Mapping the Heart for Future Health

Doctors rely on cardiovascular imaging to get vital information about patients’ heart function, their anatomy, and pertinent tissue features. It stands to reason that if imaging data from many patients with the same condition is pooled (anonymously, of course) doctors can tease out trends and patterns to help them and others better understand the condition.

It’s a viable idea, but one thing holds it back: the sheer amount of data that imaging returns can be overwhelming for healthcare providers, who struggle with how to best glean meaningful results from the reams of information. Enter phenomapping, a technique traditionally used to make sense of vast amounts of DNA information and to classify it by type. The technique is now making its way into healthcare research.

Phenomapping synthesizes many images to create phenotypes for patients with a particular disease. A phenotype groups patients by their predominant physical and biochemical characteristics. Not all patients with a particular heart condition, for example, have the same clinical profile. That is, the disease doesn’t manifest itself in the exact same way in all patients, although it can within big groups.

Researchers are now looking at the tool as a way to synthesize and make sense of cardiovascular imaging information, thus offering insight into how diseases play out subgroups of patients with the same disease.

Sanjiv Shah, a Northwestern University professor of cardiology medicine, has used phenomapping to help overcome what he calls a one-size-fits-all approach to healthcare. The approach can classify a group of patients with one disease into separate “phenotypes” within that category.

“That approach is a major challenge in the treatment of chronic medical conditions like diabetes, hypertension and heart failure,” Shah says. “The fact is, one size doesn’t always fit all.

Physicians can better tailor treatment to patients if they understand variations within the overall medical condition the patient has. This is where teasing out sub-groups of patients who manifest the same disease in different ways is helpful, Shah says.

The key is finding patterns among patients with those conditions. That’s where phenomapping comes in. For a 2015 study, he and his team used machine-learning algorithms to find patterns among 67 laboratory, electrocardiographic and echocardiographic markers from 397 patients with the heart condition HFpEF. Though phenomapping has traditionally been used to analyze genetic data, the researchers used the computer algorithms on non-genetic data gathered from patients in the university clinic, Shah says.

Current therapy doesn’t improve outcomes for the 3 million adults in the United States who have HFpEF. “Large-scale clinical trials have failed to demonstrate a significant benefit for any HFpEF treatment,” Shah says. “That was really the impetus behind the phenomapping analysis.”

The machine-learning tools led investigators to discover three distinct groups of HFpEF patients. An analysis of the groups showed that each of the three types of HFpEF has significantly different clinical profiles and levels of risk for hospitalization or death, and each demands tailored therapeutic strategies, Shah says.

In future clinical trials, the researchers plan to offer each of those three groups of patients’ specific, tailored treatments.

“The findings are a revolutionary departure from the current standard of care that lumps these patients into one broad HFpEF category,” Shah says.

Though much of the phenomapping research has been on patients with heart conditions, Shah expects other chronic diseases such as diabetes to be investigated and categorized by phenotype using the technique.

Shah says future studies using the Northwestern techniques on HFpEF patients from other hospitals are needed to verify the Northwestern results. But he believes phenomapping has much to tell us in the near future about how to best match the medication to the patient.

Jean Thilmany is an independent writer.

For Further Discussion