: Professor D. V. Coleman, Department of Histopathology and Cytology, Imperial College School of Medicine at Hammersmith Hospital, Du Cane Road, London W12 0NN, UK.
Objective To investigate the reasons why cervical smears occasionally fail to reflect the underlying pathology in the cervix even when the smear is taken at colposcopy.
Design A randomised study of three different smear-taking devices.
Setting A colposcopy clinic.
Population Women attending the colposcopy clinic.
Methods A smear was taken from 172 nulliparous and 100 multiparous women at colposcopy and the procedure was monitored on a video-imaging system. The cytological findings were compared with the biopsy report in 147 nulliparous and 85 multiparous women.
Main outcome measures Accuracy of cytology and the effect of a range of variables on the accuracy of cytology.
Results Sampling of the transformation zone was incomplete in 15% of nulliparous women and 8% of multiparous women. Univariate analysis of a range of variables including parity, type of sampling devices, completeness of sampling of the transformation zone, size of the transformation zone, size of the lesion (aceto-white area) and location of the squamo-columnar junction showed that the accuracy of cytology was influenced by all these factors except for parity and smear-taking devices. Multivariate analysis showed that the location of the squamo-columnar junction, the size of the transformation zone area, the size of the aceto-white area and the ratio of the aceto-white area to the area of the transformation zone influenced the accuracy of cytology.
Conclusions Women with large transformation zone areas (>30.03 mm2) and/or small aceto-white lesions (<7.01 mm2) are more likely to have an inaccurate cytology reports than women with small transformation zone and women with larger aceto-white areas. A ratio of the aceto-white area to the area of the transformation zone of 0.22 or less increases the risk of disagreement between the cytological and histological findings.
Several studies show that even when a cervical smear is taken at colposcopy from women with an area of CIN visible to the colposcopist, abnormal cells may not be present in the smear. This has been attributed to sampling error and has been estimated to occur in 8–12% of cases.1–3
We have investigated the extent to which these findings are due to failure of contact of the smear-taking device with the target area of the cervical epithelium.
Women attending a colposcopy clinic at the Whittington Hospital were recruited into the study consecutively. The women were stratified into two groups— nulliparous or multiparous, and each group was randomly subdivided into blocks of six subjects with two patients randomly assigned to one of three smear-taking devices within each block. This ensured that an approximately equal number of patients in the two groups were allocated to each smear-taking device.
The patients were placed in a modified lithotomy position and the cervix was exposed. A cervical scrape was taken by the colposcopist and a smear was prepared by RMA by wiping both sides of the smear-taking device on to the glass slide. The smear was fixed using a polyethylene glycol dropper fixative. The cervical smears were transported to the cytology laboratory where they were stained, labelled and screened under a light microscope at a magnification of 100 by RMA. The terminology used for reporting the cytological findings was consistent with that recommended by the British Society for Clinical Cytology (BSCC).4,5 Smears were read by RMA without knowledge of the referral smear or colposcopy/histology results. Ten percent (10%) of the smears were checked by a senior cytotechnologist.
After the smear was taken, colposcopy proceeded in the normal way with the application of 3% acetic acid, Schiller's iodine solution and punch biopsy was taken whenever abnormal areas were seen.
The whole colposcopic procedure was recorded using a colour video camera (SONY DXC-930P), video recorder (SONY SVO-9500 MDP), colour video monitor (Trinitron) and colour video printer (SONY CVP-M3E). The video film was inspected after the clinic by RMA and the colposcopist, and two images were selected for photography. The photographs were printed immediately for each case. The first photograph showed the exposed cervix. The area of contact between the spatula and the cervical epithelium within and outside the transformation zone was determined from the video film and mapped on this photograph by RMA (Fig. 1A) as shown in Fig. 1. The tracing was checked by the colposcopist. The second photograph was taken after acetic acid was applied to the cervix and the extent of the transformation zone and the abnormal areas (Fig. 1B) were mapped by the colposcopist. A transparent overlay of the aceto-white areas was prepared from (Fig. 1B) and placed over (Fig. 1A) as shown in (Fig. 1C) to determine the extent of the transformation zone sampling. Case 1619 (Fig. 2A–C) is shown as an example.
By correlating the areas mapped on A and B, we were able to identify those cases: (i) where the transformation zone was completely sampled by the smear-taking device; (ii) where the transformation zone was not completely sampled by the smear-taking device. For the latter group, the ratio of sampled to non-sampled areas was determined in the following way: the area of the cervical mucosa in contact with the smear-taking device (Fig. 2A) was traced on graph paper. The area of the transformation zone (Fig. 2B) was similarly traced. The ratio of the two areas was compared to determine the percentage of the area of the transformation zone in contact with the smear-taking devices. Other parameters recorded were area of the transformation zone, area of the lesion and the location of the squamo-columnar junction.
