Self-reported psychological stress and the risk of breast cancer: A case-control study

Abstract

The aim of this study was to examine the relationship between severe life events and breast cancer risk. This study was based on a case–control examination of 858 Polish invasive breast cancer cases and 1085 controls matched for age and place of residence. Data on life events, sociodemographic characteristic, reproductive factors, family history of breast cancer, current weight and height, and lifestyle habits were collected between January 2003 and May 2007 using a self-administered questionnaire. Odds ratios with 95% confidence intervals were estimated as the measure of the relationship between life event stress and breast cancer risk using unconditional logistic regression analyses. After adjustment for potential breast cancer risk factors, women with four to six individual major life events had 5.33 times higher risk for breast cancer, compared with those in the lowest quartile. Similarly, women with a lifetime life change score greater than 210 had about 5 times higher risk compared to women with corresponding scores in the range 0–70. Several life events (death of a close family member, personal injury or illness, imprisonment/trouble with the law, retirement) were significantly associated with breast cancer risk. These findings suggest that major life events can play an important role in the etiology of breast cancer.

Keywords::

• Breast cancer
• case-control study
• life events
• risk factors
• stress
• women

Introduction

Breast cancer is the most commonly diagnosed cancer among women, accounting for around 23% of all cancer worldwide, and is the leading cause of death from cancer in women (Jemal et al. 2008). In Poland, the age-adjusted morbidity for female breast cancer is 47.1 per 100,000 (Wojciechowska et al. 2010). Findings obtained from most epidemiological studies during the past three decades have identified risk factors that increase the likelihood of developing breast cancer. They include increasing age, early menarche, late menopause, late age at first childbirth, no breast-feeding, family history of breast cancer/genetic predisposition, and benign mammary lesion. Also obesity, low physical activity, diet rich in polyunsaturated fat, alcohol consumption, cigarette smoking, and steroid hormones are considered to be important factors in the development of this disease (Hulka and Moorman 2001; American Cancer Society 2008). However, research in this area has estimated that the recognized risk factors can explain only about 40% of the breast cancer cases (Bleiker and van der Ploeg 1999). In past decades, numerous investigators have reported a causal link between life event stress and cancer. The fact that stress can contribute to the onset of cancer was first formulated by Snow (1893). Snow's finding was based on the observation that 156 out of 250 cancer cases experienced a major life event because of the loss of a near relative.

Dalton et al. (2002) reviewed the scientific literature published between 1967 and August 2001 on association between major life events (death, family illness, divorce, job loss, bereavement), depression and a risk for cancer. These studies showed weak associations, equivocal dose–response trends, and inconsistency of the results. Also, Petticrew et al. (1994) in their earlier meta-analysis concluded that recent adverse life events did not affect breast cancer risk. In contrast, the idea that psychological stress may promote breast cancer has been supported by the results of some cohort and case–control studies (Jacobs and Bovasso 2000; Lillberg et al. 2003; Kruk and Aboul-Enein 2004; Peled et al. 2008) that controlled for many of the confounding variables, analyzed cumulation of life events, and the sample size was adequate to evaluate the relationship. In addition, earlier investigators suggested that only a few psychosocial variables have effect on tumor growth and development (Greer and Morris 1975; Cooper et al. 1989; Edwards et al. 1990), and that psychological stressors are associated with the anthropometric, behavioral, and lifestyle factors known to contribute to breast carcinomas (Cohen et al. 2007; McGregor and Antoni 2009).

As presented above, the epidemiological data of the last three decades on the life event stress–breast cancer relationship remain limited in scope, and the results from different studies frequently do not agree, thereby the evidence remains inconclusive. The inconsistent findings in the literature may be mainly due to the weakness in methods of measurement of life events, the often small sample size and selection bias, as well as to errors in measures of life events.

The objective of this retrospective study was to test the hypothesis that women with breast cancer would experience more stressful life events preceding their illness. We also hypothesized a cumulative effect of severe life events on the development of breast cancer. Two time intervals, one beginning at birth and ending one year before the cancer diagnosis, and the other beginning at birth and ending 5 years before the cancer diagnosis, were selected for the inclusion of life events to analyze.

