Article Text

Download PDFPDF

Observer agreement comparing the use of virtual slides with glass slides in the pathology review component of the POSH breast cancer cohort study
  1. Emily Clare Shaw1,
  2. Andrew M Hanby2,
  3. Kevin Wheeler3,
  4. Abeer M Shaaban2,
  5. David Poller4,
  6. Sheila Barton5,
  7. Darren Treanor2,
  8. Laura Fulford6,
  9. Rosemary A Walker7,
  10. Deirdre Ryan8,
  11. Sunil R Lakhani9,10,
  12. Clive A Wells11,
  13. Heather Roche12,
  14. Jeffrey M Theaker12,
  15. Ian O Ellis13,
  16. J Louise Jones14,
  17. Diana M Eccles15
  1. 1Department of Cellular Pathology, Southampton General Hospital, Southampton, UK
  2. 2Department of Histopathology, St James's University Hospital, Leeds, UK
  3. 3Division of Cancer Sciences, University of Southampton, Southampton, UK
  4. 4Department of Cellular Pathology, Queen Alexandra Hospital, Portsmouth, UK
  5. 5MRC Lifecourse Epidemiology Unit, University of Southampton, Southampton, UK
  6. 6Department of Cellular Pathology, East Surrey Hospital, Surrey and Sussex Healthcare NHS Trust, UK
  7. 7Department of Cancer Studies and Molecular Medicine, University of Leicester, Leicester, UK
  8. 8Department of Cellular Pathology, St Bartholomew's Hospital, London, UK
  9. 9School of Medicine and Centre for Clinical Research, Molecular and Cellular Pathology, University of Queensland, Queensland, Australia
  10. 10Department of Anatomical Pathology, Pathology, Queensland, Australia
  11. 11Department of Histopathology, St Bartholomew's Hospital, London, UK
  12. 12Department of Cellular Pathology, Southampton General Hospital, Southampton, UK
  13. 13Department of Histopathology, University of Nottingham, Nottingham, UK
  14. 14Tumour Biology Laboratory, Cancer Research UK Clinical Cancer Centre, Institute of Cancer Studies, Queen Mary's School of Medicine and Dentistry, London, UK
  15. 15Division of Cancer Sciences, University of Southampton, Southampton, UK
  1. Correspondence to Dr Emily Clare Shaw, Department of Cellular Pathology, Mail point 2, Southampton General Hospital, Tremona Road, Southampton SO16 6YD, UK; ecshaw{at}doctors.org.uk

Abstract

Aims (1) To compare the use of scanned virtual slide images (virtual microscopy) with glass slides (conventional microscopy) in the assessment of morphological characteristics of breast cancers within the setting of the Prospective study of Outcomes in Sporadic versus Hereditary breast cancer (POSH), involving a cohort of women under 40 years of age, presenting with breast cancer. (2) To assess the acceptability to histopathologists of the use of virtual slide images.

Methods 13 histopathologists from the UK and Australia participated in the POSH pathology review. The observers were asked to assess multiple morphological features such as tumour grade and type. Comparisons were made for a single observer using both virtual images and glass slides. Intra- and inter-observer variability was calculated using the κ statistic and a comparison was made between the use of each image modality.

Results Diagnostic performance with virtual slides was comparable to conventional microscopic assessment, with the measurement of agreement best for vascular invasion, necrosis and the presence of a central scar (κ=0.37–0.78), and poor for more subjective parameters such as pleomorphism, stroma, the nature of the tumour border and the degree of lymphocytic infiltrate (κ=0.1).

Conclusion Virtual slides represent an acceptable methodology for central review of breast cancer histopathology and can circumvent the need for either travel to view material, or the potential problems of sending it by post.

  • Virtual slides
  • breast cancer
  • digital pathology
  • breast pathology
  • cancer research
  • computer assisted
  • digital pathology
  • breast
  • pagets disease
  • CERB 2
  • cancer genetics
  • collagenous colitis
  • crohns disease
  • thyroid
  • gastric pathology
  • apoptosis
  • immunohistochemistry
  • steroid receptors
  • molecular oncology
  • oncology
  • molecular biology
  • molecular genetics
  • molecular pathology
  • cancer
  • histopathology
  • diagnosis
  • cytology

