Guidelines for May-June 1997

 
Proposed Guidelines for Primary Screening Instruments for Gynecologic Cytology
Developed by the Intersociety Working Group for Cytology Technologies

These guidelines were approved by the governing boards of the following six societies represented in the Intersociety Working Group for Cytology Technologies: American Society for Cytotechnology (ASCT), American Society of Clinical Pathologists (ASCP), American Society of Cytopathology (ASC), College of American Pathologists (CAP), International Academy of Cytology (IAC), and Papanicolaou Society of Cytopathology (PSC). The proposed guidelines as written reflect the current status of technologies in early 1997. These guidelines may evolve over time as newer technologies are developed.

1.0 General Considerations
This document delineates proposed guidelines in evaluating primary screening instruments for gynecological cytology (cervicovaginal screening) specimens.
     The goal of this document is to provide guidance to the FDA and other regulatory agencies as they evaluate such instruments.

1.1 Intended Use
Primary Screeners are devices which are intended to triage gynecologic cytology slides for identification of malignant or premalignant disease. Slides are scanned by these devices and are classified, either interactively or independently, into two categories: No Review Indicated (NRI), and Needs Further Screening (NFS). Slides determined by the Primary Screener to be excluded from human evaluation should not contain abnormalities from ASCUS/AGUS to invasive cancer at a rate exceeding that achievable with primary manual screening.
     Such devices may or may not Òelectronically commentÓ on specimen adequacy, endocervical component, or the presence or absence of infectious disease agents. If the devices are not capable of making these assessments then each slide should be reviewed manually for these components as clinically indicated or requested.

Discussion
The potential benefit of these devices is that they may increase the overall sensitivity without significant loss of specificity by identifying an equal or greater number of abnormal slides (again, abnormal is defined as ASCUS/AGUS+) than a manual screening alone.
     The primary criterion for evaluating these devices should be performance in correctly separating normal slides from those containing neoplastic and preneoplastic conditions. While other functions, such as evaluation of specimen adequacy, the presence of an endocervical component, etc. are important, they are not requirements sine qua non. If these devices do not perform these auxiliary functions automatically then these auxiliary functions would need to be performed manually as clinically indicated or requested. The group of slides categorized as NRI should not contain slides which are unsatisfactory. While manual evaluation for auxiliary findings can generally be performed more rapidly than evaluation of a slide for neoplasia, it should be recognized that any manual evaluation of a slide for any reason reduces the utility and cost-effectiveness of these devices.
     These devices do not compensate for smears which are inadequately prepared, are not fully representative of the ectocervix and endocervix, or are not accompanied by and interpreted in light of appropriate clinical information.

1.2 Recognized Limitations
Use of these devices should not affect the interval recommended between gynecologic cytology screenings. There is no substitute for a regular cervicovaginal cytologic screening, particularly because roughly half of all falsely negative tests are due to conditions unrelated to screening error.1

2.0 Function
Primary screener devices functionally fall into two categories which are not mutually exclusive:
     1. Interactive. Interactive devices display potentially abnormal cells or cell clusters in a computer window-display after they are scanned by a computerized microscope and analyzed by a computer for features associated with abnormalities. Cytotechnologists review these computerized images or computerized ÒabstractsÓ of each slide and make a decision whether or not to manually screen each slide based upon findings on the computer screen. If a cytotechnologist makes a decision not to review a slide further, the slide is signed out as negative with no further follow-up.
     2. Independent. These devices automatically scan slides and assign a score to each slide based on the potential for abnormalities as ascertained by the computer program. Those slides which are assigned a score higher than a predetermined threshold are subsequently rescreened by humans. Those slides which are assigned a score below the threshold are signed out as NRI without human examination.
     The result of using either type of device, interactive or independent, or the two in combination, is the same: The creation of a group of slides which are subjected to one or more human screenings, and a group of slides which are signed out as NRI and which are never manually screened by a human for the presence of ASCUS/AGUS+. The putatively negative group should contain either no or a very small number of slides which are truly high grade squamous intraepithelial lesions (HSILs) or malignant (these latter situations could have grave consequences for the patient, especially if a slide of a later stage lesion is incorrectly classified as negative).
 
