Pattern recognition in menstrual bleeding diaries by statistical cluster analysis

Area of Science:

• Gynecology
• Statistics
• Pharmacology

Background:

• Clinically relevant bleeding patterns require objective description.
• Existing methods lack treatment independence for bleeding pattern analysis.

Purpose of the Study:

• To identify a treatment-independent statistical method for describing bleeding patterns.
• To analyze menstrual bleeding diaries from clinical studies involving sex hormone drugs.

Main Methods:

• Applied four cluster analysis methods: single, average, complete linkage, and Ward's method.
• Determined the optimal number of clusters using statistical criteria (semi-partial R2, cubic cluster criterion, pseudo-F, pseudo-t2).
• Assessed the interpretability of results from a gynecological perspective.

Main Results:

• Ward's method effectively produced distinct clusters, unlike other methods.
• Identified six distinct bleeding patterns: two desirable and four undesirable.
• Successfully separated cyclic and non-cyclic bleeding patterns.

Conclusions:

• Ward's cluster analysis provides a robust, treatment-independent method for categorizing bleeding patterns.
• Facilitates comparison and categorization of medications and devices impacting menstrual bleeding.