Fax: (011) 31402396109
Liquid-based cervical cytology†
A review of the literature with methods of evidence-based medicine
Article first published online: 22 SEP 2003
Copyright © 2003 American Cancer Society
Volume 99, Issue 5, pages 263–271, 25 October 2003
How to Cite
Klinkhamer, P. J. J. M., Meerding, W. J., Rosier, P. F. W. M. and Hanselaar, A. G. J. M. (2003), Liquid-based cervical cytology. Cancer, 99: 263–271. doi: 10.1002/cncr.11673
Conducted as part of the work of a committee on cervical cytology initiated by the Netherlands Society of Pathology and the Dutch Institute for Health Care Improvement.
- Issue published online: 10 OCT 2003
- Article first published online: 22 SEP 2003
- Manuscript Accepted: 14 MAY 2003
- Manuscript Revised: 23 APR 2003
- Manuscript Received: 26 NOV 2002
- Ministry of Health, Welfare, and Sports of The Netherlands
- cervical cytology;
- cervical cancer screening;
- liquid-based cytology;
- thin layer;
The objective of the current study was to evaluate the applicability of liquid-based cytology in the Netherlands population screening program for cervical cancer.
A special committee performed an evaluation of all the available literature. Two methods were investigated: the AutoCytePrep system (currently known as ShurePath-system; TriPath Imaging, Burlington, NC) and the ThinPrep system (Cytyc, Boxborough, MA) for the detection of squamous epithelial abnormalities. All literature up to May 2000 was evaluated.
For the AutoCytePrep system, there were indications that the detection rate for atypical squamous cells of undetermined significance (ASCUS) or higher had lower sensitivity compared with conventional screening. No definitive statement could be made concerning the value of the AutoCytePrep system for the detection rate of low-grade squamous intraepithelial lesions (LSIL) or higher and high-grade squamous intraepithelial lesions (HSIL) or higher because of conflicting results. For the ThinPrep system, there were indications that the detection rate of ASCUS or higher had a higher detection rate compared with conventional screening, with slightly lower specificity. It is likely that the detection rate of LSIL or higher with the ThinPrep system had greater sensitivity compared with conventional screening with almost unchanged specificity. In addition, it is likely that the detection rate of HSIL or higher with the ThinPrep system had a higher detection rate and greater absolute sensitivity compared with conventional screening with almost unchanged relative and absolute specificity.
Further research that complies with the standards stated in the current study will be necessary to evaluate the applicability of the AutoCytePrep method. Further evaluation of the costs and benefits of the ThinPrep method should be undertaken to decide definitively whether to implement this method in the Netherlands population screening program. Cancer (Cancer Cytopathol) 2003;99;263–71. © 2003 American Cancer Society.
Taking second place only to breast cancer, cervical carcinoma is the most incident malignancy among women worldwide. Like many other industrial countries, the incidence and mortality from cervical cancer decreased considerably in The Netherlands after the introduction of screening with Pap smears. In The Netherlands, the current incidence has been reduced to 50% of the level in the years 1950–1959, whereas mortality has been reduced to < 40%.1 Microscopic evaluation of smears from the uterine cervix is considered an adequate method of detecting preinvasive lesions before symptoms develop.2, 3
High-quality cervical screening can contribute to a reduction in the incidence and mortality of cervical cancer; however, cancer can develop in screened women due to false-negative smears (sampling or diagnostic errors), patient management errors, and interval carcinomas. The adequacy of screening depends on all consecutive steps in the procedure: invitation, patient compliance, collection of the smear, correct handling and staining, microscopic evaluation, and reporting.4–6
Currently, the conventional cervical smear still is used in the Dutch population screening program. The conventional smear procedure has several drawbacks: the manual procedure of applying the cells to the glass slide is impossible to standardize; cells are distributed unevenly over the surface of the slide; often, there are thick areas of overlapping that are difficult to interpret; cells can be obscured by mucous, blood, and/or inflammatory cells; fixation often is insufficient due to air-drying of the smear after it is applied to the glass slide or due to insufficient permeation of the fixative in thick portions of the slide. Liquid-based cytology has been developed to address most of these issues and, thus, reduce the number of false-positive and false-negative smear results. Two systems that apply the liquid-based cytology concept are used widely in the United States and Europe. They are the ThinPrep system (Cytyc, Boxborough, MA) and the AutoCytePrep system (currently known as ShurePath; TriPath Imaging, Burlington, NC). The AutoCytePrep-system is currently known as ShurePath. Both systems were developed and tested in the United States and acquired approval of the U.S. Food and Drug Administration for primary screening.
