Will AI Transform Health Care or Perpetuate Bad Data?

Artificial intelligence promises to revolutionize health care. Potential benefits include improving diagnostics, uncovering new treatment options, and facilitating personalized medicine.  

But what if the data-feeding AI is imprecise because it is based on studies that are not representative of diverse populations? John D. Carpten, Ph.D., a prominent genomics researcher and chief scientific officer at cancer research and treatment institution City of Hope, has these concerns.   

"While the mapping of the human genome has ushered in a new era of precision medicine, about 95% of the data in genomic studies conducted over the last two decades comes from the genomes of white Europeans. A scant 3% comes from Asians, and less than 1% comes from African Americans or Afro-Caribbeans, Africans and the Latinx community,” Dr. Carpten says.  

It is a situation that can lead to underrepresented populations not benefiting from the targeted treatments that precision medicine aims to provide. And medical biases could perpetuate for years. 

Scientists have pointed out that clinical studies must be more inclusive to better understand health disparities. In the United States, African American men are 76% more likely to be diagnosed with — and 120% more likely to die from — prostate cancer compared to Caucasian men, according to the National Institutes of Health. Yet, this phenomenon is just beginning to be studied. 

Multiple myeloma provides another example of cancer outcomes. Although African Americans make up roughly 14% of the U.S. population, they represent about 20% of multiple myeloma cases. 

AI Needs Diverse Data 

How do these findings limit AI? Most data used to generate cancer tumor biomarkers has come from populations largely of European descent — even though differences at the molecular level in tumors are seen in individuals with different backgrounds. If these data are used for AI to extrapolate findings, scientists may instead compound and perpetuate biases and introduce misinformation, which could exacerbate treatment gaps among groups — as well as genders and life stages. 

For AI to be beneficial for biomedical research, clinical databases must include and reflect more patient populations. By incorporating as much diversity as possible into the system, findings are more likely to be accurate and equitable, reflecting the diverse healthcare needs of all communities. 

Dr. Carpten says that including more diverse patient populations in clinical trials will lead to improved health outcomes consistent with real-world medicine. This approach also enhances the accuracy and reliability of AI in health care, as it ensures that the data used to train AI models is more representative of the broader patient population. 

Also Addressing Access and Workforce Diversity 

And there are other challenges to overcome. Even if actionable cancer tumor mutations specific to patient populations are identified, the findings will have little impact on patient outcomes unless access to specialized care for those populations also improves. 

Dr. Carpten says that building a diverse medical workforce is also important, particularly in cancer research. "Having scientists and outreach partners who reflect and understand the cultural nuances of the communities they serve helps build trust, foster inclusion, and increase participation in clinical trials,” he says. 

“We will get there, one step at a time,” says Dr. Carpten. “As the Reverend Jesse Jackson said, ‘When everyone is included, everyone wins.’” 

**This article is for informational purposes only and does not substitute for professional medical advice. If you are seeking medical advice, diagnosis or treatment, please consult a medical professional or healthcare provider.