Long-term psychosocial consequences of false-positive screening mammography: a cohort study with follow-up of 12–14 years in Denmark

Abstract

Objective

To compare the long-term psychosocial consequences of mammography screening among women with breast cancer, normal results and false-positive results.

Design

A matched cohort study with follow-up of 12–14 years.

Setting

Denmark from 2004 to 2019.

Participants

1170 women who participated in the Danish mammography screening programme in 2004–2005.

Intervention

Mammography screening for women aged 50–69 years.

Outcome measures

We assessed the psychosocial consequences with the Consequences Of Screening–Breast Cancer, a condition-specific questionnaire that is psychometrically validated and encompasses 14 psychosocial dimensions.

Results

Across all 14 psychosocial outcomes, women with false-positive results averagely reported higher psychosocial consequences compared with women with normal findings. Mean differences were statistically insignificant except for the existential values scale: 0.61 (95% CI (0.15 to 1.06), p=0.009). Additionally, women with false-positive results and women diagnosed with breast cancer were affected in a dose–response manner, where women diagnosed with breast cancer were more affected than women with false-positive results.

Conclusion

Our study suggests that a false-positive mammogram is associated with increased psychosocial consequences 12–14 years after the screening. This study adds to the harms of mammography screening. The findings should be used to inform decision-making among the invited women and political and governmental decisions about mammography screening programmes.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study has a follow-up of 12–14 years.

• Psychosocial consequences were assessed at six times points and measured with a condition-specific questionnaire with adequate psychometric measurement properties.

• Screening results were only assessed at baseline; classification bias might be present due to changing exposure.

Background

Although it is more than 30 years since research established that false-positive screening mammograms cause psychosocial consequences, no studies have quantified these beyond 3 years.^{1 2}

Breast cancer screening aims to detect cancer at a localised stage to facilitate early treatment and thus potentially lower morbidity and breast cancer-specific mortality. In this pursuit, a large number of women are repeatedly screened with mammography.^{3 4} The specificity of screening mammography is not 100% and as a consequence, a considerable amount of women will receive a false-positive screening result. A false-positive mammogram refers to an initial abnormal screening that confirms non-malignant through further investigations, for example, imaging technology, needle or surgical biopsies.^{5 6} The cumulative probability of at least one false-positive mammogram ranges from 20% to 60%.^7–15 This means that in some countries, more than half of the women who participate in mammography screening will receive at least one false-positive result, while in others, it is one in five. Christiansen and colleagues estimated that the cumulative risk of a false-positive mammography in some contexts can reach almost 100% depending on, for example, age, breast density and screening frequency.¹⁶

The finding of an abnormality, even when cancer is ruled out in the further process, may give rise to considerable stress and anxiety. This is notable not only in the critical waiting period, but also in the long term. Studies have shown that women who receive false-positive results experience negative psychosocial consequences up to 3 years after the screening mammography.^{6 17–20} We have previously shown that 6 months after final diagnosis, women with false-positive results reported changes in existential values and inner calmness as great as those changes reported by women who were diagnosed with breast cancer.⁶ Additionally, we established that 3 years after final diagnosis, women with false-positive results reported greater psychosocial consequences across all outcomes compared with women with normal screening results. To the extent of our knowledge, no study has previously assessed the psychosocial consequences beyond 3 years.

This study aimed to examine the psychosocial consequences of false-positive screening mammography 12–14 years after the screening using the condition-specific questionnaire Consequences Of Screening–Breast Cancer (COS-BC).

Methods

Study population and survey administration

We conducted a cohort study among women who received abnormal mammograms in two Danish screening centres between 2004 and 2005. The women were approached by personnel at the recall clinic, informed about the study and invited to participate. The women completed the COS-BC before they had any additional examinations. Each of these women was matched with two women who had normal results the same day at the same clinic. The matched women were mailed the COS-BC. In the areas, women aged 50–69 years had been offered screening from 1990 to 1993, and women were included regardless of their stage of screening process.

