Interpretable machine learning reveals key dose trajectory patterns predicting success of acupuncture-assisted methadone tapering

Abstract

Objective: Long-term methadone maintenance treatment (MMT) requires gradual dose reduction to mitigate adverse effects; prior research has shown that acupuncture facilitates dose tapering and alleviates opioid cravings. Although medication dosage parameters hold significant clinical value in addiction medicine, the impact of early outpatient dose data on therapeutic effects currently remains unclear. This study aimed to construct a methadone dose reduction prediction model and analyze the clinical value of historical dose trajectories.

Methods: Data from two randomized trials (N = 197 patients across six Chinese MMT clinics) were analyzed, with participants grouped into the acupuncture and non-acupuncture cohorts. The primary outcome was combined outcome comprising methadone dose reduction and craving score changes. Pre-intervention dose trajectories were derived via cluster analysis and merged with baseline features. Five machine learning models were trained using SHapley Additive exPlanations (SHAP) to explain the feature contributions. Subgroup analyses linking trajectories to the effects of acupuncture were conducted.

Results: Methadone dose data were clustered into three trajectories. Model training included nine features from 11 variables. The CatBoost model achieved the best performance on the test set (area under the curve = 0.7531, accuracy = 0.8205). The SHAP summary plot revealed that the three most influential factors in methadone dose reduction were intervention type, body mass index, and dosage trajectory. Subgroup analysis showed that trajectory class 2 exhibited significantly better efficacy than class 0 at weeks 2 and 4 of acupuncture (week 2: risk difference, 20.4%, P = 0.015; week 4: risk difference, 27.5%, P = 0.013).

Conclusions: In this study, we established a trajectory-based prediction model for MMT dose reduction and demonstrated the clinical value of historical trajectories. The results suggest that acupuncture optimally supports patients with dynamic “rise-then-decline” trajectories, advancing personalized MMT strategies.

Keywords

Acupuncture / Machine learning / Methadone maintenance treatment / Predictive model / Trajectory analysis

Conflicts of interest statement

The authors declare no conflict of interest.

Funding

The study was supported by the National Natural Science Foundation of China (82174527, 82405556); the Project of First-Class Universities and High-level Dual Discipline for Guangzhou University of Chinese Medicine (A1-2601-24-414-109Z71, A1-2601-24-415-109Z50, and A1-2601-24-415-109Z76); and Zhongshan Coma and Wake-up Promotion Workshop of Traditional Chinese Medicine (A2-3105-242-109-001). The China Postdoctoral Science Foundation General Project (2024M750464).

Author contributions

All authors participated in design of this study. Baochao Fan and Yiming Chen drafted the manuscript. Liming Lu and Jingchun Zeng revised it. Baochao Fan and Yiming Chen participated in designing the search strategies. Chen Chen and Peiming Zhang participated in the data extraction. Baochao Fan, Yiming Chen and Chen Chen participated in data analyses. Liming Lu arbitrated any disagreements in the process of the study. All authors read and approved the final manuscript.

Ethical approval of studies and informed consent

This post hoc analysis of de-identified data complied with the Declaration of Helsinki and was approved by the Guangzhou University of Chinese Medicine Institutional Review Board (YJ-KY-2025-035), which waived the requirement for informed consent.

Acknowledgments

The author sincerely thanks the medical and administrative staff of Guangzhou Huiai Hospital, Guangzhou Baiyun District Maternal and Child Health Hospital, Shunde Wu Zhong Pei Hospital, The Third People’s Hospital of Foshan, The Third People’s Hospital of Zhaoqing, and Zhongshan Second People’s Hospital for providing clinical data for this study. Special thanks to Yuting Wang, Yu Dong, Dehui Nie, Chenyang Tao, and Cui Li for their valuable support in data collection and patient management. Ultimately, this work would not have been possible without the cooperation of all the patients participating in the clinical trials, and we are deeply grateful for their contributions.

Data availability

Data supporting the findings of this study will be made available from the corresponding author [e-mail].

References

[1] United Nations Publications. World Drug Report 2024:special points of interest. United Nations Publications; 2024. Available from: https://www.unodc.org/unodc/en/data-and-analysis/world-drug-report-2024.html. Accessed December 2, 2025.

[2] Launonen E, Wallace I, Kotovirta E, et al. Factors associated with non-adherence and misuse of opioid maintenance treatment medications and intoxicating drugs among Finnish maintenance treatment patients. Drug Alcohol Depend. 2016; 162:227-235.

