The NHS is a cohort of 121,700 registered nurses recruited between the ages of 30 and 55 years in 1976 and followed up via biennial mailed questionnaires (Colditz et al. 1997
). Our study population was drawn from two subsamples of the NHS cohort that was evaluated for lead exposure. The first is a sample of women who participated in a case–control study of lead exposure and hypertension (Korrick et al. 1999
). Women were invited to participate if they lived in the Greater Boston metropolitan area, Massachusetts, were not obese (body mass index < 29 kg/m2
), and did not have a history of major, chronic disease—mental health problems were not an exclusion criterion. Women who first reported a diagnosis of hypertension between 1990 and 1994 were invited to participate as cases; women free of major chronic disease participted as controls and were frequency matched to cases by 5-year age groups. In total, between 1993 and 1995, 302 NHS participants agreed to participate.
The women in the second sample were recruited for a study of lead exposure and osteoporosis. Similar eligibility criteria used for controls in the hypertension study were applied, with participants being free of chronic diseases (not including mental health problems) up to the time of recruitment (2001 through 2004), at which time they underwent their lead exposure measurements. This sample comprised 320 NHS participants. Altogether, lead content was measured in cortical (tibia) and trabecular (patella) bone in 621 and 620 women, respectively, from these two studies. After providing a complete description of the study to the participants, written informed consent was obtained before participation in each substudy. The present study was approved by the institutional review boards of the Brigham and Women’s Hospital and the Harvard School of Public Health.
Measures of psychological symptoms were obtained as part of several of the regular biennial NHS mailed questionnaires. Depressive symptoms were measured using the Mental Health Index 5-item (MHI-5) subscale, which was included on the 1992, 1996, and 2000 questionnaires. Anxiety symptoms were measured using the phobic anxiety scale of the Crown-Crisp Index (CCI) which was included on the 1988 and 2004 questionnaires. Of the 621 women who participated in the lead exposure studies, 617 completed at least one psychological symptom assessment. Of these 617 women, 613 (99.4%) completed at least one MHI-5 questionnaire, and 609 (98.7%) completed at least one CCI questionnaire. Individual women completed the MHI-5 in up to three separate study cycles for a total of 1,696 MHI-5 assessments, and the CCI in up to two separate study cycles for a total of 1,117 CCI assessments. We excluded 26 MHI-5 assessments (1.5%) and 17 CCI assessments (1.5%) for which data were missing on pack-years of smoking (n = 10 women). Thus, the final analytic sample included 603 women with 1,670 MHI-5 assessments (24 women with one, 91 with two, and 488 with three), and 599 women with 1,100 CCI assessments (98 women with one and 501 with two; ). All responses to the 2004 CCI (n = 554) occurred after bone lead measurement, whereas all responses to the 1988 CCI (n = 546) occurred before the bone lead measurements (mean ± SD, 10.0 ± 4.2 years; range, 4.4–15.9 years). Of the MHI-5 assessments, 1,102 (66.0%) occurred before the bone lead measurements (5.1 ± 3.5 years earlier; range, 0.4–12 years). Because bone lead reflects many years of past exposure, we included all psychological assessments in our primary analyses regardless of their timing with respect to bone lead measurement. Nonetheless, we also did sensitivity analyses restricted to those psychological assessments that occurred after bone lead measurement.
Timeline of MHI-5 and CCI assessments among women with bone lead measurements from study subsample 1 (A) and subsample 2 (B), as described in “Materials and Methods.”
Lead exposure assessment.
Participants visited the outpatient General Clinical Research Center of the Brigham and Women’s Hospital for measurement of lead content in their bone by K-shell X-ray fluorescence (KXRF) (Aro et al. 1994
), expressed as micrograms of lead per gram of bone mineral (micrograms per gram). Bone lead measurements were taken at each woman’s midtibial shaft and patella. These sites are targets for bone lead research, because the half-life of lead in tibia and patella differs. In a cohort of older men, the half-life of lead in patella has been estimated to be on the order of years, whereas in tibia it is estimated to be on the order of decades (Wilker et al. 2011
). Among women, however, faster bone turnover (Seeman 2003
) likely makes these half-lives shorter.
