How AI brings accessibility and equity to healthcare
The U.S. Census Bureau reports that during the onset of COVID-19 in 2020, 28 million American citizens didn’t have health insurance at any point during the year. And although many Americans did have health insurance, it often does not cover everything individuals need like mental health services and follow-up breast cancer screenings, which aren’t always covered.
This is where artificial intelligence (AI) can step in to provide quality healthcare options at a lower cost. Companies like Vara and Paradromics are already working to increase access, affordability and ultimately healthcare outcomes — and investors are paying close attention.
“AI could really solve this accessibility issue, especially now that an aging population is a big trend across developing and developed countries,” said Lu Zhang, founder of FusionFund, a venture capital firm focused on backing early-stage startups like Paradromics. “The main point is to be able to better understand the root of the disease and to achieve a highly personalized diagnostic and treatment plan.”
Those without access or with minimal health insurance coverage are often Black, Indigenous, and people of color (BIPOC) individuals and are disproportionately impoverished. The Kaiser Family Foundation (KFF) found that “In 2019, non-elderly AIAN [American Indian, Alaska Native], Hispanic, NHOPI [Native Hawaiian and other Pacific Islander] and Black people remained more likely to lack health insurance than their White counterparts.” And although programs and services like Medicaid and the Children’s Health Insurance Program help, “…they do not fully offset the difference, leaving them more likely to be uninsured.”
Insurance access was made even worse because of the 2020 pandemic, which disproportionately impacted individuals in the communities listed above with job losses and decreases in income, and therefore likely contributed to further disruptions in healthcare and medical coverage, according to KFF.
AI-powered healthcare on the horizon
“AI could improve health outcomes by up to 40% and reduce treatment costs up to 50% by improving diagnosis, increasing access to care and enabling precision medicine,” according to Harvard’s School of Public Health, If implemented correctly at scale, it could save the medical industry upwards of $150 billion in costs by 2025.
“I think we start with, for example, AI for medical imaging, AI for diagnostic or AI for medical sequencing. There’s also more discussion about how we could better improve workflow efficiency,” Zhang said. “When we talk about AI, we only think about AI algorithms, but there’s also other artificial intelligence products like AI robotics.”
Improving access and results in breast cancer screenings
Every year in the U.S., the CDC reports, on average 255,000 cases of breast cancer are diagnosed in women and 2,300 in men — and 42,000 women and 500 men die per year from the same.
As part of proactive healthcare planning and treatment, individuals, particularly women, are encouraged to have a mammogram performed annually or every few years, depending on age. Though, an important distinction particularly related to insurance coverage is regarding the type of screening they should get.
An annual mammogram is the screening most commonly covered by insurance plans as it is preventative care, according to United Healthcare, a multinational managed healthcare and insurance company.
However, if an individual goes in for an annual mammogram, for instance, and any abnormalities are found, they are then referred for a diagnostic mammogram, which is a screening that is less commonly covered by insurance, but that is used to diagnose breast cancer. And since the latter is used to make a diagnosis, more costs are typically associated with it, even if insurance covers part of it, United Healthcare notes.
The high costs for diagnosis is one reason Jonas Muff, founder and CEO of Vara, an AI-powered mammography screening platform, started his company. The company offers a software screening service that can be installed on existing machines and doesn’t require hospitals or healthcare companies to invest in substantial new equipment. Once a center adopts Vara’s technology, the main change (other than improved efficiency) is a branding partnership, which Muff noted is often simple and along the lines of, “Clinic XY powered by Vara.”
Vara’s software platform works across the workflow of a radiologist. Muff says Vara uses AI on multiple fronts. The software platform works to seamlessly filter out normal cancer-free mammograms, so the radiologist can spend more time focusing on and analyzing screenings that may have suspicious aspects. Additionally, Vara’s technology also alerts the radiologist in case they missed a potential case of cancer that might be otherwise overlooked. Muff said the team refers to this feature as Vara’s “safety net,” which, via its AI and machine learning, may more quickly spot potential cancer.
“The vision is really that every woman can afford it. The more clinics Vara is in, the more women can afford these screenings, which is then obviously very good for the patients, but ultimately, it’s also great for businesses and everyone in the cancer treatment industry,” Muff said.