The current article reports on the work of the liquid-based cytology working group, which consists of epidemiologists and pathologists from the Netherlands Pathology Society. The working group was part of an advisory committee consisting of health care scientists, pathologists, epidemiologists, and molecular biologists that was formed to evaluate new developments in cervical cytology (automated screening, liquid-based cytology, and human papillomavirus screening) and to advise on the possible implementation of these techniques in the Dutch population screening program.
The objective of the current study was to perform a systemic, evidence-based review of all known published articles on thin-layer cytology of the uterine cervix. Two thin-layer methods were assessed by calculating the (relative) sensitivity, specificity, and detection rates of the two thin-layer methods compared with the conventional method.
MATERIALS AND METHODS
Selection of Studies
Using a wide variety of relevant terms, a systematic literature search was performed using Medline starting in 1995 and ending April 30, 2000; the Cochran data base also was used starting in 1995 until Issue 3, in 1999. In addition, the personal files of committee members and reference lists of selected articles and recently published, systematic reviews were consulted.7–10 In total, 122 articles were retrieved.
After screening the articles that had abstracts written in English, appeared in peer-reviewed journals, and reported on studies of the test performance of cervical cytology using liquid-based cytology, 60 articles remained for a more detailed evaluation. Several parameters were used in the further assessment and exclusion of articles, as discussed below.
Context of smear taking
Because the current review concerned population screening, the studies described had to include the investigation of cervical cytology as a primary screening test.
Presence and type of reference test
The presence of an independent reference test with which both methods were compared was considered mandatory to reliably compare liquid-based cytology methods with conventional screening (index tests) and to calculate specificity and sensitivity. There is no universally accepted gold standard to be used as a reference test for cervical cytology, and combinations of cytologic follow-up, histologic follow-up, and panel diagnoses were accepted. The different gold-standard categories encountered were lumped into six broad categories (Table 1). Conventional cytologic diagnosis alone could not be used as a reference test; thus, such studies were excluded.
|1) Conventional cytologic diagnosis (these studies were excluded from further evaluation)|
|2) Combination of histologic follow-up and conventional cytologic follow-up with incomplete data|
|3) Panel diagnosis of the index specimens|
|4) Histologic follow-up or panel diagnosis in case of missing follow-up|
|5) Histologic follow-up of at least HSIL combined with a balanced follow-up diagnosis of all other available follow-up data|
|6) Histologic follow-up of all cases|
Spectrum of disease
Knowledge of the prevalence of disease is important in the interpretation of the accuracy of test results. Therefore, a good description of the study population was considered mandatory (low-risk population, high-risk population, referred patients, etc.).
Reporting of absolute numbers
The results of the (relative) sensitivity and specificity, using predefined thresholds, had to be provided in the study, unless the data were presentable as a 2 × 2 table to enable the calculation of these accuracy measures.
After this selection procedure, 10 studies remained: 4 studies on the AutoCytePrep method and 6 studies on the ThinPrep method.
After they were selected, the 10 studies were reviewed and categorized systematically. Special attention was paid to the items discussed below.
The characteristics of the studies, among others, included the type of sampling device, the length of follow-up, the percentage of index tests in which a reference test was available, the referral criteria, the considered threshold values, and the number of patients.
Independent evaluation of index tests and reference tests
It was considered important for all tests to be evaluated independently. This applied to both index tests, which should have been blinded to one another, and the reference test, which should have been blinded to both index tests.
Study population representative for population screening
Because the starting point of the current study was the evaluation of liquid-based cytology in cervical screening, a study was considered more relevant if the test population was representative for a screening population. We encountered studies in which the test population was a screening population, a high-risk screening population, and a population of patients referred to a gynecologist.
Two basic study designs are possible in these kinds of studies: within-subject studies, in which both methods are applied to the same individual, or between-subject studies, in which both methods are applied to different individuals. The latter design is more powerful, because it makes paired statistical comparisons possible. When both index tests are applied to the same individual (a within-subject study), the studies can be divided further into split-sample studies and residual-sample studies. A between-subject study can be either a cohort study or a prospective study. In a cohort study, special attention should be paid to the fact that both groups are completely comparable, and both groups have equal length of follow-up. Attention also was paid to whether study populations were chosen selectively or nonselectively.
