Case Study 1 — The "With" Convention Meets Risk Adjustment

Real guideline history and a real, documented enforcement area. Tier 1 for the convention's history and for the existence of the enforcement activity; qualitative throughout for magnitudes.

Background

Section 9.7 presented the "with" convention as a resolution: the coder is not inferring anything, the classification is presuming, and the presumption exists only where the Index or a Tabular instruction creates it.

That is the correct reading and it took several years and several revisions to settle, because the convention sits at the intersection of two things that push hard against each other — a coding rule that permits linkage without documentation, and a payment model in which linkage is worth money.

This case study is about what happened at that intersection.

The convention's own history

The "with" guidance has been revised and clarified more than once since ICD-10-CM took effect, and the sequence matters because it shows the Cooperating Parties responding to exactly the problem this case study is about.

The original text established the presumption: conditions linked by "with" in the Index or Tabular are coded as related even without explicit provider documentation.

Coders immediately asked the obvious question: how far does this go? If the classification presumes diabetes and chronic kidney disease are related, does it presume anything a clinician might plausibly connect?

Subsequent revisions narrowed and clarified it. The guidance was tightened to be explicit that the presumption applies where the terms appear in the classification — a code title, the Alphabetic Index, or a Tabular instructional note — and not wherever the word appears in a clinical note, and that documentation stating the conditions are unrelated defeats it.

The direction of travel is the finding. Every revision made the convention narrower and more explicit about where it applies. That is the Cooperating Parties responding to a rule being stretched.

Why it was being stretched

Because of Chapter 36's subject, arriving early.

Under risk-adjusted payment — Medicare Advantage, and increasingly other arrangements — a plan is paid a per-member amount based substantially on the diagnoses documented and reported for that member. A diabetes code with a complication and a diabetes code without one are not equivalent for this purpose; they describe different expected costs, and they are worth different amounts.

Which creates a specific and predictable pressure: a convention permitting linkage without documentation is worth money precisely to the extent it can be read broadly.

The stretched reading goes like this. The classification presumes diabetes is related to certain conditions. A clinician would find many other relationships plausible. Therefore — the argument runs — a coder may link them.

That reasoning is wrong at the first step, and §9.7 says why: the presumption is not a statement about clinical plausibility. It is an instruction attached to specific pairings that the classification has already decided about. Extending it to pairings the classification has not made is the coder supplying a clinical judgment, which is Chapter 4 §4.7's prohibition.

What the enforcement record shows

Risk-adjustment coding has been an active area of federal audit and enforcement attention for years. The OIG has published a substantial body of audit work examining whether diagnoses submitted for risk-adjustment purposes were supported by the medical records, and the Department of Justice has pursued matters involving risk-adjustment submissions. CMS operates Risk Adjustment Data Validation audits, which sample enrollees and require the medical record to support each submitted diagnosis.

The recurring finding across that body of work is not exotic. It is that diagnoses were submitted that the medical record did not support — through chart review programs that surfaced codes without adequate documentation, through addenda produced specifically to support codes, and through the inference of relationships the record did not establish.

The last one is this case study.

This book does not characterize any particular organization's conduct, and the enforcement record contains matters that were settled without admission, matters that were litigated, and matters that found no wrongdoing. What is not in dispute is that this is an active audit area and that the documentation standard applied is the one Chapter 4 §4.7 states.

What it shows

First, it demonstrates that a coding convention and a payment incentive can point in different directions, and that the convention is not the thing that moves. The "with" presumption is a classification rule. It was written to solve a classification problem — that clinicians document relationships inconsistently and the classification needs a default. It was not written to allocate payment, and reading it as though it were is a category error.

Second, it explains why §9.7's boundaries are stated so carefully. A rule that says "presume a relationship" is genuinely useful and genuinely dangerous, and the difference between those is entirely in the boundary. This book states the boundary three times in one section for that reason.

Third, it shows the Cooperating Parties doing maintenance in response to use. As in Chapter 8's Case Study 1, the guidance changed because practice revealed a problem with it. Read the current version. A description of the "with" convention written at any point since 2015 may describe a version that has since been narrowed, and that includes §9.7.

Fourth, and most practically: the safe formulation is a question, not a rule.

"Did the classification link these two, or did I?"

If a coder can point to the Index entry or the Tabular instruction, the presumption applies. If the coder is reasoning from clinical plausibility, it does not — and no amount of clinical plausibility converts one into the other.

The lesson

A presumption that lives in the classification is not a presumption you are authorized to extend.

Three carry-forwards:

Locate the linkage before you rely on it. Not "these are obviously related" — where does the classification say so? An Index entry or a Tabular instruction. If you cannot point at one, you do not have the convention.

Read the current guidance, not a summary. This particular convention has been revised repeatedly and its wording is the whole of it. Section I.A.15 of the current Guidelines.

And be especially careful where the answer is worth money. That is not a moral instruction; it is a practical one. The places where a coding rule is stretched are, without exception, the places where stretching it pays — and those are also the places that get audited. A convention applied identically whether or not it pays is a convention you can defend.

Discussion questions

1. §9.7 resolves an apparent contradiction by saying the coder is not inferring, the classification is presuming. Is that a real distinction or a formal one? Defend your answer with reference to who is accountable for the code.

2. The convention was revised repeatedly toward narrowness. What does that suggest about how a rule permitting inference should be written in the first place — and is a narrower rule always better?

3. Construct the strongest version of the stretched argument — that clinical plausibility should suffice. Then say precisely where it fails, and whether its failure is legal, logical, or both.

4. The safe formulation is "did the classification link these two, or did I?" Design the workflow control that makes a coder actually ask it. Where would it sit, and what would it look like?

5. Compare this case study with Chapter 5's Figure 5.1 (the billing macro). Both are about a rule applied at volume without a person checking each instance. What is different about the two, and which is harder to prevent?