Evaluating the added value of salivary hormones in the context of menstrual cycle staging: A machine learning approach and app-implementation

Objective

Salivary hormone assessment is commonly used in menstrual cycle studies, but its validity for accurate menstrual cycle staging has been questioned. In the present study, we explore possibilities and limitations of salivary hormone assessment for menstrual cycle staging using a machine-learning approach. Specifically, we determine, how saliva sampling should be scheduled in various scenarios to maximize prediction accuracy of menstrual cycle phases from salivary estradiol and progesterone.

Methods

We utilize a data set including daily salivary estradiol and progesterone assessment, urinary ovulation tests, as well as accurate forward and backwards counts of cycle days over 136 cycles from 68 women (age: 18–35 years). A Support Vector Machine (SVM) approach was chosen to evaluate improvements in prediction accuracy for cycle phases due to salivary hormone assessments, using a series of models designed to reflect practical scenarios in menstrual cycle research.

Results

A singular salivary hormone assessment does not significantly improve prediction of menstrual cycle phases when adequate counting methods or urinary ovulation kits are available. When no counting method is available, only progesterone, but not estradiol measurements can adequately distinguish between cycle phases, specifically progesterone works best in identifying mid-luteal sessions. However, salivary hormone assessment does significantly improve prediction of cycle phases when more than one time-point is assessed, and values can be referenced against each other. Adding a second assessment timepoint is more informative for estradiol than progesterone values, but most effective when both hormones are combined. Importantly, and contrary to common practice, prediction accuracy is highest when saliva is sampled on days near the transitions between cycle phases when counting methods do not allow for a definitive decision.

Conclusion

These results demonstrate that salivary hormone assessments are not necessary in all research designs but are useful when counting methods are inadequate or do not allow for a definitive decision, or when multiple assessment days are included. Results of our models were implemented in a web application to aid researchers in assessing the prediction accuracy of their menstrual cycle staging based on the measures they have available.