Boardman, Lori A. MD; Meinz, Heidi MD; Steinhoff, Margaret M. MD; Heber, Walter W. MSc; Blume, Jeffrey PhD
Many health care providers continue to use the endocervical curette in the evaluation of women with abnormal cervical cytology despite the known characteristics and limitations of the test. For example, in women with mildly abnormal cervical cytology, satisfactory colposcopic examinations, and normal colposcopic findings, the use of endocervical curettage may safely be omitted.1 Furthermore, not only is the test associated with an appreciable level of patient discomfort,2–5 but up to 20% of endocervical specimens obtained with the curette are also inadequate for diagnosis.3,6
Beginning in the late 1980s, these issues led investigators to compare traditional endocervical curettage with the curette to different endocervical sampling techniques, including the cytobrush. In initial studies, the sensitivity of the cytobrush was significantly greater than that of the endocervical curette, whereas the endocervical curette was associated with significantly greater specificity.3,7 By placing a sleeve over the cytobrush and thus protecting the specimen from contamination with ectocervical tissue, Gosewehr was able to demonstrate similarly high sensitivity and significantly increased specificity in comparison with the nonsleeved cytobrush.8 In none of these studies, however, was the sleeved cytobrush sample directly compared with the curette specimen, were both tests evaluated in each patient, nor was the order of sampling varied (the cytobrush consistently preceded the endocervical curette). The goal of this randomized trial was to obtain endocervical curette and brush specimens from each patient undergoing either conization or hysterectomy, with randomization designating the order of endocervical sampling, and to determine specimen adequacy as well as sensitivity and specificity of each sampling technique.
MATERIALS AND METHODS
Patients were enrolled into this randomized, clinical trial at the time of cervical conization or hysterectomy for the treatment of cervical neoplasia or carcinoma or at the time of hysterectomy for benign indications. Institutional Review Board approval was obtained in July 2000; patient enrollment began in August 2000 and was completed in March 2001. Exclusion criteria included the presence of a gross cervical lesion, cervical stenosis prohibiting the evaluation of the endocervical canal by either technique, or endocervical sampling done within 4 weeks of definitive surgery.
After obtaining informed consent, patients were randomized to one of two groups designated on the order of sampling: in the first group, cytobrush sampling was done first, then an endocervical curette followed by surgery; in the second group, the endocervical curette was done first, then the brush followed by surgery. Randomization was generated via a random number table, with results placed in sealed, opaque envelopes opened at the time of study enrollment. All patients received local, regional, or general anesthesia, depending on the type of surgery performed, prior to undergoing endocervical evaluation.
Using a Kevorkian curette, sharp curettage was performed circumferentially from the internal to the external cervical os. The specimen collected in the basket of the curette was transferred to formalin and sent to pathology. For the endocervical brush specimens, a plastic sleeve (a spinal needle cover) was placed over a cytobrush (Medscand AB, Malmo, Sweden); the sleeved cytobrush was applied to the external os, and the brush was passed into the canal, rotated 360° five times, and retracted into the sleeve. The specimen was then placed in a preservative solution (PreservCyt; Cytyc Corp., Boxborough, MA), labeled as a diagnostic specimen, and sent to pathology. Following the endocervical sampling, patients underwent either cervical conization or hysterectomy (including abdominal, vaginal, laparoscopic-assisted vaginal, or radical hysterectomies). All procedures were performed by senior resident physicians under attending supervision or by attending physicians, the latter including both general gynecologists as well as gynecologic oncologists.
A single pathologist (MMS), blinded to study group, reviewed all specimens. Endocervical specimens were assessed for adequacy for diagnosis, the presence or absence of neoplasia, and the degree of neoplasia if present. Based on clinical practice at our institution, the following criteria were used to determine endocervical specimen adequacy: five clusters of endocervical cells with a minimum of ten cells per cluster for the brush specimens and a minimum of one strip of endocervical epithelium with at least ten endocervical cells for the endocervical curette. Brush specimens were interpreted using the Bethesda System and categorized as normal, low grade, high grade, or carcinoma,9 whereas endocervical curette specimens were categorized as normal, cervical intraepithelial neoplasia (CIN) I, II, or III, or carcinoma. Surgical specimens were assessed for the presence or absence of neoplasia in the specimen as a whole as well as in the endocervical canal and for the degree of neoplasia, if such were present.
The authors performed a sample size calculation prior to study initiation. Assuming a type I error rate (α) of .05 and a power of 80% to detect an increase in specificity with the cytobrush from 35% to 75% (based on the work of Andersen3), we calculated a required sample size of 58 women (29 in each arm of the study). With respect to data analysis, demographic features of the study population were compared across groups to ensure no significant differences. For comparisons of categoric data, the Pearson χ2 statistic or Fisher exact test was used, and for continuous data the Student t test was applied. A P value of <.05 was considered significant. Approximate 95% confidence intervals (CIs) for differences in proportions were used to compare adequacy of specimens between the endocervical sampling procedures.