[3] Sharkey KM, Kurth ME, Anderson BJ, et al. Obstructive sleep apnea is more common than central sleep apnea in methadone maintenance patients with subjective sleep complaints. Drug Alcohol Depend. 2010; 108(1-2):77-83.

[4] Haber PS, Elsayed M, Espinoza D, et al. Constipation and other common symptoms reported by women and men in methadone and buprenorphine maintenance treatment. Drug Alcohol Depend. 2017; 181:132-139.

[5] Mintzer MZ, Stitzer ML. Cognitive impairment in methadone maintenance patients. Drug Alcohol Depend. 2002; 67(1):41-51.

[6] Prosser J, London ED, Galynker II. Sustained attention in patients receiving and abstinent following methadone maintenance treatment for opiate dependence: performance and neuroimaging results. Drug Alcohol Depend. 2009; 104(3):228-240.

[7] Ersche KD, Fletcher PC, Roiser JP, et al. Differences in orbitofrontal activation during decision-making between methadone-maintained opiate users, heroin users and healthy volunteers. Psychopharmacology (Berl). 2006; 188(3):364-373.

[8] Wen H, Chen R, Zhang P, et al. Acupuncture for opioid dependence patients receiving methadone maintenance treatment: a network meta-analysis. Front Psychiatry. 2021; 12:767613.

[9] Ge S, Lan J, Yi Q, et al. Acupuncture for illicit drug withdrawal syndrome: a systematic review and meta-analysis. Eur J Integr Med. 2020; 35:101096.

[10] Lu L, Zhang Y, Tang X, et al. Evidence on acupuncture therapies is underused in clinical practice and health policy. BMJ. 2022; 376:e067475.

[11] Zeng L, Tao Y, Hou W, et al. Electro-acupuncture improves psychiatric symptoms, anxiety and depression in methamphetamine addicts during abstinence: a randomized controlled trial. Medicine (Baltimore). 2018; 97(34):e11905.

[12] Meade CS, Lukas SE, McDonald LJ, et al. A randomized trial of transcutaneous electric acupoint stimulation as adjunctive treatment for opioid detoxification. J Subst Abuse Treat. 2010; 38(1):12-21.

[13] Wen H, Wei X, Ge S, et al. Clinical and economic evaluation of acupuncture for opioid-dependent patients receiving methadone maintenance treatment: the Integrative Clinical Trial and Evidence-Based Data. Front Public Health. 2021; 9:689753.

[14] Lu L, Chen C, Chen Y, et al. Effect of acupuncture for methadone reduction: a randomized clinical trial. Ann Intern Med. 2024; 177(8):1039-1047.

[15] Chen Y, Fan B, Zeng J, et al. Single-Cell RNA transcriptomics and multi-omics analyses reveal the clinical effects of acupuncture on methadone reduction. Research (Wash D C). 2025; 8:0741.

[16] Han DH, Lee S, Seo DC. Using machine learning to predict opioid misuse among U.S. adolescents. Prev Med. 2020; 130:105886.

[17] Lo-Ciganic WH, Huang JL, Zhang HH, et al. Evaluation of machine-learning algorithms for predicting opioid overdose risk among medicare beneficiaries with opioid prescriptions. JAMA Netw Open. 2019; 2(3):e190968.

[18] Campo DS, Gussler JW, Sue A, et al. Accurate spatiotemporal mapping of drug overdose deaths by machine learning of drug-related web-searches. PLoS One. 2020; 15(12):e0243622.

[19] Dong X, Rashidian S, Wang Y, et al. Machine learning based opioid overdose prediction using electronic health records. AMIA Annu Symp Proc. 2019; 2019:389-398.

[20] Moradinazar M, Farnia V, Alikhani M, et al. Factors related to relapse in patients with substance-related disorders under methadone maintenance therapy: decision tree analysis. Oman Med J. 2020; 35(1):e89.

[21] Gowin JL, Ernst M, Ball T, et al. Using neuroimaging to predict relapse in stimulant dependence: A comparison of linear and machine learning models. NeuroImage Clin. 2019; 21:101676.

[22] Wadekar AS. Understanding Opioid Use Disorder (OUD) using tree-based classifiers. Drug Alcohol Depend. 2020; 208:107839.

[23] Luan H, Zhao H, Li J, et al. Machine learning for investigation on endocrine-disrupting chemicals with gestational age and delivery time in a longitudinal cohort. Research (Wash D C). 2021; 2021:9873135.