It was recorded whether index tests were evaluated by the same laboratory and whether they were evaluated by the same or different cytotechnicians.
Conflict of interest
Mention was noted of any financial interest by one or more of the authors in the investigated methods.
The key evaluation question was formulated as follows: Compared with conventional cytology, what is the diagnostic value of liquid-based cytology in the detection of cervical abnormalities using the thresholds atypical squamous cells of undetermined significance (ASCUS) and higher, low-grade squamous intraepithelial lesions (LSIL) and higher, and high-grade squamous intraepithelial lesions (HSIL) and higher?
The absolute number of true- and false-positives and true- and false-negatives were extracted or calculated from the presented results and arranged in 2 × 2 tables. From these tables, sensitivity and specificity were calculated. Studies involving follow-up data, such as histology and cytology, rarely specify the follow-up of patients with negative results. Therefore, relative sensitivity and specificity were calculated (Tables 2, 3). In these calculations, the whole cohort was considered.
|Index test||Reference test||Reference test|
|Index test A||Index test B|
|Index test B||Index test A|
|Reference test positive||Reference test negative|
The level of evidence provided by the studies included was assessed using the criteria elaborated in Table 4. The strength of recommendation based on the level of evidence was then formulated as follows: 1) at least two independently conducted Level A studies showed similar results (it has been proven that…); 2) at least two independently conducted Level B studies showed similar results (it is likely that…); and 3) findings were not supported by sufficient Level A or Level B studies (there are clues that…).
|A||Comparison with a reference test in which predefined criteria were established for the index test and the reference test, with a good description of the test and the clinical population, using a large enough number of consecutive patients and using predefined cut-off values, with index tests and reference tests evaluated independently; impact of potential confounding factors should be controlled for, for example, by using logistic regression or group specific comparisons|
|B||Comparison with a reference test with a description of the index test and population, but not meeting the criteria for Level A|
|D||Expert opinion, for example, of committee members|
The characteristics of the 10 selected studies are summarized in Table 5. In all studies on the AutoCytePrep method and in five of six studies on the ThinPrep method, only relative sensitivity could be calculated. In one ThinPrep study, absolute sensitivity and specificity were stated; however, that study was conducted in a clinical population (see Ferenczy et al.16) (Tables 6, 7).
|Study (yr)||Design||Population||No. of women||Level of evidence||Sensitivity||Specificity||Detection rate (%)||Relative sensitivity||Relative specificity|
|Vassilakos et al., 200011||Two different; cohorts, possible selective randomization||Screening||CS, 19,923; AC, 81,120||C||—||—||HSIL +: CS, 0.23; AC, 0.39||HSIL +: 1.17||HSIL +: 1.00|
|Bishop et al., 199812||Split-sample||Screening and clinical patients||8983||B||—||—||LSIL +: CS, 0.81; AC, 0.92;||LSIL +: 1.14||LSIL +: 1.00|
|Bishop, 199713||Split-sample||Family planning,||2032||C||—||—||ASCUS +: CS, 1.62; AC, 1.33;||ASCUS +: 0.82||ASCUS +: 0.97|
|primary care,||LSIL +: CS, 1.33; AC, 1.23;||LSIL +: 0.93||LSIL +: 0.98|
|gynecology practice||HSIL +: CS, 0.39; AC, 0.15||HSIL +: 0.38||HSIL +: 1.00|
|Minge et al., 200014||Split-sample||Gynecology clinic||2156||B||—||—||LSIL +: CS, 1.90; AC, 1.76||LSIL +: 0.927||LSIL +: 0.98|
|Hutchinson et al., 199915||Split-sample||Unscreened population||8636||B||—||—||ASCUS +: CS, 5.22 TP, 6.54;||ASCUS +: 1.25||ASCUS +: 0.95|
|LSIL +: CS, 2.43; TP, 2.98||LSIL +: 1.22||LSIL +: 1.00|
|HSIL +: CS, 1.02; TP, 1.07||HSIL +: 1.05||HSIL +: 1.00|
|Ferenczy et al., 199616||Split-sample||Women referred for colposcopy||364||A2||LSIL +: CS, 70.1; TP, 78.0;||LSIL +: CS, 74.7; TP, 73.6;||—||—||—|
