“Despite the TAVR iterations, we recognize that bleeding remains a very important and perhaps also neglected issue. Indeed, no specifically developed standard algorithm existed before this to assess bleeding risk post-TAVR,” lead author Eliano Pio Navarese, MD, PhD, said in an interview.
Although bleeding rates can be as high as 9% at 30 days and between 3% and 11% in the first year, only a few studies have applied existing scores to TAVR patients, he noted.
The PREDICT-TAVR score includes six common variables and can be calculated by hand using a simple nomogram or a web-based calculator, with a dedicated website in the works, said Dr. Navarese, Nicolaus Copernicus University and SIRO MEDICINE Network, Bydgoszcz, Poland, and the University of Alberta, Edmonton.
A strength of the score is that machine-learning methods were used and the choice of variables optimized through recursive feature elimination and cross validation to remove the weakest variables, he said. Artificial intelligence, including use of random forest, naïve Bayes, and logistic regression classifiers, was also applied to the algorithms and the results cross-checked with standard multivariate analysis.
“It was a tremendous effort in terms of the analytics conducted,” Dr. Navarese said. “This is not a simple score but the integration of the most sophisticated machine learning methods and algorithms.”
Details are published in the June 14 issue of JACC: Cardiovascular Interventions.
The six variables used to calculate 30-day bleeding risk after TAVR and the points assigned to each are:
- blood hemoglobin (0-10 points)
- serum iron concentration (0-5 points)
- common femoral artery diameter (0-3 points)
- (0-3 points)
- dual antiplatelet therapy (DAPT; 0-2 points)
- oral anticoagulation therapy (0-2 points)
The six items were selected among 104 baseline variables from 5,185 consecutive patients undergoing transfemoral TAVR in the prospective RISPEVA (Registro Italiano GISE sull’Impianto di Valvola Aortica Percutanea) registry between March 2012 and December 2019, then validated in 5,043 patients in the prospective POL-TAVI (Polish Registry of Transcatheter Aortic Valve Implantation) between January 2013 and December 2019.
In the derivation cohort, 216 patients (4.2%) experienced bleeding events at 1 year, with 169 events (78%) occurring during the first 30 days.
PREDICT-TAVR exhibited high discriminatory power for bleeding events at 30 days, as reflected by an area under the curve (AUC) of 0.80 (95% confidence interval, 0.75-0.83). Internal validation by optimism bootstrap-corrected AUC was consistent at 0.79 (95% CI, 0.75-0.83).
PREDICT-TAVR also outperformed scores not developed for TAVR, such as the PARIS score for patients undergoing percutaneous coronary intervention (AUC, 0.69) and the well-validated HAS-BLED for patients receiving anticoagulation (AUC, 0.58; P < .001 for both).
In the validation cohort, the AUC for bleeding complications at 30 days was 0.78 (95% CI, 0.72-0.82) versus an AUC of 0.68 for PARIS and 0.66 for HAS-BLED.
A HAS-BLED score of 4 predicted a higher rate of severe bleeding and mortality in the year after transfemoral TAVR in the 2018 Japanese OCEAN-TAVI study.
Bleeding events by risk categories
Risk score quartiles identified as low risk were 8 points or less, as moderate risk were 8 to less than 10 points, as high risk were 10 to less than 12 points, and as very-high-risk score were above 12 points.
In the derivation cohort, 30-day bleeding events across quartiles were 0.8%, 1.1%, 2.5%, and 8.5%, respectively (overall P < .001).
Compared with the lowest quartile, bleeding risk was numerically higher for the second quartile (odds ratio, 1.75) and significantly higher in the third (OR, 2.0) and fourth (OR, 2.49) quartiles (P < .001 for both).
A landmark cumulative-event analysis showed a significantly greater risk of bleeding for the two highest quartiles up to 30 days; however, these differences were no longer significant from 30 days to 1 year, likely because of a limited number of events, the authors suggest. Similar results were seen in the validation cohort.
The number of patients in the high- and very-high-risk groups isn’t trivial, and bleeding rates reached as high as 12.6% in the highest quartile, Dr. Navarese observed. Guidelines recommend DAPT for 3 to 6 months after TAVR; however, emerging data, including a recent meta-analysis, suggest monotherapy may be a very good option.
“So, if you had a high bleeding risk and are considering postprocedural DAPT or anticoagulation, I would think twice rather than administering dual antiplatelet therapy or anticoagulation for a long time, or at least, I would consider the impact of this score on this choice,” he said.
Subgroup analyses showed AUCs ranging from 0.77 to 0.81 for subgroups such as age older than 80 years, diabetes, obesity, female sex, previous PCI, and New York Heart Association class III or IV.
Serum iron showed the highest AUC in the primary PREDICT-TAVR model; however, should iron levels be unavailable, a simplified score modeled without iron levels retained predictive power, yielding AUCs for 30-day bleeding of 0.78 in the derivation cohort and 0.75 in the validation cohort.
“PREDICT-TAVR score can impact clinical practice, not only selecting the optimal thrombotic regimen in certain high bleeding-risk populations but also to treat pre-TAVR anemia and iron deficiencies, which may affect outcomes,” Dr. Navarese said. “Of course, future prospective biological and clinical investigations are needed to elucidate the score and the role of the score’s treatable risk traits in reducing post-TAVR bleeding complications.”
Commenting for this news organization, Sunil Rao, MD, Duke University, Durham, N.C., said anemia is a covariant in many risk models for bleeding and vascular complications in PCI and acute coronary syndrome, but hemoglobin and iron levels are collinear.
“The problem I think is when you throw hemoglobin and iron in the same model, just by play of chance, one variable can knock out the other one,” he said. “So I don’t know necessarily if we need to start measuring iron on everyone. We certainly should be measuring hemoglobin, which I think most people will have, and if a patient has pre-existing anemia, that should be a red flag for us.”
Age and Society of Thoracic Surgeons (STS) risk score did not reach statistical significance in the model – likely reflecting the high-/extremely-high-risk patient population with an average STS score of 7.7 and average age of 82 years – but may become more important as TAVR is applied more widely, Dr. Rao and Zachary Wegermann, MD, Duke Clinical Research Institute, write in an accompanying editorial.
They also point out that the study was limited by a low rate of bleeding events, and, importantly, the score can’t distinguish between minor or major bleeding.
“It’s worth trying to repeat the analyses in lower-risk patients because we may find other covariates that are important,” Dr. Rao said in an interview. “The other thing we need to get to is probably being a little bit more sophisticated. The variables included in these models are the ones that are measured; they’re also the ones that are clinically apparent.”
“But there’s a whole area of genomic medicine, proteomic medicine, metabolomic medicine that, as it starts developing and becomes more and more sophisticated, my suspicion is that we’re going to get even more precise and accurate about patients’ risk, and it’s going to become more individualized, rather than just measuring variables like age and lab values,” he said.
In the meantime, having variables documented in the electronic health record, with hard stops deployed if variables aren’t measured, is “a step in the right direction,” he added.
Dr. Navarese has received research grants from Abbott, Amgen, and Medtronic and received lecture fees and honoraria from Amgen, AstraZeneca, Bayer, Pfizer, and Sanofi-Regeneron, outside the submitted work. Dr. Rao and Dr. Wegermann report no relevant financial disclosures.
A version of this article first appeared on.