- Design
- Secondary analysis of a randomised trial using NLP on counselling notes
- Population
- 76 adults treated for MDR-TB and HIV in South Africa
- Primary outcome
- Electronically monitored bedaquiline and ART adherence
- Effect
- Bedaquiline adherence fell mid-interval (β −0.048, p=0.001); ART fell mid-interval and before next session
Researchers applied natural language processing to 327 routine counselling notes from 76 people in a South African trial of combined MDR-TB and HIV treatment. Adherence to bedaquiline and antiretrovirals was measured with electronic dose monitors.
The themes in the notes, such as substance use, logistics, social support and money, did not predict adherence. Timing did. Adherence was highest just after a counselling session. Bedaquiline adherence fell in the middle period between sessions (β −0.048, p=0.001), as did antiretroviral adherence (β −0.031), and antiretroviral adherence fell further before the next visit.
The sample is small and the model fit modest, so this generates a hypothesis rather than a rule. But the pattern is plausible and cheap to act on: the gap between visits is when support is thinnest.
- Expect adherence to fade between counselling visits, not just at the start of treatment
- Consider phone or message check-ins between scheduled sessions for MDR-TB and HIV patients
- Use dose monitoring or pill counts to spot mid-interval drops early
- Address practical barriers such as transport and money at every visit
Why it matters
It suggests when adherence support is needed most, which matters for a regimen where missed doses breed resistance.
Don't overread it
This is an exploratory analysis of 76 patients; it does not show that more frequent counselling improves cure rates.
The statistics, in plain English
The β values are small daily-adherence changes, but they compound over a long regimen. With 76 people and modest model fit, the finding is a signal to test, not a measured effect of more frequent counselling.
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.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for infectious diseases, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free