Prospective Study of Leukocyte Count as a Predictor of Incident Breast, Colorectal, Endometrial, and Lung Cancer and Mortality in Postmenopausal Women

Background  The immune system and inflammation are implicated in the pathogenesis of cancer. Prospective studies linking biomarkers of inflammation with cancer incidence and mortality have been inconclusive.

Methods  To determine whether there is an independent association of white blood cell (WBC) count with incident cancer in postmenopausal women, a prospective cohort study was performed at 40 US clinical centers involving 143 748 postmenopausal women aged 50 to 79 years who were free of cancer at baseline and were enrolled in the Women's Health Initiative. The main outcome measures were incident invasive breast, colorectal, endometrial, and lung cancer.

Results  In multivariate models, there was a graded association of WBC count with incidence of all 4 types of cancer. Compared with the lowest quartile of WBC count (2.50-4.79×10⁹ cells/L), women with a WBC count in the upper quartile (6.80-15.00×10⁹ cells/L) had a statistically significantly higher risk of invasive breast cancer (hazard ratio [HR], 1.15; 95% confidence interval [CI], 1.04-1.26), colorectal cancer (HR, 1.19; 95% CI, 1.00-1.41), endometrial cancer (HR, 1.42; 95% CI, 1.12-1.79), and lung cancer (HR, 1.63; 95% CI, 1.35-1.97). The findings were similar when cancers that occurred during the first 2 years of follow-up were excluded. Statistically significant associations remained for invasive breast cancer and endometrial cancer when the analyses were limited to nonsmokers. The WBC count was also statistically significantly associated with breast cancer, lung cancer, and overall cancer mortality.

Conclusion  Postmenopausal women with higher WBC counts have a higher risk of incident invasive breast, colorectal, endometrial, and lung cancer, as well as a higher risk of breast, lung, and overall cancer mortality.

A causal link between inflammation and cancer was first hypothesized in 1863 by Virchow, who observed leukocytes in neoplastic tissues.¹ Well-known associations between cancer and certain infections (eg, chronic hepatitis B and hepatocellular carcinoma) and inflammatory conditions (eg, Crohn disease and colorectal carcinoma) have supported the notion that cancer arises at sites of chronic inflammation.¹ The apparent protection from certain types of cancer by aspirin and nonsteroidal anti-inflammatory drugs provides an additional line of evidence of a causal relation.² Researchers have begun to elucidate the molecular mechanisms by which infectious agents, inflammatory cells, and inflammatory mediators may initiate and promote cancer.^3-5

Several prospective studies^6-8 have found that the white blood cell (WBC) count is a predictor of cancer mortality, although a study⁹ among Korean men and women found no association, and another study¹⁰ found the association to be specific to smoking-related cancers. The WBC count and C-reactive protein level have been shown to be associated with an increased risk of colorectal cancer in 4 studies,^11-14 but the C-reactive protein level was unassociated with colorectal cancer in another study¹⁵ that enrolled only women. A meta-analysis¹⁶ of fibrinogen levels has also shown an association with increased risk of colorectal cancer mortality.

The Women's Health Initiative (WHI) is a multicenter prospective study of 161 808 postmenopausal women composed of diverse racial/ethnic and socioeconomic groups. At baseline, participants in the WHI underwent a WBC count, in addition to an extensive medical history and a physical examination. Because of its large size and broad representation of women from across the United States, this cohort provides an opportunity to determine whether there is an association of WBC count with specific types of cancer that are common in postmenopausal women, as well as to examine the independence of any association from other known risk factors. In this article, we describe the relation between the baseline WBC count and newly diagnosed cancers with an annualized incidence of 0.1% or higher in women enrolled in the WHI. These outcomes included invasive breast, colorectal, endometrial, and lung cancer.

