Did you miss me? Making the most of digital phenotyping data by imputing missingness with point process models

Digital phenotyping has broad clinical potential, providing low-burden objective measures of behaviour as individuals go about their day. However, progress in making clinical inferences from these data is severely challenged by the common occurrence of missing data. We investigated non-homogeneous Poisson Point Process Models (PPPMs) as a method for imputing missing temporal digital phenotyping data, considering smartphone-based activities as ‘points’. We assessed the inclusion of time-varying covariates (‘hour’ and ‘day’) in personalised PPPMs. We used participants from SMARD (n=26) for a ground truth evaluation, and PRISM (n=65) and Hersenonderzoek (n=283) for a replication analysis involving Hidden Markov Models, evaluating the effectiveness of PPPMs for imputation and their influence on downstream analysis. In the ground truth evaluation, PPPMs using one-hot encoded hour provided the highest out-of-sample likelihood. We successfully replicated findings from our prior work using this method, demonstrating that PPPMs using covariates are a promising imputation tool for digital phenotyping.