In this cohort study based on real-world data, the variation in fixation and CB processing of cytology samples by pathology laboratories in the Netherlands was assessed. We revealed that many differences exist in both the use of fixatives and of CB methods, sometimes including multiple methods within one laboratory. Correcting PD-L1 positivity rates of individual laboratories for differences in the use of fixative and CB method resulted in a reduction of the number of laboratories that differed significantly from the mean PD-L1 positivity. Moreover, the observed decrease in interlaboratory variation was considerably greater than the decrease that was seen when PD-L1 positivity rates were corrected for differences in patient and sample characteristics without the variables fixative and CB method.

First of all, the amount of different fixation and processing methods and reported combinations among the laboratories that responded to our survey is enormous: Within a total of 28 laboratories, 19 different combinations of fixation and processing cytological material into a CB could be discerned. These results are comparable to those of other studies that used surveys to assess interlaboratory variation in both fixation and CB methods, which also showed large amounts of variation in cytology processing methods between laboratories [8, 22,23,24].

Both methanol-based and ethanol-based fixatives have a potentially deleterious effect on PD-L1 immunostaining performed on CBs [15,16,17], with a risk of false-negative PD-L1 immunostaining results. Indeed, correcting for differences in cytology fixation and CB processing methods between laboratories resulted in a reduction in the number of laboratories differing significantly from the mean in PD-L1 positivity. Formalin post-fixation may reverse the negative effects of alcohol fixation to some degree [15], with some studies showing good concordance in PD-L1 positivity between histology and cytological specimens from the same tumor fixed in an alcohol-based fixative followed by formalin fixation [25,26,27]. It is unclear, however, what the maximum duration of alcohol fixation is after which formalin post-fixation is still effective, and what the most optimal formalin post-fixation time would be. CytoRich Red, containing both alcohols and formaldehyde, did not seem to have a negative effect on PD-L1 immunostaining in various studies [15, 26, 28]. Likewise, univariable logistic regression analyses in our study did not show a statistically significant difference in the odds of finding PD-L1 positivity between CytoRich Red fixation and formalin fixation, while the odds of scoring PD-L1 as positive were significantly lower in samples fixed in CytoLyt/PreservCyt without formalin post-fixation or in alcohol with formalin post-fixation compared with samples fixed in formalin.

It is possible that differences in the CB method also influence interlaboratory variation in PD-L1 positivity, regardless of the fixation method. Yet, very few studies have been published that assessed the influence of the CB method on immunostaining independently of the fixation method. In a study by Lloyd et al. [17], cytology samples processed into CBs with the Cellient-automated CB system showed optimal PD-L1 staining results compared with CB preparation according to the plasma-thromboplastin method. However, the authors advise against the use of CytoLyt as a collection medium due to the poor performance of PD-L1 immunostaining in samples collected in CytoLyt. Remarkably, CytoLyt is the collection medium of choice recommended for use with the Cellient system by the manufacturer. In our study, nearly 75% of the Cellient processed samples were fixed in CytoLyt/PreservCyt, and the remainder were fixed in an alcohol-based fixative with formalin post-fixation. None of the Cellient processed samples were fixed in formalin only. All in all, based on the available literature, it is very likely that the influence of differences in cytology processing methods on interlaboratory variation in PD-L1 positivity can be attributed mostly to differences in fixation methods.

We have reported previously that a large degree of variation in PD-L1 positivity between laboratories is problematic. Indeed, this could result in patients receiving different PD-L1 test results depending on the pathology laboratory where their material is tested [18]. Thus, efforts should be taken to keep interlaboratory variation in PD-L1 positivity to a minimum. Based on the current study, an important step to take would be to create more uniformity between laboratories in the way that cytology samples are fixed and processed, using a method that does not negatively influence immunostaining results. This desire for uniformity has been expressed by others [8, 29], too, and could prove beneficial not only to results from PD-L1 immunostaining but also to results from other immunohistochemical assays that show adverse effects of alcohol fixation, such as for progesterone receptor [30] and MIB1 [31]. External quality assessment (EQA) schemes specifically designed to assess immunocytochemistry could perhaps aid in uncovering possible technical issues and in promoting standardization [32]. Future studies should investigate which method is the preferred (combination of) cytology processing method(s) for PD-L1 testing.

Potentially, rigorous validation and optimization of immunostaining protocols that are used on cytology samples but have originally been validated on FFPE tissue samples could aid in diminishing variation as well. Unfortunately, it has been shown that validation and optimization of immunostaining protocols for cytology samples are not common practice [23, 24], even though organizations such as the College of American Pathologists (CAP) recommend that a sufficient number of cases should be tested to ensure that immunohistochemical assays achieve similar results when performed on cytological material compared to histological material [33]. No advice is given, however, on the criteria and number of specimens needed for validation, and it is stated that “separate validation of all markers on all potential cytologic specimens is generally not feasible” [33]. The type of material that should be used for validation may often not be clear either, or it may be difficult to collect enough material, especially when dealing with small cytology samples. On top of that, some laboratories receive cytological specimens from external laboratories for immunocytochemical testing, which may have been fixed and processed in a variety of ways. Moreover, with PD-L1 immunostaining, a decrease in staining intensity could result in false-negative staining results, but it is hard to determine what level of decrease in staining can still be accepted and what level would actually cause problems in clinical practice. All these factors may complicate the proper validation and optimization of PD-L1 immunohistochemical stains that are used on cytological specimens. On top of that, laboratories may use commercial assays for PD-L1 immunostaining, which use standardized protocols developed by the manufacturer that cannot simply be adjusted.

