Substance use disorders (SUD) are among the most common psychiatric disorders, with the highest mortality. Current evidence-based treatments are moderately effective, with one-year relapse rates of about 50%. Studies show considerable clinical heterogeneity in SUD patients and their response to treatment. More fruitful than developing new treatments with again moderate effectiveness is to make better use of existing evidence-based treatments.
Patient-related clinical heterogeneity asks for more personalized treatment approaches matching treatment with specific patient characteristics to improve treatment outcome. In other fields of medicine (like oncology and cardiology), staging and profiling approaches paved the way to the development of clinical decision tools for personalized medicine that reduced morbidity and mortality. These approaches have found limited translation to the field of mental health, and addiction care in particular.
Health care workers in clinical practice have to decide together with their patients what treatment strategy to pursue, for example whether this should be an intensive inpatient treatment or a less intensive outpatient trajectory. Although there is a wealth of patient-related data available, the value of this data for optimizing treatment outcomes by allocating a patient to the optimal treatment for that specific patient is unknown.
The current project aims to improve treatment outcomes in addiction care, by the evidence-based development of a decision-aid, based on insights in patient-related predictors of treatment outcome using naturalistic data. The research group is in an excellent position to implement these insights through adjustment of treatment allocation guidelines and routine outcome measures (ROM) used in clinical practice.
Identification of patient-related predictors of treatment outcome holds promise to match treatment with individual patient characteristics. Rather than conventional linear statistical methods, innovative data-driven, machine learning approaches (apt to find complex interactions between numerous predictors) might better capture the complex, non-linear dynamics of behavioral change in clinical practice.
We aim to quantitatively build data-driven prediction models that reliably prognosticates which treatment intensity should be indicated for a certain patient to optimize his treatment outcome. Such an indication can inform patients and health-care workers in their joint decision-making process.
Register data are available from SUD patients with a new treatment entry at Tactus (one of the larges addiction care facilities in The Netherlands) between 2011-2021. Data on the period 2011-2016 covered 15.588 full intake assessments, with 3-6 months follow-up routine outcome monitoring (ROM) data available for about 10.410 cases.
Primary outcome is abstinence at 3-6 months, secondary outcomes include response (25% reduction in use of primary problem substance and 25% improvement on the international classification of functioning) at 3-6 months. For outcome assessment, the Measurement for Addictions, Triage and Evaluation (MATE) will be used, available in the ROM data. Predictors will be drawn from the MATE at intake, covering 90 items (SUD severity, psychiatric comorbidity, treatment history, physical health, personality, social functioning, and demographics).
Students will be supervised by two psychiatrists from the department of psychiatry, with respectively clinical expertise in addiction care and basic experience in machine learning.
The final deliverable is a code repository on GitHub, a report and preferably a scientific publication.
Students with a major in data science, computer science, or artificial intelligence in the final stage of master level studies are invited to apply.
Interest in health care and machine learning.
Affinity with programming in R (or Python) and machine learning packages is required.
Project duration: 6 months
Location: Radboud University Medical Center
For more information, please contact Dirk Geurts