Table I     Summary of How Machine Error Compounds Human Error

Sort rate (%)

Machine sensitivity (%) 

Human sensitivity for LSIL+  (%) 

Number of additional missed LSIL+ 

cases in population of 40,000* 

20 
30 
40

99

95
90

75

75
75

7.5

37.5
75.0

*In this population, the prevalence of dysplasia/cancer (LSIL+) lesions is 2.5% and the human false negative rate is 25%. This example illustrates a non-interactive (independent) primary screening instrument. More detailed explanations are also available from member societies or working group members.
 
3.0 Metric of Performance
There are two classes of performance metric which will be considered: Approvability items and disclosure items.

Approvability
Approvability items are concerned with the performance of the device as it relates to correctly identifying NRI and NFS slides. The explicit standard is that the sensitivity of the device plus the cytotechnologist equals or exceeds the sensitivity of primary manual screening. In addition, the device plus the cytotechnologist should not cause significant loss of specificity compared to manual screening alone. Sensitivity and specificity obtained by device plus cytotechnologist and by cytotechnologist alone should be determined in a prospective, two armed (contemporaneous) adjudicated trial. Performance specifications (defined as sensitivity and positive predictive value of the instrument alone and in combination with manual screening) should be available to potential users.
     While devices need not subclassify, render specific diagnoses, or even separate completely all the normals from the abnormals, performance (sensitivity and clinical specificity) should be specified for: ASCUS/AGUS, LSIL, HSIL plus Cancer. Clearly, a missed cancer is more serious than a missed ASCUS. The sensitivity with respect to cancer and intra-epithelial lesions should equal or exceed primary manual screening, and the results should be reproducible.
     It is recognized that in a properly adjudicated, two-armed study design, it may be difficult to achieve expert consensus on which slides are truly ASCUS. Therefore, for this category it is important for manufacturers to report performance results separately and not combine them with intraepithelial lesions and carcinoma. If the increase in sensitivity is almost exclusively in ASCUS/AGUS without significantly increased sensitivity for LSIL and above, questions regarding overall clinical value should be raised. Device algorithms may need improvement or study designs may need reevaluation.
     Instruments should be expected to identify for human review the full range of glandular cell groups in the AGUS designation.
     For each category of disease (cancer, intraepithelial lesions, ASCUS and AGUS) the manufacturer should submit to the FDA contemporaneous data reflecting the Positive Predictive Value of Positive Results (PPVPR) of performance with and without use of the instrument in each clinical site.

Disclosure items
Potential users should be informed of device performance limitations with respect to: stain variability, identification of adequate endocervical and squamous components, identification of unsatisfactory slides, identification of endometrial cells, identification of infectious organisms, acceptance of slides produced by a range of sample preparation devices (smears versus liquid-based), and slide rejection rate. In addition, any patient-related criteria which would exclude specimens from device evaluation should be stated. These are all characteristics which may limit the practical utility of the device in a routine setting and should be disclosed.
     Disclosure should include data on sensitivity, specificity and PPVPR separated by diagnostic categories of ASCUS/AGUS and LSIL+.
     It would appear misleading for manufacturers to claim a new Òstandard of practiceÓ in their product advertising. Professional standards of practice are established over time by the profession.
 
Table II     Detailed Explanation of Assumptions and Results of 20% Prescreening Sort Rate

