The Breast (1998) 7.75-79 0 1998 Harcourt Brace & Co. Ltd
ORIGINAL ARTICLE
Scottish experience of double reading in the National Breast Screening Programme H. E. Deans, D. Everington, C. Cordiner, A. E. Kirkpatrick and E. Lindsay Breast Screening Centre, Aberdeen, UK S U M M A R Y. The aim of this study was to assess and quantify the results of the double reading regime in the Scottish Breast Screening Programme for the years 1992-1996, and to use this information to make a recommendation as to whether this practice should be continued. This study is a retrospective data analysis of the Scottish Breast Screening Programme (part of the UK NHSBSP). Outcome data were analysed for all women attending the Scottish Breast Screening Programme from 1 April 1992 to 31 March 1996. The number of additional cancers detected by the second reader only and the number of additional women recalled by the second reader only were assessed.Double reading resulted in the detection of 259 additional cancers by the second reader in the 4-year period. This represents 10.5% of the total cancers detected in that time. The effect of double reading on the number of recalls can only be obtained for a subset of the data. Analysis of this subset showed that an additional 4616 women (27% of the total number of women recalled) were recalled to obtain 170 additional cancers (12% of total cancers). Double reading has increased the cancer detection rate of the Scottish Breast Screening Programme. Over 10% of cancers were detected only by the second reader. This improvement in sensitivity has been maintained over the 4 years reviewed. Using a second reader resulted in a 13% increase in the cancer detection rate, although this was associated with a 37% increase in recall rate. At present we find no reason to discontinue the practice of double reading, although this will be reviewed at intervals as technological advances are introduced.
INTRODUCTION
was felt that this would be balanced by increased cancer detection and would accelerate the radiologists’ learning process. Prevalent round screening in Scotland was completed in 1994 and most centres have completed their first incident round. There are now sufficient data to evaluate the Scottish experience of double reading and to make recommendations as to whether the practice should be continued.
Although the Forrest Report’ provided the baseline for the implementation of the National Breast Screening Programme in the UK, there is little specific mention of the process of reading screening mammograms. The report observes that the purpose of the basic screen is not to make a diagnosis but to decide which women require further investigation. The report ascertained that while the reading of screening mammograms should be the responsibility of a consultant radiologist, the task might be devolved to trained non-radiologist personnel, with a radiologist reviewing difficult and suspicious films. However, in recognition of the radiology training required and of the obvious learning curve while the programme was being established, the Scottish Breast Screening Programme was funded for double reading of single view mammograms by two radiologists, a situation which did not pertain in all areas of England and Wales. Whilst it was recognized that double reading would probably lead to an increased recall rate, it Address correspondence to: H. E. Deans, Breast Foresterhill Road, Aberdeen AB25 2XF, UK
Screening
SUBJECTS AND METHODS The Breast Screening Office Radiographic Screening Form 1, ref. l/5/95 (Fig. l), is used by all Scottish centres to record screen reading results. The actual double reading regime may vary slightly from centre to centre and will have evolved as a pragmatic method of service delivery aimed at increased sensitivity. The most commonly adopted practice is that radiologists 1 and 2 view screens separately. They identify themselves on the form by their code number and mark either the normal or the review box. In the recall cases, the area of concern may be indicated on the schematic diagram on the form and the lesion may be marked on the mammogram itself. As reader 2 uses the same screening
Centre,
75
76
The Breast Reader 1
Reader 2 Radiologisr/Radlographer Ref
GMC
GMC
Radiolognt
Radiologist
Radiographer
Radiographer
Technical recall (tick)
1
II
NommllBenigdOiher Routi Rd
NormaliBenignlOther Ra*,ev(SynlpmlN)
(oblique view)
R
L
R
L
Fig. 1 The section of SBSP Form 1 completed by 2 radiologist time of initial screen reading. Reader 2 has access to the opinion reader 1.
at the of
form as reader 1 and thus may have access to the opinion of reader 1, this is not independent double reading. In theory it is completely random as to which member of the team is reader 1 or 2, but in reality, individual radiologists’ working practice and preferred time of screen reading will often result in a bias toward any individual being either a first or a second reader. Other than in Glasgow, where a triple reading policy has been adopted, patients are reviewed on the basis of any single reader recall without discussion between participating radiologists.
The Glasgow triple reading system The Glasgow Screening Programme’s catchment area covers over 40% of Scotland’s eligible women and operates the service from two static centres and 3-4 mobile units. Since the Glasgow programme started, radiological input has varied considerably both in staff numbers and in level of experience. (There have been 17 radiologists at the two Glasgow Centres, compared with a combined total of only 18 radiologists at the other 5 Centres of whom 12 have been in post since the onset of the programme.) In 1990, following routine monitoring of data against the nationally set standards, it was decided to institute a system of triple reading in an endeavour to reduce the recall rate while still maintaining a high small cancer detection rate.
