The Administrative Twin: When Billing Data Starts Making Clinical Decisions
Every patient in a modern hospital has a double: a digital profile assembled to fit the reimbursement rules. Our safety systems increasingly consult that double rather than the person in the bed. When the codes are wrong, the consequence is no longer just an audit risk.
Dr. Sven Jungmann
CEO
A resident admits a patient at three in the morning. The history is messy, the ward is full, and the admission form will not save until an infection diagnosis is chosen. She picks the first plausible entry from the list — an unspecified infection — and moves on to the next bed. It is a small act of triage, the kind every clinician commits a dozen times a night. Weeks later, the coding team will quietly correct it for the invoice. By then it has already done its damage, and no one will connect the two events.
We tend to file medical coding under finance. It exists, we assume, to satisfy the DRG system and make sure the hospital is paid; to the senior physician it is an annoyance, to the controller a lever. The quiet assumption underneath is that a wrong or generously chosen code costs us nothing worse than an audit. That assumption belonged to the paper era. In a digital hospital, the code is not a description of the decision. It increasingly is the decision.
“A coding error at admission becomes a billing error at discharge. Somewhere in between, it can quietly become a treatment error.”
Every patient now has a double
When we shape a patient's data to fit the reimbursement rules, we create a second patient alongside the first. There is the biological patient: the complicated reality in the bed. And there is what I think of as the administrative twin: the digital profile assembled to satisfy the billing logic. For years these two could drift apart with no harm done, because nothing automated was reading the twin.
That has changed. Clinical decision support systems (CDSS), interaction checkers and the first wave of predictive tools do not examine the person in the bed. They cannot. They examine the structured record — the twin. And the twin was never assembled to be clinically true. It was assembled to be billable.
How the error reaches the bedside
Return to the resident's unspecified infection code. The hospital's sepsis surveillance runs on specific coded diagnoses; it scans for them to raise an early-warning score. Faced with a generic entry, it stays silent. The safety net has a hole in it — not through any clinical incompetence, but through a moment of data expediency at three in the morning. The patient is sicker than the record claims, and the system built to catch exactly this looks straight past them.
The reverse failure is just as instructive. We encourage thorough documentation of comorbidities, because a fuller picture raises the case mix and the case mix sets the reimbursement. Code a transient dip in kidney function generously as renal insufficiency, and the money follows. But months later a prescribing or imaging tool reads that same code, concludes the kidneys are failing, and flags a needed contrast scan as unsafe. The optimisation that helped the invoice now obstructs the care.
Training tomorrow's models on yesterday's fiction
The stakes rise the moment we start training models on our own histories — and many hospitals are eager to. If a decade of records has been shaped by reimbursement, with severity quietly inflated where it paid and complications quietly understated where they were penalised, then that is the world the model learns. It does not learn human physiology. It learns the logic of the insurance contract, and it will reproduce that logic confidently, at scale, on patients who were never party to it.
Dirty data is dirty instruments
The remedy is not another layer of coding rules. It is a change in how we regard the data in the first place. A surgeon who operates with an unclean instrument has committed malpractice; we do not debate it. A clinician who reasons — or lets a system reason — from data known to be distorted is doing something closer to that than we like to admit, and we still file it under accounting.
So the audit a medical director owes the institution is no longer only the financial one. The familiar question is whether the coding lost us money. The newer and more uncomfortable question is whether it misrepresented the patient — because the administrative twin is no longer a shadow on a spreadsheet. It is increasingly the version of the patient that our machines actually treat. Keeping the two in alignment has quietly become part of clinical safety, and there is no longer a clean line between data integrity and the wellbeing of the person in the bed.