The accuracy of cytology was determined by comparing the histological and cytological findings in those women who had both a satisfactory biopsy and an adequate smear, using the biopsy findings as the gold standard. It was assumed that if there was broad agreement between the cytological and the histological findings, then the cervical smear had reflected the underlying pathological changes in the cervix. Logistic regression analysis was used to determine the factors affecting the agreement between the cytology and histology reports. Only women with adequate smears and satisfactory biopsy reports were included in the analysis.
As it was not routine to take a cervical smear prior to colposcopy at the Whittington Hospital, a consent form was prepared to explain the aim and purpose of this study. Each patient was asked to sign the consent form after a verbal explanation from the colposcopist.
The outcome measure for the study was whether the cytology and histology reports agreed, which was measured on a binary yes/no scale. Also measured as part of the study were a number of other variables, which could influence whether the reports agreed or not. These explanatory variables took the form of both categorical (e.g. parity, location of squamo-columnar junction) and continuous variables (e.g. the size of the transformation zone, size of aceto-white area).
The effect of each of the explanatory variables upon the outcome was examined. As the outcome was binary in nature, analysis was performed using logistic regression. All variables which showed even slight evidence of a significant effect upon the outcome were considered for the multivariate analysis. A backwards selection procedure was used to retain only those explanatory variables with a significant effect upon the outcome.
Two hundred and seventy-two women were investigated (Table 1). One hundred and seventy-two (63%) were nulliparous (age range 19–58 years, mean 29 years) and one hundred (37%) were multiparous (age range 20–63 years, mean 35 years). Five of the nulliparous women and eight of the multiparous women were postmenopausal.
Table 1. Efficiency of smear taking in study group women.
Age range (mean)
19–58 years (29)
20–63 years (35)
Sampling device (n = 272)
Complete sampling of transformation zone (n = 272)
Cytology results (n = 272)
Biopsy results (n = 246)
Wart virus changes
Severe glandular dysplasia
Squamo-columnar junction on ectocervix or in canal (seen)
Squamo-columnar junction high in endocervical canal (not seen)
Area of transformation zone (range)
Area of transformation zone (mean)
Area of aceto-white (range)
Area of aceto-white (mean)
Ratio area of aceto-white/transformation zone
0–100% (mean 26.9%)
0–94% (mean 23.4%)
Three commonly used smear-taking devices were evaluated— the Ayre spatula, the Aylesbury spatula and the Cervex brush (Table 1).Transformation zone coverage was complete in 146 (85%) of the 172 nulliparous women and incomplete in 26 (15%). The transformation zone was completely sampled in 92 (92%) of the 100 multiparous women and incompletely sampled in 8 (8%). There was a significant difference in the rate of incomplete sampling of the transformation zone between the two groups. (Fisher's exact test; P < 0.0001).
The area of the transformation zone in contact with the smear-taking device in the 34 cases where the sampling was incomplete ranged from 0% to 92%. In 5 of the 34 women, the transformation zone was not sampled at all because the squamo-columnar junction was located high in the endocervical canal.
The Pearson χ2 test indicated that there was a significant difference in performance between the three sampling devices in nulliparous women (P= 0.037). This difference was not observed in the multiparous women (P= 0.443).
The study was designed to collect cervical smears from 30 women for each sampling device. However, as the Cervex brush was introduced after the commencement of the study, the number of women collected for the Ayre and the Aylesbury devices exceeded the target as a consequence of randomisation. In addition, the majority of women attending the clinic were nulliparous, which again affected the number of women in each category.
The cytological findings and biopsy results in the study group patients are shown in Table 1. All the women in the study had a cervical smear taken. However, a biopsy was taken on only 246 of the 272 women (90%).
The squamo-columnar junction was seen on the ectocervix or easily seen in the endocervical canal at colposcopy in 162 (94%) nulliparous women and not seen in 10 (6%) nulliparous women due to the fact that it was high in the canal (Table 1).
The squamo-columnar junction was seen on the ectocervix or easily seen in the endocervical canal at colposcopy in 86 (86%) multiparous women and not seen in 14 (14%).
In nulliparous women the area of the transformation zone ranged from 0 to 58.2 mm2 with a mean area of 27.25 mm2 and the area of aceto-white ranged from 0 to 49.2 mm2 with a mean area of 7.83 mm2. The ratio of the aceto-white area to the area of the transformation zone in nulliparous women ranged from 0 to 100% with a mean of 26.9%(Table 1).
