Study population and data collection

The NHS II was a prospective cohort study that began in 1989; 116,608 registered female US nurses of ages 25-42 years who resided in 1 of 14 states were enrolled. The study was designed to examine prospectively the effects of oral contraceptive use and other lifestyle factors on chronic diseases, particularly cancers and cardiovascular diseases. Since 1989, these predominantly white nurses have completed biennial mailed questionnaires that include items about their health status, medical history, and known or suspected risk factors for cancer, heart disease, and other diseases. The questions include age, age at menarche, parity, age at first birth, weight, height, menopausal status, family history of breast cancer, and personal history of benign breast disease and cancer, among others. Every 2 years, follow-up questionnaires are sent to cohort members to update the information on potential risk factors, identify newly diagnosed case subjects, and record other major medical events. Further details of the cohort are described elsewhere. Response rates to the NHS II questionnaires have been at 90% throughout the follow-up period. This research was approved by the Institutional Review Board of Brigham and Women’s Hospital.

Case ascertainment and analytic definition

In 1993, the women were first asked whether they had “ever had physician-diagnosed endometriosis.” If the answer was “yes,” they were asked to report when the diagnosis had occurred (before September 1989, September 1989-May 1991, and June 1991-May 1993, which correspond to the follow-up periods) and whether it had been confirmed by laparoscopy, which is a standard surgical method for diagnosing endometriosis. These questions were asked again in each subsequent questionnaire.

As previously described, in March 1994, we conducted a study to validate self-reported endometriosis diagnosis within the NHS II cohort. Supplementary questionnaires were mailed to 200 women who were selected randomly from the then 1766 cases who had reported incident diagnosis. Among those who reported laparoscopic-confirmation and for whom records were received and reviewed (n = 105 women), a laparoscopic diagnosis of endometriosis was confirmed in 96%. However, among those women without laparoscopic confirmation (n = 26), evidence of clinical diagnosis was found in only 54% of the records. Severity data (defined by the staging system outlined by the American Society for Reproductive Medicine) suggested that most laparoscopically confirmed cases (61%) had minimal or mild disease. Requests for permission to review medical records were also sent to any woman who indicated that she had had a hysterectomy during the 2-year interval of reported endometriosis diagnosis. A diagnosis of endometriosis at the time of surgical procedure was confirmed in 80% (n = 144/181) of the records that were received. However, endometriosis was the primary indication for hysterectomy in only 6% (n = 9/163) of women for whom indication information was available.

Based on these validation results, self-reported physician-diagnosed endometriosis without laparoscopic confirmation may be misclassified substantially. In addition, allowing women who report endometriosis and a hysterectomy in the same follow-up period to be cases might yield spurious results, because it would be unclear whether the associated risk factors would be related to endometriosis or to the pathologic condition for which the hysterectomy was performed. Therefore, to reduce the magnitude of misclassification and prevent confounding by indication for hysterectomy, analyses of incident diagnosis of endometriosis were restricted to those women who reported laparoscopic confirmation of their diagnosis.

Within this restricted case definition, the relation between endometriosis and infertility status is complex. At baseline, the prevalence of infertility (defined as an attempt to become pregnant for >1 year without success) was greater among women with laparoscopic confirmation (20%) than among those who were clinically diagnosed without laparoscopic confirmation (4%). Approximately 20% of all infertile women are found to have endometriosis. Although pelvic pain information is not available in the NHS II cohort, cases of women with no infertility who have had a laparoscopic diagnosis are more likely to have experienced pelvic pain symptoms, otherwise an invasive surgical evaluation would not have been conducted. In addition, among cases with infertility, we may also assume that, had these women not attempted to become pregnant, a large proportion may never have received a laparoscopic diagnosis of endometriosis. Under these assumptions, we believe that endometriosis with infertility may be indicative of asymptomatic disease that result from other primary causes of infertility and the risk factors for endometriosis with infertility could differ from those for endometriosis without concurrent infertility. In addition, the setting of infertility evaluation introduces the potential for detection/diagnostic bias. Hence, we looked at risk factors separately by these 2 “subtypes” of endometriosis in a dichotomous outcome fashion: (1) women who had never reported infertility before their endometriosis diagnosis and (2) women with concurrent infertility. In addition, we alternatively treated infertility as a typical effect modifier, stratifying the entire study population by report of having had a clinical evaluation for infertility. Within this cohort, self-reported infertility was validated in a study of 100 randomly selected women who reported ovulatory infertility; 95% of the self-reports were confirmed through medical record review.

