We wanted to share this article as part of our commitment to keeping you and your family safe. While we do not carry hair products, it is important to have this kind of information when choosing products for your family and to understand long-term exposure to endocrine-disrupting chemicals.
Uterine cancer is one of the most common gynecologic cancers. Overall, incidence and mortality rates have increased in the United States in the past 2 decades, with more than 65 950 new cases and 12 550 deaths expected in 2022 (1,2). Exposure to excess estrogen and a hormonal imbalance of estrogen and progesterone have been identified as key risk factors for uterine cancer (3,4). Thus, it has been hypothesized that synthetic estrogenic compounds such as endocrine-disrupting chemicals (EDCs) could contribute to uterine cancer risk because of their ability to alter hormonal actions (5–11).
Hair product use, a predominant exposure pathway to various EDCs (12,13), has been associated with hormone-sensitive cancers including breast (14–19) and ovarian cancer (20–23) in previous epidemiologic studies. Hair product constituents, including formaldehyde (24–27) and formaldehyde-releasing chemicals (28–30) in some straighteners, and oxidized para-phenylenediamine and 4-aminobiphenyl in hair dyes (25,31–33), have also played a potential role in carcinogenesis, supporting an association between hair product use and cancer development.
Hair product use is common among women in the United States and Europe with more than 50% reporting ever using permanent hair dyes (19,22,34). In the Sister Study, we have previously observed a higher breast cancer incidence associated with adolescent (18) and adult use (19) of hair products and a higher ovarian cancer incidence associated with adult use of straighteners (22). However, to our knowledge, no study has investigated the influence of hair product use on uterine cancer. Therefore, this study aims to examine associations between hair product use and the age-specific hazard of uterine cancer in a large, racially and ethnically diverse cohort in the United States.
Methods
Study Population
The Sister Study is a prospective cohort that enrolled 50 884 women in 2003-2009. Participants were eligible if they were breast cancer–free women aged 35-74 years who had at least 1 sister diagnosed with breast cancer and if they lived in the United States, including Puerto Rico (35). At baseline, participants completed an interview and self-administered questionnaires that included questions about hair product use. Weight and height were measured during a home visit at baseline by trained examiners. Participants or next of kin for deceased participants are contacted annually for health updates regarding new cancer diagnoses and other health-related changes, and every 2-3 years for more detailed follow-up assessments. Response rates have been near 90% throughout follow-up (36). Data for the current analysis included person-time through September 2019 (Data Release 9.1). Written informed consent was obtained from all participants. This study is overseen by the institutional review boards of the National Institutes of Health.
We excluded women who withdrew from the study (n = 3), who self-reported a diagnosis of uterine cancer before enrollment (n = 380), had an uncertain uterine cancer history (n = 10), had an unclear timing of diagnosis relative to enrollment (n = 59), had a hysterectomy before enrollment (n = 15 585), who did not answer any hair product use questions (n = 736), and who did not contribute any follow-up time (n = 164), resulting in 33 947 eligible women.
Exposure Assessment
At baseline, participants were asked to complete a questionnaire on hair product use in the previous 12 months. Participants reported their frequency of personal use (application by themselves or others to their own hair) of 7 hair products including permanent, semipermanent, and temporary hair dyes; bleach; highlights; straighteners, relaxers, or pressing products; and hair permanents or body waves with the response options including “did not use,” “1-2 times per year,” “every 3-4 months,” “every 5-8 weeks,” “once per month,” and “more than once per month.” Additionally, frequency of nonprofessional application to others was collected for permanent hair dyes, semipermanent hair dyes, and straighteners, relaxers, or pressing products. Based on exposure distribution, we collapsed frequency variables as no more than 2 and more than 2 times per year for hair permanents and as no more than 4 and more than 4 times per year for the other hair product use among ever use. These exposure variables were also dichotomized to never and ever use. Color of dyes (“dark,” “light”) and lifetime duration of use (“did not use,” “less than 5 years,” “5-9 years,” “10 or more years”) were obtained for permanent and semipermanent hair dye use.
Outcome and Covariate Assessment
Uterine cancer cases were defined as women who reported a diagnosis of endometrial cancer, uterine sarcoma, or other types of cancer in the uterus after enrollment (n = 378). Women reporting a cancer diagnosis were asked to provide authorization to retrieve medical records. A total of 262 (69.3%) cases were confirmed using either medical records (n = 247) or death certificates indicating the primary or underlying cause of death as uterine cancer (n = 15). For those without medical confirmation, the information was obtained through self-report (n = 109) and next of kin (n = 7). The positive predictive value of self-reported uterine cancer cases in relation to medically confirmed cases is 85%.
Of 262 medically confirmed uterine cancer cases, 248 (94.7%) were classified as endometrial cancers using International Classification of Disease–10 code C54.1. Further, we used International Classification of Disease–Oncology-3 histology codes to define type I and type II endometrial cancer (37). Type I endometrial cancers are more hormone sensitive with greater estrogen receptor expression and tend to have better outcomes and survival, whereas type II cancers tend to be more clinically aggressive and have a poor prognosis (38–41).