Triple reading occurs where there is a disagreement in opinion between the first and second readers; such mammograms are then submitted to a third read. All radiologists participate as either first, second or third reader. The third reader views all triple reads without any knowledge of the previous readers’ findings. The third reader then decides whether the outcome is return to routine recall, assessment or technical recall. The software design allows only two reader opinions to be recorded, with only one opinion of recall being necessary to automatically generate a letter of recall. Hence to ensure that the correct letters are generated, only the two concordant readers’ opinions can be entered. It is therefore not possible to tabulate the centrally held Glasgow figures in a comparable format, as the recall rate and cancer detection figures for readers 1 and 2 will always be in agreement. The Glasgow centres, however, have maintained local records from which individual cancer detection (but not recall) rates have been obtained (see Table 3). Data from Form 1 (i.e. the Radiographic Screening Form), Form 2 (i.e. the Review and Assessment Clinic Form) and subsequent information from cytopathology, pathology and surgical forms are collected and collated on computer at the Central Coordinating Unit for the Scottish Breast Screening Programme at the Information and Statistics Division of the Scottish Health Service, Edinburgh. This forms the Scottish Breast Screening Database from which the statistics for this review have been obtained. Data covering the period 1 April 1992 to 31 March 1996 have been analysed. These figures include all appointment types (i.e. routine, early recall and self-GP referrals) and also both non-invasive and invasive cancers. Only one outcome of cancer per woman has been counted (i.e. if a woman has both non-invasive and invasive disease or has bilateral invasive cancers, only one invasive cancer will be registered in the results). This definition is used in the KC62 (Department of Health Returns) and for consistency has been adopted for the Scottish statistics. The figures, therefore, will underestimate the total number of cancers detected. Since the purpose of this paper is to assess the impact of a two-reader system using data available from the Central Breast Screening Database, interval cancers have not been included. Scottish interval cancer data have been analysed separately and the results are currently being prepared for publication. Most interval cancers will have been returned to routine recall by both readers, with few having been recalled for assessment by one of the readers.
RESULTS The figures considered for this review have been provided
Double reading in the SBSP 77 from the Scottish Breast Screening Programme database and the locally held Glasgow database. Table 1 shows recall rates (as a percentage of attendances) for Scotland and for Scotland excluding Glasgow over the 4-year period from 1 April 1992 to 31 March 1996. The individual recall rates for readers 1 and 2 are also shown for Scotland excluding Glasgow. From Table 1 it can be seen that reader 2 has a constantly higher recall rate when compared with reader 1. A fall in recall rates might have been expected for the following reasons: the increasing experience of radiologists; the adoption of a two-view regime for all prevalent round attenders since July 1995; and the heightened priority given to recall rate reduction motivated by a data review of 19941 1995. The UKCCCR trial of one- and two-view mammography’ indicated that this would result in a decrease of up to 15% in the recall rate for women being screened for the first time. However, at least during the introduction of this change of policy, this does not appear to have been the case as recall figures for 1995-1996 show only a slight reduction. No attempt has been made for the purposes of this paper to break down figures into prevalent or incident round recalls, although the impact of two views for prevalent round screening is currently under assessment. The lower recall rates in 1994/1995 and 1995/1996 will hopefully be maintained. Tables 2-4 demonstrate the cancers detected by only reader 1 and only reader 2, showing a persistent increase in the number of cancers detected by adopting a double reading protocol. This increase has been maintained over time and in Scotland has resulted in the detection of 259 extra cancers in 4 years, which represents 10.5% of the total cancers detected in that time. The figures quoted include non-invasive and invasive tumours. The increase in cancer detection has, however, been accompanied by increased recall rates. Tables 5-8 quantify Table