Materials and methods

Study participants

We conducted a hospital-based case–control study in the Region of Western Pomerania (Poland). All interviews were completed between January 2003 and May 2007 in two stages: January 2003–May 2004 and June 2004–May 2007. The study received Ethics Committee Approval from the Pomeranian Medical Academy (No. BN-001/254/02, December 9, 2002) in accordance with assurances approval by the Polish Department of Health and Human Services. Cases of breast cancer were women aged 28–79 years identified directly from the Szczecin Cancer Registry, prospectively selected. They were diagnosed with their first primary histologically confirmed invasive breast cancer (a malignant neoplasm that infiltrates and destroys surrounding tissue) and surgically operated between July 1, 1999, and December 30, 2006. Eligibility was limited to cases aged less than 80 years, residing in the Region of Western Pomerania and able to complete a self-administered questionnaire. The invitation letters and consent forms were sent by post by the Regional Oncology Hospital in Szczecin. During January 2003 to May 2007, a total of 3442 cases were identified. Owing to financial limitations, of 3442 cases a priori fixed percentage (70%; n = 2409) was selected (778 cases at the first stage of the study and 1631 at the second stage) for this study, using a random number generator. Of 2409 cases randomly selected, 239 were deceased, and 193 women were aged ≥ 80 years. Of the remaining 1977 cases, 1187 agreed to participate in the study, 485 did not respond, 128 were too ill, 6 refused, 62 returned a consent form without declaration, and 109 could not be located. Women who agreed to participate in the study were sent an eight-page structured questionnaire in a stamped, preaddressed envelope to complete and return. Of the 1187 cases, 881 (74.2%) completed the questionnaire, 262 (22.1%) did not respond despite two attempts, 13 (1.1%) could not be contacted, and 31 (2.6%) had too much missing data. The overall response rate for the cases was 50.6% (881 questionnaires completed out of 1740 eligible and available cases).

Female controls were frequency matched to the cases by age within 5-year age groups to yield an age distribution similar to the cases, and also by place of residence. They were required to be free of any cancer diagnosis and other chronic diseases, and aged < 80 years. The women were randomly recruited among the outpatients from ambulatories, clinics, the largest hospital in Szczecin, and four relatively smaller hospitals, located in the Region of Western Pomerania. Controls were attending, although constitutionally healthy, because they were acutely ill (e.g. suffering from cold or influenza) or in order to check their health (78.6%), or for treatment of fractures or pain (5.4%), cardiovascular disease (3.1%), back pain (2.8%), and other diseases (skin, eyes, laryngological) (10.1%); percentages of women participating are provided in brackets. We assumed that cases and controls were quite similar with regard to breast cancer screening frequency because they were recruited from the same type of medical centers with the same insurance status. Of the 1615 controls recruited, potentially eligible within the age range and being free of any cancer diagnosis, 1189 agreed to participate in the study (403 at the first stage of the study and 786 at the second stage), and 426 refused participation because of lack of interest or privacy. Among these 1189 women, 1121 completed questionnaires. The overall response rate for controls was 69.4% (1121/1615).

Finally, 858 cases (49.3%) and 1085 controls (67.2%) were included in the analyses, since information collected from 23 cases and 36 controls contained too much missing data.

Data collection

The structured questionnaire included questions about health status, sociodemographic characteristics, reproductive factors, family history of breast cancer in female relatives, current weight and height, lifestyle habits (dietary habits, tobacco smoking, alcohol consumption, use of hormones). For cases, all data obtained were currently as of the date of cancer diagnosis, and controls were assigned a comparable date. The median interval between date of diagnosis and the interview among cases was two years; the range was from 0 to 21 years ( ≤ 4 years, 666 cases; >4 to ≤ 8 years, 156 cases; >8 to ≤ 15 years, 20 cases; and >15 years, 16 cases). Controls were interviewed as soon as possible after the corresponding cases. These data have been published elsewhere (Kruk 2009). Women were also asked to declare detailed data on total physical activity. In addition, the study involved retrospective collection of self-reported life events. The life events section of the questionnaire was based on the social readjustment rating scale items developed by Holmes and Rahe (1967). The rating scale comprises a list of 43 known life events for which a score has been allocated. Events on the Holmes and Rahe scale were weighted to indicate the amount of life change generated by the event. The highest mean weight on the scale is 100 (death of husband) and the lowest weight is 11 (minor violations of the law). Nine specific life events with severe/moderate impact were chosen and included in the questionnaire. Those reported included death of husband (100), divorce (73), separation (65), death of close family member (63), personal injury or illness (53), fired at work (47), retirement (45), change in health of family member (44), imprisonment/trouble with the law (29), and a mild impact, change in residence (20) (values reported in parentheses are mean weight of an individual event). Also, long-lasting depression and a 12th item “other” were included as an open query. Participants were asked to declare which events had been experienced during their life prior to their diagnosis, and how many years and months had passed since exposure to the life events.

Statistical analysis

Relative risk estimation odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using unconditional logistic regression as the measure of the relationship between life event stress and breast cancer risk (Kleinbaum 1994). Relative risk shows the ratio of breast cancer incidence comparing each upper level of life events experience with the lowest level considered as reference.

