new risk calculator for heart failure

 secondary analysis of a large trial of patients with heart failure with preserved ejection fraction found that a predictive model (PREDICT-HFpEF) was quite accurate in predicting actual events that occurred (see chf PREDICT finerenone study JAMAcardiol2025, or doi:10.1001/jamacardio.2025.0025)

 

Details:

-- 6001 patients were enrolled in the FINEARTS study that found that finerenone, a nonsteroidal mineralocorticoid receptor antagonist that is more selective than spironolactone, had the same benefit as spironolactone in patients with heart failure with preserved ejection fraction (HFpEF), reducing the rate of unplanned hospital admissions or urgent visits for heart failure by 16% over placebo, independent of the use of SGLT-2 inhibitors: https://www.nejm.org/doi/full/10.1056/NEJMoa2407107

--the FINEARTS study was performed across 653 sites in 37 countries in adults who were at least 40 years old and had symptomatic heart failure; left ventricular ejection fractions of at least 40% (ie, patients had either mildly reduced ejection fraction (EF 40-49%) or preserved ejection fraction (at least 50%); evidence of structural heart disease (left ventricular hypertrophy or left atrial enlargement); and an elevated N-terminal BNP at least 300 pg/mL or BNP at least 100 pg/mL in those with normal sinus rhythm or NT-proBNP at least 900 pg/mL or BNP at least 300 pg/mL in those with atrial fibrillation

    -- ambulatory and hospitalized patients were included in the study, with exclusion criteria of those with an eGFR <25 by creatinine or serum potassium level >5.0 mmol/L

-- the intent of the study was to evaluate the use of the PREDICT-HFpEF risk predictions as compared to the actual observed outcomes in the FINEARTS study

 

-- The PREDICT-HFpEF model incorporated the following:  NTproBNP per log unit increase, heart failure hospitalizations in past 6 months, creatinine per 0.0565 mg/dL increments, history of diabetes, region of the world, heart failure duration >1year, not receiving SGLT-2 inhibitors, history of COPD, history of TIA/stroke, heart rate per 10 bpm increase, BMI<25, ejection fraction per 5% less than 55%, age per 5 years increase above 70 years, male sex, and NYHA functional class III/IV

 

-- patients had a mean age of 72 years, 55% were male

-- overall risk factors varied dramatically depending on which quintile of the PREDICT-HFpEF the patients were in (these quintiles that stratified the risk were for the outcomes of cardiovascular death or HF hospitalization; cardiovascular death; and all-cause death):

    -- for those in the lowest vs highest quintile, there were significantly more white patients, higher systolic pressure (132 versus 127 mmHg), lower pulse (67 versus 76), higher left ventricular ejection fraction (EF) 54% versus 51%, lower NT-proBNP 299 versus 3000, lower creatinine 1.0 versus 1.3, higher eGFR 73 versus 52, lower urinary albumin-to-creatinine ratio 64 versus 353, fewer heart failure hospitalizations in the past six months 136 versus 1073, fewer prior heart failure hospitalizations 289 versus 1153, more functional NYHA class II in 949 versus 656/fewer in lower functional NYHA class III in 249 versus 523, higher baseline Kansas City Cardiomyopathy Questionnaire (KCCQ) total score 71 versus 60, less diabetes 370 versus 640, less atrial fibrillation 71 versus 706, but more myocardial infarction 424 versus 276, and less stroke 50 versus 257

        -- overall, those at higher risk were older, more often male and had lower EF values, BMI and lower eGFR, worse NYHA heart failure functional class, and more often and more recently had HF hospitalization, more diabetes, COPD and cerebrovascular disease, took more loop diuretics, had lower blood pressure, and more atrial fibrillation

-- medications were different: quintile 1 versus quintile 5: beta blockers 987 versus 1029, ACE inhibitors 489 versus 398, ARBs 486 versus 349, sacubitril/valsartan 123 versus 183, loop diuretic 888 versus 1144, calcium blockers 462 versus 354

 

-- 3 primary outcomes: composite of cardiovascular death and total heart failure events (hospitalizations and urgent HF visits); composite of time to first occurrence of cardiovascular death or worsening HF (a prespecified sensitivity analysis); and all-cause death. PREDICT-HFpEF model performance was assessed using the Harrell C statistic, where a value of 1.0 reflected perfect correlation, 0.8-1.0 a very strong positive correlation, and 0.6-0.79 a strong positive correlation

