Peer Review #2 of "The relationship between mobile phone location sensor data and depressive symptom severity (v0.1)"

Sohrab Saeb, Emily G. Lattie, Stephen M. Schueller, Konrad P. Körding, David C. Mohr, S Abdullah, M Matthews, E Murnane, G Gay, T Choudhury, J Ayuso-Mateos, R Nuevo, E Verdes, N Naidoo, S Chatterji, E Basch, P Torda, Adams, L Canzian, M Musolesi, A Ferrari, F Charlson, R Norman, S Patten, G Freedman, C Murray, T Vos, H Whiteford, W Fleeson, S Gilbody, T Sheldon, A House, P Greenberg, A Fournier, T Sisitsky, C Pike, R Kessler, A Grunerbl, A Muaremi, V Osmani, G Bahle, S Ohler, G Troster, O Mayora, C Haring, P Lukowicz, W Chiu, O Demler, K Merikangas, E Walters, S Zhao, D Blazer, M Swartz, K Kroenke, R Spitzer, Williams Jbw, J Kwapisz, G Weiss, Moore Sa, R Likamwa, Y Liu, N Lane, L Zhong, J-k Min, J Wiese, J Hong, J Zimmerman, W Mischel, J Ho, J Duffecy, K Baron, K Lehman, L Jin, D Reifler, J Rauthmann, D Gallardo-Pujol, E Guillaume, E Todd, C Nave, R Sherman, M Ziegler, A Jones, D Funder, C Reynolds, Frank, J Ruwaard, M Ejdys, N Schrader, M Sijbrandij, H Riper, M Zhang, C Karr, M Corden, Mohr Dc ; AOpenalex Unknown 2016 · DOI · OpenAlex

Abstract

Background: Smartphones offer the hope that depression can be detected using passively collected data from phone sensors. This study aimed to replicate and extend previous work using geographic location (GPS) sensors to identify depressive symptom severity. Methods: A dataset was collected from 48 college students over a 10-week period, including GPS sensor data and the Patient Health Questionnaire 9-item (PHQ-9) to evaluate depressive symptom severity at baseline and end-of-study. GPS features were calculated over the entire study, for weekdays and weekends, and in 2-week blocks. Results: The study replicated previous findings that GPS features, including location variance, entropy, and circadian movement, were significantly correlated with PHQ-9 scores ($r$ ranging from -0.43 to -0.46, $p < .05$). These relationships were stronger when calculated from weekend data compared to weekday data. Correlations with end-of-study scores remained significant regardless of the time point used to calculate features. Discussion: Findings suggest GPS features are reliable predictors of depressive symptom severity. The varying strength of these relationships on weekends vs. weekdays suggests a moderating role of volitional movement. The ability to predict severity up to 10 weeks prior to assessment suggests potential for early warning signals.

Introduction

Depression is a leading cause of disability worldwide, yet detection often relies on patients recognizing their own symptoms and seeking help. This reactive approach leads to significant delays; in the US, the median time to treatment is eight years. Passive behavioral sensing via smartphones offers a continuous monitoring alternative. This paper addresses two main challenges: the replication of findings across different populations (e.g., from urban adults to small-town students) and the impact of temporal context (weekdays vs. weekends) on predictive accuracy.

Methods and Data

The study utilized the StudentLife dataset, tracking 48 college students over 10 weeks. Researchers extracted 11 GPS features, including:

  • Location Variance: Total geographic area covered.
  • Entropy: How time is distributed across different locations.
  • Circadian Movement: Regularity of movement following a 24-hour cycle.
  • Home Stay: Percentage of time spent at home.

Depressive symptom severity was measured using the PHQ-9 at the start (baseline) and end (follow-up) of the study.

Distribution and individual changes of PHQ-9 scores from baseline to follow-up.

To ensure statistical robustness, bootstrapping with 1,000 subsamples was used to estimate correlation coefficients ($r$) and 95% confidence intervals.

Key Results

The analysis confirmed several significant correlations between GPS features and follow-up PHQ-9 scores:

  • Circadian Movement ($r = -0.48$): Irregular daily movement patterns were associated with higher depression scores.
  • Entropy ($r = -0.46$): Limited or unevenly distributed location variety linked to higher symptom severity.
  • Location Variance ($r = -0.43$): Smaller total geographic range correlated with higher depression.
  • Home Stay ($r = 0.43$): Increased time at home was a significant indicator of severity.

