Assessing the Performance of Surgical Risk Prediction Models for TAVR

Risk models routinely used for predicting 30-day mortality risk after TAVR were found to have mediocre discrimination performance.

Risk models routinely used for predicting 30-day mortality risk after transcatheter aortic valve replacement (TAVR) were found to have mediocre discrimination performance, according to a study published in Clinical Research in Cardiology.

The investigators conducted a comparative validation study of 6 surgical and TAVR-specific risk-scoring models (LogES I, ES II, STS PROM, FRANCE-2, OBSERVANT, and GAVS-II) for predicting 30-day mortality after TAVR. A total of 2946 patients (mean age, 80.9±6.1 years) from a German cohort who underwent TAVR between 2008 and 2018 (89% transfemoral TAVR procedures; 11% transapical TAVR).

The study’s primary outcome, 30-day mortality, was 3.7% in the entire cohort (transfemoral, 3.2%; transapical, 7.5%). The mean 30-day mortality risk prediction ranged from 5.8±5.0% with the OBSERVANT model to 23.4±15.9% with the LogES I model. Discrimination performance calculated with a receiver operating characteristic analysis had c-indices ranging from 0.60 with the OBSERVANT model to 0.67 with the STS PROM model, with no significant differences observed among models, between transfemoral or transapical approach, or over time.

The STS PROM model had best numerical discrimination for transfemoral TAVR (c-index, 0.66; range of c-indices, 0.60-0.66), and performance was very similar in transapical TAVR with the LogES I, ES II, FRANCE-2, and GAVS-II models which all had a c-index of 0.67.

“All models overestimated 30-day mortality risk and were poorly calibrated, especially in high-risk patients,” noted the researchers. “No significant influence of time of procedure or device type on the results could be found.”

The study authors noted that their findings are limited to risk model performance in German or European patients who undergo TAVR, and that the results may be different in other patient populations and procedural conditions.

“This analysis clarifies that risk prediction in TAVR is still an unsolved issue,” noted the study authors. “All tested models are severely limited in their performance, and dedicated TAVR-specific models are not superior to decade-old surgical scores. Development of new or updated risk models is necessary to improve risk stratification.”

Disclosures: Some of the authors declared affiliations with the pharmaceutical industry. Please see the original reference for a full list of disclosures.


Wolf G, Shamekhi J, Al‑Kassou B, et al. Risk modeling in transcatheter aortic valve replacement remains unsolved: An external validation study in 2946 German patients [published online August 26, 2020]. Clin Res Cardiol. doi: