Constant skin temperature using AI throughout pregnancy may indicate labor onset: BMC study

A new study published in the journal of BMC Pregnancy and Childbirth showed that continuous temperature deep learning might yield clinically useful pregnancy care solutions. Instead of being assessed in days, the average inaccuracy of the current clinical expected date of delivery, or EDD, is calculated in weeks. A "term" pregnancy lasts 5 weeks, ranging from 37 to 42 weeks. There are no clinical techniques that reliably indicate whether a pregnancy is likely to start on the earlier or later side of this range, despite considerable efforts to discover biomarkers of approaching labor. 

In many species, changes in body temperature predict the commencement of labor, although this idea has not been investigated in humans. So, Chinmai Basavaraj and colleagues looked into whether constant body temperature of women fluctuates in a similar way and whether these variations may be related to their hormonal state. Lastly, they created a deep learning model that forecasted the time until labor commencement each day by utilizing temperature patterning. 

Using a wearable smart ring, this research assessed trends in continuous skin temperature data from 91 pregnant women (n = 54 spontaneous labors). This research looked at daily steroid hormone samples before delivery in a subgroup of 28 pregnancies in order to evaluate the connections between hormones and the trajectory of body temperature. Also, the deep learning model called autoencoder long short-term memory (AE-LSTM) to produce a new daily estimate of the number of days till the start of labor.

Urinary hormones and the kind of labor were linked to characteristics of the temperature shift that preceded labor. When compared to pregnancies without spontaneous labor, those that had spontaneous labor had more stable circadian rhythms, lower body temperatures, and a higher estriol to α-pregnanediol ratio.

The AE-LSTM model was trained using skin temperature data from 54 pregnancies that had spontaneous labor between 34 and 42 weeks of gestation, and an additional 37 pregnancies that experienced artificial induction of labor or Cesarean without labor were utilized for additional testing.

5-minute skin temperature data from 240 days before the day of labor commencement served as the pipeline's input. Regardless of gestational age, the AE-LSTM average error decreased below 2 days at 8 days before to labor during cross-validation. From the AE-LSTM output, labor onset windows were computed utilizing a probabilistic distribution of model error.

For these windows, 79% of spontaneous labors within a 4.6-day window at 7 days prior to actual labor and a 7.4-day window at 10 days prior to true labor had their labor start accurately predicted by AE-LSTM. Overall, constant skin temperature throughout pregnancy indicates the onset of labor and changes in hormones.

Reference:

Basavaraj, C., Grant, A. D., Aras, S. G., & Erickson, E. N. (2024). Deep learning model using continuous skin temperature data predicts labor onset. In BMC Pregnancy and Childbirth (Vol. 24, Issue 1). Springer Science and Business Media LLC. https://doi.org/10.1186/s12884-024-06862-9