Personal use of hair dye and cancer risk in a prospective cohort of Chinese women

Abstract

Although widely studied over the past 40 years, personal use of hair dye generally has not been associated with overall cancer risk. The association between hair dye use and risk of bladder and hematopoietic cancers has been less conclusive. Most hair dye studies have been case‐control studies conducted in Caucasian populations. We examined the relationship between personal hair dye use and cancer risk in a prospective cohort of 70 366 Chinese women. After an average of 7 years of follow up, 2437 women were newly diagnosed with cancer by 31 December 2005. Cox proportional hazard models were used to estimate relative risks (RR) and 95% confidence intervals (CI) of cancer risk associated with hair dye use, adjusting for potential confounding factors. Compared with women who reported no hair dye use, ever users had an overall cancer risk of 0.89 (95% CI 0.82, 0.97). No significant association was observed for several common cancers, including cancers of the breast (RR 0.93, 95% CI 0.78, 1.09), lung (RR 0.81, 95% CI 0.62, 1.09), stomach (RR 0.90, 95% CI 0.66, 1.21), and colorectum (RR 1.04, 95% CI 0.84, 1.28). We also found no significant association with most other cancers, including bladder cancer (RR 1.14, 95% CI 0.56, 2.35) and hematopoietic cancers overall (RR 0.89, 95% CI 0.59, 1.35) or their subtypes, including non‐Hodgkin lymphoma, multiple myeloma, and leukemia. We generally found no evidence of an association between personal use of hair dye and cancer risk, although our study is limited by small numbers for certain cancer types. (Cancer Sci 2009; 100: 1088–1091)

It is estimated that more than one‐third of women in the USA, Europe, and East Asia use hair dyes, with permanent hair dyes constituting the majority of the market share.^{( 1} , ^{2 )} Hair dyes have been known to contain aromatic amines, which were classified in 1993 by the International Agency for Research on Cancer (IARC) as probable bladder carcinogens in individuals with occupational exposure.^{( 2} , ³ , ^{4 )} However, both the 1993 and 2008 IARC working groups found that there was insufficient evidence to evaluate the carcinogenicity of personal use of hair dye.^{( 2} , ^{5 )} A recent comprehensive systematic review also concluded that the 25 studies conducted since the 1993 IARC decision do not provide sufficient evidence of a causal link between personal hair dye use and cancer.^{( 6 )}

Although widely studied over the past 40 years, personal use of hair dye generally has not been associated with cancer risk overall, or with risk of common cancers such as lung, breast, and colorectal cancer.^{( 6} , ⁷ , ⁸ , ^{9 )} The association between hair dye use and risk of bladder and hematopoietic cancers has been less conclusive.^{( 6} , ⁷ , ¹⁰ , ¹¹ , ¹² , ¹³ , ^{14 )} Most studies that have reported positive associations have been case‐control studies conducted in Caucasian populations. These studies have found higher risks mainly with long‐term use of dark hair dye^{( 10} , ¹¹ , ^{12 )} and hair dye use prior to 1980.^{( 15} , ¹⁶ , ^{17 )}

To further examine the association between personal hair dye use and cancer risks, we analyzed data in a prospective, population‐based cohort study of Chinese women in Shanghai, where mostly dark hair dyes are used.

Materials and Methods

Study population.  The details of the Shanghai Women's Health Study have been published previously.^{( 18 )} A roster of 81 170 women aged 40–70 years was obtained from the resident offices in seven communities located in urban Shanghai, China. A total of 75 221 (92.7%) women completed baseline surveys between 1996 and 2000. Of the 5949 non‐participants, 2407 (3.0%) refused to be interviewed, 2073 (2.6%) were absent during the study period, and 1469 (1.8%) did not participate due to other reasons including health, hearing, and speaking problems. Of the respondents, we excluded 279 women who were found to be younger than age 40 years or older than age 70 years, and 1576 women who had a prevalent cancer at baseline. A total of 73 366 women remained for this analysis.

