There is an old rule in quality management: the cost of fixing a defect rises by roughly an order of magnitude at every stage it survives. Cheap at the workstation, expensive at final inspection, ruinous after it ships.
Manufacturing internalised this decades ago and moved inspection into the line. Software followed with continuous integration, because nobody wanted to find a broken build three weeks after the commit.
Healthcare documentation is running about forty years behind, and the bill for it is now public.
The end-of-line model
In US healthcare, diagnoses recorded in clinical notes determine how insurers get paid for covering a patient. Sicker patients mean higher payments, which is deliberate design, because otherwise insurers would avoid expensive people.
The workflow built around that is almost pure end-of-line inspection. A clinician sees a patient and types a short note under time pressure. Weeks or months later, software reads that note looking for conditions that should have been recorded and were not. Corrections get made long after the encounter, by people who were not there.
That is the same shape as inspecting finished goods at the loading dock. You catch some defects. You catch them at the most expensive possible moment. And you have no way to catch the ones the record does not reveal.
What the defect rate looks like
The US Office of Inspector General has audited this, and it published its detection method before it published the results.
In December 2023 it released a toolkit describing how it identifies diagnosis codes at high risk of lacking support, including the database queries. Across its audit work it reports approximately 70 percent of those codes were not supported by the associated medical records, with some categories exceeding 90 percent.
One audit sampled 97 records where an acute stroke had been submitted. None were supported as acute strokes. In 68 of those cases the patient had genuinely had a stroke, recorded in their history, and the submitted diagnosis described it as active.
Read that as a process metric rather than a scandal. Nothing was fabricated. The information needed to get it right existed at the point of care and was not captured in a form that survived the journey.
Why post-process inspection cannot close the gap
Here is the part that maps directly onto anything you run.
The software reviewing those charts predicts. It estimates which diagnoses probably apply and ranks them by confidence. That architecture finds candidates well and has a specific blind spot: it cannot determine that something already recorded is wrong. A ranked list has no entry meaning “remove this.”
So the inspection step only ever adds. Run it repeatedly and you never get a correction, not because anyone configured it badly, but because there is no output slot for one.
Anyone who has run end-of-line QA recognises this. An inspection stage that can only flag missing components and never flag wrong ones is not a quality function. It is a completeness check wearing a quality badge.
Moving the checkpoint upstream
The correction is the same one made. Move the check to where the work happens. In this sector that is called concurrent coding, reviewing documentation during the episode of care rather than after the claim has gone out.
The clinician is still available, the patient is still in front of them, and a missing detail costs ten seconds to add rather than a retrospective review to reconstruct.
The economics are the ones you already know. Catching it in-process is cheap. Catching it at inspection is expensive. Having a regulator catch it afterwards is the third column, and that is the one currently being priced.
The generalisable version
Three questions worth asking about any quality process in your own operation.
- Where does the check happen, and how far is that from where the work is done? Every stage between the two adds cost and loses context.
- Can your check fail something, or only flag what is missing? Those are different functions and only one is quality control.
- Who has the information needed to fix a defect, and are they still available when it gets found? If not, you are reconstructing rather than correcting.
Healthcare is answering these now because a regulator published the test and then ran it. Most industries get to answer them before that happens.
