Expert consensus on feasibility and application of automatic pain assessment in routine clinical use.

backgroundPain is often difficult to assess, particularly in non-communicative patients. While artificial intelligence (AI)-based objective Automatic Pain Assessment (APA) systems are a promising solution, their clinical implementation raises essential questions, primarily regarding clinician acceptance.

methodsWe conducted a survey-to-consensus investigation on the feasibility and application of APA for clinical use. Firstly, the steering committee implemented the CHERRIES guidelines and designed a questionnaire for healthcare professionals. Given the survey results, 26 experts in pain medicine were asked to participate in a two-round consensus by rating 10 statements through a 7-point Likert scale. Consensus was defined as ≥ 75% agreement ("agree" or "completely agree"). For both phases, data was collected through online questionnaires and analyzed quantitatively.

resultsFor the survey, we collected responses from 628 healthcare professionals. The output highlighted excellent acceptance of the technology and a preference for multidimensional techniques. After two rounds, consensus was achieved on 8 out of 10 statements. Experts agreed on APA utility in supporting healthcare professionals and real-time pain monitoring. A strong consensus (96.2%) supported the need to inform patients about the use and limitations of AI systems. Adequate staff training is mandatory. Moreover, 92.3% agreed on the importance of implementing risk management, data quality control, and AI governance throughout the APA lifecycle. The experts stressed the need for internal and external validation processes and periodic updates, even for research purposes. Consensus was also reached about the importance of involving interdisciplinary stakeholders and addressing regulatory, ethical, and social implications. Multimodal inputs (e.g., physiological signals, facial expressions, speech, and clinical data) in APA systems are recommended. Additionally, APA systems should be capable of grading pain levels (e.g., via NRS), not just detecting the presence of pain. On the other hand, two statements did not reach consensus: the applicability of APA systems for acute and chronic pain conditions and their potential to improve therapeutic strategies.

conclusionAPA is viewed as a promising and potentially feasible technology for clinical pain assessment, particularly in vulnerable populations. Further research is needed to validate the dedicated tools, define applications in different clinical conditions (e.g., acute and chronic pain), and demonstrate their impact on routine clinical practice for pain management.