Do breast cancer risk factors differ among those who do and do not undertake mammography screening?

Abstract

Objectives

There is considerable interest in whether mammography screening leads to over-diagnosis of breast cancer. However self-selection into screening programmes may lead to risk differences that affect estimates of over-diagnosis. This study compares the breast cancer risk profiles of participants and non-participants of population-based mammography screening. Risk profiles are also compared between those who have and have not used private screening services.

Setting

This study involved 1162 women aged 40–84 who participated in the 2012 Health Omnibus, an annual face-to-face interview-based survey of a representative sample of the population in the state of South Australia.

Methods

Data were collected on participation in mammography screening, demographic characteristics and breast cancer risk factors (including reproductive, familial and lifestyle factors). Missing data were multiply imputed. Factors independently associated with ever having been screened were identified using multivariable logistic regression, for population-based and ad hoc, private mammography screening separately.

Results

Compared with non-participants, participants of population-based screening were more likely to have used hormone replacement therapy (odds ratio [OR] = 3.72), experienced breast biopsy or surgery (OR = 2.22), and be overweight or obese (OR = 1.57). They were less likely to be sufficiently active (OR = 0.57) or be born in a non-English speaking country (OR = 0.50) or aged under 50 (OR = 0.09). Women who were screened privately were more likely to have a family history of breast cancer (OR = 1.66) and have experienced breast biopsy or surgery (OR = 3.17) than those who had not.

Conclusions

South Australian women who participated in the population-based mammography screening have a slightly higher prevalence of breast cancer risk factors. This also applies to those who undertook private screening.

Keywords

breast neoplasm risk factors mammographic screening

Introduction

While numerous studies have examined socioeconomic and psychosocial predictors of participation in mammography screening^1–5, few have focused on whether breast cancer risk differs between participants and non-participants. Given the current debate about the extent of ‘over-diagnosis’ due to mammography screening (ie. the detection of cancers that would not have been diagnosed in a woman’s lifetime had she not participated in screening⁶), it is important to determine whether women who attend mammography screening differ in their level of breast cancer risk from those who do not. Self-selection into screening programmes may affect estimates of over-diagnosis if breast cancer risk is different among screening participants compared with non-participants.

The extent to which breast cancer risk factor profiles differ between breast screening participants and non-participants is largely unknown and may vary by population. Data from the original mammography screening trials suggest that women who did not take up the invitation to attend screening had a higher risk of death from breast cancer.⁷ However, the underlying risk of breast cancer among non-attenders was not investigated, and little information is available in current non-trial settings. Heightened concern about cancer among women with a family history of breast cancer or a personal history of benign breast disease may lead to greater participation and, consequently, higher breast cancer risk among screening participants.^8,9 Several studies have identified socioeconomic differences in mammography screening uptake.^10–12 Many established reproductive risk factors (eg. nulliparity, later age at first birth, later menopause) are more prevalent among affluent women¹³, who are more likely to undergo screening, again leading to higher breast cancer risk among screening participants. A recent Swedish study, however, found little evidence of associations between screening attendance and reproductive histories.¹⁴ Conversely, other lifestyle risk factors associated with lower socioeconomic position (eg. high alcohol consumption, physical inactivity, obesity) may be more prevalent among non-participants^15,16, leading to lower breast cancer risk among screening participants.

Local context is also important. Many studies on mammography participation have been conducted in the United States, where factors associated with socioeconomic differences may be more important due to the lack of organized population-based screening.¹⁷ Socioeconomic disparities, however, were not evident in most other countries where organized population-wide screening programmes have been established.^18,19

Australia introduced population-based mammography screening, BreastScreen Australia, in 1991, which offers free biennial mammograms targeting women aged 50–69. Women aged 40–49 and over 69 are eligible to attend, but are not been actively recruited. Participation has been reasonably stable at around 55–60% of women in the target age range.²⁰ However BreastScreen records indicate that approximately 20% of women aged 50 or older have never participated in the screening programme. Women may choose to access mammography screening through private imaging services, but little is known about the extent of screening outside the BreastScreen programme.

In this study we examine differences in breast cancer risk factor profiles among South Australian women who have ever participated in BreastScreen with those who have not. We also compare the risk factor profiles of women who have ever and never participated in screening mammography outside of BreastScreen.

Methods

Study population

Data on screening participation, breast cancer risk factors and demographic characteristics were collected from 1162 women aged 40–84, via the 2012 Health Omnibus Survey. The Health Omnibus is an annual face-to-face interview survey which uses a random stratified cluster sampling technique to recruit a representative sample of 3000 South Australian men and women from selected households.²¹ The participation rate in 2012 was 70.4%. Survey methodology and research questions were approved by the University of Adelaide Human Research Ethics Committee.

Measures

Demographic factors included age, highest education level, household income, country of birth and place of residence. Socioeconomic position (SEP) was derived from participant’s residential postcode using the Index of Relative Socioeconomic Advantage and Disadvantage for 2006²², which represents the average socioeconomic status of people living within a specified postcode district. SEP was categorized into quintiles according to the distribution for South Australia.