This article reports on a follow-up survey 12–14 years among those of the 1309 women, who were still alive (n=1170). The questionnaires were sent out on paper with a prepaid return envelope and a cover letter with a description of the study. We posted a reminder 4 weeks after the initial postage. Survey administration is further elaborated on in the 3-year follow-up article.⁶

This cohort study was conducted in Denmark, where women aged 50–69 years are invited to biennial mammography screening. The mammography screening programme was implemented in parts of Denmark in 1991 to 1993, and was nationally implemented in 2007. Denmark is a welfare state with a publicly funded healthcare system, and therefore, mammography screening and follow-up diagnostics are not associated with any direct costs for the participating women.

Questionnaire

The COS-BC is a condition-specific questionnaire developed and validated to measure the psychosocial consequences of screening mammography.^{21 22} The content validity was explored in focus groups and individual interviews, where relevance, coverage and understandability were ensured. In the qualitative development of the COS-BC, psychosocial consequences in the context of mammography screening were found to include 12 domains. These domains and items were tested with Rasch analyses and statistically confirmed as valid scales.^{21 22}

The COS-BC was the first questionnaire in the Consequences Of Screening (COS) series. The core questionnaire COS was later used as a basis to develop other questionnaires in the context of screening for life-threatening diseases.^23–25 On that basis, the COS-BC was extended from 12 to 14 dimensions, now including empathy and impulsivity.²⁶ These two scales are therefore only measured in this follow-up.

COS-BC consists of two parts. Part I encompasses eight domains of negative consequences: Sense of dejection, Anxiety, Breast examination, and Negative impact on sexuality, Behavior and Sleep, and two single items: Keeping my mind off things and Feeling less attractive. The items have response categories ranging from ‘not at all’, ‘a bit’, ‘quite a bit’ to ‘a lot’ scoring 0, 1, 2 and 3, respectively. Part II encompasses six domains of changes in psychosocial consequences after the final diagnosis: Worries about cancer, Social network, Existential values, Inner calmness, Impulsivity, and Empathy. These changes can be both positive and negative. The items are worded as a personal inquiry, for example, ‘After the examinations, my thoughts about the future are…’, with response categories ranging from more pessimistic to more optimistic (scale: Existential values). The items have five response options: ‘much less’, ‘less’, ‘same as before’, ‘more’ and ‘much more’, scoring 2, 1, 0, 1 and 2, respectively. This scoring reflects ‘same as before’ as the scale’s midpoint, which indicates no change. While part I acknowledges the negative short-term consequences, part II appreciates that these in time can become positive or remain negative.¹ Positive changes can also be considered consequences as the screening initially causes a crisis that later induces a positive response.^{1 25}

Sociodemographic information was self-reported and collected at baseline.

Statistical analysis

To examine comparability between subgroups and check for potential confounding, we compared the distribution of baseline covariates between the screening groups with the χ2 test.

We compared the means of the COS-BC scores at the various follow-up time points between women who had normal results, false-positive results and breast cancer, respectively. To estimate the mean differences between the scores in these groups, we used weighted multivariable linear models with the method of generalised estimating equations (GEE). GEE is a method that accounts for correlated and overdispersed data due to repeated measurements and weighting. We used inverse probability weighting (IPW)²⁷ to adjust for potential bias due to differential attrition. With IPW, the observations that are not missing are weighted by the inverse probability of being not missing. We estimated these probabilities in logistic regression models including the baseline covariates, screening group, previous scores on the corresponding scale and response at previous follow-ups. Analyses were adjusted for baseline covariates: living alone, employment status, social class and age (quadratic). We lowered the statistical significance level to p<0.01 to avoid type 1 errors due to multiplicity. We used SAS V.9.4 for any analysis. The level of significance, methods and analyses were prespecified before data collection.

Patient and public involvement

This study used COS-BC as an outcome; the questionnaire was developed in close collaboration with 34 women who had participated in mammography screening, to ensure that the measure was relevant, understandable and comprehensive.²¹ After this survey, we conducted a qualitative interview study with 14 women from the population, where the women explicitly appreciated the research that we did.

Findings

At the follow-up of 12–14 years, the lowest response rate was observed among the women who had breast cancer diagnosed at baseline, hereafter women who had false positives, and the highest response rate among women with normal results (figure 1). Of the 1170 women who were still alive, a total of 761 (65.0%) returned the questionnaire and were included in this study.

