The device saves power by putting the EFM32 into energy mode 2 for most of the time and only waking it up to perform measurements and to transmit them. The rf24 module also has a sleep mode, which can be activated by setting a bit in its config register. The HDC is already configured to stay in a deep sleep mode until its i2c address is transmitted. When I was still measuring 19µA on the device, I was a little concerned. 

But as it turns out, disconnecting the debug pins reduces the measured current draw significantly to a more realistic 2.6µA. There also was a little charge-discharge-curve visible after the data is transmitted. Setting the chip select pin low somehow removes this bump. I am still not sure what's the explanation for this. 

I love how the starter kit board measures and displays the current over time. You can clearly see a smaller 100µA peak where the EFM32 wakes up to request a measurement, then goes to sleep again. The second peak with the rf24 data transmission is much higher. It could be a bit more than the 3mA displayed here, but the transmission takes only a couple of milliseconds.

With the 2.6µA current consumption the device would theoretically run for more than 8 years. This does not take into account the self discharge of the coin cell (only around 1% per year actually).