As technology continues to advance, hospitals and healthcare providers have gained the ability to utilize artificial intelligence (AI) and algorithms to automate many of the complex decisions that were once only capable of being made by human beings. AI is now used to detect and diagnose diseases, screen patients, and develop various treatment protocols. While the purpose of AI was intended to streamline and simplify decision making and detection processes for hospitals and providers, it has inadvertently created health inequities for Black patients. A study published by Science in 2019 revealed that predictive healthcare algorithmic software discriminates against Black patients by distributing medical resources to white patients over Black ones.
The study analyzed data from Optum’s algorithmic AI, a health services company, which programmed its AI to predict which patients were most in need of additional care. The color-blind AI system ranked patients solely according to how much they have paid for health care in the past. Because the algorithm did not account for race when predicting which patients were most in need of healthcare services, the fact that Black patients have less health access to healthcare, and thus pay less for healthcare services, was overlooked. The study revealed that Black patients had substantially worse health despite paying $18,000 less for healthcare services per year compared to their white counterparts. The discrepancy in the amount that Black patients paid for healthcare compared to white patients is the result of Black patients having less access to healthcare services. Of the patients that Optum’s algorithm indicated were in need of additional care, only 17.7% were Black whereas researchers indicated an unbiased proportion would have been 46.5%.
While Optum’s AI is only one of many algorithmic artificial intelligence products that disproportionally disadvantaged Black patients, finding and acknowledging the present biases is the first step towards correcting them in both AI and human behavior. While excluding race-based criteria may seem like a step towards health equity, racial equality calls for the acknowledgement that there are differences between racial groups. Small changes can be made to ensure that AI software is without bias. Measuring patients’ needs by avoidable costs or higher burdens of chronic conditions eliminates racial biases found in the original AI software as these measurements are more inclusive of Black patients’ struggles. Biases in AI algorithms are not inevitable. By recognizing the biases in AI, we can start to develop new technology that accounts for the different ways in which Black patients are disproportionately treated in healthcare settings.