There is a journey most people never think about. A doctor types a few lines after your appointment, and eventually money moves from the United States Treasury to an insurance company. Between those two events sits a pipeline with several stages, each one capable of introducing error, and this year that pipeline has been under the most intense examination in its history.
Following it end to end is genuinely interesting, and it explains a great deal about a story that has been in the news without ever being properly explained.
Stage one: the note
Everything begins with what the clinician writes. A visit produces a note, usually composed in minutes, in shorthand, under time pressure. That note is the only durable record of what happened in the room.
Its quality varies enormously, and the variation matters more than most people realise. A note saying a patient “has diabetes” asserts something. A note recording that the clinician reviewed the patient’s blood sugar readings, adjusted the medication, and scheduled a follow-up demonstrates something. Downstream, only the second kind holds up.
Stage two: the translation
Next, someone converts that prose into codes. There is a standardised catalogue of more than seventy thousand entries covering every condition medicine recognises, refreshed annually, and a trained coder decides which entries the note genuinely supports.
This is harder than it sounds. Clinical shorthand is ambiguous, abbreviations mean different things in different specialties, and the coder must apply a strict standard: is there documented evidence that this condition was actively monitored, evaluated, assessed, or treated during this specific visit? A condition merely mentioned in passing does not qualify.
Stage three: the score
The codes then feed a formula. In Medicare Advantage, where private insurers cover more than thirty million older Americans, each member’s documented conditions produce a risk score, and the government pays the insurer a monthly amount scaled to that score. Someone with several serious chronic illnesses generates a higher payment than someone in good health, which is the intended design: caring for sicker people costs more.
Where the pipeline broke
Notice the vulnerability. Payment depends on stage one and stage two, both of which are performed by parties who benefit from the outcome. For years, insurers invested heavily in re-reading old charts to find additional codeable conditions, and there was no equivalent investment in finding conditions that should be removed.
The results became public this spring. Federal auditors examining three insurance plans found that 81 to 91 percent of certain sampled high-risk diagnosis codes lacked adequate support in the underlying records. A major Medicare Advantage insurer paid 117.7 million dollars to settle federal claims about its chart review programme. The government has since scaled its audit workforce to roughly two thousand certified coders running rolling quarterly reviews, with sample error rates applied across entire contracts.
The stage that changed
The most interesting repair is happening at stage two, where software has increasingly taken over the first pass of reading.
Older systems were built to search. Point them at a patient’s history and they surfaced conditions that might be codeable, which was useful and also, in hindsight, one-directional by design. The current generation is built to justify. Modern ai tools for medicare risk adjustment coding accuracy attach evidence to every suggestion: the exact sentence supporting the condition, the documentation rule it satisfies, and a record of which qualified human confirmed it. Just as importantly, they flag recorded conditions the evidence cannot support, so the review runs in both directions.
The machines read. Humans still decide, and the decision is now recorded as part of the permanent trail, because the person who submits the code is the one who answers for it.
Why the pipeline is worth understanding
Because versions of it are everywhere. Wherever public money follows records that an interested party produces, the same three stages exist: an observation, a translation, and a payment. Education funding, disability support, agricultural subsidies, carbon reporting. Each has its own note, its own coder, its own formula.
The American healthcare version simply ran at the largest scale, for the longest time, with the least checking, and therefore produced the clearest demonstration of what happens next. The pipeline does not fail dramatically. It drifts, quietly, in the direction of whoever benefits, until somebody with authority follows it end to end and asks, at every stage, where the evidence is.
You may also like
Leave a Reply