Sampling is a cornerstone in cytology and pathology

Transcription

1 Keeping Collecting Device in Liquid Medium Is Mandatory to Ensure Optimized Liquid-Based Cervical Cytologic Sampling Gilbert Bigras, MD, FRCPC, PhD, 1 Malgorzata Anna Rieder, CT, 1 Jean-marc Lambercy, MD, 2 Bernard Kunz, MD, 2 Jean-Paul Chatelain, MD, 2 Olivier Reymond, MD, 2 and Daniel Cornaz, MD 2 1 Laboratoire Cytopath (Unilabs SA), Carouge (Genève), Switzerland; and 2 Groupement des Gynécologues Vaudois, Lausanne, Switzerland Abstract Objective. To determine the impact of not keeping the collecting device in liquid-based cytology. Loss of material was computed from a pair of subsamples from the same cervical scrape. One subsample was obtained by rinsing the collecting device in one vial and the other by discarding it in another vial. Homogeneity of endocervical component was assessed between subsamples. Materials and Methods. Loss of material was analyzed with a two-way analysis of variance whose two factors were G (five gynecologists) and R (number of rinsing rotations in the first vial). Endocervical clusters were counted on slides prepared from all subsamples. Results. Globally, 37% of cellular material is lost when the collecting device is discarded. Loss of material is different among gynecologists. The more intense the rinsing process, the less the loss, but the latter is never zero and is poorly predictable. The discarded subsample often contains a greater amount of endocervical clusters. Conclusions. Discarding collecting device in liquid-based cytology reproduces one of the flaws of the conventional smear technique. Losing cellular material may have an impact on cervical cancer detection, but this still has to be evaluated with further investigations. Reprint requests to: Gilbert Bigras, MD, FRCPC, PhD, Laboratoire Cytopath (Unilabs SA), Av. Cardinal-Mermillod 36, CH-1227, Carouge (Genve), Switzerland. gbigras@unilabs.ch 2003, American Society for Colposcopy and Cervical Pathology Journal of Lower Genital Tract Disease, Volume 7, Number 3, 2003, Key Words: liquid-based cytology, sampling, collecting device Sampling is a cornerstone in cytology and pathology disciplines. This is especially true for cervical cytology [1]. Liquid-based cytology (LBC) has been introduced to improve overall cervical cancer detection by addressing different flaws of the conventional smear method. Among them are the loss of cellular material that is not transferred from the smear-taking device to the slide [2, 3] and the heterogeneity aspect of multiple subsamples created from the same scrape [3, 4]. Heterogeneity is an important issue because diagnostic key elements are not always present in subsamples [3]. Liquidbased cytology avoids the heterogeneity problem by presumably transferring all the cellular material. SurePath (formerly known as CytoRich) is the LBC used in our laboratory for cervical cytology. The protocol, furnished with the SurePath product, requires that gynecologists place just the collecting device in the vial and send the latter to the laboratory. However, some gynecologists regularly discard the collecting device, which sometimes produces unsatisfactory preparations according to the Bethesda System 2001 [5]. This observation has been the substratum of the following hypothesis: rinsing a collecting device and discarding it pro-

2 Optimizing Liquid-Based Cytologic Sampling 169 duces a loss of material. The SurePath LBC technique helped to evaluate this hypothesis because a mucus-free cell pellet, which is routinely produced, can be weighed. The cell pellet weight can then be used to compare different cervical samples. Briefly, the SurePath protocol implies that, in the laboratory, vials are vortexed and the cell suspension is then smoothly transferred in tubes over a density reagent. The cell suspension is centrifuged (2 minutes) at 200 g through the density reagent. Particles, mucus, and low-density cellular elements such as red blood cells slowly migrate toward the bottom of the tube but do not reach it. They are thereafter removed by aspiration. A second centrifugation at 800 g (10 minutes) concentrates the cell pellet at the bottom of the tube. Instead of comparing samples (with and without the collecting device) from different women, a withinsubject strategy was chosen by comparing subsamples obtained from the same cervical scrape. To achieve this, gynecologists were asked to rinse the collecting device in a first vial and to discard it in another one. From the same cervical scrape, one obtains two subsamples, whose cell pellets can be weighted and from which two slides can be prepared and analyzed. This study tries to answer the following questions: To what extent, if any, does discarding collecting device produce loss of material? Does the gynecologist influence the loss of material? Does intense rinsing transfer most material? Are both subsamples homogenous in term of endocervical component? The following protocol described in Materials and Method addresses these questions. MATERIALS AND METHODS