Assisted Primary Screening Using the Automated ThinPrep Imaging System

Anatomic Pathology / THINPREP AUTOMATED CERVICAL SCREENING Assisted Primary Screening Using the Automated ThinPrep Imaging System Charles V. Biscotti, MD, 1 Andrea E. Dawson, MD, 1 Bruce Dziura, MD, 2 Luis Galup, MD, 3 Teresa Darragh, MD, 4* Amir Rahemtulla, MD, 5* and Lisa Wills-Frank, MD 6 Key Words: ThinPrep Pap Test; Cervical screening; Automated cytology DOI: 10.1309/AGB1MJ9H5N43MEGX Abstract We report the clinical trial studies for the ThinPrep Imaging System (TIS; Cytyc, Boxborough, MA). Between December 2000 and July 2001, 10,742 ThinPrep specimens were collected at 4 US clinical sites representative of the normal clinical population of the laboratories, including screening patients and referred patients. After nonstudy screening diagnoses were completed, the vials were relabeled and randomized, and study slides were prepared and stained. TIS-trained cytotechnologists and pathologists screened the slides twice, first manually, then TISassisted after an appropriate interval. Afterward, 3 independent pathologists performed an adjudication study to determine definitive diagnoses for the nonnegative slides and 5% of the negative slides; the adjudicated diagnoses served as the gold standard for subsequent sensitivity and specificity analyses. TISassisted screening was statistically more sensitive than manual screening for atypical squamous cells of undetermined significance (ASCUS) or higher (+) and statistically equivalent for low- (LSIL)+ and high-grade squamous intraepithelial lesion (HSIL)+ diagnoses. TIS-assisted screening had equivalent specificity for ASCUS+ and LSIL+ and significantly higher specificity for HSIL+. Average cytologists daily screening rates doubled with TIS-assisted screening. The sensitivity of the TIS-assisted screening system equals or exceeds the sensitivity of manual primary screening without adversely affecting specificity, and TIS-assisted screening can improve cervical cancer screening productivity. Cost issues require further study. The low prevalence of cervical disease in regularly screened populations makes the task of the screening cytotechnologist difficult. With prevalence rates for squamous abnormalities as low as 5% to 7% in the United States, the vast majority of Papanicolaou (Pap) tests that a cytotechnologist screens are negative. Encountering slide fields with atypical cells is uncommon. 1 High vigilance is difficult to maintain under these circumstances, and fatigue can contribute to screening errors. 2 Meanwhile, a declining number of cytotechnologists makes it important to efficiently use this valuable, skilled workforce. Automation can aid the cytotechnologist in screening more accurately and productively. 2 It has the potential to increase job satisfaction and performance in this important area of preventative health care. 3,4 The ThinPrep Imaging System (Cytyc, Boxborough, MA) is an automated system that assists cytotechnologists with primary screening of ThinPrep Pap Test (Cytyc) slides. The automated system has 2 component devices. The Imager Processor performs analysis of batches of up to 250 ThinPrep slides with specialized image analysis software. For each slide, the locations of 22 microscopic fields that contain cells or cell clusters of interest are recorded. At the Review Scope (an automated microscope), the cytotechnologist reviews the 22 fields saved from each slide. If the 22 fields contain only normal cells, the entire slide can be interpreted as normal. If any abnormal cells are seen, the cytotechnologist screens the entire ThinPrep slide on the automated microscope in an organized, computer-controlled pattern. The Review Scope is equipped to mark cells on the slide for subsequent pathologist review. The ThinPrep Imaging System (Imager) underwent clinical trials between December 2000 and November 2001. The clinical trial was designed to demonstrate that Imager-assisted Am J Clin Pathol 2005;123:281-287 281 281 DOI: 10.1309/AGB1MJ9H5N43MEGX 281

