What is a computer-assisted coding system (CACS)?

Manual clinical coding vs. CACS

Clinical coding is the process of translating details from medical documents into universal medical codes. Traditionally, such coding is performed by people called coders. However, manual clinical coding is both time consuming and labor intensive. It is also error prone and requires coders to have the knowledge and skillset to ensure correct text-to-code translations.

These challenges, along with the shift toward technology-driven healthcare have increased the necessity and demand for medical coding systems in the U.S., especially with the growth in the implementation of electronic health records (EHRs) and the transition to the ICD-10-CM system for classifying and coding patient diagnoses, symptoms and procedures. A CACS reduces the time and effort involved in clinical coding.

These developments have led to the increased adoption of hybrid coding systems, in which CACS perform a majority of the medical coding, especially on routine procedures. Human coders address more complex scenarios while auditing the CACS output for accuracy and relevance.

How a CACS works

CACS use ML algorithms and NLP to automatically analyze medical documents, such as patient records, charts, lab tests/procedures and radiological exams. The software identifies key terms and concepts in the document and suggests appropriate medical codes accordingly.

The CACS also analyzes the context of a piece of text to determine whether a particular instance requires coding and to ensure that the generated codes correctly reflect the documented care provided by the healthcare provider. For example, it can determine that the term "cancer" requires coding when it's a diagnosis, but not when it refers to a "family history of cancer."

Human coders can review the generated codes and make corrections or adjustments as needed. If no changes are required, the codes are finalized and assigned to the patient record. The provider can then use the code for a variety of purposes, including the following:

• To standardize the way diagnoses, procedures and services are described.
• Simplify communication with other care providers.
• Reduce misunderstandings and errors in describing medical conditions and treatments.
• Ensure accurate billing and payment receipts.
• Speed up claims processing and reimbursements, and minimize claims denials.
• Get deeper, actionable insights from healthcare data to inform treatment plans and improve resource utilization.
• Identify improvement areas for healthcare delivery, patient care and patient safety.
• Improve reporting for public health-related purposes (e.g., to understand disease patterns).
• Maintain compliance with healthcare regulatory guidelines and mandatory requirements.

Types of CACS systems

Most CACS systems are either rules-based or ML-based.

Rules-based systems use predefined or hardcoded coding rules that determine how they analyze medical text and generate codes. These systems are smart and can automate many coding tasks, helping human coders save time and improve productivity. Additionally, the products are easier to maintain since the rules are straightforward and can be reviewed and edited as needed. Also, rules-based systems use transparent logic and are explainable, i.e., their output can be easily explained and understood.

That said, the accuracy of the output of a rules-based CACS depends on the rules programmed into them. If the rules are not comprehensive and specific enough, the software might not produce the correct codes as output. To define these comprehensive and smart rules, programmers with domain expertise are required. Such experts can be hard to find, expensive, and might require additional training to program and maintain the system.

Another limitation of rules-based systems is that they cannot handle new, exceptional or unstructured data. This is again due to the rigid, preprogrammed rules. These rules place constraints on the system, which might result in inaccurate output if the input is different or if it comes from an unstructured document.

ML-based systems do not rely on explicitly programmed rules to analyze text and generate codes. Rather, they are trained on large volumes of data, referred to as training data. Training allows them to learn to identify patterns from the data and use this understanding to automate coding and generate accurate codes. As the system takes in more new data, it self-learns and continues to produce better output over time.

Unlike rules-based systems, ML-based systems can adapt to and analyze new data to produce accurate output. They can also handle the nuances and complexities of unstructured data. However, they require large training data sets and constant model improvements to continue improving their capabilities. For small healthcare organizations, these data sets can be difficult to obtain. Also, model tweaks might be costly and require significant technological and human investments.

Another drawback of ML-based systems is that their workings can be opaque. Also, their output is not always explainable. System opacity and lack of explainability make it harder for providers to assess the fairness, accuracy, and biases of the CAC software's ML models and algorithms. Without such assessments, it's difficult to build trust and confidence in the system and its output, which can be a significant problem for high-stakes medical scenarios.

Advantages and benefits of CACS

A CACS leverages advanced automated technologies like NLP to offer several advantages over manual coding. For one, it streamlines the medical coding process. It can quickly find relevant information from both structured and unstructured documents to match the codes in ICD-10, CPT, and other systems. This speeds up clinical coding and reduces the labor costs associated with manual coding.

CACS also increases the productivity and coding efficiency of human coders so they can focus on other more strategic or complex tasks where their skills are indispensable. For example, they can randomly audit CACS output for errors or suggest improvements for system accuracy.

Another advantage of automated CACS products with built-in ML and NLP capabilities is that they provide visibility into past treatments and diagnoses. This helps with the consistent application of coding rules, which reduces coding errors and helps maintain the integrity of medical data. Better quality data enables healthcare organizations to make more accurate diagnoses and design treatment plans that are more consistent with patients' medical histories and specific current requirements. In doing so, providers can deliver better quality care and thus improve patients' health outcomes. Accurate codes also ensure that providers receive proper payment and reimbursement for their services.

The coding consistency and accuracy enabled by CACS are also vital for compliance. By assigning the correct medical codes to medical records, CACS enables healthcare providers to continuously comply with all applicable laws and regulations. Additionally, the right codes allow providers to speed up claim processing and reduce unnecessary delays.

