OAKLAND, California — Artificial-intelligence models trained on routine electronic health records may help identify some pregnancy complications earlier than they are typically diagnosed, according to research from Kaiser Permanente and collaborating scientists. The study analysed records from 442,760 pregnancies in Northern California between 2011 and 2021. Kaiser Permanente’s Division of Research published the findings on November 19.
Researchers developed models to estimate the risk of seven complications at different stages of pregnancy. The tools showed promise in identifying increased risk of gestational diabetes, gestational hypertension and preeclampsia as early as 12 weeks. The team also tested the models using data from 6,105 pregnancies in the national All of Us Research Program.
Prediction was more difficult for outcomes such as preterm birth, stillbirth, placental abruption and small-for-gestational-age birth. The models generally performed less well when applied to patients in different healthcare settings, although adapting them with additional data improved performance.
The findings suggest that information already collected during prenatal visits could help clinicians decide who may need closer monitoring, preventive care or specialist referral. However, identifying risk is not the same as preventing a complication. Researchers said further work is needed to establish whether using the models in clinical practice improves outcomes for mothers and babies.
The study also underlines the importance of testing AI tools across diverse populations and healthcare systems before routine adoption. Models trained in one region may not work equally well elsewhere because patient populations, clinical practices and record systems differ. The research is promising, but it is not a replacement for prenatal assessment by qualified clinicians. More health and medical research.