After correction for the case mix including variation in the fixative and CB method, the amount of interlaboratory variation in PD-L1 positivity was still substantial at both cutoffs. Compared to histology, tissue architecture is disrupted in cytology samples, which can complicate the recognition of tumor cells. Also, it may be a lot harder to distinguish tumor cells from inflammatory cells, especially macrophages, which may lie adjacent to or intermixed with isolated tumor cells [34, 35]. The level of experience and training of the pathologists scoring PD-L1 immunostaining on cytology in a routine clinical pathology setting may not be the same in all laboratories, all the more so because scoring of PD-L1 on cytology requires adequate training in both cytopathology and PD-L1 scoring. Structural differences between laboratories could arise, for instance, when inflammatory cells are often mistaken for tumor cells. Moreover, small tissue samples can lead to an underestimation of PD-L1 expression [36], which probably also applies to cytology samples. In fact, it has been shown that PD-L1 staining results of CBs and resection specimens are more concordant when a greater number of tumor cells were present in the CB [37]. Perhaps laboratories that structurally receive cytological samples that contain more tumor cells, for instance, because multiple passes or bigger needles are used to collect material, have higher PD-L1 positivity rates based on cytology than other laboratories. Our study, however, does not provide the data to properly investigate this hypothesis.

Of note, in our study, an association was found between sex and PD-L1 expression, with PD-L1 positivity being more likely in samples from women than from men (Table 1). While similar results have been shown by others [38, 39], various other studies did not find any association between PD-L1 expression and sex [40,41,42,43,44,45] or found that PD-L1 was more likely to be positive in men than in women [46,47,48]. These studies, however, primarily used FFPE material, mostly from surgical resections or biopsies. Our study only included cytological samples, many of which were not fixed in formalin or embedded in paraffin, which might explain the differences in results. Similarly, while some studies did not find a statistically significant association between PD-L1 expression and sampling site [49, 50], comparable to our results (Table 1), others showed that pleural and nodal metastases were more likely to express PD-L1 than primary tumors [42]. Again, the difference in results could potentially be explained by the latter study using FFPE material which largely came from biopsies or surgical resections, while our study only used cytological material fixed and processed in a variety of ways. Moreover, the differences in proportions between PD-L1 positivity and negativity of the various characteristics in Table 1 were tested through univariate analysis, which does not account for potential confounding factors. Also, since we used a large cohort in our study, small and maybe even clinically insignificant differences might be statistically significant, whereas they might not have been in studies with smaller sample sizes. These factors should be taken into account when interpreting these results.

This study has some limitations. Most importantly, even though a considerable amount of laboratories responded to our survey, we did not receive answers from all laboratories. This resulted in the exclusion of 516 patients from the PALGA data set, for whom the fixative and CB method were unknown. Given the current variation in the fixation and CB methods, it is to be expected that the overall number of methods used would only be larger, potentially resulting in a larger baseline variation among laboratories to start with. This should be considered when interpreting the analyses of the influence of variation in fixation and CB processing on interlaboratory variation in PD-L1 positivity. Second, the respondents reported varying mean fixation times. Vigliar et al. [51] showed that formalin fixation time influences PD-L1 immunostaining results on CBs. This could be the case with other fixatives, too. Unfortunately, we did not know the fixation times for individual samples in the PALGA data set and, thus, could not incorporate information on fixation time in our analyses. Third, due to a large amount of variation in cytology processing methods, especially in fixation methods, it was quite difficult to divide these various methods into larger categories. In fact, if numbers had allowed to include all methods as they were, a better correction could have been performed. However, we do feel that the distribution that we used is compatible with the currently available literature. Finally, since our study is based on real-world pathology data, we were dependent on the way that pathologists report their findings. For instance, while it would have been interesting to include an analysis of TPS on a continuous scale, the fact that various laboratories only reported TPS in categories did not allow us to do so. Also, some potentially relevant information, such as information on previous treatment, is not regularly part of pathology reports, meaning that we could not correct for potential differences between laboratories within these areas. Regarding treatment status, however, in many patients, PD-L1 testing was performed on the initial diagnostic material, either at the time of diagnosis or at a later time. We also excluded patients with more than one primary lung tumor, to avoid including data from PD-L1 tests that might have been influenced by previous treatment. We therefore expect the number of patients in which PD-L1 testing was performed solely on material collected after administration of chemotherapy to be too small to influence our results in a significant way.

To conclude, this study shows that a lot of variation exists between laboratories in the methods used for fixation and CB processing of cytological samples. We have demonstrated that these differences influence interlaboratory variation in PD-L1 positivity in NSCLC patients, with a decrease in the amount of variation when PD-L1 positivity rates are corrected for differences in fixation and CB methods. A high degree of variation in PD-L1 positivity between laboratories is problematic, because this will almost inevitably lead to patients receiving different courses of treatment depending on the laboratory where their cytological material is stained and scored for PD-L1. These results warrant the need for more research to determine the best methods of fixation and CB processing of cytology samples on which PD-L1 immunostaining is to be performed, and for harmonization of these methods between laboratories.