Study population

In 1976, the NHS was initiated when 121,700 female nurses aged 30-55 years returned a mailed questionnaire about their health and lifestyle. The NHS II was initiated in 1989 when 116,430 female nurses aged 25-42 years completed and returned a similar questionnaire. Participants provided informed consent by returning their questionnaires. In both cohorts, biennial questionnaires are used to collect updated information. Participants were asked about UI on the 2000, 2002, and 2004 NHS questionnaires and the 2001, 2003, and 2005 NHS II questionnaires. The institutional review board of Brigham and Women’s Hospital approved these studies.

For these analyses, we defined baseline as 2000 in the NHS and 2001 in the NHS II. In each cohort, 93% (NHS, 78,197/83,997; NHS II, 79,684/85,507) of the women who provided information on UI at baseline also provided UI information on at least one of the follow-up questionnaires and were eligible for these analyses. Importantly, median fluid intake was identical in women who did and did not provide follow-up UI information (2.0 L/d); thus, there is unlikely any meaningful bias because of the small loss to follow-up evaluation.

Among the 83,997 NHS participants and 85,507 NHS II participants with data on UI at baseline, we excluded 40,807 NHS participants and 43,926 NHS II participants with prevalent UI , which was defined as UI at least once per month or UI less than once per month of quantities at least enough to wet the underwear. In addition, at the beginning of each 2-year follow-up period, we excluded women with missing data on UI incidence, fluid intake, or important potential confounding factors (ie, body mass index, parity, or caffeine intake). We also excluded women with major neurologic conditions (ie, stroke, multiple sclerosis, Parkinson’s disease, or amyotrophic lateral sclerosis) or functional limitations (which were defined as difficulty climbing a flight of stairs, walking 1 block, bathing, or dressing). Thus, 65,167 women (NHS, 34,143 women; NHS II, 31,024 women) were included in these analyses.

Measurement of incident UI

In both cohorts, participants were asked, “During the last 12 months, how often have you leaked or lost control of your urine?” Questions very similar to this have been proved valid previously for the assessment of UI. Response options were never, less than once per month, 2-3 times per month, about once per week, and almost every day. Women who reported UI were then asked, “When you lose your urine, how much usually leaks?” Response options were a few drops, enough to wet your underwear, enough to wet your outer clothing, and enough to wet the floor. A reliability study among 200 participants demonstrated high reproducibility of responses to these questions.

Incident cases were defined as those women who reported UI at least once per month on any of the follow-up questionnaires. Among incident cases, frequent incontinence was defined as incontinence that occurred at least once per week.

At the first follow-up evaluation, we assessed incontinence type with the use of a supplementary questionnaire was mailed only to women with frequent incontinence because we believed that it would be easier for them to identify the precipitating circumstances of their incontinence. The items that were included that assigned the type of UI had been validated previously. In the NHS, because of the large number of women with incident frequent UI, the supplementary questionnaire was mailed to a random sample of 80% of the women (n = 2183) and completed by 84%. In the NHS II, the supplementary questionnaire was mailed to 98% of incident cases (n = 1224; 19 women were identified after the supplementary questionnaire mailing was complete and thus did not receive a mailing) and was completed by 79%. For the second follow-up period, data on incontinence type were collected directly from the main questionnaire; therefore, information on incontinence type was available from 99% of women with frequent incontinence.

Stress UI was defined as leaking primarily with coughing or sneezing, lifting things, laughing, or exercise. Urgency UI was defined as primarily leaking that was accompanied by an urge to urinate or a sudden feeling of bladder fullness. UI type was classified as mixed when women reported that stress and urgency UI symptoms were equally common.

Measurement of beverage intake

In both cohorts, food and beverage data were collected repeatedly with validated semiquantitative food frequency questionnaires (FFQs). FFQs were administered in 1980, 1984, 1986, and every 4 years thereafter in the NHS and every 4 years starting in 1991 in the NHS II. On the FFQ, participants were asked how often on average during the previous year they drank 1 standard serving of 22 different, commonly consumed beverages. There were 9 response options that ranged from “none or <1 per month” to “>6 per day.” To calculate the average daily intake of each beverage, the serving size was multiplied by the reported frequency of use. Total fluid intake was calculated as liters per day by summing intakes of all beverages.

The reproducibility and validity of the NHS and NHS II FFQs have been reported previously. In a validation study among NHS participants, there were generally high correlations between beverage intake that were assessed with the FFQ and with four 1-week diet records that were completed over a 1-year period; correlations were 0.78-0.93 for coffee, 0.68-0.88 for milk, 0.40-0.84 for soda, 0.34-0.56 for punch (noncarbonated fruit flavored drinks, Hawaiian punch [Dr Pepper/Seven Up, Inc, Plano, TX], or lemonade), 0.75-0.89 for juice, and 0.52 for water (ranges because of different correlations for specific varieties of coffee, milk, soda). In an additional validation study, a reasonable correlation (r = 0.59) between fluid intake on the FFQ and 24-hour urinary volume was reported.

Statistical analysis

For the primary analyses, we defined daily fluid intake using the participant’s report from the most recent questionnaire before the development of incontinence. Because the distribution of fluid intake was very similar in the 2 cohorts, we categorized fluid intake into quintiles that were based on the distribution of intake in the combined cohorts.

Cox proportional hazards models that were stratified by age in months and a 2-year risk period were used to calculate multivariable-adjusted hazard ratios (HRs) and their 95% confidence intervals (CIs) for each case definition (ie, any, frequent, stress, urgency, and mixed UI). Covariates in the models were potential UI risk factors that were identified from the literature, including parity (0, 1-2, ≥3 births), body mass index (kilograms per square meter, continuous), cigarette smoking (never, past, current: 1-14, 15-24, 25-34, ≥35 cigarettes per day), race (white, black, Asian, other, or missing), diabetes mellitus, caffeine intake (milligrams per day, continuous), and physical activity (metabolic equivalent-hours per week, continuous). Further adjustment for diuretic use did not change the results; thus, we did not include this variable in the multivariable models.

We conducted analyses in each cohort separately and after pooling the data from the 2 cohorts. Before combining data, we tested whether the association between fluid intake and incident UI varied between the older women in the NHS and the younger women in the NHS II by using an interaction term in the Cox proportional hazards models. For each outcome, the interaction term was not significant (all probability values for interaction ≥0.16), which indicated that the association between fluid intake and UI was not different between cohorts. We present only results from the pooled analyses. Nonetheless, analyses were adjusted for study cohort in addition to the variables described earlier.

We conducted several secondary analyses to explore possible bias in fluid intake measurement and to examine the association between fluid intake and incident UI in more detail. First, we repeated the analyses using the average of the 2 previous reports of daily fluid intake (ie, from questionnaires up to 6 years before study baseline) rather than only the baseline report to decrease measurement error in a single report and minimize the impact of any changes in fluid intake that might have occurred in response to early incontinence symptoms. In addition, we examined the associations between specific beverage types and UI risk. For these analyses, we examined risks that were associated with a 240-mL (8-ounce) increase in intake of (1) milk (including skim and whole milk), (2) juice (including apple, orange, grapefruit, tomato, and other fruit juices), (3) tea and coffee (both caffeinated and decaffeinated), (4) soda, (5) punch and lemonade, (6) alcoholic beverages, and (7) water. To isolate the association between each beverage type and risk of incident UI, the multivariable model included all 7 beverage types simultaneously as well as the covariates listed earlier.

For all analyses, 2-tailed probability values < .05 were considered statistically significant. All data were analyzed with SAS software (version 9.1; SAS Institute Inc, Cary, NC).