|HSIL +: CS, 82.7; TP, 87.5||HSIL +: CS, 61.6; TP, 57.0||—||—||—|
|Bolick and Hellman, 199817||Two different cohorts, possible selective randomization||Screening||CS, 39,408; TP, 10,694||B||—||—||LSIL +: CS, 0.15; TP, 0.37||LSIL +: 1.12||LSIL +: 1.00|
|Carpenter and Davey, 199918||Two different cohorts, possible||Patient clinic, student||7727||B||—||—||LSIL +: CS, 7.7; TP, 10.5;||LSIL +: 1.36||LSIL +: 0.99|
|selective randomization||health service, and family practice||HSIL +: CS, 2.2; TP, 2.6||HSIL +: 1.18||HSIL +: 1.00|
|Sheets et al., 199519||Split-sample||Women referred for colposcopy||445||B||—||—||LSIL +: CS, 13.7; TP, 15.0||LSIL +: 1.12||LSIL +: 0.99|
|Guidos and Selvaggi, 200020||Two different cohorts, very likely selective randomization||Screening and follow-up smears||CS, 5423; TP, 9583||B||—||—||HSIL +; CS, 0.13; TP, 0.60||HSIL +: 4.62||HSIL +: 1.00|
|Threshold||Vassilakos et al., 200011||Bishop, 199713||Bishop et al., 199812||Minge et al., 200014|
|Rel sens||Rel spec||Rel sens||Rel spec||Rel sens||Rel spec||Rel sens||Rel spec|
|Level of evidence||C||B||C||B|
|Threshold||Hutchinson et al., 199915||Bolick and Hellman, 199817||Carpenter and Davey, 199918||Sheets et al., 199519||Guidos and Selvaggi, 199920||Ferenczy et al., 199616|
|Rel sens||Rel spec||Rel sens||Rel spec||Rel sens||Rel spec||Rel sens||Rel spec||Rel sens||Rel spec||Abs. sens||Abs spec||Abs sens||Abs spec|
|Level of evidence||B||B||B||B||B||A|
Vassilakos et al., 200011
This was a cohort study in which conventional screening and the AutoCytePrep method were compared with available histology and cytology follow-up data. The results for ASCUS or higher could not be calculated from the presented data, because they were not presented as a separate histologic entity. Concerning HSIL or higher, the detection rate for the AutoCytePrep method clearly was better compared with conventional screening (0.39% vs. 0.23%, respectively), with unchanged relative specificity (1.00). No mention was made of whether the reference test (histology) was blinded to the results of the previous cytologic findings. In addition, follow-up of the cohorts seemed to differ by 6 months.
Bishop et al., 199812
This was a split-sample study in which AutoCytePrep smears and conventional smears were compared with available histologic and cytologic follow-up data. For the thresholds of ASCUS and HSIL, the sensitivity could not be calculated from the given results. For LSIL or higher, the detection rate was better for AutoCytePrep screening compared with conventional screening (0.92% vs. 0.81%, respectively), with almost unchanged specificity. Follow-up histologic evaluation probably was not blinded to the cytologic results.
In this split-sample study, AutoCytePrep smears and conventional smears were compared with available histologic and cytologic follow-up data. For the threshold of ASCUS or higher, the detection rate for AutoCytePrep screening appeared lower compared with conventional screening (1.33% vs. 1.62%), with almost unchanged specificity. For HSIL or higher, the detection rate clearly was lower compared with conventional screening (0.15% vs. 0.39%, respectively), again, with almost unchanged specificity. For LSIL, the detection rate for AutoCytePrep screening was 1.23% compared with 1.33% for conventional screening. In this study, it also seems that the reference test (histology) was not blinded to the preceding cytologic results. In addition, the follow-up data for positive AutoCytePrep results seemed less complete compared with the follow-up data for conventional results.
Minge et al., 200014
This was a split-sample study in which AutoCytePrep smears and conventional smears were compared with available histologic and cytologic follow-up data. The results for ASCUS and higher and for HSIL and higher could not be calculated from the presented data. For LSIL or higher, the detection rate for AutoCytePrep screening was lower compared with conventional screening (1.76% vs. 1.90%, respectively). In this study, it also appears that the reference test (histology) was not blinded to the preceding cytologic results. In addition, the follow-up data for positive AutoCytePrep smears and positive conventional smears seemed incomplete.
