Systematic review of the psychological consequences of false-positive screening mammograms

Incidence of breast cancer

Breast cancer is the most common cancer in the UK, with 48,034 new diagnoses in 2008. 4 It accounts for 31% of all cancers in women, with a one in nine lifetime risk. 4 The incidence of breast cancer in the separate countries within the UK can be seen in Table 1.

The number of cases of breast cancer in women has been steadily increasing in the UK over the last 30 years, with the annual incidence rising from 24,120 in 1978 to 47,693 in 2007. When the age of the women is standardised, the European age-standardised incidence rate increased by more than half (57%) over this time, from 77 per 100,000 in 1978 to 124 per 100,000 in 2008. 4 However, since the introduction of the national screening programme in 1988, the mortality rate has declined (Figure 1).

FIGURE 1.

Age-standardised European incidence and mortality rates for breast cancer in women, UK, 1975–2008. 4

The rise in incidence has been greatest in women in higher socioeconomic groups. 5 It is thought that this may be linked to their greater use of hormone replacement therapy for menopausal symptoms6 and the trend for having babies later in life. 7

When the figures are broken down by age, the effects of the screening programme can be seen by the sharp increase in incidence over this time among women aged 50–64 years. 8–10 The screening programme will detect cancers that would not have been noted in the patient's lifetime (overdiagnosis) and will bring forward the date of identification of cancer, finding it at an earlier stage, thus producing lead-time bias (Figure 2).

FIGURE 2.

Age-specific incidence rates, female breast cancer, UK, 1975–2007. 4

Mortality from breast cancer

The magnitude of the effect of mammography screening on breast cancer mortality is highly contested. Statistics for mortality from breast cancer in England are not available for 2009–10, but in 2008 a little over 10,000 women died from this disease, a rate of 26 per 100,000. 11 It can be shown that mortality from breast cancer began to decline in England at about the same time that the national breast screening programme was introduced, from about 40 per 100,000 to about 26 per 100,000 in 2008. 11 However, the exact effect that breast screening has had on breast cancer mortality is difficult to determine because it is hard to disaggregate the effects of improved treatments and other factors from the effects of screening.

However, a recent retrospective trend analysis by Autier et al. 12 has attempted to do this. Autier et al. 12 compared breast cancer mortality trends in three pairs of neighbouring European countries using WHO data from 1989 to 2006. The pairs of countries had similar demographics, quality and availability of health care. They differed in when breast cancer screening was introduced, with one country introducing it in about 1990 and the other about 10–15 years later. Autier et al. 12 calculated changes in breast cancer mortality using linear regressions of age-adjusted death rates. They found that although there was a wide difference in timing of the introduction of breast cancer screening in the pairs of countries there was a striking similarity in the rate of reduction of breast cancer mortality from 1990. They concluded that this reduction in breast cancer mortality was unlikely to have been the result of mammography screening. Similar findings have been replicated by the US Preventative Services Task Force,13 who have revised their endorsement of routine screening for women aged < 50 years. 13 In a letter to the British Medical Journal (BMJ), Bleyer14 presented a graph comparing US data with data from Autier et al. 12 Bleyer14 concluded that improved treatment, rather than screening, is the main reason for the reduction in mortality.

This research by Autier et al. 12 has been criticised by de Koning15 for being based on geographical comparisons which are unreliable and for the use of standardised all-age mortality. Further criticisms are that Autier et al. 12 have not accounted for the delay in time from the introduction of screening to realising its benefits.

Furthermore, in a recent systematic review of breast screening randomised controlled trials (RCTs), Gotzsche and Nielsen16 (n = 600,000) estimated that breast screening led to a 15% reduction in breast cancer mortality, but, conversely, that there was also a 30% increase in overtreatment of women whose cancer would never become apparent in their lifetime. 16 They estimated that, over 10 years, one woman would be saved from death by breast cancer for every 2000 women invited for screening. Additionally, 10 healthy women who would have remained undiagnosed would have been treated unnecessarily for breast cancer. On top of this, for the same cohort, at least another 200 women would go through the possible distress of a false-positive outcome. 16

Of the eight eligible trials in the Gotzsche and Nielsen review16 (New York 1963,17–19 Malmo 197620–22 and Malmo II 1978,23 Two–County 1977,24–26 Edinburgh 1978,27–29 Canada 1980,30–33 Stockholm 1981,34–37 Goteborg 198238,39 and UK Age Trial 199140–44), one was excluded from meta-analysis because the randomisation was seriously flawed and the data held to be unreliable (Edinburgh 197828,29). Gotzsche and Nielsen16 found that only three of the remaining trials had adequate randomisation. Pooling the data from these trials revealed no statistically significant benefit from screening on breast cancer deaths after 7 years {relative risk (RR) 0.93 [95% confidence interval (CI) 0.79 to 1.09]} or after 10 years (RR 0.90; 95% CI 0.79 to 1.02). When the data from these trials were combined with data from the other four suboptimally randomised trials, a statistically significant reduction in death from breast cancer was found after both 7 and 13 years [RR 0.81 (95% CI 0.72 to 0.90) and RR 0.81 (95% CI 0.74 to 0.87), respectively]. The pooled data from the adequately randomised trials similarly showed that there was no significant effect from breast screening on all-cause mortality after 7 years (RR 0.98; 95% CI 0.94 to 1.03) and after 13 years (RR 0.99; 95% CI 0.93 to 1.06). 16 Gotzsche and Nielsen16 did not present data on all-cause mortality from all the included trials because the estimates were unreliable.

FIGURE 3.