The WHI is an ongoing ethnically and geographically diverse multicenter clinical trial and observational study designed to address the leading causes of morbidity and mortality in postmenopausal women.¹⁷ Briefly, 161 808 women aged 50 to 79 years were recruited at 40 clinical centers throughout the United States. Recruitment began on September 1, 1993, and ended on December 31, 1998. The WHI clinical trials (N = 68 132) include the following 3 overlapping components: the hormone trial, dietary modification trial, and calcium/vitamin D supplementation trial. In the hormone trial, women who had undergone hysterectomy were assigned to the estrogen-alone trial (0.625 mg/d of conjugated equine estrogens vs placebo); the remaining women with a uterus were assigned to the estrogen plus progesterone trial (0.625 mg/d of conjugated equine estrogens plus 2.5 mg/d of medroxyprogesterone acetate vs placebo). In the diet modification trial, a low-fat eating pattern was compared with a usual diet. In the calcium/vitamin D supplementation trial, 1000 mg of elemental calcium plus 400 IU of vitamin D₃ daily was compared with placebo.

Participants in the observational study (n = 93 676) were women who were screened for the clinical trial but proved to be ineligible or unwilling to participate or were recruited through a direct invitation for screening into the observational study. Details of the scientific rationale, eligibility requirements, and baseline characteristics of the participants in the WHI have been published elsewhere.^17-22 All participants provided informed consent using materials approved by institutional review boards at each center.

The following participants were excluded from the original cohort of 161 808 for these analyses: 15 849 with any history of cancer except nonmelanoma skin cancer at baseline, 1681 with a missing baseline WBC count, 1401 with missing data regarding cancer history at baseline, 316 with a WBC count greater than 15.0×10⁹ cells/L, and 191 with a WBC count less than 2.5×10⁹ cells/L. Some women had more than 1 exclusion criterion, yielding a final analytic sample of 143 748. This included 79 227 participants (55.1%) in the observational study and 64 521 participants (44.9%) in the clinical trials. Analyses of endometrial cancer were limited to women who had not undergone hysterectomy at baseline (n = 85 621).

Participants underwent initial screening visits during which personal information, medical history, family history, and health-related habits were assessed using standardized questionnaires. Medication and supplement use were assessed using a medication inventory. Certified staff performed physical measurements and obtained blood samples at the baseline clinic visit. The blood collection took place in the morning after a 12-hour tobacco-free fast. The hemogram sample was collected in a tube containing the anticoagulant edetic acid. These samples were analyzed for WBC count at local laboratories at each of the 40 WHI clinical centers using standardized quality assurance procedures.

The WHI follow-up was conducted by semiannual (clinical trials) or annual (observational study) mailed self-administered questionnaires. As of March 31, 2005, response rates for years 1 through 8 medical history updates in the observational study ranged from 93% to 96%, with only 1.9% of the participants lost to follow-up. An additional 2.2% had discontinued their participation in the study, and 6.1% had died. In the clinical trials, annual follow-up visits were required; 1.2% were lost to follow-up, 3.2% had discontinued participation, and 5.3% had died.

At each semiannual or annual contact, participants were asked if they had been told by a physician for the first time that they had a new cancer or malignant tumor. For all new cancers except nonmelanoma skin cancers, trained study physician adjudicators and cancer coders, blinded to exposure status, reviewed pathology reports, discharge summaries, operative reports, radiology reports, and death certificates for all biopsies, surgical procedures, and deaths to verify any cancer outcome. Pathology reports were available and were reviewed for 98% of invasive breast cancers, 95% of colorectal cancers, 97% of endometrial cancers, and 87% of lung cancers.

To describe participant characteristics across levels of WBC count, WBC counts were categorized using quartile divisions. Cross-tabulations were examined.

Hazard ratios (HRs) and nominal 95% confidence intervals (CIs) from Cox proportional hazards regression models are reported for the outcomes of invasive breast cancer, colorectal cancer, endometrial cancer, lung cancer, and cancer mortality. The initial models were stratified on enrollment in the observational study, on study component (assignment to active hormone or placebo) in the 2 hormone trials (estrogen plus progestin and estrogen alone), or on assignment to intervention or control in the dietary modification trial and were adjusted for age and exclude an additional 651 participants without follow-up. Additional potential confounders and effect modifiers (height, parity, smoking, race/ethnicity, alcohol intake, waist to hip ratio, physical activity, menstrual history, body mass index, hormone therapy use, history of breastfeeding, bilateral oophorectomy, family history of cancer, previous hormone therapy, history of diabetes mellitus, history of benign breast disease, and use of aspirin or nonsteroidal anti-inflammatory drugs) were included in at least 1 of the multivariate models. Participants without complete case data for all covariates in a given multivariate model were excluded from that analysis. Follow-up time for each woman was accrued from enrollment to the date of cancer diagnosis, death, loss to follow-up, or administrative censoring date (March 31, 2005). The mean length of follow-up for the cohort was 7.8 years (range, 0-11.2 years).