Data on screening attendance and breast cancer risk factors were only asked of women aged 40 and over. Women were asked whether they had “ever had a screening mammogram at a BreastScreen clinic or in a BreastScreen mobile van” and whether they had “ever had a screening mammogram at a private clinic or other facility that was not run by BreastScreen”. Screening was defined as “a routine check for breast cancer when there were no specific problems or symptoms”.

Personal and behavioural risk factors included parity, age at menarche, menopause and first birth, type and duration of hormone replacement therapy (HRT) use, height, weight, level of physical activity, alcohol consumption, family history of breast cancer and personal experience of breast biopsy or surgery. Body mass index (BMI) was categorized as normal or underweight (< = 24.9 kg/m²), overweight (25–29.9 kg/m²) or obese (30+kg/m²), based on self-reported height and weight. Physical activity was assessed as the time per week spent walking, undertaking moderate activities (eg. lawn bowls, golf, gentle swimming), and undertaking vigorous activities (eg. tennis, jogging, cycling, keep fit classes), converted to metabolic equivalent time (MET-mins/week) according to IPAQ scoring protocol (short form).²³ Insufficient physical activity was classified as <600MET-mins/week. Alcohol consumption was assessed using two questions: “How often do you usually drink alcohol?” and “On a day that you drink how many standard drinks do you have?” with a visual cue-card for standard serving sizes. ‘At-risk consumption’ was classified as >7 standard drinks/week, based on recommendations for cancer prevention.²⁴ Family history of breast cancer was classified as one or more close female relatives ever having been diagnosed with breast cancer (including mother, sisters, daughters and genetically related aunts). Women were also asked whether they had “ever had a breast biopsy or other surgical procedure involving the breast for any reason”. This question was intended as a proxy measure for precancerous conditions (eg. hyperplasia); hence women with breast cancer were excluded from the positive response category.

Analysis

Analysis was restricted to women aged 40–84. Data were initially weighted by individuals’ inverse probability of selection according to area and household size, and reweighted to reflect the age profile for South Australian females, using the 2011 mid-year estimated residential population.²⁵

Due to the high rate of case-wise missingness (cases with missing data in one or more variables) we conducted analysis using both complete case and multiply imputed (MI) data. Imputation was undertaken using multiple imputation chain equations to impute 20 datasets, with 5 cycles of regression switching.²⁶ Our imputation model included all outcomes and covariates in our analytic models, as well as marital status and employment status which both predicted missingness. Logistic regression was used in imputation modelling for binary variables and truncated regression for continuous variables, with upper and lower bounds taking the highest and lowest values of survey responses. Education and income levels were modelled using ordinal logistic regression and age at menopause modelled using multinomial logistic regression (to include ‘pre-menopausal’). Validity of imputed data was checked using diagnostic plots and frequency tabulations. Analyses using MI data were restricted to cases with complete data on screening attendance.²⁷

Weighted cross tabulations were performed for each of the selected characteristics and risk factors, comparing women who had ever participated in BreastScreen with those who had never participated. Multivariable logistic regression was used to identify risk factors that were independently associated with BreastScreen participation, with all variables modelled simultaneously. Reproductive and lifestyle risk factors were modelled as dichotomous variables indicating high versus low risk categories. Similar analyses were undertaken to identify risk factors associated with participation in private mammography screening.

Three summary measures of breast cancer risk were also developed to allow comparison of overall risk among screening and non-screening groups. The first of these measures was the total number of risk factors from twelve possible factors, including insufficient physical activity, overweight or obese, tall stature, at-risk alcohol consumption, HRT use, prior biopsy/surgery, nulliparity, late age at first birth, one or more family members with breast cancer, early menarche, late menopause, and being in the highest SEP quintile. The second measure was a risk score based the Gail model²⁸ with the following modifications: hyperplasia was not included in the model (because data were not collected); women reporting any breast biopsy or surgery were categorized as having had one biopsy, with no option for multiple biopsies in the model. The third of these measures was a risk score derived from the recently published risk prediction model for breast cancer developed by Pfeiffer et al. ²⁹ This prediction model includes parity, age at first birth, menopausal status, age at menopause, length of use of estrogen-progestin HRT, other HRT use, BMI, alcohol consumption, benign breast disease and family history of breast and ovarian cancer. Risk factors were categorized as described by the authors except that any biopsy/breast procedure was substituted for benign breast disease and family history of breast cancer was substituted for family history of breast and ovarian cancer. Risk scores were calculated for each individual based on published relative risks for each risk prediction model. We used linear regression with imputed data, adjusting for age, to compare both the risk scores and mean number of risk factors, according to screening participation.

All analyses were undertaken using Stata v12.0.³⁰

Results

Demographic characteristics and risk factor profiles of survey participants are shown in Table 1 for complete case and multiply imputed data. On a case-wise basis, data were missing for 39% of our survey participants. Case-wise missingness was highest among older women, those with least education and those born in non-English speaking countries. Missing responses were particularly high for questions on household income, height and weight, age at menopause and age at menarche. Profiles did not differ substantially between the complete-case and MI datasets.

Table 1.

Demographic and risk factor profile of participants in the 2012 South Australian Health Omnibus Survey (females aged 40–84).