(a) Percentages of women taking part in mammography screening in each country. (b) Change in national breast cancer mortality rate relative to country's mean rate during 1980–5. Reproduced from Bleyer A. Breast cancer mortality is consistent with European data. BMJ 2011;343:d5630,14 with permission from BMJ Publishing Group Ltd.

These results are controversial and a lively debate continues about the benefits and harms of breast cancer screening. Gotzsche and Nielsen's16 findings have been heavily criticised by Kopans et al. ,45,46 who claim that the reduction in breast cancer mortality due to screening is of the order of 20–25%. Others are also highly critical and have estimated the reduction in breast cancer due to screening to be as much as 30%. 47,48 Indeed, one modelling study estimated the reduction in mortality due to screening to be between 28% and 65%. 49 The best methods for arriving at an accurate estimate of mortality reduction are also contested. 50 It is beyond the scope of this systematic review to attempt to resolve these differences. However, currently (2011), Professor Sir Mike Richards is undertaking a review to evaluate the benefits and harms of the NHS breast cancer screening services.

Psychological Consequences Questionnaire

The Psychological Consequences Questionnaire (PCQ)67 is a reliable and validated questionnaire that was developed specifically to measure the psychological consequences of mammography screening. It comprises two subscales, one consisting of 12 statements relating to possible negative consequences of mammography screening (i.e. how often in the past week the woman has experienced loss of sleep, change of appetite, feeling depressed, being scared, feeling tense, feeling under strain, being secretive, being irritable, withdrawing socially, having difficulty in doing ordinary activities at home and work or feeling worried about the future). The second subscale relates to potential benefits from screening and has 10 items covering feeling reassured, being able to cope better with everyday life, feeling less anxious about breast cancer, feeling more hopeful and a greater sense of well-being. The statements are scored on a four-point Likert scale ranging from not at all, rarely, some of the time to quite a lot of the time, with a range of 0–36 for the negative subscale and 0–30 for the positive subscale.

Cancer Worries Scale-Revised

The Cancer Worries Scale-Revised (CWS-R)68 is a six-item questionnaire that assesses the frequency of worries about developing cancer and how these worries affect daily mood and activities. It has been shown to be reliable and valid in at-risk populations. 68–70 The items are scored on a four-point Likert scale ranging from 1 (not at all or rarely) to 4 (almost all of the time).

Brief COPE

This is a shortened version of the COPE scale, which was developed to assess people's coping responses to stressful situations. The brief COPE71 has been used in breast cancer patients, although the work has not been published, and in hurricane survivors. The brief version has 28 items measuring a range of coping strategies (e.g. self-distraction, active coping, denial, substance use, emotional support, instrumental support, disengagement, venting, positive reframing, planning, humour, acceptance, religion and self-blame).

General Health Questionnaire

The General Health Questionnaire (GHQ) was designed as a screening tool for psychiatric illness in the context of general practice or general medical outpatients (i.e. non-psychiatric settings). 72 The GHQ covers the four domains of depression, anxiety, objectively observable behaviour and hypochondrias with 60 items. This instrument looks at recent experience and elicits responses on a four-point Likert scale using the statements less than usual, no more than usual, rather more than usual and much more than usual. A number of shorter versions have been developed: GHQ-12, GHQ-20, GHQ-28 and GHQ-30. 73 Items are scored 0–3 to give a total score or can be scored dichotomously with a particular threshold deemed to indicate that the respondent is ‘a case’.

Hospital Anxiety and Depression Scale

The Hospital Anxiety and Depression Scale (HADS) was developed to screen for psychiatric disorders in general hospital settings, excluding psychiatric wards. 74 It has two subscales, anxiety and depression, which are measured with 14 items using a four-point Likert scale (0–3). The items are totalled to give an overall score for anxiety or depression. Respondents with a score of 8–10 are considered to be ‘doubtful cases’ and ≥ 11 are considered to be ‘cases’.

Life Orientation Test-Revised

The Life Orientation Test75 is a validated measure of dispositional optimism. It has 10 items that are responded to on a five-point Likert scale, ranging from 0 (I strongly disagree) to 4 (I strongly agree).

State–Trait Anxiety Inventory

The State–Trait Anxiety Inventory (STAI) measures anxiety in adults as both a current state and an enduring trait. 76 It consists of a 20-item scale of how the respondent feels in general and a 20-item scale of how they feel now. Each item is scored on a four-point scale that ranges from ‘not at all’ to ‘very much so’. Higher scores are related to higher anxiety.

Internal validity

Consideration of internal validity addresses how well potential sources of bias and confounding are acknowledged and accounted for. Bias can be characterised as potentially undermining an experimental study in four ways: through selection bias, so that the participants in each group are dissimilar; performance bias, where the treatment of the different groups varies apart from the intervention; detection bias, which can occur if the study assessors are aware of which groups participants are in; and attrition bias, where all participants are not fully accounted for or violations of the study protocol have occurred. In particular, checks of study internal validity should address the following: whether or not there is sufficient description of the inclusion criteria, outcomes, study design, setting and the intervention to ascertain that study groups were similar in all respects and were treated in similar ways except for the intervention; if a justification for the sample size is given; if appropriate data analysis techniques were used; if dropouts and withdrawals are accounted for; if the technique used to account for missing data is described and adequate; and if assessors were blind to the group status of participants.

Another threat to validity can come from confounding. This is where an unknown agent is acting independently on the outcome being measured and the matter under investigation, so that an association appears to be occurring between the outcome measure and the matter of interest, but which is an artefact of the independent relationships.

External validity

External validity was judged according to the ability of a reader to consider the applicability of findings to a patient group and service setting. Study findings can only be generalisable if they describe a cohort that is representative of the affected population at large. Studies that appeared representative of the UK breast cancer screening population with regard to these considerations were judged to be externally valid.

