Cross Validation of Risk Scores For Patients Under
Cross Validation of Risk Scores For Patients Under
Cross Validation of Risk Scores For Patients Under
Original Research
a
Cardiovascular Research Group, Galicia Sur Health Research Institute (IISGS), Hospital Alvaro Cunqueiro, Vigo, Spain; b East Carolina University, Greenville, North Carolina; c Clinical Trials
Center, Cardiovascular Research Foundation, New York, New York; d Columbia University Irving Medical Center, NewYork-Presbyterian Hospital, New York, New York; e Division of
Cardiovascular Diseases, Lankenau Heart Institute, Philadelphia, Pennsylvania; f Advanced Heart Failure and Cardiac Transplantation Section, Vanderbilt Heart and Vascular Institute,
Nashville, Tennessee; g Institute of Cardiology, ASST Spedali Civili, Department of Medical and Surgical Specialties, Radiological Sciences and Public Health, University of Brescia, Brescia,
Italy; h Division of Cardiovascular Medicine, The Ohio State University Wexner Medical Center, Columbus, Ohio; i Clinic Barcelona, University of Barcelona, Barcelona, Spain; j Department of
Internal Medicine, Division of Cardiology, Baylor Scott & White Heart and Vascular Hospitals, Plano, Texas; k Department of Cardiology, Hospital de la Santa Creu i Sant Pau, Autonomous
University of Barcelona, Barcelona, Spain; l Los Robles Regional Medical Center, Thousand Oaks, California; m Bakersfield Heart Hospital, Bakersfield, California; n Department of Cardiology,
University Hospital of Le on, Spain; o Division of Cardiology, Department of Medicine, University of Virginia Health System, Charlottesville, Virginia; p Division of Cardiology,
on, Le
Cardiovascular and Thoracic Department, University of Turin, Citt a Della Salute e Della Scienza, Turin, Italy; q Department of Cardiology, Sahlgrenska University Hospital and Wallenberg
Laboratory, Institute of Medicine, University of Gothenburg, Gothenburg, Sweden; r Interventional Cardiology Unit, Hospital Universitario Central de Asturias, Oviedo, Spain; s Cardiovascular
Institute, Hospital Clinico San Carlos, IdISSC, Madrid, Spain; t Cardiology Department, Quebec Heart and Lung Institute, Laval University, Quebec City, Canada; u Heart Institute Hadassah-
Hebrew University Medical Center, Jerusalem, Israel; v Cardiology Department Hospital Universitario Puerta de Hierro, Madrid, Spain; w Clinical Cardiology Unit, Faculty of Medicine, IRCCS
San Raffaele Scientific Institute, Milan, Italy; x Department of Cardiothoracic Surgery, Baylor Scott & White Health, Plano, Texas; y Clinical and Interventional Cardiology Department, IRCCS
Policlinico San Donato, San Donato Milanese, Milan, Italy; z The Zena and Michael A. Wiener Cardiovascular Institute, Icahn School of Medicine at Mount Sinai, New York, New York
A B S T R A C T
Background: Risk scores may identify patients with mitral regurgitation (MR) who are at risk for adverse events, but who may still benefit from transcatheter
edge-to-edge repair (TEER). We sought to cross-validate the MitraScore and COAPT risk score to predict adverse events in patients undergoing TEER.
Methods: MitraScore validation was carried out in the COAPT population which included 614 patients with FMR who were randomized 1:1 to guideline-
directed medical therapy (GDMT) with or without TEER and were followed for 2 years. Validation of the COAPT risk score was carried out in 1007 pa-
tients from the MIVNUT registry of TEER-treated patients with both FMR and degenerative MR who were followed for a mean of 2.1 years. The predictive
value was assessed using the area under the receiver operating characteristic curve (AUC) plots. The primary outcome was all-cause mortality.
Results: The MitraScore had fair to good predictive accuracy for mortality in the overall COAPT trial population (AUC, 0.67); its accuracy was higher in
patients treated with TEER (AUC, 0.74) than GDMT alone (AUC, 0.65). The COAPT risk score had fair predictive accuracy for death in the overall MitraScore
cohort (AUC, 0.64), which was similar in patients with FMR and degenerative MR (AUC, 0.64 and 0.66, respectively). There was a consistent benefit of
treatment with TEER plus GDMT compared with GDMT alone in the COAPT trial population across all MitraScore risk strata.
Conclusions: The COAPT risk score and MitraScore are simple tools that are useful for the prediction of 2-year mortality in patients eligible for or undergoing
treatment with TEER.
Abbreviations: AUC, area under the receiver operating characteristic curve; DMR, degenerative mitral regurgitation; FMR, functional mitral regurgitation; GDMT, guideline-directed
medical therapy; HFH, heart failure hospitalization; LVEF, left ventricular ejection fraction; LVESD, left ventricular end-systolic diameter; RV, right ventricle; RVSP, right ventricular systolic
pressure; TEER, transcatheter edge-to-edge repair; TR, tricuspid regurgitation.
Keywords: COAPT; mitral regurgitation; MitraScore; mortality; risk score; transcatheter edge-to-edge repair.
* Corresponding author: gregg.stone@mountsinai.org (G.W. Stone).