Assessment of nightshift working status

Detailed information on total years during which study participants had worked on rotating nightshifts was available from the 1989 questionnaire, with updates in 1991, 1993, and 1997. The 1991 questionnaire collected information about the total number of months during which the nurse had worked rotating nightshifts for at least 3 nights per month, in addition to having worked days or evenings in that month (since June 1989). The prespecified categories for total numbers of months working on rotating nightshifts were 0, 1-4, 5-9, 10-14, 15-19, and ≥20. Identical questions were posed in 1993 and 1997 (since June 1989 and June 1995, respectively). Nightshift work information was assessed retrospectively for the time-periods 1993-1995 and 1997-1999; nurses were asked on the 2001 questionnaire about the number of months they had worked on rotating nightshifts during these time periods. Specifically, nurses were asked “During the following time periods, how many months have you worked rotating nightshifts (at least 3 nights/month in addition to other days and evenings in that month)?” and “Did you work permanent nightshifts for ≥6 months during any of these time periods?”

Statistical analysis

Data for these analyses were collected in the NHS II cohort from September 1989 to June 2005. Participants who reported the diagnosis of endometriosis or a history of infertility before September 1989 were excluded from all analyses. We excluded women with a history of infertility because of the strong correlation between infertility and a diagnosis of endometriosis by laparoscopy and to reduce the potential for a healthy worker effect. Alternatively, we conducted analyses that did not censor at the report of infertility evaluation (at baseline or during the follow-up period, as described later in this article) nor polytomize the women by mode of diagnosis (indication of surgical evaluation during an infertility work-up vs not), but rather treated infertility as a typical effect modifier, thus stratifying the entire study population into groups who had never or ever had reported having had a clinical evaluation for infertility.

Analyses were also restricted to those who were premenopausal and had intact uteri, because the occurrence of endometriosis after hysterectomy or in postmenopausal women is rare. Women with previous cancer diagnoses, other than nonmelanoma skin cancer, and women who did not report their rotating nightshift work history on the baseline 1989 questionnaire were also excluded. Person-months at risk were calculated from entry into the cohort until there was an independently confirmed death or cancer diagnosis (other than nonmelanoma skin cancer), or self-reported laparoscopically confirmed endometriosis diagnosis, hysterectomy, or the onset of menopause. Women who reported physician-diagnosed endometriosis with no laparoscopic confirmation were censored at the time of that report but were allowed to reenter the analysis population with their interim person-time, if they reported laparoscopic confirmation on a subsequent questionnaire. In addition, because infertility is correlated so strongly with diagnosis of endometriosis by laparoscopy, to minimize detection bias, we censored at self-report of clinical infertility evaluation. Therefore, in our primary analysis, our comparison group consisted of women with neither diagnosed endometriosis nor infertility, which allowed for a more homogeneous comparison group, as we have previously described in detail. In a secondary analysis, we included only women who had had an infertility evaluation to equalize the possibility of secondary detection between cases and control subjects.

Because rotating nightshift workers are less likely to entrain their circadian rhythms than are permanent nightshift workers, women who reported having worked >6 months of permanent nightshift work were excluded. Rotating shift work information was updated by the use of baseline 1989 information on total number of years having worked rotating nightshift work up to then and adding months as reported on subsequent questionnaires.

Incidence rates for each exposure category were computed as the number of incident cases divided by the person-time accumulated. Time-varying Cox proportional hazards models that treated age in months and a 2-year questionnaire period as the time scale were used to estimate multivariate incidence rate ratios (RRs) and to calculate 95% confidence intervals (CIs), after simultaneous adjustment for confounding variables. Tests for trend in ordinal categoric exposures were calculated by creating an ordinal variable in which the median value or midpoint of each category was assigned to all participants in that group. Tests for heterogeneity that compared the effect estimates among the cases of women who never reported infertility with effect estimates among those women with concurrent infertility were calculated with a Wald statistic referred to a chi-squared distribution with 1 degree of freedom.

To evaluate whether the nightshift work and endometriosis associations varied by levels of other risk factors, stratified analyses were conducted, and likelihood ratio tests that compared the model with both the main effects and the interaction terms to that with the main effects only were calculated.