The evidence for a letter is already in the record

A letter of medical necessity is an argument that a specific treatment is warranted for a specific patient under a specific policy. The facts it turns on — what was tried, what it did, what the tests showed, how long symptoms have run — are almost always already in the record. Writing one is mostly the work of finding them, arranging them against criteria somebody else wrote, and doing it at the end of a clinic day. So it is done late, or thinly, or not at all, and coverage is refused on a record that would have supported it.

What gets assembled

The draft is built from the patient's own record: diagnoses and their onset, the diagnostic results with their values, units, and reference intervals, prior therapies with dates and documented response, imaging findings, functional measures, and the treating clinician's own notes. Each of those is a retrievable object on this platform rather than text to be scraped, which is the reason the assembly is possible at all.

It is arranged against the criteria of the policy actually in force — the LCD, NCD, or commercial policy that will be applied to the request — criterion by criterion, in the order the policy states them, because a reviewer working a checklist should not have to hunt for the item they are checking.

Why it reads as this patient's letter

A template argues about a condition. This argues about a person, and the difference is the data underneath it rather than the prose on top of it.

Diagnostic results arrive here as data rather than as documents — discrete values carrying their units, their reference intervals, and the version of the report they came from, normalized across every laboratory the practice orders from through the connections that bring them in. So three results over fourteen months from three different laboratories become one trend, measured against the threshold the policy actually names, instead of three PDFs somebody has to open and retype. A drafting tool pointed at a document store cannot do that, because by the time it sees the number the number is a picture.

The history around those results is what turns a list of findings into a course: when the condition started, what was tried, in what order, for how long, and what happened each time. A reviewer is looking for the course. It is also precisely the part a template cannot supply, because it is the part that is different for every patient — and it is why two people with the same diagnosis do not warrant the same letter.

Which sets the honest limit. The letter is exactly as specific as the record connected to it. Where half a patient's history sits somewhere this platform does not reach, the draft says which parts are thin rather than writing smoothly around the gap.

Every assertion carries its source

This is the whole of it. The model is not asked what it knows about the condition; it is asked to state what this record shows and to point at where. Every clinical claim in the draft is bound to the object it came from — this result, on this date, from this laboratory; this note, by this clinician — and the citation is visible to the clinician reviewing it and retained with the letter.

An assertion that cannot be sourced does not get written. Nothing infers that a therapy failed because a later one was started, and nothing rounds a duration up to reach a threshold.

The clinician writes the letter

The draft is a draft. It is presented to the treating clinician as editable text beside its citations, it is theirs to rewrite or reject, and it is not submitted by any automated path. It leaves only when they sign it, through the same attestation the rest of the platform uses: identity verified against their own credentials, the signature bound to the exact version signed, and the whole sequence in the append-only record.

That sequence includes that a draft was machine-assembled, which model version produced it, what the clinician changed, and what they signed. A letter of medical necessity is a statement made to obtain payment. Who authored each assertion in it is not a detail.

What it does not do

It does not determine medical necessity

That is the clinician's judgment and the payer's determination. The tool assembles and cites the evidence for an argument; it does not decide that the argument is correct, and a well-cited letter for a treatment that is not indicated is a worse outcome than no letter.

It does not generate clinical facts

No symptom, duration, response, or finding appears in a draft unless it is in the record. If the record is thin the draft is thin, and it says which parts are thin rather than covering for them.

It does not promise coverage

Payers deny well-supported requests. What is claimed here is that the request reflects the record, addresses the criteria that will be applied, and is assembled in minutes rather than deferred indefinitely.

It is not a substitute for the clinician's review

A signature obtained on unread text is the failure this platform is otherwise built to prevent. The review step is not a formality and the record shows whether it happened.

Denials, and the letter's second life

Where a request is denied, the letter and its citations are already the structured basis of the appeal, and the denial reason maps back to the criterion it disputes. This is the same machinery as denials becoming retrievals: the evidence that answers the objection is identified rather than searched for, and an appeal that attaches the underlying result is a different proposition to one that restates the original assertion more loudly.

How attestation worksRevenue integrity