Factors	Total sample		Complete case		Multiply imputed		% with complete data
Factors	No.	%	No.	%	No.	%	% with complete data
All respondents (weighted no.)	1148 (1150)	–	705 (700)	–	1148 (1148)	–	61.0
Age
40–49	333	29.0	238	34.1	333	29.0	71.5
50–59	323	28.1	202	28.8	323	28.1	62.5
60–69	263	22.9	159	22.7	263	22.9	60.5
70–84	229	19.9	101	14.4	229	19.9	44.1
Education
Less than high school	170	14.8	72	10.4	170	14.8	42.4
Completed high school	418	36.4	260	37.2	418	36.4	62.2
Certificate or diploma	348	30.3	211	30.1	348	30.3	60.6
Bachelor degree	212	18.5	156	22.3	212	18.5	73.6
Missing	2
Income
< = $30,000	241	28.1	179	25.6	334	29.1	53.6
$30,001–60,000	176	20.6	143	20.4	247	21.5	57.9
$60,001–100,000	174	20.2	149	21.2	225	19.6	66.2
>$100,000	267	31.0	230	32.8	341	29.7	67.6
Not stated/missing	270
Birth place
Australia	837	72.8	516	73.7	836	72.8	61.7
Other: English speaking	193	16.8	128	18.3	193	16.8	66.3
Non-English speaking	120	10.4	56	8.0	119	10.4	47.1
Residence
Metropolitan	890	77.4	548	78.2	890	77.4	61.6
Rural	259	22.6	152	21.8	259	22.6	58.7
SEP (area)
Low	263	22.9	162	23.2	263	22.9	61.6
Low-mid	206	17.9	116	16.5	205	17.9	56.6
Mid	180	15.6	112	16.1	180	15.7	62.2
Mid-high	233	20.3	159	22.8	233	20.3	68.2
High	267	23.3	150	21.5	267	23.3	56.2
Height
<1.6 m	324	29.7	213	30.5	345	30.0	61.7
1.6–1.69 m	576	52.8	360	51.5	603	52.5	59.7
1.7+m	191	17.5	126	18.0	200	17.4	63.0
Missing	57
Weight
Under or normal	402	39.8	271	38.7	448	39.0	60.5
Overweight	330	32.6	245	34.9	371	32.3	66.0
Obese	227	27.5	184	26.4	319	28.7	57.7
Missing	139
Activity level
Low (<600 MET-mins/wk)	494	43.0	275	39.3	494	43.0	55.7
Medium (600–1499 MET-min/wk)	335	29.1	212	30.2	334	29.1	63.7
High (1500+ MET-mins/wk)	321	27.9	213	30.4	320	27.9	66.6
Alcohol consumption
Low (none)	247	21.8	131	18.7	252	21.9	52.0
Medium (< = 7drks/wk)	642	56.7	410	58.6	646	56.3	63.5
High (>7drks/wk)	244	21.6	159	22.7	250	21.8	63.6
Missing	15
Parity
No births	90	7.8	55	7.8	90	7.8	61.1
1–2 birth	614	53.6	397	56.8	614	53.5	64.7
3+births	443	38.6	248	35.4	444	38.7	55.9
Missing	4
Age at first birth
<20	159	13.9	84	12.0	160	13.9	52.5
20–24	377	32.9	220	31.4	377	32.8	58.3
25–29	296	25.8	177	25.3	295	25.7	60.0
30+	225	19.6	165	23.5	226	19.7	73.0
No births	90	7.8	55	7.8	91	7.9	60.4
Missing	4
Family history
No relatives	783	69.8	477	68.2	788	68.6	60.5
1 relative	246	21.9	156	22.3	256	22.3	60.9
2+relatives	94	8.3	67	9.5	104	9.1	64.4
Missing	26
Breast biopsy/surgery
No	887	77.2	520	74.3	884	77.0	58.8
Yes	262	22.8	180	25.7	264	23.0	68.2
Menopause age
<45	136	12.8	77	11.0	150	13.1	51.3
45–49	172	16.2	124	17.7	188	16.4	66.0
50–54	287	27.0	176	25.1	315	27.4	55.9
55+	109	10.2	67	9.6	124	10.8	54.0
Premenopausal	359	33.8	256	36.5	372	32.4	68.8
Missing	95
Menarche age
<12	155	14.0	96	14.0	161	14.0	59.6
12–14	753	68.2	474	63.9	782	68.1	60.6
15+	197	17.8	130	15.4	205	17.9	63.4
Missing	50

No. and per cent in each subcategory derived from survey-weighted data.

Counts may not total due to rounding.

Sixty-five per cent of women aged 40–84 had undergone mammography screening at BreastScreen. Based on univariate analysis with MI data (Table 2), participation varied according to age (p < 0.001), education (p = 0.018), household income (p < 0.001), height (p = 0.021), BMI (p = 0.014), age at first birth (p = 0.001) and menopausal age (p < 0.001). Women who had participated in BreastScreen were also much more likely to have used HRT (p < 0.001) and had a breast biopsy or surgery (p < 0.001).

Table 2.

BreastScreen participation according to demographic characteristics and breast cancer risk factors (South Australian women aged 40–84).