The assumption of proportionality was tested by including indicators for the upper 3 WBC quartiles, product terms between these indicators, and follow-up time and by using a likelihood ratio procedure to test for zero coefficients for the 3 product terms. The assumption was met for all of the outcomes included in the analysis. Trends across WBC quartiles were assessed by including a variable that equaled the median of the WBC values within the pertinent quartile. Corresponding P values are reported.

Stratified fully adjusted Cox proportional hazards regression models were examined for interactions of WBC count with potential confounding variables. Interactions of WBC count with age, aspirin use, hormone therapy, current and past smoking, history of cardiovascular disease, and nonsteroidal anti-inflammatory drug use were tested separately for each cancer using likelihood ratio tests and comparing fully adjusted Cox proportional hazards regression models with and without the interaction terms.

In secondary analyses by cancer location, we separately examined the WBC count as a continuous outcome and omitted current smokers and cancers diagnosed within the first 2 years of follow-up. Women who participated in the clinical trials were required to have a negative mammogram (or a negative report of a mammogram verified by WHI staff) within 12 months before entering the study. We examined the association of WBC count with invasive breast cancer in this subgroup who had undergone previous breast cancer screening with well-documented negative results. We also examined the association of the WBC count with colorectal cancer in the subgroup of women who self-reported having been screened using fecal occult blood testing, sigmoidoscopy, or colonoscopy within the previous 5 years. All analyses were performed using commercially available statistical software (SAS for Windows, version 9.1.3; SAS Institute Inc, Cary, North Carolina).

The mean age of the participants at baseline was 63 years, 7% were current smokers, 18% were of a minority race/ethnicity, and 20% regularly used aspirin. Approximately two-thirds of the participants were overweight or obese. Other baseline characteristics of the WHI participants are given by WBC quartile in Table 1. Higher WBC counts were associated with older age, aspirin use, higher parity, hypertension, diabetes mellitus, bilateral oophorectomy, current hormone use, greater body mass index, greater waist and hip circumferences, early age of menarche and menopause, and current smoking and 20 or more pack-years of smoking. Lower WBC counts were associated with physical activity, months of breastfeeding, increased alcohol consumption, and black or Asian/Pacific Islander race/ethnicity. There was little or no association of WBC count with past smoking, benign breast disease, or family history of cancer.

In age-adjusted models (Table 2), there was a graded association of WBC count with incident invasive breast cancer (4639 cases), colorectal cancer (1341 cases), endometrial cancer (766 cases), and lung cancer (1237 cases). In situ breast cancer (1070 cases) was not associated with WBC count (HR for upper vs lower quartile of WBC count, 1.01; 95% CI, 0.85-1.21; P =.98 for trend [data not shown]). In multivariate models further adjusted for potential confounders, the strength of the associations was somewhat attenuated, but significant trends with increasing WBC quartile remained. Compared with women in the lowest quartile of WBC count (2.50-4.79 × 10⁹ cells/L), women in the highest quartile of WBC count (6.80-15.00 × 10⁹ cells/L) had a 15% higher risk of invasive breast cancer, 19% higher risk of colorectal cancer (with similar results for colon and rectal cancers), 42% higher risk of endometrial cancer, and 63% higher risk of lung cancer. In multivariate models with WBC count modeled as a linear variable per 10⁹ cells/L, there was also a statistically significant association between increasing WBC count and higher risk of invasive breast cancer (HR, 1.03; 95% CI, 1.01-1.05), colorectal cancer (HR, 1.08; 95% CI, 1.04-1.12), endometrial cancer (HR, 1.07; 95% CI, 1.02-1.12), and lung cancer (HR, 1.11; 95% CI, 1.08-1.15).