Statistics from Altmetric.com

Background

Many clinical trials rely on central review of the pathology in order to provide consistent histopathological parameters to correlate with outcome data. Traditionally, histopathological assessment involves the examination of glass slides. A number of digital systems are emerging that use electronically scanned microscopic images (virtual slides). Currently these are not in widespread use in diagnostic histopathology, though they are increasingly being used in research and education. Centralised review of breast cancer material is necessary since breast cancer represents a heterogeneous group of diseases with distinct morphological and molecular profiles, with the morphology being predictive of molecular attributes and also contributing to the most widely used prognostic index, the Nottingham Prognostic Index. The assumed implication of this diversity is that predicting biological behaviour or response to a therapy requires accurate histopathological categorisation and may relate only to one of these categories. Trials where central pathological review is necessary have either circulated the slides by post (eg, the SUPREMO trial1) or employed central review, with the trial histopathologists attending a ‘central’ site to analyse batches of cases (eg, the Age trial2). The main disadvantages of these strategies are the potential for loss of or damage to slides in transit and the time inefficiency and associated expense of histopathologists travelling what are frequently large distances in order to review slides. The ability to undertake these analyses ‘virtually’ has the potential to circumvent the above difficulties, and also to involve histopathologists not in an academic setting, who have minimal or no protected research time to permit travelling for central review.

The Prospective study of Outcomes in Sporadic versus Hereditary breast cancer (POSH),3 aims to determine whether the prognosis of breast cancer patients who harbour BRCA gene mutations differs from that of matched non-carrier patients, and whether breast cancers occurring in patients with different predisposing inherited gene mutations have a consistent and distinct tumour phenotype that could help histopathologists to identify potential gene carriers. The latter aim requires review of the histological parameters of each of the study cases, but with over 2500 tumours in the study cohort, this task represents a considerable logistical undertaking. In order to facilitate efficient throughput of the cases for review and scoring, it was proposed that this analysis could be performed remotely using virtual slide scanning combined with a high capacity server and internet access to facilitate the analysis of cases remotely and involvement of expert histopathologists at a variety of locations. This pilot study was instigated to compare the level of agreement between observers for assessment of glass slides versus virtual slides in a subset of the POSH cohort (222 cases).

Materials and methods

The study used a custom-made software application named POSHviewer, created by one of the authors (KW). POSHviewer uses web-based technologies to give histopathologist or radiologist reviewers at any location access to scanned slides and radiological images from patients in the POSH study. Glass slides were scanned centrally at the Leeds Institute of Molecular Medicine using Aperio XT scanners (Aperio, Vista, California, USA) and a 40× objective lens at high resolution (0.23 microns per pixel). A commonly used method of tiling at various magnifications4 was employed in order to provide a set of images that could be accessed using standard browsers operating at normally available web access speeds. The images were stored on the Institute's servers. A range of discrete magnifications were available to the user (figures 1 and 2). To allow accurate assessment of mitotic activity, the virtual ‘field diameter’ (ie, the diameter of the screen) at 40× magnification was calculated from the known, fixed scanning resolution of the image (in dpi (dots per inch), easily converted to dots per mm). The image width of 1024 dots was independent of the monitor used to view the image, since a lower resolution monitor would merely create an image that scrolled off the edge of the screen, although the field diameter would include the scrolling section.

Figure 1

A virtual slide viewed in the POSHviewer software at baseline magnification.

Figure 2

An area of the slide shown in figure 1 at the maximum 40× magnification.

The POSHviewer system was used to present one slide from each case to two different histopathologists, provide them with a template for inputting the assessed morphological features from each slide (screenshot in figure 3), and store the data for subsequent analysis. Each time a histopathologist logged into the POSHviewer system and requested work, a case from the POSH cohort was randomly allocated to them from the outstanding cases. There was no set requirement to review a certain number of cases, so workload varied according to a user's activity on the system. The virtual slide reading took place first and then glass slides were sent out several months later to those histopathologists who indicated their willingness to perform further reading for the purposes of this study. Data from 13 pathology reviewers in the POSH study were tabulated using SPSS for Windows V.16.0 according to method of assessment (glass or POSHviewer) and compared within and between assessors for the different modalities. We also attempted to assess the acceptability of online slide assessment to the pathology reviewers in the POSH study in the form of a follow-up survey sent by email.

Figure 3

A screenshot of part of the pathologist data entry form on the POSHviewer system.

Statistical analysis

The κ statistic has been calculated (using Stata V.10.0) to compare the results between groups of observers and measure agreement. This is the method used in the National Health Service Breast Screening Programme External Quality Assessment Scheme. The level of agreement is measured and compared to that expected purely by chance. There are different interpretations of the κ score values, however the following Landis and Koch classification is in widespread use:5 no agreement to slight agreement (<0.20); fair agreement (0.21–0.40); moderate agreement (0.41–0.60); substantial agreement (0.61–0.80); near-perfect agreement (>0.81).