Biscotti et al / THINPREP AUTOMATED CERVICAL SCREENING cytotechnologist review of ThinPrep slides was at least equivalent to current practice (manual screening) for all descriptive diagnostic and specimen adequacy categories of the Bethesda System. Productivity was evaluated by recording the number of slides screened per day and the time expended. The Imager received approval from the US Food and Drug Administration in June 2003. We report the methods and results of the clinical trial. Materials and Methods Cytologic diagnoses obtained by Imager-assisted screening (Imager arm) were compared with standard laboratory screening (manual arm) of the same slides. Four laboratories in the United States participated in the study: Cleveland Clinic Foundation, Cleveland, OH; Baystate Medical Center, Springfield, MA; Quest Diagnostics, Cambridge, MA; and South Bend Medical Foundation, South Bend, IN. The study population was representative of the normal clinical population of the laboratories, including women undergoing routine screening and women who had been referred for a Pap test following a recent cervical abnormality. Between December 2000 and July 2001, 10,742 specimens were collected for the study from the routine clinical volume at the participants laboratories, using residual patient vials after routine clinical diagnoses were completed. At all 4 centers, ThinPrep slides were prepared by using a ThinPrep 2000 Processor or a ThinPrep 3000 Processor (Cytyc) and were stained with ThinPrep Stain (Cytyc s proprietary Papanicolaou stain). To guarantee that the study would include a sufficient number of high-grade squamous intraepithelial lesions (HSILs) for meaningful statistical analysis, the study set was enriched by a small number of HSILs from the routine screening load. The study pathologists were aware that HSIL cases had been seeded randomly. All specimens were processed within 3 weeks of collection. Study personnel at all 4 clinical sites had appropriate training and experience in processing and evaluating ThinPrep specimens. The cytotechnologists review scope training included a 1-hour introductory lecture followed by 3 days of instruction on the review scope. In preparation for this study, 2 cytotechnologists and 1 or 2 cytopathologists from each laboratory were familiarized with the ThinPrep Stain, and all participants passed a proficiency examination. An operator at each site was trained to process the slides through the Imager. Manual screening of the study slides was conducted first. Slides were categorized as to specimen adequacy and descriptive diagnosis according to the 1991 Bethesda system that was in use during the time frame of the study. 5 After a time lag of at least 48 days, the same study slides were screened with assistance from the Imager. The slides were randomized during each arm of the study, and all screening dots were removed before Imager-assisted screening. To minimize bias, the same cytotechnologist and the same pathologist (if the slide had been referred for pathologist screening) who performed the manual screening also performed the Imagerassisted screening; in this second screening, they were unaware of the manual screening diagnoses. Optical character recognition encoded slides ensured proper identification at each step in the screening process. As described, the Imager identified 22 fields of interest on each slide. If the cytotechnologist found no abnormalities on those 22 fields, he or she could sign out the case as negative. In all cases in which any of the 22 fields contained abnormalities, the cytotechnologist was required to screen the entire glass slide. If they chose to do so, cytotechnologists could opt to screen slides from any negative cases. All slides with herpes simplex virus changes, reactive cell changes, epithelial abnormalities, or malignant cells were referred for pathologist review. Cytotechnologists also could choose to refer negative slides. Slides that could not be processed by the Imager for any reason were excluded from the subsequent data analysis. These slides were screened manually, and the data were analyzed separately. For both the manual and Imager-aided arms, cytotechnologists were advised to screen at a rate with which they were comfortable. No stopwatch monitoring of screening occurred. Following the manual and Imager-assisted slide screening phases of the study, adjudication was performed to establish a truth diagnosis