Relationship between missing data and bipolar disorder (BD) symptom severity in digital phenotyping

Principal Investigator: Katherine E. Burdick, PhD, Jessica M. Lipschitz, PhD

Authors: Rachel Van Boxtel, BS
Lay Abstract

Background: Digital phenotyping—gathering moment-to-moment data using digital tools—has gained momentum as a way to better understand disease trajectory. However, these methods rely on participants wearing devices (e.g., Fitbits) consistently, and there are many factors that can limit an individual’s participation. Thus, missing data are common and can compromise the conclusions that can be drawn.

Methods: We collected passive sensor data (via Fitbits) and active data (bi-weekly ratings of depression and mania symptom severity) in 12 patients with bipolar disorder over 9 months.

We compared the mean depression (PHQ-8) and mania (AMRS) scores for periods of compliance (defined as wearing the Fitbit for over 75% of the biweekly period) to periods of non-compliance (under 75% Fitbit wear).

Results: Depression symptom severity was higher during periods of non-compliance and results were nearly statistically significant despite the small sample size. Manic symptoms were not significantly related to changes in compliance.

Conclusions: Results are preliminary, but they indicate a trend whereby heightened depressive symptoms in BD patients may be associated with noncompliance with passive sensor data collection. Thus, these data might not be missing at random. In the future, this question should be evaluated in larger groups of patients.

Scientific Abstract

Background: Digital phenotyping has gained momentum as a way to better understand disease trajectory. However, these methods require compliance with passive sensor data collection (e.g., wearing tracking devices like Fitbits). Missing data are common and can compromise the conclusions that can be drawn.

Methods: We collected passive sensor data (via Fitbits) and active data (bi-weekly ratings of depression and mania symptom severity) in 12 patients with bipolar disorder over 9 months.

We compared the mean depression (PHQ-9) and mania (AMRS) scores for periods of compliance (defined as wearing the Fitbit for over 75% of the biweekly period) to periods of non-compliance (under 75% Fitbit wear).

Results: Depression symptom severity was higher during periods of non-compliance and results were nearly significant despite the small sample size (T=-1.80, p= 0.05). Manic symptoms were not significantly related to changes in compliance (T=-1.39, p=0.1).

Conclusions: Results are preliminary and, while they fail to reach a <.05 significance level, they indicate a trend whereby heightened depressive symptoms in BD patients may be associated with noncompliance with passive sensor data collection. Thus, it may not be appropriate to consider these data to be missing at random. Larger sample sizes are required to fully evaluate this question.

Clinical Implications
Many initiatives are underway to use tracking devices, like Fitbits, for insight into disease trajectory of various chronic conditions. Our findings suggest missing data may be related to worsening symptoms, which shifts how researchers and clinicians should interpret such data.

If the PDF viewer does not load initially, please try refreshing the page.

Comments are closed.