In order to examine whether the effects of life events might be cumulative, the total number of life events which each woman had recorded starting at birth and ending one year before the breast cancer diagnosis for the cases and one year before selection into the study for controls (the time interval will be called “lifetime”) was calculated. In addition to separately examining the effect of the individual life events on breast cancer risk (Table II), we created two kinds of summary life events variable: various numbers of lifetime major life events (Table IV) and a weighted life change score, totaled for a women's lifetime period and for the period beginning at birth and ending 5 years before the cancer diagnosis for cases, and beginning at birth and ending 5 years before the interview for controls (Table III). In order to measure the effect of accumulated major stressful life events on breast cancer risk, three dummy variables were constructed for the nine events; for one to three, four to six, seven and more (in order as shown in the Results section), and “no major events” considered as reference. To examine the effect of life change scores on the risk, life change scores were classified into four levels: 0–70, 71–140, 141–210, and >210. The lowest score level was considered as reference against which ORs of other levels were counted. Similar classification of life events and life change scores has been successfully used in earlier studies dealing with stressful life events and breast cancer risk by Lillberg et al. (2003) and Ginsberg et al. (1996). A similar classification enables comparison of our findings with those of the previous studies. For each analysis, the ORs and 95% CIs were assessed in an age-adjusted and a multivariable model with adjustment for age and other known and suspected risk factors for breast cancer, selected a priori as potentially confounding variables. Variables considered as confounders were age at enrolment, age at menarche and age at first birth treated as continuous terms, body mass index (BMI) at enrolment [calculated from the baseline survey as the ratio of study participants' self-reported weight (kg) to self-reported height squared (m²)], education level (elementary school, middle school, high school, academy, and above), parity (0, 1, 2, ≥ 3), months of breast-feeding (0, < 6, ≥ 6), menopausal status (premenopausal and postmenopausal), use of oral contraceptives (COCP; yes, no), postmenopausal hormonal replacement therapy use (HRT; yes, no), family history of breast cancer in first-degree relatives (yes, no), ever a regular cigarette smoker (never, smoking < 10, ≥ 10 cigarettes per day), ever exposure to cigarette smoking (none, smoking by parents/husband < 20, ≥ 20 cigarettes per day), adult alcohol consumption (never, ≤ 1, ≥ 2 drinks per day), diet during adult life: animal fat consumption ( ≤ 2, ≥ 3 times per week), red meat consumption ( < 1 serving per week, 1–2 servings per week, 3–4 servings per week, ≥ 5 servings per week), vegetable and fruit consumption ( ≤ 4, 5–6, ≥ 7, servings per day), screening mammography or ultrasonographic examination of breast within the past 2 years (no, yes), and total physical activity, treated as categorical variables. In order to determine the total physical activity detailed information on household, occupational and recreational activity beginning at age 14 years and ending in the reference year (i.e. one year before cancer diagnosis for cases, one year before selection in the study for controls) were collected. The final models were adjusted only for those variables found to influence goodness of fit of the model and remained statistically significant in the multivariate analyses; they are reported in the legends of the tables. The selection of confounding variables was based on a likelihood ratio test. Models were run separately for both above considered periods of a woman's experience of stressful events. The reason for selection of these long time periods is the known estimate that the average time between the initiation of cancer and the clinical detection is long, about 15 years (Fournier 1982). Women were considered postmenopausal if their periods had stopped more than 1 year before their reference date or they had reached age ≥ 55 years and reported a lack of menstruation.

Tests for linear dose–response trends were conducted by entering experience of stress as a continuous variable in the models and they were derived from the Wald χ² test (Harrell 2001). All P values reported are from tests of two-sided hypotheses. A P value less than 0.05 was considered statistically significant. Differences between group means were examined by Student's t-test. All analyses were conducted on a PC using statistical package STATISTICA 98 (stat Soft Polsca, Kraków, Poland).

Results

The anthropometric, reproductive, and lifestyle characteristics of the study participants have been previously described (Kruk 2007) and are presented briefly in Table I. Cases were less educated, more likely to have had a family history of breast cancer, and reported shorter duration of breast-feeding compared with controls. The cases were also more likely to be sometime active and passive smokers, and users of OCP than controls. The average age at enrolment, age at menarche, age at first birth, age at menopause, frequency of alcohol intake, parity, and HRT use reported by cases and controls were similar.

Table I.  Selected characteristics of the study subjects by disease status.

