A study of cellular counting to determine minimum thresholds for adequacy for liquid-based cervical cytology using a survey and counting protocol

Survey of working practice

In November 2007, questionnaire surveys (see Appendix 1) to assess the standards and practice of reporting LBC adequacy were sent to the 56 laboratories in England, Scotland and Wales that had agreed to participate in the study. In Scotland only TP was used, in Wales only SP was used and in England both SP and TP were used. All but one (98%) of the questionnaires were returned. Of those that responded, 28 used SP (Table 1) and 27 laboratories used TP (Table 2). Of these laboratories, 15 out of 28 (54%) of the SP laboratories and 14 out of 27 (52%) of the TP laboratories also provided a copy of their standard operating procedure (SOP) for assessment of specimen adequacy.

Among the SP laboratories, 18 out of 28 (64%) stated that they use morphological criteria (epithelial cells obscured by mucus or blood; cytolysis; absence of endocervical cells in follow-up of cervical glandular intraepithelial neoplasia) to determine specimen adequacy and 13 out of 28 (46.4%) stated that they record the presence of indicators of transformation zone sampling (endocervical cells or metaplastic squamous epithelial cells). The vast majority (27 out of 28, 96.4%) assess specimen adequacy by counting a minimum number of squamous epithelial cells in adjacent ocular fields. Among those in the SP laboratories that stated the calculated/estimated minimum number of squamous epithelial cells that must be present on a slide for it to be regarded as adequate (11 out of 28), all gave a figure of at least 15,000 cells.

Of the TP laboratories surveyed, 15 out of 27 (56%) stated that they use morphological criteria (epithelial cells obscured by mucus or blood; cytolysis; absence of endocervical cells in follow-up of cervical glandular intraepithelial neoplasia) or technical criteria (sampling brush left in specimen container) to determine specimen adequacy. The presence of indicators of transformation zone sampling (endocervical cells or metaplastic squamous epithelial cells) was recorded by 7 out of 27 (25.9%) laboratories. A majority of the laboratories (20 out of 27, 74%) assess specimen adequacy by counting a minimum number of squamous epithelial cells in adjacent ocular fields. Among the TP laboratories, for the 11 out of 27 (41%) that stated the calculated/estimated minimum number of squamous epithelial cells that must be present on a slide for it to be regarded as adequate, the range of 5000–15,000 cells was counted, usually in response to local or regional guidance. The distribution of the minimum calculated/estimated number of squamous epithelial cells was as shown in Table 3.

Discussion

In summary, while the majority of laboratories using TP and SP LBC systems use morphological and, in the case of TP, technical criteria to assess specimen adequacy, most do not record the presence of indicators of transformation zone sampling. The majority of laboratories assess specimen adequacy by counting or estimating a minimum number of squamous epithelial cells. All SP laboratories require a minimum of 15,000 cells for a sample to be considered adequate, but in TP laboratories there is a wide range of minimum acceptable cellular counts (MACCs), varying from 5000 to 15,000 cells, with only 3 out of 11 laboratories using a MACC of 5000.

This variation in practice is strong justification for a study to determine a MACC for both SP and TP LBC systems and provide guidance for the NHSCSP.

The background to this study was the NICE report that highlighted the need to determine whether or not there was a threshold of cellularity which should be determined to define the adequacy of a slide. 9 In the interim, UK laboratories had adopted a figure of up to 15,000 cells as a determinant of adequate LBC cellularity for the SP system, based on the LBC pilot experience and other guidance, and this was confirmed in the questionnaire survey. For all laboratories, an estimation of total cellularity was obtained by performing representative ocular field cell counts, although a review of the submitted SOPs revealed that practice varied from one laboratory to another; that is, truly adjacent consecutive fields or fields with spaces between starting point and direction of counting. Specific guidance for Scotland, where the TP samples system is exclusively used, recommended a minimum of 10,000 well-preserved squamous epithelial cells, and this was confirmed in the survey. All SP laboratories used a minimum of 15,000 cells. In addition, a majority of both SP and TP laboratories used morphological indicators to assess specimen adequacy. A substantial minority of both SP and TP laboratories recorded the presence of indicators of transformation zone sampling, but it would appear that this criterion was not used in the majority of laboratories as an indicator of an adequate sample.