Sampling Protocol Five independent gynecologists working in private practice performed 100 cervical scrapes for annual routine screening from 100 consecutive women who came from a general clinic population (20 per gynecologist). Each gynecologist received 40 vials: half were labeled A and the others B. Each vial A was marked with a unique number, either 2, 4, 6, 8, or 10, indicating the number of rinsing rotations to perform against the inner wall of the vial A. For a given woman, a random vial A was drawn blindly from a bag containing them. The tip of the device was thereafter discarded in a second vial B. Therefore, each cervical scrape provided two subsamples, A and B. Each gynecologist provided 20 pairs of vials (four pairs of each following combinations: A2- B, A4-B, A6-B, A8-B, and A10-B). At the end, 200 vials were sent to the laboratory. Because the type of collecting device has an influence on the quality and quantity of the cervical sampling (for instance, cotton-tipped applicator and Ayre s spatula are inefficient for collecting endocervical cells [6 9]), the same collecting device (Cervex-Brush) was used for all 100 patients. Cell suspensions from vials were individually transferred in tubes using PrepMate processor. Before transfer, empty tubes were identified randomly with a number from 1 to 200 and weighted. They were chosen randomly by blindly drawing them from a bag. After transfer, tubes were centrifuged twice, according to the SurePath protocol, to obtain cell pellets. The supernatant liquid was decanted and all droplets of liquid remaining on the inner wall of the tubes were removed with a sterile cotton swab. The tubes were let to dry at room temperature to dehydrate the cell pellet. Cell pellet weight was computed by subtracting the weight of empty tube from the weight of tube containing cell pellet. A percentage of material loss (%ML) was computed using pellet weights from subsamples A and B as (B /(A + B)) 100. Two slides were prepared from subsamples A and B with Prepstain slide processor. Slides were screened by one cytotechnician, and the total number of endocervical clusters per slide were reported. Before counting, slides were identified randomly with a number from 1 to 200. To facilitate counting to obtain a reproducible estimator, only clusters of five cells and more were considered. The number of endocervical clusters in slide B minus the number of endocervical clusters in slide A was computed as GL B A. Even though individual endocervical cells and tiny clusters were not taken in account, GL B A is an homogeneity estimator because both subsamples derive from the same cervical scrape. The greater the absolute difference GL B A, the less the homogeneity between subsamples A and B. Thanks to the randomization process, the operator who weighed the cell pellets and the cytotechnician who counted the clusters had no way to identify matching subsamples A and B, nor could they identify a sample A or B. Statistical and Experimental Design The central variable of this experiment is %ML eventually produced by discarding the collecting device. Many factors may influence %ML, but two were addressed in this experiment: the intensity of the rinsing process (represented by the number of rotations in vial A) and the gynecologist himself. Both factors were

3 170 BIGRAS ET AL. evaluated simultaneously with a two-way analysis of variance. This permits evaluation of both factors with the same sample and to detect any interaction between them. Data (%ML) are therefore cross-tabulated in 25 groups (Table 1) according to the number of rotations performed and to the gynecologist who performed the subsamples A and B. The two-way analysis of variance first factor, that is the number of rotations (R), has five levels (R1=2,R2=4,R3=6,R4=8,and R5 = 10). The two-way analysis of variance second factor, that is the group of five gynecologists (G), has also five levels (G1 to G5) corresponding to each gynecologist. For each level (R1 tor5 and G1 tog5), a %ML mean was computed. These means indicate, if any, the respective influence of both factors R and G on %ML. There are two null hypotheses (H 0 ). The first is related to the factor R and implies that the rinsing process intensity has no influence on loss of cellular material; therefore %ML would be the same for R1 tor5. The second is related to the factor G and implies that for all gynecologists, %ML would be the same. Thanks to the analysis of variance, interaction between factors G and R is also determined. If no interaction is found, it is possible to appreciate individually the influence, if any, of both factors. Before using an analysis of variance, one has to determine two important assumptions related to data: normal distribution of %ML and homogeneity of variance of %ML among the 25 groups (Table 1) of the experimental design. Both assumptions were addressed with the Kolmogorov-Smirnov test for normality and with the Bartlett-Box test for homogeneity [10]. If normality and/or homogeneity cannot be assumed, an arc sine