Data collection.  After obtaining informed consent, eligible individuals first completed a self‐administered questionnaire, which included hair dye exposure information. Trained interviewers conducted in‐person interviews to review the completeness and accuracy of the self‐administered questionnaire, complete an additional questionnaire, collect blood and urine specimens, and perform anthropometric measurements at baseline. Respondents were asked, ‘In the past 3 years, have you ever used hair dye?’ Women who gave an affirmative answer were then asked, ‘Over the past year, how many times did you use hair dye?’ and ‘How many years have you used hair dye?’ Other information obtained in the baseline questionnaire included demographic information, personal medical history, family medical history, reproductive history, occupational history, physical activity, diet, and lifestyle factors including active smoking, exposure to environmental tobacco smoke, and alcohol use.

Newly diagnosed cancer cases were ascertained through biennial surveys and linkage to the Shanghai Cancer Registry, which is legally mandated in Shanghai, China. Information on the date and hospital of diagnosis was collected. Medical charts and pathology slides were collected from the diagnostic hospital and reviewed by two study cancer pathologists to verify the diagnosis. In addition, death certificate data were collected from the Shanghai Vital Statistics Unit to identify causes of death for deceased participants. All incident cancer cases identified by 31 December 2005 were included in the present analysis.

Data analyses.  We estimated the association between hair dye use and risk for all cancers combined, as well as the risk for several specific cancers. Cox proportional hazards modeling was used to compute relative risks (RR) and 95% confidence intervals (CI). The end time of the observation in the model was set as the age at diagnosis for incident cancer subjects or age at death for deceased subjects. For those who did not encounter any of the events mentioned above, the end time was set as 31 December 2005. The left truncation of the model was set to the age at baseline interview. The survival model was stratified by birth cohort intervals (5‐year) to control for age and cohort effects. Other potential confounding variables, including education and smoking status, were controlled for in the final model. Further adjustment for family history of cancer, passive smoking, alcohol consumption, mean fruit and vegetable intake, mean energy expenditure, measured body mass index (BMI), number of live births, hormone replacement therapy use, menopausal status, age at menarche, income, and marital status did not result in any material changes of the observed associations (less than 10% change in the RR), and were therefore not included in the final models. Analyses were carried out in the statistical program R (version 2.3.1, Free Software Foundation, Inc. Boston, MA, USA). All tests were two‐sided, with a significance level of 0.05.

Results

In this population, 29 076 (39.63% of the cohort) reported ever using hair dye (Table 1). Hair dye use ranged from 1 to 52 years, with a median of 3 years and a mean of 3.8 years of use (excluding non‐users). Older women in the cohort tended to have a longer duration of use. The mean years of use in women aged 55–60, 60–65, and 65–70 years were 4.4, 5.2, and 6.1 years, respectively. Approximately 9% of all hair dye users in the cohort (2689 individuals) reported 10 or more years of hair dye use. Hair dye users had a median age of 51 years whereas non‐users had a median age of 49 years. Hair dye users were more likely to be post‐menopausal (51.70%) than non‐users (47.11%). Those who used hair dye were also more likely to have used hormone replacement therapy (2.76% of hair dye users vs 1.61% of non‐users). Compared to non‐users, hair dye users had higher income levels and were more highly educated. Hair dye users were as likely to have smoked and been exposed to passive smoke as non‐users, but were slightly more likely to have consumed alcohol. Fewer than 3% of all women in this cohort smoked or consumed alcohol regularly.

Table 1.

Characteristics of hair dye users and non‐users

Characteristic	Users	Non‐users
Number	29 076	44 290
Percentage of cohort (%)	39.63	60.37
Median age at interview (years)	51	49
Post menopausal (%)	51.70	47.11
Baseline measured BMI (median)	23.40	23.05
Ever smoker (%)	2.90	2.71
Ever drinker (%)	2.52	2.08
Ever exposed to passive smoke (%)	85.46	84.27
Married (%)	89.78	88.19
Ever used HRT (%)	2.76	1.61
Live births (%)
0	2.96	3.51
1	53.46	55.75
2	25.18	18.42
3	18.40	22.32
Educational level (%)
Elementary school or less	16.99	24.28
Middle school	37.23	37.14
High school	29.52	26.89
College	16.26	11.69
Income (%)
<10 000 RMB	13.68	17.69
10 000–<20 000 RMB	36.47	39.43
20 000–<30 000 RMB	28.99	27.49
≥30 000 RMB	20.86	15.39
Mean vegetable intake (g/day)	302.52	295.11
Mean fruit intake (g/day)	285.71	260.73
Mean cumulative energy expenditure (kJ/min)	232.18	235.26

