Pulmonary adenocarcinoma (AD) has a variety of architectural patterns. Recently, a 3-tiered histological pattern-based grading system was developed for stage I lung AD, stratifying patients into low, intermediate, and high risk for recurrence. However, cytology may serve as the primary method for diagnosis in patients with inoperable disease. Attempts to correlate architecture between parallel cytological and histological preparations have not been successful. Therefore, we evaluated cytomorphologic features of previously histologically graded AD to identify features of potential prognostic significance.
One hundred and thirteen fine-needle aspirations with excised adenocarcinomas were reviewed. In the liquid-based preparation, we evaluated cell arrangements (flat sheets vs 3-D clusters vs single cells), nuclear features (size variability, shape, and contour), nucleoli (prominent or inconspicuous), presence of nuclear inclusions, chromatin (fine, coarse, or clumped), and quality of background. The features were tested by multivariate analysis to identify associations with histological grade and disease-free survival (DFS), and a cytological score was generated.
Nuclear size, chromatin pattern, and nuclear contours showed a significant association with histological grade and DFS. These features were included in the composite cytological score (range, 0-5). By grouping the cytological scores, we stratified the tumors into low (median DFS, 100%), intermediate (median DFS, 78%), and high (median DFS, 55%) rate of recurrence (P = .008). There was a good correlation with the histological grading system.
Pulmonary carcinoma is a leading cause of death worldwide.1 Approximately 60% of new cases are already at advanced clinical stage at initial diagnosis2; thus, the majority of newly diagnosed lung cancers are not candidates for curative surgery. Therefore, small biopsy specimens, especially cytology, are the only means of diagnosing these tumors and the only source of tissue available for biomarker studies.3
Adenocarcinoma is the most common of all the histological types of pulmonary carcinoma.1 In addition, pulmonary adenocarcinoma is very heterogeneous histologically, which is recognized by the current World Health Organization (WHO) classification of lung cancers, where the category of adenocarcinoma mixed subtype is described.4 In fact, it has been reported that approximately 90% of all pulmonary adenocarcinomas are classified as adenocarcinomas of mixed subtypes.5 Based on findings that certain histological patterns of pulmonary adenocarcinoma are associated with prognostic implications,5-10 a histological-based grading system was developed for stage I adenocarcinoma. This grading system stratified patients into low-, intermediate-, and high-risk categories for disease recurrence by identifying the predominant histological patterns. Essentially, grade 1 tumors are composed predominantly of the bronchioalveolar carcinoma (lepidic) pattern, grade 2 is composed of tumors with predominantly acinar and papillary patterns, and grade 3 (ie, the highest-risk group) is composed of tumors of predominantly solid and micropapillary patterns.11, 12 Although cytological features of bronchioalveolar carcinoma have been shown to correlate well with the histological pattern of a well-differentiated adenocarcinoma,13-15 an attempt to correlate specific histological types with cytological patterns showed no reliable correlation.16 This suggests that we cannot predict the histological subtype of adenocarcinoma by cytological patterns of the cell aggregates seen on smears, and a different approach to cytomorphologic prognostication may be more appropriate. Good candidates are the nuclear features of the neoplastic cells, which are generally well preserved in cytological preparations and, unlike the pattern of cell aggregates, less prone to physical disruption or alteration. In the breast, it was demonstrated that nuclear grading of carcinomas showed good correlation between cytology and tissue samples.17 In this study we attempted to correlate nuclear features of pulmonary adenocarcinoma with the histological grading system previously described.11 The cytological features with prognostic significance were used to create a scoring system for pulmonary adenocarcinoma and correlate with disease-free survival interval for patients with stage I tumors.
MATERIAL AND METHODS
The prospectively maintained surgical pathology database of Memorial Sloan-Kettering Cancer Center was searched for all patients who underwent resection for stage I primary adenocarcinoma of the lung with curative intent and had a corresponding presurgical aspiration biopsy of the lesion. The AJCC 7th edition criteria for stage I lung tumors include tumor size ≤ 3 cm surrounded by lung or visceral pleura without evidence of invasion more proximal than the lobar bronchus.18 Patients with multifocal adenocarcinoma were excluded. The patient population comprised 113 patients with an average age of 69 years (range, 38-89 years) with a male-to-female ratio of 1:2.5. Clinical postsurgery follow-up was available for all patients, with an average follow-up period of 37.6 months (range, 0-103 months). The study was conducted according to HIPAA regulations and approved by our institutional review board.
