Predicting menstrual bleeding patterns with levonorgestrel-releasing intrauterine systems

Area of Science:

• Reproductive medicine
• Contraception technology
• Medical statistics

Background:

• Levonorgestrel-releasing intrauterine systems (LNG-IUS) are widely used for contraception.
• Understanding bleeding patterns post-insertion is crucial for patient counseling and adherence.
• Predictive models can improve counseling regarding expected bleeding amount and regularity.

Purpose of the Study:

• To develop and evaluate a predictive model for bleeding patterns after LNG-IUS placement.
• To inform patient counseling on the expected amount and regularity of bleeding.
• To compare the performance of different modeling approaches.

Main Methods:

• Developed fixed-cluster and regression-tree models using pooled bleeding data from two clinical trials.
• Trained and cross-validated models on LNG-IUS 12 data, then applied to LNG-IUS 20 and LNG-IUS 8 data.
• Fixed-cluster model generated amenorrhoea, spotting, and bleeding clusters; regression-tree model assigned women to less- or more-bleeding groups.

Main Results:

• The fixed-cluster model achieved correct classification rates of 70.4% (LNG-IUS 12), 72.2% (LNG-IUS 20), and 69.0% (LNG-IUS 8).
• Cycle regularity separation was best with LNG-IUS 12 data, less optimal for LNG-IUS 20 and LNG-IUS 8.
• The regression-tree model showed high variability in bleeding group distributions.

Conclusions:

• A fixed-cluster model demonstrated superior performance in predicting bleeding patterns compared to a regression-tree model.
• The fixed-cluster model provides understandable and informative outputs for clinical counseling.
• This predictive tool can enhance patient counseling regarding LNG-IUS associated bleeding.