transformation of the %ML variable should be performed. This transformation, when needed, is appropriate for percentages like %ML. GL B A were reported on a scatterplot and compared with %ML. Statistical significance was established when p <.05. RESULTS Most of the women enrolled in the study were of reproductive age, with a mean age of 38 years. Seventythree percent of these women were younger than 45 years. All 200 slides from the 100 women were screened, and seven women had cellular anomalies. Four low-grade squamous intraepithelial lesions were found with one discordance between subsamples A and B (low-grade squamous intraepithelial lesion and normal). Three samples of atypical squamous cell of undetermined significance were found, with two discordances between subsamples A and B (atypical squamous cell of undetermined significance and normal). Table 1. Two-Way Table Rotations (R1 R5) Mean Gynecologists (G1 G5) G G G G G Mean Factor R (2,4,6,8, and 10 rotations) against factor G (5 gynecologists). Percentage of material loss (ML%) is the variable computed as (B/(A + B) 100). A and B are cell pellet weights. The collecting device was rinsed in vial A and discarded in vial B which permitted to create two cell pellets A and B from the same clinical sample. ML% means of G and R factors are, respectively, found in the right column and in the bottom line.

4 Optimizing Liquid-Based Cytologic Sampling 171 Kolmogorov-Smirnov coefficient equals 0.48 with a p >.15. The %ML is then considered originating from a normal population without any need to transform it. The Bartlett-Box F(24,1947) equals 1.12, with p =.303. The %ML is then considered homogeneous regarding variance among the 25 groups (Table 1). The analysis of variance procedure can be confidently used. The global mean of %ML is 37%. Mean of %ML varies from 32.1 ± 4.3% to 51.6 ± 4.5% according to the factor G and from 26.6 ± 3.8% to 47.8 ± 5.2% according to the factor R (Tables 1 and 2). Both factors were found statistically significant by the ANOVA to explain variation of %ML (Table 2). No interaction was found between R and G that permits one to conclude individually their effect on %ML. Figure 1 illustrates the relationship between factor R (x-axis) and %ML (yaxis). Globally, one can see that the more intense the rinsing process, the smaller the %ML. However, a null %ML is never achieved (even with R = 10), and there is a huge variation of %ML around means, which makes the prediction of %ML from R quite poor. In fact, the r 2 of Pearson is only 0.08, which reflects a poor linear relationship between %ML and R. Age did not affect cell transference: no relation was found between age and %ML either globally or for a given R factor. Figure 2 illustrates the relationship between %ML (x-axis) and GL B A (y-axis). When GL B A is more than 0, the number of endocervical clusters found in slide B (from the discarded subsample) is greater than that in slide A (from the rinsed subsample). When GL B A is less than 0, it is the opposite, and when GL B A equals 0, both slides are homogenous. One can see that homogeneity is infrequent. When %ML is less than 50%, the running mean of GL B A (gray line) is negative with however many samples with a positive GL B A value. When %ML is more than 50%, GL B A is most of the time positive, indicating a greater amount of lost endocervical clusters. DISCUSSION The five gynecologists produced different cellular loss. This is not unexpected, because it was demonstrated that performance adequacy, obtained when using conventional smear, is heterogenous among gynecologists [11, 12]. Human factors do interfere when performing a presumably standard technique. Not reaching a near zero loss of material, even with an intense rinsing procedure, is a first unexpected and interesting result. A second interesting result is that of a huge variation of loss of material for a given number of rotations. The third interesting result is the heterogeneity of the endocervical component between subsamples, with a higher number of clusters in the discarded vial especially with high %ML. One may explain these results by the nature itself of the cervical sampling: a mixture of individual cells and epithelial fragments entrapped in mucus. This mucus may stick on the collecting device and may retain epithelial elements despite the rinsing process. The amount of the mucus and its variable stickiness among Table 2. Two-Way ANOVA Results Data by factor R n Mean SD SE R1 (2 rotations) R2 (4 rotations) R3 (6 rotations) R4 (8 rotations) R5 (10 rotations) Data by factor G n Mean SD SE G G G G G Source of variation SS DF MSq F p Factor R * Factor G * Interaction R G Within groups Total Mean, percentage of material loss; SD, standard deviation; SE, standard error; SS, sum of squares; DF, degree of freedom; MSq, mean square; F, Fisher distribution test. Both factors G and R were found statistically significant (p <.05) to explain variation of loss of material (%ML). These factors are independent since no interaction was found between them.