Figure 1.

Flow chart of invited women and response rate at each of the six assessments.

Table 1 shows the distribution of baseline covariates between screening groups. The distribution did not differ significantly between groups concerning cohabitation, employment and socioeconomic status. The groups differed regarding age: highest among women with breast cancer and lowest among women with false positives.

Table 1.

Study population baseline characteristics

		Screening result				χ2 p value
		Normal	False positive	Breast cancer	Total
		(n=520)	(n=156)	(n=85)	(n=761)
		n (%)	n (%)	n (%)	n (%)
Age	50–54	146 (28.1)	76 (48.7)	18 (21.2)	240 (31.5)	<0.001*
	55–59	216 (41.5)	40 (25.7)	26 (30.6)	282 (37.1)
	60–64	113 (21.7)	22 (14.1)	27 (31.8)	162 (21.3)
	≥65	45 (8.7)	18 (11.5)	14 (16.4)	77 (10.1)
	Missing				0
Living alone (cohabitation)	No	388 (75.3)	117 (77.5)	66 (78.6)	571 (76.1)	0.73
	Yes	127 (24.7)	34 (22.5)	18 (21.4)	179 (23.9)
	Missing				11
Employment	Working†	337 (65.7)	101 (66.9)	50 (59.5)	488 (65.2)	0.80
	Unemployed	19 (3.7)	4 (2.6)	2 (2.4)	25 (3.4)
	Pensioned	157 (30.6)	46 (30.5)	32 (38.1)	235 (31.4)
	Missing				13
Socioeconomic status	I	24 (4.7)	4 (2.6)	2 (2.4)	30 (4.0)	0.39
	II	90 (17.5)	31 (20.7)	13 (15.5)	134 (17.9)
	III	113 (21.8)	31 (20.7)	14 (16.7)	158 (21.1)
	IV	178 (34.6)	49 (32.7)	27 (32.1)	254 (33.9)
	V	110 (21.4)	35 (23.3)	35 (33.3)	173 (23.1)
	Missing				12

*significance test

†Working also involves three students: two with normal results and one with a false-positive result

There was a similar pattern in mean scores of the COS-BC across all psychosocial outcomes (figures 2 and 3). In this pattern, scores of women with false positives were higher than scores of women with normal results and lower than scores of women diagnosed with breast cancer.

Figure 2.

The mean scores of the Consequences Of Screening–Breast Cancer part I (y-axis) for the three screening groups at six time points: 0, 1, 6, 18, 36 and 144–169 (12–14 years) months (x-axis).

Figure 3.

The mean scores of the Consequences Of Screening–Breast Cancer part II (y-axis) for the three screening groups at six time points: 0, 1, 6, 18, 36 and 144–169 (12–14 years) months (x-axis).

Analyses revealed a consistent trend of positive mean differences between women with false positives compared with women with normal results (table 2). These mean differences were statistically insignificant except for the existential values scale: 0.61 (95% CI (0.15 to 1.06), p=0.009) (table 2). This could be interpreted as 61% of the women who had false-positive results answered ‘less’ or ‘more’ instead of ‘same as before’ on one item on the existential values scale; all other values were fixed.

Table 2.

Adjusted and weighted analyses of COS-BC: estimated mean differences 12–14 years after the index screening

