Intrauterine devices (IUDs) are contraceptive devices that are very safe and effective. Levonorgestrel IUDs (LNG-IUDs) are associated with higher user satisfaction and continuation but may be associated with unpredictable menstrual bleeding after insertion that can include a wide variation in volume, duration, frequency, and regularity. It is estimated that more than half of individuals who use LNG-IUDs experience prolonged or frequent bleeding or both in the 3 months following insertion. To help patients know what to expect and provide more individualized guidance on bleeding changes after LNG-IUD insertion, an artificial intelligence prediction algorithm was incorporated into the MyIUS app. This study is designed to validate the real-world performance of this model at predicting bleeding intensity and regularity after LNG-IUD insertion. This was a prospective cohort study conducted across Germany, Denmark, Sweden, Spain, Mexico, and Brazil. Inclusion criteria were individuals using the app who provided 90 consecutive days of bleeding data, as well as continuing to track bleeding for 180 days after enrolling in the study. The primary outcome for this study was the performance of the bleeding intensity prediction algorithm in classifying individuals into one of 3 groups. The secondary outcome was the performance of the menstrual cycle regularity prediction algorithm classifying individuals to irregular or regular cycles in each of 3 doses of LNG-IUD. Final analysis included data from 1734 individuals. Actual bleeding intensity was classified as predominantly amenorrhea in 188 individuals, predominantly spotting in 712 individuals, and predominantly bleeding in 834 individuals. The algorithm performed with an area under the curve (AUC) of 0.81 (95% CI: 0.79-0.83), showing sufficient accuracy in a real-world setting. Bleeding intensity predictions were most accurate in participants in the predominantly bleeding category, with an overall 80.2% accuracy across the 3 dosages of LNG-IUDs. Regularity analysis excluded individuals with predominantly amenorrhea, for a total of 1,620 cases for analysis. The overall AUC for this prediction was 0.66 (95% CI: 0.63-0.68), with a better performance for individuals with regular cycles compared with irregular cycles. Sensitivity analyses showed similar results, and that individuals who dropped out would not have significantly changed the results. These results indicate that the real-world performance of the bleeding pattern prediction algorithm in the MyIUS app is valid. The AUC for bleeding intensity was high overall, indicating high performance in line with previous clinical trials. Further research should focus on the improvement of the cycle regularity prediction algorithm, as well as its validation in a real-world setting after improvements. In addition, further research should compare the performance of this algorithm in other applications and in other subgroups such as women who are or are not breastfeeding or women in different age groups. (Abstracted from Contraception. 2025 Dec:152:111201. doi: 10.1016/j.contraception.2025.111201)
Machado et al. (Wed,) studied this question.