In clinical trials in Germany, where the company was founded, Muff claims that Vara found roughly 40% of all cancers that were missed by the radiologists. To get an idea of the cost savings AI can provide in this way, Vara’s screening services in Mexico are offered for about $15, which Muff noted is typically self-pay. He said women pay for the service with their credit cards, given that they’re not insured for receiving the screenings. If they choose to have a screening performed somewhere else in private clinics without Vara, Muff claims they can expect to pay between $50 to $150 in Mexico per screening.
Personalizing diagnosis and treatments in mental health
Like breast cancer screenings, mental health care and treatment are also often left out of insurance coverage in the U.S. In fact, the National Institute of Mental Health (NIMH) reports that one in five U.S. adults live with a mental illness. However, many barriers exist among insurance plans that can often delay access to treatment for these conditions, cause individuals to travel far distances for in-network providers, or may not cover mental health treatment at all, leaving individuals to pay high out-of-pocket costs.
The National Alliance on Mental Illness (NAMI) cited the above in a 2020 blog post and stated that although measures have been taken to make care for mental health more accessible, it isn’t enough.
“The 2008 Mental Health Parity and Addiction Equity Act, Affordable Care Act and state mental health parity laws require certain healthcare plans to provide mental and physical health benefits equally. And yet, insurers are still not covering mental health care the way they should,” the post reads.
“A behavioral health office visit is over five times more likely to be out-of-network than a primary care appointment,” NAMI reports that, And additionally, in general, the organization has found individuals in need of this type of treatment report increased difficulty with “finding in-network providers and facilities for mental health care compared to general or specialty medical care. Often, going out of network was the only option for treatment. And individuals reported difficulty finding correct information about the in-network providers for their health plans.”
This can leave individuals who are in need of treatment with few options or options that are unaffordable. This is where Paradromics, an AI-powered company, hopes to bridge the gap.
Paradromics aims to develop a data interface that directly interacts with neural signals from the brain using AI and machine learning. One technology the company developing, called “Connexus Direct Data Interface,” collects a massive amount of individual neural signals with a fully implantable device designed for long-term daily service. Paradromics reports that its first clinical application is an assistive-communication device for patients who’ve lost the ability to speak or type, but the technology will likely expand to mental health diagnoses in the future.
“We can imagine a future where certain mental health diagnoses become better understood through a neurological — rather than psychiatric — framework. This type of understanding could contribute to destigmatizing these disorders,” said Matt Angle, CEO of Paradromics. “It is well-known that pharmaceutical treatments, which are broad-acting and have nonspecific action, are not universally effective and pose challenges for individualizing mental health care. Within the large category of mental illness and mood disorders, over 5 million patients in the U.S. suffer from severe, drug-resistant mental illness and could immediately benefit from new treatment modalities.
Though the technology isn’t yet commercially available, Paradromics’ goals include applications that focus on detecting and treating intractable mental illnesses. Paradromics’ devices would be surgically implanted to function and would likely be used therapeutically once a condition has been diagnosed.
“Researchers have shown that depression and mood disorders, for example, are brain-network level phenomena. Promisingly, mood states can be both decoded and modulated using implanted electrodes,” Angle said. “Already we can see clinical trials for depression using older generation brain implants (deep brain stimulators) and the capabilities to decode and modulate mood and other neuropsychiatric states will only get better when DDIs [Direct Data Interfaces] become clinically available.”
Maintaining privacy and quashing bias
While AI can help improve equity and access when insurance coverage falls short, privacy can still be a concern.
“We really need better technology solutions to show that we can protect data privacy. We should not just say whoever uses the technology should have confidentiality, but rather enhance the technology itself,” Zhang said. “For example, you can search within an encryption. That technology solution could enable us to show the public that the data has protection already. This will help them ease their concern regarding the privacy issue.”
Similarly, bias can pose an issue throughout healthcare, so training the algorithms properly, while maintaining privacy, is equally important.
“It is important that we find the right model where we take the human into full account with the training data loop and that we find the right workflow for medical experts,” Muff said, “If you train your algorithm only on data from a certain subpopulation … then it’s not guaranteed that the algorithm will work on every other population, for example. It’s important that you evaluate your algorithms on clinically relevant subtypes. If you don’t, it could do more harm than good.”