 that doesn't expect a FET, so it won't squawk about it being shorted. I have not used the LM5056 part as much, so I kind of forgot about it.

The measurement accuracy at low currents on these parts stinks. They have a 12 bit adc, and the current measurement does not go below 64 counts or so. If you are pushing 3000 counts, that 64 is insignificant, on the other hand if you are only running 150 counts, the accuracy is pretty dismal. 

Those floating ground arrangements for running a low voltage part at a high voltage creep me out. one slip with a scope ground and the smoke gets away. They can be made to work though.

Here's the other resource for reference. http://www.ti.com/lit/ug/tidu833/tidu833.pdf