In multiparous women, the area of the transformation zone ranged from 0 to 73.2 mm2 with a mean area of 38.16 mm2 and the area of aceto-white ranged from 0 to 43.07 mm2 with a mean area of 8.85 mm2. The ratio of the aceto-white area to the area of the transformation zone in multiparous women ranged from 0 to 94% with a mean of 23.4%(Table 1).
Twenty-six (26) nulliparous women from the study group were excluded from the analysis due to the fact that 11 had inadequate smear reports, 14 did not have a biopsy taken at the time of colposcopy and 1 woman had an inadequate smear report and no biopsy taken. Fifteen multiparous women from the study group were also excluded from the analysis due to the fact that 3 had inadequate smear reports and 12 did not have a biopsy taken at the time of colposcopy. Thus, cytological and histological correlation was possible in 147 nulliparous women and 85 multiparous women who had satisfactory smear and biopsy reports. The two groups were analysed separately and the results are shown in Table 2.
Table 2. The cytological and histological correlation.
The cytology was considered positive when the smear report showed borderline nuclear changes or worse and the histology was considered positive when the biopsy report showed wart virus changes or worse. Overall cytological and histological findings were in agreement in 80.6% (187/232), with no agreement in 45 cases. The result shows that the cytology and histology were in agreement in 121/147 (82%) nulliparous women and in 66/85 (77.6%) of multiparous women.
Endocervical cells were present in 140 of the 187 smears (75%) where the cytological and histological findings were in agreement and in 38 of the 45 (84%) smears where the cytological and histological findings did not agree. The presence or absence of endocervical cells in the smears had no significant effect on the accuracy of cytology in predicting the underlying pathological lesion (P= 0.238).
A summary of the results of univariate analysis of 147 nulliparous women and 85 multiparous women who had adequate cytology and histology reports are shown in Table 3. The summary statistics shows the responses in each outcome group. For the categorical explanatory variables (parity, completeness of transformation zone sampling and squamo-columnar junction location), the values reported were the number and percentage of group taking each response. For the continuous explanatory variables (the area of the transformation zone, area of aceto-white, and ratio of aceto-white area to the area of the transformation zone), the values used were the mean (and standard deviation) or median (and interquartile range), depending on the distribution of the variable, for each outcome group. The odds ratios indicating the size of effect of each explanatory variable upon the outcome were also reported in Table 3. The categorical explanatory variables represented the odds of agreement for each category relative to a baseline category, while the continuous explanatory variables represented the odds of agreement for a 10-unit increase in the variable (10 units were used as 1 unit was only a very small amount).
Table 3. Univariate analyses of the effect of the explanatory variables on the accuracy of the cytology report in 147 nulliparous and 85 multiparous women.
Cytology and histology reports not in agreement
Cytology and histology reports in agreement
Odds ratio (95% CI)
Mean (standard deviation) reported as variable found to be normally distributed.
Median (interquartile range) reported as variable found to be non-normally distributed.
Odds ratios are for a 10-unit increase in the exploratory variable.
Univariate analysis showed that when each variable was examined separately, completeness of transformation zone sampling, location of the squamo-columnar junction, the area of transformation zone (mm2), the size of aceto-white area (mm2) and the ratio of aceto-white area to transformation zone were all statistically important in influencing the agreement between the cytology and histology reports. Incomplete transformation zone sampling, when the squamo-columnar junction was not seen and a larger transformation zone size was associated with a reduction in the agreement between the cytology and histology reports. An increase in the proportion of aceto-white area to transformation zone area was associated with an increase in the agreement between reports (Table 3). Analysis of Table 3 also showed that there was no evidence that parity or the sampling devices used in the study had a significant effect upon the agreement between the cytology and histology reports.
The second stage in the analysis was to adjust each of the explanatory variables for the effect of the other explanatory variables in a multivariate analysis. A backwards selection approach was used to retain the statistically significant variables, and the final model is shown in Table 4. Multivariate analysis showed that after adjustment for the other explanatory variables, the area of the transformation zone and the ratio of the aceto-white area to the area of the transformation zone were both statistically significant. There was also slight evidence of an effect of squamo-columnar junction location (P= 0.10), so this variable was left in the final model. Women with transformation zone area greater than 30.03 mm2 were found to be more likely to have a smear that does not reflect the cervical pathology compared with women with smaller transformation zone area. Similarly, women with aceto-white area less than 7.0 mm2 were also more likely to have an inaccurate cytology report than women with larger lesions. The ratio of the aceto-white area to the area of the transformation zone of less than 22% was associated with increased risk of inaccurate cytology. An increase in the area of the transformation zone of 10 units resulted in the odds of agreement being only 0.65 times as large (or decreasing 1.5 times). Conversely, an increase of 10 units in the ratio of the aceto-white area to the area of the transformation zone measure resulted in the odds of agreement being 1.5 times higher.