Hutchinson et al., 199915
In this split-sample study, the results of ThinPrep screening and conventional screening were compared with the follow-up diagnosis, which was an integrated interpretation of all follow-up data. The detection rate for ASCUS and higher was better compared with conventional smears (6.54% vs. 5.22%, respectively) but with lower relative specificity (0.95). For HSIL and higher, the detection rate was 1.07% for ThinPrep smears compared with 1.02% for conventional smears. In addition, for LSIL, the detection rate was better for ThinPrep screening (2.98%) compared with conventional screening (2.43%). However, the reference test was not blinded to the original diagnosis.
Ferenczy et al., 199616
This was a blinded, split-sample study of a clinical population of referred women who all had histologic follow-up directly after cytologic smears were taken. Therefore, true sensitivity and specificity could be calculated. With regard to results of HSIL and higher, the sensitivity for ThinPrep screening was better compared with conventional screening (87.5% vs. 82.7%, respectively); however, specificity was lower for ThinPrep screening (57.0% vs. 61.6%). For LSIL results, greater sensitivity also was reported for ThinPrep screening (78.0% vs. 70.1%), again with slightly lower specificity (73.3% for ThinPrep vs. 74.4% for conventional smears).
Raw data were not reported in this study; therefore, other statistical parameters could not be calculated. The authors seemed to have some industrial involvement. Another problem was that no standard smear device was used.
Bolick and Hellman, 199817
This was a cohort study in which conventional screening and ThinPrep screening were compared with available histology and cytology follow-up data. The necessary calculations for the thresholds of ASCUS or higher and HSIL or higher could not be made from the data given in the report. For LSIL and higher, there was a higher detection rate for ThinPrep screening (0.37%) compared with conventional screening (0.15%), with unchanged specificity. Evaluation of histology did not seem to be blinded to preceding cytology findings. Because the individuals who took the smears were free in their choice of smear method, a bias in the study may exist. In addition, the average age of the women in the ThinPrep arm of the study was younger compared with the average age of women in the conventional arm.
Carpenter and Davey, 199918
This was a cohort study in which conventional screening and ThinPrep screening were compared with available histology and cytology follow-up data. The thresholds investigated were HSIL and higher and LSIL and higher. For HSIL or higher, the detection rates for ThinPrep and conventional screening were almost identical (2.6% vs. 2.2%, respectively), with almost unchanged specificity. For LSIL and higher, the detection rate of ThinPrep smears was 10.5%, compared with 7.7% for conventional smears.
Sheets et al., 199519
This was a split-sample study in which conventional screening and ThinPrep screening were compared with available histology and cytology follow-up data. The thresholds of ASCUS or higher and HSIL or higher could not be calculated from the data presented. With regard to LSIL or higher, the detection rate of ThinPrep screening was higher compared with conventional screening (15.0% vs. 13.7%, respectively), with almost unchanged specificity (0.99). The relative sensitivity was 1.12. It was assumed that no follow-up meant that follow-up was negative for disease. The reliability was reduced by the fact that the index tests were not blinded to one another, and the reference test was not blinded to the index tests.
Guidos and Selvaggi, 199920
This was a cohort study in which conventional screening and ThinPrep screening were compared with available histology and cytology follow-up data. For HSIL or higher, the detection rate was much better for ThinPrep screening (0.60%) compared with conventional screening (0.13%), with slightly lower relative specificity. The thresholds of ASCUS or higher and LSIL or higher could not be calculated from the data presented. It was assumed that no follow-up meant that follow-up was negative for disease.
The reliability of the results appeared to be reduced. It is possible that the 2 study arms were not comparable, which may be deduced from a 4.6 times greater detection rate and lower specificity of the ThinPrep method compared with conventional cytology. ThinPrep screening may have been applied more for high-risk smears, like follow-up smears. In addition, the individuals performing histologic evaluation were not blinded to the type of index test used.
Based in the results of this literature review, the following conclusions were drawn:
Concerning the AutoCytePrep method, there are indications that the detection rate for ASCUS or higher with AutoCytePrep has lower sensitivity compared with conventional screening (one study, Level C). No definitive statement can be made regarding the effect of the AutoCytePrep method on the detection rate of LSIL or higher because of conflicting results (two studies, Level B; one study, Level C). In addition, no definitive statement can be made regarding the effect of the AutoCytePrep method on the detection rate of HSIL or higher because of conflicting results (one study, Level B; one study, Level C). There still is insufficient evidence to show that the AutoCytePrep system can contribute to a higher detection rate of cervical abnormalities. Further research that complies with the standards stated in this study will be necessary to evaluate the accuracy of this method.