†
Co-first authors.
https://doi.org/10.1016/j.jscai.2023.101227
Received 20 July 2023; Received in revised form 16 October 2023; Accepted 30 October 2023
Available online 27 November 2023
2772-9303/© 2023 The Author(s). Published by Elsevier Inc. on behalf of the Society for Cardiovascular Angiography and Interventions Foundation. This is an open access article under
the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
2 R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227
The MitraScore population and risk score Validation of the MitraScore in the COAPT trial population. Receiver
operating characteristic (ROC) curves including the 7 clinical and 1
The MitraScore was derived from the retrospective Percutaneous echocardiographic variables of the MitraScore in the COAPT trial
Mitral Valve Repair and Nutritional Status (MIVNUT) registry of 1119 population (n ¼ 614) were generated for the following 2-year out-
patients with symptomatic MR who underwent TEER at 12 centers in comes: death, HFH, and death or HFH. The predictive value of this
Europe and Canada between 2012 and 2020 who had a mean 2.1-year model was assessed using the area under curve (AUC) of the ROC plot.
follow-up.9 FMR was present in 658 (59.3%) patients and the remainder Subgroup analysis was performed in patients assigned to the GDMT
had DMR. A total of 112 participants with unknown values for at least 1 alone group (n ¼ 312) and to the MitraClip plus GDMT group (n ¼
of the characteristics required for the COAPT risk score were excluded. 302). Kaplan-Meier survival curves were generated for the incidence
Therefore, the final MitraScore cohort from the MIVNUT registry in the of death, HFH, and the composite of death or HFH in the 3 risk strata
present study (hereafter referred to as the “MitraScore cohort”) of the MitraScore (0-2, 3-4, 5-8 points) in all patients and separately for
comprised 1007 patients. the GDMT alone and MitraClip plus GDMT groups, with differences
The MitraScore was derived by assigning 1 point to each of the 8 assessed using the log-rank statistic. Cubic splines were generated to
independent predictors: age 75 years, anemia, estimated glomerular assess the continuous relationship between the MitraScore and out-
filtration rate (eGFR) <60 mL/min/1.73 m2, LVEF <40%, peripheral ar- comes in the COAPT trial.
tery disease, chronic obstructive pulmonary disease, high diuretic dose
use (80 mg of furosemide per day or use of 2 diuretic agents Validation of the COAPT risk score in the MitraScore cohort. ROC
excluding aldosterone antagonists), and no therapy with renin- curves including the 4 clinical and 4 echocardiographic variables of
angiotensin system inhibitors.9 Patients were classified into 3 risk the modified COAPT risk score were generated in the MitraScore
groups of MitraScore: low-risk (0-2 points), moderate-risk (3-4 points), cohort (n ¼ 1007 patients undergoing TEER) for the following out-
and high-risk (5 points). comes: death, HFH, death, or HFH. The predictive value of these
models was assessed using the AUC of the ROC plots. Subgroup
analysis was performed in those with FMR (n ¼ 594) and those with
The COAPT population and risk score degenerative MR (n ¼ 413). Kaplan-Meier survival curves were
generated for the incidence of death, HFH, and the composite of
The COAPT trial enrolled 614 patients with heart failure (HF), left death or HFH in the 3 strata of the modified COAPT risk score (0-4, 5-
ventricular ejection fraction (LVEF) 20%-50%, and moderate-to-severe 8, 9-15 points) in all patients and separately for the FMR and DMR
(3þ) or severe (4þ) FMR confirmed by an independent echocardio- groups, with differences assessed using the log-rank statistic. Cubic
graphic core laboratory who remained symptomatic despite maximally splines were generated to assess the continuous relationship between
tolerated GDMT and cardiac resynchronization therapy (CRT) if appro- the COAPT risk score and outcomes in the MitraScore cohort. Additive
priate at 78 sites in the United States and Canada.3 Enrolled patients were and multiplicative interactions between treatment effect with TEER (vs
randomized 1:1 to TEER with the MitraClip plus GDMT or GDMT alone. GDMT) and MitraScore risk groups on event outcomes were evaluated
The COAPT risk score was derived using 4 clinical, 4 echocardio- using the relative excess risk due to interaction and Cox regression
graphic, and 1 treatment variable.10 Different points were assigned to models, respectively.
R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227 3
Table 1. Baseline characteristics of the MitraScore cohort and the COAPT Table 2. Baseline laboratory and echocardiographic data and therapies of
trial population. the MitraScore cohort and the COAPT trial population.