-- secondary analysis outcomes:  total heart failure events; improvement of NYHA class from baseline to 12 months; change in the Kansas City Cardiomyopathy Questionnaire (KCCQ) total symptom score from baseline to 6, 9, and 12 months; composite kidney endpoint; and cardiovascular death and all-cause mortality were also assessed (some of these were not analyzed in this study, more to come)

-- prespecified safety analyses: hyperkalemia, hypokalemia, hypotension, and elevations in the serum creatinine level

 

Results:

-- composite of cardiovascular death or HF hospitalization at 2 years: 5.1 (4.6-5.7), which corresponded to a median (IQR) risk of 13.0% (7.7%-22.2%) of patients experiencing this outcome

-- cardiovascular death: 4.7 (4.1-5.3), with 2-year median risk of 6.5% (3.7%-11.3%) of patients experiencing this outcome

-- all-cause death: 4.4 (3.8-4.9), with 2-year median risk of 9.6% (5.7%-15.6%) of patients experiencing this outcome

 

-- outcomes related to PREDICT-HFpEF score:

    -- composite of cardiovascular death or first HF hospitalization:

        -- quintile 1: 2.6 (2.1-3.2) per 100 person-years

        -- quintile 5: 23.1 (21.1- 25.4) per 100 person-years

            -- Hazard Ratio HR 8.3 (6.5-10.5)

    -- a composite of cardiovascular death and total (first and recurrent) HF events:

        -- quintile 1: 3.7 (2.9-4.8) per 100 person-years

        -- quintile 5: 39.8 (37.4- 42.4) per 100 person-years

            -- HR 9.8 (7.5- 12.7)

    -- total worsening HF events (a secondary outcome):

        -- quintile 1: 2.6 (1.9-3.5) per 100 person-years

        -- quintile 5: 32.5 (29.1- 36.3) per 100 person-years

            -- relative risk 11.0 (7.9- 15.3)

 

-- PREDICT-HFpEF model discrimination:

    -- composite of cardiovascular death or HF hospitalization: 

        -- 599 of 3003 patients (19.9%) in finerenone group

        -- 664 of 2998 patients (22.1%) in the placebo group

            -- C statistic: 0.73 (0.71-0.75) at year 1

            -- C statistic: 0.71 (0.69-0.72) at year 2

                -- of note, these C statistics were in the ” strong positive correlation” area and performed better than the NT-proBNP alone for both years, with p<0.001

    -- cardiovascular death: 

        -- 242 (8.1%) in the finerenone group

        -- 260 (8.7%) in the placebo group

            -- C statistic: 0.72 (0.68-0.75) at year one

            -- C statistic: 0.68 (0.66-0.71) at year 2

                -- also in the strong positive correlation area

    -- all cause death:

        -- 491 (16.4%) in the finerenone group

        -- 522 (17.4%) in the placebo group

            -- C statistic: 0.71 (0.69-0.74) at year one

            -- C statistic: 0.69 (0.67-0.71) at year 2

                -- also in the strong positive correlation area

 

-- Absolute risk reduction/number-needed-to-treat:

    -- the overall proportional risk reduction with finerenone to the composite of time to first cardiovascular death or worsening HF events was 16%

    --the number needed to treat to prevent patients from experiencing this outcome per 100 person-years of treatment:

        -- quintile 1: 200

        -- quintile 2: 111

        -- quintile 3: 67

        -- quintile 4: 42

        -- quintile 5: 21

 

-- safety outcomes:

   -- they found that the rates of adverse outcomes increased across risk quintiles, both for the placebo group alone and the placebo-finerenone differences

 

Commentary:

--the PREDICT-HFpEF model can be found at https://predict-hfpef.com/; it basically includes the above-mentioned items mostly as binary inputs, stating that it has been validated for those with ejection fractions ranging from 35% to 88%. 

    -- a prior study by the same lead author assessing dapagliflozin in the DELIVER study (6263 patients enrolled in the trial with validation by 4796 patients in two other trials).  They were able to develop a robust prediction model with 16 independent predictors of clinical outcomes for patients with HFpEF. This current finerenone study reaffirmed that this model was effective, though they did not include the input of whether patients had a duration of heart failure longer than one year.

 

--this PREDICT-HFpEF model did quite well in classifying patients into quintiles of risk, finding that there was an 8 to 10-fold higher risk in those in the highest risk quintile (5th) vs the lowest (1st). this study added to the prior 2 HFpEF trials that had validated the PREDICT-HFpEF risk calculator

-- the actual cardiovascular event rate in this study was much higher than others (eg DELIVER or PARAGON-HF studies), with the 5th quintile being much sicker and having more adverse outcomes, so its results may be more generalizable to the broader population of patients with heart failure

-- the results of the PREDICT-HFpEF in this finerenone study were also found to extend at least 2 years into the future, with highly significant C statistics.

