This is a continuation of a series on medical coding, billing, and reimbursement.

Previous posts are here:

• Part 1–Intro & Procedural Coding
• Part 2–E&M Coding

I had planned to move on to federal monitoring and enforcement of health care reimbursement, but decided I would be remiss not to spend a little time on the diagnosis system and how it relates to medical billing and reimbursement. For those of you weary and bleary-eyed from the last two posts, this one will be a bit less insane–our friends over at ShrinkWrapped, Dr Sanity, or SC&A would probably diagnose this system as merely neurotic, rather than psychotic. But crazy it is, nevertheless.

When you submit a claim for health care services to a federal agency (e.g. Medicare or Medicaid) or a private insurance company, you must identify not only the service which you have provided, but the reason for which the service was performed. To do this, you use a system called the ICD-9 codes. ICD stands for the International Classification of Diseases, a system initially developed by the World Health Organization for epidemiology purposes–in other words, to track and categorize diseases in different parts of the world. The “-9” part indicates the revision number, and the ICD-9 has been around for quite a few years–at least 8 to 10 years to my recollection. The system, designed for tracking epidemics and targeting world health resources, has been adopted by health care payors to standardize reimbursement, similar to the CPT service codes spoken of in the first two parts of the series (CPT=current procedural terminology, the codes used for procedures and E&M services). There are plans for an ICD-10 which have been bandied about, but their implementation date is uncertain.

If you’re thinking that a system designed to send vaccines to Africa and track outbreaks of Dengue fever may not be ideally suited to health care reimbursement in the U.S., you’re showing great promise as a student of medical coding.

The ICD system goes something like this:

The codes are alphanumeric, and take a three-dot-two format: the first three digits are a category indicator, and those after the decimal are subcategory indicators. So, for example, diabetes mellitus has a category code of 250, whereas subcategory breakdowns may have 0, 1 or 2 digits after the decimal, e.g.:

• 250.0 — Diabetes Mellitus Without Complications
• 250.00 — Diabetes – Uncomplicated – Type II
• 250.33 — Type I (Insulin Dependent Type) Diabetes Mellitus With Other Coma – Uncontrolled
• 250.72 — Diabetes Mellitus With Peripheral Circulatory Disorders, Type II or Unspecified Type, Uncontrolled

And so on.

Each category has different number of subcategories, varying from none to dozens. So, for example, prostate cancer has only one code–185–while diabetes mellitus has 50. Some categories describe disease states, some symptoms, some family history or occupational risks or problems. At last count, my database shows just shy of 17,000 ICD-9 codes.

Now, logic dictates that you should peruse this rich and rewarding treasure trove of medical maladies, pick the code which best fits the clinical picture of your patient, attach it to a CPT procedure or E&M code, submit the bill, and get paid.

Excellent assumption–you’re deductive reasoning is impeccable. And you’d be wrong.

You see, some ICD codes are better than others.

Let’s say that you submit an office visit with the diabetes code above: 250.0, “Diabetes Mellitus Without Complications.” It fits the patient’s clinical picture well, and you have a deep-seated sense of satisfaction about its appropriateness. Only one problem: the code is not considered specific enough: it has subcategories with more digits. Therefore, it is highly likely to get rejected by Medicare or the insurance company with a “more specific code available” rejection indicator. Doesn’t matter that none of the more specific categories apply to your patient–still gotta use ’em. Keep in mind that in some categories–like prostate cancer, 185 above, there are no more specific codes available–and hence, a three-digit code is acceptable. In some categories, there are at most 4 digit codes, so a four digit code (say 188.9, bladder cancer) is just peachy–but a three-digit code is not. But if there’s a 5-digit code available, your 4-digit code isn’t specific enough, and is gonna get bounced. You remember math class, where 250.0=250.00? Not true in ICD codes, no siree. Different codes entirely.

With 17,000 ICD codes, you might assume that there’s a code for pretty much every medical problem out there.

There you go again, assuming things.