Factors	Total^a	No. ever screened at BreastScreen	% ever screened at BreastScreen	p-value^b	OR (adjusted)^c	95% CI	p-value
All respondents
–	1148	748	65.1	–	–	–	–
Age group
40–49	333	78	23.5	0.09	0.06–0.15	<0.001
50–59	323	248	76.8	<0.001	1.00	–	–
60–69	263	225	85.5	1.39	0.78–2.47	0.260
70–84	229	196	85.8	1.69	0.92–3.08	0.088
Education
Less than high school	170	127	74.5	1.00	–	–
Completed high school	418	269	64.4	0.018	1.14	0.66–1.97	0.631
Certificate/diploma	348	226	64.8	1.57	0.86–2.88	0.143
Bachelor degree +	212	126	59.5	1.54	0.75–3.16	0.243
Income
<$30,000	335	262	78.2	1.00	–
$30,001–60,000	247	170	68.9	<0.001	0.98	0.52–1.86	0.955
$60,001–100,000	225	123	54.5	0.64	0.34–1.19	0.159
$100,001+	341	193	56.7	0.96	0.49–1.86	0.896
Residence
Metropolitan	889	582	65.5	0.708	1.00	–	–
Rural	259	166	64.0	0.74	0.46–1.20	0.223
Birth place
Australia	836	536	64.1	1.00	–	–
Other: English Speaking	193	135	69.7	0.515	1.03	0.62–1.70	0.920
Non-English Speaking	119	77	64.8	0.50	0.29–0.85	0.011
SEP
Low	263	161	61.1	1.00	–	–
Low-mid	205	143	69.7	1.99	1.09–3.66	0.026
Mid	180	119	66.3	0.606	1.47	0.86–2.53	0.161
Mid-high	233	147	63.1	1.33	0.75–2.38	0.329
High	267	178	66.5	1.19	0.69–2.07	0.533
Tall stature
Short/average	948	634	66.9	0.021	1.00	–	–
Tall (>1.7 m)	200	114	56.8	0.88	0.54–1.42	0.588
BMI
Underweight/normal	448	268	59.9	0.014	1.00	–	–
Overweight/obese	700	479	68.4	1.57	1.08–2.29	0.018
Physical activity
Adequate	654	436	66.6	0.290	1.00	–	–
Insufficient (<600 MET-min/week)	494	311	63.2	0.57	0.39–0.83	0.003
Alcohol risk
Low	898	590	65.7	0.484	1.00	–	–
At risk (>7drks/week)	250	157	63.0	1.06	0.68–1.64	0.810
HRT use
Never	825	452	54.8	<0.001	1.00	–	–
Ever	323	295	91.4	3.72	2.26–6.13	<0.001
Family history
No	788	506	64.2	0.444	1.00	–	–
Yes (1 + female relative)	360	242	67.1	1.35	0.92–1.98	0.124
Breast biopsy/surgery
No	943	584	61.9	<0.001	1.00	–	–
Yes	205	164	79.9	2.22	1.23–3.99	0.008
Menopausal age
<55 ^d	1019	634	62.2	<0.001	1.00	–	–
55+	129	113	87.2	1.57	0.78–3.15	0.201
Menarche age
12+	987	637	64.5	0.348	1.00	–	–
<12	161	111	69.1	1.00	0.55–1.85	0.991
Nulliparous
No	1058	692	65.4	0.514	1.00	–	–
Yes	90	56	61.9	0.77	0.44–1.37	0.378
Age at first birth
<30 (or no birth)	964	651	67.5	0.001	1.00	–	–
> = 30	184	97	52.7	0.99	0.60–1.63	0.975

counts may not total due to rounding

p-values for differences across subcategories derived from separate unadjusted logistic regression models with individual variables using MI data

adjusted odds ratio derived from weighted multivariate logistic regression analysis with MI data with all variables modelled simultaneously

includes premenopausal if aged <55

In multivariate analyses (Table 2), BreastScreen participants had lower odds of being born in a non-English speaking country (Odds Ratio [OR] = 0.50; 0.29–0.85), having a sufficient level of physical activity (OR = 0.57; 0.39–0.83) and being aged 40–49 (OR = 0.09; 0.06–0.15), and higher odds of being overweight and obese (OR = 1.57; 1.08–2.29). The strongest associations were for ever using HRT (OR = 3.72; 2.26–6.13) and having had a breast biopsy or surgery (OR = 2.22; 1.23–3.99). Results showed a curvilinear association with SEP, indicating lower participation among the lowest and highest SEP groups. No significant differences were found for household income, education level, height, age at first birth, and age at menopause, after simultaneous adjustment. Results from complete case analysis (not shown) differed only slightly from multiply imputed data, in that associations with body weight and country of birth were statistically significant in the latter but not the former analysis. In addition a larger effect size was observed for non-English speaking country of birth in the MI analysis.

Similar analyses were undertaken to examine differences in risk factor profiles in relation to use of private mammography screening services (Table 3). Screening outside of BreastScreen was reported by 24% of women aged 40–84. The proportion screening privately was similar for BreastScreen participants and non-participants (25.2% vs 23.2%, p = 0.505). From multivariable analyses of MI data, those who were screened privately had higher odds of having a bachelor degree (OR = 2.08; 1.10–3.92), being from the most socioeconomically advantaged neighbourhoods (OR = 1.70, 1.01–2.88), having a family history of breast cancer (OR = 1.66; 1.19–2.31) and having had breast biopsy or surgery (OR = 3.17; 2.20–4.55). No differences were observed by age, income level, place of residence, country of birth, or any lifestyle or reproductive risk factors. Findings were similar for complete case analysis.