One hundred and thirteen cases were identified in a 6-year period (1999-2004). The histological slides for each case were reviewed and classified according to the criteria set forth by the WHO.4 In addition, the sections were graded according to the most predominant pattern as previously described.11 In summary, grade 1 are tumors with a predominantly bronchioalveolar carcinoma component (lepidic pattern), grade 2 are tumors with predominantly acinar or papillary patterns, and grade 3 are tumors with predominantly solid or micropapillary patterns.
Cytopathologic Criteria and Review
Cytological smears of the corresponding surgical cases were reviewed by 2 cytopathologists (A.L.M., K.R.G.), a cytopathology fellow (C.S.S.), and an experienced cytotechnologist (D.E.R.). The reviewers were blinded to the risk stratification grade previously assigned to the histological sections. The cytological criteria recorded for each set of slides were smear background (presence or absence of necrosis), presence of cell groups arranged in flat sheets or in 3-dimensional clusters (predominantly, mixed, focal or absent), presence of neoplastic cells singly dispersed on the smears or liquid-based slide (single cells), presence of giant tumor cells (defined as neoplastic cell nuclei larger than the aggregate size of 15 lymphocytes), nuclear size (>5 lymphocytes or ≤5 lymphocytes), nuclear size variation (uniform size or pleomorphic), nuclear contours (smooth, convoluted, or nuclear grooves), relative prominence of nucleoli (conspicuously observed at 10× objective or inconspicuous; presence of macronucleoli was also recorded), quality of chromatin (fine and granular or coarse, as observed at 20× objective), and the presence of nuclear pseudoinclusions. Each feature was considered present if seen in the majority of neoplastic cells on the slides. When evaluating architectural features such as the presence of 3-dimensional cell clusters or flat sheets, the cases were assessed for whether a feature was represented predominantly, mixed, focal, or absent. Subsequently, for statistical analysis the groups were consolidated to “predominantly/mixed” versus “focal/absent.”
These cytological criteria were recorded separately for smears and Thin Prep slides. The air-dried smears were stained with Diff-Quik, whereas ethanol-fixed smears were stained with Papanicolaou and H&E stain. All divergent reviews were resolved by consensus through multihead microscope review.
Univariate associations between individual nuclear features and histological grading were tested using the Fisher exact test.
Disease-free survival (DFS) was defined as time from surgery to recurrence or death, whichever came first. Patients who did not recur or die during the duration of the study were censored at the time of last follow-up. Individual nuclear features were compared with respect to their DFS using the log-rank test. Those features found to be significant at a level of α = 0.1 were candidates for inclusion in a multivariate Cox proportional hazards model, which then represented the base for building a cytological scoring system to predict the risk of DFS in this population of patients. To obtain a score that is user friendly and easy to apply in clinical practice, we simplified the multivariate model as follows: first, the parameter estimates were standardized (divided by their respective variance); second, the resulting coefficients were rounded to the nearest integer in order to derive the coefficients that were ultimately summed to generate the final cytological score.
Groups of patients defined by cytological score, histological grading, and combinations of the 2 were further compared with respect to their DFS using the log-rank test. In addition, we compared the association between cytological score and histological grading using the nonparametric Kruskal-Wallis test.
All statistical analyses were performed using SAS software (SAS Institute, Cary, NC).
Across the cytomorphologic features evaluated, no major difference was seen among the various stains and preparations (data not shown). The chromatin pattern and presence of nucleoli were more difficult to assess on Diff-Quik-stained smears. Because of these observations, the data presented in this article includes analysis of data obtained from Papanicolaou-stained ThinPrep slides only.
Association of Individual Cytological Criterion With Histological Grade and DFS
The recorded cytological criteria for each tumor were compared with the assigned histological grade of the corresponding excised adenocarcinoma in an attempt to identify which cytological criteria could be helpful in determining tumor grade. As presented in Table 1, smear background (clean versus necrotic, P = .019), cellular arrangements consisting of flat sheets of cells (predominant/mixed versus focal/absent, P < .001), cellular arrangements consisting of 3-dimensional clusters (predominant/mixed versus focal/absent, P = .013), the presence of giant tumor cells (P = .007), nuclear size (P = .018), and convoluted nuclear contour (P = .001) were significantly associated with the predominant histological grading.
