An algorithm flags prescribing cascades from product information

Design

Algorithm development and internal validation

Population

69 literature-supported cascade pairs and 138 random drug pairs

Primary outcome

Sensitivity and specificity for identifying prescribing cascades

Effect

Best model: sensitivity 85.5%, specificity 99% or more, PPV 98.3%, accuracy 94.7%

A prescribing cascade happens when an adverse effect of one drug is treated with another: amlodipine oedema treated with a diuretic, or metoclopramide-induced parkinsonism treated with levodopa. Researchers in Italy built an algorithm, part of a medication review tool, that compares the adverse reactions listed in one drug's product information with the indications of another, using standard regulatory terminology.

Against 69 known cascade pairs and 138 random pairs, the best version found 85.5% of true cascades with over 99% specificity, a positive predictive value of 98.3% and accuracy of 94.7%. Broadening the matching from exact terms to related groups of terms raised sensitivity from 51% to 86% without adding false alarms.

This is internal validation on a small reference set, not evidence that using the tool improves patient outcomes. But it shows that cascades can be detected systematically from data every prescriber already has, rather than relying on memory during a busy medication review.

• When a new drug is added, ask whether it treats a symptom that an existing drug can cause.
• Common cascades include calcium channel blocker oedema treated with diuretics and antipsychotic or antiemetic parkinsonism treated with levodopa.
• Structured medication review tools may help flag these pairs in patients on many drugs.
• Deprescribing the first drug is often better than adding the second.

Why it matters

Prescribing cascades drive polypharmacy quietly, and a tool built on product information could make them visible at scale.

Don't overread it

Internally validated against 69 known pairs; it has not been shown to change prescribing or outcomes.

The statistics, in plain English

Positive predictive value of 98.3% means almost every pair the tool flagged was a real cascade in the reference set. Because the test set was built with one true pair for every two false ones, predictive values in real prescribing data, where true cascades are rarer, would be lower.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.