JMIR Publications

ABSTRACT

Background:

Brain related disorders are characterized by observable behavioral symptoms. Smartphones can passively collect objective behavioral data, avoiding recall bias. Despite promising clinical utility, analyzing smartphone data is challenging as datasets often include a range of missingness-prone temporal features.

Objective:

Hidden Markov Models (HMMs) provide interpretable, lower-dimensional temporal representations of data, allowing missingness. We aimed to investigate the HMM as a method for modeling digital phenotyping time series data.

Methods:

We applied an HMM to an aggregate dataset of smartphone measures designed to assess social functioning in healthy controls (HCs), participants with schizophrenia, Alzheimer’s disease (AD) and memory complaints. We trained the HMM on a subset of HCs (n=91) and selected a model with socially “active” and “inactive” states, then generated hidden state sequences per participant and calculated their “dwell time”, i.e. the percentage of time spent in the socially active state. Linear regression models were used to compare the dwell time to social and clinical measures in a subset of participants with available measures, and logistic regression was used to compare dwell times between diagnostic groups and HCs.

Results:

We identified lower dwell times in AD (n=26) versus withheld HCs (n=156) (odds ratio=0.9455, FDR corrected P<.001) and higher dwell times related to increased social functioning questionnaire scores in HCs, with every one percent increase in dwell time reflecting a 0.1248 increase in Social Functioning Scale score (n=12, FDR corrected P=.004). No significant relationships regarding dwell time were identified for participants with schizophrenia (n=18) or memory complaints (n=57).

Conclusions:

We found the HMM to be a practical, interpretable method for digital phenotyping analysis, providing an objective phenotype that is a possible indicator of social functioning.