Apple Watch detects pregnancy with 92% accuracy, according to new study

A recent study, backed by Apple, has unveiled a revolutionary approach to early pregnancy detection, achieving an impressive accuracy rate of 92%. This advancement utilizes artificial intelligence to analyze behavioral data gathered from iPhones and Apple Watches, representing a significant leap in the use of everyday health metrics for predictive purposes. The research, titled "Beyond Sensor Data: Foundation Models of Behavioural Data from Wearables Improve Health Predictions," introduces the Wearable Behaviour Model (WBM), which transcends conventional health tracking methods that typically depend on basic sensor inputs like heart rate or oxygen saturation. Instead, it focuses on long-term behavioral trends, including sleep quality, mobility, heart rate variability, and overall activity levels, which have already been processed by Apple's sophisticated algorithms. The WBM was refined using a staggering 2.5 billion hours of data collected from participants in the Apple Heart and Movement Study (AHMS), involving over 160,000 volunteers. To enhance the model's accuracy, researchers also compiled a specialized dataset from 430 pregnancies, incorporating data from more than 24,000 women under the age of 50 who were not expecting. Rather than merely capturing short-term biometric spikes, the model identifies subtle, cumulative behavioral changes over time. It employs an advanced AI framework known as Mamba-2, which excels at analyzing time-series data like daily habits, enabling it to detect week-by-week physiological changes that could signify pregnancy, infections, or recovery from injuries. In detecting pregnancy, the AI pinpointed significant behavioral indicators such as altered gait, decreased mobility, and sleep disturbances as key early signals. When combined with biometric information like photoplethysmography (PPG), the WBM achieved its remarkable accuracy in pregnancy detection. The implications of these findings suggest that the Apple Watch and iPhone may soon play a pivotal role in reproductive health, potentially offering a non-invasive means of early pregnancy detection as a standard feature. However, the researchers emphasize that this innovative approach is not designed to completely replace raw sensor data. Instead, they propose a hybrid model where behavioral insights enrich health contexts, while traditional sensors continue to monitor real-time data. Beyond pregnancy detection, the AI model has demonstrated promising capabilities across 57 different health prediction tasks, including the early identification of respiratory infections and valuable insights into medication adherence, such as the use of beta blockers. As Apple delves deeper into the potential of wearable data, this study underscores how behavioral AI could evolve the Apple Watch into a more intelligent and proactive health assistant.