Variables MitraScore COAPT P Variables MitraScore COAPT P value
(n ¼ 1007) (n ¼ 614) value (n ¼ 1007) (n ¼ 614)
Age, y 73.5 10.4 72.2 11.2 .02 Baseline creatininea, g/dL 1.5 1.0 1.8 1.3 <.001
75 y 505 (50.1%) 291 (47.4%) .28 Baseline eGFRb, mL/min 52.9 23.1 49.3 26.8 .005
Female sex 368 (36.5%) 221 (36.0%) .82 Baseline hemoglobinc, g/dL 12.2 1.9 12.1 1.7 .29
Race and ethnicity – Left ventricular ejection fractiond, % 39.7 15.9 31.3 9.3 <.001
White N/A 457 (74.4%) <40% 547 (54.3%) 465 (80.9%) <.001
Black N/A 88 (14.3%) 25%-35% 339 (33.7%) 250 (43.5%) <.001
Hispanic N/A 40 (6.5%) <25% 142 (14.1%) 159 (27.7%) <.001
Other N/A 29 (4.7%) LV end-systolic dimensione, cm 4.6 1.1 5.3 0.9 <.001
MR classification >5.5 cm 171 (17.0%) 236 (38.9%) <.001
Functional MR 594 (59.0%) 614 (100%) <.001 LV end-diastolic dimensionf, cm 6.2 1.8 6.2 0.7 1.0
Degenerative MR 413 (41.0%) 0 (0%) <.001 Right ventricular systolic pressureg, 49.2 14.9 44.3 13.7 <.001
NYHA functional classa <.001 mm Hg
I 3 (0.3%) 1 (0.2%) >45 mm Hg 484 (48.1%) 238 (45.1%) .27
II 147 (14.6%) 239 (39.0%) Tricuspid regurgitationh <.001
III 642 (63.7%) 322 (52.5%) 0 (none) 30 (3.0%) 27 (4.4%)
IV 215 (21.4%) 51 (8.3%) 1þ 435 (43.2%) 489 (79.6%)
NYHA III-IVa 857 (85.1%) 373 (60.8%) <.001 2þ 264 (26.2%) 92 (15.0%)
Comorbidities 3þ 182 (18.1%) 5 (0.8%)
Hypertension 720 (71.5%) 494 (80.5%) <.001 4þ 96 (9.5%) 1 (0.2%)
Hypercholesterolemia 561 (55.7%) 329 (53.6%) .40 TR 2þ 542 (53.8%) 98 (16.0%) <.001
Diabetes mellitus 351 (34.9%) 229 (37.3%) .32 Left atrial sizei <.001
Coronary artery disease 572 (56.8%) 446 (72.6%) <.001 Normal (LAVI <34 mL/m2) 42 (9.1%) 145 (24.0%)
Prior myocardial infarction N/A 316 (51.5%) – Mild-moderately dilated 19 (4.1%) 213 (35.2%)
History of coronary artery bypass 191 (19.0%) 247 (40.2%) <.001 (LAVI 34-48 mL/m2)
grafting Severely dilated (LAVI >48 mL/m2) 402 (86.9%) 247 (40.8%)
Prior percutaneous coronary 346 (35.8%) 283 (46.1%) <.001 Therapies
intervention TEER 1007 (100%) 302 (49.2%) <.001
Previous stroke 105 (10.4%) 72 (11.7%) .42 High diuretic dose 307 (30.5%) 287 (46.7%) <.001
History of transient ischemic attack N/A 43 (7.0%) – No therapy with RAS inhibitors 360 (36.7%) 203 (33.1%) .27
Peripheral vascular disease 165 (16.4%) 109 (17.8%) .48
Chronic obstructive pulmonary 226 (22.4%) 143 (23.3%) .69 Values are mean SD or n (%).
disease LAVI, left atrial volume index; LV, left ventricular; LVEF, left ventricular ejection
Atrial fibrillation 594 (59.0%) 339 (55.2%) .14 fraction; LVESD, left ventricular end-systolic diameter; MR, mitral regurgitation;
CKDb stage III 485 (48.2%) 324 (53.6%) .04 NYHA, New York Heart Association; RAS, renin-angiotensin system; RVSP, right
CKDb stage IV or greater 165 (16.4%) 142 (23.5%) <.001 ventricular systolic pressure; TEER, transcatheter edge-to-edge repair; TR,
Anemia 582 (57.8%) 144 (23.5%) <.001 tricuspid regurgitation.
a
Baseline creatinine data were available for 606 patients in the COAPT trial
Values are mean SD or n (%).
cohort. b Baseline eGFR data were available for 601 patients in the COAPT trial
CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; MR,
cohort. c Baseline hemoglobin data were available for 587 patients in the
mitral regurgitation; NYHA, New York Heart Association.
a COAPT trial cohort. d Baseline LVEF data were available for 575 patients in the
NYHA functional class data were available for 613 patients in the COAPT
COAPT trial cohort. e LVESD data were available for 607 patients in the COAPT
trial cohort. b CKD stages were determined by estimated glomerular filtration
trial cohort. f LV end-diastolic dimension data were available for 608 patients in
rate (calculated by Cockcroft-Gault equation) with stage III CKD defined as
the COAPT trial cohort. g RVSP estimates were available for 528 patients in the
eGFR 30-60 mL/min/1.73m2 and stage IV or greater CKD defined as eGFR
COAPT trial cohort (in the remainder RVSP could not be estimated because of
<30 mL/min/1.73m2. CKD data were available for 605 patients in the COAPT
the absence of a TR jet). These 86 missing values were imputed as normal RVSP
trial cohort; missing values were otherwise imputed using multiple
(ie, placed in the <45 mm Hg). h There were 15 missing values for TR in the
imputations.
