2.4. Data Analysis

For each DRSP symptom, we calculated the degree to which a symptom demonstrated an elevation in days −6 to −1 (“pre-menstruum”) from the start of the cycle relative to days +5 to +10 (“post-menstruum”). This was done for each woman for all available cycles. We subtracted the average post-menstruum score from the average pre-menstruum score and divided this score by participant-specific variance for each symptom. This essentially yielded an effect size for each woman and for each symptom [49].

We next compared neuroticism-anxiety according to diagnosis. Of note, DSM-5 diagnosis is determined as a constellation of affective, psychological, physical, behavioral, and functional symptoms (Table A1) with a “yes” or “no” determination for presence of each symptom. The effect size greater than or equal to 1 reflects presence of a symptom [49] and we applied DSM-5 diagnostic criteria, as described in the “Introduction”. We performed the Kruskal–Wallis rank sum test, with the total neuroticism-anxiety score as the outcome and group (PMDD vs. PMS vs. healthy) as the predictor. We applied Wilcoxon rank sum tests to make pairwise comparisons between group levels with Benjamini–Hochberg corrections for multiple testing.

Our main analysis was to test associations between affective/psychological premenstrual symptoms and neuroticism. Hence, we treated affective and psychological symptoms as continuous variables, reflected as the effect size. The ZKPQ neuroticism-anxiety is measured on a 0–10 scale. We employed 7 separate generalized additive models [29] to study the relationship between neuroticism-anxiety score and each of the 7 affective and psychological symptoms with the total neuroticism score as the predictor and each symptom as the outcome. We adopted “gam” function in the mgcv package in R for fitting the additive models. Specifically, we utilized a spline function for the predictor with maximum degrees of freedom being 6; the best effective degrees of freedom was chosen using generalized cross validation criterion. We first constructed unadjusted models, following which we adjusted the models for current age and current age plus age of menarche based on the literature demonstrating the relationship between duration of ovulatory cycles and premenstrual symptomatology [50]. We used the False Discovery Rate (FDR) method to correct for multiple testing with a p value less than or equal to 0.05 considered significant.