Use of uterine activity to predict preterm birth by artificial intelligence assisted models: a narrative systematic review.
Jagrati Chopra, Charlotte Maden, Isabel Howlett et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review assessed the predictive accuracy and applicability of artificial intelligence models using electrohysterography (EHG) or uterine-contraction time-series data to predict preterm birth. The authors identified 53 records and synthesised results narratively because of study heterogeneity. Excellent classification performance (ACC and/or AUC ≥0.9) was reported in 7/18 studies examining EHG features and 19/22 assessing AI classifiers. Non-linear features and deep-learning models performed best, but the review found no evidence of external validation, limiting confidence in their applicability beyond the datasets studied.
Key findings
- Excellent classification performance (ACC and/or AUC ≥0.9) was reported by 7/18 studies examining EHG features.
- Among studies assessing AI classifiers, 19/22 reported excellent classification performance (ACC and/or AUC ≥0.9).
- Non-linear EHG features outperformed linear features, while deep-learning models such as neural networks were the highest-performing classifiers.
- Unclear or high risk of bias was identified in 86.7% (46/53) of records.
Why it matters
Uterine-contraction signals may offer a route to predicting preterm birth, and the reviewed models showed promising classification performance. However, strong results within sampled datasets do not establish that models will perform reliably in other populations or clinical settings. The lack of external validation and concerns about bias mean that reported accuracy should be interpreted cautiously.
What to keep in mind
The review reports substantial heterogeneity, unclear or high risk of bias in 86.7% (46/53) of records, and no evidence of external validation. Unbalanced data and small sample sizes were also identified as concerns.