COAPT trial cohort which were treated as absent or no TR ie, TR grade 0. i LAVI
data were available for only 463 patients in the MitraScore cohort and 605
patients in the COAPT trial cohort. Missing values were otherwise imputed
Differences in baseline characteristics were evaluated using the χ2
using multiple imputations.
test for categorical variables, t test for continuous variables that were
normally distributed and the Kruskal Wallis test for continuous vari-
ables that were not normally distributed. AUC values of 0.5 to 0.6 were
considered as poor discrimination, 0.6 to 0.7 as fair discrimination, 0.7 analysis (n ¼ 1007), the average duration of follow-up was 2.1 1.8 years
to 0.8 as good discrimination, and 0.8 as excellent discrimination.11 and 59.0% of patients had FMR. During follow-up, 313 patients died
Two-sided P values <.05 were considered statistically significant. (14.6 per 100 patient-years), 297 HFH were reported (16.2 per 100
Statistical analysis was performed using SAS version 9.4 (SAS patient-years) and 452 patients died or had a HFH (24.6 per 100 patient-
Institute). years).
In the COAPT trial population, 125 (40.1%), 158 (50.6%), and
201 (64.4%) patients in the GDMT alone group and 83 (27.5%), 95
Results (31.5%), and 133 (44.0%) patients in the MitraClip plus GDMT group
experienced death, HFH, or a composite of death or HFH respec-
Patient populations and outcomes tively during the 2-year follow-up period (P < .001 for all 3
comparisons).
From the MIVNUT registry, there were no significant differences in As shown in Tables 1 and 2, the COAPT trial population
the baseline characteristics of the 1007 patients that were included in compared with the MitraScore cohort had higher proportions of
this analysis compared with the 112 patients that were excluded patients with hypertension, coronary artery disease, CKD-III, and
(Supplemental Table S1). In the MitraScore cohort included in the final CKD-IV or greater, and treatment with higher diuretic doses. The
4 R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227
COAPT population also had a lower mean LVEF and a higher mean
Table 3. AUC values (with 95% CI) of the modified COAPT risk score and the
LVESD but was slightly younger and had a lower proportion of
MitraScore for 2-year outcomes in the COAPT trial.
patients with New York Heart Association III/IVa functional class,
Outcomes Modified COAPT MitraScore AUC
history of anemia and TR 2þ, and had lower mean RVSP. There
risk score AUC
were also significant differences in the baseline characteristics be-
Overall COAPT population (n ¼ 614) tween the patients enrolled in COAPT and the FMR cohort of
Death 0.74 (0.69-0.78) 0.67 (0.63-0.72) MitraScore (Supplemental Tables S2 and S3).
HFH 0.68 (0.65-0.77) 0.57 (0.54-0.63)
Death or HFH 0.71 (0.66-0.77) 0.60 (0.57-0.65)
COAPT GDMT-alone group (n ¼ 312)
Death 0.76 (0.69-0.83) 0.65 (0.60-0.73)
External validation of MitraScore in the COAPT trial population
HFH 0.69 (0.66-0.79) 0.57 (0.51-0.70)
Death or HFH 0.71 (0.66-0.79) 0.57 (0.52-0.68)
COAPT MitraClip plus GDMT group (n ¼ 302) As shown in Table 3, the MitraScore had fair to good accuracy for
Death 0.72 (0.67-0.82) 0.74 (0.69-0.80) predicting 2-year mortality in the overall COAPT trial population (AUC
HFH 0.74 (0.68-0.82) 0.57 (0.53-0.67) 0.67). Its predictive accuracy was numerically better among COAPT pa-
Death or HFH 0.76 (0.69-0.83) 0.65 (0.59-0.71)
tients who underwent TEER than those who were treated with GDMT
AUC, area under the receiver operating characteristic curve; GDMT, guideline- alone (AUC 0.74 vs 0.65, respectively, P ¼.06). The AUC for the MitraScore
directed medical therapy; HFH, heart failure hospitalization. in the COAPT TEER plus GDMT group was similar to that of the modified
Central Illustration.
Distribution of MitraScore and COAPT risk score. (A) Frequency distribution of MitraScore in the COAPT trial population superimposed over a cubic spline analysis demonstrating the
continuous relationship between the MitraScore and the primary outcome or 2-year death in the COAPT trial population. (B) Frequency distribution of the modified COAPT risk score in
the MitraScore cohort superimposed over a cubic spline analysis demonstrating the continuous relationship between the modified COAPT risk score and primary outcome of death at a
mean follow-up of 2.1 years in the MitraScore cohort. AUC, area under curve; eGFR, estimated glomerular filtration rate; LVESD, left ventricular end-systolic dimension; NYHA, New York
Heart Association; RAS, renin-angiotensin system; RVSP, right ventricular systolic pressure.
R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227 5
Figure 1.
Kaplan-Meier hazard curves in the COAPT trial population according to MitraScore risk groups. (A) Mortality. (B) Heart failure hospitalization. (C) The composite of death or heart
failure hospitalization. Top: All patients; Center: GDMT alone group; Bottom: MitraClip plus GDMT group. GDMT, guideline-directed medical therapy.
COAPT risk score (0.74 and 0.72 respectively). The MitraScore had poor FMR and DMR subgroups. Supplemental Table S4 shows the AUC for the
accuracy for predicting 2-year HFH in the overall COAPT trial population modified COAPT risk score in predicting 1-year mortality in the COAPT
and in those treated with TEER plus GDMTor GDMT alone (AUC 0.57 in all trial and MitraScore populations.