Psychological impact

The papers from the OPCERG study in this domain span 5 years, with measures being taken at 1 month (T1), 5 months (T2) and 35 months (T3) from the last screening appointment. This study is a multicentre prospective cohort at 13 NHSBSP clinics in England and Scotland. The papers follow the same cohort of women over this time but the methods used appear to vary at different time points and it is not always clear on what basis the women included in the study have been selected or how the main outcome measure, the PCQ, has been used. The study includes women with normal mammograms as the control group and compares them with women with false-positive outcomes according to the process used in their further assessment (another mammogram, FNA or biopsy) or if they had been placed on early recall of 6 or 12 months following further assessment.

The first publication by OPCERG, by Ong et al. 104 (n = 877), reported primarily on the adverse psychological consequences of being placed on early recall (6 or 12 months recall rather than 3 years) at T1. These effects were measured using the PCQ negative scale which has 12 items on a four-point Likert scale (scored 0–3). 67 Participants were considered to be ‘cases’ if they responded positively to at least one item.

The study by Ong et al. 104 is a moderately good-quality study that used appropriate methods to address its aims. The authors reported almost all the key criteria specified by the STROBE statement. 81 However, they omitted to provide demographic data for all participants, not just those on early recall, which makes it difficult to interpret the results. They also failed to fully discuss the limitations of their study and its generalisability to other situations.

Following on from the study by Ong et al. ,104 Brett et al. 103 (n = 284) recruited from the same pool of women to find out what difference, if any, a further 4 months had on how these women were feeling after their false-positive mammogram. This was also 1 month before those on early recall were due for another screening mammogram. In this prospective cohort study, the previous studies' results (Ong et al. 104) were used as the baseline measures for comparison with the same subgroups. This study only included 12 centres, as one centre did not put any women on early recall.

Although this was a fairly good-quality study,103 it did not quite meet the same standards of reporting as the previous one. 104 Omissions include not considering potential sources of bias, giving an explanation of how missing data were handled and not providing demographic information.

The latest publication from OPCERG on the population at normal risk of breast cancer is by Brett and Austoker59 (n = 505). This study follows the same cohort of women as Brett et al. 103 in 13 NHSBSP clinics in England and Scotland. They took measures of adverse psychological consequences with the PCQ at 35 months after participants' last assessment (i.e. 1 month before their next routine screening mammogram was due). Brett and Austoker59 deemed that a total score ≥ 12 (range 0–36) on the PCQ showed negative psychological consequences and so it is only the percentage of scores ≥ 12 that are reported. It is not clear why they have chosen this total score as the cut-off point for showing psychological harm. The original validation paper by Cockburn et al. 67 does not have this cut-off but indicates that the bottom quartile of scores represent no dysfunction, the second quartile mild psychological disturbance, the third quartile moderate disturbance and the top quartile marked disturbance.

Brett and Austoker59 also used an ad hoc questionnaire to measure satisfaction with the breast screening service and assess factors that may influence women's level of anxiety. They also asked women about their intention to attend their next routine mammography screening. This last outcome was compared with their actual attendance.

Again, this is a generally well-described study but with similar omissions as before: no explanation of how missing data were dealt with or consideration of the role bias may have played in the results. However, they do provide some demographic information about the participants (marital status, home ownership and educational level).

Prior to the work by the Oxford team, Sutton et al. 55 (n = 1021) conducted a retrospective cohort study that was primarily interested in the levels of anxiety experienced by women attending routine mammography screening who had normal outcomes. However, they also looked at anxiety levels in women who had false-positive results. Nine months after the pre-screening baseline, they asked these women and others with normal results to retrospectively reflect on how anxious they had felt at six stages of the screening process: (1) receiving the invitation; (2) waiting at the clinic for the mammogram; (3) at the clinic after the mammogram; (4) waiting for the results; (5) after reading the results letter; and (6) now. These measures were taken on a three-point ad hoc scale, ranging from not anxious (1), a bit anxious (2) to very anxious (3). The retrospective results are only reported numerically at stages 3–5, therefore these data have been extracted. The data at other time points are reported in graphical form only and so could not be reliably extracted.

The baseline measures were taken using different measures (STAI76 and the GHQ72); however, the results of the women with false-positive mammograms are not disaggregated from the rest of the participants and so are not reportable.

This is a poorly designed and executed study. Initially, a cohort were recruited and followed up through the screening process. Additionally, another sample group was recruited which overlapped with the original sample group. A subset that contained participants from both groups was used for the main analysis (which compared women with normal mammograms with non-attenders and is therefore outside the scope of this review), while the whole of the first sample was used in the retrospective analysis of anxiety. There was no consideration of bias, confounding or the weakness of the design that relied on memory going back 9 months using an unvalidated measure. The results of the false-positive substudy are only partially numerically reported, the whole being shown in a figure that does not allow accurate disaggregation. Furthermore, there is no accounting of attrition or information about how missing data were dealt with. There is, however, acknowledgement that if they had used a disease-specific measure, such as the PCQ, they might have found different results.

Another similarly poor-quality study was conducted by Bull and Campbell106 in 1991 (n = 750). They followed a cohort of women who had attended mammography screening from their invitation to screening until 6 weeks after they had received the results of their follow-up assessment. The women were divided into four groups: (A) those invited for screening; (B) those with a normal mammogram; (C) those who went for further assessment with a mammogram, ultrasound or FNA; and (D) those who had a surgical biopsy for further assessment. The sample in groups A and B overlapped. However, no rationale is given for the design of the study. The outcomes measured were anxiety and depression using HADS,74 frequency of breast self-examination and impressions of the screening programme using an open question in an ad hoc questionnaire.

