Effectiveness of clinical breast examination as a ‘stand-alone’ screening modality: an overview of systematic reviews

Selection criteria for studies were based on the PICOS framework (PICOS – Population, intervention, comparator, outcome, study type) as follows: (1) Population: women aged 18+ years without a high-risk of breast cancer; (2) Intervention: CBE; (3) Comparator: no screening or other screening modalities; (4) Outcomes included were in relation to benefits (mortality and stage of detected tumour), harms (false-positive rate, over diagnosis, and overtreatment) and accuracy (sensitivity, specificity, positive predicted value, and negative predicted value); (5) Study type: systematic review with or without meta-analysis (Details of inclusion and exclusion criteria are presented in Appendix 1, Supplement materials).

Five bibliographic databases were searched: MEDLINE (via Ovid, 1946-present), EMBASE (via Ovid, 1947-present), Scopus (2004-present), Web of Science (1900-present), and the Cochrane Database of Systematic Reviews (1992-present) in October 29, 2019. Search inquiries did not apply a time limit, but an English language-only restriction was applied. Key words such as ‘physical examination’, ‘palpation’, ‘breast neoplasms’, and ‘breast cancer’ were used in the searches (Detailed search strategies for all databases in Appendix 2, Supplement materials). Additional potential papers were retrieved from the reference lists of included studies and websites of relevant organizations such as WHO, International Agency for Research on Cancer (IARC), CTFPHC, USPSTF, and ACS. Grey literature such as published reports were included.

All citations resulting from the searches were imported into EndNote X8 reference manager. After removing duplicated citations, a selection process was conducted in three stages including 1) Title and abstract screening, 2) Full-text review, and 3) Quality assessment. The AMSTAR 2 tool (A MeaSurement Tool to Assess systematic Reviews) was used to ascertain the quality of eligible systematic reviews before including them in the synthesis. Overall appraisal regarding the confidence in the results of the review included four categories “high”, “moderate”, “low” and “critically low” [24]. We included reviews with a high or moderate rating. AMSTAR 2 is interlinked with PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) statement published in 2009 [24]. Reviews published prior 2009, however, may receive a lower AMSTAR 2 rating due to changes in reporting requirements for systematic reviews over time. The review team considered and discussed this issue and the general lack of evidence about CBE. Our discussion led us to the decision to include reviews prior to 2009 with a low rating (due at least partly to reporting requirements rather the methodological quality per se) in order to avoid excluding potentially valuable information based on reporting as distinct from actual quality issues. However, reviews with a rating of critically low (content validity should not be relied on) were excluded (Ratings for all eligible reviews are presented in Appendix 3, Supplement materials).

A data extraction form created by the research team was pilot tested on three randomly selected included studies (~ 20% of all included studies) and refined accordingly. Relative risk of mortality reduction and downstaging effect (from advanced cancer to early stage cancer) were the primary measure for the benefits of CBE. False-positive rate was the primary measure for the harms of CBE. Accuracy of CBE were measured by sensitivity, specificity, and positive predicted value. In addition to the outcomes listed above (benefits, harms, and accuracy of CBE), extracted data also included general information on the articles’ objectives, design, search strategy, included studies, and quality assessment of the systematic reviews as well as their strengths and limitations. Decisions about which studies to include (during tittle/abstract screening and full-text review), AMSTAR rating, and data extraction were taken independently by two authors (TTN and NTQN) following the registered protocol. Disagreements were resolved by discussion with a third reviewer (CON).

Findings from included reviews were organized and presented (narrative synthesis) by pre-determined outcomes. In cases where a review was published as a peer-review article and as a published full report (grey literature), we synthesised and discussed related evidence from both sources. We also synthesised evidence from any updated reviews from their previous versions. When the results presented in a review were unclear, we examined original studies from which results were derived. Summaries of subgroup analysis in terms of age and race were provided where available.

None of the included reviews reported any ‘direct’ evidence (evidence from RCTs that compared CBE with no screening) of a benefit from CBE on breast cancer mortality [4, 6, 7, 27‐37]. However, we deemed it important to also consider ‘indirect’ evidence in order to assess fully the effects of CBE on mortality. That is, evidence that came from RCT comparing CBE with CBE + MMR, case-control studies comparing CBE with no screening, and RCT comparing CBE + MMR with no screening.

Firstly, using data from the Canadian National Breast Screening Study 2 (CNBSS-2) trial, five reviews shared the same assessment that well-performed CBE could provide the same effect on mortality reduction as mammography [30, 32‐34, 36]. In the 1980 CNBSS-2 trial, the breast cancer mortality rate was similar between the intervention (19,711 women received CBE + MMR yearly) and control arm (19,694 women received CBE yearly) with relative risk (RR) = 0.97 (95% CI: 0.62–1.52) [30, 32‐34, 36].

