Decreased step count prior to the first visit for MDD treatment: a retrospective, observational, longitudinal cohort study of continuously measured walking activity obtained from smartphones

Abstract

IntroductionMajor depressive disorder (MDD) is a common debilitating psychiatric condition and a major cause of productivity loss in workers. Using intermittent, subjective indicators, previous studies have shown that physical activity can predict lower levels of depressive symptoms. However, there is an unmet need for continuous and objective measures to identify MDD development before it results in productivity loss. The aim of this study was to elucidate the association between continuously measured walking activity and the development of MDD.MethodsThis retrospective, observational, longitudinal cohort study used health insurance claims data. Individuals aged 20–74 years were included if they had a record of MDD diagnosis and daily step count data for the 60 days before and after the first recorded MDD-related visit, which was defined as the index date. Multivariate analysis was conducted to compare 7-day moving averages of step counts on each day of the analysis period with the mean step count on the index date. Joinpoint regression analysis was used to determine when the trajectory of the moving step count average changed (inflection point).ResultsIn total, 2,143 patients with a mean age of 41.2 (standard deviation [SD]: 10.6) years were included. The majority of patients were men (69.5%) and employed full-time (94.1%). Antidepressants were prescribed for 59.2% of patients. The 7-day moving average step count decreased from 6,310 (SD: 3758) at day −60 to 5,879 (SD: 3183) at the index date (first recorded MDD-related visit), and then increased to 6,062 (SD: 4029) at day +60. Compared with the index date, the 7-day moving average of step counts was significantly higher at days −60 to −1, +23 to +33, and + 42 to +60, and significantly lower at days +2 and + 3. Joinpoint regression analysis of 7-day moving average step counts from day −60 to day 0 identified an inflection point at day −14.ConclusionIn working-age Japanese people, a formal diagnosis of MDD was preceded by a notable decline in daily step counts by approximately 2 weeks. MDD diagnosis and (presumed) treatment were followed by a gradual increase in daily step counts.

From smartphone data to clinically relevant predictions: A systematic review of digital phenotyping methods in depression

“…Our decision to present analyses in this section was also informed by whether the overall study had already had key analyses addressed in earlier sections of the Results. Fujino, Tokuda, & Fujimoto (2023) investigated changes in smartphone step count surrounding MDD-related medical visits in a group-based analysis, finding that mean daily step count tended to decrease in the two weeks before the visits. Zou, et al, 2023 predicted treatment response in MDD patients 10 weeks in advance, achieving an AUROC (Area Under the Receiver Operating Characteristic Curve) of 0.65.…”

“…For example, by mentioning the use of “rich sensor data” (Cao, et al, 2020), or stating that analyses were only performed when sufficient data was deemed to be available (Braund, et al, 2022). Other studies explicitly stated thresholds for data inclusion (Bai, et al, 2021; Faurholt-Jepsen, et al, 2022; Fujino, Tokuda, & Fujimoto, 2023; Kathan, et al, 2022; Knights, Shen, Mysliwiec, & DuBois, 2023; Laiou, et al, 2022; Sun, et al, 2023; Zhang, et al, 2021; Zhang, et al, 2022; Zou, et al, 2023); if too high a percentage of data for a sample was missing then the sample was excluded. There was no consistent selection of thresholds for data inclusion across studies.…”

“…Moreover, few studies aimed to predict responses across time points. Of note, Fujino, Tokuda, & Fujimoto (2023) carried out a time-resolved analysis that investigated changes in step count surrounding MDD-related medical visits. Finally, a small number of studies investigated individual differences, demonstrating differences between participants of different ages and employment statuses (e.g., in time spent at home (Zhang, et al, 2022; Laiou, et al, 2022)), as well as some gender differences (e.g., in location entropy and residential location count (Zhang, et al, 2022)).…”

“…As is the case for other health applications, an open question in digital phenotyping research is whether to develop individual or group models, or perhaps combinations of the two. This review identified studies that predominantly applied group models, with individual factors commonly addressed in models as covariates (Faurholt-Jepsen, et al, 2022; Fujino, Tokuda, & Fujimoto, 2023; Zhang, et al, 2021; Zhang, et al, 2022), or as predictors of interest themselves (Kim, et al, 2023; Laiou, et al, 2022; Pellegrini, et al, 2022). Three studies compared group models to individual models (Cho, et al, 2019; Kathan, et al, 2022; Pedrelli, et al, 2020).…”