This study had a number of reporting flaws; in particular, there was no comment on the possibility of bias from the design and conduct of the study, nor was there any explanation of how missing data were accounted for. The only description of the participants was their age and no information was given about attrition or reasons for this. The discussion speculates at length about possible reasons for the findings, but neglects to mention any limitations of the study. Bull and Campbell106 also fail to comment on the generalisability of their findings.

The prospective cohort study by Ellman et al. 105 (n = 752) is of similarly poor quality. Ellman et al. 's105 aim was to compare the psychiatric morbidity experienced by women who had attended breast screening or a review clinic and had one of five results (group A, normal; group B, false-positive; group C, symptomatic with benign result; group D, symptomatic or recalled women who did have cancer; or group E, women with a history of breast cancer with or without symptoms). Only women in groups A and B were within the inclusion criteria of this systematic review and had their results reported. Morbidity was measured with GHQ-2872 either at the breast screening clinic (groups A and B) or the review clinic (groups C–E) before participants saw the clinician and was measured again 3 months later. An ad hoc questionnaire seeking opinions about the clinic and their experience of treatment was given to participants at follow-up.

Ellman et al. 's study105 failed to show an understanding of the limitations of their design. There was no consideration of potential sources of bias, no explanation of how missing data were accounted for or how this might affect the results. Although the tables of results were clearly presented and the results summarised in the discussion, there was no consideration of how generalisable they might be.

Psychological impact with family history of breast cancer

Although the population in this section is not part of the general breast cancer screening population or programme, it is interesting to look at this research and compare the results with that of the general population. The OPCERG and others collaborated in the Psychological Impact of Mammography Screening (PIMMS) research programme to look at the effects of having a false-positive screening mammogram in a population of women with a FHBC. The three papers reported in this section are from the PIMMS group.

Tyndel et al. 101 published the first paper in the series in 2007 (n = 2321). This was a prospective cohort study of women aged 35–49 years who had a moderate-to-high risk of breast cancer due to their family history and were on an annual screening programme. They were interested in comparing the psychological impact of having a false-positive mammogram with that of having a normal one within this group. Women who had previously had cancer and those with a family history of ovarian cancer were excluded. Disease-specific outcomes (PCQ and CWS-R68) were measured 1 month before participants' screening mammogram, 1 month after the ‘all-clear’ and 6 months after assessment.

This is a reasonably good-quality observational study. It is well-designed and well-reported; almost all the quality indicators in the STROBE checklist are met. The only substantial omission by Tyndel et al. 101 is a failure to report the extent of missing data and how these items were accounted for in the analysis.

The follow-on study is by Brain et al. 102 (n = 1286) who conducted a multiple regression analysis of the participants of the Tyndel et al. study101 to find out which pre-screening variables were predictive of cancer-specific distress. Initially, they conducted partial correlations of pre-screening variables on a number of scales (CWS-R,68 brief COPE,71 perceived risk of cancer and dispositional optimism) to see how they correlated with cancer worry at 1 and 6 months after assessment. This was followed by a hierarchical multiple regression to show predictive associations between these baseline variables and cancer worry at 1 and 6 months.

This second study from the PIMMS group was also of reasonable quality. The main criticism is the same: a failure to report how many data were missing and how missing data were dealt with in the analysis.

A qualitative interview study was nested within the PIMMS research. However, this was published only as the summary of a conference poster. The summary is reasonably detailed and, although it does not give the level of detail that would be expected in a qualitative research paper, it has been included because it is part of an included study and it gives some insight into the results found by the quantitative research. This interview study was conducted by Clements et al. 107 on 58 women who were part of the PIMMS cohort study (false-positive result = 22, normal result = 36).

As this study was published only as the summary of a poster, many of the expected quality criteria are unreported. However, the design is suitable for the research question and the thematic method of analysis appropriate. There is a lack of detail about the results and their interpretation as well as a lack of a theoretical framework and justification for the methods used. Therefore, it is not possible to comment thoroughly on the robustness of this study.

Impact of false-positive mammograms on returning for routine screening

The studies in this section have been subdivided into those that report on participants' actual returning for future breast screening and those that report on the intention to reattend.

Interventions to reduce the impact of being recalled for further assessment

No studies were found that directly addressed the issue of testing an intervention to relieve the negative psychological consequences of false-positive mammograms. However, two studies were found that looked at the information needs of women prior to further assessment, one from OPCERG (1994–7)110–112 and one from Smith et al. 113 These studies do not disaggregate the data according to the outcome of the recall (i.e. distinguish between those who had false-positive and true-positive outcomes). This breaches the inclusion criteria for this systematic review. However, these studies are included because at the time of recall the final outcome is unknown to the women and there is no reason to suspect that, prior to the event, women with false-positive outcomes would have different information needs from those with other outcomes.

From 1994 to 1998 Austoker, Ong and others from OPCERG published a series of three papers about their research into the information needs of women who were recalled for further assessment following routine mammography screening, plus national guidelines on behalf of NHSBSP. This research will be presented first followed by a study by Smith et al. 113 of recall invitation letters.

The original work that OPCERG studies came from is by Austoker and Ong112 (n = 1493). Their aim was to assess the need for written information among women who had been recalled for assessment following routine screening mammography. They did this by conducting a multicentre study of eight breast screening centres in the UK.

