Shared decision making using digital twins in knee osteoarthritis care: a randomized clinical trial of an AI-enabled decision aid versus education alone on decision quality, physical function, and user experience
Prakash Jayakumar, Paul J. Rathouz, Eugenia Lin, Zoe Trutner, Lauren M. Uhler, John P. Andrawis, Karl Koenig, Joel Tsevat, Kevin J. Bozic
Abstract
Background Patient decision aids (DAs) improve decision quality during shared decision-making (SDM) for patients seeking care for knee osteoarthritis (OA). However, few DAs incorporate the ‘digital twin' concept where comprehensive data are applied to computational models to generate dynamic virtual simulations and predictions to augment decision-making in real-time. We developed an artificial intelligence-enabled DA (AI-DA) that generated digital twins using patient reported outcome measurements (PROMs) and clinical data to enhance SDM by providing personalized predictions of risks and benefits for patients with knee OA considering total knee arthroplasty (TKA). We assessed the impact of the AI-DA on patient- and process-level outcomes. Methods We performed a randomized open-label clinical trial at a single, university-affiliated orthopaedic clinic in the USA involving patients with knee OA between February 2021 and November 2022. Patients received a full AI-DA incorporating patient education, preference assessment, and person-specific benefit:risk predictions of TKA (intervention group) or patient education only (control group). Outcomes included the Knee OA Decision Quality Instrument (K-DQI) (primary outcome), CollaboRATE SDM survey, Decision Conflict Scale (DCS), Decision Regret Scale (DRS), and Knee Injury and Osteoarthritis Outcome Score Joint Replacement (KOOS JR) for knee-specific health at 3 and 6 months post-randomization, satisfaction, appointment duration, and TKA rates. This study is registered with ClinicalTrials.gov, NCT04805554. Findings The analytic sample comprised 101 patients ([mean [SD], 64.9 [10.1] years; 54 [54%] women]) in the intervention group and 100 patients (mean [SD] age 63.4 [8] years; 60 [60%] women) in the control group. The intervention group reported higher decision quality (mean [SD] K-DQI: 84.4 [25.2] versus 71.4 [29.8], P = 0.0011), lower decision conflict (DCS: 1.0 [3.1] versus 3.3 [5.8], P = 0.0029), lower decision regret at 6–9 months (DRS: 18.2 [19.5] versus 27.2 [24.2], P = 0.0051), better knee-specific health (KOOS JR: 69.5 [17.3] versus 47 [18.4], P < 0.0001) at 6–9 months, and greater treatment concordance (91% versus 76%, P = 0.0043). SDM scores, knee health at 3 months, patient and clinician satisfaction, appointment duration, TKA rates, and decision regret at 3 months were similar between groups. Interpretation AI-DAs provide a more personalized, data-augmented SDM experience that can improve decision quality and longer-term health-related outcomes in patients with knee OA considering TKA. Funding Agency for Healthcare Research and Quality Grant (R21HS027037).