Scale	Mean difference
	False positive compared with normal		False positive compared with breast cancer		Breast cancer compared with normal
	Mean $∆$ (95% CI)	P value	Mean $∆$ (95% CI)	P value	Mean $∆$ (95% CI)	P value
1. Sense of dejection (0–18)	0.31 (−0.23 to 0.84)	0.26	−0.71 (−1.66 to 0.25)	0.14	1.01 (0.15 to 1.88)	0.02
2. Anxiety (0–18)	0.44 (−0.11 to 0.98)	0.11	−0.13 (−1.00 to 0.75)	0.78	0.56 (−0.18 to 1.31)	0.13
3. Negative impact on behaviour (0–21)	0.55 (−0.08 to 1.17)	0.08	−0.00 (−1.00 to 1.00)	0.99	0.55 (−0.28 to 1.38)	0.19
4. Negative impact on sleep (0–12)	0.17 (−0.25 to 0.59)	0.43	0.01 (−0.63 to 0.65)	0.98	0.16 (−0.39 to 0.71)	0.56
5. Breast examination (0–6)	0.19 (−0.07 to 0.45)	0.14	−0.03 (−0.37 to 0.31)	0.87	0.22 (−0.05 to 0.49)	0.11
6. Negative impact on sexuality (0–6)	0.41 (0.01 to 0.82)	0.04	−0.55 (−1.19 to 0.08)	0.09	0.96 (0.47 to 1.47)	<0.001*
7. Felt less attractive (0–3)	0.12 (−0.02 to 0.26)	0.098	−0.11 (−0.33 to 0.12)	0.35	0.23 (0.04 to 0.42)	0.01*
8. Keeping my mind off things (0–3)	0.01 (−0.10 to 0.13)	0.85	−0.08 (−0.28 to 0.12)	0.43	0.09 (−0.08 to 0.27)	0.30
9. Worries about breast cancer (0–4)	0.07 (−0.16 to 0.31)	0.55	0.07 (−0.16 to 0.29)	0.55	–	–
10. Inner calm (0–6)	0.17 (−0.08 to 0.41)	0.18	−0.18 (−0.54 to 0.17)	0.31	0.35 (0.02 to 0.68)	0.04
11. Social network (0–6)	0.02 (−0.10 to 0.13)	0.80	−0.21 (−0.44 to 0.03)	0.09	0.22 (−0.02 to 0.46)	0.07
12. Existential values (0–12)	0.61 (0.15 to 1.06)	0.009*	−0.50 (−1.15 to 0.16)	0.13	1.10 (0.51 to 1.70)	<0.001*
13. Impulsivity (0–12)	0.33 (0.01 to 0.65)	0.04	0.09 (−0.41 to 0.59)	0.72	0.24 (−0.18 to 0.27)	0.26
14. Empathy (0–6)	0.26 (0.02 to 0.52)	0.04	−0.13 (−0.54 to 0.28)	0.52	0.40 (0.04 to 0.75)	0.02

*significant test

COS-BC, Consequences Of Screening–Breast Cancer.

There was no consistent trend in mean differences between women with false positives and women with breast cancer. No mean differences were statistically significant and p values were high (table 2).

There was also a consistent trend of positive mean differences between women with breast cancer compared with women with normal results. All differences were statistically insignificant except for the sexuality and existential values scales (p<0.001) (table 2).

Discussion

Our results indicated that women with false-positive screening mammograms were still affected by their screening results 12–14 years after. There was a consistent dose–response association across most COS-BC scales where women diagnosed with breast cancer were responsive to the same scales as women with false positives. This means that there is an association between the severity of the screening result and the psychosocial consequence. This biological gradient suggests a causal relation.

In this dose–response association, women with false positives and breast cancer both seemed to be affected more than women with normal results, but women with false positives to a lesser degree than women diagnosed with breast cancer. Mean differences were small and mostly statistically insignificant. However, the consistent trends in estimates and the dose–response association across COS-BC scales made the results plausible as they expressed a psychological gradient and thus were less likely to be due to chance.

Compared with our previous analyses, the mean differences have progressed toward null, which indicated that the psychosocial consequences have subsided over time.⁶ This was in accordance with trends from previous research; however, no studies had quantified psychosocial consequences beyond 3 years.¹⁷

Strengths and limitations

A strength of this study was the use of the condition-specific questionnaire COS-BC, which has proven content validity and adequate psychometric properties.²¹ Other studies have used patient-reported outcome measures (PROMs) that have not been developed or tested in a screening context and thus have no proven content validity.^{17 20 28} When PROMs lack content validity or other important psychometric properties, the PROM will have low power to detect the psychological trait or will measure it inaccurately.^{29 30} Confirmatory psychometric analyses cannot make up for inadequate content validity.^31–34

Another strength was that we included women with breast cancer as a relevant benchmark, which revealed the dose–response association.