for the slides to calculate the sensitivity and specificity of manual and Imager-assisted screening. During the adjudication, all concordant cases with a diagnosis of atypical squamous cells of undetermined significance (ASCUS) or higher (+; 361 cases), all cases with discordant diagnoses of 1 level or greater (639 cases), and a randomly selected 5% of the concordant negative cases (428 cases) were reviewed by 3 independent pathologists (T.D., A.R., L.W.). Two were from university medical practices, and one was from a private medical center. For a consensus diagnosis, at least 2 of the 3 pathologists diagnoses had to agree. Cases lacking an initial consensus diagnosis were reviewed by the 3 pathologists at a multiheaded microscope. In 6 of the original 361 concordant ASCUS+ cases, the 3 reviewers did not reach a consensus diagnosis; these cases were removed from the adjudicated data set. Specimen adequacy was recorded for the manual and Imager-assisted screening according to the 1991 Bethesda categories. For all concordant unsatisfactory cases, all slides in which the manual and Imager-assisted specimen adequacy classifications did not agree and a random 5% sample of cases classified as satisfactory or satisfactory but limited by, an independent cytotechnologist reviewed the slides to decide a definitive classification. A biostatistician conducted the statistical analyses using the diagnoses from the adjudication phase of the study as the true 282 Am J Clin Pathol 2005;123:281-287 282 DOI: 10.1309/AGB1MJ9H5N43MEGX

Anatomic Pathology / ORIGINAL ARTICLE diagnosis of each slide. The adjudicated data defined true-positive and true-negative cases for the following descriptive diagnostic categories: negative, ASCUS, atypical glandular cells of undetermined significance (AGUS), low-grade squamous intraepithelial lesion (LSIL), HSIL, squamous cell carcinoma, and adenocarcinoma; and for the following specimen adequacy categories: satisfactory for evaluation, satisfactory but limited by, and unsatisfactory. Sensitivity and specificity for the various diagnostic categories, together with their 95% confidence intervals, were calculated for the manual screening and the Imagerassisted screening arms of the study, as were the differences between the sensitivity and specificity. The 95% confidence intervals for the differences in the sensitivity and specificity were based on the normal approximation for correlated data. 6 Results Of the 10,359 slides prepared for the study at the 4 study sites, 732 slides could not be processed. Excessive air bubbles (faulty coverslipping) was by far the most common problem (328 cases), in addition to which there were a variety of machine- or slide-related problems, each involving small numbers of cases. Of the 9,627 remaining slides, 77 slides were determined to be unsatisfactory for evaluation on manual and/or Imager screening. Thus, there were 9,550 cases (92.19%) remaining for analyses. Imager-assisted diagnoses agreed with manual diagnoses in 8,911 (93.31%) of 9,550 cases. Not surprisingly, agreement was relatively poor for the ASCUS category. Imager-assisted screening agreed with 108 (32.0%) of the 337 manual screening ASCUS cases. Imager-assisted diagnoses agreed with 145 (60.9%) of 238 and 105 (69.5%) of 151 manual screening LSIL and HSIL cases, respectively. The imager arm had 43 (10.9%) negative diagnoses among the 396 cases classified in the manual arm as LSIL+, including 29 (12.2%) of 238 LSIL cases and 12 (7.9%) of 151 HSIL cases. By comparison, the manual screening study arm had 31 (8.0%) negative diagnoses among 386 cases classified as LSIL+ in the Imager arm, including 24 (10.0%) of 241 LSIL cases and 6 (4.3%) of 140 HSIL cases. Adjudication of the study allowed the comparison of Imager-assisted and manual diagnoses to a definitive diagnosis for every nonnegative slide. The adjudication process included a 5% sample of the concordant negative slides (n = 428) to assess false-negative rates. When this sample was adjudicated, 3 cases (0.7%) were diagnosed as ASCUS by consensus. As a consequence, a multiple imputation calculation was applied to adjust the numbers of true negative and true positive slides for the adjudicated data analysis. 