Who Is the Author of a Clinical Note?

A signed clinical note used to mean that a physician had read its contents and judged them to be true. That may no longer be reliably true

For the past few months I have spent a lot of my time reading clinical notes written by other physicians. The thing I kept noticing, once I had read enough of them, was how little of the average note had been written for the patient being seen that day. Most of it was already in the chart before the visit ever started, carried forward from earlier notes or filled in by the record itself, with only a line or two added fresh for the patient in the room. That is what made me start to wonder whether the physician who signs a note is still the one who wrote it.

My first instinct was to read this as something the individual physicians were doing wrong. That turned out to be unfair to them, because the pattern is far too consistent to come down to who happens to be writing. Copying the previous note forward is how documentation is produced now. When O’Donnell and colleagues surveyed physicians who wrote inpatient notes in 2008, 90% were already using copy and paste, and 70% used it almost always or most of the time when writing a daily note. Wrenn and colleagues found that more than half the text in a typical note had been carried over from earlier ones. The most careful look came from Wang and colleagues. They went through 23,630 hospital progress notes and traced where every word in them had come from. The physician had typed only about 18% of the note herself. Nearly half had been copied from earlier notes, and the record had filled in the rest on its own. This is worst in the hospital, where the same note is rewritten on the same patient day after day, and gentler in the clinic. But it happens everywhere, because the moment a note can be assembled from text that already exists, it will be.

I want to be fair to these physicians, because they are working in a position that leaves them almost no room to do anything else. A single note now has to satisfy the billing rules, hold up if a malpractice lawyer ever reads it, and still be useful to whoever cares for the patient next. The physician has a few minutes to produce something that does all three. The software makes it possible to reproduce the previous note with a single click. A note written under those conditions is what you would expect, and it tells you nothing about how much the physician cares about the patient.

So I had been blaming the wrong thing. Once I saw that, I had to go back and question the rest of what I was assuming. I had been treating copied text as proof that no thought had gone into it, and that is not right either. Almost none of that carried-forward material was invented. When the patient’s history was first written down, a physician had sat with the patient and worked it out. The medications on the list were prescribed because someone judged them necessary. Almost all of it reflected someone’s judgment at the moment it first entered the record. A physician can pull that old paragraph forward today, read it carefully, decide it still describes the patient, and leave it in place. When she does, the copied paragraph reflects her judgment as much as anything she typed from nothing. The copying, on its own, is not the problem. That the original work was genuine is exactly what hides the problem, because it means the note is full of content that was once true and reasoned, with nothing to show whether anyone has looked at it since.

The trouble is what happens when the two kinds of copied text end up next to each other. The paragraph the physician read and confirmed today, and the paragraph she carried forward without reading, sit in the same note, in the same words, under the same signature. Nothing sets one apart from the other. A finding she checked this morning looks exactly like a finding that has been sitting untouched in the chart for a year. Reading the note afterward, I cannot tell which is which, and neither can anyone who reads it later.

This is what made me stop asking who wrote the note and start asking what the signature on it still stands for. A physician’s signature has always meant two things at once. It means she is the one accountable for what the note says. It also means she looked at the patient’s situation and judged the contents to be true. The first meaning is intact, and you can always tell whose name is on a note and who answers for it. The second is the one that has weakened. The signature now covers the lines that were reviewed for this visit and the lines that were carried forward without review, with no distinction between them, so it no longer tells you that any particular statement in the note was checked against this patient that day.

None of this would matter much if the note stayed where it was written, but a note never does. It is the first place the next physician looks to find out who the patient is. Someone sees a new physician, or returns to her own after several months, and that physician builds a picture of her almost entirely from the notes already in the chart, reading every line as a true account of her. She has no way to separate the lines someone confirmed from the lines someone left in place. She gives them all the same weight, which is what the note asks her to do.

The other thing I came to see is harder to notice. When a line gets carried from note to note, it does not only survive in the record. It multiplies through it. One observation, made a single time, can end up written into the chart ten times over. A physician who sees the same finding repeated across a year of notes reads the repetition as agreement, as though ten different people had each looked and reached the same conclusion. In a chart built by copying, those ten appearances can all come from one, carried forward nine times by people who never looked at it again. The repetition has added no new information, and it looks identical to the kind of repetition we treat as confirmation. A claim that was weak to begin with comes to look solid, for no reason other than how often it appears.

This is also where the systems now being built to read the medical record begin to fail. A great deal of what medicine is trying to do depends on software that can take in a patient’s whole history and reason over it, and software like that can only ever be as reliable as the record it reads. It has the first problem immediately, because it cannot tell a reviewed line from an unreviewed one any better than a person can. The second problem is worse for the software than it is for us. A system that decides how much to trust a finding by counting how often it appears will read those ten copied lines as ten separate confirmations. The more often a wrong statement has been copied, the more certain the software becomes that it is right.

There is evidence that the worst version of this harms patients. When an error gets carried forward without anyone checking it against the patient, it passes into the next note as established fact, and a later physician acts on something that was never true. In one study of diagnostic errors in primary care, Singh and colleagues found that copying and pasting had played a part in roughly 2.6% of the errors they examined. It is a small share, and it is enough to show that the mechanism is real.

I do not want any of this taken as an argument against reusing what is already in the chart, because that would make no sense. A laboratory value the physician has read and confirmed is far better carried into the next note than typed in again from memory. A good note is, in large part, a careful reassembly of what is already known about the patient. The reuse was never the problem. The problem is that reusing something after you have checked it, and reusing it without checking it at all, look exactly the same on the page, so the one thing a reader most needs to know about any given line, whether anyone has looked at it recently, is the one thing the note does not record.

A clinical note used to be a record of the decisions a physician had made and still stood behind for a particular patient on a particular day. It is still read that way, by the next physician and, more and more, by the machines we are teaching to read it. Whether the decision behind any single line was ever made, or was made once and never looked at again, is something the note can no longer tell you. Everything that treats the note as a record of present judgment now depends on an assumption that has quietly stopped being safe to make.