Researchers from Kaiser Permanente and the Massachusetts Institute of Technology developed artificial intelligence models that can identify patients at higher risk for three common pregnancy complications using only routinely collected electronic health record data. The work, published in Communications Health, reports the models can flag elevated risk for gestational diabetes, gestational hypertension and preeclampsia as early as 12 weeks into pregnancy.
The research team trained a suite of 49 machine learning models on a dataset covering 442,760 pregnancies. According to the study report, the models were built to rely solely on routine EHR inputs rather than requiring additional specialized tests or external data sources. The source material states the models can identify higher-risk patients by about the end of the first trimester — roughly 12 weeks of gestation.
Details such as the specific EHR variables used, the breakdown of the 49 model types, model training procedures, and exact performance measurements (for example, sensitivity, specificity or area under the curve) were not reported in the source summary. Likewise, the source did not report demographic breakdowns of the dataset, external validation results, or direct comparisons to existing clinical risk-assessment tools.
Early identification of risk for gestational diabetes, gestational hypertension and preeclampsia may allow clinicians to monitor patients more closely, tailor prenatal care, and consider earlier or targeted interventions. These conditions are associated with complications for both mother and fetus when not recognized and managed appropriately.
Because the models use only routinely available EHR data, they could potentially be integrated into existing clinical workflows without the need for additional testing or data collection. The source does not specify any planned implementation or pilot deployments within Kaiser Permanente or elsewhere, nor does it provide information on clinician acceptance, patient impact, or cost implications.
Machine learning approaches have been increasingly explored to support risk stratification in pregnancy, leveraging large clinical datasets to identify patterns that might not be apparent from traditional risk scores. This study adds to that body of research by focusing on common complications and by emphasizing use of routine EHR data alone.
The source notes the research collaboration between a large integrated health system and an academic technology institute, but does not describe funding sources, ethics review details, or whether the models were tested prospectively in clinical practice.
The summary available in the source provides core findings but omits several details readers and clinicians typically look for when evaluating clinical AI tools. Missing information in the source includes:
Because these items were not reported in the source, readers should view the study findings as an initial report of model development rather than definitive evidence of clinical benefit or readiness for deployment.
The source article presents the development and training of the models and their ability to flag elevated risk by about 12 weeks but does not describe subsequent steps. The report does not state whether the authors have published follow-up validation work, intend to run prospective studies, or plan integration into clinical workflows at Kaiser Permanente or other health systems.
Future work typically required before clinical use would include external validation across diverse populations, assessment of real-world impact on care and outcomes, clinician-facing implementation studies, and monitoring for fairness across demographic groups. The source does not indicate which, if any, of these steps are under way for these specific models.
Kaiser Permanente and MIT researchers trained 49 machine learning models on a large dataset of 442,760 pregnancies and reported that models using only routine EHR data can flag higher risk of gestational diabetes, gestational hypertension and preeclampsia as early as 12 weeks into pregnancy. The source provides limited methodological and performance detail and does not report on prospective validation or planned clinical deployment.
For clinicians and health system leaders interested in clinical AI, the study signals a continued trend toward leveraging existing EHR data for early risk detection in pregnancy. However, the source lacks the detailed evidence typically needed to assess readiness for clinical use, and additional validation and implementation information would be needed to judge practical impact.
Personalise this feed
Your specialty. Your sources. Your digest.
All set up in under 2 minutes.
Personalise this feed
Your specialty. Your sources. Your digest.
All set up in under 2 minutes.