Lumpers and splitters: the bumpy road to precision medicine

The eminent cardiologist Attilio Maseri once said that there are lumpers and splitters in medicine. As an example he argued that anaemia has many different phenotypes and causes and accordingly also their treatment differs substantially. However, in cardiovascular (CV) medicine we were lumpers for many years. This was initially a very successful approach, but may not be that successful in the long run. Indeed, splitters initially did not getting very far as homeopathy shows: Samuel Hahnemann (1755–1843), its founder, argued that every patient is different and accordingly requires an individualized treatment.1 Unfortunately, medicine was not mature enough at that time to pursue such a bold concept and homeopathy never made it to clinical effectiveness. Still, the debate between lumpers and splitters continued through decades and centuries. For example, in 1837 a big debate took place at the Académie Royale of Medicine in Paris between medicine’s first epidemiologist, Pierre Charles Alexandre Louis, who successfully analysed the cholera epidemic in Paris using a statistical approach (Ce sont le nombres qui cont!). In contrast, his opponent Risueño d’Amador, said L’homme moyen n’existe pas (the average human does not exist). He argued that if one would know the average shoe size of French citizens, such shoes would not fit to most of them.2 For the management of patients with heart failure, particularly those with reduced left ventricular ejection fraction (LVEF) or HFrEF, as first described by William Withering in 1785 as dropsy,3 lumping was amazingly successful. Although a reduced pump function can occur after a myocardial infarction,4 due to cardiomyopathies caused by a number of gene variants,5 due to storage diseases or deposition of false-folded proteins,6 due to excessive alcohol consumption or triggered by chemotherapeutic agents.7 the pioneering heart failure trials lumped all such patients together to evaluate the effectiveness of angiotensin converting enzyme (ACE)-inhibition.8 The success of this approach led to later trials using beta-blockers, such as bisoprolol,9 carvedilol,10 or metoprolol11—and again lumping was effective. Mineralocorticoid receptor antagonists came next12 and again it worked as it did with the latest drug class, i.e. the angiotensin receptor/neprilysin inhibitor or ARNI valsartan/sacubutril.13 The inclusion criteria of such HFrEF trials were based primarily on an ejection fraction of 35% or 40% or less. This cut-off was arbitrarily chosen by the trialists involved to assure proper event rates and an easy definition of recruitable patients. Most physicians would consider already an LVEF below 50% as abnormal; but the former definition worked for trials. The forgotten patients with mild HFrEF have been rediscovered in the most recent ESC Practice Guidelines on Heart Failure as heart failure with mid-range ejection fraction or HFmrEF14 as if they would represent a different disease. However, whether this tongue breaker is a true entity is uncertain: (i) imaging modalities are not as precise to reliably distinguish a LVEF of 38% and 44% (indeed a variability of 6% has been reported with 3D and >10% with 2D echocardiography);15 (ii) LVEF changes depending on the volume load, and (iii) it can also be altered by drugs other than those for heart failure. Recent registries, comparing HFrEF and HFmrEF showed that major adverse cardiovascular events (MACE) are higher in the former than in the latter.16 Therefore, it is reasonable to assume that HFmrEF is an early form of HFrEF and not a true entity. In fact, it is increasingly recognized that heart failure is a heterogeneous syndrome in which disease progression is associated with a dynamic functional and structural changes that cannot be described solely by LVEF.17 Thus, the next step calls for splitters along different lines: cardiomyopathies and their genetic origin is being increasingly unravelled and depending on the genetic mutation, outcomes can differ substantially, no matter whether they would classify as HFmrEF or HFrEF or even heart failure with preserved EF (HFpEF). For instance, Lamin A/C cardiomyopathies often require heart transplantation,18 while titin cardiomyopathy19 or pathogenic variants in the MYH7/MYBPC3 gene20 are prone to sudden cardiac death. To make things more complex, a number of gene mutations can cause more than one phenotype, most likely due to epigenetics and post-transcriptional modification. Thus, understanding both genetics and epigenetics will allow for personalized risk stratification and individualized management, particularly as regards sudden cardiac death and its prevention by implantable devices. Indeed, in the prevention of sudden cardiac death lumping failed miserably. For both ischaemic