5 172 BIGRAS ET AL. Figure 1. Scatterplot of the number of rotations performed against the inner wall of the vial A (x-axis) against the percentage of material loss (%ML) (y-axis). Each small gray box represents the standard error interval around the mean of %ML. The more intense the rinsing process, the less the cellular loss. However, the loss is never zero and there is a great variation of loss for a given number of rotations performed in vial A. women may explain the huge variation of cellular transference for a given number of rotations. Because endocervical cells are the mucus producers, they may be more intimately admixed with mucus and therefore less easily transferred by the rinsing process. However, when the collecting device is left in the liquid medium, with time, the mucus may dissolve in the liquid medium, which may liberate all cellular elements. LBC testing was created to address aspects of preanalytical errors. For the Papanicolaou technique, most preanalytical errors are poor sampling and poor fixation. The LBC techniques take fixation out of hands of the people who obtain the clinical sample and have shown remarkable improvements in cytologic morphology and sample adequacy. These techniques also improve problems with obscuring blood or microbial contamination. It has been proven that the Papanicolaou technique may lose a substantial amount of material with transference [3]. It has been assumed that LBC techniques retain most or all material. However, this study demonstrates that this assumption is valid as long as the collecting device is kept in the vial. From these findings, one important unanswered question remains: Does an increase in the amount of transferred cells (especially the endocervical cells) translate into a more sensitive test? The discordances found in subsamples A and B from this study are anecdotal. However, one finds in the literature conclusions that may support a positive answer to this question. For instance, it was demonstrated that endocervical cells were found significantly more often in dysplastic smears than in negative smears [13]. In another study, the rate of cervical intraepithelial neoplasia diagnosis was influenced by an adequate sampling of the endocervix [14]. Finally, it has been shown that, for conventional smears, additional slides improve the identification of dysplasia [15]. To test the hypothetical improvement of sensitivity, one should repeat the same experiment, especially by using subsamples from the same cervical scrape, with a selected population known to have a high prevalence of cervical intraepithelial neoplasia. There are other reasons to keep the maximum amount of material and especially the endocervical component. The greater the amount of material, the more tests one can perform from the same sample. Apart cytologic examination, microbiological identifications

6 Optimizing Liquid-Based Cytologic Sampling 173 Figure 2. Scatterplot of the percentage of material loss (x-axis) against GL B A (y-axis). GL B A is the number of endocervical clusters found in slide B (from the discarded subsample) minus the number of endocervical clusters found in slide A (from the rinsed subsample). The gray line represents a running mean of GL B A with a period of 10. Subsamples are not homogenous. When there is an important loss of material (more than 50%), there is a trend preferentially to lose endocervical clusters. with molecular techniques are increasingly performed (i.e., human papillomavirus and chlamydia trachomatis). It has been shown that the sensitivity of chlamydia trachomatis detection test is affected by the presence of endocervical cells: the rate of positivity for chlamydia trachomatis was 10.8% (71 of 655 specimens) among specimens containing endocervical cells, whereas it was only 3.6% (35 of 978 specimens) among specimens lacking endocervical cells (p <.0001) [16]. Interestingly