Skip to content

Causal Forest Estimation of Treatment Heterogeneity: Identifying Subgroups of Women Benefiting Most from Mindfulness-Based Stress Reduction for Chronic Pain

Tiago Figueiredo, Laura Benítez, Petra Nováková

2026 DOI: 10.61838/kman.pwj.5192 (opens in new tab)

Study at a glance

AI-extracted from the abstract
Characteristics Randomized controlled trial
Sample size 284
Population Adult women (mean age 51.4 years) residing in Spain with chronic non-malignant pain
Intervention Mindfulness-Based Stress Reduction
Duration 8-week program
Measures Brief Pain Inventory, Hospital Anxiety and Depression Scale, Five Facet Mindfulness Questionnaire
Topics Meditation
Key findings MBSR produced a statistically significant average reduction in pain interference (ΔM = -1.24), but causal forest estimation revealed substantial heterogeneity, with individual effects ranging from -3.10 to +0.45. The authors report that the greatest benefit occurred among women with elevated baseline anxiety and depression, lower non-judging mindfulness, and shorter chronic pain duration.

Abstract

Objective: The objective of this study was to apply a causal forest machine learning algorithm to estimate the conditional average treatment effects of Mindfulness-Based Stress Reduction on pain interference and identify the specific baseline characteristics of women with chronic pain who derive the greatest therapeutic benefit. Methods and Materials: This randomized controlled trial included a sample of N=284adult women (mean age M=51.4years) residing in Spain and suffering from chronic non-malignant pain. Participants were randomly assigned to an 8-week Mindfulness-Based Stress Reduction program or a treatment-as-usual control group. Baseline sociodemographic and psychological covariates were assessed using standardized psychometric instruments, including the Brief Pain Inventory, Hospital Anxiety and Depression Scale, and the Five Facet Mindfulness Questionnaire. To move beyond the traditional Average Treatment Effect (ATE), a causal forest algorithm was utilized to estimate the Conditional Average Treatment Effect (CATE) for pain interference, allowing for the empirical identification of high-benefit and low-benefit patient subgroups based on their multidimensional baseline profiles.

Findings: While the standard analysis revealed a statistically significant Average Treatment Effect for pain interference (ΔM=-1.24), the causal forest estimation uncovered substantial treatment heterogeneity, with individual CATEvalues ranging widely from -3.10to +0.45. Variable importance metrics derived from the machine learning model identified baseline anxiety as the strongest predictor of treatment benefit, followed closely by the non-judging of inner experience facet of dispositional mindfulness, and the total duration of the chronic pain condition. Specifically, the high-benefit subgroup—which experienced the most clinically significant reductions in pain interference (τ(x)≈-3.10)—was characterized by elevated baseline anxiety and depression, distinctly lower baseline scores in non-judging mindfulness, and a comparatively shorter history of chronic pain.

Conclusion: Mindfulness-Based Stress Reduction is not uniformly effective for all chronic pain patients; rather, it yields the maximum reduction in pain interference specifically among women presenting with high emotional distress, highly judgmental psychological traits, and shorter pain durations, highlighting the critical necessity of precision behavioral medicine.