	Cases (n = 858)	Controls (n = 1085)	P value^*
Characteristics	Number, %
Education level
Elementary school	262 (30.5)	253 (23.3)
Middle school	339 (39.5)	379 (34.9)
High school (academy, university)	257 (30.0)	453 (41.8)	< 0.0001
First-degree relative breast cancer	138 (16.1)	82 (7.6)	< 0.0001
Ever users of OCP	153 (17.8)	158 (14.6)	0.01
Ever users of HRT	301 (35.1)	371 (34.2)	0.68
Ever regular cigarette smoker	413 (48.2)	355 (32.8)	< 0.0001
Exposure to cigarette smoke	458 (55.7)	347 (32.6)	< 0.0001
	Mean (SD)
Age at entry (years)	55.3 (9.7)	54.8 (9.5)	0.24
BMI (kg/m^2)	26.2 (4.7)	25.3 (4.1)	0.0001
Age at menarche (years)	13.82 (1.61)	13.80 (1.57)	0.76
Parity	1.90 (1.10)	1.91 (1.03)	0.80
Age at first birth (years)	23.3 (3.94)	24.0 (3.95)	0.09
Age at menopause (years)	48.7 (5.3)	49.2 (4.65)	0.07
Breast-feeding (months)	6.6 (7.52)	10.8 (10.37)	< 0.0001
Alcohol drinking (drinks/week)^†	1.20 (0.87)	1.14 (0.95)	0.27
Lifetime recreational activity (h/wk/yr)	4.17 (6.34)	5.27 (4.59)	< 0.0001
Lifetime household activity (h/wk/yr)	16.6 (11.8)	19.4 (11.29)	< 0.0001
Lifetime occupational activity (h/wk/yr)	24.4 (10.09)	26.2 (12.81)	0.0026

SD, standard deviation

^* Student's t-test differences in means frequency distributions; COCP, oral contraceptive pills; HRT, hormonal replacement therapy

^† One alcoholic drink, tin of beer or a small bottle, 125 ml of wine or 30 g of high-grade alcohols. Values are mean (SD) or the percent of total reporting (%).

Compared with controls, cases had lower average values for each type of physical activity starting at age 14 years ending in the reference year, and higher BMI at enrolment. The mean number ± SD of lifetime severe and moderate life events reported by cases and controls was 1.67 ± 1.52 and 1.11 ± 1.20, respectively, P < 0.0001, and the weighted mean life change scores were 83.5 ± 79.2 and 59.0 ± 60.9, respectively, P < 0.0001.

Table II gives the results of the analyses conducted for the relationship between individual life events accumulated during the lifetime and breast cancer risk. Regardless of disease status, the women most frequently reported experiencing the death of a close family member, personal injury or illness, injury or illness of a close family member, and the death of a husband. As expected, cases experienced psychological stress more frequently than controls with respect to divorce/separation, death of a close family member, personal injury or illness, change in health of a family member, imprisonment/trouble with the law, being tired at work, and retirement (P ≤ 0.027). Table II also shows age- and multivariable-adjusted risk estimation for associations of lifetime individual life event categories with the breast cancer incidence. As can be seen, the multivariable-adjusted ORs differed from these adjusted for age only. The highest statistically significant increased ORs were observed for severe personal illness (OR = 3.95, 95% CI: 2.39–6.55), imprisonment/trouble with the law (OR = 3.74, 95% CI: 1.99–7.03), death of a close family member (OR = 3.14, 95% CI: 2.18–4.54). Also, increased risk by about 2-fold for breast cancer was observed for injury or illness of a close family member, job loss, and retirement. These risks decreased slightly after further adjustment for the remaining life events, and two of them (injury or illness of a close family member and job loss) became statistically insignificant. Although a number of respondents reporting life events marked as “other” was notable (16.5% cases and 8.2% controls) with a case–control difference, the number of responses in each categories was too small to allow statistical comparison. The most reported “other” stressful events were child divorce, work problems, being at variance with relatives, change in number of family get-togethers, and financial troubles. These events were often defined with little accuracy, which made it difficult to describe events mean weights on the Holmes and Rahe scale. It was also found, in a separate analysis, that chronic depression had an enhancing effect on the development of breast cancer. Adjusted for age, the OR associated with depression was 2.63 (95% CI: 1.23–5.56). Multivariable-adjusted ORs were also increased but they did not reach statistical significance (OR = 1.99, 95% CI: 0.84–4.63 and OR = 1.1, 95% CI: 0.53–3.02; adjusted as shown in Table II). We did not find statistical significance between the stress caused by change in a woman's residence [mean weight = 20 on the Holmes and Rahe scale (1967)], multivariable-adjusted OR = 1.30 (95% CI: 0.60–3.85; 56 cases and 52 controls).

Table II.  Odds ratios and 95% confidence intervals for breast cancer by categories of self-reported major life events.