in an older similar study, Kellogg et al. demonstrated an improved polymerase chain reaction detection of Chlamydia trachomatis when the collecting device was not discarded but sent directly to the laboratory. When the collecting device was discarded, the polymerase chain reaction detection was found to be poor, as was the adequacy of the sample in term of endocervical component [17]. In conclusion, using LBC and discarding the collecting device may reproduce one of the flaws of conventional smear technique. Further investigations are needed to measure a genuine negative impact of losing material on cervical cancer detection. In the meantime, it would seem reasonable to optimize the sampling process by simply placing the collecting device in the liquid medium. Acknowledgments The authors thank Mr. Armand Schrumpf for his technical support and Dr. Carole Williamson for manuscript revision. REFERENCES 1. Baandrup U, Bishop JW, Bonfiglio TA, Branca M, Hutchinson ML, Laverty CR, et al. Sampling, sampling errors and specimen preparation. Acta Cytol 2000;44: Al-Awadhi R, Byrne M, Coleman DV. The preparation of additional smears from a cervical scrape: impact on the rate of detection of cervical neoplasia. Cytopathology 2001;12: Goodman A, Hutchinson ML. Cell surplus on sampling devices after routine cervical cytologic smears. A study of residual cell populations. J Reprod Med 1996;41: Hutchinson ML, Isenstein LM, Goodman A, Hurley AA, Douglass KL, Mui KK, et al. Homogeneous sampling accounts for the increased diagnostic accuracy using the Thin- Prep Processor. Am J Clin Pathol 1994;101: Solomon D, Davey D, Kurman R, Moriarty A, O Connor D, Prey M, et al. The 2001 Bethesda System: terminology for reporting results of cervical cytology. JAMA 2002;287: Helderman G, Graham L, Cannon D, Waters K, Feller D. Comparing two sampling techniques for endocervical cell recovery on Pap smears. Nurse Pract 1990;15:30 2.

7 174 BIGRAS ET AL. 7. Lo L, Jordan J. Comparative yield of endocervical and metaplastic cells. Two sampling techniques: wooden spatula and cytology brush. Can Fam Physician 1995;41: Martin-Hirsch P, Jarvis G, Kitchener H, Lilford R. Collection devices for obtaining cervical cytology samples (Cochrane Review). In: The Cochrane Library (ISSN X), Issue 2, 2000 (Document reference: CD001036). Oxford: Update Software. Updated quarterly. 9. Martin-Hirsch P, Lilford R, Jarvis G, Kitchener HC. Efficacy of cervical-smear collection devices: a systematic review and meta-analysis. Lancet 1999;354: Sokal RR, Rohlf FJ. Assumptions of analysis of variance. In: Sokal RR, Rohlf FJ, eds. Biometry: The Principles and Practice of Statistics in Biological Research. 3rd ed. New York: W.H. Freeman; 1995: Celasun B. Presence of endocervical cells and number of slides in cervicovaginal smears: differences in performance between gynecologists. Acta Cytol 2001;45: Kane BR, Berger MS, Lisney M. Pap smear adequacy: the role of clinician experience. Fam Med 1997;29: Szarewski A, Curran G, Edwards R, Cuzick J, Kocjan G, Bounds W. Guillebaud comparison of four cytologic sampling techniques in a large family planning center. Acta Cytol 1993;37: Schettino F, Sideri M, Cangini L, Candiani M, Zannoni E, Maggi R, et al. Endocervical detection of CIN. Cytobrush versus cotton. Eur J Gynaecol Oncol 1993;14: Al-Awadhi R, Byrne M, Coleman DV. The preparation of additional smears from a cervical scrape: impact on the rate of detection of cervical neoplasia. Cytopathology 2001;12: Loeffelholz MJ, Jirsa SJ, Teske RK, Woods JN. Effect of endocervical specimen adequacy on ligase chain reaction detection of Chlamydia trachomatis. J Clin Microbiol 2001; 39: Kellogg JA, Seiple JW, Klinedinst JL, Stroll ES, Cavanaugh SH. Improved PCR detection of Chlamydia trachomatis by using an altered method of specimen transport and highquality endocervical specimens. J Clin Microbiol 1995;33: