March 26, 2023

When the first draft of the human genome was sequenced in 2003, it cost nearly $3 billion. Today, a human genome can be sequenced for just a couple hundred dollars. On Tuesday, Sept. 20, Harvard Medical School hosted its annual Precision Medicine Symposium, which focused on the ethical development and deployment of genetic screening to predict people’s risk of developing various diseases.
During an afternoon session, panelists described that by comparing the genomes of large groups of patients with and without a particular disease, scientists can develop models used to calculate an individual’s so-called polygenic risk score — their relative chance of developing common diseases based on a holistic read of their genome. These risk scores can be combined with information about an individual’s lifestyle factors like diet, exercise, or smoking to enhance their predictive value. The hope is that polygenic risk scores will allow doctors and patients to then prevent, mitigate, or prepare early for illness.
Right now, however, genetic screening in clinical settings is largely reserved for patients who actively seek it out, such as in the case of screening for rare diseases among expectant parents. But some scientists, clinicians, and companies are pushing for the widespread use of these tests, say, for common disease. For example, companies like 23andMe offer direct consumer genetic testing for people to explore their effects of their ancestry on disease risk.
But before scientists scale the research and technology, panelists at the conference underscored the need for more diverse datasets to ensure that risk scores are applicable to all kinds of people.
“I see the single greatest limitation in genetics for precision medicine for common diseases as the fact that our datasets are so vastly Eurocentric,” says panelist Alicia R. Martin, a scientist at the Broad Institute.
The largest and most commonly used dataset is the UK Biobank, which includes half a million people, most of whom are white, explained panelist Carlos D. Bustamante, a professor of biomedical data science at Stanford University. When risk scores are calculated from Eurocentric datasets, they come with inherent limitations. Applying these models to non-European patients can compound health disparities due to their inaccuracies.
“There’s a different type of urgency because [polygenic risk scores] are being proposed to be applied to very large swaths of the population,” says Arjun R. Manrai, assistant professor of biomedical informatics at Harvard Medical School and organizer of the conference. “They’re being trained and built for very common conditions, things like Type 2 diabetes, coronary artery disease, and obesity.”
Yet, diversifying datasets will require reshaping genetic studies at every step, from creative recruitment strategies to standardizing analyses. And Martin and Bustamante both know the challenges of this process firsthand.
Martin leads studies which look for genetic risk factors for neuropsychiatric illnesses like schizophrenia, psychosis, and bipolar disorder in African populations. She notes that it’s taken her team seven years to collect 40,000 samples. In addition to confronting technical hurdles such as protocol standardization and ethics committee reviews, researchers have faced deep mistrust from patients.
Bustamante is part of a team building a project called Biobank of the Americas. They hope to compile data from 10 million people by collecting their Covid tests. Since the pandemic has affected people across all the “nooks and crannies of the healthcare system,” their patient sample is more diverse — he says that about 70 percent of people have had no problem with researchers keeping their Covid sample.
“We unfortunately don’t have as much money as either 23andMe or the UK Government or the NIH,” Bustamante says. “We’ve just got some freezers in Miami.”
Though they lack large-scale funding, Bustamante is excited about their grassroots approach to collecting genetic material. The same strategy could be expanded to taking samples from blood banks and bone marrow registries, he says.
Stella Aslibekyan, a genetic epidemiologist at 23AndMe, says that in terms of raw numbers, 23AndMe has large samples of data from minority populations — but that percentage-wise, it’s still not representative of the broader population.
“Because we have 12 million people in absolute numbers, we actually have pretty big cohorts of African American individuals, of Latinx individuals, and of different individuals belonging to different ethnic groups,” Aslibekyan says. However, as a percentage, 75 percent of their population is still of European descent, revealing a gaping need for diversity if the goal is to overcome mistrust and set up inclusive research infrastructure.
“We have a lot of work to do in terms of reaching out and building trust with the communities that had historically been hurt by medical research, and specifically genetic research,” Aslibekyan says. In the meantime, she says that they have rigid criteria for releasing a polygenic risk score to consumers: “If it performs poorly in at least one ethnic group, we don’t release it at all, because we don’t want to perpetuate existing disparities that exist in that space.”
By the end of the discussion, all of the panelists concurred that despite increased interest in genetics research and improved financial accessibility, the field needs to see equal gains in diversity to reflect national or global representation.
While researchers work on expanding diversity, clinicians need to pay attention to how they communicate test results, says Aniruddh Patel, a cardiologist at Massachusetts General Hospital.
Patel says he didn’t know what polygenic risk scores were when he graduated medical school in 2014 and hypothesizes that most primary care clinicians are not equipped to have robust conversations about these results.
Communicating the implications of risk scores is even more challenging in the absence of clear thresholds for health or disease status. The name of the conference — “The New Normal” — hints at this ambiguity: for example, should a patient with 30 percent risk of a cardiovascular condition be given the same advice as a patient with 80 percent risk?
While researchers may know scores come with caveats and could be imprecise, Patel explains that the bar is even higher in practice because “you’re actually practicing and giving recommendations that they apply to everybody.”
– Akila V. Muthukumar can be reached at Follow her on Twitter @akila29m.


Leave a Reply