7 This statistical technique provided a more unbiased estimate of the sensitivity and specificity of the 2 screening methods. The distribution of the original manual and Imager-assisted screening diagnoses were tabulated against truth diagnoses from the adjudication study Table 1. Of the 786 true-negative cases in the adjudication study, Imager-assisted diagnoses were negative in 587 cases (74.7%), and the manual diagnoses were negative in 570 cases (72.5%). The Imager arm had 199 false-positive diagnoses (25.3%; ASCUS+) compared with 216 false-positive manualarm screening diagnoses (27.5%). For the 251 true ASCUS cases, the Imager-assisted diagnosis was ASCUS in 142 cases (56.6%), and the manual diagnosis was ASCUS in 103 cases (41.0%). The Imagerassisted method had 45 false-negative ASCUS cases; the manual method had 83 false-negative ASCUS cases. For both methods, 63 and 64 cases, respectively, of true ASCUS cases had been diagnosed as LSIL+. There were 10 cases adjudicated to be AGUS (not shown in Table 1). These 10 cases were classified by Imager-assisted Table 1 Distributions of Manual and Imager-Assisted Screening Review Diagnoses Compared With Gold Standard Adjudicated Diagnoses * Adjudicated Diagnoses Negative ASCUS LSIL HSIL Unadjudicated Diagnoses Manual Imager Manual Imager Manual Imager Manual Imager Negative 570 (72.5) 587 (74.7) 83 (33.1) 45 (17.9) 21 (8.9) 17 (7.2) 6 (4.3) 2 (1.4) ASCUS 183 (23.3) 173 (22.0) 103 (41.0) 142 (56.6) 40 (16.9) 50 (21.2) 8 (5.8) 9 (6.5) AGUS 7 (0.9) 5 (0.6) 1 (0.4) 1 (0.4) 0 (0.0) 0 (0.0) 1 (0.7) 0 (0.0) LSIL 15 (1.9) 16 (2.0) 48 (19.1) 50 (19.9) 152 (64.4) 155 (65.7) 21 (15.2) 17 (12.3) HSIL 9 (1.1) 5 (0.6) 15 (6.0) 12 (4.8) 23 (9.7) 14 (5.9) 100 (72.5) 108 (78.3) Carcinoma 2 (0.3) 0 (0.0) 1 (0.4) 1 (0.4) 0 (0.0) 0 (0.0) 2 (1.4) 2 (1.4) Total 786 786 251 251 236 236 138 138 AGUS, atypical glandular cells of undetermined significance; ASCUS, atypical squamous cells of undetermined significance; HSIL, high-grade squamous intraepithelial lesion; LSIL, low-grade squamous intraepithelial lesion. * Data are given as number (percentage). Data for adjudicated diagnoses of AGUS and carcinoma not shown; found in text. For information about the Imager, see the text. Am J Clin Pathol 2005;123:281-287 283 283 DOI: 10.1309/AGB1MJ9H5N43MEGX 283

Biscotti et al / THINPREP AUTOMATED CERVICAL SCREENING screening as follows: negative, 4; ASCUS, 1; AGUS, 4; and squamous cell carcinoma, 1. Manual screening classified these 10 cases as follows: negative, 2; ASCUS, 2; AGUS, 3; HSIL, 1; and squamous cell carcinoma, 2. In the adjudication review, there were 236 true LSIL cases. The initial diagnoses were LSIL in 155 Imager-arm cases (65.7%) and 152 (64.4%) in the manual arm. Among the true LSIL slides, there were 67 (28.4%) that had been undercalled in the Imager arm, including 17 (7.2%) that were falsenegative diagnoses; for the manual arm, there were 61 undercalled cases (25.8%) and 21 false-negative LSIL cases (8.9%). For the 138 true HSIL cases, there were 108 (78.3%) that concurred in the Imager arm and 100 (72.5%) that concurred in the manual arm. True HSIL slides were undercalled in 28 Imager-assisted diagnoses (20.3%), including 2 false-negative cases (1.4%), and undercalled in 36 manual diagnoses (26.1%), including 6 false-negative cases (4.3%). In the set of true HSIL cases, there also were 2 cases that had been called HSIL by one method and squamous cell carcinoma by the other and 1 true HSIL case that had been diagnosed as squamous cell carcinoma by both methods. After adjudication, there was only 1 true case of squamous cell carcinoma (not shown in Table 1). This case was interpreted as squamous cell carcinoma on manual-arm screening and as HSIL on the Imager arm. A separate study of the ability of the Imager to detect invasive cancer is reported elsewhere. 