[24] Mullin S, Zola J, Lee R, et al. Longitudinal K-means approaches to clustering and analyzing EHR opioid use trajectories for clinical subtypes. J Biomed Inform. 2021; 122:103889.

[25] Bao Y, Ding H, Zhang Z, et al. Intelligent acupuncture: data-driven revolution of traditional Chinese medicine. Acupunct Herb Med. 2023; 3(4):271-284.

[26] Liu X, Gong T. Artificial Intelligence and Evidence-Based Research Will Promote the Development of Traditional Medicine. Acupunct Herb Med. 2024; 4(1):134-135.

[27] American Psychiatric Association. D, American Psychiatric Association DS: Diagnostic and Statistical Manual of Mental Disorders. Washington, DC: American Psychiatric Association; 2013. p. DSM-5.

[28] Dong Y, Fan B, Yan E, et al. Decision tree model based prediction of the efficacy of acupuncture in methadone maintenance treatment. Front Neurol. 2022; 13:956255.

[29] Smith CBR, Brands B, Lacroix S, et al. Methadone Maintenance Treatment Program Standards and Clinical Guidelines. 4th ed. Toronto: College of Physicians and Surgeons of Ontario Publishing; 2011.

[30] SAMHSA/CSAT Treatment Improvement Protocols: Medications for Opioid Use Disorder: for Healthcare and Addiction Professionals, Policymakers, Patients, and Families. Rockville, (MD): Substance Abuse and Mental Health Services Administration. US; 2018.

[31] Heinzerling KG, Swanson AN, Kim S, et al. Randomized, double-blind, placebo-controlled trial of modafinil for the treatment of methamphetamine dependence. Drug Alcohol Depend. 2010; 109(1-3):20-29.

[32] Rezaei F, Emami M, Zahed S, et al. Sustained-release methylphenidate in methamphetamine dependence treatment: a double-blind and placebo-controlled trial. Daru. 2015; 23(1):2.

[33] Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: machine learning in Python. J Mach Learn Res. 2012; 12:2825-2830.

[34] Paparrizos J, Gravano L. Proceedings of the 2015 ACM SIGMOD international conference on management of data; 2015: k-shape: efficient and accurate clustering of time series. p.1855-1870.

[35] Lundberg SM, Lee SA. A unified approach to interpreting model predictions. Adv Neural Inf Process Syst. 2017; 30:4765-4774.

[36] Nosyk B, Sun H, Evans E, et al. Defining dosing pattern characteristics of successful tapers following methadone maintenance treatment: results from a population-based retrospective cohort study. Addiction. 2012; 107(9):1621-1629.

[37] Trafton JA, Minkel J, Humphreys K. Determining effective methadone doses for individual opioid-dependent patients. PLoS Med. 2006; 3(3):e80.

[38] Prokhorenkova L, Gusev G, Vorobev A, et al. CatBoost: unbiased boosting with categorical features. Adv Neural Inf Process Syst. 2018:6638-6648.

[39] Hancock JT, Khoshgoftaar TM. CatBoost for big data: an interdisciplinary review. J Big Data. 2020; 7(1):94.

[40] Faggiano F, Vigna-Taglianti F, Versino E, et al. Methadone maintenance at different dosages for opioid dependence. Cochrane Database Syst Rev. 2003(3):CD002208.

[41] Cui CL, Wu LZ, Luo F. Acupuncture for the treatment of drug addiction. Neurochem Res. 2008; 33(10):2013-2022.

[42] Talal AH, Ding Y, Venuto CS, et al. Toward precision prescribing for methadone: determinants of methadone deposition. PLoS One. 2020; 15(4):e0231467.

[43] Baldassarri SR, Beitel M, Zinchuk A, et al. Correlates of sleep quality and excessive daytime sleepiness in people with opioid use disorder receiving methadone treatment. Sleep Breath. 2020; 24(4):1729-1737.

[44] Sason A, Adelson M, Schreiber S, et al. The prevalence of constipation and its relation to sweet taste preference among patients receiving methadone maintenance treatment. Drug Alcohol Depend. 2021; 225:108836.

[45] Zhang X, Zhang X, Li X, et al. Roadmap for modern development of Chinese medicine: Traditional, evidence-based, and digital-intelligent paths. Sci Traditi Chin Med. 2023; 2(1):14-19.