Choosing the Diagnosis

“There are only three things that are important in medicine: diagnosis, diagnosis, diagnosis.”

William Osler, The Principles and Practice of Medicine

Osler’s aphorism about the importance of the diagnosis applies to the electronic health record (EHR) almost as much as it did to doctoring in the late 19^th century. Diagnoses have three functions in the EHR: identification of the nature of the patient’s illness, justification for orders, and support for billing. In the best of circumstances one diagnosis would fulfill all three needs; however, order entry and billing level of service have undue influence over our choices. As physicians, we have lost some control over these decisions. We may make the diagnosis in the exam room, but we choose the diagnosis in the EHR. The only way to tolerate the frustration is to be amused.

The pool of ICD-10 diagnoses in the EHR is expansive. Some diagnoses have too many options; others are simply absurd. Pain is the largest category, including every anatomic site and permutation imaginable. There are lists of aggravating and relieving factors, association with bodily functions, attribution to numerous procedures or implantable devices, adherence (or not) to pain management contracts and intensities for every integer on both the verbal and non-verbal pain scales. There are even specific diagnoses for patients who have “deficient knowledge” of their pain, codifying a type of poor historian.

Trauma, in particular, yields a frightening but entertaining litany of injuries. One can scroll through the multitude of body parts from which a patient could suffer a bite, or wound, or retain a foreign body within. There are 67 choices for a stab wound to the ankle.

To the dismay of geriatricians and good fortune for personal injury attorneys, there are scads of ways to fall, from the mundane to the bizarre. Only slipping on a banana peel is missing.

Fall:

on scissors (despite parental admonitions not to run)

downhill (like Jack and Jill)

from a moving sidewalk (Watch your Step!)

off a cliff (“Why don’t you go jump…”)

into a storm drain or manhole cover (Grandma was right next to me…)

from a haystack (but found the needle)

from other gliding type pedestrian conveyance (AI hallucination?)

slipping, tripping, and stumbling (Chaplinesque)

collision with (nonpowered) inflatable craft or other watercraft or object (Huh?)

into a bucket of water causing other injury (circus divers)

assaulted by steam or hot vapors (vegetable cooker with agency?)

slipping on ice (completes the three physical states of H₂0)

Smoking cessation has some strange entries. I hadn’t felt threatened by former smokers until I saw this list:

Smoking cessation:

former trivial smokers (less than one per day)

tolerant former smokers

aggressive former smokers

militant former smokers

Apparently, there is a continuum of nicotine addiction from tolerance to militancy in former smokers analogous to the stages of quitting. I should tread lightly when reinforcing smoking cessation.

For order entry in our hospital-based internal medicine practice my preferred diagnosis is the one that best reflects my rationale, but sometimes the appropriate diagnosis doesn’t compute. For example, the diagnosis that allows me to order a screening bone density study for osteoporosis is not the intuitive “screening” or “osteoporosis risk” but “asymptomatic postmenopausal status,” a condition that an infinite number of monkeys using AI would never consider. Chest CT for lung cancer screening presents a similar problem. The magic words are “personal history of tobacco use, presenting hazards to health.” Prompting for the USPSTF criteria comes later. Choosing the justification for vaccines is my favorite. “Immunization due” works for all of them. It’s versatile and comes close to two authoritative universal justifications one can only dream about: “Because I said so.” or “Because she needs it.” (Patients who decline immunizations also should have to justify their decisions.)

Billing presents its own problems with diagnoses. We want our group to prosper, but despite innumerable training sessions, most of us are bewildered by the Byzantine criteria for determining the level of service. We do what works. When my documentation is incomplete or lacks the required criteria, I receive a coding query from a billing specialist. It’s often a question about a visit I had long forgotten. A response is mandatory; the message cannot be forwarded or deleted. Often, the question cannot be answered. A few examples:

The patient has had breast, colon and skin cancer, all in remission.

Which is primary?

(“All are primary” is rejected.)

An inpatient had a BUN/creatinine of 30/1.8 mg/dl. The hemoglobin was 11.4 g/dl. The only other data point is a creatinine of 1.1 mg/dl from five years previous.

Is this AKI or CKD?

(Uncertainty is not appreciated.)

A frequent query is the categorization of malnutrition and the repletion of the second tier chemistries, magnesium and phosphate.

Treating starvation and hypophosphatemia must be a billing bonanza.

Artificial intelligence promises to erode our diagnostic autonomy. The diagnostic accuracy of large language models has equaled or outperformed clinicians in recent studies using sample cases. Eventually AI will usurp our role as diagnostician. Technology might not replace us entirely, but it could be the most important thing in medicine.

Todd Stern MD is a general internist at the University of Chicago.