8 When the total numbers of true abnormal cases are summed (ASCUS, LSIL, and HSIL cases from Table 1 plus AGUS and squamous cell carcinoma cases not shown in Table 1), there were 636 true abnormal cases. Of these truepositive cases, there were 68 (10.7%) false-negative Imagerassisted diagnoses compared with 112 (17.6%) false-negative diagnoses by manual screening. This represents a statistically significant improvement in the false-negative proportion using imager-assisted screening. Sensitivity and Specificity By using the adjudicated diagnosis as the truth, sensitivity and specificity were calculated with exact 95% confidence intervals for the 2 screening methods in each diagnostic category. Differences in sensitivity and specificity also were calculated and analyzed for statistical significance. Table 2 gives this statistical analysis. The Imager arm was more sensitive for diagnoses of ASCUS+. Imager-assisted and manual screening had equivalent specificity for ASCUS+ diagnoses. Thus, Imager-assisted screening had significantly fewer false-negative diagnoses among the true ASCUS+ cases without increasing the number of false-positive diagnoses. For LSIL+, the sensitivity and specificity rates of Imagerassisted screening and manual screening were statistically equivalent. The sensitivity and specificity data for HSIL+ diagnoses showed that Imager-assisted and manual screening had equivalent sensitivity. Imager-assisted screening had significantly better specificity for HSIL+. Specimen Adequacy and Detection of Infections Bethesda System 1991 categorization of specimen adequacy was used. The comparison of Imager-assisted screening and manual screening slide adequacy from all study sites is shown in Table 3. When the specimen adequacy from the initial screening was reviewed by the independent cytotechnologist, there were 58 slides adjudicated to be true unsatisfactory for evaluation. Manual screening correctly identified 18 of these 58 cases, and Imager screening correctly identified 8 of the 58. Both screening methods showed similar detection rates for infections. Specifically, the Imager arm detected 101 infections: Table 2 Sensitivity and Specificity of Manual and Imager-Assisted Screening of ThinPrep Slides for Bethesda Diagnostic Categories of ASCUS, LSIL, and HSIL * Sensitivity Specificity Manual Imager Difference Manual Imager Difference ASCUS+ 75.6 (72.2-78.8) 82.0 (78.8-84.8) +6.4 (2.6-10.0) 97.6 (97.2-97.9) 97.8 (97.4-98.1) +0.2 ( 0.2 to 0.6) LSIL+ 79.7 (75.3-83.7) 79.2 (74.7-83.2) 0.5 ( 5.0 to 4.0) 99.0 (98.8-99.2) 99.1 (98.9-99.3) +0.09 ( 0.1 to 0.3) HSIL+ 74.1 (66.0-81.2) 79.9 (72.2-86.2) +5.8 ( 1.1 to 12.6) 99.4 (99.2-99.6) 99.6 (99.5-99.7) +0.2 (0.06-0.4) AGUS, atypical glandular cells of undetermined significance; ASCUS, atypical squamous cells of undetermined significance; HSIL, high-grade squamous intraepithelial lesion; LSIL, low-grade squamous intraepithelial lesion; +, or higher. * Results are given as percentage (95% confidence interval). Sensitivity and specificity are based on the gold standard of the adjudicated diagnoses, adjusted by statistical imputation for the 5% sample of negatives in the adjudication review (see text). For information about the Imager, see the text. Sensitivity is the percentage of true diagnoses that were classified as such in the initial, unadjudicated screening. For ASCUS+, sensitivity is the percentage of true cases of ASCUS and higher (ASCUS, AGUS, LSIL, HSIL, squamous cell carcinoma, and adenocarcinoma) that were classified as ASCUS and higher in each arm of the study. For LSIL+, sensitivity is the percentage of true cases of LSIL and higher (LSIL, HSIL, squamous cell carcinoma, and adenocarcinoma) that were classified as LSIL and higher in each arm of the study. For HSIL+, sensitivity is the percentage of true cases of HSIL and higher (HSIL, squamous cell carcinoma, and adenocarcinoma) that were classified as HSIL and higher in each arm of the study. Specificity for ASCUS+ is the percentage of true negative slides that were classified in each arm of the study as negative. Specificity for LSIL+ is the number of slides with a true diagnosis less than LSIL (negative, ASCUS, AGUS) that were classified lower than LSIL in each arm of the study. Specificity for HSIL+ is the number of slides with a true diagnosis less than HSIL (negative, ASCUS, AGUS, LSIL) that were classified lower than HSIL in each arm of the study. 284 Am J Clin Pathol 2005;123:281-287 284 DOI: 10.1309/AGB1MJ9H5N43MEGX

