A novel approach for constructing personalized networks from longitudinal perceived causal relations

Abstract

Personalized networks of psychological symptoms aim to advance therapy by identifying treatment targets for specific patients. Statistical relations in such networks can be estimated from intensive longitudinal data, but their causal interpretation is limited by strong statistical assumptions. An alternative is to create networks from patient perceptions, which comes with other limitations such as retrospective bias. We introduce the Longitudinal Perceived Causal Problem Networks (L-PECAN) approach to address both these concerns. 20 participants screening positive for depression completed 4 weeks day of brief daily assessments of perceived symptom interactions. Quality criteria of this new method are introduced, answering questions such as “Which symptoms should be included in networks?”, “How many datapoints need to be collected to achieve stable networks?”, and “Does the network change over time?”. Accordingly, about 40% of respondents achieved stable networks and only few respondents exhibited network structure that changed during the assessment period. The method was time-efficient (on average 7.4 min per day), and well received. Overall, L-PECAN addresses several of the prevailing issues found in statistical networks and therefore provides a clinically meaningful method for personalization.