Responder Classification of Patients Engaged With a Digital Medicine System Based on Data Quality Footprints
This poster was presented at the 30th annual Psych Congress, held Sept. 16-19, 2017, in New Orleans, Louisiana.
The convergence of technology and traditional pharmaceutical-based therapy promises to beckon a new era of patient treatment options. However, an appropriate definition of a 'responder' to this type of treatment does not currently exist.
In this work, we leverage data collected from two clinical studies in serious mental illness (SMI) with Otsuka's digital medicine (DM) system (NCT02722967, NCT02219009) to: i) propose an initial data quality footprint as a means for clustering patients into appropriate 'responder' sub-groups; ii) explore clinical, demographic, and actimetry-based metrics that further distinguish one group from another; and iii) assess the potential for predicting data-based response from available data.
Three core patient sub-groups were identified, representing high-, moderate-, and low-performers (responders): These sub-groups also displayed distinct patterns in actimetry, clinical, and demographic variables. Random forest models were able classify responder-group at roughly 89% accuracy with 8-week retrospective data, and 77% accuracy using data from the first two weeks of data.
Our work represents a novel way of defining a 'responder' for patients engaged with DM systems, provides insights as to initial characteristics/behaviors of patients who are successfully engaging, and provides evidence that prediction of engagement may be made as early as two weeks after the initiation of treatment.