Results

The number of cases reported by pathology reviewers varied widely: three histopathologists reported just one case each using the POSHviewer application, and the three most avid users reported 130, 109 and 67 cases, respectively.

The results were analysed in four separate datasets (n=number of ‘reads’); see figures 4 and 5:

Figure 4

Assessment using the POSHviewer (Pv) program versus glass slides. Assessment by the same reader (green) or different readers (blue) is compared for multiple morphological features examined in breast cancers in the POSH study patients. This figure is produced in colour in the online journal; please visit the website to view the colour figure.

Figure 5

Use of the POSHviewer program (green) versus glass slides (blue) by the same reader to assess multiple morphological features examined in breast cancers in POSH study patients. The CIs cover a wider range of values for glass slides than the POSHviewer due to the numbers assessed being lower. This figure is produced in colour in the online journal; please visit the website to view the colour figure.

  • Glass slides versus POSHviewer: same readers (n=69).

  • Glass slides versus POSHviewer: different readers (n=86).

  • Glass slides versus glass slides: different readers (n=20).

  • POSHviewer versus POSHviewer: different readers (n=222).

The best agreement between different modalities and the same modalities was seen for vascular invasion (κ scores for datasets 0.66 and 0.55 vs 0.78 and 0.58), necrosis (κ scores 0.45 and 0.50 vs 0.56 and 0.47) and the presence of a central scar within the tumour (κ scores 0.43 and 0.37 vs 0.46 and 0.40). There was little difference between observers or modalities. The lowest level of agreement (with 95% CIs crossing zero) was seen for different readers assessing glass slides for:

  • Overall grade (κ=0.34; CI –0.036 to 0.715).

  • Tubule formation (κ=0.167; CI –0.225 to 0.559).

  • Pleomorphism (κ=0.211; CI –0.212 to 0.633).

  • Characteristics of the stroma (κ=0.040; CI –0.309 to 0.389).

When comparing glass slides to POSHviewer for different readers, poor agreement was seen for:

  • Pleomorphism (κ=0.150; CI –0.048 to 0.348).

  • Stroma (κ=0.150; CI –0.009 to 0.309).

  • Lymphocytic infiltrate (κ=0.117; CI –0.028 to 0.262).

  • Tumour border (κ=0.145; CI –0.067 to 0.357).

The third component of grade—mitotic rate—showed moderate inter-observer variation for different readers using glass slides (κ=0.42; CI 0.11 to 0.73) or virtual slides (κ=0.35; CI 0.26 to 0.44) only and moderate inter-modality variation for the same reader assessing glass and virtual slides (κ=0.37; CI 0.2 to 0.53).

Nine of 13 pathology reviewers registered as users on the POSHviewer system replied to the follow-up survey designed to assess the acceptability of virtual slides to the user, giving a 69% response rate. Most accessed the program from a work computer, half of which were university-owned and half of which were NHS-owned. The majority of those who had used a non-work computer as well as a work one found the program slower on the work computer, suggesting that network speed was a limiting factor in health service and university settings. Further information obtained from the survey is shown in table 1.

Table 1

Discussion

We present data to support the view that virtual microscopy provides a valid and reliable medium for histopathologists to assess certain morphological features in scanned histological specimens. The κ scores for the morphological features demonstrating best agreement between different observers and reporting modalities (presence of tumour necrosis, central scar and vascular invasion) provide evidence that in the hands of these assessors, these features could be confidently and consistently identified using either conventional or virtual microscopy. These findings are consistent with published data generated by the National Health Service Breast Screening Programme External Quality Assessment,6 in which these features have been shown to have only moderate degrees of inter-observer agreement. Inter-observer agreement of assessment of stromal characteristics was poor for both conventional microscopy alone (κ=0.040) and conventional versus virtual microscopy (κ=0.15). This may be because this feature has only recently been recognised as being of potential importance in classifying breast cancers and does not form part of the recommended Royal College of Pathologists minimum dataset for breast cancer reporting.7 It is likely to be the case that histopathologists are not used to describing stromal characteristics and do not have access to well defined criteria for doing so, unlike those in existence for the standard dataset items.