3 groups). As seen in the Central Illustration and Figure 1A, there was a
continuous increase in the primary outcome of 2-year mortality in the
COAPT trial population with increasing MitraScore values. As also seen in FMR patients undergoing TEER
Figure 1, there was a steady increase in 2-year mortality with increasing
MitraScore values in both the TEER plus GDMT group as well as the GDMT Restricting the analyses to patients with FMR undergoing TEER in
alone group of the COAPT trial population. Figure 1 also demonstrates the COAPT trial, the MitraScore had good accuracy (AUC 0.74) in
that the MitraScore was associated with HFH and the composite of death
or HFH in the overall population as well as in the TEER plus GDMT group
but not in the GDMT alone group. Supplemental Table S4 shows the AUC Table 4. AUC values (with 95% CI) of the modified COAPT risk score and the
for the MitraScore in predicting 1-year mortality in the MitraScore and MitraScore for 2.1-year outcomes in the MitraScore cohort.
COAPT trial populations. Outcomes Modified COAPT MitraScore AUC
risk score AUC
Figure 2.
Kaplan-Meier hazard curves in the MitraScore population according to COAPT risk groups. (A) Mortality. (B) Heart failure hospitalization. (C) The composite of death or heart failure
hospitalization. Top: All patients; Center: DMR group; Bottom: FMR group. DMR, degenerative mitral regurgitation; FMR, functional mitral regurgitation; HFH, heart failure
hospitalization.
predicting 2-year all-cause mortality, similar to that of the COAPT risk MitraScore demonstrated fair to good accuracy for the prediction of
score (AUC 0.72) (Table 3). However, the COAPT risk score had mortality in the external validation cohorts. The predictive utility of the
numerically better accuracy in predicting the 2-year rate of HFH MitraScore in the COAPT trial population was better among TEER-
compared to the MitraScore in the COAPT trial population (AUC 0.74 vs treated patients than in those treated with GDMT alone. The Mitra-
0.57, respectively). Among patients with FMR undergoing TEER in the Score had lower accuracy for predicting HFH than mortality.
MIVNUT registry, the COAPT risk score had fair accuracy (AUC 0.64) to Conversely, the COAPT risk score showed fair predictive accuracy in
predict 2.1-year all-cause mortality, similar to that of the MitraScore the MitraScore cohort (all of whom had TEER) for predicting both
(AUC 0.68), and at least as good predictive accuracy for the 2.1-year death and HFH, including both FMR and DMR patients (even though
rate of HFH (AUC 0.61 vs 0.58, respectively) (Table 4). the COAPT risk score was derived from a pure FMR population). TEER
had a positive effect among randomized patients in the COAPT trial in
reducing death and HFH in all risk categories of the MitraScore
Benefit of TEER in the COAPT trial population according to compared with GDMT alone, although the magnitude of reduction
MitraScore risk score groups appeared to be greater in the lower risk strata.
TEER is presently indicated for select patients with FMR (principally
There were no significant interactions between treatment with TEER those meeting COAPT criteria) and for those with DMR at high surgical
plus GDMT compared with GDMT alone and the risk strata of the risk.1,2 However, patients still have high rates of early and late adverse
MitraScore for the absolute and relative risk reduction in 2-year outcomes events after TEER principally due to their associated comorbidities and
in patients from the COAPT trial (Figure 3). However, by cubic spline underlying degree of ventricular dysfunction. In this regard, scoring
analysis, the greatest risk reductions in death and the composite of death systems may assist heart teams and patients in shared decision-making
or HFH were observed in the lower risk strata of the MitraScore (Figure 4). as well as create awareness in patients and family members regarding
the expected outcome of a procedure. Choosing cost-effective thera-
pies is also important in the contemporary era of precision medicine,12 a
Discussion consideration for which risk score can also assist.
An important prerequisite for the acceptance of generalizable risk
This present analysis reports the cross-validation of 2 recently stratification models is the demonstration of external validity. The pre-
developed risk scores to predict clinical events in patients with MR sent study demonstrates that both the COAPT risk score and the
and in those undergoing TEER. Both the COAPT risk score and MitraScore have acceptable accuracy in predicting survival in patients
R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227 7
Figure 3.
Absolute and relative risk reductions in 2-year out-
comes with TEER plus GDMT compared with GDMT
alone in the COAPT trial population by risk strata of
the MitraScore. Top: Absolute risk reduction. P value
for additive interaction (for ARR) between treatment and
MitraScore risk group on death ¼ .41; P value for addi-
tive interaction between treatment and MitraScore risk
group on HFH ¼ .09; P value for additive interaction
between treatment and MitraScore risk group on death
or HFH ¼ .62. Bottom: Relative risk reduction. P value for
multiplicative interaction (for RRR) between treatment
and MitraScore risk group on death ¼ .09; P value for
multiplicative interaction between treatment and Mitra-
Score risk group on HFH ¼ .62; P value for multiplicative
interaction between treatment and MitraScore risk
group on death or HFH ¼ .15. ARR, absolute risk
reduction; GDMT, guideline-directed medical therapy;
HFH, heart failure hospitalization; TEER, transcatheter
edge-to-edge repair.