Assumptions:
  Population Assumptions:
    Population 40,000
    Prevalence of LSIL+ in percent 2.50%
    Prevalence of LSIL+ as a number 1,000
  Machine Performance Assumptions: 
    Percentage of slides reviewed by humans 80%
    Percentage of slides never seen by humans 20% If humans review 80%, they never see 20%
    Sensitivity of machine to LSIL+ slides 99% Better than human, on the face of it
    Percentage of LSIL+ slides never seen by humans  (false negative of the machine)  1% An inevitable consequence of previous line
    Number of LSIL+ slides never seen by a human (completely missed!)  10 Again, COMPLETELY MISSED. ZERO chance of finding these slides 
    Number of LSIL+ slides which will be part of 80% pool reviewed by humans, and have the potential for being identified  990 Only 75% would be ultimately identified (see human sensitivity below) 
  Human Performance Assumptions:
    Human sensitivity  75%
    False negative rate (1-sensitivity)  25%
Results:
  Human Screening (the way things are done now):
    Truly LSIL+ slides found by humans alone 750 False negative rate times number of LSIL+ slides in population 
    Truly LSIL+ slides missed by humans alone 250 Sensitivity times number of LSIL+ slides in population
  Machine Prescreening:
    LSIL+ slides machine fails to bring to attention of humans (therefore never seen by humans)  10 These misses occur entirely in the 20% of slides which humans never see. This is a direct consequence of the 99% sensitivity (1% false negative rate of the machine).
    Slides humans miss among slides humans actually screen 247.5 This is product of the human false negative rate (25%)  times the number of abnormal slides (990) available for human review. Remember, humans look at only 80% of slides which do contain 99% of abnormals. 
    TOTAL missed slides by machine + human  257.5
    False negative rate using human + machine  25.75% Total missed slides divided by total LSIL+ slides in population
  The Bottom Line:
    Missed LSIL+ slides by human alone  250.0
    Missed LSIL+ slides by human/machine combo  257.5
    Number of additional missed LSIL+ slides by use of machine + human instead of human alone 7.5 
  CONCLUSION: 
    Use of Machine which finds 99% of abnormals while reducing total number of slides to be examined by 20% results in 7.5 additional missed slides in a population of 40,000 slides with a 2.5% prevalence rate of LSIL+

 
 
 
Table III     Detailed Explanation of Assumptions and Results of 30% Prescreening Sort Rate

Assumptions:
  Population Assumptions:
    Population 40,000
    Prevalence of LSIL+ in percent 2.50%
    Prevalence of LSIL+ as a number 1,000
  Machine Performance Assumptions: 
    Percentage of slides reviewed by humans 70%
    Percentage of slides never seen by humans 30% If humans review 70%, they never see 30%
    Sensitivity of machine to LSIL+ slides 95% Better than human, on the face of it
    Percentage of LSIL+ slides never seen by humans  (false negative of the machine)  5% An inevitable consequence of previous line
    Number of LSIL+ slides never seen by a human (completely missed!)  50 Again, COMPLETELY MISSED. ZERO chance of finding these slides 
    Number of LSIL+ slides which will be part of 80% pool reviewed by humans, and have the potential for being identified  950 Only 75% would be ultimately identified (see human sensitivity below) 
  Human Performance Assumptions:
    Human sensitivity  75%
    False negative rate (1-sensitivity)  25%
Results:
  Human Screening (the way things are done now):
    Truly LSIL+ slides found by humans alone 750 False negative rate times number of LSIL+ slides in population 
    Truly LSIL+ slides missed by humans alone 250 Sensitivity times number of LSIL+ slides in population
  Machine Prescreening:
    LSIL+ slides machine fails to bring to attention of humans (therefore never seen by humans)  50 These misses occur entirely in the 30% of slides which humans never see. This is a direct consequence of the 95% sensitivity (5% false negative rate of the machine).
    Slides humans miss among slides humans actually screen 237.5 This is product of the human false negative rate (25%)  times the number of abnormal slides (950) available for human review. Remember, humans look at only 70% of slides which do contain 95% of abnormals. 
    TOTAL missed slides by machine + human  287.5
    False negative rate using human + machine  28.75% Total missed slides divided by total LSIL+ slides in population
  The Bottom Line:
    Missed LSIL+ slides by human alone  250.0
    Missed LSIL+ slides by human/machine combo  287.5
    Number of additional missed LSIL+ slides by use of machine + human instead of human alone 37.5 
  CONCLUSION: 
    Use of Machine which finds 95% of abnormals while reducing total number of slides to be examined by 30% results in 37.5 additional missed slides in a population of 40,000 slides with a 2.5% prevalence rate of LSIL+

 
 
Table IV     Detailed Explanation of Assumptions and Results of 40% Prescreening Sort Rate

