Labeling self-tracked menstrual health records with hidden semi-Markov models

Abstract

AbstractGlobally, millions of women track their menstrual cycle and fertility via smartphone-based health apps, generating multivariate time series with frequent missing data. To leverage data from self-tracking tools in epidemiological studies on fertility or the menstrual cycle’s effects on diseases and symptoms, it is critical to have methods for identifying reproductive events, e.g. ovulation, pregnancy losses or births. We present two coupled hidden semi-Markov models that adapt to changes in tracking behavior, explicitly capture variable– and state– dependent missingness, allow for variables of different type, and quantify uncertainty. The accuracy on synthetic data reaches 98% with no missing data, 90% with realistic missingness, and 94% accuracy on our partially labeled real-world time series. Our method also accurately predicts cycle length by learning user characteristics. It is publicly available (HiddenSemiMarkov R package) and transferable to any health time series, including self-reported symptoms and occasional tests.