She Was Not Non-Compliant, She Had Two Correct Instructions

The discharge packet was six pages. Page two said, in a box, HOLD warfarin until INR checked. Page five, in the medication list generated by a different system, said warfarin 5 mg, take one tablet daily, continue. The pharmacy label on the bottle in her bag, printed three weeks earlier, also said take one tablet daily.

She took one tablet daily. She was readmitted eleven days later.

In the chart, the readmission is coded with a note that reads, in part, patient non-adherent to discharge instructions. I have read that note perhaps thirty times. It is not a lie. She was, in the strict sense, non-adherent to one of the two instructions she was given, and there was no way on the paper in front of her to tell which one was newer.

I want to make a claim that is narrower than the one usually made in this territory, because the broad version — the system fails patients — is true, unfalsifiable and useless. The narrow version is this: non-compliance is not an observation about a patient. It is a conclusion we reach when the instruction was defective and we have no field in which to record that.

Consider what has to be true for the word to be accurate. A patient is non-compliant when they understood what was asked, were able to do it, and did not. That is a real category and it is not empty; I have patients who understand the plan perfectly, can afford it, and do not follow it, and some of them have reasons that are none of my business. Non-compliance in that sense is worth naming because it changes what you do next.

Now look at what it actually gets applied to. It is applied when the instruction was ambiguous. It is applied when two documents disagreed. It is applied when the medication was not affordable in the second week, when the transport to the follow-up did not exist, when the summary was written at a reading level she does not read at, and — this is the one I see most — when the patient did exactly what she was told by the most recent thing that spoke to her in a voice she trusted, which was a pharmacy label, because a pharmacy label is a short authoritative sentence and a six-page packet is not.

Every one of those is a different problem with a different fix, and all of them arrive in the record wearing the same word.

Here is why the word rather than the underlying event, because I can hear the objection that this is semantics. It is not semantics; it is where the causal arrow points. Non-compliance locates the failure inside the patient. Once it is there, the intervention that follows is aimed at the patient — education, counselling, a phone call reminding her to take her medication, in some systems a note that follows her and changes how the next clinician reads her. Nothing at all is aimed at the document, which is where the defect was, and which will produce the identical contradiction for the next patient discharged on warfarin that afternoon.

I know this because I went and looked. Our discharge summary pulls the medication list from the active-orders table and the free-text instructions from the physician's note, and those two are not reconciled by anything except a human reading both. There is no check. There has never been a check. Nobody chose that; it is what happens when two systems are procured six years apart and interfaced by exporting one into a template.

Now the objection I take most seriously, because I have been on the other side of it.

The people writing these summaries have twelve minutes. Asking for a second, patient-facing document — plain language, one page, the three things that changed — sounds obviously right and has been tried repeatedly, and the honest record of those trials is mixed. A second document is a third place for the instruction to live, and a third place is a third opportunity to disagree with the other two. There are published cases of patient-facing summaries introducing errors that the clinical summary did not have. Anyone who has not sat with that finding is not being serious.

I accept it entirely, and it does not touch the claim, because I am not arguing for a second document. I am arguing that the contradiction between the two we already have is detectable by a computer in about four milliseconds, and that nobody has asked for it, and that the reason nobody has asked for it is that the resulting readmissions are already fully explained by a word.

There is a second case, and it is the one that turned this from an irritation into something I think about at night.

Readmission penalties. Our hospital, like most American hospitals of its size, is financially penalised for readmissions inside thirty days for certain conditions. That policy was designed to make institutions care about what happens after discharge, and to a real extent it worked — I have watched follow-up call programmes and transport vouchers appear that would never have been funded on clinical argument alone.

But a penalty creates a demand for explanations, and the cheapest available explanation for a readmission is that the patient did not do what they were told. That is not a conspiracy and no one in my building is falsifying anything. It is that when a case is reviewed and the reviewer sees a note saying non-adherent, the review stops there, because the box has an answer in it. I have sat in those reviews. The question asked is what happened, and non-adherent is a complete-sounding answer to what happened, and it takes an additional forty minutes and somebody's willingness to be annoying to discover that page two and page five disagreed.

So the policy that was meant to make the institution responsible for the transition has, through the ordinary operation of people trying to close cases, generated an industrial demand for the exact word that moves responsibility back out of the institution. I do not think that was foreseeable and I do not think it makes the policy wrong. I think it means the policy needed a companion field and does not have one.

That is my whole proposal and it is smaller than the essay: a required structured field on any readmission review, before non-adherence can be selected, asking whether the discharge documentation contained conflicting medication instructions — answerable by the computer, not the reviewer. Not a plan, not a programme. A field that has to be false before the convenient word is available.

I want to be careful about what I am claiming for it. I do not know the size of the effect. My belief that these contradictions are common rests on our own chart audit, which covered one service line over four months and found the medication list and the free-text instructions in disagreement in a number I would not publish without a statistician standing next to me. It was not a small number. It was large enough that I stopped being able to read the word in a note without checking.

She is fine. She came back, the bleed was managed, she went home with a one-page sheet that I wrote by hand because I did not trust the packet, which is exactly the artisanal workaround that lets a systemic defect survive another decade. Somebody discharged on warfarin from our floor this afternoon got the six pages.