Validating the HFA-PEFF Score – Or how to Define a Disease?

This article refers to ‘Validation of the HFA-PEFF score for the diagnosis of heart failure with preserved ejection fraction’ by A. Barandiarán Aizpurua et al., published in this issue on pages 413–421. If we conduct empirical treatment trials in the situation where in fact there is no unifying disease, but rather a profusion of different entities presenting similarly, then the trial is more likely to fail, as some only some of the sub-types may respond to the particular therapy under study. This may be what we have seen in heart failure with preserved ejection fraction (HFpEF) trials to date, and it is a wake-up call to be more precise in how we select HFpEF patients for recruitment into treatment trials. Being precise does not necessarily mean being more comprehensively evaluated, however. It may in contrast mean restricting our definition of HFpEF to a narrower group of patients, such as those with recent documented pulmonary oedema who have been shown to have an ongoing clinical requirement for loop diuretics and elevated natriuretic peptide (NP) levels. Before we discuss the papers in this issue of the Journal concerning two recently proposed scores for HFpEF, it may be useful to consider different types of diseases and how we define them. For some illnesses we know the precise pathogenetic sequence, for example infections or cancer. These we can consider simplistically as ‘categorical diseases’ in that you either have them or you do not. You may need to apply the most accurate diagnostic test to detect the aetiological agent or feature, such as histological analysis of biopsy material or DNA sequencing to detect infective agents, but ultimately disease presence is a yes/no answer. Other diseases are defined by a change in relative function, and as such we may call them ‘numerical diseases’. Most degenerative diseases of old age are numerical disorders – osteoporosis, liver disease, heart failure (HF), hypertension, hyperlipidaemia, type 2 diabetes, renal failure, chronic obstructive pulmonary disease, etc. They are not either present or absent as for the categorical illnesses, they are graded and may worsen or improve, or progress or regress with time. The aetiology of this second type of disease can be varied or multiple, and the illness state is defined by a relative loss of organ function, either defined by symptoms, or prognostic effect, or response to therapy. Clearly there are gradations all the way from normality, through mild and moderate dysfunction to severe disease, often with no clearly identifiable thresholds in between. Debates may run for years or decades about disease, such as what is normal blood pressure in an older person, or even the famous half century old Pickering–Platt debate about whether hypertension is a disease or instead the upper range of a physiological variable that carries clinical consequences.1 One notable difference between categorical diseases and numerical diseases is that for the former the aetiology is crucial and treatments often target its eradication, whereas for the latter the aetiology may be of lesser importance, what may matter more is the body's response and secondary damage, such as arterial occlusion in hyperlipaemia and hypertension or neurohormonal overactivity in HF. It is often the extent or consequences of organ dysfunction that matters most. For example we treat hypertension, HF, chronic obstructive pulmonary disease, renal failure, and liver failure similarly irrespective of the original aetiology, whereas we would never contemplate doing this for infections or most cancers. When we consider HF, most historical definitions recognised this as numerical disease, presenting with typical symptoms (fatigue and/or dyspnoea) associated with a documented reduction in the pumping capacity of the heart. These definitions have undergone multiple refinements over the years, but by and large we have not required any specific diagnostic test or aetiology to diagnose HF. One significant development was when clinical trials, after the first few, began to enrich for patients with a reduced left ventricular ejection fraction (LVEF), not initially because it was felt they were of different inherent nature, but because it was recognised they were at increased risk of mortality and hence clinical trial power would be enhanced by having more of these patients recruited. We later recognised that we needed to differentiate HF with reduced ejection fraction (HFrEF) from all HF, because most of our treatments had been proven, because of this trial strategy, only for HFrEF and not for all HF.2 Reading the most recent guidelines for HF we see that the definition of HF has undergone very little change from that of the 1970s, apart from us specifically calling out a subset of HF with this pathognomic reduction in ejection fraction as HFrEF, leaving aside whether we now suggest the threshold should be 40%, 50%, or even 55%. Yes, the guidelines say that measurement of exercise capacity, haemodynamic evaluation and NPs can help in selected patients, but ultimately the diagnosis of HFrEF is clinical, consisting of typical signs and symptoms associated with a reduction in the systolic function of the left ventricle. We do not even need to determine if the cause is myocardial infarction, chronic ischaemia without infarction, or a specific cardiomyopathy. By and large we consider them all together, because by and large, our trials have shown similar efficacy in all aetiologies of HFrEF studied. Why is HFpEF any different? If HFpEF had responded to the same therapies, and in a similar way as HFrEF, we would have the same simple diagnostic threshold