Table 3.

Participation in private screening (outside of BreastScreen) according to demographic characteristics and breast cancer risk factors (South Australian women aged 40–84).

Factors	Total^a	No. ever screened privately	% ever screened privately	p-value^b	OR (adjusted)^c	95% CI	p-value
All respondents
1130	270	23.9	–	–	–	–
Breast Screen
Non-participants	398	100	25.2	0.505	–	–	–
Participants	732	170	23.2	–	–	–
Age
40–49	330	71	21.4	0.86	0.55–1.36	0.502
50–59	320	80	25.1	0.648	1.00	–	–
60–69	260	71	27.2	1.07	0.66–1.73	0.786
70–84	220	48	21.8	0.90	0.53–1.54	0.710
Education
<Secondary school	167	32	18.9	1.00	–	–
Secondary school	408	82	20.2	<0.001	1.11	0.65–1.87	0.710
Certificate or diploma	345	86	24.9	1.36	0.78–2.40	0.279
Bachelor degree	211	70	33.1	2.08	1.10–3.92	0.024
Income
< = $30,000	323	74	23.0	1.00
$30,001–60,000	245	50	20.6	0.168	0.79	0.47–1.33	0.378
$60,001–100,000	225	51	22.5	0.85	0.44–1.62	0.613
>$100,000	338	95	28.0	0.90	0.50–1.62	0.716
Birth place
Australia	824	191	23.2	1.00	–	–
Other: English speaking	188	45	23.8	0.279	1.04	0.67–1.62	0.846
Non-English speaking	119	34	28.8	1.27	0.76–2.12	0.352
Residence
Metropolitan	878	219	24.9	0.169	1.00	–	–
Rural	252	51	20.4	0.96	0.62–1.48	0.840
SEP (area)
Low	256	43	16.8	1.00	–	–
Low–mid	201	42	20.8	1.28	0.75–2.16	0.362
Mid	178	43	23.9	<0.001	1.53	0.89–2.64	0.124
Mid-high	230	64	28.0	1.73	1.02–2.92	0.042
High	266	78	29.4	1.70	1.01–2.88	0.047
Tall stature
Short/average	993	242	24.4	0.374	1.00	–	–
Tall (>1.7 m)	197	42	21.1	0.78	0.49–1.26	0.318
BMI
Underweight/normal	437	111	25.5	0.375	1.00	–	–
Overweight/obese	693	159	22.9	0.92	0.71–1.39	0.969
Physical activity
Adequate	647	161	24.9	0.410	1.00	–	–
Insufficient (<600 MET-min/week)	483	109	22.5	0.92	0.66–1.28	0.623
Alcohol risk.
Low	884	209	23.6	0.698	1.00	–	–
At risk (>7drks/week)	246	61	24.9	1.03	0.70–1.52	0.865
HRT use
Never	815	181	22.2	0.048	1.00	–	–
Ever	315	89	28.2	1.24	0.87–1.79	0.237
Family history
No	780	164	21.0	0.003	1.00	–	–
Yes (1+female relative)	350	105	30.1	1.66	1.19–2.31	0.003
Breast biopsy/surgery
No	942	180	19.4	<0.001	1.00	–	–
Yes	203	90	44.3	3.17	2.20–4.55	0.001
Menopausal age
<55^d	1003	240	23.9	0.917	1.00	–	–
55+	127	30	23.5	0.82	0.49–1.38	0.460
Menarche age
<15	972	236	24.3	0.462	1.00	–	–
15+	158	33	21.3	0.84	0.51–1.39	0.492
Nulliparous
No	1041	253	24.3	0.215	1.00	–	–
Yes	89	17	18.5	0.64	0.36–1.13	0.124
Age at first birth
<30 (+no birth)	946	219	23.1	0.235	1.00	–	–
> = 30	183	51	27.7	1.12	0.73–1.73	0.611

counts may not total due to rounding

p-values for differences across subcategories derived from separate unadjusted logistic regression models with individual variables using MI data

adjusted odds ratio derived from weighted multivariable logistic regression analysis with MI data

includes premenopausal if aged <55

Differences in summary measures of breast cancer risk according to screening participation are shown in Table 4. Both the mean number of risk factors and the mean risk score, based on the Pfeiffer model, differed significantly between BreastScreen participants and non-participants, as well as between women who had attended screening outside BreastScreen compared with those who had not. Risk scores, based on the Gail model, did not differ significantly between BreastScreen participants and non-participants, but were higher for women who participated in private screening compared with those who had not. These results applied to all women (aged 40–84) as well as those within the target age range (aged 50–69).

Table 4.

Comparison of summary measures of breast cancer risk according to participation at BreastScreen or other private screening service.

