Whose Lives Do Our Datasets Represent?
We also need to ask whose lives our datasets represent, and whether the methods we develop work across different populations and settings.
A model trained to recognise sleep or activity in one population may perform less accurately in another. Working patterns, transport, household tasks and sleep routines can differ, changing the signals a device records. Before we can interpret what those patterns might mean for mental health, we need confidence that we are measuring them accurately.
The discussion highlighted that diverse datasets are essential, alongside local testing and, where needed, adaptation of the methods.
Practical Realities in Low-Resource Settings
There are practical considerations too. In some low-resource settings, the cost of devices, access to compatible phones, reliable electricity and affordable connectivity can make participation harder. Studies need to budget for charging, repairs and technical support, and consider devices that can store data without a constant internet connection. Instructions and support also need to suit local languages and levels of familiarity with the technology.
These circumstances vary considerably within and between low- and middle-income countries. Working with local researchers and communities is therefore essential to choosing an approach that fits the setting.
Building Evidence Alongside Trust
The potential is exciting. Realising it means building evidence alongside trust, with people involved from the start.
One thing to flag is that the opening line refers to “that question”, so it reads as though a question or title sits just before it. If that’s not already in place, the first section may need a lead-in.