More recently, pending the results of this study, quality assurance guidance for the NHSCSP has recommended that an adequate liquid-based sample is defined as one that contains the minimum level of squamous epithelial cellularity necessary to ensure a squamous abnormality detection rate equivalent to that offered by conventional smears. 14

It is clear from this survey that differences in practice exist, that these differences are not evidence based, and that it is timely to develop evidence-based practice guidelines. Developing the evidence required to produce such guidance was the purpose of this study, as described in Objectives 2, 3 and 4, which in turn cover counting methodology, cell counts from all of the study laboratories on a standard slide set and the use of cell dilution to evaluate the effect of reducing overall cell counts, as well as dyskaryotic cell counts, on the chances of detection.

Method

The objective was to establish a reliable method for rapidly estimating the cellularity of a LBC sample for SP and TP slides. Cells were included in the count if they were intact, mature or parabasal squamous cells with nuclei, even if the nuclei were pale. Syncytial aggregates of squamous cells, as seen in cytolysis, were counted according to the number of nuclei they contained, even if the cytoplasmic margins of individual cells were not identifiable. Any free nuclei, and any anucleate squamous cells or fragments of squamous cytoplasm were not counted. If no cellular material was present then a zero value was recorded. If exceptionally thick groups of cells were present, an estimate of cellularity was used. A full quadrant of a high-power microscope field (×40 objective) contains approximately 1000 small parabasal squamous cells and 750 mature squamous cells. Cells at the edge of the field were counted if the entire circumference of the nucleus was seen (otherwise they were not counted).

The cell count for a slide was obtained by counting the number of squamous cells in 10 fields of view. The total cell count for a slide was then computed as

The SOP by which the 10 fields of view were selected was as follows. An assessor had a choice of one of four starting points to begin the slide count. Assuming that the deposit represented a clock face, a starting position on the perimeter at 12, 3, 6, or 9 o’clock was chosen for the first field that was neither hypo- nor hypercellular. After counting cells in the starting field of view, the assessors then moved along a radius towards the centre of the slide in steps determined by the width of the field of view counting a further nine fields. For the SP LBC system, adjacent fields of view were counted, whereas for the TP system every second field was counted (see Appendices 3 and 4).

Reliability study of squamous cell counting procedure

A reliability study was carried out to assess the consistency of the cell counting procedure. This considered two situations. The first was a pair of assessors using the same starting position (12, 3, 6, or 9 o’clock), and the second was a pair of assessors using different starting positions. For each LBC system, 30 slides were assessed by three experienced cytopathologists (LT, MD and JS). LT performed the cell count procedure for all four starting points, completing 120 cell counts for each system. So as to mimic the everyday laboratory procedure, JS and MD performed just one cell count for each slide using their choice of starting position based on the cell counting SOP, thereby completing 30 cell counts each for each system. This gave data that enabled the repeatability of the procedure to be assessed when using the same starting point or different starting points.

Statistical analysis (total cell counting)

Reliability of cell counting was assessed using an intraclass correlation coefficient (ICC). In this setting, the ICC estimates the proportion of the total variance of the cell count between slides. In the ideal situation, where there is no measurement error, that is no disagreement between assessors/cytopathologists, the ICC will be equal to 1. Alternatively, if all the measurement was noise, so that nothing is being learnt regarding the true cellularity, the ICC would be equal to 0. An ICC closer to 1 therefore represents higher reliability. To calculate the ICCs provided by three assessors using same/different starting positions, nine pairs of results (four pairs JS with LT, four pairs MD with LT, and one pair JS and MD) were created for each slide and corresponding confidence intervals (CIs) calculated using a bootstrap procedure. A bootstrap procedure is performed by randomly sampling with replacement of each pair of results until a sample of equivalent size to the original data is created. This is repeated 1000 times. Each sample is then analysed and the results are combined together to give robust standard errors. 15 For each LBC system, the total cellularity of 30 slides was assessed, with LT completing 120 cell counts and JS and MD each completing 30. JS and MD used the same starting point for 8 of the 30 SP slides and 15 of the 30 TP slides.

