Navigating the complexities of health care billing and reimbursement can often feel like dealing with a convoluted system that is only understood by a select few.
Frequently, these processes are also marred by inaccuracies and insufficient oversight, and they tend to prioritize the interests of the companies involved.
However, an Arizona State University professor says that artificial intelligence has the potential to transform the industry’s paradigm, making it more advantageous to clients and reducing the likelihood of errors.
A recent study co-writtenMohammad Amin Morid, assistant professor in the Leavey School of Business at Santa Clara University, is Sheng’s co-author on the study. by Olivia Liu Sheng reveals how deep learning can reduce billing errors, especially for high-need patients.
Sheng, W. P. Carey Distinguished Chair and a professor in the Department of Information Systems in the W. P. Carey School of Business, spoke with ASU News about the findings.
Note: Answers have been edited for length and/or clarity.
Olivia Liu Sheng
Question: What inspired you to write your current paper?
Answer: While AI has extensively impacted and transformed diagnostic, prognostic and treatment decision support, the potential of AI to enhance decision making and policy setting for the stakeholders of health insurance is underexplored.
The reform of health insurance to implement fee-for-value has been underway in the U.S. Risk adjustment models, which critically rely on the predictions of annual health care expenditures in adjusting health capitation amounts and payments for individuals, suffer from inaccurate predictions for high-need individuals, e.g., older individuals or young but sicker patients.
While the potential of enhancing risk adjustment models for such patients via principled AI innovations is promising, the limitation of accessing proprietary health insurance claims data for data-driven research has resulted in a glaring related research gap.
Q: Why is medical/insurance billing such a highly complicated system?
A: Because the spectrum of health insurance players and payers — e.g., government agencies, private insurers, providers and patients — have varying goals, standards and processes, medical insurance billing and other operations have been complex and opaque. Research that provides insights into the impact of the limitations and improvements of risk management models over heterogeneous patient populations can shed some light on certain parts of the ecosystem of health insurance.
Q: In your paper you contend that AI can improve overpayments and underpayments for patient care. How so?
A: Underpayments and overpayments denote the discrepancies between reimbursements and the preset capitation amounts by patients’ health insurance plans. These discrepancies could be caused by inaccurate patient cost predictions for heterogeneous patients underlying the decisions on capitation amounts.
Our paper proposes and illustrates an advanced deep neural network framework adept at distinguishing the complicated insurance billing codes and paying differential attention to heterogeneous medical journeys can significantly reduce such discrepancies.
Whereas patient cost predictions are essential in setting adequate capitation amounts and fee schedules, and reducing overpayments and underpayments, they could also potentially affect insurers’ decisions on insurance premiums to properly manage revenues as well as those on resource allocations to enhance care quality and coverage for patients.
Q: What is your future research focus on AI?
A: This study is a step toward ethical AI by enhancing the fairness of AI for risk adjustment models. Such amazing benefits of AI have triggered widespread deep transformations in all industries and sectors. At the same time, the harms and risks of AI caused by algorithmic biases, adversarial attacks during the AI life cycle and post-deployment misuses have surfaced and are disruptive to the sustainability and benefits of AI.
At the Center for AI and Data Analytics, a team of researchers and I are proposing a Mindful AI Framework to support more ethical, responsible and trustworthy AI implementations. Principled innovations and mindful AI implementations rely on sector-oriented governance strategies, design principles and auditing and regulatory criteria for AI technology in each sector. Hence, a main research stream of this team is to evaluate and enhance mindful AI governance strategies, control guidelines and design principles specific to several key industries in Arizona.