In most low-resource labour wards a midwife checks the baby's heartbeat now and then by listening with a Pinard stethoscope or hand-held Doppler; how often, when relative to a contraction, and for how long must she listen to catch the heart-rate drops that warn of a baby short of oxygen?
- Good it would do
- 3 / 5
- Can AI help
- 3 / 5
- How neglected
- 4 / 5
- Can we check it
- 3 / 5
Why we're going ahead
Accepted on a narrowed question: a benchmark, on the open CTU-UHB labour recordings, of what share of labelled late decelerations each common way of intermittent listening catches (WHO single 1-minute count, 60 s after the contraction, ACNM multiple short counts, Sholapurkar 2022), framed for researchers and never as a clinical schedule. It is a weak Accept: Good done is 3 because benefit needs bedside validation and a guideline decision, the open data hold only the last hour of labour, and both outside critics warn that replayed monitor output may not reflect what a midwife hears.
Why it's stuck
The World Health Organization's 2018 intrapartum care guidance recommends intermittent listening with a Doppler or Pinard stethoscope for healthy women in labour, but rates the certainty of the evidence for that recommendation as very low (WHO evidence table, NCBI Bookshelf NBK513805). A 2015 review by Sholapurkar (Journal of Clinical Medicine Research 7(12):919-925) argues that the usual guideline method, listening for 60 seconds after a contraction every 15 to 30 minutes, may miss many late decelerations (the dangerous heart-rate drops), because many of them reach their lowest point before the contraction ends; it concludes that 'the current methodology of IA may be flawed'.
What we'd try
The CTU-UHB database on PhysioNet is open (Open Data Commons Attribution licence): 552 intrapartum cardiotocography recordings, selected from 9,164 recorded at University Hospital Brno in 2010-2012, each with the baby's heart rate, the contraction signal, and the umbilical artery blood result (pH, base excess) and Apgar score. Treat each continuous trace as the truth and replay thousands of listening schedules on it: interval (5 to 60 minutes), timing (before, during, after a contraction), listening length (15 to 60 seconds), and realistic human counting error.
Do you know the answer?
One reply from someone in the field can settle what our AI can't.
Members can read the full record for this problem, check its sources and comment. Open the member view →