Summary risk measures	Breast Screen			Other private screening
	Participant	Non-participant	Participant	Non-participant
	Mean (95%CI)	Mean (95%CI)	p-value^d	Mean (95%CI)	Mean (95%CI)	p-value^d
Total number of risk factors^a
(count of: At-risk alcohol, overweight/obese, tall stature, inactive, HRT use, nulliparious, late first birth, early menarche, late menopause, family history breast cancer, breast biopsy, high socioeconomic position)
40–84	3.10 (2.99–3.21)	2.61 (2.46–2.76)	<0.001	3.20 (3.02–3.38)	2.85 (2.74–2.95)	0.001
50–69	3.21 (3.07–3.35)	2.70 (2.44–2.97)	0.004	3.37 (3.12–3.60)	3.04 (2.89–3.19)	0.022
Gail risk prediction model^b
Age at first birth, age at menarche, breast biopsy, family history breast cancer
40–84	2.22 (2.06–2.37)	2.17 (2.00–2.35)	0.728	2.47 (2.25–2.69)	2.12 (1.98–2.26)	0.009
50–69	2.20 (2.04–2.35)	2.30 (1.96–2.64)	0.603	2.50 (2.21–2.80)	2.12 (1.97–2.28)	0.025
Pfeiffer risk prediction model^c
BMI, Alcohol use, HRT, age at first birth, parity, menopausal age, breast biopsy, family history breast cancer
40–84	2.21 (2.15–2.27)	1.95 (1.88–2.01)	<0.001	2.43 (2.32–2.54)	2.02 (1.97–2.07)	<0.001
50–69	2.25 (2.17–2.32)	1.91 (1.79–2.04)	<0.001	2.50 (2.36–2.64)	2.08 (2.01–2.15)	<0.001

^amean total number of listed risk factors

^bmean risk score based on adapted Gail breast cancer risk prediction model – excluding hyperplasia and combines 1 or more biopsy rather than specifying the number of biopsies

^cmean risk score based on Pfeiffer breast cancer risk prediction model, replacing family history of breast cancer for family history of breast or ovarian cancer

^b,crisk scores above assume all women have the same base line risk of breast cancer irrespective of age

^dp-values derived from regression models of mean score (or total number of risk factors) by screening attendance, adjusted for age, using multiply imputed datasets

no. in models for Breast screen participation: n(40–84) = 1148; n(50–69) = 587

no. in models for participation in private screening; n(40–84) = 1130; n(50–69) = 581

Discussion

This study indicates that, for South Australian women, breast cancer risk factor profiles differ between those who have and have not participated in population-based mammography screening. Participation was less likely among women younger than 50 who are not actively targeted by BreastScreen, those born in non-English speaking countries and those who were not sufficiently active, while the likelihood of having used HRT and of having experienced breast biopsy or surgery was much more likely among BreastScreen participants. Contrary to expectations, BreastScreen participants were more likely to have been overweight or obese, and no statistically significant differences were noted for family history of breast cancer.

The association between HRT use and BreastScreen participation may be due to increased contact with medical practitioners (to obtain or renew prescriptions). Due to concerns about increased risk of breast cancer many doctors may have recommended screening to patients using HRT. There is consistent evidence that a doctor’s recommendation is a strong predicator of screening attendance.¹

As observed in other studies⁹, a history of breast biopsy or surgery was associated with screening participation. However, determining the causal direction is difficult, given the cross-sectional design of this study. Women who attend BreastScreen would have a greater chance of biopsy due to investigation of abnormal mammographic lesions.

While our finding of increased participation among overweight and obese women contradicts other studies showing lower participation^15,31,32, it is consistent with findings from a recent US study.³³ One explanation may be that lighter women with smaller breasts feel more confident in detecting breast abnormalities through self-examination and therefore see less benefit from attending mammography screening.

Also in contrast to other studies⁹, including a previous South Australian study³⁴, we did not observe any association between family history of breast cancer and participation in population-based screening. Additional analyses using different post-hoc definitions of family history (eg. two or more close female relatives diagnosed at any age and/or one relative diagnosed before age 50) also failed to show any statistically significant differences, though odds were consistently elevated.

While there appears to be a relationship between income and having participated in the Breast Screen programme, this was only observed for univariate analysis. The association with income did not persist when controlling for other covariates in multivariate analysis. We believe the association with income is confounded by age. Older women will have had more opportunity to have ever participated in Breast Screening. They are also more likely to have lower household incomes given many will be pensioners or self-funded retirees with modest annual incomes.

Undergoing mammography screening outside of BreastScreen, however, was associated with family history of breast cancer, as well as higher education and higher SEP. There was also a strong association between private screening and breast biopsy/surgery. It is possible that some women confused screening and diagnostic mammography, which may explain the greater odds of biopsy/surgery among those who reported being screened privately. Alternatively those with prior breast problems may seek additional security by undertaking screening more frequently using both public and private services. This may also apply to women with a family history of breast cancer. Additionally women with some premalignant diagnoses, such as atypical ductal hyperplasia, may be discharged from population based screening programmes as a matter of policy, because their risk exceeds population level risks. The association with higher levels of education and SEP suggests those with sufficient financial resources may also seek additional security through private screening. We did not find evidence that BreastScreen non-participants were more (or less) likely undergo mammography elsewhere.

