Fragmented prescribing

Why can every prescriber be individually right and the patient still end up on a dangerous list of drugs?

Picture an eighty-two-year-old on eleven medicines. It is tempting to look for the culprit: a careless doctor, a lazy pharmacist, someone who did not read the notes. But go and audit each prescription against the reason it was written, and you will usually find that every single one was defensible. The cardiologist's drug was right for the heart. The rheumatologist's was right for the joints. The hospital team's was right for the infection that put her in the ward.

So the puzzle is not incompetence. It is this: how does a set of individually correct decisions add up to a harmful one, with no step at which anyone was wrong?

Start with what each prescriber can actually see. A specialist is asked a bounded question — manage the rhythm, manage the pain — and answers it well. What they hold is their own problem, their own drug, and whatever list the patient or the referral letter happened to supply. What they do not hold is the whole.

Now ask the awkward follow-up: whose job is the whole? Say it out loud and the answer is uncomfortable. The general practitioner is nominally the coordinator, but she did not start most of these drugs, is reading a specialist's reasoning second-hand, and stopping someone else's medicine carries a strange asymmetry — if she leaves it and harm follows, that is the system's fault; if she stops it and harm follows, it is hers. So the whole list has no owner with both the information and the authority to act on it.

Notice what kind of failure that is. Nobody made an error. The decomposition made the error: the problem was cut into pieces along organ boundaries, each piece was solved correctly, and the pieces do not recombine into a solution because the harm lives in the interactions between them, which is exactly the part no piece contains.

Then consider how the list grows, because it is not linear. Each new drug can interact with every drug already there, so the number of pairs rises roughly as the square of the count — ten drugs give forty-five pairs. Most pairs are harmless. But the reviewer's workload grows quadratically while the reviewing time stays fifteen minutes.

There is a nastier mechanism too, and it is the one that turns fragmentation into a ratchet. A drug produces a side effect. The side effect presents to a different clinician, who has no reason to read it as a drug effect and every reason to read it as a new disease — so it gets its own drug. This is the prescribing cascade, described by Rochon and Gurwitz, and the textbook examples are almost comic once you see them: a calcium channel blocker causes ankle swelling, which is treated as fluid overload with a diuretic; a drug for nausea causes tremor and stiffness, which is treated as Parkinson's disease. The cascade only runs because the two encounters are separate. One clinician seeing both events in sequence would probably catch it.

And some harms are not attributable to any single drug at all. Anticholinergic effects are the clearest case: several ordinary medicines each carry a mild anticholinergic action, and the concern is the sum — higher cumulative burden is associated with confusion and falls in older people, though how much of that association is causal is still argued. There is no offending prescription to find. Each contribution was small and reasonable. Only the total is dangerous, and the total is precisely the quantity nobody computes.

So what would a fix have to do? At minimum, reconstruct the shared object that everyone has been writing to blind. That is medication reconciliation: at every transition of care — admission, transfer, discharge — somebody sits down, compares the patient's own account with the pharmacy record and the referral list, and produces one verified list, with each discrepancy resolved deliberately rather than by default. It is a formal safety requirement in many systems, including a Joint Commission national patient safety goal.

Does it work? Honestly: partly, and it matters which part. Reconciliation reliably finds discrepancies — studies of hospital admissions routinely find unintended differences in a large fraction of patients — and it reliably reduces them. Whether that translates into fewer readmissions or deaths is much less settled, and trials have been mixed.

Why might that be? Because reconciliation fixes the record, not the authority. It answers "what is she actually taking?" It does not answer "should she be taking it?", and it does not create anyone empowered to say no on behalf of the whole patient. That second problem needs different tools — explicit deprescribing criteria such as Beers or STOPP/START, structured medication review, someone whose job is the list itself. An accurate list of eleven drugs is still eleven drugs.

Two corrections are worth making explicitly.

First, this is not an argument that polypharmacy is bad. Many older people genuinely need many drugs, and under-treatment is a real harm too. The claim is narrower: that beyond a certain count, nobody is evaluating the list as a list, so its quality drifts free of anyone's judgement.

Second, it is tempting to read the problem as missing information — if only every prescriber could see the full record, this would go away. Shared records help, and they help most with the cascade, since a visible start date next to a new symptom is often the whole diagnosis. But information is necessary and not sufficient. A prescriber who can see all eleven drugs, has fifteen minutes, is answering a different question, and would be second-guessing four colleagues, will usually still change nothing. Coordination failures are about decision rights as much as about data.

• Deprescribing: what makes stopping a drug so much harder than starting one.
• Whether the same shape explains other multi-specialist failures, such as conflicting advice after complex surgery.
• The reverse risk: programmes that strip out drugs the patient genuinely needed.

Every term the collection defines is gathered in the glossary.