Dr Simon Peck
In his book “Profiles of the Future: An Inquiry into the Limits of the Possible”, Arthur C. Clarke made the following, widely quoted observation:
“Any sufficiently advanced technology is indistinguishable from magic”.
In today’s world, there are many kinds of advanced technologies ranging from nuclear fission to computer chips. One area that continues to fascinate is medicine and its associated technologies. Even though our understanding of medicine and the human body has advanced over the millennia, to the outside observer the process of treating the human body using technology remains as mysterious as ever.
This mystery extends from the procedures themselves to medical jargon sprinkled with abbreviations and snippets of ancient Greek. Given the nature of medical procedures, it is no wonder that it is difficult to describe such procedures and treatments to a listener not well versed in the medical arts. Listening to a conversation between medical professionals can sound like a foreign language. Making sense of the anatomical terms, abbreviations and technical jargon used is a challenge.
Medical coding standards have been established to deal with this challenge. They exist to communicate conditions and procedures to non-medical consumers of information. The two most obvious uses of medical coding are medical outcomes (clinical) research and medical insurance reimbursement. These two kinds of coding are not necessarily compatible.
This bulletin will first discuss the two different categories of encoding (morbidities and interventions) and then the different uses of the encoded information (clinical vs reimbursement). I will argue that coding for reimbursement is inherently incompatible with coding for medical outcomes research use and illustrate the argument with real-life examples.
As a provider of analytics software to the Healthcare Industry, we deal with many kinds of medical coding when processing medical insurance claims.
The two broad categories are morbidity coding (describing the diagnosis of a medical condition such as "W06.5: Fall involving hammock") and services (describing the procedures carried out by healthcare providers, such as "Q0113: Pinworm examinations").
In addition to new versions being released over time, each version has geographic variations and clinical modifications. Clinical modifications typically add granularity, and geographic variations add local variations. Some providers have, over time, developed their own variants of the main codesets usually in order to add granularity in their particular areas of expertise.
We also occasionally see claims with no encoding at all – from an analytics provider perspective, these are an absolute nightmare. Without the ability to quickly analyse standardised data, it is challenging to spot FWA (Fraud, Waste and Abuse) patterns.
It is possible to use tools like NLP (Natural Language Processing) to infer encoding from medical descriptions. Kirontech offers an NLP solution that mitigates problems arising with from poorly encoded datasets. However, over the long term it is important to look at generating encoding at source; lack of encoding oftens masks other problems in the claim management pipeline. Consequently, a strong claim management pipeline always starts with properly encoding services and diagnoses.
The almost universally adopted standard for encoding diagnoses is ICD (International Classification of Diseases). It is a very easy-to-use and intuitive system that allows recording of symptoms, diagnoses and coexisting problems in a logical hierarchical format.
Initially, ICD used to be known as the "International List of Causes of Death". The name was changed to the current in 1949. Apart from acquiring a slightly happier-sounding name, the ICD sets starting from 1949 contain descriptions of medical conditions in addition to the causes of death.
The WHO has recently launched ICD version 11, but ICD 10 is the most prevalent worldwide at the time of writing. However, other versions (ICD-9 and even ICD-8) continue to be used in some jurisdictions and by certain providers. As an analytics provider, we must, of course, consistently deal with this variety.
For each version, regional variations incorporate additional items not found in the core versions. For example, ICD-10-AM (the Australian modification) contains the following gem of a code: "X27.0: Contact with Platypus". Interestingly, ICD-10-AM is also used in Romania and Saudi Arabia, even though the incidence of platypus contact in each country is presumably rather low.
The second part of the clinical coding process is intervention coding – recording the tests and treatments given to patients. This, unfortunately, is far less standardised than the recording of morbidities, and coding systems vary from country to country.
In the UK, the NHS records interventions using OPCS Classification System (Office of Population Censuses and Surveys), and the private sector uses CCSD codes, that are a cut-down version oif the OPCS codes. For many years, I was a board member of CCSD, and we tried to keep the coding system simple and clear and use the minimum number of classifications necessary.