Regarding the ThinPrep method, there are indications that the detection rate of ASCUS or higher with ThinPrep is greater than the detection rate with conventional screening, with slightly lower specificity (one study, Level B). It is likely that the detection rate of LSIL or higher with the ThinPrep method has a greater sensitivity compared with conventional screening, with almost unchanged specificity (one study, Level A; four studies, Level B). In addition, it is likely that the detection rate of HSIL or higher with the ThinPrep method is greater, as is the absolute sensitivity, compared with conventional screening, with almost unchanged relative and absolute specificity (one study, Level A; three studies, Level B). The implementation of the ThinPrep system will result in greater sensitivity and almost unchanged specificity.
Many published studies could not be included in the current report because of methodological shortcomings. In 50 of 60 originally selected studies, serious flaws in the study design were detected. Furthermore, in many studies, it proved impossible to summarize the data in a 2 × 2 table to allow calculation of (relative) specificity and specificity or to calculate detection rates.
None of the published studies that used a cohort design allocated study participants randomly to both study arms. Therefore, prevalence rates between the study arms may differ, which influences the detection rates of the two methods.
Furthermore, many studies used different reference tests (gold standards). Some used conventional screening as a reference test, and this was considered unacceptable. Some studies used only a consensus opinion as a reference test. Thus, in such studies, it could not be determined with certainty whether the additional cytologic abnormalities that were detected with liquid-based cytology would be confirmed by histology. Again, other studies used different combinations of panel diagnosis, cytologic follow-up, and histology as reference tests, which may have led to a bias, because only a subset of diagnoses had histologic follow-up. None of the studies included a thorough follow-up of negative smears.
In addition, in most studies, it seems that the histologic verification was not blinded in any of the study arms to the original cytologic diagnosis or to the method by which cytology was obtained. This was true in particular for studies in which routine histologic diagnosis was used for follow-up. Finally, some studies could not be included because essential information necessary to evaluate the methodology was not provided.
Other aspects beyond diagnostic value of the comparison between liquid-based cytology and conventional smears in screening programs were not evaluated systematically in this review. The other aspects include cervical smear quality (the presence of endocervical cells, the number of rejected smears); screening time, and the possibility of applying extradiagnostic techniques on the basis of liquid-based cytology. Issues like increased material costs and excess waste due to containers and liquids were not evaluated. However, these issues play a role in the decision about converting our national screening program from the conventional smear technique to liquid-based cytology.
Two other large studies have been presented on the subject of liquid-based cytology in cervical cytology. In a large American study,8 it was concluded that estimates of the sensitivity of the conventional Pap test were biased in most studies. Those investigators found that, based on the least biased studies, the sensitivity was 51%, which was much lower than is believed generally. Newer technologies may improve sensitivity compared with conventional Pap screening; however, the American authors found that there were no precise estimates for the effect on specificity. Under assumptions that favored improved initial screening technologies and rescreening technologies, they found that either approach could result in an acceptable cost per life-year saved at 3-year Pap screening intervals. However, the imprecision of estimates of effectiveness and costs of the new technologies made drawing firm conclusions about their relative cost-effectiveness problematic in that study.
A study performed in New Zealand9 concluded that the clinical effectiveness of shortened ThinPrep and AutoCytePrep methods for detecting high-grade abnormalities could not be determined reliably from the current evidence. Moreover, those authors found that it was not possible to determine whether one device had advantages over another in terms of considered outcomes. They concluded that valid estimates of test sensitivities and specificity of these devices would have to await further research using better designs. In addition, the introduction of new devices for cervical screening could not be recommended for the New Zealand national cervical cancer screening program at the time of the report.