Our findings suggest that participants of population-based mammography screening have a slightly higher risk of developing breast cancer than non-participants. Both HRT use and being overweight are associated with a moderately increased risk of breast cancer.^35,36 Likewise, increased likelihood of breast procedures may indicate a higher prevalence of premalignant abnormalities, and potentially greater risk of breast cancer.³⁷ Furthermore, women from non-English speaking countries (who participate less) tend to have lower risk of breast cancer.³⁸ Only the observed differences in physical activity levels would potentially lead to reduced risk among screening participants.³⁹

Comparison of the summary risk measures supports this conclusion. We found that BreastScreen participants, on average, had a greater number of breast cancer risk factors and a modestly elevated risk of developing breast cancer based on risk scores using the Pfeiffer’s risk prediction model. Risk differences were greater still in relation to participation in private mammography screening. While we did not find statistically significant differences in risk scores based on the Gail model, this is not unexpected, given that there was no difference between BreastScreen participants and non-participants for family history and age at first birth, on which the Gail model is largely based.

This study has a number of limitations. Measures of both screening attendance and risk factors may be inaccurate, due to our reliance on self-reported data. Recall and response bias may also have led to erroneous findings. Because BreastScreen is a well-recognized service, we believe that recall of attendance at BreastScreen would have been reasonably accurate. However participants may have over-reported attendance at private screening services if they did not distinguish between screening and diagnostic mammography. This study did not include all probable risk factors (eg. breast feeding, oral contraceptive use) or explore potential risk factors where evidence is less clear (eg. tobacco smoking, fat intake). Nor did it account for interaction between certain risk factors (eg. effect modification by BMI on HRT effects⁴⁰). Also the cross-sectional design does not allow causal inferences to be determined. This is particularly pertinent to disentangling whether increased breast procedures result from, or are a motivation for, screening participation. However, even when a history of biopsy/surgery is excluded from the calculation of the summary risk scores, we still observed significantly higher mean risk scores among women who have participated in Breast Screen using the Pfeiffer model and mean total count of risk factors.

Finally, our summary measures of risk were relatively crude. Using the total number of risk factors as an indicator of the risk profile does not account for the differing strength of association between various risk factors and the risk of developing breast cancer. Also due to the lack of data on women’s history of pre-malignant disease we used a modified version of the Gail model which did not included hyperplasia or multiple biopsy procedures. Furthermore, the recently published Pfeiffer risk prediction model has not been validated in an Australian population and we have applied it as if the background level of risk has not changed over time. Therefore results should be considered indicative rather than definitive.

The strengths of this study are that we collected a range of risk factor data from a randomly selected sample of women who are representative of the eligible screening population for screening, and used multiple imputation to increase precision and reduce potential biases due to the exclusion of a large proportion of cases with missing data.

Conclusion

Our findings suggest that South Australian women who participate in population-based screening have a slightly worse risk factor profile than those who do not. For the most part this can be attributed to greater HRT use among BreastScreen participants. These findings may not be generalizable to other settings, given variations in risk factor prevalence and/or organization of screening services in different countries. This highlights the need to understand differences in breast cancer risk profiles within the local context, especially in relation to evaluating benefits and risks of mammography screening including over-diagnosis.

Footnotes

Funding

This study was funded by the National Breast Cancer Foundation through a Novel Concept Award (#NC-11-53).

Conflict of interest

Authors have no conflict of interest.

Acknowledgements

We wish to thank Ms Lora Dal Grande and Mr Graeme Tucker for their assistance in relation to survey weighting. We also wish to acknowledge the contribution of women who took part in the 2012 HOS survey. Ms Beckmann is the recipient of a NHMRC PhD scholarship. Prof John Lynch is supported by a NHMRC Australian Fellowship.

References

Schueler

Chu

Smith-Bindman

. Factors associated with mammography utilization: a systematic quantitative review of the literature. J Womens Health 2008; 17(9):1477–98.

von Euler-Chelpin

Olsen

Njor

. Socio-demographic determinants of participation in mammography screening. Int J Cancer 2008; 122(2):418–23.

Smith

Hochhalter

Ahn

. Utilization of screening mammography among middle-aged and older women. J Womens Health 2011; 20(11):1619–26.

Vernon

Laville

Jackson

. Participation in breast screening programs: a review. Soc Sci Med 1990; 30(10):1107–18.

Meissner

Breen

Taubman

. Which women aren't getting mammograms and why? (United States). Cancer Causes Control 2007; 18(1):61–70.

Paci

Duffy

. Overdiagnosis and overtreatment of breast cancer: overdiagnosis and overtreatment in service screening. Breast Cancer Res 2005; 7(6):266–70.

Duffy

Cuzick

Tabar

. Correcting for non-compliance in case-control studies to evaluate cancer screening programmes. Journal of the Royal Statistical Society Series C (Applied Stastistics) 2002; 51(2):235–43.

Hay

McCaul

Magnan

. Does worry about breast cancer predict screening behaviors? A meta-analysis of the prospective evidence. Prev Med 2006; 42(6):401–8.

McCaul

Branstetter

Schroeder

. What is the relationship between breast cancer risk and mammography screening? A meta-analytic review. Health psychology: official journal of the Division of Health Psychology, American Psychological Association 1996; 15(6):423–9.

10.

Kim

Jang

. Socioeconomic disparities in breast cancer screening among US women: trends from 2000 to 2005. Journal of preventive medicine and public health 2008; 41(3):186–94.

11.

Kjellen

von Euler-Chelpin

. Socioeconomic status as determinant for participation in mammography screening: assessing the difference between using women's own versus their partner's. International journal of public health 2010; 55(3):209–15.

12.

