P5407Predicting adherence to internet-delivered cognitive behaviour therapy for comorbid symptoms of depression and anxiety after myocardial infarction

Abstract

Background: Psychotherapeutic treatment for the subgroup of patients with MI that also suffer from comorbid symptoms of anxiety and/or depression (MI-ANXDEP) is part of cardiac rehabilitation (CR). Adherence to a range of treat-ments and lifestyle advice is crucial for risk reduction in these patients. Under-standing the relative importance of predictors of adherence to internet-delivered cognitive behaviour therapy (ICBT) for MI-ANXDEP could improve their targeted care. Purpose: We estimated the relative importance of a range of established and novel predictors of adherence to ICBT for MI-ANXDEP patients. Method: The study sample consisted of 90 MI-ANXDEP patients (58.4 years, 62% men) recruited from 25 hospitals in Sweden who were randomised to active treatment in the ICBT trial U-CARE Heart. Time-point of prediction was at com-pletion of the first homework assignment (HWA), and adherence was gauged at the end of treatment (48% adherers). Adherence was defined as completing at least the first two HWAs within the 14-week treatment period. A supervised machine learning (ML) procedure, applying 3x10 cross-validated recursive feature elimination with a random forest model as internal classifier, estimated the relative importance of predictors for adherence from a range of patient demographic, clinical, and linguistic variables that were available at completion of the first HWA. Result: Out of 34 potential predictors, ML selected an optimal set of 19 predictors (Accuracy 0.64, 95% CI 0.61-0.68). The strongest predictors for being classified as adherent were in order of relative importance (1) higher self-rated cardiac fear (CAQ fear), (2) female sex, (3) more words used by the patient to answer the first homework assignment (Number of words), (4) higher self-rated general cardiac anxiety (CAQ total), and (5) a higher rate of words used by the patient that were identical with words prompted by the first homework assignment (Number of mutual words), as depicted in the figure. [Figure Presented] Conclusion(s): It is of clinical importance to understand poor adherence to ICBT treatment in the high risk MI-ANXDEP subpopulation. Higher cardiac anxiety and female sex were the strongest predictors for adherence. A novel finding was that linguistic variables were important for predicting adherence, particularly the num-ber of words used may signify the degree of personal investment and motiva-tion for treatment, and the number of mutual words used may be a proxy for therapeutic alliance within the treatment. Education had no predictive value. Future research should investigate potential causal mechanisms, and whether these findings replicate outside of Sweden, in larger samples, and for similar eHealth treatments. Funding Acknowledgements: The Swedish Research Council for Health, Work-ing Life, and Welfare [2014-4947], the Vårdal foundation [2014-0114], U-CARE [2009-1093].