Datasets of Smartphone Modalities for Depression Assessment: A Scoping Review

As mobile sensing research for depression assessment is rapidly expanding, it is vital to uncover fundamental patterns and overarching limitations in the datasets available for such research, especially as many analyses are conducted on the same datasets. As such, we conduct a scoping review by identifying 80 datasets through the start of 2024 that contain smartphone modalities and depression labels. These datasets originated from 72 manuscripts and approximately 60 research groups. The most collected smartphone modalities included location and/or activity for 68.75%, communication logs for 47.5%, phone use for 41.25%, vocal utterances for 30%, and WiFi and/or Bluetooth connectivity for 28.75% of the datasets. The PHQ-8 and PHQ-9 (Patient Health Questionnaire) were the most popular screening instruments, administered for 53.75% of datasets. Of the 80 datasets, 31.25% recruited from student populations, 22.5% from patient populations, and 7.5% recruited crowdsourced workers. Excluding the nine datasets that did not report demographics, 73% reported a majority of women participants. Given the importance of datasets on analysis outcomes, this scoping review is an invaluable resource to inform the state of science and guide future mobile health research.