Life event (Holmes and Rahe weight)	Cases(N = 858, %)	Controls(N = 1085, %)	OR (95%CI)
			Age-adjusted	Multivariate^†	Multivariate^‡
Death of husband (100)^*	121(14.1)	134(12.3)	1.31(0.77–2.27)	1.42(0.76–2.70)	1.30(0.63–2.35)
Divorce (73) or separation (65)	102(11.9)	99(9.1)	1.35(1.01–1.82)	1.36(0.97–1.92)	1.68(0.80–1.63)
Death of a close family member (63)	382(44.2)	348(32.1)	2.56(1.87–3.51)	3.14(2.18–4.54)	2.48(1.70–3.64)
Personal injury or illness (53)	211(24.6)	144(13.3)	4.28(2.75–6.69)	3.95(2.39–6.55)	2.60(1.63–4.62)
Change in health of family member (44)	183(21.3)	168(15.5)	1.95(1.26–3.00)	2.46(1.49–4.04)	1.60(0.95–2.70)
Imprisonment, trouble with the law (29)	39(4.5)	19(1.8)	2.66(1.52–4.63)	3.74(1.99–7.03)	2.94(1.56–5.75)
Loss of a job (47)	84(9.8)	72(6.6)	2.52(1.29–4.80)	2.12(1.01–4.49)	1.72(0.72–3.39)
Retirement (45)	95(11.1)	62(5.7)	2.03(1.45–2.85)	2.13(1.44–3.15)	1.52(1.08–2.45)

OR, odds ratio; CI, 95% confidence interval

^* P for the case–control difference >0.05

^† Adjusted for age, lifetime recreational physical activity, place of residence, education, age at menarche, breast-feeding, family history of breast cancer, exposure to cigarette smoke, alcohol consumption, intake of vegetables and fruits

^‡ Adjusted as multivariate and further for the remaining items of life events. Values given in parentheses mean weight of an individual event on the Holmes and Rahe scale.

Table III presents the results for the age-adjusted and multivariable ORs and 95% CIs of breast cancer according to different life change scores. Findings are shown for a woman's lifetime and events that occurred in the time interval beginning at birth and ending 5 years before the cancer diagnosis. Statistically significant increased risks were observed for both considered periods among women with life change score greater than 70 compared with women at the baseline level. The risk increased with increasing score levels on the life change scale, and the results were statistically significant and supported by a trend-test for both logistic regression models. Results from an age-adjusted model suggest that women with lifetime life change score in the range 71–140 (the second score level) appear to be about 1.44 times more likely to develop breast cancer than those with corresponding scores less than 71. Women with life change scores greater than 210 (the highest score level) appear to be 3.88 times more likely to develop breast cancer compared with women at the baseline level. Further adjustment for other potential confounders than age changed the amplitude of ORs by >10% only for the highest score level, regardless of the considered period of life. Results were generally similar for events occurring in both time intervals that were considered. Estimation of a trend-test also provided support for an apparent increase in the risk of breast cancer for increasing life change scores.

Table III.  Breast cancer odds ratios and 95% confidence intervals according to life change scores.

Period of life	Score range	Cases (N = 858)	Controls (N = 1085)	Age adjusted OR (95% CI)	Multivariable-adjusted OR (95%CI)^*
Lifetime
	0-70	410	673	1.00	1.00
	71-140	206	234	1.44(1.15–1.81)	1.56(1.20–2.02)
	141-210	141	136	1.68(1.28–2.19)	1.71(1.25–2.33)
	>210	101	42	3.88(2.65–5.70)	5.09(3.41–8.50)
P trend				P < 0.0001	P < 0.0001
Beginning at birth and ending 5 years before the cancer diagnosis
	0-70	529	798	1.00	1.00
	71-140	177	195	1.37(1.08–1.73)	1.33(1.03–1.72)
	141-210	98	73	2.01(1.45–2.79)	2.03(1.41–2.91)
	>210	54	19	4.26(2.48–7.31)	5.07(2.80–9.18)
P trend				P < 0.0001	P < 0.0001

OR, odds ratio; CI, 95% confidence interval

^* Adjusted for age, lifetime recreational physical activity, place of residence, education, age at menarche, breast-feeding, sleeping time, family history of breast cancer, fruit consumption, alcohol consumption, and exposure to cigarette smoke. P trend: a test for linear trend calculated from logistic regression by modeling the number of life change scores as the continuous variable.

The relationship between summary major life events and breast cancer risk is given in Table IV. In both age-adjusted and multivariate-adjusted analyses, the number of stressful life events was significantly associated with the breast cancer incidence. A clear tendency toward a higher risk of breast cancer with an increasing impact on the number of events could be observed. The age-adjusted OR for breast cancer was 2.33 (95% CI: 1.88–2.90) for women with one to three major life events compared with women reporting zero major stressful events. Women who reported experience of four to six major life events had a significantly increased OR of 4.69 (95% CI: 3.21–6.87) compared with women at the baseline level. The number of women who had declared experience of seven or more crude life events was too small to allow statistical comparison. After further adjustment for known breast cancer risk factors, those are potentially confounding factors, the risk estimates become somewhat higher.