1
Recall
Scotland excluding Glasgow Reader 1 Reader 2
Overall
1992-1993 1993-1994 1994-1995 1995-1996
7.9 8.2 7.3 6.7
5.2 4.4 4.8 4.7
7.1 6.1 6.8 6.4
Single
Cancers detected only by reader 1
1992-1993 1993-1994 1994-1995 1995-1996 Total
Table
reader-detected
6.2 5.5 5.9 5.8
cancers,
Cancers detected only by reader 2
Single reader-detected
4
Scotland
excluding
Year
Cancers detected only by reader 1
Percentage of total cancers detected
1992-1993 1993-1994 1994-1995 1995-1996
6 10 2 7
40 44 48 38
12.2 12.3 13.7 11.3
329 357 351 336
Total
25
170
12.4
1373
Glasgow Percentage of total cancers detected
3 1 3 5
24 24 21 20
7.5 7.8 9.3 8.1
320 306 226 248
12
89
8.1
1100
Single reader-detected
cancers,
Total cancers detected
Scotland
Year
Cancers detected only by reader 1
1992-1993 1993-1994 1994-1995 1995-1996
9 11 5 12
64 68 69 58
9.9 10.3 12.0 9.9
649 663 577 584
Total
37
259
10.5
2473
Cancers detected only by reader 2
Percentage of total cancers detected
Total cancers detected
this in relation to the extra cancers detected. These figures relate to Scotland excluding Glasgow, as the latter’s triple reading system precludes a similar analysis from the central database. Table 5 shows the total number of women recalled and cancers detected. Tables 6 and 7 show data relating to readers 1 and 2, respectively. The effect of double reading (obtained by comparing Table 5 with Table 6) is given in Table 8. The use of the second reader has resulted in an extra 170 cancers (12% of the total cancers) detected in the 4-year period. To obtain these extra cancers, 4616 additional women were recalled, which represents 27% of the total number recalled in the same 4 years. The double reading Table 5 Glasgow
1992-1993 1993-1994 1994-1995 1995-1996 Total
Glasgow
cancer,
Cancers detected only by reader 2
Combined No. recalled
Overall
2
3
rates as % of attendances
Scotland Year
Table
Table Year
Table
6
Total cancers detected
Total
Recall rate (%)
No. of cancers
1 and 2, Scotland
excluding
Cancer detection rate (per 1000 women)
No. of cancers per 1000 women recalled
4486 4188 4335 3967
7.1 6.1 6.8 6.4
329 357 351 336
5.2 5.2 5.5 5.4
73.3 85.2 81.0 84.7
16976
6.6
1373
5.3
80.9
Reader No. recalled
1992-1993 1993-1994 1994-1995 1995-1996
data for readers
1 individual Recall rate (%I
data, Scotland No. of cancers
excluding
Glasgow
Cancer detection rate (per 1000 women)
No. of cancers per 1000 women recalled
3314 3018 3075 2953
5.2 4.4 4.8 4.7
289 313 303 298
4.6 4.6 4.8 4.8
87.2 103.7 98.5 100.9
12360
4.8
1203
4.7
97.3
78
The Breast
Table
7
Reader
2 individual
No. recalled 1992-1993 1993-1994 1994-1995 1995-1996 Total
Table 8 Glasgow
Recall rate (%)
data, Scotland No. of cancers
excluding
Glasgow
Cancer detection rate (per 1000 women)
No. of cancers per 1000 women recalled
3922 3775 3788 3616
6.2 5.5 5.9 5.8
323 347 349 329
5.1 5.0 5.5 5.3
82.4 91.9 92.1 91.0
15101
5.9
1348
5.2
89.3
Only
reader
2, i.e. ‘gain’
Additional no. recalled
of second
Additional no. of cancer
reader,
Scotland
excluding
Additional no. of cancer 1000 women recalled
1992-1993 1993-1994 1994-1995 1995-1996
1172 1170 1260 1014
40 44 48 38
34.1 37.6 38.1 37.5
Total
4616
170
36.8
per
process has therefore increased the cancer detection rate by 13% (5.3 compared with 4.7 per 1000 women screened) and there has been an associated increase in the recall rate of 37% (4.8% compared with 6.6%). There has been no trend over time with regard to any of the outcome measures shown in the Tables. Tables 6 and 7 demonstrate the higher recall and cancer detection rates of reader 2 compared with reader 1. There are proportionately fewer cancers in the additional women recalled by reader 2. This is shown by the last columns in the tables which give the number of cancers detected per 1000 women recalled. This can be seen more clearly in Table 8 which concentrates only on those women recalled by reader 2. The lower figure of 37 additional cancers per 1000 additional women recalled (compared to 97 and 89 cancers per 1000 women recalled for reader 1 and reader 2) is not surprising as one might expect these tumours to be more subtle in their presentation. Without the security of the second reader process, reader 1 may have recalled more women with these more subtle signs. The effect that this would have on the cancer detection rate of reader 1 is an unknown quantity. For this analysis, however, it is assumed that the behaviour of reader 1 would not alter in a true single reader situation.