Statistical analysis (thresholds of cellularity)

We began by comparing the agreement of assessors in determining if the cellularity was above or below the MACC. Agreement was measured using a kappa coefficient, and the same pairing and bootstrapping procedure as for the previous analysis using the ICC. The kappa coefficient is a measure of reliability for categorical data corresponding to the ICC used for the total cell count and has a similar interpretation. A kappa coefficient of 1 implies complete agreement between the two ratings, so there is no measurement error, whereas a kappa coefficient of 0 implies no agreement beyond that as a result of chance.

Results

Discussion

The estimates of the kappa coefficient for cell counting are systematically lower for TP than for SP, but it should also be noted that the CIs for TP are wide and some overlap the CIs for SP. This difference between the systems may be explained by the smaller proportion of slides below the threshold used for TP. Low prevalence of slides below the threshold may explain the lower values of kappa, as binary scales with a prevalence closer to either 0 or 1 tend to have smaller values of the kappa coefficient than scales with prevalence closer to 0.5. 16 This property of the kappa coefficient has implications for the values of kappa observed for the SP threshold. In service settings, the prevalence of slides with cellularity below the threshold needs to be low if the screening method is to have utility. If the reliability study for SP were repeated in a sample with a more realistic proportion of subjects being below the threshold, it is likely that the kappa coefficient would be rather lower. Hence, the low value of kappa observed for TP may be more realistic than the value of kappa observed for SP. This has implications for the application of a strict threshold to this method of cell counting.

There is some evidence from Table 4 of only a small proportion of substantial disagreements between assessors. Considering first SP, the number of occasions for which the two assessors disagreed substantially, that is, where one assessment was less than 10,000 and the other was greater than 15,000, was only three (see Table 4). Similarly, for TP, there were only eight (see Table 4) pairs of assessments where one assessor scored less than 5000 and the other above 10,000.

This study has some limitations. First, the sample size used for either system is small, with each being based on only 30 slides. A study with a larger sample size would be needed to gain a more precise estimate of reliability. A second issue is the representativeness of the slides used for the reliability exercise. The proportion of assessments with a cell count below 15,000 cells for SP or below 5000 cells for TP was higher than would be found in a representative service sample of all slides. Nonetheless, it should be noted that cell counting is likely to be applied as a formal technique only in samples that have been pre-screened ‘informally’. The sample may, therefore, be rather more representative of samples to which one might actually apply formal cell counting. One avenue for further work would, therefore, be to compare the reliability of a combined formal and ‘informal’ assessment of slide cellularity.

A methodology for estimating the total number of epithelial cells on a slide by counting the number of epithelial cells within 10 fields of view and combining them to give an estimate of the total cell count has been developed and evaluated in this study. To assess the reproducibility of this method, a reliability study was carried out which revealed an ICC between 0.547 (95% CI –0.456 to 0.638), representing moderate agreement in evaluation of TP samples, and 0.741 (95% CI 0.700 to 0.784), representing strong agreement in evaluation of SP samples. This difference might be expected because of the more uniform distribution of cells on SP slides, and it may also explain why in TP samples the point estimates of reliability in the comparisons of the same starting position are less than in different starting positions. For the same reason, it is surprising that the ICCs for the two assessors who followed the SOP strictly regardless of the choice of starting position for counting were 0.305 (95% CI 0.000 to 0.633 – fair agreement) and 0.650 (95% CI 0.441 to 0.858 – moderate agreement) for SP and TP, respectively. It should also be noted that the cell counts were performed by three senior experienced consultant cytopathologists, whereas in routine practice it is expected that these counts would be performed by primary screening staff, either cytology screeners or biomedical scientists, and these staff groups have different microscopic and morphological interpretive skill sets. It might have been better if this part of the study had been undertaken by primary screening staff in order to provide assurance that the counting methodology was appropriate and utilisable in routine clinical practice. However, in routine practice, formal cell counting as described is rarely required in SP samples, as the vast majority of samples are clearly adequate in terms of a naked-eye assessment, when the colour and density of the cell deposit strongly contrasts with inadequate or potentially inadequate samples. Inadequate samples show very pale-stained or virtually invisible cell deposits, and this is confirmed on low-power microscopic examination.