Table IV.  Breast cancer odds ratios and 95% confidence intervals according to number of major stressful lifetime life events.

Total number of major life events	Cases (N = 858)	Controls (N = 1085)	Age adjusted OR (95% CI)	Multivariable-adjusted OR^*
Not reporting major life events	156	395	1.00	1.00
1–3 major life events	585	633	2.33 (1.88–2.90)	2.58 (2.0–3.34)
4–6 major life events	103	54	4.69 (3.21–6.87)	5.33 (4.01–8.21)
7–9 major life events	14	3	–	–

OR, odds ratio; CI, 95% confidence interval; P trend < 0.0001

^* Adjusted for total recreational physical activity, place of residence, education, age at menarche, breast-feeding, sleeping time, family history of breast cancer, vitamin, fruit consumption, and alcohol intake. Nine life events were identified as major on the basis of the life event weights as follows: the death of a husband, divorce, separation, the death of a close relative or friend, personal injury or illness, change in health of family member, imprisonment/trouble with the law, loss of job, and retirement. P trend: a test for linear trend calculated from logistic regression by modeling the number of major life events as the continuous variable.

The age-adjusted OR for the disease was 2.34 (95% CI: 1.88–2.90) per one event increase in the range of one to three life events (death of husband, divorce, separation), and OR = 2.17 (95% CI: 1.79–2.62) in the range of four to six life events (death of husband, divorce, separation, death of a close family member, personal illness or injury, change in health of family member). After adjustment for other potential risks of breast cancer, the OR estimates slightly increased (OR = 2.50, 95% CI: 2.0–3.3, and OR = 2.88, 95% CI: 2.74–3.71) (data not shown).

Discussion

In this retrospective case–control study of 858 cases and 1085 controls based on self-reported life events, the main finding was the presence of a positive relationship between major life events and breast cancer. We found that the total life event scores accumulated over lifetime was, on average, higher for breast cancer cases than for controls. Women with the highest scores were five times more likely to develop the disease than those at the baseline level. Also, there was a higher accumulation of the individual life events of those with the greatest severity, chosen a priori from the Holmes and Rahe (1967) social readjustment rating scale, among cases than controls. Women with four to six major life events (the third tertile) had about 5 times increased risk compared with women not reporting such experience. We found that death of a close family member, personal illness or injury, imprisonment/trouble with the law, and retirement were each significantly related to increased risk.

Yet, the relationship between life events and breast cancer risk is still a highly controversial topic. Some researchers found the linkage between life events and diagnosis of breast cancer, others failed to observe any linkage. Analysis of prospective studies that have investigated this relationship shows mixed results. Jacobs and Bovasso (2000), for example, found an association between maternal death in childhood and breast cancer (OR = 2.56, 95% CI: 1.59–4.35) among a US cohort study of 1213 women (29 cases, 10 died of breast cancer) followed up for 15 years with interviews. The authors did not observe a relation for recent life events. However, they did not adjust risk for the reproductive factors. In turn, Lillberg et al. (2003) in the Finnish Twin Cohort Study (10,808 women, 180 cases) found that multivariable-adjusted risk was significantly increased for divorce or separation (hazard ratio, HR = 2.26, 95% CI: 1.25–4.07), death of a close relative or friend (HR = 1.36, 95% CI: 1.00–1.86), and death of a husband (HR = 2.00, 95% CI: 1.03–3.88). The next Norwegian cohort study by Kvikstad and Vatten (1996) (4491 cases and 44,910 controls) found slightly increased adjusted risk for widowed women, which was statistically insignificant (OR = 1.13, 95% CI: 0.94–1.36) (similar to findings in our study). In contrast, studies relating to register data only, not having data on several life events and without data on known breast cancer risk factors have not shown a significant association between breast cancer risk and divorce (Ewertz 1986; Kvikstad et al. 1994), death of a husband (Jones et al. 1984; Ewertz 1986), death of a child (Jones et al. 1984; Kvikstad and Vatten 1996), and cancer in a child (Johansen and Olsen 1997). Also, the recent study by Surtees et al. (2009) examining moderately or extremely upsetting life events, excluding personal illness, during the 5 years preceding assessment, found no evidence for association of social stress with development of breast cancer. The results presented by us are consistent with several case–control studies (Bremond et al. 1986; Cooper et al. 1989; Forsen 1991; Geyer 1991; Fox et al. 1994; Chen et al. 1995; Ginsberg et al. 1996; Kruk and Aboul-Enein 2004; Peled et al. 2008) that have found an increased breast cancer risk among women with one or more life events in the severest categories or in women with a high number of life change scores. Among the most comparable studies, Ginsberg et al. (1996) observed a significant increase of breast cancer in women with a high number of 10-year life change scores. They noted that women whose life change scores were greater than 210 had 4.67 times increased risk than women with scores ≤ 70, and a test-trend confirmed an increase in breast cancer ORs with increasing score levels. Similarly, the study by Geyer (1991) that had examined the role of life events in breast cancer among 97 women with a suspicious breast lump (33 diagnosed as having breast cancer) found that severe life events were reported most often in the cancer group. The recent study by Peled et al. (2008) (255 cases, 367 controls under the age of 45 years) observed that accumulation of more than one severe life event is positively associated with breast cancer (OR = 1.62, 95% CI: 1.09–2.40). In contrast, findings from some other studies (Roberts et al. 1996; Protheore et al. 1999; Price et al. 2001; Ollonen et al. 2005) reported no association between stressful life events and risk of breast cancer. The findings from our study are also consistent with our previous research (257 breast cases operated on during 1993 to 1998, 565 controls) that showed that accumulation of major life events increased breast cancer risk (OR = 3.93, 95% CI: 1.88–6.18) compared with those who reported no stress. We could not find significant results when we separately examined the effect of a job stress, stress of daily activity, or depression. Little is known about the role that depression plays in the etiology of breast cancer. Of a few studies concerning this relation, three found significantly increased risk (Gallo et al. 2000; Jacobs and Bovasso 2000; Peled et al. 2008), one study insignificantly increased risk (Hahn and Petitti 1988) (similarly to our observation), whereas the study by Hjerl et al. (1999) noted no association.

