11 SEP 2026 — A study published in the BMJ on 10 September ranks the 24 prescribing cascades that happen most often in a real population of older adults. A prescribing cascade is what it sounds like: a drug causes a side effect, the side effect is read as a new condition, and a second drug is prescribed to treat it. The pharmacology here was already known. What was missing was which of the known chains actually occur.
The clearest example
Non-steroidal anti-inflammatory drugs raise blood pressure in some people. A patient taking an NSAID for joint pain returns with a higher reading, and leaves with an antihypertensive.
Nothing in that sequence involves an error anybody could point to at the time. The blood pressure reading is real. Treating raised blood pressure is correct medicine. The problem is invisible from inside the consultation. The cause was in the last prescription, and the patient now takes two drugs where one, or none, would have done.
Statins and iron supplements are among the other triggers the researchers name.
What is new, and what is not
The concept is not new, and the authors do not claim otherwise. An international expert panel — 12 specialists in internal medicine, geriatric medicine and clinical pharmacology — had already assembled a list of 65 potentially inappropriate prescribing cascades. A related tool, ThinkCascades, was published in 2022.
The contribution here is prevalence. Using population-level prescription data from ICES, Ontario's health data institute, the researchers tested each of the 65 against three things: how common the first drug is in the population, how often the second drug follows it, and how strong the association between the two is. That reduced 65 candidates to 24 that matter at population scale.
Turning a list of things that can happen into a ranked list of things that do is undervalued. A clinician cannot hold 65 patterns in mind during a ten-minute appointment; 24 ordered by frequency is short enough to become a habit.
Potentially is the word doing the work
The study calls these potentially inappropriate prescribing cascades, and the qualifier marks the honest boundary of what the method can see.
Administrative prescription data shows that drug B was dispensed after drug A. It does not show why. It cannot see the consultation, the clinician's reasoning, or the possibility that the patient's blood pressure was going to rise anyway and the NSAID is a coincidence in the timeline. Some proportion of every one of these 24 sequences is correct medicine, arrived at deliberately, and the data cannot separate those cases from the cascades.
Read the finding accordingly. It identifies sequences worth a second look, at a rate that makes looking worthwhile, rather than errors. Any account of this research that says patients are being given unnecessary drugs 24 different ways has overstated what the authors claimed.
Why nobody catches it
The structural reason is that prescribing and reviewing are usually done by different people at different times.
The drug that starts a cascade may have been prescribed years earlier, by a different clinician, in a different setting, for a reason nobody in the current room was present for. The new symptom looks like a new problem because that is how it shows up. A cascade is visible only to someone holding the whole medication list and the dates it was assembled, and that view is rarer than it should be.
This is also why the effect compounds with age. Every additional long-term medication adds another candidate first link, and the list of things a new symptom might be a side effect of grows faster than the list itself.
What this changes for a patient
The most useful piece of information is not on the label. It is in the history: the date.
A cascade becomes detectable when someone knows when each medication was started, and what it was started for. The clinician can then ask whether a new symptom appeared after a recent change. A medication list without dates cannot answer that. Keeping the record is something a patient or a family member can do without any clinical knowledge, and it is the input the system most often lacks.
The researchers point to closer medication reviews, pharmacist involvement and automated alerts as the mechanisms most likely to help. The first two of those are easier to ask for than they seem.
This research is not a reason to stop a medication. Stopping a drug that is working is a direct harm, and a larger one than anything described here. The cascades in this study are identified precisely because the individual prescribing decisions look reasonable. The finding supports a conversation with a prescriber, not a decision taken without one.
What to watch
Whether the ranked list gets into the software. A finding about which sequences are common is exactly the kind that belongs in a prescribing system rather than in a journal — an alert that fires when drug B is prescribed to a patient who started drug A three weeks ago is a mechanical implementation of the paper's entire contribution.
The obstacle is well documented and not technical. Prescribing systems already generate more alerts than clinicians can attend to; a new alert competes for an attention budget that is already overdrawn. A ranked list is useful in that context for the same reason it is useful in a consultation: it says which 24 out of 65 are worth interrupting someone for.
View author profile → · Editorial policy