Secondly, two reviews considered evidence from case-control studies [7, 29]. USPSTF 2009 review included an US case-control study of women aged 40–65 years who had obtained a CBE in the last three to 5 years - it reported no effect of CBE alone on mortality (OR = 0.94, 95% CI: 0.79–1.12 and OR = 0.8, 95% CI: 0.59–1.08, respectively) [29]. JNCC 2016 review discussed the same US study plus another case-control study from Japan which found favourable results for CBE that one CBE within 5 years among asymptomatic women aged 30+ had a protective effect (OR = 0.45, 95% CI: 0.22–0.89) [7]. While presenting contradictory results, evidence from the Japanese study received less criticism than the US study [29].

Thirdly, using the data from the US Health Insurance Plan (HIP) trial (yearly MMR + CBE vs no screening), Barton et al. (1999) identified a 30% mortality reduction in intervention arm where 45% of cancer cases were detected by CBE alone [32]. It should be noted that at the time of this trial, mammography was not well developed compared to current practice and this underdevelopment might explain the high rate of cancer cases that were detected by CBE alone as well as the large contribution of CBE to the mortality reduction in this 1963 trial.

Only one of the included reviews [36] considered down staging as an assessment outcome for the effectiveness of CBE. In this IARC-conducted review, evidence for a down staging effect came from three RCTs comparing CBE with no screening (1996 Philippines trial, 1998 Mumbai trial, and 2006 Kerala trial). The results in the IARC review were unclear and did not include results from the 2000 Cairo trial and 2010 Sudan trial. Thus, we decided to examine the original five RCTs.

With one exception (Sudan trial), all trials showed a shift to a lower stage in the detection of tumours at diagnosis that was statistically significant (Fig. 2 and Appendix 5, Supplement materials). The longest follow-up trial (three rounds of screening from 1998 to 2005 in Mumbai) reported that the relative risk of having cancer at an advanced stage in the control group compared to the screened group was 1.68 (95% CI: 1.14–2.47) [23, 36]. Similar results were observed in the Kerala trial where the RR was 1.51 (95% CI: 1.13–2.04) [22, 36]. The absolute risk difference in the proportion of advanced stage cancers (stage III & IV) was at 17% (Philippines trial) and 47% (Cairo trial) higher in the control group compared to screened group [20, 36, 38]. The Sudan trial reported an absolute risk difference of 33.4% and calculated RR at 1.20 (95% CI: 0.29–4.95) which was not statistically significant though this might have been due to the short follow-up of the trial [19].

Four reviews looked at the FPR of CBE, of which, two reported ranges from 0.9 to 5.7% in RCTs and 2.2 to 5% in cohort studies [35, 36]. The remaining two reviews used the results of one 10-year cohort study in US to report that 13.4% of women had false-positive result on CBE at least once (CBE was provided biennially) and the cumulative risk of a false-positive result after 10 CBE was 22.3% [33, 37].

A wide range of values for the sensitivity of CBE was reported in 8/11 reviews though six of them reported a range from 40 to 69% with a pooled result of 54.1% [4, 7, 30, 32, 33]. One review reported a much higher upper boundary of 85% which came from a cohort study in Japan [36]. In contrast, another review recorded a much lower range from 28 to 36% which was derived from three cohort studies conducted in US during 1988–1998 [37]. In these studies’ setting, unlike in RCTs, the physicians (including nurses and radiologists) who performed CBE did not receive any training about the technique beforehand. Thus, quality of the procedure was questionable and might lead to the much lower sensitivity. Two reviews noted that sensitivity was higher for women in their 40s compared to women in their 50s (4–5% higher) [7, 32]. Regarding race, one review reported higher sensitivity among Asian women compared to Caucasian women (88% vs 35%, adjusted p = 0.04) [36]. Likewise, only one review described a difference in sensitivity when CBE was performed as a ‘stand-alone’ screening modality and when it was performed in conjunction with mammography (range: 63–69% versus 40–69%, respectively) [32].

The specificity of CBE was higher than 93% in all but one review [7, 30, 32, 36, 37]. The 2002 USPSTF review reported a slightly lower specificity of 88% based on a cohort study in US [33]. Barton et al. (1999) calculated a pooled result of 94% (95% CI: 90–97%) [32]. Sub-analysis for specificity of CBE was not reported in any included reviews.

