Prognostic value of a non-invasive haemodynamic congestion index and traffic-light risk score for 90-day non-elective hospitalisation following right-heart catheterisation
A S Sauer, M F Fudim, T M MartynAbstract
Background
Invasive filling pressures measured during right-heart catheterisation (RHC) reflect haemodynamic congestion and are associated with subsequent heart failure (HF) events. We defined a haemodynamic congestion index (HCI = RAP + 4×PCWP) to emphasise left-sided filling pressure (PCWP) relative to right atrial pressure (RAP). Using non-invasive multisensor recordings acquired concomitantly with RHC, we trained machine-learning models to estimate (i) continuous HCI and (ii) PCWP- and RAP-based ‘traffic-light’ classifiers.
Purpose
To assess the prognostic value of non-invasively estimated congestion index (HCI) and PCWP/RAP traffic-light classification for 90-day non-elective hospitalisation after RHC, and to test whether combining approaches improves risk stratification.
Methods
In CAPTURE-HF (analysis set n=1,206; 90-day follow-up), reference HCI was derived from invasive RAP and PCWP. Reference PCWP traffic-light was classified into <15 (green), 15–20 (yellow) and >20 mmHg (red). Reference RAP traffic-light classes were <5 (green), 5–10 (yellow) and >10 mmHg (red). An integrated congestion score (ICS; 0 to 4) combined RAP (5/10 mmHg) and PCWP (15/20 mmHg) traffic-lights (a score of 4 = red RAP and red PCWP; a score of 0 = green RAP and green PCWP). Using non-invasive multisensor features, we generated cross-validated (CV) estimates (3 seeds; 5-fold CV) for HCI (regression) and traffic-light class (multiclass classification). We compared 90-day event rates across estimated HCI quintiles, estimated traffic-light classes, and hybrid strategies (ICS; and top quintile vs bottom quintile estimated HCI among not classified as red patients).
Results
Overall, 14.4% had a non-elective hospitalisation within 90 days. Estimated HCI quintiles showed a monotonic risk gradient (Q5 vs Q1: 20.6% vs 9.8%; risk ratio 2.10). Estimated traffic-light class separated risk (green 11.4%, yellow 15.7%, red 20.9%; p<0.005). Hybrid strategies, which combined both congestion index values and traffic-light classes, strengthened stratification: the highest ICS group had the greatest event rate (22.9%); among patients not classified as red (PCWP ≤20 mmHg), high estimated HCI identified a higher-risk subgroup (19.0% vs 11.9%; p<0.005).
Conclusions
Non-invasively estimated HCI and a PCWP- and RAP-based traffic-light classification stratified 90-day non-elective hospitalisation risk after RHC. The red traffic-light classes provide a rule-in signal, while continuous HCI adds incremental stratification among non-red patients. Combining traffic-lights with HCI (ICS and hybrid strategies) showed the strongest prognostic separation. Further validation in an independent HF cohort is warranted. If validated prospectively, this approach may have potential to aid risk stratification and inform discharge decision-making and early post-discharge management.CAPTURE-HF Phase 1 Cohort CharacteristicFor image description, please refer to the figure legend and surrounding text.Prognostic Value of HCI (Q5, Q1) – Non-eFor image description, please refer to the figure legend and surrounding text.