Assumptions:
  Population Assumptions:
    Population 40,000
    Prevalence of LSIL+ in percent 2.50%
    Prevalence of LSIL+ as a number 1,000
  Machine Performance Assumptions: 
    Percentage of slides reviewed by humans 60%
    Percentage of slides never seen by humans 40% If humans review 60%, they never see 40%
    Sensitivity of machine to LSIL+ slides 90% Better than human, on the face of it
    Percentage of LSIL+ slides never seen by humans  (false negative of the machine)  10% An inevitable consequence of previous line
    Number of LSIL+ slides never seen by a human (completely missed!)  100 Again, COMPLETELY MISSED. ZERO chance of finding these slides 
    Number of LSIL+ slides which will be part of 80% pool reviewed by humans, and have the potential for being identified  900 Only 75% would be ultimately identified (see human sensitivity below) 
  Human Performance Assumptions:
    Human sensitivity  75%
    False negative rate (1-sensitivity)  25%
Results:
  Human Screening (the way things are done now):
    Truly LSIL+ slides found by humans alone 750 False negative rate times number of LSIL+ slides in population 
    Truly LSIL+ slides missed by humans alone 250 Sensitivity times number of LSIL+ slides in population
  Machine Prescreening:
    LSIL+ slides machine fails to bring to attention of humans (therefore never seen by humans)  100 These misses occur entirely in the 40% of slides which humans never see. This is a direct consequence of the 90% sensitivity (10% false negative rate of the machine).
    Slides humans miss among slides humans actually screen 225 This is product of the human false negative rate (25%)  times the number of abnormal slides (900) available for human review. Remember, humans look at only 60% of slides which do contain 90% of abnormals. 
    TOTAL missed slides by machine + human  325
    False negative rate using human + machine  32.50% Total missed slides divided by total LSIL+ slides in population
  The Bottom Line:
    Missed LSIL+ slides by human alone  250.0
    Missed LSIL+ slides by human/machine combo  325
    Number of additional missed LSIL+ slides by use of machine + human instead of human alone 75.0 
  CONCLUSION: 
    Use of Machine which finds 90% of abnormals while reducing total number of slides to be examined by 40% results in 75 additional missed slides in a population of 40,000 slides with a 2.5% prevalence rate of LSIL+

 

4.0 Design Requirements for a Protocol
A study which is used to determine performance should have the following elements:
     1. Two-armed prospective design: Device performance in a primary screening modality should be compared to contemporaneous primary manual screening for the same slide population.
     2. Discordant study results from the two study arms should be adjudicated by an independent panel consisting of experienced cytology professionals. The goal of the adjudication process is to achieve a consensus. If a consensus is not achieved at the level of cancer or intraepithelial lesions, the slide should be excluded from the study. For an adjudicated diagnosis of ASCUS, at least two of three panel members should agree in the consensus phase.
     Generally, in determining sensitivity, it is important to use adjudicated cytology as a gold standard* and not biopsy because:

 3. Positive predictive value should also be considered, enabling an approximation of clinical specificity. This requires use of tissue biopsy (using a consensus diagnoses) of a statistically significant subset of patients with a positive cytologic diagnosis as the gold standard for this aspect of the study.
     4. Using the adjudicated cytology results from 2 (above), or histology from 3 (above), a comparison of the receiver operator curve (ROC) characteristics of the two study arms is recommended.
     5. Bias resulting from differences in cytotechnologist ability in different arms of the study should be avoided. For example, for interactive devices, cytotechnologists who perform primary manual screening and cytotechnologists who examine the interactive computer screens should possess comparable experience and abilities.
     6. The study arm evaluating the device should replicate intended use conditions in order to properly assess cytotechnologist and pathologist performance with respect to sensitivity and specificity.
For illustrative examples of the effects of how machine error can compound human error, see Tables IÐIV.
     7. The trial design should allow for separate analyses of sensitivity, specificity and positive predictive value for three categories of cytologic diagnosis: ASCUS/AGUS, LSIL, and HSIL plus cancer. By separating ASCUS/AGUS into a class by itself, the increased inter-observer variation in ASCUS/ AGUS can be uncoupled from the more consistent performance of humans in classifying intraepithelial lesions and cancer.

