Problem Statement

A significant problem related to women’s health research is that there is a gap in our understanding of the relationship between menstrual cycles and diabetes, particularly how hormone fluctuations might influence glucose levels. While Dexcom CGMs can illustrate high resolution fluctuations of glucose levels at 5-minute intervals, this information alone is insufficient for analyzing how menstrual cycles can impact glucose regulation and diabetes progression. Additionally, current datasets often lack information about hormones like estradiol and testosterone, which are required in order to gain a full understanding of the role that menstrual cycles have in the development and diagnosis of diabetes.

Furthermore, the ability to draw meaningful conclusions from glucose or other hormonal data is hindered by factors such as small sample size, reliance on self-reported data, and the lack of existing studies that focus on the role that menstrual cycles play on the onset of diabetes. This lack of high-quality, comprehensive data makes it challenging to draw concrete conclusions or properly inform diagnosis and treatment of diabetes from sex-based trends found in CGM data.

Methods

To start testing our theory that signals corresponding to the menstrual cycle can be found in glucose data, we first analyzed our personal CGM data collected by Dexcom sensors. We will present Saara’s data, which is the most complete, as a case study in the results section (Subproject 1).

Moving forward, our next step was to look at our larger dataset from Dexcom (Subproject 2). Specifically, we had one year of data for 16,000 diabetic individuals, split into 10,000 females in their 30s and 2,000 each of females in their 60s, males in their 30s, and males in their 60s. Our dataset was also split evenly between T1D and T2D.

Once we took a look at our own data and discussed with our advisor, we came up with the following daily metrics to analyze in the larger dataset: 5th percentile (p5), median, and 95th percentile (p95) glucose values; number of peaks per day; number of minutes out of range (OOR); and percent time OOR. For the last three metrics, we categorized a peak/out of range value as greater than 180 mg/dL, which is the standard threshold used by Dexcom. We collected these values across the ~1 year of data we had for each individual, then ran that through a Lomb-Scargle algorithm developed within the Smarr Lab. Lomb-Scargle works by fitting a sinusoidal model to the data at each possible frequency (i.e. number of days), with a larger power reflecting a better fit. Additionally, we utilized a classification algorithm developed by the Smarr Lab to get thresholds for cyclicity, using males as a control. The results of this pipeline yielded a “cyclic range” of 22-43 day signals with a power above 7.5. The significance of the Lomb-Scargle values for each group were then computed using the Chi-Square test, which compared observed vs. expected counts in the cyclic range.

For all of these metrics, we can infer a possible link to menstrual rhythms if high-powered 22-43 day Lomb-Scargle signals are enriched for females in their 30s when compared to other groups. Younger females are the only group that should have a high prevalence of menstrual cyclicity. Specifically, if the effects for this group are greater than in males in their 30s, we can rule out age effects, and if they are greater than in females in their 60s, we can rule out sex effects.

Leader: Honieh Hemati