IJCS | Volume 31, Nº6, November / December 2018

655 Mesquita et al. HFPEF phenotypes Int J Cardiovasc Sci. 2018;31(6)652-661 Review Article by machine learning algorithms. Machine learning is a field of artificial intelligence, in which a computer is programmed to learn the relationship between the objects of study by data processing and accumulate experience with previous problem-solving approaches. Machine learning algorithms are classified into supervised and unsupervised. While supervised learning is focused on outcome prevention, unsupervised learning aims to infer intrinsic structures of the data. Therefore, in this approach, a large volume of data can be analyzed and mutually exclusive categories of HFPEF can be obtained by phenotype mapping of the syndrome and grouping of patients in subgroups according to phenotypic characteristics. Phenotypic classification of patients with HFPEF would be helpful to the development of clinical trials on therapeutic strategies specific to each phenotypic profile. 12 A recent studywas the first to identify phenogroups of a heart disease, and the first to use the machine learning technique as an approach to solve the heterogeneity of a cardiovascular syndrome by phenotype analysis. 33 The data analyzed for patients’ classification using the machine learning approach included clinical variables, physical features, laboratory data and electrocardiogram and echocardiogram parameters. Although the patients shared many clinical characteristics, they were classified in three subgroups with distinct characteristics and prognosis: 33 • Group 1, composed of younger patients, with moderate diastolic function and relatively normal BNP levels. These patients have the mildest myocardial remodeling, electrical dysfunction and hemodynamic change, although 65% of them had moderate diastolic dysfunction and elevated pulmonary capillary pressure (PCP) and pulmonary artery systolic pressure. • Group 2 involved obese, diabetic patients, with a high prevalence of sleep apnea and impaired left ventricular relaxation. This group showed the highest PCP and highest pulmonary vascular resistance. • Group 3 was composed of older patients, with significant chronic kidney disease and pulmonary hypertension. Inthisphenogroup, amoreseveremyocardial remodeling and electric dysfunction was observed, with a longer QRS-T interval, higher relative thickness of cardiac walls, higher left ventricular mass index, higher E/e’ ratio and worse right ventricular function. 33 In addition, different phenogroups haddifferent clinical course and outcomes, and distinct risk stratification. Prognosiswas divided into the following categories: death, hospitalization for non-cardiac causes, hospitalization for cardiac causes, and hospitalization for HF. In group 1, the most frequent prognostic factors were hospitalization for cardiovascular and hospitalization for non-cardiovascular diseases; in group 2, hospitalization for non-cardiovascular causes andHF, and in group 3, themost prevalent outcome was death, followed by hospitalization for HF. 33 However, although ideally the subgroups should be mutually excluding, some patients had overlapping clinical features, especially in the analysis of group 1 patients (Figure 1). Even so, this was a pioneer study in the phenotyping of complex cardiovascular syndromes. 33 In light of the above, one may infer that the use of the machine learning tool in international centerswouldprovide new, essential information on HFPEF epidemiology. Considering that this study was conducted in a North American setting, 33 it is expected that results observed in the subgroups are different from those in South America. Therefore, application of the technique in LatinAmerican prevalence studies is paramount for future phenotype mapping of HFPEF in Brazil. Treatment Classical therapeutic approach of HFPEF has not reduced mortality and morbidity rates of these patients. Thus, considerable differences between the phenogroups indicate the importance of a specific therapeutic approach, since advances in therapies have been so far hampered by such phenotypic complexity. To deal with that, new therapies that have a direct effect on signaling cascades involved in the pathophysiology of the HFPEF have been proposed. 34,35 Today, these therapies varied from signaling pathways of systemic inflammation to myocardial elasticity, and additional therapies to different comorbidities associated with the same pattern of phenotypic predisposition to the disease. Aging, obesity, systemic hypertension, type 2 diabetes mellitus, kidney failure, and sleep apnea can trigger a chronic systemic inflammation that affects the myocardium and other organs. The patient may have pulmonary hypertension, sodium retention and impaired oxygen extraction by skeletal muscles. For patients with pulmonary congestion or metabolic risk, it is recommended the use of diuretics, statins, organic nitrites /nitrates, energy-intake restriction, stimulants of PKG pathway, and extracellular matrix- stimulating agents, like spironolactone.