The patterns of relapse and abstinence: using machine learning to identify a multidimensional signature of long-term outcome after inpatient alcohol withdrawal treatment.
A machine learning approach to identify a multidimensional signature associated with relapse and long-term outcome in alcohol dependence treatment.
In this observational naturalistic study, inpatients with alcohol dependence received qualified detoxification plus CBT (Cognitive Behavioral Therapy) and were followed up 6-months after discharge to assess abstinence and drinking behavior. Cross-validated multivariate sparse partial least squares analysis (SPLS) was used to investigate the relationship between clinical features and four long-term outcome variables.
Germany.
152 patients (on average 47.8 years old, 72% male) with alcohol dependence, who received inpatient qualified detoxification plus CBT.
35 clinical features were used to cover all three phases of inpatient treatment (pre-, within-, post-treatment). Among these, sociodemographic characteristics, ICD-10 psychiatric diagnoses, previous detoxification treatments, and somatic measurements as well as inpatient treatment setting such as withdrawal medication, liver ultrasound, further information about the patients´ stay, and post-inpatient care were assessed. The four outcome dimensions included: continuous abstinence, abstinence at follow up, daily alcohol consumption, and days of abstinence after discharge.
Six months after withdrawal treatment 46% of the patients achieved continuous abstinence. Socioeconomic, clinical and somatic features across the treatment timeline were analyzed and summarized into a multivariate signature associated with long-term treatment outcome. Thereby, the SPLS algorithm identified regular completion of withdrawal treatment, higher education, and employment status to be most strongly associated with a positive outcome. Alcohol-related hepatic and hematopoietic damage, number of previous withdrawal treatments and living in a shelter were most profoundly associated with a negative outcome.
Conceiving treatment outcome as a multidimensional signature and moving beyond simple binary classifications of relapse versus abstinence may improve the understanding of relapse pathways and support more individualized treatment strategies.
In this observational naturalistic study, inpatients with alcohol dependence received qualified detoxification plus CBT (Cognitive Behavioral Therapy) and were followed up 6-months after discharge to assess abstinence and drinking behavior. Cross-validated multivariate sparse partial least squares analysis (SPLS) was used to investigate the relationship between clinical features and four long-term outcome variables.
Germany.
152 patients (on average 47.8 years old, 72% male) with alcohol dependence, who received inpatient qualified detoxification plus CBT.
35 clinical features were used to cover all three phases of inpatient treatment (pre-, within-, post-treatment). Among these, sociodemographic characteristics, ICD-10 psychiatric diagnoses, previous detoxification treatments, and somatic measurements as well as inpatient treatment setting such as withdrawal medication, liver ultrasound, further information about the patients´ stay, and post-inpatient care were assessed. The four outcome dimensions included: continuous abstinence, abstinence at follow up, daily alcohol consumption, and days of abstinence after discharge.
Six months after withdrawal treatment 46% of the patients achieved continuous abstinence. Socioeconomic, clinical and somatic features across the treatment timeline were analyzed and summarized into a multivariate signature associated with long-term treatment outcome. Thereby, the SPLS algorithm identified regular completion of withdrawal treatment, higher education, and employment status to be most strongly associated with a positive outcome. Alcohol-related hepatic and hematopoietic damage, number of previous withdrawal treatments and living in a shelter were most profoundly associated with a negative outcome.
Conceiving treatment outcome as a multidimensional signature and moving beyond simple binary classifications of relapse versus abstinence may improve the understanding of relapse pathways and support more individualized treatment strategies.
Authors
Raabe Raabe, Brechtel Brechtel, Lugmair Lugmair, Weiser Weiser, Schiltz Schiltz, Koutsouleris Koutsouleris, Falkai Falkai, Hoch Hoch, Pogarell Pogarell, Koller Koller, Popovic Popovic
View on Pubmed