First, 484 women were invited for a questionnaire interview after recall and before assessment. Two weeks after assessment these and other women (total n = 1493) were given an ad hoc questionnaire to assess their response to being recalled and how worrying or reassuring the messages contained in recall literature were. Second, Austoker and Ong112 also collected samples of the literature sent to women from the different centres and found a wide variation in the amount of information given and the language used. They assessed the breadth of information given in the recall letters and leaflets and whether or not the women thought they had been given sufficient information on matters such as the reason for recall, the location of the centre, who could come with them, how long the appointment would take, what tests they would undergo, who they would see, when they would know the outcome and how to get more information. They also assessed the language used in the information literature and categorised it according to whether the participants found it to be particularly worrying or reassuring. Simple descriptive statistics are reported together with a few examples of comments made in the open questions.

This is a well-presented and clearly reported study. However, Austoker and Ong112 failed to gather information about the demographic profile of their participants; given that this was a large study, it would have been very interesting to see what light this information might have shone on the results.

Following on from this, and based on their original development of criteria for evaluating breast screening information material, Ong et al. 111 (n = 84) evaluated the health education literature in 84 breast screening units in the UK. They considered both the initial screening letter given to women and the recall letter and leaflet sent to those with suspicious mammograms. The evaluation criteria were divided into words and phrases that were particularly worrying and those that were particularly stress relieving. The number of centres that mentioned items in their literature in these criteria was recorded.

As this is an evaluation of written material with simple descriptive statistics, many of the STROBE quality criteria do not apply. The methods and results were clearly described and comprehensively discussed. The main criticism is that there is a lack of discussion of the limitations of the study.

This study led into the most recent of the primary research papers, which is by Ong and Austoker110 (n = 1493). They used an ad hoc questionnaire to elicit views of women about their experience of being recalled, in particular the quality of communication at the clinic (who was available to talk to at the clinic and related information needs). The questionnaire also had open questions that allowed a free response that underwent a discourse analysis.

Overall, this is a well-conducted and clearly reported survey. However, there are some omissions. The authors have failed to report the demographic characteristics of the participants or discussed what effect these may have had on the results, given no indication of how bias or confounding factors may have played a role in the results and have not considered other limitations of their study. Furthermore, the methods and results of the discourse analysis are not described, except to say that the most frequently occurring items or those deemed to be most important by the analysts are tabulated.

The most recent publication in this series is by Ong et al. 114 and is the national guidelines for improving the quality of written information sent to women who are recalled for assessment. The guidelines are based on the findings of the previous three papers and give advice on the content and wording of assessment invitation letters. The research evidence that is included in the report is a conflation of the findings of the studies and does not contain new evidence. Therefore, these data have not been extracted as they are already in the primary research paper data extraction forms. The guidelines are mentioned here for completeness.

The final study in this section is a poorly described survey of women's satisfaction with the Leicestershire Breast Screening Service in 1989–90 by Smith et al. 113 (n = 103). This includes assessing the adequacy of information given to women recalled for further assessment. Their aim was to compare three different versions of a recall letter giving different amounts of information, including telephone access to a breast care nurse, to see which was most acceptable to recalled women.

This study is described as an audit, but it goes further than that as it includes experimental testing of the different forms of the recall letter. The reporting of the study is very inadequate. Not enough information about the methods used is provided; it is often difficult to reconcile the text with the tables and there is no acknowledgement of the limitations of their methods.

Psychological impact

In the first paper from OPCERG, by Ong et al. 104 (n = 877), adverse psychological consequences were measured at T1 (1 month after the last screening appointment) using the negative PCQ subscale. The proportion of participants within each of the screening outcomes was recorded. The results are presented simply as proportions; therefore, the review authors (MB) have calculated RRs to provide greater insight into the relationship between false-positive and normal mammograms with adverse psychological consequences. Table 8 shows the outcomes at T1. These results show an increased risk of psychological distress at 1 month after the last screening appointment for women who had a false-positive result compared with women who had a normal mammogram. The risk of distress increases in line with the intrusiveness of the assessment process, so that women who had another mammogram had a RR of 1.71 (95% CI 1.24 to 2.35), whereas women who had a biopsy were at the greatest risk, RR 2.96 (95% CI 2.19 to 4.01). Those put on early recall or who had a FNA also showed increased distress, RR 2.13 (95% CI 1.58 to 2.87) and RR 1.97 (95% CI 1.44 to 2.69), respectively.

The next paper in this series is by Brett et al. 103 (n = 284) and reports outcomes 5 months after the last screening appointment (T2). Although all the women included at T2 were respondents at T1, it appears that they are a selected subset of Ong et al. 's104 participants, who were matched with T1 respondents placed on 6 months early recall, using undisclosed criteria. A comparison between T1 results in Brett et al. 103 (the selected subgroup) and T1 in Ong et al. 104 indicates that the Brett et al. 's103 subset has a higher proportion of adverse PCQ scores than the original study found (compare Table 8 with column 2 of Table 9), thus leading to higher RR estimates in the study by Brett et al. 103 than in that by Ong et al. 104

There is another source of uncertainty about the comparability of Brett et al. 's T1 results103 with Ong et al. 's T1 results:104 it is unclear what the proportion of women experiencing adverse psychological events represents. In the earlier paper (Ong et al. 104) the proportions represent those experiencing at least one psychological event on the PCQ and in the last paper in the series (Brett and Austoker59) it is the proportion of women scoring ≥ 12 on the PCQ. However, Brett et al. 103 do not disclose the method used for calculating the proportion. One clue is that they report that the PCQ is scored 1–4 rather than the conventional 0–3. If they have transposed the scores in this way, then a score ≥ 12 would be equivalent to a score of at least 1 on the original paper (Ong et al. 104) where the PCQ is scored 0–3. Assuming that the scores are directly comparable in this way it can be seen that, for all screening outcome categories, the level of psychological adverse effects has reduced over time (T1–T2) for the matched subset (see Table 9). This suggests that the matched subset in Brett et al. 103 is not representative of the population in Ong et al. 104 which would introduce a potential bias leading to overestimates of adverse psychological consequences at T1 for women having false-positive mammograms.

