This is an analysis of data from the baseline and first 6 annual follow-up visits (1995–2002) of women who reported incontinence during the Study of Women’s Health Across the Nation (SWAN). SWAN is a multi-center, multi-ethnic, prospective study of mid-life women and the menopause transition(5
). Seven clinical sites (Boston, Massachusetts; Chicago, Illinois, the Detroit area, Michigan; Los Angeles, California; Newark, New Jersey; Pittsburgh, Pennsylvania; and Oakland, California) identified 16,065 community-based women aged 40–55 years by random digit dialing, snowball, and/or list-based sampling and screened them for eligibility for the cohort study. From this large sample, each of the sites recruited about 450 women to include a Caucasian group and one designated minority group (African American at four sites, and Chinese, Japanese and Hispanics at one site each) for a total cohort of 3302 women. In SWAN, these minority groups were over-sampled to allow sufficient numbers for the planned analyses by group. Eligibility criteria for the SWAN cohort were age 42–52 years and self-identification as one of five racial/ethnic groups to be studied. The exclusion criteria were inability to speak English, Spanish, Japanese, or Cantonese, no menstrual period in greater than 3 months or hysterectomy and/or bilateral oophorectomy prior to enrollment, and the current use of oral contraceptives, estrogens, progestins, or luteinizing hormone agonists. SWAN was approved by Institutional Review Boards at all sites and all women gave informed consent.
In a self-administered questionnaire at each annual follow-up visit, women were asked: “In the past year (or since your last study visit), have you ever leaked even a small amount of urine involuntarily?” Participants listed the frequency of their incontinence as “almost daily/daily” (daily), “several days per week” (weekly), “less than one day per week” (monthly), “less than once a month” or “none.” Because we consider incontinence occurring less than once a month not to be clinically significant and to have a higher misclassification rate, we combined this category with “no incontinence” or to create the category of no regular incontinence.
We categorized type of incontinence as “stress” if participants reported leakage with “coughing, laughing, sneezing, jogging, jumping, with physical activity or picking up an object from the floor” or as “urge” if participants reported leakage “when you have the urge to void and can’t reach the toilet fast enough.” We defined prevalent incontinence as incontinence reported at baseline. First reports of incontinence at any of the 6 annual follow-up visits were considered to be incident incontinence.
We defined improving and worsening in incontinence in two ways. First, we evaluated change in incontinence reporting from year to year. Incontinence was considered improving if the frequency of incontinence decreased from one annual visit to the next, i.e.: from daily to weekly or less, from weekly to monthly or less, or from monthly to no regular incontinence. We considered incontinence worsening if the frequency of incontinence increased from one annual visit to the next, i.e.: from no regular incontinence (after a previous report of incontinence) to monthly or more, from monthly to weekly or more or from weekly to daily. We defined no change in incontinence as the same reported level of frequency from one annual visit to the next. Second, we evaluated change over the full 6 years of follow up in the following manner. From one year to the next, a change from less frequent incontinence to more frequent incontinence was assigned a numerical value of +1, a change from more frequent incontinence to less frequent incontinence a value of −1, and no change in frequency, a value of 0. We summed these values of change over the 6 years. Women whose scores were less than 0 were considered to have improved, those whose scores were greater than 0 were considered to have worsened, and those with a score of 0 were considered to have no change. A no change score could represent either consistent reporting of the same frequency of incontinence from year to year or variable reporting of frequency from year to year that summed to 0. We sub-divided this group into “no change with low reporting variance” and “no change with high reporting variance” respectively. Similarly, we determined the proportion of women who only worsened (ie: never reported improvement from year to year) and only improved (ie: never reported worsening from year to year) over the six years.
Main Time-Dependent Covariates
Our main independent variables in this analysis were menopause transition category and hormone use determined in the year previous to the reported change in incontinence frequency. We classified menopause status from questions assessing menstrual bleeding patterns on annual follow-up interview questionnaires. The definitions of natural menopause transition categories used by SWAN are as follows: Pre-menopause: less than three months of amenorrhea and no menstrual irregularities in the previous year; Early Peri-menopause: less than three months of amenorrhea and some menstrual irregularities in previous year; Late Peri-menopause: three to 11 months of amenorrhea; Post-menopause: 12 consecutive months of amenorrhea with no apparent medical cause. SWAN defined surgical menopause as hysterectomy (with or without oophorectomy) or bilateral oophorectomy.
Women who started using hormones in pre-, early peri- and late peri-menopause were considered to be hormone users with an unclassifiable menopausal status. We divided post- and surgical menopausal women into those using and not using hormone therapy when numbers in each category were sufficient for analysis. To account for change in menopausal status between years, we classified women as having transitioned from one menopausal stage to another, started using hormones, stopped using hormones, or as having no change in their status. Finally, from the baseline and year 6 menopause status we created categorical variables to reflect the change in status over the 6 years: from pre-/early peri-menopause to pre-/early peri-menopause, to late peri-menopause, to post-menopause, to surgical menopause and to unclassifiable menopause status.
Weight and waist/hip ratio changes were also main variables of interest. Certified staff used calibrated scales and a stadiometer to measure height, weight, and waist and hip circumferences. We calculated baseline body mass index (BMI) as weight in kilograms/(height in meters)2 and waist to hip ratio (WHR) as waist circumference in centimeters/hip circumferences in centimeters. We defined weight gain from year to year as an increase in weight per pound; gain in WHR was an increase by 0.1 units.