Despite the pronounced benefits of the long follow-up period, it might also expose the study to classification bias. Women would typically have been to several screening tests between baseline and present follow-up, which implies that they could have had different screening results. Some women might also have been screened before the study baseline. Notably, previous or subsequent false-positive and true-positive results would increase the psychosocial consequences measured at this follow-up. Women with false positives are about three times more likely to develop breast cancer compared with normal results, however, still rare in absolute numbers.¹⁷ This would overestimate the mean difference between women with false positives and women with normal results. However, 20%–60% of women will, throughout screening participation, receive false positives, which would underestimate the mean difference between women with false positives and women with normal results.7 8 10 14 15 The probability of women shifting exposure status from normal to false positive is, nevertheless, much higher than from false positive to breast cancer. This might indicate that the mean difference is more likely to be underestimated rather than overestimated. Yet, we did not conduct any analysis to test these hypotheses. Instead, we evaluated the psychosocial consequences of a specific false-positive screening result. Hence, if we are to adjust in some way for intermediate screening results, we would lose the interpretability of the effects as long term.

A pitfall in our study was that non-response was different between groups. Those who dropped out were more likely to be those of low socioeconomic status, have competing health concerns or experience greater negative psychosocial consequences. However, we accounted for potential differential attrition by IPW, and thus, the generalisability of results was not compromised.

We conducted an interview study within this population shortly after the survey.³⁵ The study revealed that women made a great deal out of not being worried, were resistant to the fact that screening could cause harm and were eager to disprove the psychosocial consequences of mammography screening.³⁵ In a survey study, this can be considered a faking-good bias. Generally, women appreciated screening, overestimated the benefits and underestimated the harms.^36–38 Therefore, information about harms may be discordant with existing ideas about the benefits of screening.³⁹ Studies have shown that women who had false-positive screening results appreciate screening more after their final diagnosis,²⁶ which would make bias differential and mean differences biased towards null.

This was a sample of Danish women that was representative of Danish women who participate in screening. Findings may be generalisable to a setting similar to Denmark, such as Sweden and Norway, but might not be to countries with different healthcare systems or cultures.

Implications

First, we argue that the identified psychosocial changes should be considered important and relevant due to the large number of women who participate in mammography screening and the high cumulative risk of receiving a false-positive result. In a screening setting where apparently healthy citizens are affected, a minimal important difference should be evaluated differently than in a clinical setting where patients are treated and cared for.

Our results should be considered at two levels: (1) at the societal level when assessing the balance between the benefits and harms of screening mammography and (2) at the individual level when women are invited to screening and have to decide whether or not to participate.

On a societal level, these findings affect the benefit–harm ratio that informs policymaking. The psychosocial harms of false-positive mammograms now seem to become more comprehensive than previously assumed. Evidence has pointed to the fact that the benefits of screening mammography have decreased, which further affect the benefit–harm ratio.⁴⁰

At the individual level, this information should be included in information pamphlets for potential participants, so they are informed about the potential benefits and harms, and thereby can make an informed evidence-based choice.^41–46

While our study concludes long-term psychosocial consequences of false-positive screening mammography, it does not inform why. It should preferably be explored ethnographically, their experience with receiving the test results, and how and why their lives are still affected more than a decade after the false-positive mammography screening.

Data availability statement

No data are available. Please contact the corresponding author at emma.gram@sund.ku.dk.

Ethics statements

Patient consent for publication

Not required.

Ethics approval

This study involves human participants and was approved by the Danish Data Protection Agency (2007-41-0777). According to Danish Law on health research (§14, 2), questionnaire surveys do not require ethical approval. The legal information is available in Danish at the official website for Danish legal information: https://www.retsinformation.dk/eli/lta/2020/1338%23id3be671ac-0c7d-4363-bf57-173502ff61dc and at the official National Center for Ethics: https://nationaltcenterforetik.dk/ansoegerguide/overblik/hvad-skal-jeg-anmeldehttps://nationaltcenterforetik.dk/ansoegerguide/overblik/hvad-skal-jeg-anmelde.