Medical Coding With Artificial Intelligence Is The Future
Updated: Sep 10, 2021
Medical diagnostic and treatment coding has always been a difficult task. Even in simpler times, it was difficult to translate a patient's complicated symptoms and a clinician's efforts to address them into a clear and unambiguous classification code. Hospitals and health insurance companies now want a lot of information about what went wrong with a patient and what efforts were done to treat it.
Complexity increases as the number of codes increases.
The World Health Organization's ICD-10 (tenth version of International Classification of Disease codes) is the current international standard for medical coding (WHO). ICD10 has almost 14,000 diagnosis codes. In May 2019, the latest update of this worldwide standard, ICD-11, was formally adopted by WHO member states. ICD-11 will be implemented in January 2022 by WHO member states, including the United States. The new ICD-11 has over 55,000 diagnostic codes, four times the number of diagnostic codes contained in the WHO’s ICD-10.
In truth, at least in the United States, there are many more codes than the ones listed above. There are around 140,000 classification codes in an improved version of IDC-10 that is particular to use in the United States, with approximately 70,000 characters for diagnosis and another 70,000 codes for classifying therapies. Given that the US version also includes treatment codes and has previously contained a larger number of diagnostic codes, we estimate the improved version of IDC-11 that will be tailored to use in the US to have at least many times the number of codes in the WHO version of IDC-11.
No human being can recall all of the disease and treatment codes, especially since the number of codes has grown to tens of thousands over the decades. Medical coders have used "code books" to hunt up the correct code for designating a condition or treatment for decades. The procedure was obviously slowed by thumbing through a code reference book. It's not merely a case of looking for the proper code. There are challenges with interpretation. There are typically multiple ways to code a diagnosis or treatment with ICD-10 and previous versions of the classification scheme, and the medical coder must choose the most relevant options.
As a means of dealing with the rising complexity of coding diagnoses and treatments, the use of computer-assisted coding systems has progressively risen across the healthcare industry over the last 20 years. More recent versions of computer-assisted coding systems have incorporated cutting-edge machine learning techniques and other aspects of artificial intelligence to improve the system's ability to analyze clinical documentation—charts and notes—and determine which codes are applicable to a given case. To discover and check the right codes, some medical coders are increasingly collaborating with AI-enhanced computer-assisted coding systems.
AI-Assisted Coding and Mellissa Meyers
Melissa Meyers is a medical coder with 20 years of experience who lives in Southern California. She used to work for a corporation that owned several hospitals, but now she works for an acute care facility in a metropolitan area. She works from home, putting in eight hours per day to complete a particular amount of patient charts. She focuses on outpatient therapy, which frequently include outpatient procedures.
Meyers is well aware of the rising complexity of coding and is a strong proponent of her employer's AI-assisted com coding system, which offers codes for her to review. “It's come to the point where I can't recall everything—right side, left side, fracture displaced or not.” “AI only goes so far,” she cautions. For example, the system may read text from a chart document, note that the patient has congestive heart failure, and choose that disease as a diagnostic and reimbursement code. But that diagnosis is in the patient's past, not what he or she is currently being treated for. She explains, "Sometimes I'm shocked at how exact the system's coding is." “However, there are situations when it makes no sense.”
When Meyers opens a chart, there are codes on the left side of each page with pointers to where the code originated in the chart report. Some coders don't bother reading the patient chart from beginning to end, but Meyers thinks it's crucial. “Perhaps I'm a touch old fashioned,” she acknowledges, “but when I read it, it's more accurate.” “The system makes you faster,” she admits, “but it may also make you a little lazy.”
Some patient cases are straightforward to code, while others are more difficult. “I can check all the codes and get through it in a few minutes if it's just an appendectomy for a healthy patient,” Meyers adds. This is despite the fact that even a basic surgery requires numerous parts on a chart, such as the patient's physical examination, anesthesia, pathology, and so on. She does, however, point out that if its a surgery on a 75-year-old man who has end-stage kidney disease, diabetes, and cancer, she has to document their medical history and what medications they're on, which takes a lot of time. Medical history codes are also significant because if a patient has several diagnoses, the physician will have to spend more time with them. These 'assessment and management' codes are crucial for properly reimbursing the physician and the hospital. Meyers and other coders are subject to a 95% coding quality standard, which is audited every three months to ensure they are meeting it.
Meyers was skeptical of AI-enhanced coding when she first started using it a few years ago, fearing that it might throw her out of work. She now believes that will never happen, and that human coders will always be required. She points out that medical coding is extremely difficult because there are so many variables and scenarios to consider. She feels that coding will never be totally automated due to its intricacy. She's effectively turned into a code supervisor and auditor, double-checking the codes the system assigns and ensuring that the system's recommendations are appropriate for the situation. All coders, she believes, will eventually transfer to auditor and supervisor jobs in the AI-enabled coding system. The AI system makes coders far too productive to ignore it.
Meyers holds a “Associate of Science in Health Information Technology” degree, which she earned over the course of two years. She also has a number of coding certificates, both general and specific to her specialization disciplines, such as emergency medicine. Regular continuing education units and examinations are required to keep certificates current.
However, not all coders have received this level of education. Meyers claims that there are a slew of “sketchy” schools that offer medical coding training online. They frequently exaggerate the likelihood of a profitable job—paying up to $100,000 per year—if a student joins a six-month coding school. Another tempting feature of these professions is the ability to work from home.
The issue is that hospitals and coding service companies prefer experienced coders over inexperienced newcomers. AI makes the more simple and straightforward coding decisions; more complex coding decisions and audits require specialists. According to Meyers, the "newbies" may be certified, but without past experience, they have a tough time finding work. To make them productive, their employers would have to provide far too much on-the-job training. Both the AAPC (formerly the American Academy of Professional Coders) and AHIMA (American Health Information Management Association) have Facebook sites where their members can discuss issues related to medical coding.
Coding, on the other hand, is a good job for Meyers, especially with the help of AI. She likes it and finds it to be rather well-paid. Her ability to work from home at any time of day or night gives her a great deal of flexibility. If she didn't enjoy her current career, she'd be approached by headhunters about different coding opportunities all the time. Only entry-level medical coders, according to Meyers, are affected by AI, as are those who refuse to learn the new skills required to work with a smart computer.-