Table 2.

Therapeutic evidence matrix: clinical effect estimates, mechanistic directness, and inference boundaries.

Intervention or target	Clinical evidence by objective	Mechanistic evidence and directness	Defensible inference and remaining gap
SGLT2 inhibitors	Randomized-trial meta-analyses produced closely similar point estimates that differed in statistical significance: RR 0.888 (95% CI 0.806–0.978) in a network meta-analysis including standard-of-care comparators (59), RR 0.90 (0.80–1.01) in a placebo-controlled analysis (60), and RR 0.86 (0.77–0.96) in a larger placebo-controlled analysis of 52 trials, in which the effect was preserved in trials enrolling heart failure with reduced EF but not mildly reduced or preserved EF (P = 0.01) (61). AF was usually an adverse-event outcome, and the network meta-analysis found a significant reduction only for dapagliflozin (59). Post-ablation: a propensity-matched diabetes registry reported HR 0.63 (0.44–0.90); its study-level meta-analysis reported OR 0.61 (0.54–0.69), but treatment was not randomized (62). The DARE-AF randomized trial (n = 200), which excluded diabetes, heart failure, and chronic kidney disease, found no reduction in 3-month early recurrence (HR 1.11, 0.66–1.86) (63).	A (human atrial fibroblasts; non-metabolic endpoints): Empagliflozin attenuated transforming growth factor-β1-induced contraction and ECM remodeling in fibroblasts cultured from surgical atrial tissue (64) and reduced migration/procollagen I/III in commercial human atrial fibroblasts through NHE-linked pH/Ca2+ signaling (65). Neither study measured fibroblast metabolism. B (human atrial tissue): In right atrial appendages (n = 80; AF n = 18), inflammatory remodeling was associated with SGLT2 expression and ex vivo SGLT2 inhibition reduced ROS; immunofluorescence localized SGLT2 predominantly to endothelial cells and proinflammatory macrophages, with weaker cardiomyocyte colocalization, and no fibroblast marker was included (66). C (atrial animal models): Atrial animal studies report less fibrosis and, in diabetic models, lower ROS, conduction abnormalities, and inducibility (67–69).	Direct cultured-cell effects were measured on contraction, migration, and matrix output rather than metabolism, so they do not establish in situ correction of fibroblast metabolism. Human atrial tissue evidence is not fibroblast-specific and comes from a single manufacturer-supported study (66). Because DARE-AF excluded diabetes, heart failure, and chronic kidney disease, its neutral result does not test substrate modification in a metabolically loaded atrium. Clinical AF outcomes do not demonstrate regression of established human atrial fibrosis.
GLP-1 receptor agonists	Earlier randomized-trial meta-analysis: no overall AF association (70). Obesity-focused randomized-trial meta-analysis: incident AF RR 0.82 (95% CI 0.70–0.96; I2 = 0%) (71), an estimate for which possible data duplication has been raised in published correspondence (72). STEP-HFpEF was not powered for incident AF (73). Post-ablation meta-analysis, conservative random-effects estimate: HR 0.78 (0.61–0.99; I2 = 75%); a reconstructed individual-patient analysis in the same report gave HR 0.53 (0.38–0.72). Evidence mainly observational (74).	B (human atrial tissue): In the 2026 right-atrial-appendage study, inflammatory remodeling was associated with GLP-1 receptor expression and GLP-1 receptor modulation reduced ROS ex vivo; no fibroblast-specific mechanism was shown (66). Separate atrial-animal and non-atrial cardiac studies (C/D: atrial animal models and ventricular or general cardiac fibroblasts) report antifibrotic, electrophysiological, or cardiolipin effects (75–80).	No human atrial fibroblast target engagement has been shown. Clinical associations cannot be separated from weight, glycemic, hemodynamic, inflammatory, and heart-rate effects without prospective mechanistic trials.
Weight and risk-factor management	Structured weight/risk-factor programmes improved symptoms and structural measures in mixed AF populations (81, 82). Post-ablation SORT-AF primary endpoint: OR 1.14 (95% CI 0.37–3.61; neutral) (83). ARREST-AF: recurrent-arrhythmia HR 0.53 (0.32–0.89) with bundled management (84).	These interventions modify multiple systemic and paracrine inputs simultaneously. No category A (human atrial fibroblast) study has demonstrated metabolic correction in human atrial fibroblasts.	Supports context-specific risk-factor management, but weight loss alone and bundled management are not interchangeable. Neither proves regression of established interstitial fibrosis.
PFKFB3 and GLS1	No AF clinical efficacy studies identified in the cited evidence set.	Glycolysis and glutaminolysis are supported mainly in non-atrial cardiac or extracardiac fibroblasts (D/E) (8, 9, 20–23, 36).	Candidate experimental nodes only. Atrium-specific, fibroblast-restricted perturbation must show target engagement, fibrosis change, and AF change while monitoring metabolic compensation.
Fatty-acid metabolism, NOX4, and YAP/TAZ	No AF clinical efficacy studies identified in the cited evidence set.	Separate non-atrial cardiac and extracardiac studies (D/E) support individual roles in activation or mechanotransduction (26, 31–35).	Do not constitute a unified atrial fibroblast pathway. Systemic inhibition may affect multiple cell types; atrial delivery, safety, and reversibility remain unresolved.

Evidence categories are defined in Table 1. Clinical efficacy and mechanistic target engagement are evaluated separately; effect estimates are reported as published and should not be compared directly across heterogeneous designs or endpoints. AF, atrial fibrillation; CI, confidence interval; ECM, extracellular matrix; EF, ejection fraction; GLP-1RA, glucagon-like peptide-1 receptor agonist; GLS1, glutaminase 1; HR, hazard ratio; NHE, sodium–hydrogen exchanger; NOX4, NADPH oxidase 4; OR, odds ratio; PFKFB3, 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase 3; ROS, reactive oxygen species; RR, risk ratio; SGLT2, sodium–glucose cotransporter 2; TAZ, transcriptional coactivator with PDZ-binding motif; YAP, Yes-associated protein.