Ethical and feasibility constraints underlying level C recommendations in ESC guidelines: an AI-assisted mapping study

Background ESC clinical practice guidelines grade recommendations by Class (I–III) and Level of Evidence (A–C). Level C recommendations mostly rely on expert consensus, which may be conceptualized as collective empirical knowledge, derived from repeated clinical practice and shared across the professional community, and are frequently interpreted as “low evidence”, yet many may remain untested because randomised controlled trials (RCTs) are ethically unacceptable and/or operationally impractical. Purpose To quantify recommendation Class and Level of Evidence across recent ESC guidelines and, among Level C recommendations, to map ethical and feasibility constraints for RCTs. Methods All formal recommendations with explicitly stated Class and Level of Evidence were extracted from 21 ESC guidelines (2021–2025) using an AI-assisted extraction pipeline with predefined parsing rules, previously externally validated against human assessment in 36 non-ESC cardiology guidelines. Classes IIa/IIb were combined (Class II). Level C recommendations were classified into four RCT-constraint groups: Group 1 (ethical & feasible), Group 2 (ethical but not feasible), Group 3 (feasible but not ethical), and Group 4 (neither). Ethical and feasibility constraints (e.g., rarity/logistics) were mapped by AI using a predefined framework informed by the CONSORT extension for pilot/feasibility trials, as well as the Declaration of Helsinki and CIOMS documents. Results A total of 3643 recommendations were analysed. Class distribution: Class I 1698 (46.61%), Class II 1689 (46.36%), Class III 256 (7.03%). Levels of evidence: Level A 498 (13.67%), Level B 1395 (38.29%), Level C 1750 (48.04%). Within Level C, Group 1 accounted for 425 (24.29%), Group 2 205 (11.71%), Group 3 1018 (58.17%), and Group 4 102 (5.83%), corresponding to 11.67%, 5.63%, 27.94%, and 2.80% of all recommendations, respectively. Conclusions Nearly half of ESC guideline recommendations are Level C. Most Level C recommendations were classified as feasible but not ethically randomisable (Group 3), suggesting ethical constraints represent a major barrier to RCT-based evidence upgrading for many guideline questions. Manual verification of a random subset ensured consistency in classification. This framework may support research prioritisation and encourage alternative evidence-generation strategies when RCTs are not acceptable. In addition, these findings highlight that expert consensus-based recommendations often reflect ethical constraints rather than a lack of scientific rigour, underscoring the continuing role of clinical judgement in guideline development. Finally, it indicates the importance of good clinical practice and its dissemination across the professional community.