148 Chapter 7 First, we plotted the percentage of migraine on each day of the menstrual cycle for all women. Menstrual cycles were then divided into two segments based on the onset of menstruation and the ovulation dates and standardized to 28 days. Standardization was applied solely for visualization in graphs. The pre-ovulation segment was standardized to days 1–14 and the post-ovulation segment was standardized to days 15–28. Subsequently, for the graphs, the migraine attack incidence was calculated for each standardized day. Additional plots were made following the same steps of standardization with migraine incidence subdivided into migraine attacks with aura and migraine attacks without aura, as a previous study has shown differing incidences across the menstrual cycle.12 These plots were made using data of the whole cohort and another using only the data from women diagnosed with migraine with aura. We conducted our analyses using three statistical approaches: (1) mixed-effects logistic regression, (2) case-crossover, and (3) self-controlled case series (SCCS). The mixed-effects model was selected because it allows to account for patient intervariability by including individual participants as random effects, while ovulation and perimenstrual windows were included as fixed effects. The outcome was start of a migraine attack, including a differentiation between attacks with and without aura. The self-controlled designs eliminate time-invariant confounders by allowing participants to serve as their own controls. In the case-crossover design, we stratified the data per patient, starting with the occurrence of migraine attacks and then examining whether these were preceded by days within the ovulatory window. Additionally, we conducted a sub-analysis excluding all days during the perimenstrual window, as this period is a known risk factor for migraine attacks.5 The perimenstrual window is defined as days -2 to +3 relative to the onset of menstruation, with day 1 marking the start of menstruation and no day 0 included.13 For the SCCS model, patient time was further segmented into four periods: the ovulatory window (days -14 to -9), the perimenstrual window (days -2 to +3), the luteal phase (following the ovulatory window and ending with the onset of perimenstrual window), and the follicular phase (following the perimenstrual window and ending before the ovulatory window). This method allows for the calculation of migraine risk across different phases of the menstrual cycle. In two separate analyses, we used both the follicular and luteal phases as reference periods. Calculations for the SCCS model were performed using conditional Poisson regression. To assess model
RkJQdWJsaXNoZXIy MjY0ODMw