Working Group Members
Members of the 1996-1997 Intersociety Working Group for Cytology Technologies included Carlos Bedrossian, M.D. (PSC representative), Thomas Bonfiglio, M.D. (ASCP representative), Denis Coble, EdD, CT (ASCT representative), Diane Davey, M.D. (Co-Chair, ASC representative), Martha Hutchinson, M.D. (ASC representative), Edward Kaufman, M.D. (ad hoc), Paul Krieger, M.D. (CAP representative), Dina Mody, M.D. (CAP representative), Stephen Raab, M.D. (Co-Chair, PSC representative), Ibrahim Ramzy, M.D. (IAC representative), Dorothy Rosenthal, M.D. (IAC representative), Patricia Saigo, M.D. (ASCP representative), Janet Schumann, CT (ad hoc), Diane Solomon, M.D. (ad hoc), Theresa Somrak, J.D., CT (ASCP staff), and Sue Zaleski, SCT (ASCT representative).

Reference

  1. U.S. Dept. of HHSÑPublic Health Service. Improving the Quality of Clinician Pap Smear Technique and Management, Client Pap Smear Education, and the Evaluation of Pap Smear Laboratory Testing. Sept. 1989, pp A3-A7.

Appendix I

Information for Pathologists and Cytotechnologists on New Cytology Technologies
Developed by the Intersociety Working Group for Cytology Technologies

About 15,700 new cases of invasive cervical cancer are diagnosed annually in the United States, and almost 5,000 American women die. About half of the women diagnosed with cervical cancer have never had a Pap screening exam and another 10% have not had one for at least five years. Still others have had suboptimal frequency of screening. Therefore, the most important way to decrease cervical cancer morbidity and mortality is to encourage cervical/ vaginal cytology at regular intervals.
     False negative Pap smears in patients with cervical cancer can be divided into sampling and laboratory errors. Many false negative Pap smear results are due to sampling problems, that is, abnormal cells are not present on the smear when rescreened. In some of the cases, abnormal cells may be obtained by the sampling device but are obscured on the smear by blood and/or inflammation. Laboratory false negative results may be a result of screening errors, interpretive errors, or a paucity of well-visualized abnormal cells.
     New cytology technologies available or under development aim to decrease false negative Pap smear results by addressing one or more of the factors responsible. As laboratory errors are the primary ones addressed, and these account for the minority of all cervical cancers, such technologies cannot be expected to have a major impact on cancer rates without concurrent emphasis on regular screening of all women at risk.
     Factors to consider in evaluation of new technologies include sensitivity, specificity, reproducibility of diagnoses, how the technology affects the cytologist performance in daily laboratory practice (vigilance), economics and workload issues, specimen adequacy, and types of specimens appropriate for the new device. False negative, false positive, and positive predictive value calculations are useful. It is recommended that new technologies used in an adjunctive mode be documented on the cytology report. Other automated technologies may be documented in laboratory procedures or records.

A. Specimen Quality and Preparation
Monolayer or thin layer technologies aim to improve specimen adequacy by decreasing obscuring factors and improving the visualization of any abnormal cells present in the sample. These devices may have an impact on multiple causes of the false negative Pap smear, including sampling, screening, and interpretation. The sampling (collection) device is placed in a vial of fixative instead of being smeared onto a slide, and preparations from the liquid specimen are prepared within the laboratory. The Cytyc ThinPrep 2000 System passes the cell suspension through a polycarbonate filter until a specified density of cells has been deposited, and then touches the filter to a slide. The AutoCyte
CytoRich uses a density gradient technique to separate the cells of interest, and then allows cells to settle onto a slide.
     Only Cytyc CorporationÕs ThinPrep 2000 has received FDA approval at this date. Either a ÒbroomÓ type sample collection device or a plastic spatula/endocervical brush combination must be utilized. Implementation requires a training course for those individuals who will be interpreting the slides and laboratory validation procedures. Supplies used in the ThinPrep 2000 System must be those designed and supplied by Cytyc. Only one specimen is processed at a time. A multisample device is being designed and tested.
     The FDA trial for the ThinPrep 2000 showed that this system can be used as a replacement for conventional Pap smear preparation. The FDA also recently approved new labeling for ThinPrep to make the following claims: 1) ThinPrep is significantly more effective than the conventional Pap smear for the detection of LSIL and more severe lesions, and 2) Specimen quality with the ThinPrep is significantly improved over that of the conventional Pap smear preparation. The tables following the references, which were taken from the labeling information, compare the ThinPrep to the conventional smear.
     Costs involved: ThinPrep 2000 device, materials including fixative, filter, slides, and slide preparation time. Telephone 1-888-THINPREP or 1-800-442-9892 for more information.
     The AutoCyte CorporationÕs CytoRich device (formerly Roche) is nearing completion of its clinical trial and should be submitted for FDA review for use on gynecologic specimens in the near future (nongynecologic cytology use does not need FDA approval). This instrument prepares and stains a batch of 48 slides in one run. For further information call 1-800-426-2176.
     Finally, the thin layer preparations have not yet received FDA approval for automated rescreening devices such as PAPNET and AutoPap 300. It might be anticipated that monolayer preparations would facilitate computerized image analysis, but additional studies and possible software modifications are required prior to FDA submission and approval.

