Cynthia Chude, PhD Candidate, The Wharton School
Abstract: Rising healthcare costs remain a significant challenge in the U.S., with one major contributor being the increasing incidence of amputations due to vascular diseases. The number of vascular disease-related amputations in the U.S. is projected to double by 2050, reaching an estimated 3.6 million cases. In 2023 alone, Medicare’s annual expenditure exceeded $900 billion, with $24.5 billion (~2.72%) allocated to amputation procedures and post-amputation care. Unfortunately, nearly 50% of individuals who undergo vascular disease-related amputations die within five years, a mortality rate higher than that of breast, colon, or prostate cancer.
Amputations disproportionately impact men, Black, and Hispanic individuals. These patients often experience a diminished quality of life, depression, and loss of employment. Comprehensive Limb Salvage programs involving innovative limb salvaging procedures have proven effective in reducing amputation rates by 36% to 86%. Despite this success, there are currently no standardized national guidelines for screening for peripheral arterial disease and identifying patients who are candidates for limb salvaging procedures.
This project aims to develop a machine-learning algorithm to more effectively identify candidates for limb salvage procedures, thereby reducing the number of unnecessary amputations. Additionally, it seeks to contribute to the AI literature on mitigating bias, addressing gaps in research that primarily focus on correcting biases in existing algorithms. For example, studies like Ziad et al. (2019) have highlighted racial bias in healthcare algorithms. However, they have not explored how AI can address innate provider biases, which are shaped by their lived experiences and influenced by reimbursement practices. Previous work has shown that AI algorithms can inadvertently down-code risk factors for Black and female patients, exacerbating disparities in treatment decisions and costs. This project shifts the focus to addressing human biases, particularly in the context of PAD patients, where provider bias may lead to the upcoding of risk factors for Black and male patients, resulting in disparities in procedural choices. Additionally, it examines how systemic provider factors, including financial incentives, influence biased decisions that lead to poor health outcomes.