Pulmonary carcinoids are classified as typical or atypical by assessing necrosis and mitoses, which usually cannot be adequately assessed on small biopsies. Ki-67 is not currently used to grade pulmonary carcinoids, but it may be helpful to determine preliminary grade in biopsies. However, the rate at which Ki-67 could underestimate or overestimate grade on small biopsies has not been well studied.
To compare Ki-67 labeling obtained on small biopsies to subsequent resection.
Ki-67 was performed on paired biopsy and resection specimens from 55 patients. Slides were scanned using Aperio ScanScope. Labeling index was determined using automated hot spot and tumor tracing methods.
The study included 41 typical and 14 atypical carcinoids. Atypical carcinoids were larger and had more distant metastases. Death from disease occurred in 3 patients (all had atypical carcinoids). Median hot spot Ki-67 labeling index was greater in resection compared with biopsy by 0.7% (P = .02). Median tumor tracing Ki-67 was lower in resection compared with biopsy by 0.5% (P < .001). Receiver-operating characteristic analysis showed similar hot spot Ki-67 cutoffs to predict atypical histology (3.5% for biopsy, 3.6% for resection; area under the curve [AUC], 0.75 and 0.74, respectively). Different optimal cutoffs were needed for tracing method based on biopsy (2.1%; AUC, 0.75) compared with resection (1.0%; AUC, 0.67).
Hot spot Ki-67 tends to underestimate grade on small biopsies, whereas grade is overestimated by tumor tracing. Hot spot Ki-67 cutoff of 3.5% predicted atypical histology for both biopsy and resection. Different biopsy and resection cutoffs were necessary for tumor tracing, which would make clinical implementation more difficult.
Pulmonary neuroendocrine tumors are a heterogeneous group of neoplasms that includes carcinoid tumors, which have somewhat unpredictable but relatively indolent behavior, as well as small cell and large cell neuroendocrine carcinomas, which are usually very aggressive. Per World Health Organization (WHO) criteria, pulmonary carcinoids are classified as typical or atypical by mitotic count and assessing for necrosis.1 Typical carcinoid tumors (TCTs) have ≤1 mitotic figure per 2 mm2 and absent necrosis, whereas atypical carcinoid tumors (ACTs) have 2 to 10 mitoses per 2 mm2 and/or presence of necrosis. This distinction is important because ACTs have a poorer prognosis, with increased risk of recurrence, and nodal and distant metastasis.2–7 Sampling issues usually preclude definitive classification as typical or atypical based on small biopsies, unless there are obvious atypical features.
Ki-67 labeling index is incorporated into the 3-tiered grading system for neuroendocrine neoplasms of the gastrointestinal and pancreaticobiliary tract (grade 1 tumors have Ki-67 ≤3% by hot spot method; grade 2 tumors have Ki-67 between 3% and 20%; grade 3 neoplasms have Ki-67 >20%).8 Although Ki-67 is not currently used in the WHO grading scheme for pulmonary carcinoids, many studies have shown that ACTs have increased Ki-67, and that increased Ki-67 is generally associated with poorer prognosis.2,9–24 However, the use of Ki-67 labeling index in the evaluation of pulmonary carcinoid tumors is controversial and not standardized, because there are conflicting data regarding the added value to the current WHO system. A Ki-67 cutoff value to divide TCT from ACT also has not been clearly defined, which is likely in part due to the highly variable methods used to determine Ki-67 index, and the wide range of cutoffs that have been proposed. One important technical consideration is the method used to establish Ki-67 labeling index, which has important implications for any cutoff that might be established. Ki-67 may be assessed by concentrating on areas of highest activity, the “hot spot” method, or by counting all available tumor cells on the stain, the “tumor tracing method.” The hot spot method has the advantage of identifying the most proliferative area, but it could underestimate Ki-67 on a small biopsy because a “hotter spot” could be missed because of sampling. Tumor tracing has the advantage of providing an “average” Ki-67 throughout the tumor, and is therefore a very stable measure, but it could overestimate Ki-67 on a small biopsy if an area of relatively high proliferation is sampled. As might be expected, values established by hot spot counting are routinely higher than tumor tracing values, and thus any proposed cutoffs would be expected to be quite different depending on the method used.
Despite these shortcomings, Ki-67 has particular appeal in the evaluation of small biopsies of pulmonary carcinoids, where it may be helpful to determine preliminary grade, or to estimate grade when surgical excision is not clinically feasible (high stage, comorbidities, etc). The rate at which Ki-67 could underestimate or overestimate grade on small biopsies is not fully known. One study using manual hot spot Ki-67 index showed good concordance between biopsies and resections.25 The goal of this study is to compare automated Ki-67 index of preoperative biopsies to subsequent resection specimens, in order to determine how often they are different. Additionally, we sought to determine whether hot spot or tumor tracing method might provide the best preliminary grading information.
