https://orcid.org/0000-0002-1793-5297
This editorial refers to ‘Novel biomarker-driven prognostic models to predict morbidity and mortality in chronic heart failure: the EMPEROR-Reduced trial’, by S.J. Pocock et al., https://doi.org/10.1093/eurheartj/ehab579.
Graphical presentation of the eight key predictors of the primary composing endpoint in EMPEROR-Reduced. HF, heart failure; NT-proBNP, N-terminal probrain natriuretic peptide; hs-cTnT, high-sensitivity cardiac troponin T.
Do you routinely use a risk score to quantify the risk of death and the risk of heart failure (HF) readmission in your patients? Probably, most of us have not done so until now. Some of the arguments that we had been using for years to excuse our medical inertia are no longer valid given the new evidence reported in the current issue of the European Heart Journal by Pocock et al.1
HF is a heterogeneous syndrome with an ongoing rise in prevalence driven at least partly by the ageing society.2 Equally, HF is a growing social and economic burden in terms of high mortality and morbidity, and excessive healthcare costs.3 However, the prognosis of HF has become variable, depending on HF aetiology, healthcare determinants, and disease-modifying therapies.4 In order to better allocate healthcare resources and to overcome uncertainties about the intensity of management, several prognostic models have been developed.1 These should assist physicians in order to implement more standardized and individualized decision-making and to improve the current mostly subjective and empirical process.5
Stuart Pocock and colleagues present a biomarker-driven prognostic tool for patients with chronic HF and reduced left ventricular ejection fraction (HFrEF).1 The authors used data from the Empagliflozin Outcome Trial in Patients with Chronic Heart Failure and a Reduced Ejection Fraction (EMPEROR-Reduced) to develop multivariable Cox regression models for the primary composite outcome of (i) HF hospitalization or cardiovascular death, (ii) all-cause death, and (iii) cardiovascular mortality.
As most would agree that it is not wise to challenge Professor Pocock on statistics, we will keep that section very short. Eight out of 33 pre-selected variables were key predictors of the primary composite endpoint (Graphical Abstract). As well as clinical parameters such as systolic blood pressure, heart rate, NYHA class III or IV, and the presence of peripheral oedema, parameters describing a patient’s disease course such as the most recent HF hospitalization and a long-standing history of HF were also included. However, N-terminal probrain natriuretic peptide (NT-proBNP) and high-sensitivity cardiac troponin T (hs-cTnT) as biomarkers representing two important pathophysiological pathways (haemodynamic cardiac stress and cardiomyocyte injury) in HF were the dominant predictors of the primary endpoint.6,7 The overall primary outcome risk score including eight significant key predictors and randomization to empagliflozin discriminated well (c-statistic 0.729), with a steep gradient in the composite endpoint between risk deciles. Patients in the highest decile of risk had an event rate nine times higher when compared with the lowest decile. Regarding the model’s calibration, there was a good agreement between the observed and predicted patient risk in each decile. NT-proBNP and hs-cTnT were also the dominant predictors regarding all-cause or cardiovascular mortality, followed by NYHA class III or IV as well as an ischaemic aetiology of HF. With both achieving a c-statistic of 0.69, the overall models for cardiovascular and all-cause mortality showed a good discrimination as well. An external validation in the BIOSTAT-CHF cohort showed similar c-statistics.
Interestingly, the authors examined the addition of further variables in addition to eight key predictors for the composite endpoint and found no meaningful gain in c-statistic (0.729 vs. 0.733) when compared with the prior model. Given the vast number of emerging HF risk scores, this highlights the first out of six criteria that may help to put data on HF risk scores into clinical perspective and assess their potential to ultimately improve patient management8–10 (Table 1). Ideally a risk score should demonstrate the highest possible goodness of fit while exhibiting the smallest possible complexity to guarantee ease of use (1). This aspect leads to requirements for predictor variables which should be fast and routinely available and should assess disease severity as objectively as possible rather than depending on the subjective judgement of a physician to avoid interobserver variation. With regards to the EMPEROR-Reduced risk score, the included variables mostly meet these requirements with the exception of NYHA class which is less objective (2). In addition, reliable clinical models should be developed using a large and high-quality dataset possibly even obtained from a multicentre study such as in EMPEROR-Reduced (3). Furthermore, as performed by Pocock and colleagues, a newly developed model should be externally validated before introduction into clinical practice can be considered (4). Ideally, a model should be developed in the setting in which it will be clinically applied. Regrading chronic HF, using data from hospitalized or ambulatory patients would be preferable, whereas in acute HF data obtained in the emergency department would be ideal to avoid selection bias11 (5). Direct comparison of a new score with an established score in the same dataset is recommended to document prognostic equivalence or even superiority. This was performed by application of the PREDICT-HF risk score to the EMPEROR-Reduced data set, which showed superiority of the EMPEROR-Reduced score (6).
