Increasing the Value of Digital Phenotyping Through Reducing Missingness: A Retrospective Analysis

Abstract

ObjectivesDigital phenotyping methods present a scalable tool to realize the potential of personalized medicine. But underlying this potential is the need for digital phenotyping data to represent accurate and precise health measurements. This requires a focus on the data quality of digital phenotyping and assessing the nature of the smartphone data used to derive clinical and health-related features.DesignRetrospective cohorts. Representing the largest combined dataset of smartphone digital phenotyping, we report on the impact of sampling frequency, active engagement with the app, phone type (Android vs Apple), gender, and study protocol features may have on missingness / data quality.SettingmindLAMP smartphone app digital phenotyping studies run at BIDMC between May 2019 and March 2022Participants1178 people who partook in mindLAMP studiesMain outcome measuresRates of missing digital phenotyping data.ResultsMissingness from sensors in digital phenotyping is related to active user engagement with the app. There are small but notable differences in missingness between phone models and genders. Datasets with high degrees of missingness can generate incorrect behavioral features that may lead to faulty clinical interpretations.ConclusionsDigital phenotyping data quality is a moving target that requires ongoing technical and protocol efforts to minimize missingness. Adding run-in periods, education with hands-on support, and tools to easily monitor data coverage are all productive strategies studies can utilize today.Strengths and Limitations of this Study○Methods are informed by a large sample of participants in digital phenotyping studies.○Methods can be replicated by others given the open-source nature of the app and code.○Methods are informed by only mindLAMP studies from one team which is a limitation.

Digital Phenotyping Data to Predict Symptom Improvement and App Personalization: Protocol for a Prospective Study

“…Like our prior studies, our research is fully reproducible. We offer details of our recruitment process and procedures in this paper that outlines details of our recruitment, screening, and data coverage procedures [ 12 ]. The mindLAMP app remains open-source software currently deployed at over 50 clinical sites worldwide, and our algorithms are also publicly accessible via GitHub [ 27 ].…”

“…This run-in period will serve to screen out participants whose devices are not able to capture digital phenotyping data or do not engage with the app at all, and give the study coordinators time to verify that informed consent is signed and dated correctly. The run-in period is designed to help improve overall digital data coverage that is important for validation of the predictive model [ 12 ]. After these 3 days, participants who have completed the required surveys and have sufficient GPS data will be moved to the enrollment period of the study.…”

“…To enable scalable research, we will build upon the digital study infrastructure used in our prior studies [ 12 ]. All parts of the study will be automated via workers implemented in Python.…”

The Feasibility of Implementing Remote Measurement Technologies in Psychological Treatment for Depression: Mixed Methods Study on Engagement

“…The passive data streams most vulnerable to missing data were the GPS and Bluetooth sensors. However, other passive sensing studies on mental health have found the opposite pattern, with more data being available from GPS than from accelerometers [ 15 , 20 , 37 ]. Sensor noncollection can occur for multiple reasons, including participants turning off the data permissions or the sensor itself.…”

“…This, in turn, has implications for the integrity of the constructed variables and for understanding the potential sources of biases in the data. Sparse active data points on mood questionnaires can affect how ground truth is determined, whereas less passive data can result in inaccuracies in how features are derived and the resulting data analysis (refer to Currey and Torous [ 20 ] for an example of this).…”