AI Breakthrough in Pregnancy Risk Assessment: Uncovering Hidden Dangers
In a revolutionary study published in BMC Pregnancy and Childbirth, researchers used artificial intelligence (AI) to analyse nearly 10,000 pregnancies, revealing previously unknown combinations of risk variables associated with major adverse outcomes, such as stillbirth. The study also discovered that infants receiving identical treatment according to clinical recommendations could have a tenfold variation in risk.
"This is a significant step forward in personalised risk assessment and pregnancy care," says Dr. Nathan Blue, the study's principal author and assistant professor of obstetrics and gynaecology at the University of Utah. "Our AI model helped identify a really unexpected combination of factors associated with higher risk."
The researchers began with an existing dataset of 9,558 pregnancies from around the country, which contained comprehensive information on social and physical variables such as pregnant women's levels of social support, blood pressure, medical history, and foetal weight. Using AI to analyse the data, the researchers discovered unexpected combinations of maternal and foetal traits associated with poor pregnancy outcomes.
One intriguing discovery was that female foetuses, who are typically at a lower risk of problems than male foetuses, are at a higher risk if the pregnant woman has pre-existing diabetes. According to Dr. Blue, this previously unnoticed trend indicates the AI model's ability to identify hidden threats. "It identified a potential risk indicator that even the most seasoned clinicians struggled to identify."
We estimate pregnancy risks using a wide range of variables, including maternal health and ultrasound data. While experienced clinicians can consider these aspects when making personalised treatment decisions, even the finest doctors may be unable to quantify how they arrived at their final conclusions. Human factors like bias, temperament, and sleep deprivation can gently sway judgement calls away from optimal care.
To overcome these problems, the researchers used a model known as "explainable AI," which provides users with the estimated risk for a particular combination of pregnancy factors, as well as information on which variables contributed to the risk estimation and how much. Unlike the more common "closed box" AI, which is nearly impenetrable even to specialists, the explainable model "shows its work," identifying areas of bias that can be rectified.
The researchers were particularly interested in establishing more accurate risk estimates for foetuses in the bottom 10% of the weight range but not the bottom 3%. These newborns are little enough to cause concern but large enough to be completely healthy. Current professional recommendations recommend intensive medical monitoring for all such pregnancies, which can be a substantial emotional and financial strain.
However, the study discovered that within this foetal weight class, the risk of a poor pregnancy outcome varied greatly, ranging from no riskier than an ordinary pregnancy to nearly ten times the average. The risk was determined by a number of factors, including foetal sex, the presence or absence of preexisting diabetes, and foetal abnormalities, such as a heart problem.
Dr. Blue emphasises that the study only found correlations between variables and does not provide insight into what causes unfavourable outcomes. But an AI risk-assessment tool might be much better than clinical "gut checks" because it would help clinicians make well-informed, fair, and repeatable suggestions.
The researchers must still test and validate their model in additional populations to ensure it can forecast danger in real-world scenarios. Dr. Blue is optimistic that an explicable, AI-based model will eventually aid personalised risk assessment and therapy during pregnancy.
"AI models can essentially estimate a risk that is specific to a given person's context, and they can do it transparently and reproducibly, which is what our brains can't do," Dr. Blue explains. "This kind of ability would be transformational across our field."
The study's findings underscore AI's potential to transform prenatal care by offering more precise and personalised risk assessments. This could enhance outcomes for both mothers and newborns, ultimately altering the treatment and monitoring of pregnancy.