Zackrisson

Andersson

Manjer

. Non-attendance in breast cancer screening is associated with unfavourable socio-economic circumstances and advanced carcinoma. Int J Cancer 2004; 108(5):754–60.

13.

dos Santos Silva I, Beral V. Socioeconomic differences in reproductive behaviour. In: Kogevinas M, Pearce N, Susser M, Boffett P, editors. Social Inequalities and Cancer. Lyon International Agency for Research on Cancer; 1997.

14.

Lagerlund

Sontrop

Zackrisson

. Do reproductive and hormonal risk factors for breast cancer associate with attendance at mammography screening? Cancer Causes Control 2013 Epub 2013/06/06.

15.

Lopez-de-Andres

Martin-Lopez

Hernandez-Barrera

. Predictors of breast and cervical cancer screening in a Spanish metropolitan area. J Womens Health 2010; 19(9):1675–81.

16.

Coughlin

Uhler

Hall

. Nonadherence to breast and cervical cancer screening: what are the linkages to chronic disease risk? Preventing chronic disease 2004; 1(1):A04–A04.

17.

Miles

Cockburn

Smith

. A perspective from countries using organized screening programs. Cancer 2004; 101(5 Suppl):1201–13.

18.

Walsh

Silles

O'Neill

. The importance of socio-economic variables in cancer screening participation: a comparison between population-based and opportunistic screening in the EU-15. Health policy 2011; 101(3):269–76.

19.

Palencia

Espelt

Rodriguez-Sanz

. Socio-economic inequalities in breast and cervical cancer screening practices in Europe: influence of the type of screening program. Int J Epidemiol 2010; 39(3):757–65.

20.

BreastScreen SA. The impact of BreastScreen SA, 20 year report 1989–2008: with trend data for 1989-2004. Adelaide, Government of South Australia 2010.

21.

Wilson

Wakefield

Taylor

. The South Australian Health Omibus Survey. Health Promotion J Aust 1992; 2: 47–9.

22.

Australian Bureau of Statistics. Information Paper: An Introduction to Socio-Economic Indexes for Areas (SEIFA), 2006 Canberra ABS; 2008.

23.

The IPAQ Group. Guidelines for data processing and analysis of the International Physical Activity Questionnaire (IPAQ) - Short and long forms. 2005 November 2005 [cited; Available from: www.ipaq.ki.se/scoring.pdf.

24.

World Cancer Research Fund, American Institute for Cancer Research. Second Expert Report, Food, Nutrition, Physical Activity, and the Prevention of Cancer: A Global Perspective, Washington, DC: AICR, 2007.

25.

Australian Bureau of Statistics. 3101.0 - Australian Demographic Statisitics, June 2012, Canberra: ABS, 2012.

26.

White

Royston

Wood

. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med 2011; 30(4):377–99.

27.

von Hippel

. Regression with missing Ys: An improved strategy for analyzing multiply imputed data. Sociological Methodology 2007; 37(1):83–117.

28.

Gail

Costantino

Bryant

. Weighing the risks and benefits of tamoxifen treatment for preventing breast cancer. J Natl Cancer Inst 1999; 91(21):1829–46.

29.

Pfeiffer

Park

Kreimer

. Risk prediction for breast, endometrial, and ovarian cancer in white women aged 50 y or older: derivation and validation from population-based cohort studies. PLoS Medicine 2013; 10(7):e1001492–e1001492.

30.

StataCorp. Stata Statisitical Software: Release 12. College Station, TX: StatCorp LP, 2010.

31.

Wee

McCarthy

Davis

. Obesity and breast cancer screening. Journal of general internal medicine 2004; 19(4):324–31.

32.

Fontaine

Heo

Allison

. Body weight and cancer screening among women. Journal of women's health & gender-based medicine 2001; 10(5):463–70.

33.

Berz

Sikov

Colvin

. ‘Weighing in' on screening mammography. Breast Cancer Res Treat 2009; 114(3):569–74.

34.

Roder

Houssami

Farshid

. Population screening and intensity of screening are associated with reduced breast cancer mortality: evidence of efficacy of mammography screening in Australia. Breast Cancer Res Treat 2008; 108(3):409–16.

35.

Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormone replacement therapy: collaborative reanalysis of data from 51 epidemiological studies of 52,705 women with breast cancer and 108,411 women without breast cancer. Collaborative Group on Hormonal Factors in Breast Cancer. Lancet 1997; 350(9084):1047–59.

36.

van den Brandt

Spiegelman

Yaun

. Pooled analysis of prospective cohort studies on height, weight, and breast cancer risk. Am J Epidemiol 2000; 152(6):514–27.

37.

Hartmann

Sellers

Frost

. Benign breast disease and the risk of breast cancer. N Engl J Med 2005; 353(3):229–37.

38.

Arnold

Razum

Coebergh

. Cancer risk diversity in non-western migrants to Europe: An overview of the literature. Eur J Cancer 2010; 46(14):2647–59.

39.

Zhang

Kang

. Physical activity and risk of breast cancer: a meta-analysis of prospective studies. Breast Cancer Res Treat 2013; 137(3):869–82.

40.

Narod

. Hormone replacement therapy and the risk of breast cancer. Nat Rev Clin Oncol 2011; 8(11):669–76.