Table 4. Multivariate analysis after adjusting each of the explanatory variables for the effect of the other explanatory variables on the accuracy of the cytology report.
Odds ratios are for a 10-unit increase in the explanatory variable.
The multivariate analysis indicated that the ratio of the aceto-white area to the area of the transformation zone, the area of the transformation zone and squamo-columnar junction location were found to have some influence on agreement. Larger area of the transformation zone and squamo-columnar junction location not seen decreased the chance of agreement, while the higher the ratio of the aceto-white area to the area of the transformation zone measure, the greater the chance of agreement.
The study shows that, even when cervical smears are taken under colposcopic control, sampling of the transformation zone may be incomplete. Univariate analysis showed that a number of variables were found to influence the agreement between the cytology and histology reports. These included the completeness of transformation zone sampling, visualisation of the squamo-columnar junction, the size of the transformation zone (mm2), the size of aceto-white area (mm2) and the proportion of aceto-white area to transformation zone area. However, multivariate analysis confirmed that the size of the transformation zone, size of lesion and the ratio of the aceto-white area to the transformation zone area as well as visualisation of the squamo-columnar junction, were the most important factors influencing the agreement between the cytology and histology agreement. The larger the size of the transformation zone the less chance of agreement, while the higher the ratio of the aceto-white area to the transformation zone area the greater the chance of agreement. The initial effects of the other variables can be explained by their relationship with these two variables. Surprisingly, we found that parity and type of smear-taking device (Ayre spatula, Aylesbury spatula and Cervex brush) did not significantly affect the ability of the cervical smear to reflect the underlying pathological changes in the cervix.
Our findings were consistent with the findings of Jarmulowicz et al.6 who measured the area of CIN in laser cone biopsies and correlated this to the cytological findings. They found that the larger the area of CIN in the histological sections, the greater the chance of agreement between the cytology and histology reports.
This study provided an opportunity to determine whether the presence of endocervical cells in the smear improved the accuracy of cytology in predicting the underlying pathological lesion. This has been a controversial point for many years7,8 and our study supports the view of Kivlahan and Ingram that the presence of endocervical cells in a cervical smear does not significantly affect the rate of detection of CIN. We noted that even when the transformation zone was completely sampled, the cytology results did not always correspond to the biopsy findings. This could be accounted for by the fact that abnormal cells on the smear-taking device were not transferred to the glass slide. We have shown that this occurs in 11% of cases.9 In order to determine whether this was a factor in the present study, one additional smear was prepared from the original scraped material from 43 women; unexpectedly, there was no difference in the cytology content of the conventional and the additional smear from the 43 cases investigated in this way, possibly because the smear-taking devices used in this study differed from that used in our earlier investigation.
It has also been suggested that when samples are taken at colposcopy, less pressure is applied to the cervix than when a smear is taken under direct vision in a well women clinic, as the aim of the colposcopist is to minimise the risk of contact bleeding which could obscure his or her view of the cervix. It has been proposed that this gentle approach could account for the failure of abnormal cells to be present in smears taken at colposcopy (EM, personal communication). Although it can be argued that a reduction in the amount of pressure applied to the cervix may increase the risk of an inadequate smear, it will not affect the coverage of the transformation zone, which we have shown to be less successful in multiparous compared with nulliparous women. It should also be noted that we have excluded the inadequate smears from our analysis.
To the best of our knowledge, this is the first study to investigate smear-taking technique using video-imaging techniques. The results are of importance to smear takers who are constantly exhorted to sample the whole of the transformation zone.10 Clearly, this is not always achieved even when the whole of the transformation zone is visualised under optimal conditions.
Our study also shows that women with large transformation zone areas (>30.0 mm2) and or small aceto-white lesions (<7.0 mm2) are more likely to have an inaccurate cytology report than women with small transformation zone and women with larger aceto-white areas. We also found that the ratio of the aceto-white area to the area of the transformation zone is important in determining the accuracy of the cytology report and a ratio of 0.22 or less increases the risk of disagreement between the cytological and histological findings.
The authors would like to thank Mr Paul Bassett for assisting us with the statistical analysis.