with severe MR considered for TEER. A major advantage of both scoring Other risk scores have been developed to predict outcomes in pa-
systems is that they are derived from clinical and echocardiographic tients with MR but have additional limitations. The GRASP (Getting
data that are readily available prior to the procedure, without additional Reduction of mitrAl inSufficiency by Percutaneous clip implantation)
costs beyond the standard of care. The MitraScore uses 7 preprocedural score,7 derived from a small cohort of patients undergoing TEER, is
clinical and laboratory-based measures and 1 echocardiographic vari- computationally complex as it requires weighting and logarithmic terms
able (LVEF) for mortality prediction. Despite its simplicity, the Mitra- for the predictor variables, and cannot be implemented easily. The
Score has shown a good ability to predict death in both patients with MITRALITY risk score for patients undergoing TEER was developed with
FMR and DMR undergoing TEER. Perhaps not surprisingly, the Mitra- machine learning techniques and uses 6 baseline clinical features for
Score (derived from a patient population that had undergone TEER), prediction.6 However, despite good predictive value (AUC 0.78 for
showed better predictive accuracy in the TEER plus GDMT group of the mortality), this score is also computationally complex, requiring an on-
COAPT trial population (AUC 0.74 for mortality) than the GDMT alone line calculator. Additionally, the variables included in this score are not
group (AUC 0.65 for mortality). In contrast, the 9-variable COAPT risk specific for TEER and reflect general factors prognostic for most cardiac
score showed good predictive ability for 2-year all-cause mortality in procedures. In contrast to these risk scores, the COAPT risk score and
both the TEER plus GDMT as well as GDMT alone groups of the COAPT the MitraScore offer advantages since they comprise readily available
trial population (AUC 0.72 and 0.76, respectively [Table 3]), and also variables and include disease-specific variables that predict post-TEER
was moderately accurate for the prediction of HFH and the composite outcomes, and from spline analysis may identify which patients benefit
of death or HFH from COAPT. Moreover, although the COAPT risk score most from TEER, as also previously reported from the COAPT risk score
was derived from an FMR patient population, it showed fair predictive derivation data.10 Nonetheless, acknowledging the limitations of
accuracy for both mortality and HFH in both the FMR and DMR groups cross-study comparisons, future studies are warranted to examine the
of the MitraScore cohort. This wider utility may be explained by the relative utility of these risk scores compared with the COAPT risk score
inclusion of risk variables in the COAPT risk score that are known to and MitraScore from a common external study population.
impact outcomes in both FMR and DMR, including pulmonary hyper- The present findings highlight the importance of patient selection to
tension, the presence of significant TR, reduced LVEF, and LV optimize the outcomes of the TEER procedure. Nonetheless, the
dilatation.13–15 benefit from TEER is also realized in patients with advanced disease, ie,
8 R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227
Figure 4.
Cubic spline analysis relating the MitraScore to the absolute
and relative risks of 2-year outcomes in patients treated with
MitraClip plus GDMT vs GDMT alone. Top: Death. Center: HFH.
Bottom. Death or HFH. The figure shows restricted cubic splines
with 3 knots located at risk score values of 2, 3, and 5. HR (blue
line and blue shadow for 95% CI) and absolute risk difference (red
line and red shadow for 95% CI) for the 2-year rate of outcomes
(death, HFH, death or HFH) after treatment with MitraClip plus
GDMT compared with GDMT alone across the range of Mitra-
Score values. P values for nonlinearity in the HR relationship were
0.21, 0.80, and 0.49 for death, HFH, and death or HFH respec-
tively indicating that the relative treatment effects of the MitraClip
plus GDMT compared with GDMT alone were independent of the
risk score. GDMT, guideline-directed medical therapy; HFH, heart
failure hospitalization; HR, hazard ratio.
those in the “high-risk” categories of the MitraScore and COAPT risk terms of life expectancy, but the rate of HFH may be reduced and
score. Knowledge of potential outcomes with continued GDMT vs quality of life may be improved.16,17 Although a formal comparison
benefit from TEER in these “high-risk” populations can enable patients between the MitraScore and COAPT risk score has not been performed
and physicians to make an informed decision about the best course of as it would require an additional external validation dataset, the main
action. In some cases, there may only be a modest benefit expected in advantage of the former is the simplicity of the score, and of the latter
R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227 9
the inclusion of echocardiographic MR-specific variables and improved grant from the National Institutes of Health/National Heart, Lung, and
prediction for HFH (as well as its greater utility in a GDMT-alone–treated Blood Institute to Columbia University Irving Medical Center
patient population). Both scores have now been externally validated (T32HL007854). JoAnn Lindenfeld reports research grant support from
and consequently represent valid instruments for evaluating patients AstraZeneca, and consulting income from Abbott Vascular, AstraZeneca,
with severe MR referred for consideration of TEER. CVRx, Edwards Lifesciences, RESMED, Relypsa, Boehringer Ingelheim,
and V-Wave. William Abraham reports research grant support from
Abbott Vascular, and National Institutes of Health–National Heart, Lung,
Limitations and Blood Institute, consulting income from Abbott Vascular, and
speaker honoraria from Impulse Dynamics. Xavier Freixa has served as a