B. Computer-Assisted Pap Smear Screening Devices
Two automated devices have been approved by the FDA for quality control and/or adjunctive (supplemental) rescreening of smears. Both devices aim primarily to decrease screening false negative errors, but there may be some impact on interpretive problems as well.
     1.PAPNET (Neuromedical Systems Incorporated [NSI]) is an interactive system approved for conventionally prepared Pap smears judged negative by initial manual screening. The smear is shipped to an NSI facility by the clinical laboratory, where it is scanned on an automated computerized microscope which utilizes algorithms and neural networks, and 128 digitized images of the cells/cell groups scored by the software as most likely to be abnormal are saved on a digital tape. The slide and tape are returned to the original laboratory which has been equipped with a computer station video monitor, where the images are viewed. This is done by cytotechnologists and/or pathologists employed by the original laboratory trained in this technology. The cases which are considered negative upon viewing the tape are signed out, while potential epithelial cell abnormalities are reviewed and signed out manually. Recent/future adaptations may allow the microscope which is connected to the computer to go to the exact location where the cells are located (electronic dotting).
     The PAPNET FDA study was a retrospective analysis of previous negative smears in patients with biopsy-proven HSIL or cancer, and 32% of
patients were found to have abnormalities on
PAPNET-assisted review. Of the control negative slides reviewed by PAPNET, 4.8% were judged abnormal, and 69% of these revised diagnoses were ASCUS. A new marketing claim recently approved by the FDA involves a comparison between manual rescreening and PAPNET rescreening. This was a nonparallel historical (retrospective) study at three sites. Manual rescreen detected errors in 0.6% of 13,761 rescreened cases (years 1985-1992), while PAPNET detected errors in 6.2% of 2293 specimens (years 1975-1991). The majority of detected errors were ASCUS/AGUS by both methods. NSI claims that PAPNET-assisted review detects 7.1´ more false negatives when compared with manual rescreening.
     Costs involved: Tape viewing computerized station and attachments for microscope, per slide payment to Neuromedical Systems for off-site screening, additional time for packaging slides as well as time spent by technologist/pathologist viewing tape and manual rescreening if indicated. Call 1-800-368-3630 or 1-800-PAPNET-4 for more information.
     2. The AutoPap 300 QC System (NeoPath Incorporated) is approved for rescreening negative conventional smears, and if performed on all negative smears, can replace a laboratoryÕs quality control random rescreening. It is currently a noninteractive system which uses algorithm classifiers to provide a slide evaluation score of 0 to 1.0 on each slide. Slides receiving scores above a predetermined threshold are to be manually rescreened. The technology aims to segregate the abnormal cases above the threshold. The FDA study demonstrated an approximate 5 times improved sensitivity for significant abnormalities compared to those which would be detected in a 10% random rescreen. Also, 77% of biopsy-confirmed HSIL and cancer cases fell above the threshold in two institutions. Staining and coverslipping of slides need to be optimized for the instrument to score the slides. Future enhancements may include electronic dotting.
     Costs involved: Lease/purchase of instrument or per slide charge, time spent to load machine and manually rescreen slides, time spent to optimize stain/preparation. Call 1-800-NEOPATH for more information.
     3. Cytology support/assistance devices: These automated devices aid the cytotechnologist/ pathologist in screening the entire slide and automatically recording information. They include a microscope and computer, and some include a complete workstation with the following: automated stage, specimen identification, slide loading, automatic dotting, data management. Two companies marketing such devices are AccuMed International (AcCell Series 2000, phone 1-800-650-2228), and CompuCyte Corporation (Pathfinder System, phone 1-800-840-1303).

