IHA-25

Estimates suggest that billing errors may cost patients and the healthcare system anywhere from $10 – $20 billion annually. Students on the IHA team will integrate AI techniques to develop a tool that assists auditors in identifying improperly billed patient services, assigning proper billing codes, and predicting such errors to prevent recurrence.

Abstract:

Accurately determining medical billing codes and billing correctly is very challenging due to the complexity and specificity required in coding medical procedures and diagnoses. The coding system, Current Procedural Terminology (CPT), involves a vast array of codes with nuanced distinctions, demanding precision to reflect the exact nature of the medical services provided. Errors can arise from ambiguous documentation, variations in clinical terminology, and human transcription errors. Overbilling errors in the system result in patients and their insurers being charged incorrectly and require significant time and effort for all involved to correct after the fact and can lead to expensive audits for the medical system. Underbilling costs the medical provider in lost income for services that they have provided, impacting the bottom line.   

AI has the potential to significantly improve the accuracy and efficiency of clinical billing systems by automating the coding process and reducing human error. Machine learning algorithms can analyze vast amounts of clinical data and documentation to identify the correct codes with high precision, adapting to changes in coding standards and payer requirements in real-time. AI can also assist in flagging discrepancies or inconsistencies in the documentation, thereby enhancing the accuracy of the bills generated. By streamlining the coding process, AI can reduce the administrative burden on healthcare professionals, allowing them to focus more on patient care while potentially lowering the incidence of billing errors and accelerating claim approvals.

Given access to large data sets of claim records, audit reports, and Electronic Medical Records (EMR), several AI techniques are well-suited to identify, correct, and prevent billing errors. Students on the IHA team will identify macro-trends, investigate patterns and anomalies, and generate insights from the data that are actionable by IHA full time employees (FTEs). The goal is to start small with a single CPT code and particular diagnosis that has high probability of misbilling. The team will look at historical data related to this CPT code and apply AI methodologies for this single use case and build out a proof-of-concept deliverable before expanding to additional CPT codes. A successful project will result in time-savings for the IHA audit team, capturing of missed revenue, and prediction of error-prone cases.  

The team will investigate techniques and come up with a strategic approach to address the use case(s). Together, these techniques should be able to enhance billing precision, compliance, and efficiency, ultimately streamlining the healthcare billing process. The team will likely incorporate more than one of the following techniques: 

Natural Language Processing (NLP): NLP can analyze and interpret unstructured clinical documentation that exists in EMR data, extracting relevant information such as but not limited to diagnoses, symptoms, and treatment plans to determine applicable billing codes. Various NLP techniques can be explored to extract relevant data from the clinical notes and provide structed data for use in the machine learning model. Labeled data helps train models to understand and extract relevant information from text accurately.

Machine Learning (ML) Classification Model: ML classification models can be trained on historical claims data and auditor reports to predict the appropriate codes for new cases. IHA is interested in exploring supervised learning algorithms such as Random Forest, Gradient Boosting Machines (GBM), Support Vector Machines (SVM) or other appropriate algorithms for this context. Labeled data will be available to train the models.

Rule-Based Systems: IHA is interested in exploring combining the ML classification model with a rule-based system to   ensure that coding adheres to established guidelines and payer requirements. This system will validate codes against predefined rules and highlight potential discrepancies for auditors to review. Expert auditors will be available to provide the established guidelines and requirements.