The results of the final logistic regression models suggest several breast cancer risk factors, among others, lifetime physical activity, that are related to the disease. Also, our findings showed that, when the effect of an individual life event on the risk is estimated, regression models should be controlled for the remaining major life events experienced by a woman. Unfortunately, most of the studies linking stress events to breast carcinogenesis have focused on severe psychological factors but lack of information on key breast cancer confounders that may contribute to cancer or adjustment for other life events. In addition, a majority of studies dealing with this association have tended to concentrate on major life events that took place within 3 or 5 years before the cancer diagnosis, and a 10-year period or longer is a rare case (Ginsberg et al. 1996; Kruk and Aboul-Enein 2004; Peled et al. 2008). Concerning time period in this study we considered lifetime history for all severe life events which the subjects experienced. In this respect, the study is, in part, similar to the study of Peled et al. (2008) and differs from a majority of studies that examined life events exposure within 3–5 years before cancer diagnosis. Stress can cause or promote cancer development indirectly, e.g. by initiation of high fat diets, alcohol intake, or tobacco smoking (Cassileth 1996). A period of several years can be too short to exert effect on tumor development, since breast cancer can take much longer to reach a clinical outcome (Devitt 1986), and carcinogenesis may be at a stage where life events do not have effects. The literature data on the period of tumor development are scarce. For example, the time interval between tumor development and clinical detection reported by Rush (1978) has been hypothesized to be at least 5 years. This suggestion was confirmed by Fournier (1982) who reported that an average value of this interval reaches 15 years.

This study also shows that the impact of life events or life change scores does not appear to be simply additive on breast cancer development; this finding is consistent with conclusions of Burke and Goodkin (1997). We observed significant associations between life change scores and the development of breast cancer and similar values of ORs over both time intervals considered. Also trend tests were apparent; increasing score levels resulted in higher breast cancer ratios. The data suggest that a shorter period (4 years) does not result in an important diminution of the effect of life event scores on the breast cancer risk ratios. This may be because both time intervals were long enough to contribute to tumor development. This finding is in line with the finding of Ginsberg et al. (1996) who detected that statistically significant breast cancer risk increases with increasing score levels over a 10-year period and non-significant increased risk over a 2-year period prior to participants' interviews. Our finding may explain a lack of evidence for an association between life events and breast cancer risk in the studies limited to examination of life events exposure within 3–5 years before cancer diagnosis.

The relationship between major life events and breast cancer risk is biologically plausible and multiple mechanisms may be operative. However, the linkage has not been proved despite a long history of research. It has been hypothesized that psychological changes associated with life events influence development of cancer through immune downregulation (Cohen and Herbert 1996; Kiecolt-Glaser and Glaser 1999). There is evidence that chronic stress increases cortisol level, which disturbs immune function (Lutgendorf et al. 1999; Kemeny and Schedlowski 2007). It is speculated that immune suppression can accelerate development of cancer as immunological defense is important for elimination of mutated cells. Life stress may also promote breast cancer through DNA damage, faulty DNA repair, or inhibition of apoptosis. In addition, stress may alter endocrine characteristics (Forlenza and Baum 2000) that are considered to be important in the development of reproductive cancer (Rowse et al. 1992; Westerlind 2003).