and definition and we would not separate them in routine practice. In the absence of a low ejection fraction, however, and recognising the trials in HFpEF have not been as positive as those in HFrEF, we have recognised we have a problem. Some of the patients labelled with HFpEF may not have had any real heart disorder at all, they may merely have mild cardiac dysfunction and their symptomatic limitation or their frequent hospital admissions may be related to non-cardiac comorbidities, such as lung disease or arrhythmias, and that this may be the reason they have not responded to effective HFrEF therapies. Sadly, in many of our trials the financial incentive to recruit patients has meant a small percentage of centres are found to have recruited fake or falsely documented patients. Possibly a significantly larger number may also have flexibly interpreted entry criteria to allow patients to be recruited to the trial when they should not have been. This is particularly true when an entry criterion is open to interpretation, such as New York Heart Association class or symptom severity, which cannot be easily independently verified. This means that some of the HFpEF trials in the past may have missed their primary endpoint because of the recruitment of non-real HFpEF patients, rather than because true HFpEF does not respond to therapy. It is against this background that we consider two important recently proposed diagnostic scores for HFpEF, the H2F-PEF3 and HFA-PEFF4 scores. Both have undergone independent validity testing, for H2F-PEF5 and for HFA-PEFF in another a paper in this issue.6 Together in this issue are also two insightful editorial comments concerning these papers.7, 8 Both praise the attempts to define a scoring system, so there can follow some standardisation of the definition of HFpEF. One common feature of all these articles, however, is the acceptance of the gold standard being expert opinion, guided ultimately in difficult cases, by estimation of invasively measured haemodynamic measurements demonstrating raised intracardiac increased filling pressures, either at rest or on exercise. It is accepted by all these papers that this feature is the core abnormality of HFpEF. This may be true, but we cannot as yet say we have proven this. It would depend on us knowing there is a disease related to increased intra-cardiac filling pressures and that we are seeking practical solutions to detecting this clinically that can realistically and usefully be applied both in clinical practice and in clinical trials to assess therapies. We should perhaps take a step back and ask if this is indeed the most crucial feature of HF or even of HFrEF. If we look at the example of HFrEF, we accept that raised intracardiac increased filling pressures, either at rest or on exercise, are common, but not universal in diuretic-treated patients, and it has never been demonstrated that this is indeed the crucial feature that explains any of the premier triad of symptom generation, conferral of risk or prediction of responses to therapy. A much stronger case can be made for LVEF reduction, neurohormonal activation or NP elevation predicting the last two and secondary changes in skeletal muscles and vascular function explaining the first.9 So why do we promote elevated filling pressures to such prominence; is it because we need something objective? Maybe we should focus instead on some other aspects of the disease entity. Both published HFpEF scores are designed as numerical scoring systems to estimate the likelihood of an ultimate HFpEF diagnosis, and as has been pointed out by both commentaries, the validation studies of both are affected by an expectation bias, if the ‘gold standard’ is an expert panel decision, then any score that includes features which the expert panel trusts as indicating HFpEF will predict the diagnosis more accurately. But, this does not mean it is indeed an accurate diagnosis. What we need is a way to define, detect and recruit patients with HFpEF in a way that is far simpler than the information the expert panels would like to make their diagnosis. It needs to be boiled down to very few concepts or measurements to be clinically useful. In this regard both scores are disappointing. The H2F-PEF score is a weighted score based on six variables (obesity, atrial fibrillation, age > 60 years, treatment with ≥2 antihypertensives, echocardiographic E/e' ratio >9, and echocardiographic pulmonary artery systolic pressure >35 mmHg) and thus cannot be considered in any way a diagnostic test, for it is in reality 60% a risk marker score (age, hypertension, atrial fibrillation) with the addition of 40% weight given to two isolated echocardiographic markers (diastolic function and pulmonary artery pressure), neither of which are fundamental to the HF syndrome. Critical elements of any HF diagnosis (symptoms, congestion, exercise intolerance, or neurohormonal activation) are entirely absent. The HFA-PEFF score is also too complex to be used as a way to diagnose HFpEF for trials or epidemiological purposes, for it assumes patients will require pre-testing evaluation, initial evaluation, specialised testing and then possibly final aetiological evaluation. The total cost of a full evaluation to make the diagnosis may be multiple 1000 euros, if several of the required steps are needed. Whist this may indeed be best practice to fully evaluate an individual patient, it is impractical as a diagnostic definition for a disease, or to be used an as entry criterion for a clinical trial in HFpEF. For this we need a much simpler system. Taking the examples of successful criteria for other ‘numerical diseases’, the definition of both hypertension