Furthermore, there is disagreement between tables within the paper by Brett et al. 103 about the benign biopsy and early recall results. In their table 1, 61% of women placed on early recall and 59% of women who had a biopsy show adverse psychological consequences, but in their table 2 these figures are reversed, although we are told that the difference between the scores at T1 and T2 was lower for women placed on early recall than for those who had a biopsy. Therefore, it has been assumed that table 2 is correct. The authors were contacted about this but no reply was received.

At T2 the results show that for all groups of women the level of distress had reduced. The greatest decrease in distress was for those women who had a biopsy at assessment (−30%). However, the RR of distress compared with those with normal mammograms had increased for all false-positive subgroups. Those women who had a biopsy continued to be at a greater risk of distress than those with a normal mammogram (RR 6.33; 95% CI 2.59 to 15.50). Those women who were put on 6 months early recall were also at a high RR (6.10; 95% CI 2.56 to 14.54). This is probably because they were only 1 month away from their next screening appointment (see Table 9).

Brett et al. 103 also conducted a logistic regression to find out which personal characteristics might be influencing the results. This showed that only having psychological consequences at 1 month (OR 5.82; 95% CI 2.70 to 12.56) and the kind of further investigation that women had (OR 4.40; 95% CI 1.35 to 14.35) were related to having psychological consequences at 5 months (Table 10).

In the final paper in the series, Brett and Austoker59 (n = 505) report outcomes at 35 months (T3) from the last screening appointment and compare them with the same participants at T1. It appears that this cohort has not been matched, as in the previous paper by Brett et al. ,103 but is the same cohort as in the first paper by Ong et al. 104 However, in this last paper, cases are those with a PCQ score ≥ 12. If the assumption of comparability is made between methods used to calculate T1 and T3 scores, then the proportion of women with normal mammograms experiencing psychological adverse events is similar at T1 (26%) and T3 (25%), although these results are higher than at T2 (10%). For those women with false-positive mammograms (other screening outcomes) the proportion experiencing adverse outcomes has reduced over time, although for those who were assessed with FNA the proportion who were distressed at T3 (45%) is similar that at T2 (44%) (Table 11).

An examination of the RR of experiencing psychological consequences at T3 and T1 show that this risk has diminished substantially for those assessed with a further mammogram, RR 1.28 (95% CI 0.82 to 2.00), as they were not statistically significantly more distressed than women who had a normal mammogram. Also, women who were placed on 6-months early recall showed a 23% reduction in their RR of distress at T3, RR 1.82 (95% CI 1.22 to 2.72). However, women who were assessed by FNA had the least reduction in RR (0.28) with a 35-month RR of 1.80 (95% CI 1.17 to 2.77). Those who had a biopsy maintained their status of having the highest RR of distress at T3, RR 2.07 (95% CI 1.22 to 3.52). Table 11 shows the results comparing T1 and T3.

The results from the OPCERG study indicate that there is an enduring relationship, lasting at least 3 years, between having a false-positive mammogram and exhibiting negative psychological consequences. This effect is shown to be in proportion to the degree of invasiveness of the assessment procedure, with 52% of women who were assessed by biopsy, 45% assessed by FNA, 32% assessed by further mammogram and 46% of those placed on 6 months early recall, compared with 25% of those with a normal mammogram, still experiencing distress 3 years after their last screening appointment. The numbers of distressed women may partly be explained by the impending date for their next screening appointment, although the proportions are still likely to reflect their assessment procedure.

The forest plot in Figure 6 gives an overview of the RR of psychological distress from a false-positive mammogram in relation to time and the method of assessment when compared with a normal mammogram.

FIGURE 6.

Forest plot of the RRs of negative psychological consequences from having a false-positive mammogram compared with a normal one by type of false-positive assessment at T1 (1 month after assessment), T2 (5 months after assessment) and T3 (35 months after assessment), measured with the PCQ.

The results from the ad hoc questionnaire, given at 35 months (T3), reported by Brett and Austoker59 shed some light on factors that may have influenced the level of distress experienced. Table 12 shows the items that were statistically significantly correlated with psychological distress at 1 month before the next routine screening (T3). The highest correlation with distress is with the lack of opportunity to talk to someone after the screening appointment (r = 0.35). The other most highly correlated factors are the waiting time between screening and assessment (r = 0.30), communication problems at the screening appointment due to anxiety (r = 0.29), fear of radiation (r = 0.28), amount of written information (r = 0.28), the performance of health workers (r = 0.27) and the unnecessary worry caused by the last breast screen (r = 0.26).

Preceding the work of OPCERG, Sutton et al. 55 (n = 1021) published results from their retrospective cohort study. These results show that, on their ad hoc questionnaire, when looking back over the previous 9 months, women who had false-positive mammograms were more likely to report that they were more anxious than women with normal mammograms at three time points: at the clinic after the mammogram, mean difference 0.24 (95% CI 0.03 to 0.45); while waiting for the results letter, 0.25 (95% CI 0.02 to 0.48); and after reading the results letter, 1.69 (95% CI 1.54 to 1.84). Unfortunately, Sutton et al. 55 do not numerically report how anxious the two groups were at the time they completed the questionnaire; these results are only presented as a graph which it is not possible to accurately read. The findings from this study are questionable as, apart from other methodological considerations of bias, the experience of having a false-positive mammogram may have coloured the women's view of how they felt at the time of screening and confounded their responses. It would seem unlikely that there should be any genuine difference between the two groups prior to receiving the results letter when it might be expected to find a difference in anxiety levels. A summary of the results from this study can be found in Table 13.