The different morphological features identified as showing the poorest agreement may be those that are particularly difficult to assess using computer-based modalities. However, the scores between different readers assessing glass slides only showed similarly poor agreement, suggesting that this represents inter-observer variability independent of the modality. The numbers of cases assessed using conventional microscopy were lower than those for virtual microscopy, due to the design of this pilot study with the emphasis on facilitating use of digital technology. The current study is limited to the assessment of cases with malignant pathology, predominantly high-grade due to the nature of the disease occurring in this cohort.

The data indicate better agreement on the degree of tubule formation between different reviewers using POSHviewer (κ=0.54; CI 0.43 to 0.65) as opposed to glass slides (κ=0.17; CI –0.23 to 0.56). This is consistent with the findings of a previous study into the use of virtual microscopy in a bowel cancer screening programme, in which there was an improvement in inter-observer agreement with the use of virtual microscopy rather than glass slides to assess the extent of a villous component in an adenoma.8 This finding lends further support to the assertion that the whole slide view provided as the on-screen starting point in virtual microscopy permits superior assessment of the microscopic architecture of a lesion when compared with conventional microscopy.

Our overall experience of inviting colleagues to use this technology was that a minority of histopathologists enthusiastically engaged with the program and assessed a large number of cases using it, whereas most histopathologists tried it for few cases but indicated that they felt more confident using conventional microscopy. Specific problems identified as a result of the follow-up survey were deterioration of the image with ‘pixellation’ while trying to navigate around the screen, for example, to assess the overall growth pattern or identify areas with the highest mitotic count, and also assessment of fine morphological detail such as the presence of nucleoli and identification and counting of mitotic figures.

Our data also demonstrates the willingness of some histopathologists to use virtual slide technology and that the minimum necessary technological requirements are already in place in some institutions.

Using virtual slides can lead to cost savings as well as avoidance of the logistical obstacles involved in setting up a central review meeting and reimbursing travel, accommodation and subsistence for the attending histopathologists. In previous similar work in the area of BRCA gene mutations,9 the international Breast Cancer Linkage Consortium arranged central histopathological review of breast cancer cases from 440 patients and 547 controls. This involved international collaboration by seven expert breast histopathologists: five from Europe in a 4-day first sitting and the original five plus an additional two from the USA and Australia in a 3-day second sitting. In our opinion, including international expertise is ideal but can be prohibitively expensive; the virtual slide methodology circumvents this issue and contributes to an improvement in the quality of the review. The slides in this study were scanned at a central location; however since then it has been possible to extend scanning to two other locations and successfully input scanned images from all three different systems—Aperio, Ariol (Leica, Wetzlar, Germany) and Olympus Dotslide (Olympus, Southend-on-Sea, UK)—to the central database. In the authors' experience, an increasing number of diagnostic pathology departments are acquiring the necessary capabilities to undertake virtual slide production.

The current state of virtual microscopy technology has been the subject of several review articles.10–14 Virtual slides have not been approved or widely adopted for diagnostic pathology to date. A recent study examining the use of virtual slides in diagnostic cervical cytology revealed that the cytologists strongly preferred glass slides over virtual slides and that ability to focus through the entire depth of the slide was a significant requirement, producing an average file size of 7.5 GB per case.15 There are an increasing number of commercial companies showing activity in this area, and some EQA schemes (eg, urological pathology and bowel cancer screening programme) already use online images and submission. Virtual slides are used most commonly as a conference, research and educational tool. Although the potential merits in the research setting are clear, the main drawback of this emerging medium in the diagnostic setting is that acceptability to histopathologists in general has yet to be established, and this has been our experience in the POSH study to date. There are considerable issues surrounding the availability and cost of the hardware and software needed to view this material, particularly on networked workplace computers.

Take-home messages

  • In the current study, diagnostic performance with virtual slides was comparable to conventional microscopic assessment of the morphological features of breast cancer.

  • A follow-up survey indicated a generally positive disposition of the study histopathologists towards this emerging modality, but some obstacles to implementation need to be addressed before wider adoption of virtual slide technology.

  • This provides evidence that virtual slides represent an acceptable methodology for assessment of some of the morphological features of breast cancer.

Acknowledgments

We would like to acknowledge input from the Digital Pathology Project team at the Leeds Institute of Molecular Medicine, University of Leeds as well as funding from Cancer Research UK and the Breast Cancer Campaign, and assistance with study administration and block handling from Lorraine Durcan and William Goodison.

References

Footnotes

  • Funding This study was supported by Cancer Research UK and the Breast Cancer Campaign.

  • Competing interests None.

  • Ethics approval Ethics approval was provided by the South and West Multi-Centre Research Ethics Committee (reference MREC/00/6/69).

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.