When we began measuring the women’s bone lead, we used an instrument developed by ABIOMED Inc. (Danvers, MA) (Burger et al. 1990
). In 1999, we replaced our prototype ABIOMED instrument with an upgraded instrument designed to improve measurement precision (Aro et al. 1994
). Intercalibration data from persons who were measured on both instruments demonstrated a linear relationship between the two measurements with a slope of 0.87. Using this correction factor, we are able to combine data from our prototype and upgraded KXRF machines (Nie et al. 2008
). To reduce the impact of any additional scaling differences in these readings on our epidemiologic analyses, we included in all of our bone lead regression models a term for lead substudy, which effectively adjusts for instrument, because women from the hypertension substudy were assessed on the ABIOMED instrument, and women from the osteoporosis substudy were assessed on the upgraded instrument.
Psychological symptom assessment.
Symptoms of depression and anxiety were assessed with the MHI-5 and the CCI, respectively, each of which has been validated and has an extensive history in research and been used in populations of a similar age as ours (Albert et al. 2005
; Beusterien et al. 1996
; Burgess et al. 1987
; Friedman et al. 2005
). The MHI-5 is a five-item subscale derived from the Short Form-36 health status survey designed to capture, among other aspects of psychological functioning, psychological distress versus well-being (Ware and Sherbourne 1992
). The MHI-5 asks respondents how much of the time over the past month (all, most, good bit, some, little, or none; ranked 1–6) they felt nervous, felt so down that nothing could cheer them up, felt calm and peaceful, felt down and blue, or felt happy. The two positively worded questions are reverse coded so that lower scores indicate more depressive symptoms; all item scores are summed, and then the sum is rescaled to obtain a total score ranging from 0 to 100 (Ware et al. 2000
). An MHI-5 score < 60 denotes the presence of severe depressive symptoms and predicts major depression as identified using the Mini-International Neuropsychiatric Interview Major Depressive Episode module with high sensitivity and specificity among adults ≥ 65 years of age (Friedman et al. 2005
). We also used an alternative cutoff score of < 53, which has been validated for identifying major depression in younger populations (Berwick et al. 1991
; Holmes 1998
The phobic anxiety scale of the CCI measures personality symptoms of phobic anxiety (Crown and Crisp 1966
). It is a brief self-rating inventory of eight questions on common phobias—such as fear of enclosed spaces, illness, going out alone, heights, and crowds—with two yes or no questions (scored 2 or 0) and six three-level response questions (e.g., never/sometimes/often or not at all/moderately/very; scored 0/1/2). Scores range from 0 to 16, with higher scores corresponding with higher levels of phobic anxiety. The CCI has been validated in psychiatric outpatient clinic settings and found to discriminate patients with anxiety disorders and agoraphobia from those with other mental health disorders (Mavissakalian and Michelson 1981
). The validity of the phobic anxiety subscale of the CCI in the NHS population has been tested previously, where scores were shown to be associated with use of tranquilizer medications (Albert et al. 2005
). A CCI score of ≥ 4 has been related to several adverse outcomes in the NHS population, for example, Parkinson’s disease (Weisskopf et al. 2003
), coronary heart disease, and sudden coronary death (Kawachi et al. 1994
; Whang et al. 2009
We used the generalized linear model framework to analyze repeated outcome measurements and used an unstructured covariance matrix to account for correlations in scores within individuals. Primary analyses treated MHI-5 scores as a continuous variable. Because the distribution of scores on the CCI was skewed, we used generalized estimating equations to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for scoring ≥ 4 on any CCI measurement (i.e., high phobic anxiety; n
= 188 women, 244 assessments). We conducted additional analyses using this same approach to estimate ORs for scoring < 60 on the MHI-5 (i.e., severe depressive symptoms; n
= 96 women, 121 assessments) or the alternative cutoff of < 53. When using dichotomous outcomes, additional sensitivity analyses were performed that included as “cases” women taking antidepressant or antianxiety medication who did not meet criteria for major depression by MHI-5 scores (n