In terms of estimating LBC cellularity, there appears to be a better interassessor agreement with SP than with TP, and this may be related to the qualitative differences between the way the cells are presented on the slide. There also appear to be differences in reproducibility in terms of counting; however, it is not possible to state categorically that a specific method (same or different starting position) achieves greater reliability. Given the variation inherent in cell counting, it would perhaps be optimal if an agreed counting method were agreed for use in the NHSCSP.

Method

A survey was carried out to determine the distribution of cellularity of samples classified as inadequate, negative or abnormal. The objectives of the survey were:

• to determine the distribution of cellularity of samples classified as inadequate, negative or abnormal

• to investigate the threshold used by different laboratories to determine adequate cellularity

• to investigate the cellularity distribution of samples classified as cytology negative and HPV positive.

The 56 laboratories which were involved in the survey (28 SP and 28 TP) were asked to submit 20 consecutive cervical screening LBC cases from each of the categories of inadequate, mild dyskaryosis and high-grade dyskaryosis (moderate dyskaryosis and above), and a further 50 consecutive negative cases. Some laboratories varied slightly from the requested numbers.

It was not feasible for all material from the slide survey to be assessed for cellularity at a single centre. The task of cell counting was, therefore, distributed across all participating laboratories. In order to reduce potential bias as a result of interlaboratory variability in the assessment of cellularity, each laboratory was sent a set of slides with similar composition by type (inadequate, mild dyskaryosis, high-grade dyskaryosis or negative) from the other laboratories using the same LBC system. Slides from the originating laboratories were relabelled with an anonymous code and repackaged, randomly sorted within the laboratory and by slide type before being systematically allocated to laboratories using a computer-generated pre-prepared list and then sent to the participating laboratory for cell counting. Each laboratory received approximately four slides from every other laboratory using the same system, with approximately 20 slides from each of the inadequate, mild dyskaryosis and high-grade dyskaryosis classifications and 50 slides classed as negative. Each laboratory was asked to nominate primary screening staff to carry out the cell counting. One laboratory withdrew from the study at this stage and their batch of slides was randomly reassigned to other laboratories in the study. All slides in the study sets were assessed for the presence of transformation zone indicators and each had a formal cell count according to the study cell counting SOP (see Appendices 3 and 4; see also Objective 2: development and test of cell counting methodology). Cell counts were recorded directly into a database to minimise data errors and the database was returned on completion to the study centre (Liverpool) with the slide sets.

Slides classified as cytology negative but HPV positive might be considered to contain a small proportion of cytological false negatives. Furthermore, if HPV primary screening is introduced, it is likely to require women who are HPV positive to undergo reflex cytology in order to triage ongoing management. To investigate the cellularity distribution of such samples classified as cytology negative but HPV positive, 1200 cases from the ARTISTIC (A Randomised Trial In Screening To Improve Cytology) study17 previously documented as cytology negative and HPV positive were also subject to cell counting. A comparison of the cell count distribution with both the negative and mildly dyskaryotic cases from the slide survey would help to determine whether or not their potentially false-negative cytology relates to their cellularity. The ARTISTIC study used the TP LBC system; therefore, these cell counts were compared with those for negative and mildly dyskaryotic slides from the 28 TP laboratories.

