Healthcare leaders call for better datasets to power AI innovations

As innovations powered by artificial intelligence infiltrate the healthcare system, stakeholders on Tuesday suggested how to navigate such tools without exacerbating inequities.

At a summit held by Northwell Health and startup studio Aegis Ventures, Aike Ho—a partner at Acme Capital, an early-stage venture firm in San Francisco—said AI-powered solutions tend to replicate biases and perpetuate inequities because they are often built on clinical-care protocols, which historically have been developed and tested only on specific populations.

Local startups are helping shrink the gap. Ho said the firm invested in women’s health startup Tia, which is based in the Flatiron District, in part because it is collecting more comprehensive data and rebuilding clinical care protocols to better fit women.

Ho cited a question on Tia’s intake form that asks if a patient has ever experienced sexual abuse or assault, both of which have been linked to a wide range of health problems.

Northwell is improving its data on maternal health in Black women by working with 100 faith-based collaborators in Black communities, said Mark Solazzo, president of strategic initiatives and chief operating officer.

Data related to a patient’s neighborhood of residence and employer type should also be included, said Empire BlueCross BlueShield CEO Alan Murray, adding that insurance companies can work with their employer clients to make those datasets more robust.

Download Modern Healthcare’s app to stay informed when industry news breaks.

Stakeholders at different points of care then need to better connect that data to paint a complete picture of patients’ health that all their providers can access and use to inform care decisions, Murray said.

However, those datasets need to be accompanied by context on how the data was compiled and labeled to result in better AI models, said Mark Michalski, worldwide head of healthcare and life sciences for Amazon Web Services.

Michalski said the first step is for innovators to articulate exactly what they want to build, so they can spot failures. They also need to develop algorithms that can offer explanations, not just answers.

“I don’t think anyone’s cracked this,” he said.

This story first appeared in our sister publication, Crain's New York Business.