Treatment Adherence in Digital Psychotherapy Using Machine Learning to Predict Patient No-shows

Abstract: Background: Untreated patients and discontinuity in treatments are problems that mental health care is facing. Even though people seek care, there is still a pattern of patients who do not attend their scheduled appointments, referred to as No-shows. Noshows result in prolonged waiting times for patients and decreased efficiency and workflow for healthcare professionals. Moreover, causing great financial costs and losses for the healthcare sector. Using machine learning to predict potential No-shows beforehand could be a possible solution to minimize No-shows, while enhancing treatment adherence. Aim: The aim is to explore the best-performing algorithm for No-show predictions in digital psychotherapy. Furthermore, gaining a deeper knowledge of common behaviors in patient demographic and appointment data that may explain the reasons behind No-shows in digital mental health care services. Methods: A quantitative experimental research methodology and design with an inductive approach were utilized, incorporating computational methods, tools, and techniques. The Knowledge Discovery in Databases process was used as a guidance in the data mining process. Results: An observational relationship was found between No-shows and the following features age, day of the week of the appointment, date in a month of the appointment, month of the appointment, and waiting time. The best-performing algorithms to predict No-shows were Gradient Boosting Decision Tree and Random Forest. The date in a month was the most impactful feature for both classifiers, followed by the appointment month, the day of the week, and the number of waiting days. Conclusion: Machine learning has the potential to predict No-shows in digital psychotherapy and can be used to identify the underlying factors and patterns behind Noshows while providing useful information to support and improve digital mental health care delivery, treatment adherence, and patient outcomes. Thus, predicting No-shows beforehand is highly relevant for enhancing treatment adherence in digital psychotherapy and mental health care.