The results of Bull and Campbell's (n = 750) cohort study106 are ambiguous. The authors' ad hoc questionnaire of the frequency of breast self-examination found that 6 weeks after receiving the ‘all-clear’ women who had a false-positive mammogram examined their breasts more frequently than women who had a normal mammogram. Women who had a biopsy at assessment were more likely to self-examine once a week or more than those who had less invasive assessments. This may indicate that the women with false-positive results were more anxious about having breast cancer than those with normal results, although there is uncertainty about the meaning of very frequent breast self-examination. These results can be seen in Table 14.

When the results of both false-positive groups are combined there is a clear link between frequency of breast self-examination and having a false-positive mammogram, with the RR increasing with the frequency of self-examination (Table 15).

However, Bull and Campbell's106 results with the HADS tell a different story. They show no statistically significant difference in anxiety or depression between the groups at 6 weeks after assessment, with the majority in all groups being within the normal range on both subscales. Borderline anxiety and depression were higher in the false-positive group that did not have a biopsy, depression (12%) or anxiety (12%) than in the false-positive group that had a biopsy, depression (6%) or anxiety (8%). This may have been because a biopsy was seen as a more conclusive declaration of health than other methods of assessment (Table 16).

When the false-positive groups are combined, the RR of experiencing anxiety or depression is not statistically significantly greater in the false-positive group (Table 17).

The prospective cohort study by Ellman et al. 105 (n = 752) used the GHQ-28,72 another generic instrument, to measure psychological morbidity at the screening visit and 3 months later. They found that there were no statistically significant differences between women who had normal mammograms and those with false-positive ones. Scores ≥ 4 on the GHQ-28 are deemed to be ‘cases’ (Table 18).

The distribution of scores on the four subscales of somatic, anxiety, social dysfunction and depression can be seen in Table 19. These show that the proportion of scores above ‘case’ level declined in both groups between baseline and follow-up, with the greatest decline in the false-positive group for the somatic and anxiety subscales (40% to 26% and 44% to 29%, respectively).

When the RRs are calculated for psychological morbidity these again show that there is no statistically significant difference between those women with and those without false-positive mammograms at baseline and 3 months later (Table 20).

The forest plot in Figure 7 shows the results from Bull and Campbell106 and Ellman et al. 105 Here it can be seen that, although none of the results show clinical levels of general anxiety and depression, there is a trend in that direction.

FIGURE 7.

Relative risk of suffering clinical levels of general anxiety and depression following a false-positive mammogram compared with a normal mammogram, measured by Bull and Campbell106 (HADS) and Ellman et al. 105 (GHQ-28).

Psychological impact with family history of breast cancer

The PIMMS Management group's first results are from Tyndel et al. 's101 (n = 2321) study in the FHBC population. When their within-study group results are considered, the women who had a normal mammogram showed a statistically significant decrease in their distress levels between T1 (1 month before screening) and T2 (1 month after screening) on both measures (CWS-R and PCQ), whereas those in the recall group did not. However, when the between T2-and-T3 (6 months after screening) scores are compared, both groups show statistically significant reductions in distress over this 5-month period on both measures (Table 22).

The between-group scores are harder to interpret owing to the potential for bias to be introduced by lack of randomisation and demographic data that indicates that there are differences between the groups. The false-positive group have statistically significantly greater proportions of participants with biological children (p = < 0.05), a high-risk family history (p = < 0.05) and post-mammography hospital attendance for symptoms (p = < 0.05). The only statistically significant difference in the between-group scores is with the PCQ at T2, when those in the recall group {mean [standard deviation (SD)] 7.1 (7.44)} were more distressed than those in the normal result group [mean (SD) 4.08 (6.19)], this difference was no longer statistically significant 5 months later at T3 (see Table 22). However, these results, like those from the other studies, have not been adjusted for potential confounders (items on the PCQ and CWS-R). Tyndel et al. 101 present adjusted results (for the potential confounders listed above) in graphical form only (which could not be accurately transposed) that indicate that if these are considered then the recall group showed the greatest decrease in distress between T2 and T3 and the normal result group between T1 and T2 on both measures. The adjusted results were statistically significant; the level of significance is not reported.

Unusually, Tyndel et al. 101 also took measures on the positive subscale of the PCQ to see if there were any benefits from having a false-positive mammogram. They found that women who were recalled scored statistically significantly more highly at T2 [mean (SD) 13.02 (7.6)] than those with normal mammograms [mean (SD) 10.81 (6.9)]. However, this effect had diminished and was not statistically significant by T3. Additionally, when they had received their results at T2 and T3, women were asked whether their opinion of the benefits of breast screening had changed since their last visit. Women who were recalled were statistically significantly more likely to feel positive about the benefits of screening at both time points than those with normal mammograms, T2 = OR 3.16 (95% CI 2.14 to 4.70) and T3 = OR 2.35 (95% CI 1.53 to 3.61) (Table 23).

Overall, the levels of distress were similar in both groups although at T2, on the PCQ, the false-positive group showed greater distress. Additionally, 1 month after the all-clear (T2) those women with false-positive mammograms saw statistically significantly more benefits from screening and were more positive about the benefits of screening at T2 and 6 months later (T3).

Brain et al. 102 (n = 1286) then used the results of Tyndel et al. 101 to investigate which factors, evident at pre-screening, predicted cancer distress at 1 month (T2) and 6 months (T3) after assessment. The results of hierarchical multiple regression showed that, among others, cancer worry at 1 month after screening was predicted by having a false-positive mammogram (p < 0.05). This was no longer the case at 6 months after the all-clear. The model accounted for 61% of the variance at T2 and 57% at T3. Table 24 gives all the variables predictive of cancer distress in the FHBC population.

