Multimodal interactive assistance for inclusive collection of patient-reported outcome measures: A pragmatic cross-over randomised controlled trial
In brief
AI speech assistance did not ease questionnaire effort in 90 rehabilitation patients
In a trial of 90 rehabilitation patients, AI speech assistance increased average mental effort during an 82-item questionnaire compared with completing it unaided, though responses varied between users. About half said the system improved concentration and motivation, while one-third intended to use it; whether optional assistance can help specific patients remains unclear.
- Journal
- Digital health (Q2)
- Published
- 9 October 2026
- Study design
- Randomized controlled trial
- Evidence level
- Level 1, High (CEBM 1b)
- Authors
- Manuela Marquardt, Philipp Graf, Stefan Hillmann, Jan-Niklas Voigt-Antons, Philipp Harnisch, Navid Ashrafi, et al.
- PMID
- 43048053
- DOI
- 10.1177/20552076261484852
Why clinicians should know about it
- Picked for Health Informatics (top studies of the week, 22 November 2026): Pragmatic test of AI assistance, PROM usability, and clinician-facing workflow
Abstract
OBJECTIVES: Patient-reported outcome measures (PROMs) support patient-centred care and quality improvement, yet completing these questionnaires can be burdensome. We developed a multimodal assistance system using artificial intelligence methods for speech processing to enable interactive questionnaire completion. The system was implemented across three embodiments (robot, virtual agent, and voice-only), each paired with a questionnaire tablet and iteratively refined through participatory co-design. METHODS: In a pragmatic randomised cross-over trial at a German outpatient rehabilitation clinic, 90 participants with psychosomatic or neurological conditions completed a eighty-two item PROM questionnaire split into two 41-item halves, once with and once without AI speech assistance. Order of questionnaire halves and assistant embodiment were randomised. The primary outcome was subjective mental effort (Rating Scale Mental Effort, RSME); secondary outcomes included engagement, disclosure, user experience, and ratings of assistance. We analysed data with linear mixed models and repeated-measures ANOVAs. RESULTS: Assisted PROM completion was associated with higher subjective effort compared with unassisted completion, and this effect varied by general health. Between-person differences explained most of the variance (ICC = 0.75). Effort was positively associated with psychosomatic indication, poorer general health and concentration barriers, and negatively with openness to technology. Assistant embodiment showed no significant effect in this sample. System evaluations were heterogeneous: about half reported improved concentration and motivation and one-third intention to use the system. DISCUSSION: Multimodal AI-assisted PROM collection did not reduce mean subjective effort in this rehabilitation sample, but its effects varied between users. The heterogeneous evaluations suggest that flexible, optional, and equity-oriented implementation may be needed to address heterogeneous user needs.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.