Anatomic Pathology / ORIGINAL ARTICLE 8 Trichomonas vaginalis cases (0.08%), 31 Candida infections (0.32%), 60 cases (0.62%) with a predominance of coccobacilli, and 1 case each of Actinomyces and herpesvirus infection. By comparison, the manual screening arm detected 128 infections: 8 T vaginalis cases (0.08%), 47 cases of Candida infection (0.49%), 71 cases (0.74%) with a predominance of coccobacilli, and 1 case each of Actinomyces and herpesvirus infection. Manual screening showed a higher rate of benign cellular changes (405 cases) than did Imager-assisted screening (293 cases). Cytologist Screening Rates Cytotechnologists daily screening rates were determined by tracking hours spent screening and number of slides screened per day. Daily screening time varied, depending on the availability of the Review Scope and scheduling within the cytotechnologists routine workloads. Table 4 summarizes the average cytotechnologist screening rates, extrapolated to represent an 8-hour workday. Based on the totals from all 4 study sites, Imager-assisted screening typically doubled the average cytotechnologist screening rate. Discussion In the present study, Imager-assisted primary screening compared favorably with manual ThinPrep screening. By using adjudicated cytologic diagnoses as the truth for sensitivity and specificity calculations and combining the data from all 4 study sites, all statistical comparisons demonstrated that Imager-assisted screening had at least equivalent sensitivity and specificity at every threshold of abnormality (ASCUS+, LSIL+, HSIL+). Imager-assisted screening had significantly higher sensitivity (ie, lower false-negative rate) than manual screening for ASCUS+ abnormalities and equivalent sensitivity at the LSIL+ and HSIL+ levels. Imager-assisted screening had equivalent specificity for ASCUS + and LSIL+ and significantly higher specificity for HSIL+. These results show that Table 3 Unadjudicated Marginal Frequency Summary of Specimen Adequacy Results for All Sites Combined in 9,627 Cases * Descriptive Diagnosis Manual Review Imager Review Satisfactory for evaluation 7,375 (76.61) 7,346 (76.31) Satisfactory but limited by 2,186 (22.71) 2,252 (23.39) Endocervical component absent 1,196 (12.42) 1,397 (14.51) Scant squamous epithelial component 92 (0.96) 102 (1.06) Obscuring blood 45 (0.47) 17 (0.18) Obscuring inflammation 69 (0.72) 68 (0.71) No clinical history 982 (10.20) 933 (9.69) Cytolysis 4 (0.04) 2 (0.02) Other 6 (0.06) 33 (0.34) Unsatisfactory for evaluation 66 (0.69) 29 (0.30) Endocervical component absent 6 (0.06) 0 (0.00) Scant squamous epithelial component 35 (0.36) 22 (0.23) Obscuring blood 17 (0.18) 2 (0.02) Obscuring inflammation 8 (0.08) 5 (0.05) No clinical history 2 (0.02) 2 (0.02) Cytolysis 0 (0.00) 0 (0.00) Other 2 (0.02) 0 (0.00) * Data are given as number (percentage). For information about the Imager, see the text. Table 4 Average Cytotechnologist Screening Rates for Manual and Imager-Assisted Screening Extrapolated to an 8-Hour Workday Average Screening Rate Site Cytotechnologist Manual TIS Assisted Percentage Difference 1 1 60 153 255 1 2 78 154 197 2 1 71 110 155 2 2 66 109 165 3 1 82 230 280 3 2 78 178 228 4 1 75 150 200 4 2 63 126 200 All sites combined 72 151 210 TIS, ThinPrep Imaging System (see text for proprietary information). Am J Clin Pathol 2005;123:281-287 285 285 DOI: 10.1309/AGB1MJ9H5N43MEGX 285

Biscotti et al / THINPREP AUTOMATED CERVICAL SCREENING the sensitivity of the Imager-assisted screening system equals or exceeds the sensitivity of manual primary screening without decreasing specificity. Comparison of cervical cancer screening methods must include evaluation of sensitivity for HSIL detection. Left undetected or untreated, most LSIL spontaneously regresses. 9 Undetected HSIL, however, has a much greater likelihood of progression to invasive carcinoma. 10 Given the importance of detecting HSIL, it is an unfortunate paradox of Pap testing that HSIL can be more difficult to recognize than LSIL. 11 LSIL lesions shed larger cells with larger, often more atypical nuclei. By contrast, HSIL lesions can shed small cells with relatively subtle nuclear features. In the present study, Imager-assisted screening identified 79.7% of the adjudicated HSIL cases (110/138) compared with manual screening that identified 73.9% (102/138). Statistically, the 2 methods had equivalent sensitivity for HSIL+ diagnoses. After adjudication of all original nonnegative screening diagnoses in the screening study, there were 6 true false-negative HSIL diagnoses on the manual arm and 2 true false-negative HSIL diagnoses on the Imager arm. The difference was not statistically significant. False-positive diagnoses can have adverse consequences owing to unnecessary follow-up testing. For example, current standards of medical practice recommend repeated Pap tests, human papillomavirus testing, or colposcopy after an ASCUS diagnosis. 