Researchers Develop Machine Learning Algorithm to Predict Threatened Miscarriage
A new study shows how routine blood tests and machine learning can help predict the risk of threatened miscarriage early in pregnancy.
As laboratory diagnostics continue to evolve alongside artificial intelligence (AI) and machine learning (ML), a new study out of China highlights the growing potential of routine blood testing as a predictive tool in obstetric care. Researchers have developed a machine learning model capable of identifying early signs of threatened miscarriage using only standard blood test results.
For lab leaders, this is a key opportunity to use existing lab data and infrastructure to support early pregnancy risk detection. Properly standardized blood data can help shift diagnostics upstream, potentially identifying issues before symptoms appear.
A threatened miscarriage (also called threatened abortion) is a medical condition in early pregnancy where a woman experiences vaginal bleeding before 20 weeks gestation, but the cervix remains closed and the pregnancy is still viable—meaning the fetus is alive with a detectable heartbeat.
The results of the study, “Predicting the Risk of Threatened Abortion Using Machine Learning Methods: A Comparative Study,” were published in BMC Pregnancy and Childbirth in August.
To perform the research, scientists from the Second Affiliated Hospital of Shaanxi University of Chinese Medicine, collected medical records and analyzed data from 1,764 patients with threatened miscarriage and 1,489 healthy control subjects. Blood test data of the study participants were collected and a data preprocessing tactic called the Z-score normalization technique, also known as standardization, was applied to routine blood indicators. The high-level, general-purpose programming language called Python was used to facilitate the data transformation of eight different ML algorithms. The performance of those ML models were evaluated by calculating their area under the curve (AUC) values, which represents the overall performance of the various models.
Machine Learning Unlocks Predictive Power of Routine Blood Tests
The ML algorithm known as Deep Neural Network (DNN) achieved the best predictive performance with an AUC value of 96.76% That model also had the top metrics for accuracy at 91.88%, specificity (91.62%) and sensitivity (92.11%).
According to the study, approximately 25% of all pregnancies involve a threatened miscarriage and women who experience this condition are 2.5 times more likely to experience a miscarriage compared to healthy pregnancies.
“We’re not sure why this happens in some pregnancies but not in others,’ said Lisa Jackson, MD, assistant clinical professor in the Raquel and Jaime Gilinski Department of Obstetrics, Gynecology and Reproductive Science at Icahn School of Medicine at Mount Sinai, in an interview with The Bump.
Lisa Jackson, MD, assistant clinical professor in the Raquel and Jaime Gilinski Department of Obstetrics, Gynecology and Reproductive Science at Icahn School of Medicine at Mount Sinai noted, “The good news is that the majority of the time, bleeding in early pregnancy with the presence of fetal heart tones and a closed cervix doesn’t result in miscarriage.” (Photo credit: Icahn School of Medicine at Mount Sinai)
“For most women, the outlook is good, and they go on to have normal pregnancies,” added women’s health expert Jennifer Wider, MD.
Neither Jackson nor Wider were affiliated with the study out of China.
The study links incidents of threatened miscarriage with various factors, including chromosomes, genes, hormonal imbalances, immune factors, and maternal health as well as environmental influences. The researchers also stated in their study, “Women experiencing threatened abortion often encounter significant stress and require multiple clinical assessments to accurately determine their pregnancy status. Threatened abortion not only poses a threat to the health of pregnant women but may also affect their future fertility.”
The researchers believe their ML algorithm has immense clinical value and can assist doctors in accurately identifying high-risk patients. They suggest exploring the application of the DNN model in different clinical settings and patient populations and incorporating more diverse data sources to optimize its predictive capabilities.
A New Role for Clinical Labs in Reproductive Health
The researchers concluded, “Our research on constructing a prediction model for threatened abortion through routine blood tests has revealed the great potential of ML algorithms in detecting threatened abortion. This algorithm is expected to analyze routine blood data to identify at-risk pregnancies at an early stage, significantly improving the early detection of this common pregnancy complication. It will assist healthcare providers in intervening earlier and reducing the incidence of (threatened) abortion.”
More research and trials are needed before any ML model may be used for the early detection of threatened miscarriage incidents. However, this latest study does suggest the potential of machine learning algorithms could be beneficial in timely predictions of the medical condition, which would improve patient outcomes.
For laboratories, this research underscores the untapped value in the routine data already being collected daily. As machine learning models like the deep neural network in this study continue to demonstrate clinical relevance, lab leaders are uniquely positioned to facilitate their integration—by ensuring high-quality, normalized data flows and supporting interdisciplinary collaboration between data scientists and clinicians.
Moving forward, blood testing won’t just confirm conditions, it may soon help predict and prevent them. With strategic investment in data analytics and model validation, labs can transform from diagnostic endpoints into proactive partners in patient care.
—JP Schlingman
