How to guess fetal blood pressure

Physiologists tell us there are limits to the ability of fetal heart rate monitoring (of any type) to detect the difference between a fetus with hypoxia who has made adequate adjustments and is coping, and one who is unable to make adequate adjustments and is experiencing damage in important tissues (read more here). One of those important tissues is the muscle of the heart. Low oxygen levels reduce the strength of cardiac contractions and blood pressure falls. So having a way to measure fetal blood pressure would be handy. Popping a wee blood pressure cuff on the fetus on isn’t an option though.

John Tolladay, a member of Christopher Lear’s fetal physiology research team, has recently published a paper about whether it is possible to use artificial intelligence to find a way to estimate what is happening with fetal blood pressure using heart rate patterns (2023). (I know – clever idea, right!) Their paper is free to read so you might like to check it out yourself.

How did they test the idea?

Data from previous studies of fetal sheep were used. In these previous experiments, sensors were placed that could measure not just heart rate, but also oxygen levels and blood pressure, and a device was used to intermittently compress, then release, the umbilical cord. This simulated the fall in blood flow that happens with contractions in labour. A series of computer models were developed and then tested to find one that reliably predicted dangerously low levels of blood pressure based on fetal heart rate data.

The final model was able to correctly identify when blood pressure fell below 30 mmHg 34% of the time, while correctly identifying blood pressures that didn’t fall this low 94% of the time.

How does it work for humans?

Taking the model that worked well for their sheep, they applied the model to a database of over 50,000 CTG recordings with known outcomes, called the Oxford dataset. The final 90 mins of each recording were analysed as it would be expected that this would most accurately relate to perinatal outcome. Fetal blood pressures (obviously) were not recorded, so instead they looked to see if the model could tell when any one of stillbirth, neonatal death, neonatal seizures, encephalopathy, intubation or cardiac massage, followed by admission to neonatal intensive care occurred. (This of course assumes those affected by one of these outcomes had low blood pressure in the last 90 minutes of labour, and those who did not had normal blood pressures. Whether that is true or not is unknown.)

For CTG recordings where one or more poor outcomes had occurred there was a statistically significant difference in the estimated blood pressure, compared to CTG recordings where no poor outcome occurred. The difference was only 0.5 mmHg however – not something clinically useful. This research offers a potential starting point for new ways of thinking about fetal physiology and for developing new tools to accurately reflect fetal wellbeing in labour. But for now – we still have no tools to accurately detect when expedited birth will improve perinatal outcomes.

Reference

Tolladay, J., Lear, C. A., Bennet, L., Gunn, A. J., & Georgieva, A. (2023). Prediction of fetal blood pressure during labour with deep learning techniques. Bioengineering, 10(7). https://doi.org/10.3390/bioengineering10070775