This study has several potential limitations as well as strengths. As for most retrospective case–control studies, our findings were susceptible to recall bias and selection bias. The overall response rate was lower among cases than among controls (see Materials and Methods section), but it is unlikely that the observed strong positive relationships between stress and breast cancer reflect differential participation rates by the study cases and controls with high rates of severe life events. Other authors may have achieved a higher response rate from cases in case–control studies, but our study had the advantage of a larger number of cases examined, e.g. by comparison with the study of Geyer (1991), Ginsberg et al. (1996), or Roberts et al. (1996). A further limitation of this study is that the controls were not randomly selected, but they were outpatients from ambulatories and hospitals, however they were free of chronic diseases. To address the issue of selection bias, we compared the cases and controls with those who participated in a large population-based case–control study in Poland [2502 controls, mean age 55.9 (SD 10.1) years] (Garcia-Glasas et al. 2006) that evaluated the breast cancer risk factors by tumor characteristics. We found that cases and controls in both studies were similar in a majority of characteristics (age at menarche, parity, age at first birth, age at menopause, family history of breast cancer). In addition, the study controls were similar in several medical, lifestyle, and social profiles with the Polish Population Survey (GUS 2006). Also the relations between life events and breast cancer found in women examined previously (Kruk and Aboul-Enein 2004) argue against any strong selection bias. Moreover, the fact that several significant effects were detected including evidence of a dose–response effect and high degree of the examined association argues against strong selection bias in the data. Recall error is of particular concern for lifetime evaluation of data in retrospective studies (Jenkins et al. 1979) because validity of life event interviews depends on a subject's memory. Additionally, it has been reported that cases more frequently than controls believe that experience of severe stress is a risk for breast cancer (Lillberg et al. 2003). In our study, we evaluated the relation to major life events, the most severe, or disruptive ones. Funch and Marshall (1984) and Roberts et al. (1996) showed that the severity of events is linked to reporting reliability and high reproducibility, because major life events show little change in recall over time. This provides reassurance that recall bias was minimized. In addition, our subjects were assessed on sources of stress together with anthropometric, reproductive, and life style factors using a questionnaire that tended to estimate mainly a relationship between physical activity and breast cancer risk. Thus, the subjects' reports were not influenced by knowledge of the stress-breast cancer estimation as a breast cancer risk. In addition, there is still limited awareness of the effect of stress on breast cancer development among Polish women. Also the life events section of the questionnaire was generally consistent in scope with other studies that were based on the Holmes–Rahe social readjustment rating scale (Cooper et al. 1989; Edwards et al. 1990; Forsen 1991; Roberts et al. 1996). Although the questionnaire applied in this study was not directly validated in the examined subjects with respect to stress, its validity has been previously examined (Brugha and Cragg 1990). Another noteworthy limitation is different time periods between cancer diagnosis and retrospective life events assessment. Our median time from breast cancer diagnosis to interview was two years. It is possible that in this study cases recently diagnosed with breast cancer might have recalled past severe event exposures better than those diagnosed several years earlier. However, Maruti et al. (2009) in their “posthoc analyses” did not find any evidence for greater “differential recall bias” when they stratified results for two different time intervals ( ≤ 2 or >2 years) between a diagnosis and interview. A further limitation of the study is that we relied on self-reported life events, which will be imperfect. However, Chasan-Taber et al. (2002) found that reliability of self-reported lifetime exercise was of high reproducibility. Misclassification of the exposure variables, due to inaccurate self-reported life events, may have influenced OR values, although information was collected from cases and controls using the same method. Like most studies, we did not have information on interactions between a severe event or immune parameters and personality (Burke and Goodkin 1997; Segerstrom 2003).

Strengths of this study include a large sample size, histological confirmation of breast cancer, well-characterized participants, and detailed information on potentially confounding factors. Effects of confounds were minimized by utilizing a range of covariates. As well as obtaining information about age when experiencing stress, the study evaluated two measures of the stress experience, i.e. accumulation of the individual life events and total life event scores, finding similar tendencies in breast cancer risk behaviors. Our results agree with the recent meta-analysis of Santos et al. (2009) who found a statistically significant relationship between stressful life events of high intensity and breast cancer.

In conclusion, the results of this study further substantiate a positive relationship between severe life events and the development of breast cancer, confirming and extending our previous research (Kruk and Aboul-Enein 2004). Our findings indicate a cumulative effect of life events on the development of breast cancer, and underline the importance of adjusting for possible risk factors and analyses of life events across long time intervals. These results may be important for health care actions aimed at early detection and prevention of breast cancer.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.