Predicting Premature Treatment Termination in Inpatient Psychotherapy: A Machine Learning Approach

University Hospital Heidelberg

Status: Completed

Conditions: Inpatient Psychotherapy; Machine Learning; Dropout Prediction; Premature Treatment Termination

Treatments: Behavioral: Psychotherapy

Study type: Observational

Funder types: Other

Identifiers

NCT06042595

Dropout-Prediction-2023

Details and patient eligibility

About

The study aims to develop a prediction model of premature treatment termination in psychosomatic hospitals using a machine learning approach.

Full description

The aim of the study is to identify risk factors that lead to or predict premature treatment termination in psychosomatic hospitals. In the long-term, the study shall help to develop more precise prediction models that can enhance communication between therapists and patients about potential dropout and- if necessary- adaption of treatment in using a feedback loop.

Since it is still not clear which variables play a major role in predicting treatment termination in psychosomatic hospitals, the study design is exploratory and includes a broad range of intake patient characteristics. The purpose of this study is hereby, to develop a prediction model based on the information that are routinely assessed at intake. Therefore, three kind of variables are planned to be included: (1) demographic and other clinical variables (e.g. age, gender, ICD-10 diagnoses), (2) psychological questionnaire data (e.g. PHQ, SF-12, EB-45, IIP-32, OPD-SFK), and (3) physiological data (e.g. routine laboratory data, blood pressure). For the study, all patients that started inpatient psychotherapy at the medical centre Heidelberg between 2015 and January 2022 will be included, resulting in a sample size of approximately N = 2000. As the average dropout rate based on meta analytical results is around 20%, one can assume that up to 400 patients prematurely dropped out of treatment.

To calculate the prediction model, it is planned to use a machine learning approach which is highly functional in big data sets. Using a Random Forest Model for binary outcomes (regular treatment length vs. premature treatment termination) it is envisioned to identify variables that contribute to the prediction of premature treatment termination at intake. Additionally, waiting list effects will be considered by taking into account the waiting duration between the initial intake interview and the moment of the hospital admission. Therefore, the study will, for the first time, investigate a prediction model for premature treatment termination in inpatient psychotherapy including clinically relevant physiological data as well as waiting time effects in preparation of the psychosomatic treatment.

Enrollment: 2,023 patients

Sex: All

Ages: 18+ years old

Volunteers: No Healthy Volunteers

Inclusion criteria

• patients of at least 18 years of age
• included in inpatient psychotherapy treatment program in a hospital for psychosomatic medicine
• provided information about admission and discharge date

Exclusion criteria

• bipolar, acute psychotic or substance abuse disorder