DISCUSSION Published assessments of other double reading programmes have provided differing results and conclusions. Ciatto et al3 found only a 4.6% increase in cancer detection in association with a 15% increase in recall rate, but the Scottish experience is more in line with that of several other
groups,4-7 where double reading has produced a lo-14% increase in cancer detection. These groups employ a variety of double reading regimes, some involving completely independent double reading and others adopting a consensus method, and this makes comparison of every element of the double reading process difficult. The Scottish Breast Screening Programme is now well established and subject to rigorous quality assurance with continuous audit of individual performance, cancer detection rates and interval cancers. For a radiologist involved in the screening programme, there is nothing more disappointing and distressing than the discovery of an interval or incident round cancer that might have been detected at an earlier screen, and for that reason alone the increase in detection rate afforded by double reading might be deemed a justifiable practice. There are also less tangible benefits derived from double reading. Radiologists appreciate the close collaboration and support of their colleagues, the enhancement of team strengths and the sharing of opinion and experience. The team approach is much valued in both the screen reading and review clinic situations where there is a feeling of being less isolated through having the reassurance of one’s colleagues’ input. Whether there is any actual medicolegal benefit in having two opinions as opposed to one is uncertain. In the symptomatic breast service, double reading is not common practice, although in this situation mammography comprises only one part of a breast assessment. Double reading is a well-established practice in the UK Breast Screening Programme, a recent paper estimating that 76% of all screening films are currently double-read.’ Whilst the benefits of double reading are likely to be greater at the introduction of a national programme, the necessity to continue the practice has become a topic of considerable debate and was questioned at the 1996 Royal College of Radiologists Breast Group Meeting. In a climate of increasingly scarce resources and with the imperative for a costeffective service, it is only correct that double reading should be evaluated and that consideration be given to any other strategies (such as reducing the screening interval or two-view mammography)’ that might produce similar or improved cancer detection rates. Radiologists tend not to view the double reading practice in terms of financial cost, but it is recognized that cost is a fundamental element of health gain, and an objective economic analysis of the double reading in the Scottish Programme has been undertaken and forms the basis of a tandem paper. The Scottish practice has always involved double reading by radiologists but it has been suggested that similar results might be obtained from a radiographer/radiologist duo or a variety of other skill mix options.8~‘0It may be that a particular radiologist performs consistently better as a first or second reader and that fixed pairings might be advantageous. The logistics of organizing such a system around other commit-
Double reading in the SBSP ments, may make this difficult to implement, although it is recommended that performance indicators are reviewed locally, taking this into account. The variation in double reading practice is a further area of interest. Several centres have reported an increase in cancer detection associated with a decrease in recall rates using the consensus approach.5,6,8The Scottish experience of this is limited to Glasgow, where a triple reading system was implemented primarily in an attempt to reduce a high recall rate. As for the other Scottish centres, we see no great benefit in reverting to an independent double reading regime. By knowing that a woman has been recalled by reader 1, reader 2 can quickly decide whether or not he or she agrees in the knowledge that the woman will be reviewed regardless. Reader 2’s efforts can then be concentrated on those not recalled, as it is in this group that there is the potential to detect additional cancers. It is the Scottish radiological experience that double reading is an effective practice which has resulted in the detection of 259 extra cancers, i.e. 10.5% of all cancers detected over 4 consecutive years. The increase in cancer detection produced by double reading has been maintained over the time span of this study and there is, at present, no evidence to support discontinuation of this practice in favour of an alternative strategy.
Acknowledgements We thank all our Scottish Breast Screening colleagues of SBSP Information Systems Project Team at Trinity
and the members Park House,
79
Edinburgh for their cooperation and support throughout the production of this paper. We also thank Dr Freda Alexander and Jan Warner, the SBSP National Coordinators, for their input, and our sincere thanks to Elaine Thomson for her secretarial help and unlimited patience.
References 1. Forrest P. Breast cancer screening: report to the Health Ministers of England, Wales, Scotland and Northern Ireland. London: HMSO, 1986. 2. Wald N J, Murphy P, Major P et al. UKCCCR multicentre randomised controlled trial of one and two view mammography in breast cancer screening. BMJ 1995; 311: 1189-l 193. 3. Ciatto S, Turco M R D, Morrone D et al. Independent double reading of screening mammograms. Journal of Medical Screening 1995; 2: 99-101. 4. Thurfjell E L, Anders L K, Taube A A S. Benefit of independent double reading in a population-based mammography screening program. Radiology 1994; 191: 241-244. 5. Antinnen I, Pamilo M, Soiva M, Roiha M. Double reading of mammography screening films - one radiologist or two?. Clin Radio1 1993; 48: 414-421. 6. Warren R M L, Duffy S W. Comparison of single reading with double reading of mammograms and change in effectiveness with experience. Br J Radio1 1995; 69: 958-962. 7. Anderson E D C, Muir B B, Walsh J S, Kirkpatrick A E. Efficacy of double reading mammograms in breast screening. Clin Radio1 1994: 49: 248-25 1. 8. Wells J C, Cooke J. Film reading practice of UK breast screening units. The Breast 1996; 5: 404-409. 9. Blanks R G, Moss S M, Wallis M Cl. A comparison of two view and one view mammography in the detection of small invasive cancers: results from the National Health Service breast screening programme. Journal of Medical Screening 1996; 3: 200-203. 10. Pauli R, Hammond S, Cooke J et al. Comparison of radiographer/radiologist double film reading with single reading in breast cancer screening. Journal of Medical Screening 1996; 3: 18-22.