Skip to main content
Live
Main content

Kaiser Permanente AI models flag three pregnancy complications by week 12

Trained on 442,760 pregnancies, the Kaiser Permanente and MIT models beat NIH and SMFM risk checklists for gestational diabetes, hypertension and preeclampsia.

Jaeden Schafer
Editor in Chief · · 4 min read
Kaiser Permanente AI models flag three pregnancy complications by week 12

Kaiser Permanente and MIT researchers have built AI models that flag patients at higher risk for gestational diabetes, gestational hypertension and preeclampsia as early as 12 weeks into pregnancy, using only data already entered into routine electronic health records. The Kaiser Permanente study, published in Communications Health, trained the models on 442,760 pregnancies seen at Kaiser Permanente Northern California between 2011 and 2021.

The team built 49 machine learning models, covering seven complications at seven timepoints spaced six weeks apart. The models were trained on 80% of the patient records and tested on the remaining 20%.

At 12 weeks, the models scored an AUC of 0.80 for gestational diabetes, 0.79 for gestational hypertension and 0.74 for preeclampsia, on a scale where 1.0 is perfect. Those early models outperformed existing risk checklists from the NIH and the Society for Maternal-Fetal Medicine.

“Our study shows that we may be able to use information already collected during routine prenatal care”
— Yeyi Zhu, Research scientist, Kaiser Permanente Division of Research

Key facts

  • 01Kaiser Permanente and MIT models flag gestational diabetes, gestational hypertension and preeclampsia risk as early as 12 weeks.
  • 02The 49 models were trained on 442,760 pregnancies at Kaiser Permanente Northern California from 2011 to 2021.
  • 03At 12 weeks, AUC scores were 0.80 for gestational diabetes, 0.79 for gestational hypertension and 0.74 for preeclampsia.
  • 04Counseling the top 20% of flagged patients for gestational diabetes would save about $10.8M across the test group.
  • 05Accuracy dropped on 6,105 pregnancies from the NIH All of Us program; local adaptation narrowed the gap.

Complications that occur before or at delivery were harder to predict. Models for preterm birth, stillbirth, placental abruption and small-for-gestational-age births scored between 0.62 and 0.68 even at 24 weeks. The researchers said adding ultrasound or other test measurements might help.

The researchers also modeled costs. Steering the top 20% of flagged patients into gestational diabetes lifestyle counseling would save about $10.8 million across the study's test group, and aspirin prophylaxis for preeclampsia was also cost-effective. A cervical-length screening pathway for preterm birth was not.

“From a clinical perspective, knowing that a patient is at increased risk is most valuable when we have something we can offer to reduce that risk”
— Mara Greenberg, Director, Kaiser Permanente Northern California Regional Perinatal Service Center

To test how well the models carry over to other populations, the team ran them on 6,105 pregnancies from the NIH's All of Us Research Program. Performance dropped on that data. Adapting the models with All of Us data narrowed the gap but did not close it.

“We found that adapting the models using data from All of Us improved their predictive performance”
— Marzyeh Ghassemi, Associate professor, MIT

The models also performed worse for Black patients on preeclampsia and gestational hypertension. The researchers said this may stem partly from early-pregnancy blood pressure being recorded earlier and more often for white patients.

Yeyi Zhu, a research scientist at the Kaiser Permanente Division of Research, and Marzyeh Ghassemi, an associate professor at MIT and a principal investigator with the MIT Jameel Clinic, were co-senior authors. The work was funded by the National Heart, Lung, and Blood Institute and the National Science Foundation Graduate Research Fellowship Program.

Related · from this week
Recovered ELIZA source code reframes the origin story of the chatbot
Jaeden Schafer · 5 min read →

Next, the Kaiser Permanente team plans to test whether adding ultrasound measurements, clinical notes and other biomarkers improves predictions, and whether using the models in clinical practice improves outcomes for mothers and infants.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from News

Recovered ELIZA source code reframes the origin story of the chatbot
Analysis

Recovered ELIZA source code reframes the origin story of the chatbot

A new book pulls the 1966 program's code from the MIT Archives, revealing multiple ELIZAs and a template for today's AI intimacy problem.

Jaeden Schafer5 min read
OpenAI logo
News

OpenAI puts Codex agents to work fixing ChatGPT slowdowns

Codex agents ran more than 1,700 investigations in three months and found root causes in 600+ cases, with engineers still approving every fix.

Jaeden Schafer5 min read
Apple job listing shows it is building an AI agent to run the Mac
News

Apple job listing shows it is building an AI agent to run the Mac

The agent would use apps, browse the web and write code, and Apple is reportedly eyeing a launch as soon as the end of next year.

Jaeden Schafer5 min read