So you could sample the data with the boron that wakes from an interrupt, but does not connect to cellular except after 10 mins when you do and send the data - this should be more efficient than running a separate MCU board?
Look at this (search and there are many others on the same topic) thread