Statistical analysis (slide comparisons)

The SP and TP total squamous cell counts across all slides and within each slide diagnosis (inadequate, high grade, mild dyskaryosis and negative) were compared using an independent-samples t-test. Owing to the highly skewed nature of the cell counts, a square root transformation was performed prior to the comparison tests. In order to describe the cell count distribution, continuous cell counts for both LBC methods were categorised as was deemed suitable by the cell count distribution and displayed in a cross-tabulation with slide diagnosis. For comparison, all TP tables also include cell counts from the cytopathology-negative/HPV-positive ARTISTIC data set. 17

Assessing consistency across reading laboratories within inadequate and negative slides is important, and comparisons of cellularity for inadequate slides across reading laboratories were performed. A one-way analysis of variance (ANOVA) compared mean cellularity between laboratories for inadequate slides and a chi-squared test compared the proportion of slides with cellularity above the cut-off points 15,000 and 5000 between laboratories. Formal statistical testing is difficult to interpret when comparing large numbers of units because of the large number of post-hoc pairwise comparisons that can be made. The ICC was performed as an alternative measure of heterogeneity in order to assess the magnitude of variation between laboratories in either the mean cellularity of inadequate slides or the proportion of inadequate slides above the MACC threshold when compared with the total variation. The ICC estimates the proportion of the total variation that occurs between laboratories. In this context, an ICC of 0 indicates that there is no variation between laboratories above that as a result of sampling variation. Small sample sizes when comparing the cellularity of negative slides by laboratory means that the expected cell counts are likely to be small, hence a Fisher’s exact test may replace the chi-squared test.

Results

Almost all laboratories submitted the minimum number of requested slides (110 slides). Two TP laboratories produced 123 slides and three SP laboratories produced 123, 125 and 136 slides. In all, 3110 cell counts were carried out on SP slides and 3176 cell counts on TP slides, totalling 6286 slides.

Key findings

1. The mean counts for inadequate samples at around 14,000 and 11,000 for SP and TP, respectively, are lower than those for negative, mild dyskaryosis and high-grade dyskaryosis samples, which are around 50,000 for both SP and TP.

2. A MACC cut-off point at 15,000 for SP would include only around 25% of the slides reported as inadequate, as well as 97.5%, 97.7% and 98.6% of the slides reported as high-grade dyskaryosis, low-grade dyskaryosis and negative, respectively. This suggests that a SP MACC cut-off of 15,000 could be confirmed as a sensible count reflecting current reporting practices. With regard to TP, however, the MACC recommendation of 5000 would include 56.7% of inadequate samples, and 98.2%, 98.4% and 98.7% of slides reported as high-grade dyskaryosis, low-grade dyskaryosis and negative, respectively. Even above a MACC cut-off of 10,000, which is used in Scotland, 32% of slides reported as inadequate would be included along with 93.5%, 96.4% and 94.2% of high-grade dyskaryosis, low-grade dyskaryosis and negative slides.

Discussion

The results of this cell counting exercise reveal a number of useful findings. The first is that, overall, the mean cell counts for slides classified as inadequate are around 14,000 and 11,000 for SP and TP, respectively. These are considerably lower than for slides reported either negative or abnormal, at around 50,000 for both SP and TP. It should be recognised, however, that the SDs around these means are wide. The second finding relates to the relationship between the MACC and the slide results. A MACC cut-off point of 15,000 for SP would include around 25% of the slides reported as inadequate, as well as 97.5%, 97.7% and 98.6% of slides reported as high-grade dyskaryosis, low-grade dyskaryosis and negative, respectively. This suggests that the widely accepted SP MACC of 15,000 is confirmed as a sensible count reflecting current practice.

With regards to TP, recommended count of TBS of 5000 would include more than 50% of the inadequate samples, and 98.8%, 98.4% and 98.7% of high-grade dyskaryosis, low-grade dyskaryosis and negative samples, respectively. These results, which reflect a cross-section of reporting practice in England, also suggest that a MACC for TP of 10,000, as is currently recommended in Scotland, would still exclude a substantial proportion of slides reported as inadequate and it would also exclude slightly more negative slides. The proportion of TP slides with cellularity below 5000 and 10,000 in this study read as adequate by participating laboratories can be compared with data from the study by McQueen and Duvall. 11 They reported that, among slides read as adequate (normal or dyskaryotic), 2.5% and 6.5% had a cellularity of less than 5000 and 10,000, respectively. Our data for the same categories were somewhat lower, at 0.9% and 4%, respectively. The finding from our data, that 25% of TP slides read as inadequate had a cellularity between 5000 and 10,000, suggests that, by reducing the MACC from 10,000 (as used in Scotland and other laboratories in England), the TP inadequate rate could be lowered from 2.5% in Scotland18 to below 2%.