Given that a woman’s initial reported frequency of incontinence is likely to impact whether her incontinence improves or worsens, we controlled for this effect by treating frequency of incontinence at first report as an independent variable. Race and ethnicity were self-defined. We estimated socioeconomic status by level of difficulty paying for basics (food, heat and shelter). SWAN obtained self-reported diabetes, hypertension, fibroids, obstetrical history, smoking history and medication use, including hormone use by interview and general health status by the self-administered questionnaire. SWAN used an adaptation of the SF-36(6
), the Center for Epidemiological Studies-Depression (CES-D) scale(7
), and the Medical Outcomes Study (MOS) Social Support Survey(8
) and Life Stressors and Social Resources Inventory (LISRES)(9
). Life stressors were assessed with a scale derived from the Psychiatric Epidemiology Research Interview(10
) and modified to include events particularly relevant for middle-aged women or for those living in low socioeconomic environments, involving job, family, financial, and illness/death events. Anxiety symptoms were measured by a summed score of days in the past two weeks in which certain symptoms were experienced (irritability, feeling tense or nervous, pounding or racing hear, feeling fear for no reason); anxious was defined by a score of 4 or more(11
), For our time-dependent covariates, we evaluated whether new onset medical conditions and change in these scales and events from year to year were associated with change in incontinence frequency.
For data that were collected annually (for example, social support, depressive symptoms, and weight changes) we examined the degree of change for each variable over the 6 years by plotting their values. For those variables noted to change over time, we created summary variables describing the patterns of change. For example, we categorized weight gain and loss as a change of greater than 5% from baseline weight at year 6. For WHR, gain was defined as an increase of greater than 7.5% or loss more than 2.5% from baseline WHR. For variables whose change could be bi-directional, we accounted for variability in the change over time (for example, weight cycling) by entering the standard deviation of change over the 6 years into our models.
Drop outs included continent women who were deceased, discontinued the study, or could not be contacted for two or more consecutive visits. Incontinent women were retained in the analysis as long as they contributed data. When a woman was missing data on frequency and type of incontinence from one or two visits, we imputed values as follows. If one missing value occurred at baseline, we imputed the value from the subsequent visit. If the missing value occurred at year 6, we imputed the value at the previous visit. If the missing values were between concordant values, we imputed that concordant value. If one missing value occurred between two discordant values, we randomly assigned it the value of the previous or subsequent visit. If two missing values occurred in sequence between two discordant ones, we randomly assigned the missing values the previous and subsequent observed values. We did not impute for three or more missing data points. Overall, we imputed incontinence frequency for 808 women at 1150 visits (5.0% of all visits) and incontinence type for 447 women at 572 visits (2.5% of all visits). When weight or WHR were missing for one or two visits, the values were imputed as the mean between the two known values (weight: 263 women at 394 visits, 1.7% of all visits; WHR: 257 women at 374 visits, 1.6% of all visits).
To assess the robustness of our results, we developed our main multivariable analyses with and without imputed values. Additionally, because we had a disproportionate and systematic loss to follow-up of the Caucasian and Hispanic women from the New Jersey site in year 6, we developed separate models with and without these participants. As the results of our models were similar, we present our final models with imputed data and with New Jersey year 6 data included.
We checked each continuous covariate for normality by conducting either a Q-Q plot or a Lilliefors test, an adaptation of the Kolmogorov-Smirnov test. The social support score was the only covariate that violated the normality. Distributions of incontinence and other important variables for women who remained in (analysis cohort) and who dropped out of (drop-out cohort) our study were then compared using the t-test or Mann-Whitney-U statistic (in the case of social support) for continuous and chi-squared or proportional tests for categorical data.
After selecting candidate variables based on the literature and a priori hypotheses, we used multivariable analysis to examine which factors were associated with improving and worsening incontinence. To build our models, we used backward and forward stepwise processes entering variables based on our hypotheses or based on associations with the incontinence outcome at a p-value of 0.10 or less. We used SAS 9.1, SAS Institute Inc., Cary, NC, USA.
We used Generalized Estimating Equations (GEE), a methodology for dealing with repeated measures in longitudinal data, to evaluate factors associated with improving or worsening incontinence from year to year. The use of GEEs is based on a marginal modeling approach(12
) for making inferences about the effects of independent variables on the outcomes. While this approach does not explicitly model correlation structure between repeated observations on the same woman, it accommodates the correlation. We developed models to compare the following: improving with worsening/no change in incontinence as the reference, worsening incontinence with improving/no change in incontinence as the reference. For women who reported only stress or urge incontinence at two or more visits, we developed models for improving and worsening stress or urge incontinence between those visits. Because our main objective was to evaluate the effect of menopausal status and change in status on incontinence independent of age, we forced these factors into our final models. Model stability was tested by respectively forcing in menopausal status, change in status and age without significant changes in the point estimates of the other variables. Estimated correlation matrices showed no significant co-linearity among our independent variables, including between menopausal status and age. We chose our final models based on the lowest QIC (quasi-likelihood under independence model criterion) and the related QICu statistic which indicate the best fit for GEE models(13
We also used regression modeling to evaluate factors associated with change in frequency of incontinence over the entire 6 year period using the 6-year summary outcomes and independent variables described above. We assumed that low variance in reporting no change in incontinence would be a stable reference group. We compared odds of high variance in reporting no change in incontinence, odds of improving with low variance no change/worsening incontinence and odds of worsening with low variance no change/improving incontinence.