In plain language
Standardized therapy protocols work for many people, but a large minority — often 30% to 50% — do not achieve meaningful improvement, and average effect sizes have plateaued for decades. One promising response is to personalize treatment by identifying which psychological processes matter most for each individual. This tutorial paper shows, step by step, how three advanced analytic methods — iARIMAX, iBoruta, and a new time-series machine-learning method called tsBoruta — can identify personalized treatment processes, using hair-pulling disorder (trichotillomania) as the empirical case.
The data came from 54 adults meeting DSM-5 criteria for trichotillomania who completed ecological momentary assessment surveys three times a day for 30 days (up to 90 prompts per person), reporting on hair-pulling and psychological processes such as cognitive fixation, valued action, and anxiety. The authors modeled each person's time series individually, compared what the three methods concluded at both the group and individual level, and assessed each method's ability to detect linear and nonlinear associations.
All three methods converged on cognitive fixation as a key group-level predictor of hair-pulling, but the individual-level picture was strikingly heterogeneous: most participants had their own unique combination of relevant process-outcome links. The paper argues for combining standardized protocols with personalized interventions, and provides a practical tutorial showing that tsBoruta offers a powerful, balanced approach to modeling complexity in clinical time-series data for treatment planning.
Key findings
- All three methods (iARIMAX, iBoruta, tsBoruta) confirmed cognitive fixation as a core aggregate-level predictor of hair-pulling in trichotillomania.
- Associations that appeared strong in single-process iARIMAX models became more modest — but still meaningful — once multiple processes were modeled together.
- tsBoruta, which respects the time-series structure of the data, was more conservative than iBoruta in confirming significant individual-level effects, reducing the risk of spurious findings.
- 61.11% of participants showed a unique combination of relevant process-outcome links, demonstrating substantial systematic heterogeneity across individuals.
- Targeting just three key processes — cognitive fixation, valued action, and anxiety — could potentially benefit 52 of the 54 individuals in the sample.
- The paper provides a full methodological tutorial (with code) for applying tsBoruta to ecological momentary assessment data for idionomic, process-based treatment planning.
How to cite
APA
Sahdra, B. K., Woolley, M. G., Hernández, C., Li, W., Hayes, S. C., Ciarrochi, J., Twohig, M. P., & Levin, M. E. (2026). Time series machine learning for idionomic process-based treatment planning: A tutorial on tsBoruta. Journal of Contextual Behavioral Science, 40, 100983. https://doi.org/10.1016/j.jcbs.2026.100983
BibTeX
@article{sahdra2026time,
title = {Time series machine learning for idionomic process-based treatment planning: A tutorial on tsBoruta},
author = {Sahdra, Baljinder K. and Woolley, Mercedes G. and Hern{\'a}ndez, Crist{\'o}bal and Li, William and Hayes, Steven C. and Ciarrochi, Joseph and Twohig, Michael P. and Levin, Michael E.},
journal = {Journal of Contextual Behavioral Science},
year = {2026},
volume = {40},
pages = {100983},
doi = {10.1016/j.jcbs.2026.100983}
}
Related work
- All publications by Joseph Ciarrochi (searchable, with free PDFs)
- Process-Based Therapy & Idionomic Analysis
Author: Joseph Ciarrochi (ORCID 0000-0003-0471-8100). Free copy hosted with permission for scholarly use. Please cite the published version via the DOI above.