Background and Purpose: Causality assessment of adverse drug reactions (ADRs) is crucial for patient safety, yet traditional methods like the Naranjo algorithm remain vulnerable to human subjectivity. This study evaluated the diagnostic concordance and agreement between human clinical reviewers and the Claude AI model (version 3.5 Sonnet) using a structured Naranjo algorithmic approach to analyze cardiovascular adverse events.

Methodology: A retrospective comparative analysis was performed on 60 clinically validated cardiovascular ADR case reports retrieved from peer-reviewed literature indexed in PubMed, Embase, and Google Scholar (2022–2025). Cases were systematically coded for demographic data, severity, preventability, and System Organ Class (SOC) using MedDRA vocabulary. Statistical concordance was determined using a 3 times 3 confusion matrix and Cohen’s kappa coefficient (kappa) with its 95% confidence interval.

Results: Human evaluators classified ADRs as probable (83.3%), possible (13.3%), and definite (3.3%). Claude AI categorized the same cases as probable (85.0%), possible (11.7%), and definite (3.3%). Overall, the two methods demonstrated a 95.0% concordance rate (57/60 cases). Statistical analysis revealed an almost perfect agreement, with a Cohen’s kappa of 0.818 (95% CI: 0.732 to 0.895 , P < 0.001). Statins were the most frequently implicated drug class (35.0%), primarily associated with musculoskeletal disorders such as rhabdomyolysis. A higher prevalence was observed among females (55.0%), and patients aged 65–74 years constituted the largest age cohort (33.3%). The majority of cases ranged from serious to severe, and 95.0% of the reactions were deemed preventable. Discrepancy analysis (5.0%) revealed that the AI model struggled with complex, non-temporal clinical details.

Conclusion: Claude AI demonstrates strong alignment with human expert evaluation when utilizing structured scoring systems, acting as a highly reliable digital assistant to minimize subjectivity and accelerate signal detection workflows in pharmacovigilance.

Keywords: Adverse Drug Reactions (ADRs), Naranjo Algorithm, Artificial Intelligence (AI), Cardiovascular Drugs.