Table 1. Association Between Cytological Features and Predominant Histological Grade
P values were calculated using Fisher's exact test.
Abbreviations: lym, lymphocyte nuclei.
Predominant or mixed
Predominant or mixed
Giant tumor cells
Large (>5 lym)
Small (≤5 lym)
Next, we evaluated the cytological criteria for potential association with DFS. As indicated in Table 2, small nuclear size (median DFS, 7.7 years compared with 5.5 years for patients with large nuclear size; P = .005) and fine chromatin (median DFS, not reached; P = .045) were significantly associated with improved DFS. In addition, convoluted nuclear contours (P = .050) and predominant/mixed flat sheets (P = .073) exhibited trends toward better DFS.
Table 2. Association Between Cytological Features and Disease-Free Survival (DFS)
Abbreviations: DFS, disease-free survival; lym, lymphocyte nuclei; NR, not reached.
Predominant or mixed
Predominant or mixed
Giant tumor cells
Large (>5 lym)
Small (≤5 lym)
The 4 variables (nuclear size, chromatin quality, nuclear contours, and predominant/mixed flat sheets) associated with DFS at a significance level of α = 0.1 were included in a multivariate model in order to generate a cytological score easy to use in clinical practice. Variables (eg, smear background and giant cells) with poor performance in the DFS analysis were excluded from the multivariate model.
In the presence of the other variables, predominant/mixed flat sheets had a markedly reduced effect on the risk of DFS and therefore were excluded from the model. The updated multivariate model is presented in Table 3; it represents the basis of the nuclear grading system involving 3 nuclear features: nuclear contour, size, and chromatin quality. According to this score, a tumor receives 1 point if the nuclear contours are convoluted and 2 points for any of the following 2 occurrences: large nuclear size (larger than 5 lymphocytes) or dark, granular, or coarse chromatin. Points are assigned if the features are present in the majority of tumor cells. The final score equals the sum of all points and can range from 0 points (if the patient has no morphologic risk factors) to 5 points (if a patient has all 3 of the morphologic risk factors). The application of the scoring system is illustrated in Figure 1 using cases selected from the study.
Table 3. Multivariate Model for the Association Between Disease-Free Survival and Specific Nuclear Features (Overall Cytological Score Equals Sum of the Score Coefficients)
Hazard Ratio (95% CI)
Abbreviations: lym, lymphocyte nuclei.
Small (≤5 lym)
Large (>5 lym)
Correlation of Cytological Scores and DFS
Figure 2a presents DFS curves by individual nuclear score from all patients studied. Based on this, we further simplified the scoring system such that 3 groups were identified: tumors with low risk of recurrence or death, corresponding to scores of 0-1; tumors with intermediate risk of recurrence, corresponding to scores of 2-3; and tumors with high risk of recurrence, corresponding to scores of 4-5. DFS at 5 years was 100% in the low-risk group, 78% in the intermediate-risk group (95% CI, 62%-97%), and 55% in the high-risk group (95% CI, 39%-79%); P = .008 (Fig. 2b).
Correlation of Cytological and Histological Scores of Pulmonary Adenocarcinomas
As illustrated in Figure 3, there was a significant association between cytological score and histological grading, as defined by the predominant architectural pattern of the excised tumor (P < .001). Histologically classified well-differentiated adenocarcinomas clustered with low cytological scores (median score, 0; range, 0-2), whereas poorly differentiated tumors had higher cytological scores (median score, 5; range, 2-5). In contrast, the histological grade 2 tumors showed a larger distribution of scores ranging across cytological scores 0-5 (median score, 3).
Correlation of the cytological scores with a histological scoring system based on the 2 most predominant histological patterns11 was also attempted. However, the association of cytological scores and histological risk categories was less pronounced (data not shown).
We investigated whether the cytological score can bring further insight to the relationship between morphological classification and DFS. A stratification based on the most predominant histological pattern identified 10 patients with low risk of recurrence (DFS at 5 years, 100%), 84 with intermediate risk of recurrence (DFS at 5 years, 78%), and 19 patients with high risk of recurrence or death (DFS at 5 years, 44%); see Figure 4a. The high-risk group had a significantly higher probability of recurrence or death, compared with the low-risk (P = .044) and intermediate-risk (P = .010) groups, but the low- and intermediate-risk groups were not statistically different (P = .19).
