The process of codifying clinical activity for administrative and financial purposes is critical for a high functioning health system. This complex and arcane mechanism is still in the waiting room of digital health reform. ICD-10 diagnosis codes1 are a critical piece of the hospital reimbursement jigsaw, yet their selection and assignment is still subject to the vagaries of human decision making. Somehow, the funding system has been made so complex that the default setting is to penalise hospitals through under-payment if they rely on the fallibility of humans.

Until now, the process relied solely on human doctors, human clinical coders and human accountants. The doctors write the case notes while caring for the patient, the coders read them, abstract clinical terms, assign the ICD-10 diagnosis codes and group them for each episode of care. Finally, the accountants send the bill and claim the hospital’s payment.

The advent of artificial intelligence (AI) in health system decision making raises new possibilities to accomplish this vital task.

Not human-centred design

But first, let’s investigate my premise. Speaking with CEOs and CIOs of 30 different healthcare organisations around Australia has confirmed my suspicions. ICD-10 codes are used inconsistently—a real problem if you’re trying to run a hospital based on your forecast revenue as a healthcare administrator. Health Information Managers confirmed that clinical coders are regulated by the Australian Coding Standards, however the clinical coding varies by individual, by hospital and by state. The challenge is nothing new. In fact, research has moved on from the question of whether there are errors in ICD-10 coding, to studies that “examine potential sources of errors at each step of the described inpatient International Classification of Diseases (ICD) coding process”. The Australian executives Accenture spoke to confirmed there is a problem with the way Activity-Based Funding (ABF) is coded, grouped and billed—it’s a wicked problem and many attempted solutions have failed.

The “Speed versus Quality” dilemma

The fundamental challenge is healthcare’s age-old speed versus quality dilemma. The bottom line is that if you code more, quality suffers, and if you code better, productivity suffers. As Accenture global health lead, Kaveh Safavi has pointed out in his blog titled Breaking Healthcare’s Iron Triangle, access, affordability and effectiveness no longer need to be competing priorities. Intelligent ICD-10 coding is an opportunity to deliver simultaneous improvement in cost and productivity while facilitating revenue assurance for hospitals. We call this: “faster and smarter.” This is no small matter—we’re talking about the accuracy of health payments totalling billions of dollars across the world, with $50bn being paid out annually based on coding in Australia alone.

The Health Information Managers we spoke to estimated that variability in terms of coding assignment is 20 to 30 percent. That’s not to say there is a 20 to 30 percent saving available, but rather that activity and revenues can be significantly more accurately predicted and assured. Better documentation of diagnosis coding would result in better health system data and better revenue assurance—knowing that hospitals are claiming and getting paid the right revenue. Furthermore, better assurance leads to reduction in costly, onerous, hospital audits which are currently overwhelming both payers and providers. This is not about wastage in crude terms, it is about uncertainty and lack of assurance for activity-based payments.

Australia suffers an undersupply of skilled, certified clinical coders. The fulcrum of speed versus quality means episodes of care often get under-coded. So the outcome of a hip replacement procedure might be impacted by diagnoses of hypertension, peripheral vascular disease, atherosclerosis and osteoarthritis, conditions which may not be coded if the doctor did not document them, if the hand writing was illegible or if the clinical coder did not assign them. The principal diagnosis must be accompanied by its attendant complications and co-morbidities. This might involve a difference of hundreds or thousands of dollars on a typical patient journey through the hospital. Given that there is under-coding across a significant proportion of hospital cases, the lack of revenue assurance typically results in tens of millions in revenue uncertainty per annum, per hospital.

Much health system research, reporting and policy making relies on accurate diagnosis coding. Improvements in code assignment would give a more accurate picture of the status of Australian healthcare, resulting in improved decision making and resource allocation for health system infrastructure, workforce, education, public health and population health management.

In my next blog I’ll reveal how Accenture has piloted an innovative solution based on a human + machine model, and has achieved some very encouraging results. Get in touch if you have any thoughts or questions.


1 - used in 192 countries around the world.

Ian Manovel

Managing Director – Innovation Principal Director Health, ANZ

Subscribe to Accenture's Insight Driven Health Blog Subscribe to Accenture's Insight Driven Health Blog