AI Helps Hospitals Predict DRG-Based Claims Reimbursement, Costs

July 15, 2021 - Artificial intelligence enabled researchers to accurately predict claims reimbursement based on diagnosis-related group (DRG) shortly after inpatient admission, according to a study published in npj Digital Medicine. The model requires no additional manual coding and could support hospitals in improving cost estimations and resource allocation.

Under the Inpatient Prospective Payment System (IPPS), each inpatient case is categorized into a DRG. Each DRG has a fixed payment rate based on the average cost of resources used to treat that specific category, according to CMS.

Teaching hospitals, hospitals that treat a high percentage of low-income patients, and unusually costly cases may warrant an increase in payment amounts. As a result, hospitals typically calculate DRG payments post-discharge.

Since hospitals review DRG payments retrospectively, it is difficult to make predictions and adjust for active patients. Researchers set out to streamline the process and enable early estimation of DRG costs using a deep learning-based natural language processing (NLP) model that analyzes clinical notes to make estimations.

The model can adapt to variations and does not require additional manual coding efforts, which makes it a promising tool for supporting smoother operations and real-time decision-making in hospitals.

Researchers used the MIMIC-III dataset, which includes 40,000 patients admitted to critical care units at the Beth Israel Deaconess Medical Center in Boston.

“By leveraging a large set of patient data, deep neural network models have the potential to identify important diagnostic indicators from the raw data and encapsulate clinical patterns,” the study explained.

“Since each DRG group corresponds to a defined weight representing the expected payment, the modeling results can be applied to estimate DRG-based inpatient cost at the hospital level.”

The model enabled early classification of Medicare severity-DRG (MS-DRG) and all patient refined-DRG (APR-DRG), which was later used to estimate cost for patient populations. The NLP model also assessed cost indicators like case mix index (CMI), which averages DRG payment weights of a patient group and is commonly used to estimate costs.

The prediction model contained the CMI error under 15 percent for APR-DRG and under 8 percent for MS-DRG based on the first day of ICU admission.

“The promise of big data to support managing high-risk patients and save costs in the high-spending healthcare industry is clear from the research,” the study authors wrote in an accompanying article.

“As is the value of applying big data approaches to clinical text data to provide clinical decision support, but its use in hospital operational planning had previously been unexplored.”

Artificial intelligence’s capabilities are vast, and its applications in the medical field continue to grow. A recent study found that machine learning, artificial intelligence, and predictive analytics can accurately predict COVID-19 severity and risk factors.

The model accurately predicted clinical severity using data from the first 24 hours of a patient’s hospital stay. Authors also concluded that demographics and comorbidities were linked to higher clinical severity.

Recent analysis also revealed that artificial intelligence could be applied to anemia-related lab tests in connection to COVID-19. The research uncovered a link between patients readmitted to the hospital for long-term COVID-19 symptoms and their likelihood of developing anemia before or after infection.