In order to gain greater understanding of their results and how women valued being in the annual FHBC screening programme, the PIMMS Management Group conducted an interview study of their participants who did (n = 22) and did not (n = 36) have a false-positive mammogram. 107 As this work has been published only as a poster summary, it is only possible to report an overview of their findings.

The thematic analysis of the interviews by Clements et al. 107 (n = 58) showed that being part of the FHBC screening programme helped to relieve fear of breast cancer and resulted in women feeling more in control of their family history. Women believed that taking part in screening would enable the earlier detection of cancer and that this would lead to a positive outcome. They also believed that a mammogram was more likely to detect early-stage cancer than actually is the case. They thought that an all-clear result meant that they did not have cancer. Although women who had a false-positive result were initially distressed, when they received the all-clear they had increased feelings of reassurance and security and a greater faith in the screening process than those with an initial all-clear result. They felt that being recalled was positive proof that screening worked.

Impact of a false-positive mammogram on returning for routine screening

Interventions to reduce the impact of false-positive mammograms

As mentioned in Interventions to reduce the impact of being recalled for further assessment we found no studies that directly addressed this issue. However, two studies were found that looked at the information needs of women who are recalled after screening. The initial results from the multicentre OPCERG study by Austoker and Ong112 (n = 1493) found that 92% of women were distressed or very distressed when they received their recall letter for assessment following a screening mammogram. Although no standardised measures of anxiety were taken, the quotes from the answers to the open questions in Table 32 give an indication of the range of responses.

Further results from the questionnaire compared the responses of women who had been given particular items of information in recall letters or leaflets with those who had not. Austoker and Ong112 found that for all items women wanted more information; this was especially the case when the item was not mentioned in the recall literature. The item that the greatest proportion of women wanted more information about was the reason for their recall (item mentioned in recall literature 36%, item not mentioned in recall literature 46%) (Table 33).

Austoker and Ong112 then compared the level of distress at being recalled with the reported need for more information using an ad hoc questionnaire. They found that women who were distressed or very distressed at being recalled reported a greater need for more information than women who were somewhat or not at all distressed at the prospect of further assessment. The items with the strongest relationship between distress and information need were ‘wanting to have more information about why they were recalled’, RR 1.84 (95% CI 1.58 to 2.14), and ‘more information about how to get more information’, RR 1.55 (95% CI 1.27 to 1.89) (Table 34).

Austoker and Ong112 also examined the information in the initial invitation to screening and compared levels of distress, using an ad hoc questionnaire, with whether or not the initial invitation referred to the possibility of recall. They found that when this was not mentioned in the letter or leaflet that women were more likely to be distressed or very distressed if they were recalled, RR 0.76 (95% CI 0.58 to 0.99). They also found an additional benefit from having an information leaflet as well as a recall letter. The RR of finding an aspect of the information about recall reassuring increased statistically significantly when participants were also given an information leaflet, RR 4.04 (95% CI 3.10 to 5.26). Having an additional leaflet also increased women's belief that they understood the assessment procedure, RR 1.27 (95% CI 1.22 to 1.33).

Austoker and Ong112 categorised words and phrases in the recall letter and information sheet, according to responses in the questionnaire and women's additional comments, to produce lists of information that were either reassuring or worrying (Table 35).

The study that followed on from this initial research into the information needs of recalled women, by Ong et al. ,111 used the categories developed above to evaluate the recall information letters and leaflets sent out by all the assessment centres in the UK. Of the total 87 centres, 84 (97%) sent their materials for evaluation.

The authors found that 99% of the information sent out with the initial screening invitation referred to the possibility of recall (Table 36).

When Ong et al. 111 compared the list of ‘worrying information’ with that contained in the recall letters and leaflets they found that 54% of the literature included one of these items (Table 37).

An examination of stress-relieving information found that 83% of recall literature contained at least one item (Table 38).

The data from the Austoker and Ong112 study was further analysed by Ong and Austoker110 (n = 1493) to find out women's views of the quality of communication at the assessment clinic, how this related to their level of distress and what role a breast care nurse might play in mitigating this distress. Table 39 shows that there was a strong link between whether or not participants would have liked to have talked with someone at the assessment centre [RR 1.42 (95% CI 1.15 to 1.74)], whether or not they thought they were given enough information about their physical examination [RR 1.71 (95% CI 1.02 to 2.86)] or X-rays [RR 2.34 (95% CI 1.47 to 3.73)] and whether they were distressed/very distressed or somewhat/not distressed about being recalled.

The results in Table 40 clearly show that the availability of a breast care nurse at an assessment centre greatly increased the probability that participants would have talked to somebody at the centre about the reason for their recall, RR 0.62 (95% CI 0.59 to 0.66). Where the opportunity to talk to someone was not available women were far more likely to have liked to talk to someone, RR 7.46 (95% CI 4.57 to 12.16). Women were also more likely to have had their assessment tests explained to them and not to need more information about their tests if a breast care nurse was available at the assessment centre (see Table 40).

The final study in this section is the survey of user satisfaction with the Leicestershire breast screening service by Smith et al. 113 (n = 103). Table 41 shows that they found that 75% of participants were upset or very upset when they received their recall letter.

The results of the survey about how satisfied the women were with information given in different versions of the recall letter about the reason for their recall and what would happen at the clinic showed that satisfaction varied considerably between the versions of the letter. Women were most satisfied with letter three that gave the most information about what would happen at the clinic and least satisfied with letter one that gave minimal information (Table 42).

When participants were asked if they would have telephoned a breast care nurse if one was available, 98% said they would and 100% of the women who received letter two with the breast care nurse's telephone number used this facility (Table 43).