Discordance between PAH risk scores in systemic sclerosis–associated pulmonary arterial hypertension

Background risk stratification is central to treatment decisions in pulmonary arterial hypertension (PAH). Several validated tools are widely used, including ESC/ERS, REVEAL, COMPERA, and French approaches. However, these models were primarily derived from heterogeneous PAH populations and may not perform consistently in systemic sclerosis–associated PAH (SSc-PAH), a phenotype with distinct clinical characteristics and outcomes. Aim to quantify agreement between commonly used PAH risk scores in SSc-PAH and explore differences in risk distribution, calibration, and prognostic performance. Methods we analysed 678 patients with SSc-PAH from the Spanish REHAP registry. ESC/ERS (3-strata), REVEAL 2.0, REVEAL Lite 2, COMPERA 2.0 (4-strata), and French non-invasive scores were applied cross-sectionally at registry inclusion using available clinical data. Agreement between ordinal classifications was assessed using quadratic-weighted Cohen’s kappa (κw). Risk distribution, calibration against expected mortality, and transplant-free survival (TFS) were analysed. Results the cohort was predominantly female (85%), with median age 63 years and median follow-up 3.6 years. Observed mortality (death or transplantation) was 6.6% at 1 year, 23.3% at 3 years, and 43.6% at 5 years. Risk distributions differed markedly across tools. ESC/ERS classified most patients as intermediate risk (~90%), whereas REVEAL-based scores allocated substantially more patients to high risk (up to ~50%). COMPERA 2.0 mainly assigned patients to intermediate strata, while the French score showed a more balanced distribution. Agreement between scales was limited, with pairwise κw values generally low (≈0.10–0.20), indicating only slight concordance. Concordance was mainly observed for high-risk assignments, whereas agreement for low-risk categories was minimal and discordance was greatest within intermediate strata. Many patients changed category depending on the score applied (Figure). Calibration analyses revealed systematic differences between expected and observed mortality. REVEAL-based and French scores tended to overestimate 1-year mortality across several strata, whereas ESC/ERS showed better calibration in incident patients but miscalibration when prevalent cases were included. Despite discordant classification, all tools stratified TFS to some extent. REVEAL 2.0, REVEAL Lite 2 and the French score showed the highest discriminatory performance, whereas ESC/ERS displayed more limited predictive accuracy. Conclusions established PAH risk tools show poor inter-scale agreement in SSc-PAH, particularly within intermediate strata, leading to inconsistent risk labelling for individual patients. Differences in calibration and discrimination suggest that these tools are not fully interchangeable in this population and support the need for disease-specific calibration strategies.