CACS can also ensure the accuracy of the codes selected, even if the codes or healthcare regulations change (which they frequently do). To keep up with changes and ensure coding accuracy, a healthcare facility simply needs to update the CACS. By removing the need to train human coders on new codes and technological updates, CACS systems enable providers to save time and money, while also maintaining compliance and avoiding the costs of noncompliance, such as fines or penalties.

Limitations of CACS

The cost and the chance for errors are potential disadvantages of CACS. A CACS that is unchecked by human coders can be less accurate (i.e., produce less accurate coding output) compared to a hybrid system that uses a combination of the CACS and manual coders.

Accuracy can also be affected when new medical codes emerge. To minimize errors and ensure accurate output, the CACS must be updated every time new codes are added. It's also important to test the system for accuracy and immediately flag errors, particularly in scenarios involving high volumes of scanned documents or paper records (or both). The EHR system being used might also affect coding accuracy so, testing the CACS and, if required, combining it with manual coding, is essential.

CACS is best suited for coding smaller, highly structured records like radiological exams or ER visits. However, for larger records, like inpatient visits, or for documents containing unstructured data, like images or clinical notes, a CACS might not generate accurate codes. In such situations, either manual coding or a combination of manual coding/CACS might be a better strategy. Human coders can review the CACS output and also manually review the documents to determine if the CACS missed any information.

Key features to look for when selecting a CACS

When selecting a CACS product, healthcare organizations and providers must look for software that includes the following features and capabilities:

• It should seamlessly integrate with the provider's EHR documents and images to ensure that it acquires all the necessary information and produces accurate codes.
• It should also integrate with existing clinical and administrative applications and workflows, including billing and insurance to enable seamless data-sharing.
• It should include the organization's must-have functionalities, such as work queues, security features and user-friendly interfaces.
• It should be able to assign the correct codes and modifiers depending on the services provided, date of service and patient.
• It should allow the organization to set a customizable time frame for document storage based on factors like current coding backlog, bandwidth requirements, memory costs, etc.
• Ideally, it should be scalable and include expansion capabilities.
• Ideally, it should maintain a list of the organization's providers (physician master index) to eliminate the need for manually reproducing this data.
• The product should include security safeguards to meet the Health Insurance Portability and Accountability Act's (HIPAA) security requirements.
• The vendor should provide regular upgrades and security patches to protect the organization's data and the security of its patients.

The selected CACS should also include reporting capabilities. It should provide detailed and downloadable reports that enable the healthcare provider to review the performance and return on investment of the CACS. These might include reports on the following:

• CAC acceptance rates (per coder or per department).
• CAC coder rejection rates (per coder or per department).
• New CAC charges captured.
• DNFC (Discharge Not Final Coded) report (by days and dollars).
• Trending by inpatient procedure code.
• Trending by outpatient diagnosis code.
• Trending by outpatient CPT code.
• Narrative summary of statistics.
• Other reports that help with clinical documentation improvement.

To choose the best possible product, buyers should request selected vendors to provide product demonstrations using real-life examples. Vendors should be able to show how their product decreases the average time needed to code a case, speeds up tedious manual tasks like code lookup and data entry, accurately captures all charges, and how it can potentially reduce additional work during claim submissions.

Before finalizing a CACS purchase, buyers should also ask vendors for important information, such as the following:

• Vendor's org size.
• Number of years in business.
• Number of successful installations.
• User group information.
• References from and contact information of other clients.
• Standard response time to issues or queries.
• Details about the vendor's NLP architecture, CAC database, and CAC reporting functions and features.

This information can help with vendor and product evaluations and enable providers to select the CACS that best suits their requirements, workflows and budget.

Popular CACS vendors

Numerous companies develop CACS products for use by healthcare organizations and providers. Some well-known vendors are listed below:

• Solventum. Solventum's CACS solutions leverage AI to seamlessly and automatically code medical records so healthcare organizations can deliver better care, reduce time-to-bill and minimize denials. Coding is based on Solventum's AI-powered, clinically driven review process, expert content, regulatory and compliance expertise, and insights-driven quality and patient classifications.
• Notable Systems. Its HIPAA-compliant CACS leverages three main technologies -- ML, NLP and optical character recognition -- to extract, classify and route critical information for automated medical coding. The product enables DME (durable medical equipment) and HME (home medical equipment) providers to streamline billing and reimbursement workflows, reduce write-offs and improve business operations.
• Optum. Powered by patented clinical language intelligence (CLI) technology, Optum's CACS automates coding and streamlines coding workflows. It, therefore, improves the accuracy and functionality of ambulatory coding, reduces denial rates and helps speed up revenue recovery.
• AGS Health. AGS offers an AI platform with a built-in NLP engine to automate manual coding workflows. The platform automatically suggests billable codes (ICD-10-CM, PCS, CPT) from different types of clinical documents, offers one-click coder validation and acceptance, and self-improves based on coder actions.
• Dolbey. Dolbey's Fusion CAC platform integrates AI and ML technologies for autonomous coding and automated prioritization that simplify medical coding and auditing. The product suggests ICD-10 and CPT codes for all patient types and includes modules for all coding-related workflows, including auditing, denial management tracking, and clinical documentation integrity (CDI).

Explore how a partnership highlights the key advantages of AI in medical coding and auditing within the revenue cycle.