= 63 women; 84 observations) or high phobic anxiety by CCI scores (n
= 56 women; 56 observations)—along with women who were cases based on their symptom scores. For women missing one or two CCI responses (n
= 55 women, 57 observations) or one MHI-5 response (n
= 44 women, 46 observations), we used the standard approach of imputing the total score for the scale by dividing their score by the fraction of questions answered and rounding to the nearest integer (DeVellis 1991
; Ware et al. 2000
We specifically addressed the potential impact of menopause status in separate analyses. Menopause is associated with increased bone turnover that mobilizes stored lead from bone into circulating blood, increasing blood lead levels (Korrick et al. 2002
). An increased rate of bone remodeling means that the lead concentration in bone is increasingly in flux during this period, which may compromise the utility of the bone lead concentration as a measure of cumulative lead exposure. Bone turnover with menopause is reduced in women who have taken HRT (Nelson et al. 2002
). Therefore, we conducted additional analyses restricted to women either who were premenopausal at bone lead measurement (n
= 45) or who had consistently taken HRT between menopause and bone lead measurement (n
= 97). Menopausal status was based on each woman’s self report on the regular NHS biennial questionnaires of permanent cessation of natural menses, as well as the age at which this occurred. For women who underwent hysterectomy without bilateral oophorectomy, a life table was used to assign an age at menopause based on the date of surgery, smoking, and hormonal status.
We performed separate analyses for patella and tibia bone lead biomarkers. Bone lead tertiles were calculated based on the distributions among women who were premenopausal or consistently on HRT between menopause and bone lead measurements. Covariates in our models were obtained from biennial questionnaires and included age (years) at psychological symptom measurement, education (registered nurse, bachelor’s degree, master’s or doctorate degree), alcohol consumption (tertiles of grams per day: < 1.8, 1.8–8.69, ≥ 8.7), and pack-years of smoking (0 and tertiles among smokers: 1–8.9, 9–21.9, ≥ 22), because these variables have been associated with lead exposure and psychological symptoms (Bjelland et al. 2008
; Boden and Fergusson 2011
; Hu et al. 1996
; Kessler et al. 1994
; Korrick et al. 2002
; Theppeang et al. 2008
; Wiesbeck et al. 2008
). We also included husband’s education (high school or less, college education, graduate school, or not married) and employment status at psychological symptom measurement (retired/homemaker, part-time worker, full-time worker) as indicators of socioeconomic status, which may be related to both lead exposure and psychological symptoms (Kessler et al. 1994
; Theppeang et al. 2008
). We adjusted for substudy to account for both the different KXRF machines used in the two substudies, as well as secular trends in lead exposure and possibly psychological symptom reporting, and age (years) at bone lead measurement, which is related to measured bone lead concentrations (Wilker et al. 2011
). About half of the participants were women from a case–control study of hypertension. Because the selection of hypertension cases and controls could have introduced selection bias, we conducted additional sensitivity analyses adjusting for hypertensive status in this substudy. Hypertension cases were identified based on self-reported physician diagnosis on the NHS biennial questionnaires. We also conducted sensitivity analyses adding adjustment for whether the nurse’s parents owned their own home when she was born, an additional socioeconomic status indicator that reflects status in childhood. The women responded to items on some covariate data in more than one questionnaire cycle, and for our analyses, we used responses from the NHS questionnaire closest to the psychological symptom assessment. We used missing indicator variables for the small amount of missing data. Tests for trend with increasing lead concentration were computed by including a continuous term for lead that was formed by assigning to each woman the median lead concentration of the tertile in which she was classified. This approach minimizes the influence of extreme exposure values. A two-sided p
-value ≤ 0.05 was considered statistically significant. We conducted all analyses in SAS (version 9; SAS Institute Inc., Cary, NC).