MATERIALS AND METHODS
Institutional pathology archives (1997–2017) were searched for patients with biopsy showing pulmonary carcinoid and subsequent resection. Diagnoses were confirmed and agreed upon by 2 pulmonary pathologists, with classification using 2015 WHO criteria applied to the resected tumor. Ki-67 immunostaining (clone MIB1, Dako, Carpinteria, California) was performed on formalin-fixed, paraffin-embedded tissue sections from both biopsy/cell block and resection specimens (1 representative whole tissue section was stained for resection specimens to include any areas of increased mitotic activity, if applicable). Slides were scanned at ×20 magnification on the Aperio ScanScope AT Turbo brightfield instrument (Leica Biosystems, Buffalo Grove, Illinois) at a resolution of 0.50 microns per pixel. The images were 24-bit contiguous standard pyramid tiled TIFFs compressed via JPEG with a quality setting of 70. Cases that had insufficient tumor on the Ki-67 stain (fewer than 300 cells) or complete lack of Ki-67 staining (lack of internal positive control) were excluded. The biopsies from the final study group included 38 transbronchial biopsies, 11 needle core biopsies, and 6 cytology cell blocks.
Ki-67 labeling index was determined via digital image analysis by an experienced technologist, using both hot spot and tumor tracing methods. Aperio ImageScope Software (Leica Biosystems) was used. Tumor areas were circled by a pathologist on corresponding hematoxylin-eosin slides for reference, to ensure only tumor areas were scored. For the tumor tracing method, using the reference hematoxylin-eosin, a minimum 85% of tumor was traced with a digital pen tool to indicate the region of analysis. Care was taken, either in the tracing process or by using a negative pen tool, to eliminate tissue folds and avoid staining artifacts. For hot spot method, a second annotation layer was added to the image and 10 fixed-size boxes were placed on hottest-staining region. Each box was 107 × 106 μm, for a total analyzed hot spot area of 0.11 mm2. A nuclear algorithm was used to analyze the selected tissue. Manual assessment of Ki-67 was performed in 10% of cases by a pathologist as a quality control measure, which included verification of placement of hot spot boxes, confirmation of excluded nontumor areas/cells, and counting of at least 5 high-power fields for a manual estimate of Ki-67 both in hot spot areas and at random; no discrepancies were identified from the automated reads.
Follow-up information was obtained from clinical records. Descriptive statistics were employed to summarize the data (frequencies and percentages, means and SDs, or medians and ranges, as appropriate). Patient characteristics were compared between those with TCT versus ACT with Fisher exact tests (categoric variables) or Wilcoxon rank sum tests (ordinal or continuous variables). Within each of the 2 methods (hot spot, tracing), Ki-67 was compared between paired biopsy and resection samples with Wilcoxon signed rank tests, and the median differences were reported. Additionally, the Pearson correlation coefficient was used to further characterize the association between the methods. For each method and specimen type, receiver-operating characteristic (ROC) analysis was performed to identify a Ki-67 cutoff to jointly optimize the sensitivity and specificity for predicting ACT, and the area under the curve (AUC) was reported. P values less than .05 were considered statistically significant. Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, North Carolina) and R.26
RESULTS
Demographic, clinical, and pathologic information is summarized in Table 1. Based on WHO criteria, 41 tumors were TCT (74.5%; Figure 1, A) and 14 were ACT (25.5%; Figure 1, B and C). Sex and age were similar between groups. The ACTs were significantly larger than TCTs (mean, 4.0 versus 2.6 cm; P = .03). The rate of lymph node metastasis at the time of surgery was not significantly different (8 of 41 [19.5%] for TCT; 4 of 14 [28.6%] for ACT; P = .48). Three patients had distant metastases at the time of diagnosis, all of whom had ACT metastatic to the liver and had concomitant mediastinal lymph node metastases. The rate of distant metastasis at the time of diagnosis was significantly higher in ACT (3 of 14 [21.4%] for ACT versus 0 of 41 for TCT; P = .01). Of 55 patients, 49 were treated with surgery alone (89.1%). Adjuvant treatment regimens included radiotherapy and chemotherapy (cisplatin, etoposide) in 1 patient with ACT; salvage cisplatin and etoposide in 1 patient with ACT; and salvage sandostatin in 2 patients with ACT and 2 patients with TCT (1 with metastatic disease, 1 with diffuse idiopathic pulmonary neuroendocrine cell hyperplasia). Median follow-up was 48.8 months (range, 11 days to 191 months). Recurrence/metastases occurred after surgery in 5 of 15 ACT patients (2 of whom also had liver metastases at diagnosis), and 1 of 40 TCT patients. An additional 3 patients with TCT had either multifocal carcinoid tumor (1 patient with 2 endobronchial TCTs in separate lobes) or multiple carcinoid tumors in the setting of diffuse idiopathic pulmonary neuroendocrine cell hyperplasia. Sites of late metastasis included liver (4), bone (2), pericardium/heart (1), lung (1), eye (choroid) and orbit (1), lymph nodes (1), spleen (1) and soft tissue (1). Three patients died of disease (17, 73, and 75 months after diagnosis), all of whom had ACT.