and non-ischaemic cardiomyopathies, guidelines recommended an ejection fraction of 30% to 35% or less as an indication for an implantable cardioverter-defibrillator (ICD). Yet as Myerburg et al. showed in a seminal paper21 only 1 in 5 sudden cardiac deaths per year occur in such patients. Moreover, unlike patients with ischaemic, those with non-ischaemic cardiomyopathy and LVEF of 35% or less overall do not benefit from an ICD.22 Interestingly, the degree of myocardial fibrosis appears prognostically important.23 Indeed, this group of patients appears to benefit from an ICD, while those without fibrosis do not.24 Here, a trial lumping all patients together was overall negative, but stimulated research into subgroups that may benefit from such an intervention. These hypothesis-generating results are now prospectively tested in a randomized trial. Eventually, such a personalized approach will not only allow for better risk assessment, but also for specific genetic interventions. Indeed, a seminal study in human embryonic tissue already provided a proof-of-concept of such an approach by correcting a mutation of MYBPC3 gene using Crisper Cas9 technology.25 This is likely a door opener for precision medicine in the field of cardiomyopathies, potentially providing cure rather than palliative treatment. It is not only a reduced LVEF that is a problem, too high an LVEF is also harmful. Indeed, there is a ‘U-shape’ curve of MACE including mortality and LVEF with the ‘sweet spot’ being around 60%.26 This questions the current LVEF definition of HFpEF. Also, HFpEF is likely a mixed bag ranging from an aging heart to amyloidosis. Of note, with aging the LV gets smaller and the left atrium bigger and LVEF tends to increase even in healthy elderly. In some, probably those with myocardial fibrosis or deposits or renal failure among other comorbidities, filling pressure, natriuretic peptides, and pulmonary pressure increase and symptoms of HFpEF develop. However, while lumping worked in HFrEF, it so far was not effective in HFpEF.27–29 Indeed, in the TOPCAT trial, mainly those with mildly reduced LVEF benefited from spirolocatone, further arguing for a narrower definition of HFpEF.30 Just recently, the large PARAGON trial using valsartan/sacubitril narrowly missed its composite primary endpoint of reducing CV death and heart failure hospitalizations.31 Of note, a subanalysis suggested thst those with an LVEF below the median - i.e. those with early HFrEF rather than HFpEF - did benefit. In contrast, dividing out those with amyloidosis provided the first effective treatment for a specific form of HFpEF, i.e. Tafamidis in Transthyretin Amyloid Cardiomyopathy.32 Thus, splitting maybe the way to go in HFpEF as lumping did not work. Congenital heart disease used the reverse approach—every patient is different. And indeed, there is some truth to it as many such lesions and their combination are quite diverse. Contrary to expectations individualized management of such patients initially made important progress thanks to understanding haemodynamics, advanced imaging and pioneering surgeons and paediatric cardiologists such as Alfred Blalok, Helen Brooke Taussig, Ake Senning among others. However, medical treatment remained eminence-based.33 Indeed, large randomized trials are rare as the mindset of this specialty is dominated by splitters rather than lumpers. While there are many randomized trials in pulmonary hypertension,34 just a few have been performed in congenital heart disease, i.e. in Eisenmenger Syndrome. The BREATHE-5 trial, found the endothelin antagonist bosentan to be safe and effective in lowering pulmonary vascular resistance and improving the 6-min walk test.35 Subsequent studies, with bosentan and the phosphodiesterase inhibitor sildenafil suggested even improved survival.36,37 However, the most recent MAESTRO trial with macitentan was neutral in its primary endpoint of 6 min walking distance.38 Finally, the prostacyclin agonist selexipag showed promise in a small study.39 These trial is just a start; many trials must follow, if this field wants to progress further. When President Dwight D. Eisenhower had his famous heart attack on 23 September 1955,40 there was only one type of infarction and not much to do about it except tender loving care. Lumping everything together did work: defibrillators,41,42 beta-blockers,43 and aspirin44 reduced mortality substantially. But then thrombolysis worked only in those with coronary occlusion and not in those with non-complete coronary lesions. As a consequence, different guidelines evolved for patients with ST-segment elevation myocardial infarction (STEMI45) and those with non-STEMI.46 Beyond that, the 4th Definition Universal of Myocardial Infarction (2018)47 even distinguishes five types of infarction as well as myocardial