Smartphone accelerometer data as a proxy for clinical data in modeling of bipolar disorder symptom trajectory

Abstract

Being able to track and predict fluctuations in symptoms of mental health disorders such as bipolar disorder outside the clinic walls is critical for expanding access to care for the global population. To that end, we analyze a dataset of 291 individuals from a smartphone app targeted at bipolar disorder, which contains rich details about their smartphone interactions (including typing dynamics and accelerometer motion) collected everyday over several months, along with more traditional clinical features. The aim is to evaluate whether smartphone accelerometer data could serve as a proxy for traditional clinical data, either by itself or in combination with typing dynamics. Results show that accelerometer data improves the predictive performance of machine learning models by nearly 5% over those previously reported in the literature based only on clinical data and typing dynamics. This suggests it is possible to elicit essentially the same “information” about bipolar symptomology using different data sources, in a variety of settings.

A Novel Approach to Clustering Accelerometer Data for Application in Passive Predictions of Changes in Depression Severity

“…Accelerometer features were calculated from the clustered data to determine their efficacy in predicting clinically relevant changes in PHQ score when combined with typing and clinical features. Model accuracy ranged from 94 to 95.5% (2% standard deviation), with 96 to 98% AUC, similar to that reported in [ 33 ]. Importantly, the accelerometer features calculated from the cluster labels contributed greatly to model predictions, even performing better than some otherwise highly ranked clinical features previously reported in the literature [ 12 , 49 ], as can be seen in Table 2 .…”

“…For comparison, missing values were accounted for via two methods: (1) imputation and (2) filtering out of individuals with missing data. Generally, models using imputation performed a few percentage points higher than filtering in terms of accuracy and area under the ROC curve (AUC), which is consistent with previous results [ 12 , 33 ]. In total, there were 295 individuals available in this dataset when using imputation, but only 100 left when using filtering.…”

“…All data collected are de-identified. The data have been used in previous BiAffect studies that found relationships between smartphone keyboard typing patterns and mood disorders [ 12 , 20 , 21 , 22 , 29 , 30 , 31 , 32 , 33 ].…”

Unveiling the digital phenotype: A protocol for a prospective study on physical activity behavior in community-dwelling older adults

“…The cognitive testing segment, focusing on reaction time, has undergone clinical evaluation and demonstrated reliability and validity, comparing favorably to the standard Computerized Test of Information Processing assessment. However, the capacity of SWAY to function consistently across various mobile devices and operating systems is yet to be validated (52,53,54), therefore it will be use to collect all the data.…”

CoDuBoTeSS: A Pilot Study to Eradicate Counterfeit Drugs via a Blockchain Tracer Support System on the Nigerian Frontier

“…Additionally, tracing and linking knowledge to materials cum constituents for each drug manufactured, without bias becomes difficult due to the intricacy of stakeholder interactions. The lack of a secure decentralization bank as split across various stakeholders often degrades performance and ensure access difficulty by stakeholders to records [33]- [35]. Song et al [36] access to medical product records via efficient tracing of drug constituents/composition along its supply chain helps to prevent negative incidents/impacts of counterfeit drugs.…”