Turns out, there’s a fair amount of clinical situations for which there is no ICD code even remotely close to the clinical problem. And medical science, being a fast-moving endeavor, conforms poorly to ossified code categories, with new clinical situations and diagnoses appearing regularly–while the ICD codes get updated just once a year. And diagnosis codes outdate: the code you used last year may be invalid this year, or be supplanted by more specific subcategories–leading to the dreaded “more specific code available” error. Of course, though the codes are required to bill federal programs and health insurers, they are not free: you must purchase a license to use them–from our good friends at the AMA. Books and CD’s of these codes typically cost hundreds of dollars a year–and come with restrictions on their use (“This product is licensed for use on a single computer terminal. Use over a network or on multiple machines constitutes a violation of this license.” etc. etc.) And of course, the AMA and their resellers absolve themselves of any liability for errors or misprints–even though they may cost you a chunk of change–or even expose you to risk of fraud.

And remember our old friend edits? Yeah, the “gotchas” of procedural coding, where you can’t bill procedure A with procedure B because A is a component of B, or they are mutually exclusive. Well, not to be outdone, there are ICD code edits as well. Tons of them. And unlike the CCI edits (the massive list of paired procedure codes published by the feds), finding the ICD edits can be difficult or impossible. ICD edits work like this: If you do procedure A, and bill it with ICD diagnosis code B, the bill will be paid only if it is a valid ICD code for the procedure. Makes sense–kinda–at first glance: shouldn’t bill for a hysterectomy using an enlarged prostate diagnosis code. The fact that this is a largely unneccessary precaution seems to go unnoticed. More importantly, however, the feds and the insurance carriers use ICD codes as a means for rejecting payment–based on rules which are difficult, or even impossible–to determine prior to submitting the claim.

Medicare, for example, contracts with regional insurance companies to administer claims. Each of these “Medicare carriers” may have their own set of internal ICD edits–so a procedure with diagnosis A may get paid in New York but rejected in Nevada. The Medicare carriers at least publish their edits–although it is generally necessary to plow through pages of “newsletters” or poorly-designed web sites to find them. Health insurance companies, on the other hand, frequently refuse to publish their ICD edits at all, keeping them entirely black box. You find out only when your claim is rejected–and only then, when you call them (and wait on hold for an hour) to ask why they rejected it. And when you ask what ICD codes are permitted to get paid for this procedure, they refuse to tell you: “it’s not our job to practice medicine.” Yeah, right–but that’s exactly what they’re doing: restricting medical procedures and services by denial of payment.

Now lest you think I believe there are absolutely no redeeming values to the ICD diagnosis system, I must inform you of their most endearing characteristic: they make excellent cocktail-party conversation starters.

There are large–and largely-useless–categories of ICD codes which describe occupational or other atypical situations; these rarely are valid codes for medical reimbursement, but provide for lighthearted chatter abundantly sufficient to win the heart of the most jaded socialite over a glass of fine Riesling and Brie. To wit:

• E804.1 — Fall in, on, or From Railway Train Injuring Passenger on Railway
• E81.7 — Non-collision Motor Vehicle Traffic Accident While Boarding or Alighting
• E810.4 — Motor Vehicle Traffic Accident Involving Collision With Train Injuring Occupant of Streetcar
• E82.7 — Animal-drawn Vehicle Accident
• E820.6 — Non-traffic Accident Involving Motor-Driven Snow Vehicle Injuring Pedal Cyclist
• E83.3 — Fall on Stairs or Ladders in Watercraft
• E84.5 — Accident Involving Spacecraft
• E88.46 — Fall From Commode
• E828.4 — Accident Involving Animal Being Ridden Injuring Occupant of Streetcar

So there you have it–ICD-9 coding in a nutshell. Time for a lunch break–your seats are assigned in the dining area down the hall. Remember, you may order the soup only if your sandwich is not ham or turkey; and you may order soup and a sandwich, or desert and a sandwich, but not soup and dessert: they are mutually exclusive, unless you are diabetic, in which case you may have neither…

After lunch, on to federal regulation of medical billing.