Most of the world outside of the UK uses either regionally developed coding systems or HCPCS / CPT codes that originate in the US. HCPCS are public codes (the free access is mandated in the United States by Health Insurance Portability and Accountability Act, 1996), whereas CPT are copyrighted by AMA (American Medical Association). Both HCPCS and CPT originate in the United States. The United States is in many ways the global leader in data standardisation for medical coding, primarily due to the prevalence of private medical insurance and the need to standardise the services provided by different providers for Medicare and Medicaid.
The ACHI (Australian Classification of Health Interventions) is Australia's service coding standard.
Having briefly covered the two categories of medical coding, let us turn to the objectives of medical coding and see how the different codesets are used in practice.
One of the uses of clinical coding is to enable healthcare purchasers to pay for treatment episodes. We call this "Reimbursement Coding". We will refer to the non-reimbursement coding as “Medical Outcomes Research”, which includes areas as such clinical activity recording, research, management, plus many more. The key differentiator is the consumer of the encoded information: consumers of Reimbursement Coding are the payors (typically an insurance company). On the other hand, consumers of Medical Outcomes Research encoding range from academic institutions to management of hospitals.
The two main objectives for a medical coding system are as follows:
Objective 1: Coding for Reimbursement must be:
Objective 2: Coding for Medical Outcomes Research must be:
In other words, an effective coding system for reimbursement sacrifices flexibility for clarity and detail for simplicity. It is a difference in granularity. One way to bridge the granularity gap between the two objectives is by collecting detailed data and using this data to assign episodes to simple cost bands such as DRG bands in the USA and Australia or HRG (historically used in the UK by the NHS).
However, the use of cost bands only partly addresses the underlying conflict - there is a more fundamental problem which cannot be overcome. The fundamental problem is this:
"coding for reimbursement corrupts the data as healthcare providers choose codes which are the most lucrative, rather than the most descriptive."
This is not a marginal problem affecting a handful of fringe providers.
Every audit I have undertaken has consistently revealed the same result: miscoding for reimbursement pervades the entire data set. Whilst not all providers miscode episodes intentionally, there are still quite a few for whom miscoding is the norm. To compound the problem, an entire cottage industry has mushroomed over the last 30 years to help healthcare providers maximise reimbursement. In the USA, many companies specialise in reimbursement maximisation, and this practice is also becoming more prominent in the UK. Whilst these companies market their services in terms of "making sure you claim everything you can" or "helping you to understand the nuances of each payor", many of them in fact, do go the extra mile, teaching creative billing tactics such as coding the same service differently depending on who is paying. This is clearly not conducive to any meaningful research when it comes to medical outcomes, even though it's highly conducive to research when it comes to FWA!
Years ago, I was invited to an NHS coding conference as a panellist and speaker. In my presentation, I remember discussing this exact point. The initial reaction was disagreement from the academic coders and even those with an audit focus. However, in the subsequent discussions and using real-world examples, I managed to convince a few of them.
Let's work through that basic argument here.
This is a complex subject and one on which I could and may one day write a book. However, some of the basic strategies include:
As a little anecdotal story, my favourite misrepresentation case involved a particularly costly surgery which was not covered by insurance. The surgeon's claim was referred to our investigation team, and we spotted numerous inconsistencies in the medical information. This led us to suggest to the surgeon that his medical report was inconsistent with the facts. He took this suggestion very much to heart, so he decided to sue me personally for defamation. Unfortunately, perhaps in the red mist of rage – he did not check the package he intended to send very carefully as it included a letter from the patient offering him some theatre tickets and a fine bottle of wine if he "got the claim through".
This case was of considerable financial value. By the time I received the parcel, it had been referred to the Financial Services Ombudsman. Following internal discussions, we decided to include the letters in our evidence. The existence of prima facie evidence of bribery put an end to his legal adventurism. The Ombudsman refused to deal with the matter, referring it instead to the courts. Unsurprisingly, the surgeon quickly withdrew from the proceedings. The customer ended up paying the bill, their policy was canceled, and the surgeon himself was sanctioned.
So – having discussed different kinds of coding and their uses in detail, we can perhaps offer the following summary:
As always, we would be happy to hear views from others on any of the above.
Dr Simon Peck, Chief medical Advisor, Kirontech UK Ltd