To examine whether subclinical cancer could have caused inflammation and raised the WBC count, we looked at models excluding 1044 invasive breast cancers, 320 colorectal cancers, 185 endometrial cancers, and 260 lung cancers that occurred during the first 2 years of follow-up. The point estimates of the HRs, CIs, and P values were almost identical, and trends across quartiles were statistically significant, despite the reduced power.

After excluding current smokers from the multivariate analyses, there were almost identical associations between WBC count and invasive breast cancer (upper quartile HR, 1.14; 95% CI, 1.03-1.26; P =.01 for trend) and endometrial cancer (upper quartile HR, 1.45; 95% CI, 1.14-1.84; P =.001) but a slightly weaker and statistically nonsignificant association with colorectal cancer (upper quartile HR, 1.16; 95% CI, 0.97-1.39; P =.07). For breast, colorectal, or endometrial cancer, there were no statistically significant interactions between WBC count and smoking status (current smoker vs past smoker vs never smoker). There were only 198 lung cancers that occurred in never smokers, and there was no association of WBC count with lung cancer in this subgroup. There was a statistically significantly increased risk of lung cancer in women with a WBC count in the upper quartile among past smokers (HR, 1.65; 95% CI, 1.30-2.10) and current smokers (HR, 1.96; 95% CI, 1.15-3.32). However, in the lung cancer models, a formal test of interaction between the WBC count and pack-years of past and current smoking was statistically nonsignificant (P =.33). No significant interactions with the WBC count were observed for the following variables: age, aspirin use, hormone therapy, history of cardiovascular disease, and nonsteroidal anti-inflammatory drug use.

Among women with complete covariate information who had a documented negative breast cancer screening test at baseline in the clinical trials, 1553 invasive breast cancers occurred during the follow-up period. Compared with the overall results in women in the highest WBC quartile, women in the clinical trials in the highest WBC quartile had a similar risk for invasive breast cancer, but the results were no longer statistically significant (HR, 1.12; 95% CI, 0.96-1.30; P =.18). Among women who at enrollment reported having undergone colorectal cancer screening within the previous 5 years, 647 colorectal cancers occurred during follow-up. Compared with the overall results, women in this subgroup with WBC counts in the highest WBC quartile had a somewhat lower risk for colorectal cancer, and there was no statistically significant association of WBC count with incident colorectal cancer (HR, 1.13; 95% CI, 0.89-1.42; P =.23).

Cause-specific cancer mortality is given in Table 3. Invasive breast cancer mortality remained more than 2-fold elevated in the highest WBC quartile, with little difference between age-adjusted and multivariate-adjusted models. Colorectal cancer mortality had a borderline association with WBC quartile, which was also not altered substantially by multivariate adjustment. Compared with other cancers, there were few cases of endometrial cancer mortality (n = 48), and the statistically significant trend in mortality with WBC quartile in the age-adjusted model was not seen after multivariate adjustment. Age-adjusted lung cancer mortality was more than 3-fold elevated among women in the highest WBC quartile. After adjustment for smoking, there was marked attenuation of the association, but there was still a statistically significant 65% increase in lung cancer mortality in the highest WBC quartile. Nonlung cancer and total cancer mortality also showed statistically significant associations with WBC quartile.

These findings demonstrate for the first time (to our knowledge) that WBC count is associated with incident invasive breast, colorectal, endometrial, and lung cancer among postmenopausal women, providing further support for the hypothesis of a causal link between inflammation and the initiation, promotion, and progression of these types of tumors. The strongest associations were observed for endometrial and lung cancer incidence. We also found a strong association of WBC count with invasive breast and lung cancer mortality, as well as overall cancer mortality, supporting previous studies^6-8 that found a positive association between WBC count and cancer mortality.

Adjustment of multivariate models for known risk factors, exclusion of incident cases early in the follow-up period, and limitation of the analyses to nonsmokers diminish the chance that our findings are a result of confounding or reverse causality, especially for breast and endometrial cancer. We also tested models limited to women who had previously undergone breast cancer screening (verified mammography reports) and colorectal cancer screening (self-report only). Because these analyses included only one-third to one-half as many cancer cases, power was much diminished, and there were no statistically significant associations of WBC count with these cancers in this analysis. Because we did not demonstrate that an elevated WBC count is present before breast and colorectal cancer is detectable by screening, definitive evidence of a causal relation is not present.