Criteria for risk scores and biomarkers that may help to put data on heart failure risk scores into clinical perspective and assess their potential to ultimately improve patient management
| Criteria for risk scores | Criteria for biomarkers |
|---|---|
| 1. High goodness of fit and low complexity. | 1. Assignment to a pathophysiological pathway in heart failure. |
| 2. Variables: fast and routinely available, objectively collectable. | 2. Pre-analytical and analytical aspects. |
| 3. Developed in a large and high-quality dataset. | 3. High accuracy for prognosis. |
| 4. External validation in a different cohort. | 4. Incremental value compared with available biomarkers. |
| 5. Development in the setting in which application is planned. | |
| 6. Direct comparison with an established score. |
| Criteria for risk scores | Criteria for biomarkers |
|---|---|
| 1. High goodness of fit and low complexity. | 1. Assignment to a pathophysiological pathway in heart failure. |
| 2. Variables: fast and routinely available, objectively collectable. | 2. Pre-analytical and analytical aspects. |
| 3. Developed in a large and high-quality dataset. | 3. High accuracy for prognosis. |
| 4. External validation in a different cohort. | 4. Incremental value compared with available biomarkers. |
| 5. Development in the setting in which application is planned. | |
| 6. Direct comparison with an established score. |
Criteria for risk scores and biomarkers that may help to put data on heart failure risk scores into clinical perspective and assess their potential to ultimately improve patient management
| Criteria for risk scores | Criteria for biomarkers |
|---|---|
| 1. High goodness of fit and low complexity. | 1. Assignment to a pathophysiological pathway in heart failure. |
| 2. Variables: fast and routinely available, objectively collectable. | 2. Pre-analytical and analytical aspects. |
| 3. Developed in a large and high-quality dataset. | 3. High accuracy for prognosis. |
| 4. External validation in a different cohort. | 4. Incremental value compared with available biomarkers. |
| 5. Development in the setting in which application is planned. | |
| 6. Direct comparison with an established score. |
| Criteria for risk scores | Criteria for biomarkers |
|---|---|
| 1. High goodness of fit and low complexity. | 1. Assignment to a pathophysiological pathway in heart failure. |
| 2. Variables: fast and routinely available, objectively collectable. | 2. Pre-analytical and analytical aspects. |
| 3. Developed in a large and high-quality dataset. | 3. High accuracy for prognosis. |
| 4. External validation in a different cohort. | 4. Incremental value compared with available biomarkers. |
| 5. Development in the setting in which application is planned. | |
| 6. Direct comparison with an established score. |
Pocock and colleagues also emphasize the central role of NT-proBNP and hs-cTnT in the risk prediction of HFrEF. Given the constantly rising number of cardiovascular biomarkers reported to be associated with mortality and rehospitalization in HF, four criteria may help to identify those most useful for a more individualized patient management (Table 1). First, and of crucial importance, a prognostic biomarker should be assigned to one of the pathophysiological pathways involved in the development and progression of HF6 (1). In addition, the following analytical and pre-analytical aspects should be met: sample preparation should be easily implementable into routine clinical practice, and reasonable analytical precision needs to be demonstrated to ensure reproducibility of results. Turn-around time and the cost of measurement should be reasonable (2). Furthermore, a promising biomarker should provide high accuracy for prognosis. Some biomarkers as single variables only have modest accuracy, but may provide some additional value in multivariable models.12 As these models only rarely reflect integrated clinical judgement, physicians should remain critical whenever the new biomarker itself lacks high accuracy (3). In addition, an incremental value is necessary to justify a possible application of a novel biomarker in routine clinical practice. A novel biomarker should add new and clinically useful information rather than recapitulating information already available at the bedside or via the use of inexpensive routinely available biomarkers13 (4).
The EMPEROR-Reduced prognostic model represents an easily implementable prognostic assessment of patients with HFrEF by using the combination of NT-proBNP and hs-cTnT with additional fast and routinely available variables. This is in full agreement with the fact that biomarkers have improved the understanding of HF pathophysiology and can therefore contribute to adjust and individualize patients’ management. However, despite the important role of hs-cTnT and NT-proBNP in HF as a heterogeneous syndrome with various phenotypes, a biomarker approach alone seems rather insufficient for an accurate risk prediction.14 Therefore, the combination with easily available clinical parameters such as vital signs is even more promising to assist clinicians in standardized decision-making. Novel biomarker-driven prognostic models such as the EMPEROR-Reduced model could help to customize the intensity of management and to better allocate healthcare resources in HF. Therefore, they represent a possible next step to a more individualized medicine for HF patients. Moreover, communicating the quantified risk to the patient may help to maintain/increase adherence with treatment.
What could be the next steps? Obviously, the model exclusively applies to patients with our favourite HF phenotype—HFrEF. Given the recent completion of EMPEROR-Preserved, Professor Pocock and colleagues hopefully very soon will be able to examine whether this model or a slightly modified version also works well in patients with other HF phenotypes. The authors appropriately mention that variables not captured or under-represented in their study, including frailty, social deprivation, and dementia, also have important prognostic impact and will need to be addressed by specific interventions.
As with most other innovations in clinical medicine, successful implementation of the EMPEROR-Reduced score also depends on several factors beyond science. Therefore, it is highly appreciated that an Online Risk Calculator is provided in the Online Supplement that can be incorporated into the electronic medical record of your patients.1 Let us start gathering our clinical experience with it!
Conflict of interest: C.M. has received research grants from the Swiss National Science Foundation, the Swiss Heart Foundation, the University of Basel, the University Hospital Basel, Abbott, Beckman Coulter, BRAHMS, Idorsia, Novartis, Quidel, Roche, Siemens, Singulex, and Sphingotec, as well as speaker/consulting honoraria from Amgen, Astra Zeneca, Bayer, Boehringer Ingelheim, Daiichi Sankyo, Idorsia, Novartis, Osler, Roche, Sanofi, Siemens, and Singulex, all paid to the institution. D.W. has no conflicts to declare.
The opinions expressed in this article are not necessarily those of the Editors of the European Heart Journal or of the European Society of Cardiology.