The "Weekend Effect"

GPS features were significantly stronger predictors when derived from weekend data. For instance, Weekend Circadian Movement showed a strong correlation ($r = -0.51$), whereas the weekday correlation was notably weaker ($r = -0.28$). This suggests that "volitional" movement—unconstrained by weekday obligations like classes—more accurately reflects internal motivational states.

Early Warning Potential

GPS patterns from the start of the study remained significantly correlated with depression scores measured at the end. This indicates that changes in mobility may serve as early warning signals, potentially preceding a patient's conscious awareness of worsening symptoms.

Discussion

This study successfully replicated previous findings, suggesting that features like location variance and circadian rhythm are robust "digital biomarkers" for depression. However, limitations include a small, predominantly male sample size and data sparsity issues. Future work should refine the "volitional movement" hypothesis by incorporating personalized schedules. These findings pave the way for "just-in-time" clinical interventions triggered by passive sensor data.

Summary updated

Abstract

Background.Smartphones offer the hope that depression can be detected using passively collected data from the phone sensors.The aim of this study was to replicate and extend previous work using geographic location (GPS) sensors to identify depressive symptom severity.Methods.We used a dataset collected from 48 college students over a 10-week period, which included GPS phone sensor data and the Patient Health Questionnaire 9-item (PHQ-9) to evaluate depressive symptom severity at baseline and end-of-study.GPS features were calculated over the entire study, for weekdays and weekends, and in 2-week blocks.Results.The results of this study replicated our previous findings that a number of GPS features, including location variance, entropy, and circadian movement, were significantly correlated with PHQ-9 scores (r's ranging from -0.43 to -0.46, p-values<.05).We also found that these relationships were stronger when GPS features were calculated from weekend, compared to weekday, data.Although the correlation between baseline PHQ-9 scores with 2-week GPS features diminished as we moved further from baseline, correlations with the end-of-study scores remained significant regardless of the time point used to calculate the features.Discussion.Our findings were consistent with past research demonstrating that GPS features may be an important and reliable predictor of depressive symptom severity.The varying strength of these relationships on weekends and weekdays suggests the role of weekend/weekday as a moderating variable.The finding that GPS features predict depressive symptom severity up to 10 weeks prior to assessment suggests that GPS features may have the potential as early warning signals of depression.
BibTeX
@inproceedings{saeb2016peer,
  author = {Sohrab Saeb and Emily G. Lattie and Stephen M. Schueller and Konrad P. Körding and David C. Mohr and S Abdullah and M Matthews and E Murnane and G Gay and T Choudhury and J Ayuso-Mateos and R Nuevo and E Verdes and N Naidoo and S Chatterji and E Basch and P Torda and Adams and L Canzian and M Musolesi and A Ferrari and F Charlson and R Norman and S Patten and G Freedman and C Murray and T Vos and H Whiteford and W Fleeson and S Gilbody and T Sheldon and A House and P Greenberg and A Fournier and T Sisitsky and C Pike and R Kessler and A Grunerbl and A Muaremi and V Osmani and G Bahle and S Ohler and G Troster and O Mayora and C Haring and P Lukowicz and R Kessler and W Chiu and O Demler and K Merikangas and E Walters and R Kessler and S Zhao and D Blazer and M Swartz and K Kroenke and R Spitzer and Williams Jbw and J Kwapisz and G Weiss and Moore Sa and R Likamwa and Y Liu and N Lane and L Zhong and J-K Min and J Wiese and J Hong and J Zimmerman and W Mischel and D Mohr and J Ho and J Duffecy and K Baron and K Lehman and L Jin and D Reifler and J Rauthmann and D Gallardo-Pujol and E Guillaume and E Todd and C Nave and R Sherman and M Ziegler and A Jones and D Funder and C Reynolds and Frank and J Ruwaard and M Ejdys and N Schrader and M Sijbrandij and H Riper and S Saeb and M Zhang and C Karr and S Schueller and M Corden and K Kording and Mohr Dc ; A},
  title = {{Peer Review \#2 of "The relationship between mobile phone location sensor data and depressive symptom severity (v0.1)"}},
  booktitle = {Openalex Unknown},
  year = {2016},
  doi = {10.7287/peerj.2537v0.1/reviews/2},
}
Figure