12 Follow-up medical testing inconveniences patients and exposes them to risks that should be avoided. Thus, a new, automated primary screening system must not compromise specificity. 13 In the present study, Imager-assisted screening had equivalent specificity compared with manual screening for true-negative diagnoses. Our results demonstrate that Imager-assisted primary screening has equivalent or improved diagnostic accuracy compared with manual screening. These results compare favorably with studies of other automated assisted primary screening systems. 3,14-16 Improved diagnostic accuracy with automated assisted primary screening is not surprising. Manual Pap test screening is labor-intensive. Because of the low prevalence of cervical disease and the scarcity of diagnostic cells even on slides from women with a cervical abnormality, most of a cytologist s screening effort is expended on negative cases and on microscope fields containing no abnormalities. Automated assisted primary screening allows cytotechnologists to focus on key areas of each slide. For negative slides, this facilitates rapid diagnosis and conserves time. When abnormalities are present, automation allows the cytotechnologist to locate diagnostically important cells and clusters immediately, alerting him or her to screen the entire slide. Because of this, we expect that Imager-assisted screening should reduce false-negative results associated with the relatively rare event detection that otherwise occurs with manual screening. 2,3 The cytotechnologists who participated in the study preferred Imager-assisted primary screening, mainly because it reduced fatigue and improved their work environment. The cytotechnologists also cited the ergonomic advantages of keypad control of the motorized stage and the automated dotting as factors contributing to decreased fatigue. Use of the Imager allowed cytotechnologists to approximately double slide throughput, with increases ranging from 55% to 180%. Improved screening productivity also has been reported with location-guided screening using other automated systems. 3 Clinical Laboratory Improvement Amendments of 1988 workload regulations now recognize manufacturers labeling for workload levels. The US Food and Drug Administration and Centers for Medicare and Medicaid Services have approved a daily cytologist screening workload of up to 200 slides per day when using the Imager. 17 Increased laboratory throughput with automated assisted primary screening should help laboratories cope with the current and projected shortage of cytotechnologists. 2-4 Cost and compliance issues will affect the impact of automated assisted primary screening systems. In the United States, it is estimated that 50% of cervical carcinomas affect patients who have never had a Pap test, and another 25% to 33% affect inadequately screened patients. 18 Therefore, efforts to improve regular screening compliance must continue, and the cost issues associated with improvements in technology must be evaluated for their effects on screening compliance. Improvements in technology can positively impact cervical cancer screening and can prove to be clinically effective and cost-effective. 