The TP slides included HPV-positive cytologically negative samples from the ARTISTIC study. 17 Looking to the future, this category will constitute the large majority of cytology which will continue to be used to triage women who screen HPV positive. It is clear from these results that the cellularity of slides reported as negative in women whose HPV status is unknown is almost identical to that found in the ARTISTIC slides. This suggests that the results of this study will be applicable to a possible future role of cytology based on HPV status. Consideration is also given to an alternative MACC of 10,000 for TP in the dilution study that follows, when detection rates are analysed.

Introduction

The total number of squamous cells in a cervical sample may influence the probability of detection of abnormality by screeners. To investigate this, serial dilution was used to produce slides from established cases of high- or low-grade abnormality. These slides are referred to as unmixed dilutions. In addition, the relative proportion of dyskaryotic cells compared with normal squamous cells may also influence the probability of detection of abnormality by screeners. This was investigated by producing slides with varying ratios of dyskaryotic to total squamous cells but with a similar background cellularity, which are referred to as mixed dilutions. The unmixed and mixed dilutions can both be thought of as true-positive cases that should be detected by screening as either low- or high-grade dyskaryosis.

Methods

A total of 176 SP and 176 TP cases were selected from material routinely accessioned at the Royal Liverpool University Hospital and the Manchester Cytology Centre that displayed a range of histologically confirmed low- and high-grade cytological abnormalities.

Figure 1 outlines the slide preparation structure for both LBC methods. Diluted preparations were made from each source sample. Seven serial dilutions were made from half of the cases and referred to as ‘unmixed dilutions’. The range of dilutions from a cellularity of 5000–10,000 to over 55,000 was skewed towards preparations of lower cellularity, as these were expected to have higher false-negative rates. The remaining half of the cases were serially mixed with known negative cases in varying proportions to establish sets of slides containing different numbers of abnormal cells ranging from < 25 to over 1600. These slides are mixed with normal samples in order to dilute the abnormal cells with normal cells and were, therefore, referred to as ‘mixed dilutions’ (see Appendix 6). In total, 2400 new slides were prepared for each LBC system.

FIGURE 1.

Flow diagram outlining slide preparation structure for each LBC method SP/TP. a, Slides with no dyskaryotic cells were removed and replaced. These replaced slides were removed from the main analysis because of a coding error.

Results

A total of 2400 samples from one of four original sources (inadequate, negative, unmixed dilutions and mixed dilutions) were sent to each of 3 out of 24 laboratories for a morphological assessment – high-grade dyskaryosis, low-grade dyskaryosis, negative and inadequate – resulting in 7200 results.

Detection of abnormal cytology

Table 16 gives the distribution of ‘low- or high-grade’ assessments and the detection rate stratified by specimen cellularity for the unmixed dilutions. With each slide assessed by three laboratories, each could be 0, 1, 2 or 3 assessments as ‘low or high grade’. Suppose the frequency of 1, 2 or 3 positive assessments for a particular stratum of cellularity are f1, f2, and f3. Table 16 gives these frequencies for each strata. The detection rate is, therefore,

where n is the number of slides in the strata. For example, if we consider the band of cellularity 0–4999 for SP in Table 16, there are 10 slides, of which one slide had no assessments as ‘low or high grade’, one slide had one ‘low- or high-grade’ assessment, three slides had two assessments, and five slides were assessed as ‘low or high grade’ by all three laboratories. These frequencies combine to give a detection rate for ‘low or high grade’ for this stratum equal to

For both SP and TP, the lowest band of cellularity had the lowest rate of detection of ‘low- or high-grade’ abnormality.