We further stratified the group with intermediate histological risk based on cytological score: favorable (score, 0-3) versus unfavorable (score, 4-5) cytological prognosis (Fig. 4b), resulting in 2 groups with different outcomes: DFS at 5 years was 91% for the intermediate histology/favorable cytological prognosis group compared with 54% for the intermediate histology/unfavorable cytological prognosis group (P = .004). The group with intermediate histology/unfavorable cytological prognosis was no different in its DFS outcomes compared with the group with high histological risk.
In this study, we have demonstrated that objective tumor grading with prognostic significance in pulmonary adenocarcinoma is possible in cytological material. The proposed grading system is easy to perform because it is based on only 3 criteria: nuclear size, chromatin pattern, and nuclear contours. These parameters are familiar to and commonly used by cytopathologists when describing malignant tumor cells. Moreover, nuclear size and atypia/contour abnormalities are components of commonly used histological grading systems, such as those for breast and renal cortical carcinomas.19, 20 In contrast to prior studies on surgical resections for lung cancer, which found that nuclear size variability and giant nuclei correlated with a worse prognosis,21 we did not find these 2 characteristics to correlate with DFS, and they were not included in the grading system. We also did not assess for mitotic activity because, in general, this parameter is difficult to assess in cytological specimens and is too labor intensive. Importantly, mitotic figure counting in smears has never been validated against a histological mitotic figure counting.
Although we found a correlation between our cytological score and a histological score based on the predominant histological architecture pattern, cytoarchitectural features were not used ultimately in the scoring system because they were not significantly associated with DFS. This finding is in contrast to the histological scoring system, which accounts for the presence of mixed histological patterns. Thus, our analysis may lend further support for the observation that the histological architectural heterogeneity seen in pulmonary adenocarcinoma cannot be ascertained in samples obtained by fine-needle aspiration biopsy,16 and thus cytoarchitecture cannot be used as a surrogate for defining the level of histological differentiation. Instead, our study suggests that nuclear features may correlate with histological differentiation, but they perform best in a model with 2 tiers, unlike the 3 tiers typically used in describing histological differentiation (well, moderate, and poor). The proposed nuclear grade best separates the histological well-differentiated (lepidic pattern predominant) from poorly differentiated (solid and micropapillary patterns predominant) adenocarcinomas. When nuclear grading by cytology is applied to the histology-defined intermediate-grade category (acinar and papillary patterns predominant), we could further divide these tumors into 2 groups: low and high risk of recurrence or death of disease.
We acknowledge that this study was performed exclusively on stage I adenocarcinomas, where the majority of tumors undergo surgical resection and it could be argued that histological grading can be achieved in the histological resection. Studying nuclear grade in this limited population was necessary to test its ability as a prognostic marker in a population subject to uniform therapeutic guidelines. Extension of nuclear grading for prognosis to the majority of patients for whom cytology specimens are the only diagnostic material is the logical next step.
We have not challenged the reproducibility of this system, which is essential for the validation of any potentially useful grading system. The features selected for the grading system such as nuclear size are not overly prone to subjective analysis. In prior studies evaluating grading systems for breast carcinoma, nuclear grade has been shown to be reproducible.22 Our chosen internal standard for estimating relative nuclear size, the lymphocyte nucleus, is ubiquitous and should be easy to use.
We developed our nuclear grading system using Papanicolaou-stained Thinprep slides, but we observed similarly good preservation of chromatin and nuclear features in well-preserved Papanicolaou-stained smears. In practice, a possible limitation is when alcohol-fixed slides show air-drying artifact or if there is availability of slides stained only with Romanovsky stain (Diff-Quik). In both these situations, chromatin patterns and nuclear contours are more difficult to evaluate. However, with the current trend toward greater utilization of liquid-based cytology, this probably will not represent a real problem, and in laboratories where liquid-based cytology is not feasible, it is possible to use Papanicolaou-stained smears.
Overall, our goal in developing a grading system is to provide prognostic information that may affect therapeutic decision making, clinical follow-up interval, and clinical management. Further studies to validate the prognostic value of this grading system in patients with stage I lung cancers that have other medical conditions that prevent them from undergoing surgery may be especially useful.
In summary, we identified several nuclear features that can be easily tested in cytological specimens and can be combined into a grading system that (1) correlates well with histological grading and (2) shows promise as a prognostic tool in stage I lung adenocarcinoma.