References

  1. Acta Cytologica 1996;40:1Ð142 (Entire issue devoted to new technologies).
  2. Acta Cytologica 1997;41:1Ð223 (Most of issue devoted to these technologies).
  3. Birdsong GG: Automated screening of cervical cytology specimens. Hum Pathol 1996;27:468Ð481
  4. Check WA: Pap devices seek to star in clinical galaxy. CAP today 1995 (Sept)
  5. Gupta PK: American Society of Cytopathology (ASC) Statement on Technical Devices for Innovations in Cervical Cytology Screening. Acta Cytol 1996;40:604

Appendix II

Evaluation of Automated Systems
     

Number of abnormal cases missed 

FN=

Total abnormal cases 

Number of negatives called abnormal 

FP=

Total negative cases 

Broken down by severity of lesion
What is the gold standard? Expert review, consensus, rescreening, biopsy, HPV

Evaluation of Automated Systems

ASC Position Statement

ASC Position Statement

ThinPrep 3 Category Analysis
 

Conventional smear  


ThinPrep

Neg  

Atypical 

LSIL+ 

Total  


Neg 
Atypical 
LSIL+ 
Total 

5224 

331 
125 
5680 

298 

132 
 99 
529 

71 

 54 
413 
538 

5593 

 517 
637 
6747 

Atypical=ASCUS/AGUS 
 

ThinPrep Specimen Adequacy
 

Conventional smear  


ThinPrep

SAT  

SBLB  

Unsat 

Total  


SAT 
SBLB 
Unsat 
Total 

4316 

722 
63 
5101 

1302 

665 
41 
2008 

38 

44 
32 
114 

5656 

1431 
136 
7223 

 

ThinPrep Clinical Trial

PAPNET study (Neuromedical)

PAPNET QC Claim

PAPNET (Neuromedical)

AutoPap 300 (Neopath)

AutoPap Study (Neopath)