injury. Splitting proved very effective and today mortality of acute coronary syndromes (ACS) has dropped to a record low.40 In addition, new entities such as acute myocarditis, Takotsubo Syndrome,40, 48 coronary dissection,49 and the enigmatic myocardial infarction with non-obstructed coronary arteries (MINOCA)50 that still wait for a proper understanding of their causes and effective management further expand the spectrum of ACS—the door is open for further personalization. There is even room for dividing STEMI further: intracoronary optical coherence tomography revealed that plaque rupture is not the only cause of ACS.51 Indeed, often endothelial erosion is the underlying cause. Whether stenting of such lesions after percutaneous reopening of the occlusion is the right approach has been questioned.52 Possibly, such patients require just interventional thrombus removal, antithrombotics, massive low-density lipoprotein cholesterol (LDL-C) reduction, and no stent.53 Finally, antithrombotic management offers room for personalized medicine. Conceptionally, patients with STEMI would require an anticoagulant as an occlusive clot can only form with substantial fibrin formation. However, exposing all STEMI patients to a novel oral anticoagulant (NOAC) provided confusing results: while a low-dose NOAC reduced ischaemic events, it increased bleeding.54 Of note, thrombus firmness varies among patients with STEMI. While some lyse their clots rapidly, others require much more time.55 Thrombus lysis time is highly predictive of future outcomes. In the PLATO trial, thrombus firmness was predictive of mortality, while loose clots were predictive of bleeding.56 Thus, the propensity of some STEMI patients to form solid clots may provide the basis for personalized antithrombotic management specifically in ‘thrombotic occluders’. Prevention started with a very simple, general approach; i.e. the introduction of hygiene in the 19th century. Although very unspecific this approach was probably the most successful for human health, as it substantially reduced the incidence and prevalence of numerous infections that had killed thousands in the past. In the 20th century non-communicable diseases became important, while their causes remained uncertain. When Franklin D. Roosevelt, the 32nd President of the United States, died on 13 April 1945 with blood pressures of 300/190 mmHg, neither the role of hypertension, nor its management was known.57 The Framingham Heart Study, initiated by his successor Harry S. Truman, provided the first evidence on the role of specific CV risk factors such as high blood pressure, lipids, diabetes and smoking,58 and recently even pain.59 While general measures, such as a low salt and low-fat diet proved ineffective, specific drugs for hypertension60 and lipid disorders61 proved successful. As it turned out lipids are diverse molecules. HMG-coenzyme-reductase inhibitors, i.e. statins, were a breakthrough and allowed for a marked reduction in LDL-C, myocardial infarction, stroke, and death.62 However, there are many remaining CV risks. Personalized prevention focusses on lipoprotein(a) that is genetically determined and important in young patients with coronary artery disease or ACS and in those with repetitive procedures in spite of controlled LDL-C levels.63 While PCSK9 inhibitors do lower lipoprotein(a) moderatly,64 antisense technology is able to reduce it by up to 80%.65 Finally, diabetics with elevated triglycerides who often have low HDL-C and LDL-C levels benefit from unsaturated fatty acids such as icosapent ethyl, a highly purified eicosapentaenoic acid. In the REDUCE-IT trial it not only reduced triglycerides, but also MACE by 25%.66 Antisense technology may provide a further personalized approach.67,68 Thus, personalized prevention started to address unmet CV needs by refining approaches according to individual risk profile. Finally, inflammation is a major driver of atherosclerosis in those with elevated C-reactive protein (CRP) levels. In patients after ACS, the interleukin-1β antagonist Canakinumab reduced MACE by 15%. In contrast, a low dose of the less specific anti-inflammatory drug methotrexate proved ineffective in the CIRT trial.69 Thus, in patients with high CRP, atherosclerosis is specifically driven via the NLRP3 inflammasome-interleukin pathway—another success of splitting. Thus, progress in medicine evolves on a long and winding road, meandering between lumping and splitting to the benefit of populations and individual patients—precision medicine is still ahead, but within reach. Conflict of interest: There are no specific conflicts related to this manuscript, but the author acknowledges that over the last 3 decades he has worked with most pharmaceutical and device companies that allowed for the progress described in this article.