Multiple medications exist for the same condition and not everyone is suitable to take the same drug for the same health problem. Machine learning is advancing in its efficacy for assessing and helping treat patients in various ways.
This instalment, which is focused on machine learning, will delve into a new study published in the journal Heart, which examines the use of machine learning to identify cohorts of a population that may benefit from specific medications. Next month’s column will explore a second study on the use of machine learning for optimising treatment.
The study
In a study published in Heart, Sotomi et al (2023) examined the use of machine learning to identify cohorts of a population that may benefit from specific medications. The team had previously established that a machine learningbased clustering model was able to classify heart failure with preserved ejection fraction (HFpEF) into four distinct phenotypes. Depending on the phenotype, the person may suit a particular medication, and thus, specific medications may be found to have good efficacy in each of the four phenotypes. Therefore, the researchers assessed whether this was the case, as this could have positive implications for targeting certain populations. By administering correct treatment at an earlier stage in their heart failure, it may be possible to improve longer-term outcomes.
Sotomi et al (2023) conducted a post-hoc analyses of the PURSUIT-HFpEF registry—a prospective, multicentred, observational study. The team evaluated the clinical effectiveness of each of the following four types of medication given after discharge in the four different phenotypes: angiotensin-converting enzyme inhibitors (ACEi) or angiotensin-receptor blockers (ARB), beta blockers, mineralocorticoid-receptor antagonists (MRA), and statins. The team analysed for the primary endpoint of all-cause death and hospitalisation due to heart failure. They examined the data of a total of 1231 patients. The median duration for follow-up was 734 days. The team found that 528 patients, a total of 48% of the sample, experienced the primary endpoint. The team established that learning-based clustering may have potential in identifying populations for which specific medications may be effective. For example, the study found evidence that there is good efficacy of MRA, ACEi or ARB and statin for specific phenotypes of HFpEF.
Sotomi et al (2023) summarised their findings as follows:
- Machine learning-based clustering may have the potential to identify populations in which specific medications may be effective
- None of the four medications evaluated in this analysis had a significant effect on clinical outcomes in phenotype 1
- MRA significantly improved clinical outcomes in phenotype 2
- ACEi or ARB and statin therapy significantly improved clinical outcomes in phenotype 3
- Beta blockers tended to worsen the clinical outcomes in phenotype 4.
Phenotypes
Sotomi and colleagues went on to discuss the phenotypes. They described phenotype 1 as ‘rhythm trouble’, with a low comorbidity associated with it, and mainly only worsens heart failure through atrial fibrillation, with a mostly benign prognosis. They described ACEi/ARB blockers, beta blockers, MRA and statins being mostly lacking in efficacy for this group of patients, who may only benefit from a combination of antiarrhythmic drugs and catheter ablation.
However, for phenotype 2, MRA was found to be particularly significant for outcome improvement. This phenotype shows cardiac hypertrophy and hypertension, which is common in HFpEF, and is therefore described as ‘ventricular-arterial uncoupling’. Research has often used this phenotype as the focus, and inflammation is the main element of its pathophysiological development. Sotomi et al (2023) describe their previous research that found eplerenone to be of use in reducing structural alteration and diastolic dysfunction independent of lower blood pressure. They found that the upregulation of the mineralocorticoid receptor may have a central role in the pathogenesis of HFpEF, which may be why MRA, but not ACEi/ARB, significantly improved clinical outcomes in this phenotype. The team stated MRA may be effective in blocking upregulated myocardial mineralocorticoid receptors in this particular phenotype. Sotomi et al (2023) also point to other research that found MRA being exclusively effective in the phenogroup characterised by obesity, diabetes, chronic kidney disease, concentric left ventricular hypertrophy, high renin and biomarkers of tumour necrosis factor-alpha-mediated inflammation, liver fibrosis and tissue remodeling.
Phenotype 3 was found to benefit from ACEi/ARB, where the group was mainly characterised by people with high rates of chronic kidney disease and frailty. This particular type of medication was previously seen to be effective in frail patients; ACEi or ARB are well studied medications for hypertension, showing good efficacy for renal and cardiovascular protection in chronic kidney disease. The team stated that it is difficult to identify exactly why the drug has good efficacy in frailty and chronic kidney disease, but the effect can be multifactorial. Statins were also seen to have particularly good efficacy in this phenotype. The phenotype itself is described by Sotomi et al (2023) as ‘low output and systemic congestion’, which is characterised by increased gamma-glutamyl transferase and bilirubin, hypotension and bradycardia, carrying a relatively low burden of comorbidities (hypertension, diabetes and dyslipidaemia). The effectiveness of statin therapy was therefore unexpected. Statin therapy in HFpEF has never been evaluated in a randomised trial, but Sotomi et al (2023) explained that there have been some observational studies describing its effectiveness in this group of patients. The good efficacy may be due to improvement of endothelial function, increase in arterial distensibility, regression of cardiac hypertrophy, fibrosis, anti-inflammatory and immunomodulatory effects. However, these mechanisms did not explain the specific effectiveness of these drugs in this phenotype. The specific effect of statins on pulmonary hypertension may partially explain why there is significant effectiveness in phenotype 3, as it had the highest level of pulmonary artery systolic pressure. There will need to be further research as to what the precise mechanisms are, as these remain unknown and this warrants investigation. Therefore, the team recommended large-scale prospective studies to reconfirm the findings.
Phenotype 4 was seen to not benefit from beta blockers. In fact, the drugs may make the clinical outcomes worse in this phenotype. Specific features of this phenotype include the worst nutritional status, highest level of frailty and the worsening of infection-triggered heart failure (Sotomi et al, 2023). This might be due to an increase in central blood pressure caused by reflected pressure waves. Prolonged diastolic filling increases ventricular volumes and pressures, increasing ventricular load, which leads to elevated B-type natriuretic peptide and N-terminal pro B-type natriuretic peptide levels. However, the reason why beta blockers worsened outcomes exclusively in phenotype 4 remains to be clarified, the authors stated (Sotomi et al, 2023). The authors speculated that cardiac sympathetic activity in this phenotype may be more attenuated than in the other phenotypes, and therefore patients may be more prone to chronotropic incompetence due to this phenotype having a higher proportion of elderly patients, with a higher frailty score and a worse nutritional status. The use of beta blockers may further attenuate cardiac sympathetic activity and thus, could worsen their chronotropic incompetence. This might be why the use of beta blockers in this group resulted in a worse prognosis.
Conclusion
Caution should be applied when interpreting the results as most are based on hypotheses, and only four types of drugs were assessed. Cardiac rehabiltation and self-management were not analysed. Mechanisms of the treatments found to be effective also need to be established. However, overall, the study does indicate that machine learning-based clustering may be beneficial in identifying populations in which specific medications may be effective.
Machine learning is increasingly being adapted to aid in the treatment and prevention of physical and mental health conditions. As times goes on, more tools will be developed that help to more precisely direct the specialists responsible, to identifying the treatment that works most for the patient, based on their unique set of characteristics and circumstances.