[[Depression]] [[Digital Phenotyping]] Source [[Sun et al. - 2023 - Challenges in Using mHealth Data From Smartphones .pdf]] **Challenges**: 1. Patient engagement is important to ensure long-term comprehensive data collection 2. What constitutes an acceptable threshold of missing data? 3. How to distinguish what features are useful in tracking how health changes over time, and any way to identify people who are at higher risk? 4. How can we understand how depression affects people differently, though we are measuring the same signals? **Summary**: 1. The researchers determined that in a 14-day data collection period, having data for at least 8 days is sufficient to ensure that the analysis remains valid and reliable. 2. Cross-section and Longitudinal data may seem contradictory: Simpson's Paradox. 1. **Sleep Onset Time**: This feature had a better correlation with PHQ-8 scores when looked at cross-sectionally (i.e., at a single point in time) than longitudinally (over a period of time). This suggests that ==the time when participants fell asleep was more closely linked to their depression scores at a specific time rather than over time.== 2. Disturbed sleep: HOWEVER, over time (looking at longitudinally), patient having disturbed sleep: the amount of time patients spent awake after first falling asleep was more consistently related to their levels of depression. 3. So, depending on the purpose and intention, we must consider how to analyse the data. (Cross section or longitudinal) **There are three cluster of patient with depression** Out of 623 patients. Cluster 1 : Participants who sleep longer, walk less and wake up later when depressed Cluster 2 : Only show marginal changes in behaviour during depression Cluster 3: Individuals reduce the time and frequency of smartphone use when depressed. --- My Questions - Borrowing the idea from Victor Frankl's Dimensional Ontology. Due to the heterogeneity of mental illness, although patients are diagnosed with "depression", it may have different "causes", or it may manifest differently. 1. ![[Book - The Will to Meaning#^3999db]] - Do we need to have different algorithms to detect patterns for depression and psychosis? Can this really be trans-diagnostic? - They use Smartphone GPS, Fitbit step count, Fitbit sleep, and smartphone use. - [[DP - Sleep]], [[DP - Accelerometry]] - to deduce physical activity, [[DP - Device usage]] useful for depression --- Related - [[Paper - Correlations between objective behavioural features collected from mobile and wearable devices and depressive mood symptoms in patients with affective disorders systematic review]]