For SP, the detection rate increased from 73% for cellularity between 0 and 4999, to a maximum of 94% for cellularity above 75,000. It is noteworthy that in SP, for slides above the cellularity of 15,000, there was unanimity among the three readers in well over 70% of cases. Similarly, it is of note that, in the case of the 11 slides with cellularity above 75,000, there was unanimity among the three assessors in nine cases. For TP, the detection rate increased from 80% for cellularity between 0 and 2499 to a maximum of 97% at cellularity between 50,000 and 74,999, dropping to 80% for cellularity above 75,000. In TP, unanimity of three readings reached 83% and 85% in slides with cellularity over 5000 and 10,000, respectively. For cellularity between 50,000 and 74,999, there was unanimity of the three assessments for 45 of the 48 slides.

Table 17 presents analyses to investigate the effect of hypo- and hypercellularity on the detection of low-grade or high-grade abnormality. This analysis reports the OR associated with either hypocellular (lowest 10%) or hypercellular (highest 10%) compared with the middle 80% range. The hypocellular threshold at 10% approximated to 5000 and 15,000 for TP and SP, respectively. The hypercellular threshold approximated to 58,000 for both systems. ORs below 1 indicate a reduction in the detection rate compared with the middle 80%. These ORs were estimated using the logistic GEE regression. There was evidence that hypocellularity reduced the detection rate for both LBC systems as the OR of detection was 0.474 (95% CI 0.230 to 0.976; p = 0.043) for SP and 0.250 (95% CI 0.108 to 0.577; p = 0.001) for TP. The detection rates appear to be increased in the hypercellular band for SP with an OR equal to 3.136 (95% CI 1.199 to 8.199; p = 0.020). For TP, the detection rate was reduced (OR 0.366, 95% CI 0.161 to 0.832, p = 0.016). The difference between the two systems may be explained by the numbers of dyskaryotic cells in hypercellular slides.

Table 18 summarises the percentage of assessments in each cytological category by cellularity and LBC system for slides prepared as unmixed dilutions. Two ORs were estimated using the logistic GEE regression. The ‘non-low or high grade’ versus ‘low or high grade’, and ‘non-negative’ versus ‘negative’, were compared with the cellularity cut-offs (15,000 and 5000) for each LBC system. The ‘low- or high-grade’ percentages for SP are 81.8% for slides below 15,000 and 89.0% for slides above 15,000 (OR 0.56, 95% CI 0.34 to 0.91; p = 0.020). Corresponding TP results are in 81.6% of slides assessed as low grade or high grade where cellularity was below 5000, compared with 91.7% for slides above 5000 (OR 0.40, 95% CI 0.23 to 0.71; p = 0.016). This indicates that in both systems cellularity greater than 15,000 in SP and 5000 in TP reduces the likelihood of failing to detect ‘low- or high-grade’ dyskaryosis by approximately 56% and 40% for SP and TP.

The proportion assessed as negative is higher for higher cellularity, with the non-negative compared with negative ORs is 0.410 (95% CI 0.360 to 0.850; p = 0.021) and 0.606 (95% 0.217 to 1.695; p = 0.195) for SP and TP, respectively. When cellularity is greater than 15,000 in SP and 5000 in TP, the likelihood of a non-negative outcome is reduced by 41% and 61%, respectively.

Discussion

This prospective study of the effect of serial dilution of samples containing dyskaryotic cells confirmed that there was significant reduction in detection of dyskaryotic cells when squamous epithelial cell counts were below 15,000 and 5000 for SP and TP, respectively. This is the first study to have analysed both LBC systems and utilised a very broad-based evaluation across a large number of cytology laboratories providing services for the cervical screening programmes in the UK. The results support the use of a MACC of 15,000 cells for an adequate SP sample and 5000 cells for an adequate TP sample.

The confidence limits for the detection rates where cellularity is below the MACC are wide because of the small numbers (Table 16) and there is some uncertainty regarding the determination of cellularity at the MACC (see Table 4).