Appendix III

Intersociety Working Group for Cytology Technologies
 
Name  Organization  Address  Telephone  FAX  E-mail Address

Diane Davey, M.D. Co-Chair American Society of Cytopathology  Univ. of Kentucky 
800 Rose Street MS117
Lexington, KY 40536-0084
606-257-5357 606-323-2094 ddavey@pop.uky.edu
Martha Hutchinson, M.D. American Society of Cytopathology Dept of Path/Lab Med
Women & Infants Hosp.
101 Dudley St.,
Providence, RI 02906
401-453-7939 401-453-7689  75321.3156@compuserve.com
Paul Krieger, M.D.  College of American
Pathologists
Corning Clin. Labs.
1 Malcolm Ave.
Teterboro, NJ 07608
201-393-5436 201-393-6127
Dina Mody, M.D.  College of American
Pathologists
Baylor College of Med.
One Baylor Plaza 
Houston, TX 77030
713-790-5903 713-793-1473 dinam@path.bcm.tmc.edu
Carlos Bedrossian, M.D.  Papanicolaou Society of
Cytopathology
Detroit Med Ctr.
4707 St. Antoine Blvd.
Detroit, MI 48201
313-745-0834 313-993-8894  Bedrosian2@aol.com
Stephen S. Raab, M.D.
Co-Chair
Papanicolaou Society of 
Cytopathology
Univ. of Iowa
Hospitals & Clinics
Iowa City, IA 52242
319-356-4164 319-356-8470 Stephen-Raab@UIOWA.edu
Thomas Bonfiglio, M.D.  American Society of
Clinical Pahtologists
Univ. of Rochester
Medical Center
Rochester, NY 14642
716-275-3184 716-273-1027 tbonfiglio@pathology.rochester.edu
Patricia Saigo, M.D. American Society of 
Clinical Pathologists
Memorial Sloan-
Kettering Cancer Ctr.
1275 York Ave.
New York, NY 10021
212-639-5902  212-639-6318  saigop@mskcc.org
Theresa Somrak, JD, 
CT(ASCP)
American Society of
Clinical Pathologists
(Staff)
Cytopath. Education
Consortium, 2100 W.
Harrison Street
Chicago, IL 60612
312-738-4851 312-738-9798 theresas@ascp.org
Sue Zaleski
SCT(ASCP)
American Society
for Cytotechnology
Univ. of Iowa
Hospitals and Clinics 
220 Hawkins Dr. 5222
RCP
Iowa City, IA 52242
319-356-3976 319-356-8470 sue-zaleski@uiowa.edu
Janet Schumann, CT
(ASCP)
American Society 
for Cytotechnology
(Alternate)
4101 Lake Boone Trail,
Suite 201
Raleigh, NC 27607
919-787-5181
203-452-7811
203-459-2532
919-787-4916
203-459-2532
Denis A. Coble, EdD,
CT (ASCP)
American Society
for Cytotechnology
Univ. of Conn. 
Allied Health U-101
  358 Mansfield Road
  Storrs, CT 06269-2101
860-486-0014 860-486-4191
Dorothy Rosenthal, M.D. International Academy
of Cytology
Johns Hopkins Hosp.
600 N. Wolfe St.
Baltimore, MD 21287
410-955-1180 410-614-9556  drosenth@pathlan.path.jhu.edu
Ibrahim Ramzy, M.D. International Academy
of Cytology
Baylor College of
Medicine
One Baylor Plaza
Houston, TX 77030
713-790-4508 713-793-1473  iramzy@bcm.tmc.edu
Kenneth Noller, M.D.  ACOG Univ. of Massachusetts
Medical Center
119 Belmont Street,
Worcester, MA 01605-2903
508-793-6266 508-793-6063
Edward Kaufman,
M.D. 
Ad Hoc SmithKline Beecham
1201 South Collegeville 
Road, Collegeville,
PA 19426
610-454-6182 610-983-2010 edward.kaufman@sb.com
Lee Hilborne, M.D. Ad Hoc 11116 Montana Ave.   
Los Angeles, CA 90049
310-825-5656 310-794-9218 Lhilborn@UCLA.Edu
LEEH@Rand.Org
Diane Solomon, M.D. Ad Hoc NCI
EPN Room 233 301-402-6211 
6130 Executive Blvd.
Rockville, MD 20852
301-496-6355 301-480-9939 dianesol@box-d.nih.gov

    

Editorial Comment
The Committee on Cytology Automation of the International Academy of Cytology published guidelines for designers of automated systems1 that are worthy of repetition upon this occasion. The standards are divided into two major groups: (I) Mandatory Conditions and (II) Highly Desirable Features:

I. Mandatory Conditions (Conditio sine qua non)
     I.1. The system shall not be designed or constructed so as to pass as negative any sample that contains malignant tumor cells;
     I.2. The system shall not flag more Òfalse alarmsÓ on normal cells than could be readily handled by  visual manual review;
     I.3. The system shall not use up the entire sample or render the sample unusable for classic microscopic review; the pathologist must be able to examine the sample after routine staining;
     I.4. The system shall yield reproducible results on repeated scannings of the same sample (within appropriate confidence limits);
     I.5. The system shall have an internal calibration standard for quality control; and
     I.6. The system shall identify the inadequate (or empty) slide.

II. Highly Desirable Features (Required for Clinical Application)
     II.1. The system should be able to clearly demonstrate the item that led to an ÒalarmÓ for subsequent review by a human observer;
     II.2. The system should include dysplastic cells in the alarm group and should not restrict itself to the identification of frankly malignant tumor cells;
     II.3. The system should detect and identify contamination and artifacts as such to avoid unnecessary human review;
     II.4. The preprocessing of the sample required by a given system should be convenient and inexpensive for the clinician and laboratory (worst case: a system requiring specific preprocessing that is more expensive than the entire routine classic workup and evaluation while not offering improved diagnostic quality);
     II.5. Infectious organisms, such as trichomonads and fungi, as well as Òfootprints,Ó as in herpes
inclusions and koilocytosis, should be identified; and
     II.6. The system should operate cost-effectively.

Reference

  1. International Academy of Cytology: Specifications for automated cytodiagnostic systems proposed by the IAC. Acta Cytol 1984;28:582