PPV was documented in 5/11 reviews and ranged from 1 to 14% [30, 32, 34, 36]. One review reported a result of 46% which came from the 1963 HIP trial [33]. The pooled result calculated by Barton et al. (1999) was 10.6% (95% CI: 5.8–19.2) [32].

While there is no ‘direct’ evidence (from RCTs compared CBE with no screening) of an effect of CBE on mortality, the ‘indirect’ evidence (from case-control and RCTs compared CBE with other screening modalities) is indicative that CBE may be effective in reducing breast cancer mortality. CBE may have the desired effect in some settings and/or be as effective as mammography. The results illustrate the importance of context and culture. For example, Japan is the only country that has had a long-term, nation-wide population-based screening program using CBE (1987–2003) [7]. Given its longevity, studies conducted in this context may provide particularly valuable information. Included reviews reported results from a case-control study (Kanemura S, 1999) in which the odds of mortality for a group who were screened by CBE was lower than a group that did not receive screening [7]. We found similar results for a comparative analysis of age-adjusted death rates (ADR) due to breast cancer between Japanese municipalities with high- and low-coverage screening rates. Reduction in breast cancer ADR was 51.7% (p < 0.05) in the municipalities with coverage-rates higher than 50% compared to control municipalities (coverage < 10%) [39]. Another example is from the HIP trial which was conducted in 1960s when mammography was not well developed and thus, created an opportunity for a greater contribution from CBE to mortality reduction. A similar opportunity is also present in LMICs where mammography is absent due to inaccessibility.

Down staging was ignored as an outcome of CBE by most systematic reviews. Given the inaccessibility of and/or unaffordability of treatments for late staged cancer that may help explain why LMICs have considerably lower survival rates compared to HICs (from 13 to 50% in LMICs compared to 80% or higher in HICs) [14, 40, 41], early detection may be particularly relevant in this context. Thus, if CBE screening can contribute significantly to the shift towards early stages of tumour at diagnosis, this outcome is promising and worthy of investigation in LMICs for whom a national screening programme based on mammography would be prohibitively expensive.

Evidence in favour of a down staging effect came from both RCTs and observational studies. In RCTs, the absolute difference in proportion of advanced-stage cancer between intervention and control group ranged from 17 to 47%. Analysis from 11 regions of Japan showed 6% difference in late stage diagnosis between patients who were found to have breast cancer by mass-screening and matched patients with breast cancer detected in out-patient clinics [42]. A 5-year pilot program in Malaysia in which community nurses performed CBE in their monthly visit to women in the community brought a 40% reduction in advanced-stage breast cancer [43]. Similarly, a 3-year pilot started in 2009 in Tanzania witnessed the difference in proportion of advanced-stage cancer between a control village and an intervention village where CBE was provided to 99% of women was 9, 23, and 17% in 2009, 2010, and 2011 respectively [44]. Evidence across these studies is consistent and comparable with a recent modelling study in which the down staging shift was assumed at 25% [45].

Systematic reviews concurred that CBE has similar specificity with mammography (at 93–97%) but lower sensitivity (40–69% vs 77–95% respectively) though it is important to note that higher sensitivity does not guarantee a reduction in mortality as can be seen from the results of CNBSS-2. In the trade-off for sensitivity, CBE has lower false-positive rate compared to mammography (1–5% for CBE vs 7–12% for mammography) [29]. Cumulative FPR after 10 year was recorded as 22.3% for CBE and as high as 50–60% for mammography. Reviews also reported that CBE sensitivity is higher in younger women (40–49 vs 50–59 years old), Asian women, and when it is applied as a stand-alone screening modality (compared to the CBE + MMR combination). Thus, in the settings where these three aspects present (younger women with a higher risk of getting BC, a large population of Asian ethnicity, and CBE can be applied only as stand-alone screening modality), CBE screening may be of particular interest.

This is the first overview of systematic reviews to examine the effectiveness of CBE as a stand-alone screening modality. It is only the second review that is dedicated to CBE per se (rather than spreading the investigative focus to other screening modalities such as mammography, ultrasound, and magnetic resonance imaging-MRI). Our comprehensive and systematic approach to identification, selection, appraisal, and data extraction followed the methodological guidelines by the Cochrane Handbook of Systematic Reviews [46]. Any systematic review that may have been missed is due likely to an indexing issue related to the databases (e.g. reviews that were not indexed in the category of “systematic review”) rather than oversights in the search strategy. Assessment of the quality of reviews was performed using the AMSTAR 2 checklist and the collective expertise of the review team. However, an AMSTAR rating is subjective and depends on the quality of study reporting. Another limitation is that we were unable to include studies from non-English speaking countries, especially given our emphasis on LMICs.