Author + information
- Marco Canepa, MD, PhD,
- Luigi Tavazzi, MD, PhD,
- Aldo P. Maggioni, MD∗ (, )
- on behalf of the ESC HF Long Term Registry Investigators
- ↵∗ANMCO Research Center, Via La Marmora, 34, 50121 Firenze, Italy
We thank Dr. Bayés-Genís and Lupón for their interest in our recent work (1), in which we showed the limited performance of 4 major available prognostic risk scores (CHARM [Candesartan in Heart Failure–Assessment of Mortality and Morbidity], GISSI-HF [Grupo Italiano per lo Studio della Streptochinasi nell’infarto Miocardico-Heart Failure], MAGGIC [Meta-analysis Global Group in Chronic Heart Failure], and SHFM [Seattle Heart Failure Model]) and their scarce use in daily clinical practice of European cardiologists. The important limitations of currently available prognostic risk models have been examined in detail in our discussion (1) and in the accompanying editorial by Simpson and McMurray (2).
As a potential improvement over these tools, Drs. Bayés-Genís and Lupón put forward some additive values of the Barcelona Bio-Heart Failure Risk Calculator, a prognostic score designed by the same investigators to estimate risk of death or heart failure (HF) hospitalization for up to 5 years. This tool was originally developed from a single-center cohort of 864 chronic HF patients (3), and an updated version was recently proposed in a research letter (4). This score has the peculiarity of incorporating, beyond classical risk factors, 3 HF biomarkers (i.e., high-sensitivity cardiac troponin T, ST2, and N-terminal pro–B-type natriuretic peptide), and it is designed to run with the availability of 0, 1, 2, or 3 of them. We concur with Bayés-Genís and Lupón on the need to include, when possible, pathophysiological biomarkers in prognostic scores to get them to be more clinically pregnant and useful in practice. However, despite some enthusiastic performance reported in the original work with an area under the curve (AUC) >0.80 (3), a later application of this score to a subset of 1,934 participants from the PARADIGM-HF (Prospective Comparison of ARNI [Angiotensin Receptor–Neprilysin Inhibitor] with ACEI [Angiotensin-Converting–Enzyme Inhibitor] to Determine Impact on Global Mortality and Morbidity in Heart Failure) trial, for whom the 3 biomarkers were available, showed a maximal AUC of just 0.70 for risk of both death and HF hospitalization at 2 years (4). This performance is comparable to those described in our paper and elsewhere (1).
What is the value of adding variables, even relevant ones such as cardiac biomarkers, if the performance of the score remains the same? Adding variables to logistic models may increase AUC because of overfitting rather than increased accuracy, and this risk should always be considered. In addition, in the most updated version of the Barcelona Bio-HF Calculator, the investigators included beta values for ARNI treatment that “were derived from the benefit observed in the PARADIGM-HF trial.” This same approach was used by the SHFM investigators for device therapy and beta-blocker medications and was not confirmed in our study (1). Indeed, the effects sizes seen in large randomized clinical trials are notoriously greater than those observed in real-world populations; therefore, they should be cautiously translated to patients because of the important risk of prognostic underestimation, as we reported for the SHFM (1).
As the accompanying editorial notes in its title (2), we agree it is time for a reboot in prognostic modeling of HF. We think this reboot should begin with designing different models for specific subsets of HF patients (e.g., HF with reduced ejection fraction vs. HF with preserved ejection fraction, ischemic vs. nonischemic, and low vs. high comorbidity burden) and outcomes (in particular distinguishing sudden cardiac death from death due to worsening HF and/or pump failure). To this goal, proper phenotyping of patients and ascertainment and adjudication of outcomes in future registries will become fundamental. These new scores will not only need to be validated in external populations, but will also require impact analysis to determine whether their use is better than the usual care (1).
Please note: The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Canepa M.,
- Fonseca C.,
- Chioncel O.,
- et al.
- Simpson J.,
- McMurray J.J.V.
- Lupon J.,
- Simpson J.,
- McMurray J.J.V.,
- et al.