Behavioral Rigidity in Residential Treatment for Substance Use Disorder (RIG-TD) (RIG-TD)

This is an observational, longitudinal intensive study with repeated measures (weekly assessments) conducted in a residential therapeutic community in Mexico. Participants are adults (N=34, age 18-60) with substance use disorder and co-occurring emotional dysregulation or behavioral maladaptation, receiving standard psychosocial treatment (not assigned by the research protocol). Inclusion requires a minimum stay of four weeks, at least three complete measurement time points, and ability to complete self-reports. Exclusion criteria: severe neurocognitive impairment, acute psychotic symptoms, or incomplete longitudinal records.

Measures are collected at the beginning of each weekly therapy session:

Automatic Thoughts Questionnaire-8 (ATQ-8): frequency of negative automatic thoughts (higher scores indicate more frequent negative automatic thoughts).

Behavioral Activation for Depression Scale (BADS): subscales of Activation (goal-directed activity) and Avoidance/Rumination (experiential avoidance).

From these, a Behavioral Rigidity Index (BRI) is computed in two versions:

BRI-Simple (BRI-S): sum of standardized intraindividual scores of ATQ-8 + BADS-Avoidance - BADS-Activation.

BRI-Pondered (BRI-P): weighted combination derived from principal component analysis of the three standardized scores. The first principal component explained the majority of the common variance, supporting unidimensionality of the construct.

Additionally, natural language processing (NLP) pipelines are applied to transcribed therapy sessions and self-registries. Text is tokenized, lemmatized, and normalized by document length. Four linguistic domains are extracted:

Affective polarity: density of negative/positive terms.

Linguistic absolutism: frequency of extreme quantifiers (e.g., "always", "never", "everything", "nothing").

Grammatical agency: proportion of active vs. passive voice sentences.

Lexical complexity: type-token ratio and entropy.

A Linguistic Rigidity Index (LRI) is created by standardizing and averaging absolutism scores and inverted negative polarity (higher LRI = greater linguistic rigidity).

Statistical analysis uses Bayesian multilevel models with random intercepts by participant. The outcome variable is intraindividual standardized ATQ-8 (ATQ8_total_z). Predictors include time in treatment, BRI (contemporary and lagged by one week), and LRI. Model comparison will be performed using information criteria appropriate for multilevel models. Secondary analyses explore dynamic systems metrics derived from the state space defined by ATQ-8 and BRI-P, such as deviation from typical states and return speed after disturbances.

All procedures comply with the Declaration of Helsinki. The protocol was approved by the institutional ethics committee (Comité de Ética en Investigación de la Comunidad Terapéutica Under The Tree). Participants gave informed consent for anonymized use of their clinical data for research purposes.