Predicting early treatment termination in opioid use disorder: a machine learning model for risk stratification

Background: Half of patients with opioid use disorder (OUD) discontinue opioid agonist treatment within the first year. Identifying premature termination risk at intake could enable targeted clinical responses. Objectives: To develop and externally validate a machine learning model predicting premature treatment termination in OUD; to derive a clinical risk stratification system; and to identify variables most associated with predicted probability. Methods: Retrospective data from the Information System of the Andalusian (South Spain) Plan on Drugs and Addictions (SiPASDA) covered 11,475 patients (86.2% male) admitted to 121 public outpatient centers between 2015 and 2023 (90.8% termination rate). A CatBoost classifier was developed (n = 10,328) and externally validated (n = 1147). Predicted probabilities were discretized into low, moderate- and high‑risk strata. Shapley Additive exPlanations (SHAP) analysis quantified global and stratum-specific predictor contributions. Results: CRIBOOST achieved an area under the receiver operating characteristic curve (AUC–ROC) of 0.652 ± 0.018 in cross-validation and 0.646 externally. Termination rates rose monotonically across strata: 34.6% (low); 69.6% (moderate); 80.8% (high). Housing instability was the strongest positive predictor (SHAP= +0.569; +0.601 in high‑risk). Judicial referral was most protective (SHAP = −0.412), attenuating from low (−0.553) to high (−0.254); employment showed a similar gradient. Mood disorder (SHAP = +0.470) and benzodiazepine use disorder contributed positively, escalating with risk. Conclusions: CRIBOOST stratifies OUD patients into risk groups at admission using routine intake data, enabling clinicians to target retention-focused interventions where most needed and supporting efficient resource allocation. Stratum-specific SHAP contributions reveal which factors drive risk within each group, informing intervention design.

Authors

Publication Details

Journal
Figshare
Published
2026-11-20
DOI
https://doi.org/10.6084/m9.figshare.35080732
Primary Topic
Opioid Use Disorder Treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Predicting early treatment termination in opioid use disorder: a machine learning model for risk stratification

Marta Narváez‐Camargo, Francisco A. Márquez, Cinta Mancheño‐Velasco, Óscar Lozano-Rojas et al.
2 links
Figshare
Opioid Use Disorder Treatment
article

Predicting early treatment termination in opioid use disorder: a machine learning model for risk stratification

Marta Narváez‐Camargo, Francisco A. Márquez, Cinta Mancheño‐Velasco, Óscar Lozano-Rojas, Antonio Moreno-Moreno
article en

Abstract

Background: Half of patients with opioid use disorder (OUD) discontinue opioid agonist treatment within the first year. Identifying premature termination risk at intake could enable targeted clinical responses. Objectives: To develop and externally validate a machine learning model predicting premature treatment termination in OUD; to derive a clinical risk stratification system; and to identify variables most associated with predicted probability. Methods: Retrospective data from the Information System of the Andalusian (South Spain) Plan on Drugs and Addictions (SiPASDA) covered 11,475 patients (86.2% male) admitted to 121 public outpatient centers between 2015 and 2023 (90.8% termination rate). A CatBoost classifier was developed (n = 10,328) and externally validated (n = 1147). Predicted probabilities were discretized into low, moderate- and high‑risk strata. Shapley Additive exPlanations (SHAP) analysis quantified global and stratum-specific predictor contributions. Results: CRIBOOST achieved an area under the receiver operating characteristic curve (AUC–ROC) of 0.652 ± 0.018 in cross-validation and 0.646 externally. Termination rates rose monotonically across strata: 34.6% (low); 69.6% (moderate); 80.8% (high). Housing instability was the strongest positive predictor (SHAP= +0.569; +0.601 in high‑risk). Judicial referral was most protective (SHAP = −0.412), attenuating from low (−0.553) to high (−0.254); employment showed a similar gradient. Mood disorder (SHAP = +0.470) and benzodiazepine use disorder contributed positively, escalating with risk. Conclusions: CRIBOOST stratifies OUD patients into risk groups at admission using routine intake data, enabling clinicians to target retention-focused interventions where most needed and supporting efficient resource allocation. Stratum-specific SHAP contributions reveal which factors drive risk within each group, informing intervention design.

Figshare
Good health and well-being
Opioid Use Disorder Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Predicting early treatment termination in opioid use disorder: a machine learning model for risk stratification — Marta Narváez‐Camargo, Francisco A. Márquez, et al. · Figshare (2026) | TGRS Research Map | TGRS