19 Our results suggest that the ThinPrep Imaging System assisted method can improve cervical cancer screening cytology by increasing diagnostic accuracy and improving screening productivity. Cost issues for this new technology require further study. From the 1 Department of Anatomic Pathology, Cleveland Clinic Foundation, Cleveland, OH; 2 Department of Pathology, Baystate Medical Center, New England Pathology Associates, Springfield, MA; 3 South Bend Medical Foundation, South Bend, IN; 4 University of California San Francisco/Mt Zion Medical Center; 5 Union Hospital, Lynn, MA; and 6 Clinical Pathology Associates, Louisville, KY, Supported by Cytyc, Boxborough, MA. Address reprint requests to Dr Biscotti: The Cleveland Clinic Foundation, 9500 Euclid Ave, Cleveland, OH 44195. * Dr Darragh is a consultant for the Cytyc Speakers Bureau, and Dr Rahemtulla has received remuneration from Cytyc as a consultant. Acknowledgments: We acknowledge the essential contributions of the study coordinators, cytotechnologists, and specimen adequacy adjudicator. We also gratefully acknowledge David Zahniser, PhD, for commitment to developing the Imaging System. 286 Am J Clin Pathol 2005;123:281-287 286 DOI: 10.1309/AGB1MJ9H5N43MEGX

Anatomic Pathology / ORIGINAL ARTICLE Study coordinators were Kathy Dzuira, Baystate Medical Center, Springfield, MA; Sandy Dolar, the Cleveland Clinic Foundation; and Debbie Hillsdon-Smith, South Bend Medical Foundation. Cytotechnologists were Jeannette Marchand and Cindy Struthers, Baystate Medical Center; Dawn O Brien and Debbie Sabo, the Cleveland Clinic Foundation; and Ed Lawson and Julie O Keefe, South Bend Medical Foundation. The specimen adequacy adjudicator was Mary Elizabeth Reilly, CT(ASCP), Women and Infant Hospital of Rhode Island, Providence. References 1. Davey DD, Woodhouse S, Styer P, et al. Quality management in gynecologic cytology using interlaboratory comparison. Arch Pathol Lab Med. 2000;124:672-681. 2. Birdsong GG. Automated screening of cervical cytology specimens. Hum Pathol. 1996;27:468-481. 3. Wilbur DC, Parker EM, Foti JA. Location-guided screening of liquid-based cervical cytology specimens: a potential improvement in accuracy and productivity is demonstrated in a preclinical feasibility trial. Am J Clin Pathol. 2002;118:399-407. 4. Allen KA. Cytologist shortage harms patient health. ASCT News. 2002;22:33-35. 5. Kurman RJ, Solomon D. The Bethesda System for Reporting Cervical/Vaginal Cytologic Diagnoses: Definitions, Criteria, and Explanatory Note for Terminology and Specimen Inadequacy. New York, NY: Springer-Verlag; 1994. 6. Fleiss JL. Statistical Methods for Rates and Proportions. New York, NY: John Wiley; 1981. 7. Hawkins DM, Garrett JA, Stephenson B. Some issues of diagnostic tests using an imperfect gold standard. Stat Med. 2001;20:1987-2001. 8. Hutchinson M, Lapen D, Werneke S, et al. Cancer detection with the ThinPrep Imaging System [abstract]. Acta Cytol. 2003;47:858. 9. Nasiell K, Roger V, Nasiell M. Behavior of mild cervical dysplasia during long-term follow-up. Obstet Gynecol. 1986;67:665-669. 10. Stoler MH. Human papillomaviruses and cervical neoplasia: a model for carcinogenesis. Int J Gynecol Pathol. 2000;19:16-28. 11. DeMay RM. Common problems in Papanicolaou smear interpretation. Arch Pathol Lab Med. 1997;121:229-238. 12. Cox JT, Massad LS, Twiggs LB, et al. ASCCP-Sponsored Consensus Conference: 2001 Consensus Guidelines for the Management of Women With Cervical Cytological Abnormalities. JAMA. 2002;287:2120-2129. 13. Intersociety Working Group for Cytology Technologies. Proposed guidelines for primary screening instruments for gynecologic cytology. Am J Clin Pathol. 1998;109:10-15. 14. Doornewaard H, van der Schouw YT, van der Graaf Y, et al. The diagnostic value of computer-assisted primary cervical smear screening: a longitudinal cohort study. Mod Pathol. 1999;12:995-1000. 15. Bishop JW, Kaufman RH, Taylor DA. Multicenter comparison of manual and automated screening of AutoCyte gynecologic applications. Acta Cytol. 1999;43:34-38. 16. Sherman ME, Schiffman M, Herrero R, et al. Performance of a semiautomated Papanicolaou smear screening system: results of a population-based study conducted in Guanacaste, Costa Rica. Cancer. 1998;84:273-280. 17. Summary of Safety and Effectiveness Data [ThinPrep Imaging System]. Available at: http://www.fda.gov/cdrh/pdf2/ P020002b.pdf. Accessed November 30, 2004. 18. Stoler MH. Advances in cervical screening technology. Mod Pathol. 2000;13:275-284. 19. Montz FJ, Farber FL, Bristow RE, et al. Impact of increasing Papanicolaou test sensitivity and compliance: a modeled cost and outcomes study. Obstet Gynecol. 2001;97:781-788. Am J Clin Pathol 2005;123:281-287 287 287 DOI: 10.1309/AGB1MJ9H5N43MEGX 287