First, the methods used to derive and validate the models were proctor for Abbott Vascular and has received consultant fees for Abbott
simple and based on C-statistic or AUC. We did not perform more so- Vascular. Paul Grayburn has received grant support and consulting fees
phisticated analyses such as the net reclassification index or the inte- from Edwards Lifesciences and Neochord and grant support from Boston
grated discrimination index. Nonetheless, the AUC is a widely accepted Scientific, Medtronic, and Tendyne. Dabit Arzamendi has received
tool to assess predictive models. Second, the strength of both scores is consultant fees for Abbott Vascular and Edwards Lifesciences. Saibal Kar
that the majority of predictors in our final models are measures that are has been a consultant for Boston Scientific, W.L. Gore, and Medtronic
routinely available. However, the overall predictive ability of both risk and holds equity in Valcare. D. Scott Lim has received research grant
scores was moderate at best (although somewhat better in the FMR TEER support from Abbott, Boston Scientific, Edwards Lifesciences, and
treatment subgroup). Although other risk scores may perform well or Medtronic; and consultant/advisory board fees from Ancora, Philips,
slightly better,6 the computational complexity of these scores challenges LagonaTech, Valgen, and Venus. Luis Nombela-Franco has served as a
their practical use. Third, many of the variables included in the COAPT proctor for Abbott Vascular and has received speaker fees from Edwards
risk score and MitraScore are dynamically affected by HF medications Lifesciences and Boston Scientific. Josep Rod es-Cabau holds the
and CRT; however, it is assumed that all patients are treated with maxi- Research Chair “Fondation Famille Jacques Larivi ere” for the develop-
mally tolerated GDMT (including CRT as appropriate) prior to TEER ment of structural heart disease interventions (Laval University) and has
(which was confirmed by a central eligibility committee in COAPT). RVSP received institutional research grants and consultant-speaker fees from
and TR, which are included only in the COAPT risk score, are sensitive to Edwards Lifesciences, Medtronic, and Abbott. Michael Mack served as
volume status and may be elevated in patients with decompensated HF. co-primary investigator for the PARTNER trial for Edwards Lifesciences
However, the COAPT trial only included stabilized patients after maximal and COAPT trial for Abbott and as study chair for the APOLLO trial for
GDMT optimization, including diuretics. Neither score includes a mea- Medtronic; all activities were unpaid. Francesco Bedogni is a proctor for
sure of natriuretic peptides, an important prognostic biomarker. Several Boston Scientific. Gregg W. Stone has received speaker honoraria from
parameters such as posttreatment reduction in MR grade, severity of Medtronic, Pulnovo, Infraredx, and Abiomed; has served as a consultant
residual MR, posttreatment RVSP, right ventricular (RV) function, and to Valfix, TherOx, Robocath, HeartFlow, Ablative Solutions, Vectorious,
RV-pulmonary artery coupling which afford additional prognostic insights Miracor, Neovasc, Ancora, Elucid Bio, Occlutech, CorFlow, Apollo Ther-
in this patient population18,19 were not included in either risk score. The apeutics, Impulse Dynamics, Cardiomech, Gore, Amgen, Adona Medi-
extent to which these variables alter or further improve the predictive cal, Millennia Biopharma; and has equity/options from Ancora, Cagent,
accuracy of the COAPT risk score and the MitraScore is unknown. We also Applied Therapeutics, Biostar family of funds, SpectraWave, Orchestra
did not explore whether either score could be simplified by removing 1 or Biomed, Aria, Cardiac Success, Valfix, Xenter. Gregg Stone’s daughter is
more variables without substantially affecting predictive accuracy. Finally, an employee at IQVIA. Institutional disclosure: Gregg Stone’s employer,
there are important differences between the COAPT and MIVNUT co- Mount Sinai Hospital, receives research support from Abbott, Abiomed,
horts: the COAPT trial only included patients with FMR, but this cohort Bioventrix, Cardiovascular Systems Inc, Phillips, Biosense-Webster,
underwent treatment with GDMT alone as well as TEER plus GDMT, Shockwave Medical, Vascular Dynamics, Pulnovo, and V-Wave. All
whereas the MIVNUT registry included both DMR and FMR patients, but other authors have reported that they have no relationships relevant to
all were treated with TEER. Despite the notable variations between the 2 the contents of this paper to disclose.
study cohorts, each score showed its potential for generalizability by
demonstrating at least moderate predictive accuracy in the other cohort. Funding sources
The present analysis provides external validation of both the COAPT Ethics statement and patient consent
risk score and the MitraScore in independent cohorts, with both risk
scores exhibiting overall fair to good predictive value for 2-year mortality Both the MIVNUT registry and the COAPT trial were conducted
after TEER. The COAPT risk score also provided fair to good accuracy in according to the relevant ethical guidelines and all patients provided
predicting 2-year HFH and was equally prognostic in both DMR and FMR written informed consent.
patients undergoing TEER, as well as in FMR patients treated with GDMT
alone. Both risk scores are simple to use in practice, being readily derived
Supplementary material
from clinical, laboratory, and echocardiographic variables that are
routinely measured as standard of care, and both provide utility for the
To access the supplementary material accompanying this article,
prediction of outcomes in patients eligible for or undergoing treatment
visit the online version of the Journal of the Society for Cardiovascular
with TEER. Use of these risk scores may aid in the early identification and
Angiography & Interventions at 10.1016/j.jscai.2023.101227.
referral of appropriate patients for transcatheter MR therapies.
