Autonomous vs Computer-Aided Medical Coding
Blog: AI for Electronic Health Records
Target audience: Beginner
Estimated reading time: 10'
This article provides a synopsis of the technology involved and the advantages and expenses associated with autonomous and computer-aided medical coding. It is designed to be accessible without requiring prior expertise in healthcare processes, yet healthcare administrators, practitioners, and billing personnel are likely to find it particularly advantageous.
Table of contents
Health Revenue Cycle Management
We'll begin by examining revenue cycle management and its critical role in medical coding, a key element in charging for healthcare services.
Revenue cycle management (RCM) in healthcare is a vital financial process that encompasses the administrative and clinical functions essential for the handling of claims, processing payments, and generating revenue. This intricate process involves the identification, management, and collection of patient service revenue [ref 1].
The advent of Electronic Health Records (EHR) systems has significantly enhanced the efficiency of healthcare revenue cycle management. Many healthcare organizations now rely on technology to monitor the entire lifecycle of claims, secure payments, and manage denials effectively.
There are several key components to revenue cycle management:
- Pre-authorization for procedure reimbursement
- Billing with standardized medical codes
- Collection of payments
Medical coding
Medical coding transforms the contents of a clinical narrative or patient record into a string of codes, succinctly capturing the diagnosis and procedures (such as medication administration, recommendations, follow-up plans) relevant to a patient’s medical visit. These codes are compiled into a claim, which is then forwarded to one or several health insurance providers.
A typical claim includes diverse types of medical codes. The most prevalent are:
- Current Procedural Terminology (CPT): This is a standardized vocabulary or system for coding medical services and procedures, facilitating streamlined and precise reporting by health providers [ref 2].
- International Classification of Diseases (ICD): This system is employed by healthcare providers for classifying and communicating diseases, symptoms, or clinical findings that are relevant to the diagnosis of the patient [ref 3].
The major complexity in creating a claim arises from the extensive number of codes available—around 19,000 CPT codes and 71,000 ICD-10 codes that must be accurately pinpointed in a patient’s medical record. The recent ICD taxonomy update to ICD-11 expands the diagnostic codes even further, by as much as four times!
Note: The following billing workflow is simplified for the sake of illustrating the pros and cons of autonomous and computer-assisted medical coding.
Simplified workflow of the billing process for medical service
The steps of the manual generation of a claim from a patient chart is as follow:
- The billing/RCM company loads clinical notes/patient charts from the provider EMR system
- Medical coders extract the clinical codes (CPT, ICD, ...) from the clinical note given the patient and provider data
- Coders formulate the claim for the service rendered
- The claim may be audited for quality purpose
- The claim is submitted to the health insurance provider usually through a clearing house.
Computer-assisted coding
The goal of a computer-assisted coding (CAC) system is to expedite the billing workflow. It does so by automatically extracting medical codes (such as CPT and ICD) from clinical documentation and creating or submitting a payment claim [ref 4]. Human coders then review and, if necessary, adjust the proposed claim and its corresponding codes.
Additionally, CAC systems can aid coding professionals by pinpointing and emphasizing specific segments or critical terms within the patient's record that corroborate the proposed claim and its codes.
Simplified workflow of the computer assisted medical coding
This level of automation cuts the time to produce a health insurance claim by up to 60%. The efficiency of the process relies on the synergy between the computer assisted software/service and the human coders.
CAC alters the billing process as follow:
- The billing/RCM company load clinical notes/patient charts from the provider EMR system
- The computer assisted coding software processes the patient chart
- The system suggests/predicts a claim
- The human coder validates and corrects the suggested claim if necessary
- The claim is forwarded to the health insurance provider through a clearing house.
The technology behind Computer-assisted coding platform combines coding rules and heuristics (domain expertise), Natural Language Processing (NLP) and machine learning models.
Autonomous coding
The purpose of an autonomous coding engine is to create and deploy a software solution that by-passes entirely human coders.
The workflow for the fully automated medical coding service is
- The autonomous coding engine loads a clinical note from the provider EMR system
- The engine generates a claim
- The claim is submitted to the health insurance carrier.
In contrast to computer-assisted coding, autonomous coding directly produces a claim using a blend of information from the provider and patient, as well as the transcription of the doctor's visit [ref 5]. This direct conversion of clinical notes into a health insurance claim depends on advanced deep learning models, including Large Language Models (LLMs) like those behind ChatGPT, BERT, and XLNet.
Models
The comparison between autonomous and computer-assisted coding may use many characteristics depending on the stakeholder. The following table highlights some of these characteristics.
Generic comparative table for autonomous vs computer assisted coding
Computer-assisted programs sift through patient charts to gather an array of medical codes, which are then compiled to formulate the required claim. The claim creation process can be fully automated for straightforward diagnoses and procedures. However, when dealing with more intricate health services that involve referrals, the expertise of a human coder might be necessary to finalize, assess, and approve the claim prior to its submission.
The intricate architecture of large language models, often comprising over 100 million parameters that power autonomous coding tools, demands substantial data input and computational power. Yet, these autonomous coding solutions are capable of directly producing a claim, which streamlines the implementation and integration within the existing billing workflow.
Securing the patient data necessary to precisely train any machine learning model presents a considerable hurdle, heightened by HIPAA regulations and patient privacy issues. This obstacle becomes even more pronounced for autonomous coding systems, as they necessitate an exceptionally voluminous training dataset.
References
[2] CPT codes - Centers for Medicare & Medicaid Services[3] Comprehensive Listing ICD-10-CM - Centers for Disease Control and Preventions
[4] Computer-assisted coding WHIMA
[5] Building the Autonomous Coding Ecosystem: What HIM Leaders Need to Know
Abbreviations
- CAC: Computer-Assisted Coding
- RCM: Revenue Cycle Management
- EHR: Electronic Health Records
- CPT: Current Procedural Terminology (American Medical Association)
- ICD: International Classification of Diseases
- NLP: Natural Language Processing
- LLM: Large Language Model
- BERT: Bi-directional embedding representations for Transformer
- ChatGPT: Chat Generative Pre-trained
Comments
Post a Comment