-- as a background here, this model was much better than the ability of clinicians to predict outcomes, or the reliance on NT-proBNP levels or NYHA heart failure functional assessments (this study found, for example, that even with the milder class 2 NYHA group, there was a very large variation in predicted and actual heart failure outcomes)

 

-- this model is quite different from the newly developed PREVENT risk calculator, a risk calculator that was to include both patients with cardiovascular disease as well as heart failure. The new features of the PREVENT risk calculator over the prior cardiovascular risk predictors (eg ASCVD risk calculators, which also did not have a heart failure function) is that it optionally included other predictors, some clinical (urinary albumin-to-creatinine ratio, hemoglobin A1c) as well as several that are social (assessment of percent living in poverty in the area, education level, percent single-parent households, percent living in rented housing, percent living in overcrowded housing, percent having a car, percent being <65yo and unemployed, and the ZIP Code as a surrogates for "social deprivation" (https://gmodestmedblogs.blogspot.com/2024/07/prevent-new-cardiac-risk-factor.html)

-- not surprisingly, social issues such as stress also can affect heart failure itself, both directly and indirectly, with stress reduction interventions improving outcomes: https://pmc.ncbi.nlm.nih.gov/articles/PMC8026548/ . But these “non-medical” factors are not incorporated into the PREDICT-HFpEF used above

 

Limitations:

-- though there was a quite large (8- to 10-fold) difference in both predicted and actual risk (per the finerenone study) from quintiles 1 vs 5, there were much smaller differences between the less extreme quintile differences (eg comparing quintile 1 to quintile 2, or quintile 4 vs 5), and these more subtle differences have much less significantly different predictions (some clearly not significant, especially for cardiovascular death per their graphs, though quintile 5 is pretty dramatically worse for all outcomes as compared to the other 4 quintiles)

-- as noted above, the PREDICT-HFpEF risk model does not include psychosocial or demographic risk factors that certainly reflect very important issues in disease development as well as access to optimal clinical care that affect disease outcomes. The PREVENT calculator for cardiovascular disease and heart failure is a much more inclusive and holistic model of where we should be heading in disease risk calculators. And these more complete calculators would reveal the very important adverse psychosocial/demographic situations that need to be addressed to help the overall population

-- the cohorts in this and prior studies used to validate the results of the PREDICT-HFpEF risk calculator were all in clinical trials. There is a selection bias here in terms of how patients were recruited (eg by trial recruiters working in cardiology departments, or over the internet…), what patients decided to participate (those going to tertiary centers for their care, those who had more medical literacy, those from more urban communities….), etc, all of which might limit generalizability of the results to other groups

-- there were intrinsic exclusions to a finerenone study in that those with more renal failure or hyperkalemia were excluded. This would limit the generalizability of the results to those with these problems, many who would not be in this study but still may come to clinic with heart failure….

-- also the use of finerenone in the trial also distorts the potential generalizability of the results because of its renal and electrolyte effects

-- they did not assess one variable (whether patients had a duration of heart failure longer than one year) from the model, which might have altered the results somewhat

 

so,

-- this PREDICT-HFpEF risk model does seem to have quite a good prediction assessment for those with heart failure who have a left ventricular ejection fraction >40% and are symptomatic, vs just the subjective NYHA (New York Heart Association) functional classification that focuses only physical limitations and does not include many known medical risk factors

    -- this is an important step forward, since our clinical assessments or our relying on single factors (NT-proBNP results) is much poorer as a predictor of heart failure control and subsequent in-patient or out-patient visits than this risk calculator.

-- the advantage of a good calculator is that we can use the results to alter therapy to prevent these heart failure exacerbations, by  making sure that the patients are on optimal guideline-directed medical care

 

geoff

-----------------------------------

If you would like to be on the regular email list for upcoming blogs, please contact me at gmodest@bidmc.harvard.edu

to get access to all of the blogs:  go to http://gmodestmedblogs.blogspot.com/ to see the blogs in reverse chronological order

or you can just click on the magnifying glass on top right, then type in a name in the search box and get all the blogs with that name in them

 

    -- another similar approach but without the completeness of this predictor was a study finding that tracking and attending to increased BNP measurements in patients with chronic kidney disease but no hospitalizations for heart failure led to improvements in both heart failure and kidney disease outcomes:  https://gmodestmedblogs.blogspot.com/2023/12/routine-bnp-assessment-helpful-for.html