1. Otto CM, Nishimura RA, Bonow RO, et al. 2020 ACC/AHA guideline for the
Rodrigo Estevez-Loureiro is a consultant for Abbott Vascular, Edwards management of patients with valvular heart disease: a report of the American
Lifesciences, and Boston Scientific. Mahesh Madhavan is supported by a College of Cardiology/American Heart Association Joint Committee on Clinical
10 R. Est
evez-Loureiro et al. / Journal of the Society for Cardiovascular Angiography & Interventions 3 (2024) 101227
Practice Guidelines. Circulation. 2021;143(5):e72–e227. https://doi.org/10.1161/ 11. Pencina MJ, D’Agostino Sr RB. Evaluating discrimination of risk prediction
CIR.0000000000000923 models: the C statistic. JAMA. 2015;314(10):1063–1064. https://doi.org/10.1001
2. Vahanian A, Beyersdorf F, Praz F, et al. 2021 ESC/EACTS Guidelines for the /jama.2015.11082
management of valvular heart disease. Eur Heart J. 2022;43(7):561–632. https:// 12. Adamo M, Capodanno D, Cannata S, et al. Comparison of three contemporary surgical
doi.org/10.1093/eurheartj/ehab395 scores for predicting all-cause mortality of patients undergoing percutaneous mitral
3. Stone GW, Lindenfeld J, Abraham WT, et al. Transcatheter mitral-valve repair in valve repair with the MitraClip system (from the multicenter GRASP-IT registry). Am J
patients with heart failure. N Engl J Med. 2018;379(24):2307–2318. https:// Cardiol. 2015;115(1):107–112. https://doi.org/10.1016/j.amjcard.2014.09.051
doi.org/10.1056/NEJMoa1806640 13. David TE, Armstrong S, McCrindle BW, Manlhiot C. Late outcomes of mitral valve
4. Obadia JF, Messika-Zeitoun D, Leurent G, et al. Percutaneous repair or medical repair for mitral regurgitation due to degenerative disease. Circulation. 2013;
treatment for secondary mitral regurgitation. N Engl J Med. 2018;379(24): 127(14):1485–1492. https://doi.org/10.1161/CIRCULATIONAHA.112.000699
2297–2306. https://doi.org/10.1056/NEJMoa1805374 14. Magne J, Lancellotti P, Pi erard LA. Exercise pulmonary hypertension in
5. Mack MJ, Lindenfeld J, Abraham WT, et al. 3-year outcomes of transcatheter mitral asymptomatic degenerative mitral regurgitation. Circulation. 2010;122(1):33–41.
valve repair in patients with heart failure. J Am Coll Cardiol. 2021;77(8):1029–1040. https://doi.org/10.1161/CIRCULATIONAHA.110.938241
https://doi.org/10.1016/j.jacc.2020.12.047 15. Essayagh B, Antoine C, Benfari G, et al. Functional tricuspid regurgitation of
6. Zweck E, Spieker M, Horn P, et al. Machine learning identifies clinical parameters to degenerative mitral valve disease: a crucial determinant of survival. Eur Heart J.
predict mortality in patients undergoing transcatheter mitral valve repair. JACC 2020;41(20):1918–1929. https://doi.org/10.1093/eurheartj/ehaa192
Cardiovasc Interv. 2021;14(18):2027–2036. https://doi.org/10.1016/j.jcin.2021. 16. Lindenfeld J, Abraham WT, Grayburn PA, et al. Association of effective regurgitation
06.039 orifice area to left ventricular end-diastolic volume ratio with transcatheter mitral valve
7. Buccheri S, Capodanno D, Barbanti M, et al. A risk model for prediction of 1-year repair outcomes: a secondary analysis of the COAPT trial. JAMA Cardiol. 2021;6(4):
mortality in patients undergoing MitraClip implantation. Am J Cardiol. 2017; 427–436. https://doi.org/10.1001/jamacardio.2020.7200
119(9):1443–1449. https://doi.org/10.1016/j.amjcard.2017.01.024 17. Godino C, Munaf o A, Scotti A, et al. MitraClip in secondary mitral regurgitation as a
8. Iliadis C, Spieker M, Kavsur R, et al. "Get with the Guidelines Heart Failure bridge to heart transplantation: 1-year outcomes from the International MitraBridge
Risk Score" for mortality prediction in patients undergoing MitraClip. Clin Registry. J Heart Lung Transplant. 2020;39(12):1353–1362. https://doi.org/10.1016/
Res Cardiol. 2021;110(12):1871–1880. https://doi.org/10.1007/s00392-021- j.healun.2020.09.005
01804-3 18. Ben-Yehuda O, Shahim B, Chen S, et al. Pulmonary hypertension in transcatheter
9. Raposeiras-Roubin S, Adamo M, Freixa X, et al. A Score to assess mortality after mitral valve repair for secondary mitral regurgitation: the COAPT trial. J Am Coll
percutaneous mitral valve repair. J Am Coll Cardiol. 2022;79(6):562–573. https:// Cardiol. 2020;76(22):2595–2606. https://doi.org/10.1016/j.jacc.2020.09.609
doi.org/10.1016/j.jacc.2021.11.041 19. Kar S, Mack MJ, Lindenfeld J, et al. Relationship between residual mitral
10. Shah N, Madhavan MV, Gray WA, et al. Prediction of death or HF hospitalization in regurgitation and clinical and quality-of-life outcomes after transcatheter and
patients with severe FMR: the COAPT risk Score. JACC Cardiovasc Interv. 2022; medical treatments in heart failure: COAPT trial. Circulation. 2021;144(6):
15(19):1893–1905. https://doi.org/10.1016/j.jcin.2022.08.005 426–437. https://doi.org/10.1161/CIRCULATIONAHA.120.053061