and diabetes has changed over time as they move the definitional threshold lower and ever closer to the normal range, as treatments have started to show benefit from therapy at this new diagnostic cut-off. For osteoporosis the threshold of a bone mineral density ≥2.5 standard deviations below that of healthy young adults is accepted and only requires one test. For hypertension several blood pressure measurements is all that is needed. For diabetes only fasting sugars or glycosylated haemoglobin values are required. For renal failure, now reframed as severe chronic kidney disease, only a measurement of glomerular filtration rate is needed. Why are we making the diagnosis of HFpEF so enormously complex? Is it truly so complex a disorder, or is in fact not one disorder, but several and that is the reason we are becoming so comprehensively enmeshed in expensive tests. In an attempt to unravel this, lets break down what we mean by HFpEF. We find HFpEF in two circumstances, as a chronic disease burden in an outpatient or primary care setting, or as an acute presentation leading to hospital admission. Our estimates of the epidemiological prevalence of HFpEF come from the same two main routes of presentation. In both settings HFpEF represents about half of all HF cases.10, 11 We may think the real disease burden of HFpEF is stable HF in the general population in the presence of preserved LVEF, but actually the diagnosis is rarely made in this setting, more often the diagnosis is only finally established by an acute HF admission. The true burden of the HFpEF population is largely unknown, and in our present state of knowledge unknowable, because the evaluation of exertional symptoms cannot be completed in a large number of subjects in the primary care setting and for many elderly patients, health care resources do not allow full evaluation of such patients in a specialist centre. In the acute setting, however, an HF diagnosis can made more easily, satisfying some simple clinical signs and symptom criteria, supplemented by a response to therapy (most frequently diuretics). Here the diagnosis is more robust as more diagnostic tests have become available and all patients have crossed a threshold of requiring admission and cardiac therapy. It is thus possible to focus on the substantial subset of HF patients who present with a documented episode of acute decompensated HF and subsequently recruit them into adequately powered treatment trials, either during the admission or shortly thereafter. Even so there will be errors of mis-diagnosis of the aetiology of the syndrome between being cardiac or non-cardiac, but despite these reservations we can establish a decent disease burden of suspected HFpEF patients who have at some stage had an admission ascribed to acute HF. If we restricted this type of HFpEF to people who have presented in the recent past with documented pulmonary congestion, perhaps supplemented or confirmed by elevated NPs and certain echocardiographic abnormalities such as left ventricular hypertrophy or left atrial enlargement, then we would have a high-risk population to study. This would be much simpler and arguably would lead to more likely positive trials because the patients are enriched by having a prior admission with decompensated HF and by having the other risk features mentioned above. So, my plea would be not for more complex definitions of HFpEF that attempt to cover all patients, but rather a simpler, more easy to document definition of a disease: ‘recent documented acute congestive HF associated with a preserved ejection fraction’. I would recruit patients between 3 days and 90 days post-stabilisation and enrich for elevated NPs and possibly other simple factors previous trials suggest increase event rates, such as left ventricular hypertrophy or left atrial enlargement. The previous HFpEF trials (I-Preserve, PEP-CHF, CHARM-Preserved, TOPCAT and PARAGON-HF) which have failed, all recruited at least a proportion of patients without any previous hospitalisation for HF, let alone a recent one. I suspect, but it can never be proven that if they have recruited only such patients, they may have been positive, especially as many came very close to reaching their primary endpoint. We could accelerate our study of potential HFpEF therapies, not by rejecting all those treatments that have failed in our HFpEF trials to date, but by looking at these large trial databases to find patient features that predicted a better response to therapy and repeating the study focussing on these patients alone, testing again candesartan, perindopril, spironolactone or sacubitril/valsartan or even nebivolol that was tested in the SENIORS study that had subset of such patients. At least one is subject to follow-up trials SPIRRIT (ClinicalTrials.gov Identifier: NCT02901184) and SPIRIT-HF (EudraCT Number: 2017–000697-11) for spironolactone. Sadly, we may wait until there is commercial interest in repeating such studies. Lastly as we attempt to define the most important fundamental processes in HFpEF, my own personal guess for the first positive HFpEF trial will be a sodium–glucose co-transporter 2 inhibitor, for these agents act as effective diuretic agents and may prevent HF hospitalisations also in HFpEF as they have been shown to both in diabetes and HFrEF. Conflict of interest: A.J.S.C. declares no conflicts related to this work. Outside of this work, in the last 3 years, A.J.S.C. declares having received honoraria and/or lecture fees from AstraZeneca, Bayer, Menarini, Novartis, Nutricia, Servier, Vifor, Actimed, Cardiac Dimensions, CVRx, Enopace, Faraday, Gore, Impulse Dynamics, Respicardia, Stealth Peptides, V-Wave, Corvia, Arena, ESN Cleer.