Abstract
<jats:p>Background: Smartphone enabled remote patient monitoring has the potential to complement conventional follow-up in inflammatory arthritis. We previously presented the finger fold index (FFI) derived from hand photographs as ratio of automated detected proximal interphalangeal (PIP) joint diameter and surface of dorsal finger folds as a digital biomarker for clinical joint swelling and disease activity in rheumatoid arthritis (RA) and psoriatic arthritis (PsA). Objective: To evaluate the feasibility, image quality, patient engagement, and clinical utility of both HCP- and patient-collected hand photographs integrated into a national rheumatology registry, and to assess the performance of the FFI as an image-derived digital biomarker for clinical joint swelling of the proximal interphalangeal joints in a real-world arthritis cohort. Methods: In this longitudinal multicenter study, a photo function with written instructions were integrated into the Swiss Clinical Quality Management in Rheumatic Diseases (SCQM) registry and their mySCQM mobile application, respectively. Patients with RA or PsA contributed longitudinal smartphone photographs together with patient-reported outcomes (PROs) via the mySCQM mobile application while health care professionals (HCPs) acquired images during routine visits. After manual quality assessment, images were processed using an automated computer vision pipeline to derive the FFI, a digital biomarker based on dorsal finger-fold morphology. Image quality was evaluated for both HCP- and participant-collected photographs, and patient engagement was assessed. Associations between FFI, clinical proximal interphalangeal (PIP) joint swelling, RADAI-5, DAS28-CRP and longitudinal changes were assessed. A generalized linear mixed model was used to estimate the association between FFI and joint swelling while accounting for repeated measures and within-subject correlations. Results: Between 2023 and 2025, 374 RA and PsA patients were included. HCPs captured 977 hand images while 174 patients collected 1228 hand images via the mySCQM app. Patients demonstrated sustained engagement after instruction, contributing a mean of seven images during data collection. Following quality control, 1729 hand images comprising 4048 PIP joints were included for analysis. Image quality was comparable between patient-acquired and HCP-acquired photographs; 78.3% of the patient-acquired vs. 73.5% of the HCP-acquired hand images were suitable to run the ML-model. 23.1% of the cropped joints had to be removed after the running of the FFI algorithm due to false diameter or finger fold detection e.g. due to wrong hand positioning. In images taken by HCPs, mean FFI and DAS28-CRP were weakly but significantly correlated (Spearmans ρ = 0.164; 95% CI [0.004 to 0.317]; p = 0.039). Conversely, RADAI-5 scores did not correlate with the mean FFI in RA patients (r = 0.007, p = 0.932, 95% CI [-0.169-0.183]). At follow-up visits, clinical swelling resolved in 40 joints, of which in 68.0% the direction of the delta FFI was concordant with the clinical change. In contrast, 13 joints developed incident clinical swelling, of which 87.5% had a direction of the delta FFI that was concordant with the clinical change. However, the GLMM showed no significant associations between swelling and joint location or time-varying FFI, and no evidence of interaction between FFI and PIP joint. Conclusion: Integration of patient self-imaging into a remote monitoring application for inflammatory arthritis is feasible and achieves image quality comparable to clinician acquired photographs. The FFI derived from collected images shows association with clinical joint swelling and disease activity scores, but not PROs. In a substantial proportion of images, the FFI algorithm could not be applied because of insufficient image quality. More standardized image acquisition and further refinement of the FFI algorithm are warranted.</jats:p>