The weaker and statistically nonsignificant relation of WBC count with colorectal cancer in nonsmokers could be because of chance or because the overall association is caused by elevated WBC counts in smokers. Smoking is a risk factor for colorectal adenomas,²³ and our data show that current smoking is also strongly related to WBC count. Similarly, confounding of the relation of WBC count and lung cancer by smoking is a possibility, although we adjusted for current smoking, past smoking, and pack-years of smoking in the multivariate analysis.

In addition to the epidemiological evidence, extensive basic research supports a role of the immune system and inflammation in the pathogenesis of cancer. Most tumors are heavily infiltrated by inflammatory cells, which in turn produce cytokines and chemokines that regulate the growth, migration, and differentiation of tumor and stromal cells.^1,2,24,25 Several genetic polymorphisms in cytokine genes have been associated with an increased risk for colorectal adenomas, a precursor of colorectal cancer.²⁶ Neutrophils, eosinophils, and mononuclear cells produce reactive oxygen and nitrogen species that further damage cellular DNA, RNA, lipids, and proteins.²⁷ Proteases, matrix metalloproteinases, angiogenic factors, and other soluble mediators produced by leukocytes affect cell survival and tissue remodeling.²⁸ Of the cytokines, tumor necrosis factor α may play a central role as a tumor promoter, as well as in the interaction between tumor and inflammatory cells, in part through induction of nuclear factor κB–signaling pathways.²⁹ Cancer cells may use the same adhesion molecules and trophic factors made by inflammatory cells in homing during metastasis.^2,3 Therefore, inflammation seems to play a key role in all stages of cancer from initiation to distant spread.

Several limitations of this analysis must be considered. Differential cell counts were not performed for the WHI, so we do not have information about what types of leukocytes were more prone to be elevated in women who subsequently developed cancer. Only 1 measurement of WBC count was performed, and the analyses were performed in 40 local laboratories on automated counters. Multiple measurements in a central laboratory would have reduced measurement error and increased the precision of our results; therefore, our present results are likely to be underestimates of the true associations due to nondifferential misclassification. The participants in the WHI were generally healthy, well-educated, postmenopausal women; therefore, our results may not apply to the general population or to men, despite the broad geographic representation of the 40 clinical centers.

In summary, we demonstrated that postmenopausal women with higher WBC counts have a higher risk of incident invasive breast, colorectal, endometrial, and lung cancer, as well as a higher risk of breast, lung, and overall cancer mortality. Before these findings can be applied clinically, they should be replicated in other populations, preferably with 2 measurements of WBC count, and expanded to include other biomarkers of inflammation that may be more specifically linked to underlying causal mechanisms of neoplasia.

Correspondence: Karen L. Margolis, MD, MPH, HealthPartners Research Foundation, PO Box 1524, Mail Stop 21111R, Minneapolis, MN 55440-1524 (Karen.L.Margolis@HealthPartners.com).

Accepted for Publication: March 23, 2007.

Author Contributions: Dr Margolis had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Margolis and Thomson. Acquisition of data: Margolis, Thomson, and McTiernan. Analysis and interpretation of data: Margolis, Rodabough, Thomson, and McTiernan. Drafting of the manuscript: Margolis and Thomson. Critical revision of the manuscript for important intellectual content: Rodabough, Lopez, and McTiernan. Statistical analysis: Rodabough. Obtained funding: Margolis. Administrative, technical, and material support: Thomson and McTiernan. Study supervision: Thomson and McTiernan.

Financial Disclosure: None reported.

Funding/Support: The WHI program is funded by the National Heart, Lung, and Blood Institute, US Department of Health and Human Services (Dr Margolis).

Group Information: A full list of the members of the Women's Health Initiative Research Group was published in Arch Intern Med (2007;167[9]:901).

Role of the Sponsor: The funding source participated in the design and conduct of the study and approved the final manuscript but was not involved in the analysis or interpretation of the data.