It has been suggested that SP requires a higher MACC because of the smaller cell deposit area, although this is not a proportionate increase (TP, 15.9 cells/mm²; SP,112 cells/mm²), supporting Duval’s21 suggestion that the MACC may be dependent on the preparation method either because TP preferentially enriches and SP depletes the preparation in abnormal cells or because small numbers of abnormal cells are more difficult to detect in SP than in TP preparations. The latter could well be the case because the increased squamous epithelial cell density in SP preparations makes sparse abnormal cells difficult to detect under routine screening conditions. 22

All of the high-grade cases included in this study were confirmed histologically as CIN 2 or CIN 3; some, but not all, of the low-grade cases had histological confirmation of CIN 1. Participating laboratories were asked to assess slides as inadequate, negative, low- or high-grade dyskaryosis, but no attempt was made to correlate the grade of dyskaryosis with the grade proffered by the submitting laboratory. There are a number of reasons for this decision. Cytological preparations reported as high-grade (severe) squamous dyskaryosis will often contain the complete gamut of changes from low to high grade and the relative proportion of those grades is hugely variable. As a consequence, residual dyskaryosis in highly diluted preparations from high-grade lesions may be of low grade. Hence, when the dilutions were counted, all dyskaryotic cells were counted, irrespective of their individual grade. This does not detract from the overall purpose of the study, which was to determine what proportion of abnormal cells would be detected, not missed, and either triaged by HPV testing for mild abnormalities or referred directly to colposcopy if high grade.

Each of the outcomes in this study suggests that a MACC of 15,000 and 5000 for SP and TP, respectively, would be associated with greater unanimity of reading and higher detection rates for cytological abnormalities. Raising the MACC for TP to 10,000 would not appear to be relevant according to the results of the unmixed-dilution study.

Detection of abnormal cytology

Table 22 gives the distribution of ‘low- or high-grade’ assessments and the detection rate stratified by specimen cellularity for the mixed dilutions. For SP, the detection rate was highest at 58% for cellularity between 15,000 and 24,999, and decreased to 41% for cellularity above 50,000. TP detection rates tended to be higher, at an average of 80%, and increased from 70% for cellularity between 5000 and 7499 to a maximum of 81% for cellularity 50,000–74,999 and 80% for above 75,000. For cellularity below 5000, limited sample sizes of two and four slides meant an inconsistent set of detection rates, hence the large CIs. The same applies to cellularity below 10,000.

To investigate how varying the dyskaryotic cell count with total cell count affects the detection of an abnormal slide, Table 23 gives the detection rates of a ‘low- or high’-grade slide given varying total and dyskaryotic cell counts. In addition, rate ratios are given for each combination of total and dyskaryotic cell count when compared with the lowest total cell count group (SP = 15,000/TP = 5000) and the highest dyskaryotic cell count (> 50). A rate ratio greater than 1 indicates an increased rate of detection whereas a rate ratio less than 1 indicates a decrease.

For SP, detection rates appear to decrease as total cell count increases and increase as dyskaryotic cell count increases. This resulted in the largest detection rate (0.768) occurring when total cell count was below 25,000 and dyskaryotic cell count higher than 50, and the smallest detection rate (0.139) when the total cell count was higher than 50,000 and the dyskaryotic cell count lower than 25. TP tended to show an increase in detection as dyskaryotic cell count increased, but stayed constant as total cell count increased.

To investigate if trends were present in the ordinal variables described in Table 23, a logistic GEE regression model was fitted to assess the likelihood of detecting a ‘low or high grade’ versus ‘non-low or high grade’. The results in Table 24 reveal that in both LBC methods the interaction between total and dyskaryotic cell count was fitted and found to be non-significant, indicating that a change in total and dyskaryotic cell counts independently affects the odds of ‘low- or high-grade’ result, hence the interaction was removed in the subsequent model.

Once the interaction was removed, SP results confirmed a significant decrease in the likelihood of detecting ‘low- or high-grade’ outcome as cellularity increased (OR 0.512, 95% CI 0.328 to 0.798) and a significant increase as dyskaryotic cell count increased (OR 4.949, 95% CI 3.603 to 6.798). For TP, there was a slight increase in the detection rate as total cellularity increased, although this was not significant (p-value 0.214). The detection rate increased with an increase in dyskaryotic cell count (OR 4.315, 95% CI 3.103 to 6.000).