For the purposes of this study, we dichotomized the endocervical and surgical specimen results as either normal or neoplastic (the latter included any degree of dysplasia, CIN, or malignancy). In the initial, unmatched analysis, sensitivity and specificity were computed using standard formulas with cone or hysterectomy results considered indicative of true disease status. Because the correlation between subject is likely positive, this analysis will be conservative (ie, a wider CI will result from ignoring the matching). However, we may account for the natural matching in the study design (as both sampling methods applied to each participant) by using a generalized estimating equations model for binary data with logit link function. To compare the sensitivity (specificity) of the two tests, SAS/GENMOD software (SAS Institute Inc., Cary, NC) was used to model the probability of a neoplastic (non-neoplastic) test outcome as a function of test type (endocervical curette versus brush) using a disease indicator to identify patients who had disease on the final surgical specimen. Because of the sparseness of the specificity data (everyone who was actually negative tested negative), we were only able to obtain comparisons for the sensitivity of the two tests. We report odds ratios (ORs) with approximate 95% CIs. For each regression, an exchangeable (compound symmetric) correlation structure was used to model the correlation of responses within each patient (ie, binary responses for individual patients were assumed equally correlated).10 Analyses were done using SAS 6.12 (SAS Institute, Cary, NC).
Sixty-two eligible women were randomized to one of the two study groups based on sampling order. Of the 31 women in the “brush first” group, two were excluded in the final analysis because of missing “gold standard” information. One patient withdrew following randomization, and the second failed to keep her appointment for conization. Of the 31 patients in the “endocervical curette first” group, one patient underwent endocervical sampling with the cytobrush only secondary to a procedural error. The results of this patient's cytobrush sampling and cone specimen are included. A total of 59 endocervical curettes, 60 brush samples, and 60 surgical specimens were therefore available for analysis.
Demographic characteristics of the study population are included in Table 1. The study groups were comparable with respect to patient age, gravidity, parity, insurance status, marital status, tobacco use, and history of neoplasia or treatment for neoplasia. However, the “brush first” group had a significantly higher proportion of white women (79% versus 39%) than the “endocervical curette first” group (P = .006). We did not adjust for race in our statistical analysis, because the sample size was sufficiently small that such a stratification would be uninformative statistically and, more importantly, we did not believe that race would have an effect on the final outcome of the study.
In terms of surgical indications, 76% of the “brush first” group and 71% of the “endocervical curette first” group had a preoperative biopsy with CIN II or III. There were no statistically significant differences between study groups in terms of indication for procedure or type of procedure performed (Table 2).
Specimen adequacy did differ between the two sampling methods (Table 3). Of the 59 endocervical curettes, 13 (22%) were inadequate. Of the 60 cytobrush specimens, one (2%) was inadequate (95% CI for mean difference between the two procedures: 9%, 31%). With respect to the curette specimens only, the order of sampling (“endocervical curette first” versus “brush first”) did not affect specimen adequacy, as 8 of 30 (27%) samples collected with the curette first were inadequate, compared with 5 of 29 (17%) samples collected with the curette second (95% CI for the mean difference: −12%, 30%).
Calculation of the sensitivity and specificity of the sampling procedures was done with exclusion of inadequate samples (Table 3). In the unmatched analysis, the sensitivities of the sleeved brush (44%) and endocervical curette (32%) were not significantly different (RR 1.38, 95% CI 0.65, 2.94). The endocervical curette was slightly more specific compared with the brush: 100% for the endocervical curette and 88% for the brush (RR 1.13, 95% CI 1.00, 1.28). In the matched analysis, however, the sleeved endocervical brush was a more sensitive sampling method than was the curette (OR 2.04, 95% CI 0.98, 4.22 compared with OR 1.68, 95% CI 0.51, 5.56 for the unmatched analysis). Matching improved the efficiency in the estimation of sensitivity. A matched analysis comparing test specificity could not be performed owing to problems of estimation in the computation of generalized estimating equations parameter estimates secondary to the lack of patients with false-positive endocervical curettes.
The role of endocervical curettage during routine colposcopy remains controversial, in part based on the problems caused by false-negative results (leading to under-treatment) and false-positive findings (resulting in overtreatment).11 In published studies in which endocervical curettage was compared with cytobrush sampling, reported sensitivities of the brush ranged from 77% to 93%, whereas those of the curette ranged from 36% to 64%.2,3,5,7,12–14 Although the cytobrush appeared to result in greater sensitivity, the same could not be said for specificity. Andersen,3 and later Hoffman,7 reported specificities on the order of 26–38%; with the addition of a sleeve over the cytobrush, Gosewehr demonstrated an improvement in specificity to 90% with maintenance of excellent sensitivity. Gosewehr, however, compared the sleeved cytobrush with the nonsleeved cytobrush and not to the endocervical curette.8
Although the American Society for Colposcopy and Cervical Pathology advocates sampling of the endocervical canal with either the curette or sleeved endocervical brush,11 the two procedures have not been directly compared. This trial was designed to evaluate the two techniques in a prospective fashion using the gold standard of histologic evidence of endocervical neoplasia in conization or hysterectomy specimens as the outcome. The strengths of this study include its split-sample design with each patient undergoing both procedures, randomization based on procedure order (to control for the effect of sampling order), and blinding of both the patient and study pathologist to the order of endocervical sampling (to reduce possible detection bias). Our study hypothesis was that, with pathologic specimens available on each patient studied, the sleeved cytobrush, when directly compared with the curette, would perform equivalently, if not better, in the evaluation of the endocervical canal.