A total of 2437 incident cancer cases were identified during follow up through 31 December 2005. Compared with women who never used hair dye, ever users had a RR of 0.89 (95% CI 0.82, 0.97) for subsequent cancer (Table 2). We did not find an association between hair dye use and the risk of bladder cancer (RR 1.14, 95% CI 0.56, 2.35), hematopoietic cancers overall (RR 0.89, 95% CI 0.59, 1.35) or their subtypes, including non‐Hodgkin lymphoma (NHL), multiple myeloma, and leukemia. No significant association was observed for several common cancers, including cancers of the breast (RR 0.93, 95% CI 0.78, 1.09), lung (RR 0.81, 95% CI 0.62, 1.09), stomach (RR 0.90, 95% CI 0.66, 1.21), and colorectum (RR 1.04, 95% CI 0.84, 1.28). However, a significant inverse association was observed for thyroid cancer, with an estimated risk reduction of nearly 60% among hair dye users (RR 0.42, 95% CI 0.25, 0.69).

Table 2.

Relative risks ^* (RR) and 95% confidence intervals (CI) for the association of hair dye use (ever/never) with risk of major cancer sites in the Shanghai Women's Health Study

Cancer site	No. non‐user cases †	No. user cases ‡	RR	95% CI
All cancer	1536	901	0.89	0.82, 0.97
Brain	24	15	0.96	0.49, 1.84
Thyroid	68	20	0.42	0.25, 0.69
Breast	358	234	0.93	0.78, 1.09
Lung	164	85	0.81	0.62, 1.09
Stomach	121	67	0.90	0.66, 1.21
Pancreatic	52	27	0.88	0.55, 1.41
Liver	67	33	0.83	0.54, 1.27
Colorectal	234	156	1.04	0.84, 1.28
Kidney	31	23	1.11	0.64, 1.92
Bladder	19	13	1.14	0.56, 2.35
Ovarian	62	38	0.89	0.59, 1.35
Uterine	71	54	1.10	0.77, 1.58
Hematopoietic	64	35	0.89	0.59, 1.35
Non‐Hodgkin lymphoma	31	20	1.09	0.61, 1.92
Multiple myeloma	12	6	0.84	0.31, 2.27
Leukemia	20	9	0.68	0.31, 1.51

^* RR were derived using Cox proportional hazards models, adjusted for age, education, and smoking duration in pack/years. The average number of person years was 7.31.

^† The number of non‐user controls was 42 739.

^‡ The number of user controls was 28 166.

We found no consistent pattern between cancer risk and years of hair dye use for all cancers combined, or for any individual cancers (Table 3). An excess risk of borderline significance was observed for ovarian cancer among those who used hair dye for 5–9 years (RR 1.67, 95% CI 0.87, 3.19), and for bladder cancer among hair dye users of 3–4 years (RR 2.16, 95% CI 0.89, 5.22). For thyroid cancer, a 50–60% reduction in risk was associated with hair dye use across all duration of use categories, with the associations reaching significance for short‐term users.

Table 3.

Relative risks ^* (RR) and 95% confidence intervals (CI) for duration of hair dye use and risk of major cancer sites in the Shanghai Women's Health Study

Cancer site	No. non‐user cases	1–2 years			3–4 years			5–9 years			≥10 years
		No. user cases	RR	95% CI	No. user cases	RR	95% CI	No. user cases	RR	95% CI	No. user cases	RR	95% CI
All cancer	1536	386	0.89	0.79, 0.99	237	0.81	0.71, 0.93	161	0.96	0.82, 1.14	116	0.90	0.74, 1.09
Brain	24	7	1.03	0.44, 2.42	2	0.45	0.11, 1.95	3	1.16	0.35, 3.92	3	1.49	0.44, 5.02
Thyroid	68	10	0.45	0.23, 0.87	5	0.35	0.14, 0.87	3	0.40	0.13, 1.28	2	0.44	0.11, 1.83
Breast	358	105	0.90	0.72, 1.12	66	0.87	0.66, 1.13	37	0.91	0.65, 1.29	26	1.00	0.67, 1.50
Lung	164	30	0.70	0.47, 1.04	21	0.70	0.44, 1.11	14	0.78	0.45, 1.35	20	1.29	0.81, 2.06
Stomach	121	29	0.94	0.62, 1.41	16	0.76	0.45, 1.29	17	1.35	0.81, 2.25	5	0.47	0.19, 1.15
Pancreatic	53	14	1.17	0.64, 2.12	6	0.70	0.30, 1.64	3	0.54	0.17, 1.74	3	0.55	0.17, 1.78
Liver	67	14	0.85	0.47, 1.52	12	1.10	0.59, 2.05	5	0.76	0.31, 1.91	2	0.35	0.09, 1.43
Colorectal	234	66	1.09	0.83, 1.44	42	1.00	0.71, 1.39	30	1.14	0.78, 1.68	18	0.79	0.49, 1.28
Kidney	31	13	1.46	0.76, 2.81	5	0.82	0.32, 2.13	1	0.29	0.04, 2.13	4	1.50	0.52, 4.32
Bladder	19	3	0.59	0.17, 2.00	7	2.16	0.89, 5.22	2	1.14	0.26, 4.95	1	0.75	0.10, 5.66
Ovarian	62	10	0.51	0.26, 0.99	11	0.87	0.46, 1.66	11	1.67	0.87, 3.19	6	1.44	0.61, 3.38
Uterine	71	26	1.18	0.75, 1.85	16	1.08	0.62, 1.87	7	0.88	0.40, 1.92	5	0.98	0.39, 2.44
Hematopoietic	64	12	0.70	0.37, 1.30	13	1.18	0.65, 2.16	7	1.12	0.51, 2.46	3	0.59	0.18, 1.90
Non‐Hodgkin lymphoma	31	8	1.03	0.47, 2.26	7	1.36	0.59, 3.13	3	0.99	0.30, 3.28	2	0.74	0.17, 3.11
Multiple myeloma	12	2	0.69	0.15, 3.14	2	1.03	0.23, 4.65	2	1.63	0.36, 7.37	0
Leukemia	20	2	0.31	0.07, 1.31	4	1.05	0.36, 3.12	2	1.05	0.24, 4.56	1	0.89	0.12, 6.76

^* RR were derived using Cox proportional hazards models, adjusting for age, education, and smoking duration in pack/years. The average number of person years was 7.31.

We further examined cancer risks among 486 women who reported using hair dye for 20 or more years. After adjusting for age and education, the RR for all cancers combined in this group was 1.00 (95% CI 0.67, 1.49). A significant excess risk among hair dye users was found for ovarian cancer (RR 5.95, 95% CI 2.13–16.68; N = 4 cases). No significant association was observed for other cancers (data not shown).

We carried out additional sub‐analyses for selected cancers. For lung cancer, we stratified results by smoking status but found no significant association between hair dye use and cancer in either smokers or non‐smokers. When we stratified by menopausal status, we found no association between breast or ovarian cancer and hair dye use in either pre‐ or post‐menopausal women.

We carried out additional tests to determine whether frequent hair dye users (>12 times per year) had a different RR of cancer than less‐frequent users. The difference in RR between these two groups was very slight. Finally, we excluded cases diagnosed during the first 2 years of follow up and found no material difference in the overall risk of cancer.

Discussion

In this large prospective study, we generally found no association between hair dye use and risk of overall cancer. We found no evidence that hair dye use increases the risk of bladder cancer or hematopoietic cancer, the two malignancies most frequently linked with hair dye use in the literature. A significant increased risk for ovarian cancer was associated with long‐term hair dye use (≥20 years), but the result was based on a very small number of cases. We have no explanation for the significant inverse association observed for thyroid cancer, but chance variation or failure to adjust for confounding factors not measured in our study are possible explanations. Although the lack of a dose response in risk with duration of hair dye use tends to argue against a causal relationship, further assessment of the link between hair dye use and thyroid cancer risk may be warranted.

Our findings are consistent with previous cohort and case‐control studies that have generally reported no association or a non‐significant inverse association between hair dye use and cancer.^{( 8} , ⁹ , ¹⁰ , ¹¹ , ¹⁹ , ^{20 )} The most recent systematic review concluded that the 25 studies conducted since the 1993 IARC review do not provide sufficient evidence of a causal link between hair dye and cancer.^{( 6 )} The CPS‐II study, which was the largest cohort to examine this association, initially reported a statistically significant inverse association almost identical to our findings;^{( 11 )} however, no association was observed after additional follow up.^{( 10 )}

Despite several studies that have reported null associations;^{( 8} , ¹⁹ , ²¹ , ^{22 )} hair dye has continued to be a concern because a number of case‐control studies and a meta‐analysis have found links between personal hair dye use and bladder cancer.^{( 23} , ²⁴ , ^{25 )} Hair dye use has also been suggested as a risk factor for NHL,^{( 9} , ¹⁰ , ¹² , ^{17 )} leukemia,^{( 26} , ²⁷ , ^{28 )} multiple myeloma,^{( 12} , ^{29 )} and myelodysplasia,^{( 30 )} although several other studies found no association with hematopoietic cancers.^{( 15} , ²⁰ , ³¹ , ^{32 )} Statistically significant increases in risk have been found in case‐control studies of hair dye use and glioma,^{( 33} , ^{34 )} ovarian cancer,^{( 35 )} and cancer of the salivary gland.^{( 36 )} A recent meta‐analysis concluded that there was no strong evidence of an association between hair dye use and risk of breast, bladder, and hematopoietic malignancies; however, this analysis did not consider specific exposure measurements such as duration, color, and frequency of use.^{( 7 )}

Long‐term use of dark dye has been associated with increased cancer risk in a few studies.^{( 10} , ^{11 )} However, these studies have had limited power due to the low prevalence of long‐term exposure to dark hair dyes.^{( 13} , ^{14 )} Most studies have been conducted in Caucasian women who use a range of light and dark colors, and where users of black dye have generally accounted for less than 1% of study participants.^{( 11} , ³³ , ^{34 )} Two studies in a Japanese population reported associations between hair dye use and myelodysplastic syndrome.^{( 30} , ^{37 )} Although we did not collect data on hair dye color, it is likely that our study population of mostly middle‐aged and elderly Chinese women used primarily dark or black dye. However, we were unable to obtain data published in the scientific literature documenting the use of hair dye by type in China. To the extent that this population used primarily dark hair dye, a type that has been most commonly linked to cancer risk, our observation of no association even among long‐term users suggests that personal hair dye use is probably not a strong cancer risk factor.

Recently, a few studies have found associations between hair dye use and NHL and bladder cancer in subsets of the study population, such as individuals who are N‐acetyltransferase 2 slow acetylators,^{( 25 )} N‐acetyltransferase 2 rapid acetylators,^{( 16 )} and individuals who used hair dyes prior to 1980.^{( 15} , ¹⁶ , ^{17 )} Our analysis did not evaluate associations in susceptible subgroups of the population, although overall we found no association among hair dye users of 20 or more years. Whether hair dye use increases cancer risk in certain susceptible subgroups and whether particular chemical components of hair dyes pose more risk than others may warrant further investigation.

Our study was limited by the brief nature of the questions on hair dye use in the baseline questionnaire. No questions were asked about the color or type of hair dye used (e.g. permanent, semipermanent). It is also possible that our data underestimates exposure to hair dye because women who had not used hair dye in the 3 years prior to interview were classified as non‐users, even if they used hair dye previously. This misclassification is likely non‐differential and may tend to bias the risk estimates towards the null. Despite the large number of hair dye users in our cohort, the number of cases of specific cancers was small, thus limiting the possibility for a more detailed analysis. The small numbers also produced unstable risk estimates for some less common cancers.

In conclusion, we found no evidence of an association between personal use of hair dye and overall cancer risk in this large prospective study in Chinese women. Further studies are warranted to clarify whether personal use of hair dye increases the risk of specific cancers in susceptible individuals.