PSY-ICD study: AI phenotyping identifies device-focused distress overlooked by conventional mood assessment in ICD recipients
D Castagno, A Olivero, R Marcato, B Contato, M Miniotti, L Boccia, E Cavallone, F Ferraris, A Saglietto, M Matta, V Dusi, M Anselmino, C Giustetto, G M De Ferrari, P LeombruniAbstract
Background
Psychological distress is frequent among implantable cardioverter-defibrillator (ICD) recipients. However, screening strategies mainly focused on anxiety and depression may fail to capture distress specifically related to living with the device.
Purpose
Within the PSY-ICD study, we applied unsupervised artificial intelligence methods to identify distinct psychological phenotypes in ICD recipients and to explore whether device-related distress may remain undetected by conventional mood-based assessment.
Methods
Consecutive ICD recipients (single- or dual-chamber, CRT-D, subcutaneous or extravascular systems) assessed between March 2025 and February 2026 completed validated questionnaires evaluating mood symptoms (DASS-21), psychological well-being (WHO-5), euthymia, demoralisation (DS-6), mental pain (MPQ), and device-related adjustment and acceptance (IDAS, FPAS). Patients with severe cognitive impairment (pathological MMSE) were excluded. Unsupervised k-means clustering (k=4) was applied to standardised psychometric summary scores (z-scores). Identified phenotypes were compared according to sex, comorbidity burden (Charlson index), timing of enrolment (at implant vs follow-up), and device therapies in the previous year (ATP and/or shocks).
Results
A total of 268 patients were included (median age 62 years; 73% male). AI clustering identified four distinct phenotypes: Resilient (49%), Device-focused distress (24%), Moderate affective distress (19%), and Severe/global distress (8%). The Device-focused distress phenotype was characterised by markedly elevated device-related distress (IDAS distress 20.4; FPAS distress 14.3; FPAS total 49.0) despite relatively low mood symptom scores (DASS stress 6.3; anxiety 3.3; depression 3.7), suggesting that mood-only screening may underestimate clinically relevant suffering (see Figure). Device therapies in the previous year were more frequent in the Severe/global distress phenotype (28.6%) compared with the other groups (6.9–15.6%; p=0.020). Enrolment at the time of implantation was strongly associated with phenotype distribution (29.7% in Device-focused distress vs 3.8% in Resilient; p<0.001, see Figure). Charlson comorbidity index did not differ across phenotypes (p=0.63). Female sex was more prevalent in Severe/global distress compared with Resilient patients (45.5% vs 21.4%; p=0.039).
Conclusions
In the PSY-ICD study, AI-based phenotyping revealed a distinct subgroup displaying prominent device-related distress despite low anxiety and depression scores. These results underscore the limitations of mood-only screening and support the integration of device-specific psychological assessment for earlier recognition and more personalised follow-up in the long-term management of ICD recipients.Heatmap of standardised psychometric scores