2 - (SYM 62) Using Baseline Data to Identify Heterogeneity in Within-person Processes: The Use of Longitudinal Recursive Partitioning

In intensive longitudinal studies, it is common to analyze between person differences through the inclusion of grand mean centered variables. However, particularly in clinical research, a large number of constructs and characteristics are assessed at baseline, which often goes unutilized in intensive longitudinal modeling. In this talk I will discuss how one novel algorithm from machine learning, longitudinal recursive partitioning, can allow researchers to better utilize baseline data to examine between person differences in within-person processes. To demonstrate this, I detail results from an ecological momentary assessment study, which assessed individuals with past month suicidal ideation (SI) four times per day. The final sample, after requiring each person to have at least twenty responses, included data from 35 participants. Further, for the analyses, we excluded any responses that were more than 12 hours apart, resulting in a total of 2,030 unique responses. During the study period 27 unique participants reported SI ( > 1), and non-zero SI was reported on 30.6% (n = 520) of prompts. The multilevel model included active SI as an outcome, a random intercept, with an autoregressive effect and a cross-lag effect for lagged versions of several salient measures of emotional distress (i.e., negative and positive affect; perceived burdensomeness; thwarted belongingness). For baseline variables, we used the following assessments: current mental health treatment engagement, prior psychiatric hospitalization, gender, race, ethnicity, and prior two-week depression symptoms (via the Patient Health Questionnaire-9; PHQ-9) . Using the longRPart2 package in R, the final tree had splits on PHQ-9 at a score of 27, and between identifying as White and all other races. Those higher in baseline depression had higher degrees of inertia (autoregressive effects), while those lower in baseline depression had stronger cross-lag effects, particularly for thwarted belongingness. Those that identified as American Indian, Asian or Black had the strongest effects for perceived burdensomeness and were the only group to have a negative effect for positive affect. In summary, we showed how baseline assessments can be used to identify meaningful heterogeneity to within-person processes, shedding further light on the complexity to understanding momentary fluctuations to SI.