The current study supports the results presented in both of those review studies. Reviews generally describe yesterday's science. Our decision to include studies that were published up to May 2000 was arbitrary and was due only to practical necessity. We are aware that the scientific evidence on the effectiveness of liquid-based cytology is accumulating and that more recent trials have been and will be published. However, the issue is whether this will change our conclusions. A literature search similar to that used in the current study was performed in September 2002 and found that two trials on liquid-based cytology with histology as a reference test were published after May, 2000: one on the ThinPrep method21 and one on the AutoCytePrep method.22 Both trials included only women who were referred because of abnormal smears. In the study by Park et al., ThinPrep slides were slightly less sensitive compared with the conventional Pap test and slightly more specific. In the trial of Bergeron et al., AutoCytePrep slides were assessed as being more sensitive, but that was primarily due to a large proportion of inadequate conventional smears (11.6%, compared with 0.8% in AutoCytePrep slides), all of which yielded negative results but were assessed as being abnormal on histology. In normal practice, these women would receive repeat smears, and not all would be overlooked.
The current study was not designed to include a specific evaluation of the detection of glandular lesions. Although the incidence of these lesions seems to be increasing, it remains a relatively small portion of premalignant cervical lesions. Some studies indicate an increase in sensitivity for glandular lesions with liquid-based cytology preparations, with unchanged or better specificity.23, 24
The organization of the Dutch cervical population screening program is different from the screening practices in the United States, where most studies originate. It is organized and paid for nationally, with a screening interval of 3 years, combined with measures to discourage tests outside the program. In addition, the screening workload per screener is different from standards in the United States, with an average of 5500 slides screened per screener per year. Many requirements are in place for screeners to participate actively in follow-up and quality control. Slides considered inadequate, on average, include < 1% of all slides. All of these factors may influence the possibility of making further improvements in the sensitivity and specificity of the screening program in The Netherlands. An increase in sensitivity as a result of the introduction of liquid-based cytology, as discussed above, may be balanced with the extra costs and waste of the system. This cost-based argument, of course, is redundant when, in any system, it can be demonstrated that sensitivity and specificity are significantly and relevantly better. On the basis of this review of existing knowledge, the working group believes that it is justified to conduct a specific study evaluating the added diagnostic value of the introduction of liquid-based cytology into the Dutch cervical screening program, including a medical technology assessment.
The authors thank Marc Arbyn, M.D., Ph.D. (Scientific Institute of Public Health, Brussels, Belgium) and professor Dirk Ruiter, M.D., Ph.D. (University Medical Center Nijmegen, Nijmegen, The Netherlands) for their valuable criticism in reviewing this study.
- 1Effects and costs of cervical cancer screening [Ph.D. thesis]. Rotterdam: Erasmus University, 1998..
- 2International Agency for Research on Cancer (IARC) working group on evaluation of cervical cancer screening programmes. Screening for squamous cervical cancer: duration of low risk after negative results of cervical cytology and its implication for screening policies. BMJ. 1986; 293: 659–664.
- 3Anonymous. National Institutes of Health consensus conference on cervical cancer. Bethesda, Maryland, April 1–3, 1996. J Natl Cancer Inst Monogr. 1996; 21: 1–148.
- 4Parameters of quality control in cervical cytodiagnosis. In: WiedGL, KeeblerCM, RosenthalDL, SchenckU, SomrakTM, VooijsGP, editors. Compendium on quality assurance, proficiency testing, and workload limitation in clinical cytology. Chicago: International Academy of Cytology, 1995: 95–108., , , .
- 7Liquid-based cytology in cervical screening. A report by the School of Health And Related Research (ScHARR), the University of Sheffield, for the NCCHTA on behalf of NICE. Sheffield, United Kingdom: Trent Institute for Health Services Research, 2000., , .
- 8Agency for Health Care Policy and Research. Evaluation of cervical cytology. Evidence report/technology assessment no. 5. Rockville: Agency for Health Care Policy and Research, 1999.
- 9Effectiveness and cost effectiveness of automated and semi-automated cervical screening devices: a systematic review. New Zealand health technology assessment report. Volume 3 Christchurch: Clearing House for Health Outcomes and Health Technology Assessment, 2000..
- 10A systematic review of the role of human papillomavirus testing within a cervical screening programme. Health Technol Assess. 1999; 3: 1–196., , , et al.
- 15Utility of liquid-based cytology for cervical carcinoma screening: results of a population-based study conducted in a region of Costa Rica with a high incidence of cervical carcinoma. Cancer. 1999; 87: 48–55., , , et al.
- 19Colposcopically directed biopsies provide a basis for comparing the accuracy of ThinPrep and Papanicolaou smears. J Gynecol Tech. 1995; 1: 27–33., , , , .