Although we found excellent specificity with both modalities, the sensitivity of both the curette and sleeved cytobrush was poor. Although the low sensitivity of the curette has been documented previously, the reported sensitivity of the cytobrush has been shown by multiple authors to be in excess of 77%.3,7,8 One possible explanation for our findings may be that only a fraction of the endocervical canal was sampled, and the areas of disease were missed. If this were the case and the brush specimens were considered adequate, the sensitivity would be decreased, a result of the increase in the false-negative rate. On review of the literature, however, our definition of specimen adequacy for the brush samples was neither vague2,13 nor parsimonious.14,15 Another possible explanation for the low sensitivity of the sleeved brush may lie in the order of sampling. If the first endocervical sampling procedure reduced the amount of available dysplastic tissue, we would expect the sensitivities to increase for both modalities if the sensitivities were calculated based only on first-pass specimens. This was not the case, for the sensitivity calculated using just first-pass specimens was only 50% for the curette as well as the cytobrush.
Based on our findings, women whose endocervical sampling with either the curette or cytobrush reveals an adequate amount of normal endocervical tissue can safely be assumed to have no disease in the cervical canal. Only 2% of the sleeved brush specimens, but over 20% of the curette specimens, were considered inadequate for diagnosis. The major clinical disadvantage to the use of the curette, then, lies in the high rate of inadequate specimens. This rate is consistent with previous reports. For example, nearly 30% of the curette samples obtained in Andersen's study contained either scant or no endocervical tissue, and the false-negative rate of the scant endocervical curette specimens approached 80%.3 Endocervical sampling with the sleeved cytobrush, then, may be the preferred modality for sampling the canal in women for whom this procedure is indicated.
1. Williams DL, Dietrich C, McBroom J. Endocervical curettage when colposcopic examination is satisfactory and normal. Obstet Gynecol 2000;95:801–3.
2. Weitzman GA, Korhonen MO, Reeves KO, Irwin JF, Carter TS, Kaufman RH. Endocervical brush cytology: An alternative to endocervical curettage? J Reprod Med 1988;33:677–83.
3. Andersen W, Frierson H, Barber S, Tabbarah S, Taylor P, Underwood P. Sensitivity and specificity of endocervical curettage and the endocervical brush for the evaluation of the endocervical canal. Am J Obstet Gynecol 1988;159:702–7.
4. Oliveira MM, Farias-Eisner RP, Pitkin RM. Endocervical sampling by Kevorkian curette or Pipelle aspiration device: A randomized comparison. Am J Obstet Gynecol 1995;172:1889–94.
5. Holmstrom SW, Burke JJ, Pare L, Smith D, Gallup DG. A prospective randomized comparison of the endocervical brush and endocervical curette. Obstet Gynecol 2002;99:4S.
6. Martin D, Umpierre SA, Villamarzo G, Sanchez O, Sanchez J, Carrodeguas J, et al. Comparison of the endocervical brush and the endocervical curettage for the evaluation of the endocervical canal. P R Health Sci J 1995;14:195–7.
7. Hoffman MS, Sterghos S, Gordy LW, Gunasekaran S, Cavanagh D. Evaluation of the cervical canal with the endocervical brush. Obstet Gynecol 1993;82:573–7.
8. Gosewehr JA, Julian TM, O'Connell BJ. Improving the cytobrush as an aid in the evaluation of the abnormal Papanicolaou test. Obstet Gynecol 1991;78:440–3.
9. Kurman RJ, Solomon D. The Bethesda system for reporting cervical/vaginal cytologic diagnoses: Definitions, criteria and explanatory notes for terminology and specimen adequacy. New York: Springer-Verlag, 1994.
10. Lesisenring W, Pepe MS, Longton G. A marginal regression modeling framework for evaluating medical diagnostic tests. Stat Med 1997;16:1263–81.
11. Cox JT. ASCCP practice guidelines: Endocervical curettage. J Lower Genital Tract Dis 1997;1:251–6.
12. Klam S, Arseneau J, Mansour N, Franco E, Ferenczy A. Comparison of endocervical curettage and endocervical brushing. Obstet Gynecol 2000;96:90–4.
13. Dunn TS, Stevens-Simon C, Moeller LD, Miekle S. Comparing endocervical curettage and endocervical brush at colposcopy. J Lower Genital Tract Dis 2000;4:76–8.
14. Morgensen ST, Bak M, Dueholm M, Frost L, Knoblauch NO, Praest J, et al. Cytobrush and endocervical curettage in the diagnosis of dysplasia and malignancy of the uterine cervix. Acta Obstet Gynecol Scand 1997;76:69–73.
15. Tate KM, Strickland JL. A randomized controlled trial to evaluate the use of the endocervical brush after endocervical curettage. Obstet Gynecol 1997;90:715–7.