Other covariates collected at baseline included self-reported age, race and ethnicity (African American/Black including Hispanic/Latina, Hispanic/Latina non-Black, non-Hispanic White, and all others, including Asian/Pacific Islander or American Indian), physical activity (metabolic equivalent hours per week), smoking status (never, past, or current), alcohol consumption (never or past, current <1 drink per day, current ≥1 drink per day), educational attainment (high school or less, some college, college and above), and occupational history working in beauty salons or barbershops. Reproductive history included age at menarche (younger than 13 years, 13 years and older), menopausal status (premenopausal and postmenopausal), parity (0, 1, 2, ≥3), oral contraceptive use (none, <2, 2 to <10, ≥10 years), hormone replacement therapy use (none, estrogen alone, estrogen plus progesterone but never estrogen alone). Body mass index (BMI) was calculated by height and weight measured at baseline (35).
Statistical Analysis
We estimated pairwise correlations using Spearman correlation coefficients among frequency metrics for hair product use. Associations between hair product use and uterine cancer were assessed using Cox proportional hazards models with age as the timescale. Women were followed from enrollment until uterine cancer diagnosis and were considered censored at the earliest event including hysterectomy, loss to follow-up, death, or end of follow-up. The 95% confidence intervals (CIs) were calculated using cluster-robust sandwich estimators to account for relations among participants (2872 women had >1 family member enrolled). Cox models were adjusted for an a priori selected list of confounders including race and ethnicity, educational attainment, BMI (restricted cubic spline with knots at the 5th, 35th, 65th, and 95th percentile, kg/m2), physical activity, menopausal status at enrollment, parity, smoking status, alcohol consumption, oral contraceptive use duration, hormone replacement therapy, and age at menarche. Proportional hazards assumptions were evaluated by Wald tests of covariate-by-attained-age interaction terms. Ptrend was assessed using Wald tests for continuous variables of hair product use frequency and duration.
We considered whether associations between hair product use and uterine cancer varied by race and ethnicity, obesity, and physical activity. Stratum specific hazard ratios (HRs) were estimated by augmenting our primary model with hair product-by-modifier interaction terms, and heterogeneity was tested using Wald tests. Hazard ratios by race and ethnicity were only estimated for African American/Black and non-Hispanic White participants because of the small number of women in other racial and ethnic groups. To estimate the cumulative risk of developing uterine cancer by age 70 years, we used the Breslow method to create baseline hazard functions for uterine cancer and competing risks, and a modified Aalen-Johansen estimator for cumulative risk for all participants, standardized to the covariate distribution of the study population (42). From this, we calculated absolute differences and numbers needed to harm for use of straighteners.
We also evaluated the associations by different cancer subtypes, including pre- and postmenopausal uterine cancers, endometrial cancer, and types I and II endometrial cancers. Women who were premenopausal at enrollment were at risk for premenopausal uterine cancer until menopause, at which time they were considered at risk for postmenopausal uterine cancer. Women became at risk for postmenopausal uterine cancer at either age at enrollment or age at menopause, whichever was later.
Several sensitivity analyses were conducted to evaluate the robustness of our findings. First, we restricted the outcome to medically confirmed uterine cancer cases. Second, we removed physical activity from the model because physical activity may act as a confounder and/or a mediator. Costly and time-consuming hairstyling practice may be a barrier to physical activity, whereas women engaging in intense physical activities may change their hairstyling practices (43,44). We excluded 1048 women who had worked in beauty salons or barbershops to eliminate the potential impact of occupational exposure. We considered simultaneous adjustment for other personal hair product use, including frequently used products and all products. Further, the first year of follow-up was excluded to assess the possibility of reverse causation. Finally, associations with use of straighteners or hair permanents as a combined exposure were estimated because these terms have been used interchangeably in African American and/or Black communities to describe chemical products used to change the texture of hair (45,46).
Because of sample size limitations, the frequency, stratified, and sensitivity analyses are only reported for the products more frequently used, including personal use of permanent dyes, semi-permanent dyes, straighteners, and hair permanents. Estimates with fewer than 5 cases in any stratum in the statistical model are not reported. Complete-case analyses were done with 557-1316 (1.6%-3.9%) women being excluded in the main analyses because of missingness in any variable. All analyses were conducted in R version 4.1.0.
Results
We followed participants for a mean of 10.9 years. The study population consists of 7.4% Black/African American, 4.4% Hispanic/Latina non-Black, 85.6% non-Hispanic White, and 2.5% all other race and ethnicity. Participants had high educational attainment on average (55.8% college degree or above). Compared with the cohort overall, uterine cancer cases tended to be older with an earlier age at menarche, a higher BMI, and lower physical activity (Table 1). The participants who ever used straighteners were mostly African American/Black (59.9%) and tended to be younger with a higher BMI and lower physical activity than those never used (Supplementary Table 1, available online). The frequencies of use of different hair products were weakly correlated with each other (correlation coefficients: −0.10 to 0.39).