Abstract:Introduction Late Life Depression (LLD) is characterized by low remission rates using currently available treatments. Therefore, it is crucial for clinicians to have personalized tools to better identify older adults who may have a greater chance of success when deciding on a given treatment. In particular, psychotherapy is an important and under-utilized treatment option for older adults who may not tolerate medications or have drug interactions that limit pharmacotherapy. In this study, we aimed to use machine-learning methods to identify pre-treatment characteristics that are predictive of psychotherapy treatment success (i.e. Depression Remission after psychotherapy treatment). Methods We analyzed data from a previously described 12-week trial of weekly Problem Solving Therapy (PST) in 96 participants with Major Depressive Disorder who were 65 years or older (Kassel et al. 2022). Using machine-learning methods, we examined 20 pre-treatment characteristics of interest across four categories: 1) Demographic Factors, 2) Depression History, 3) Depression Symptom Characteristics, and 4) Cognitive Test Scores. The Gradient Boosted Method (GBM) was used to determine the relative importance of each characteristic as a possible predictor, which guided development of specific, short (depth=3) Decision Tree models to predict Depression Remission (17-item Hamilton Depression Score < 7). Given sample size limitations, these models were trained on the same sample to which they were applied for prediction. Results Three Decision Tree models were significant at the p<0.001 threshold: Age and Depression History, pre-treatment Episodic Memory Function, and All 20 Pre-treatment Characteristics Available (the resulting optimal model only utilized 5 of the 20 characteristics across the 4 categories of available pre-treatment characteristics). Amongst the three significant models: Classification Accuracy was 73.6-77.4% and AUC was 0.763-0.789. These 3 models offered high Specificity (82.8-95.1%) and Negative Predictive Value (75.3-82.8%) and relatively lower Sensitivity (34.5-63.3%) and Positive Predictive Value (61.5-76.9%). Age and Depression History (See Figure 1) was the simplest of the three significant models with only five Leaves (i.e. Terminal Nodes). In contrast, models derived from the pre-treatment Cognitive Domains of Executive Function, Processing Speed, and Language Function were each not significantly predictive of Depression Remission at the p<0.001 threshold. Models derived from pre-treatment Depression Symptom Characteristics were also not significantly predictive at this threshold. Conclusions We have derived an Age and Depression History Decision Tree model (Figure 1) that would only require assessment of 3 simple pre-treatment characteristics to improve prediction of psychotherapy success: correctly classifying about three quarters of the participants in our sample. The pre-treatment Episodic Memory Function Decision Tree model had a similar classification rate but does require cognitive testing. Both models were superior to models derived only from other Cognitive Domains, including models of Executive Function. However, validation of these models is needed and will require testing their predictive value in larger, more representative samples. These results demonstrate a novel machine-learning based prediction strategy that is simple for clinicians to use and thus may hold promise in personalizing difficult treatment decisions for LLD.

Making the most out of timeseries symptom data: A machine learning study on symptom predictions of internet-based CBT

Methodological choices and clinical usefulness for machine learning predictions of outcome in Internet-based cognitive behavioural therapy

Predicting treatment failure in regular care Internet-Delivered Cognitive Behavior Therapy for depression and anxiety using only weekly symptom measures.

Predicting therapy outcome in a digital mental health intervention for depression and anxiety: A machine learning approach

Predicting long-term outcome of Internet-delivered cognitive behavior therapy for social anxiety disorder using fMRI and support vector machine learning

Early change in specific depression symptoms and later outcome in internet-delivered psychotherapy for depression: A cohort study and cross-lagged network analysis

What the future holds: Machine learning to predict success in psychotherapy

Can we predict who will benefit from cognitive-behavioural therapy? A systematic review and meta-analysis of machine learning studies

The Precision in Psychiatry (PIP) study: Testing an internet-based methodology for accelerating research in treatment prediction and personalisation

Automatic structure estimation of predictive models for symptom development

Dataset size versus homogeneity: A machine learning study on pooling intervention data in e-mental health dropout predictions

Psychosis Prognosis Predictor: A continuous and uncertainty‐aware prediction of treatment outcome in first‐episode psychosis

Predicting CBT modality, treatment participation, and reliable improvements for individuals with anxiety and depression in a specialized mental health centre: a retrospective population-based cohort study

Predicting short-term suicidal thoughts in adolescents using machine learning: developing decision tools to identify daily level risk after hospitalization

Developing Clinically Relevant Decision Tree Models Using Machine-Learning to Predict Psychotherapy Success in Late Life Depression

A deep learning quantification of patient specificity as a predictor of session attendance and treatment response to internet-enabled cognitive behavioural therapy for common mental health disorders

Finding the Best Match - a Case Study on the (Text-)Feature and Model Choice in Digital Mental Health Interventions

Prediction of electroconvulsive therapy outcome: A network analysis approach

A psychological symptom based machine learning model for clinical evaluation of irritable bowel syndrome

A machine-learning framework for robust and reliable prediction of short- and long-term treatment response in initially antipsychotic-naïve schizophrenia patients based on multimodal neuropsychiatric data

Temporal prediction of suicidal ideation in an ecological momentary assessment study with recurrent neural networks