![Figure 1. Typical carcinoid tumor (A), showing classic trabecular growth and rosette formation, with finely stippled chromatin. No necrosis or increased mitotic activity is observed. Atypical carcinoid tumor, showing increased mitotic activity (B, arrows) and focal necrosis (C). Representative levels of Ki-67 labeling index are pictured, including <1% (D), 2% (E), and >10% (F) (hematoxylin-eosin, original magnification ×200 [A through C]; original magnification ×200 [D through F]).](https://allen.silverchair-cdn.com/allen/content_public/journal/aplm/144/8/10.5858_arpa.2019-0374-oa/1/m_i1543-2165-144-8-982-f01.png?Expires=1753804360&Signature=uvIRXfQPRGxiPn9tA1ngNFKoJZ4T3tDix3yAQO38vzEjwOxurEMeiwcjNJ8xE5leIt1PQ0MGrjrmYHMdrQ1ReZbbOkGdygwsc20WScgDmHb5xjAu2SEykzXXpttOcF6X99FAsUUpQGn0H0GoYCcx1utZmtGhAb7M2iHK2HwPcw2TdEOcnqZs38pli6sM7oqrQMpp~yu2q6z0kAAqxbxd9BgvSHnUvUO6cTMOlfNcMHheauCl6Fgc0eZmonkScMKHpQMXcF8YxmQBUpM-qLU~XJbUowmEugygmwW-G5MgY85HthlrJ8OuA4ibKWqVJHX7UyL2A7w38bPz5rwkfqe7eg__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Typical carcinoid tumor (A), showing classic trabecular growth and rosette formation, with finely stippled chromatin. No necrosis or increased mitotic activity is observed. Atypical carcinoid tumor, showing increased mitotic activity (B, arrows) and focal necrosis (C). Representative levels of Ki-67 labeling index are pictured, including <1% (D), 2% (E), and >10% (F) (hematoxylin-eosin, original magnification ×200 [A through C]; original magnification ×200 [D through F]).
Typical carcinoid tumor (A), showing classic trabecular growth and rosette formation, with finely stippled chromatin. No necrosis or increased mitotic activity is observed. Atypical carcinoid tumor, showing increased mitotic activity (B, arrows) and focal necrosis (C). Representative levels of Ki-67 labeling index are pictured, including <1% (D), 2% (E), and >10% (F) (hematoxylin-eosin, original magnification ×200 [A through C]; original magnification ×200 [D through F]).
Median and range of Ki-67 results are summarized in Table 2, including comparison between biopsy and resection specimens (Figure 1, D through F), and comparison between hot spot and tumor tracing methods. As expected, based on the method of Ki-67 evaluation, the median value of the hot spot method was greater than the tumor tracing method for biopsy and resection specimens. The differences observed between biopsy and resection values by tumor tracing (Figure 2, A) and hot spot (Figure 2, B) methods are summarized in Table 3. Tumor tracing Ki-67 was lower on the resection compared with biopsy by a median of 0.5%, which was significant (P < .001), and this significance was maintained when TCT (median, 0.4%; P < .001) and ACT (median, 0.7%; P = .009) were considered separately. The difference between biopsy and resection for tumor tracing Ki-67 was more pronounced in ACT (0.7%) versus TCT (0.4%), but this was not significant (P = .25). The hot spot Ki-67 index was greater in the resection compared with biopsy by a median of 0.7%, which was statistically significant (P = .02), although the difference between biopsy and resection did not reach significance when TCT (P = .11) and ACT (P = .10) were considered separately. The median difference was greater in ACT (1.3%) compared with TCT (0.7%), but this difference was not significant (P = .28). The correlation between Ki-67 values determined by hot spot and tumor tracing methods was stronger in biopsy compared with resection (Pearson correlation coefficient 0.88 for biopsies, 0.70 for resections). The Pearson correlation coefficient between biopsy and resection Ki-67 for the hot spot method was 0.65, whereas it was 0.58 for the tumor tracing method.
Correlation between Ki-67 labeling index values on biopsy and resection, using tumor tracing (A) and hot spot (B) methods.
Correlation between Ki-67 labeling index values on biopsy and resection, using tumor tracing (A) and hot spot (B) methods.
The ROC analysis was performed to determine what Ki-67 cutoff best discriminated between histologic classification of TCT versus ACT (Figure 3). For the hot spot method, ROC analysis of both biopsy and resection showed a similar cutoff to optimize sensitivity and specificity (3.5% for biopsy [Figure 3, C] and 3.6% for resection [Figure 3, D]; AUCs of 0.75 and 0.74, respectively). However, the ROC analysis of biopsy versus resection gave quite different optimal cutoffs for the tracing method, with a proposed cutoff of 2.1% based on the biopsy data (AUC, 0.75; Figure 3, A), but a lower cutoff of 1.03% generated based on resection data (AUC, 0.67, Figure 3, B). The distribution of Ki-67 labeling index for TCT and ACT around the cutoffs determined by ROC analysis is illustrated in Figure 4. Ki-67 labeling index determined on resection specimens of tumors near the mitotic rate cutoff between TCT and ACT is summarized in Table 4, which includes TCT cases with 1 mitotic figure per 2 mm2 (n = 14) and ACT with 2 to 3 mitoses per 2 mm2 (n = 9); although the median Ki-67 was higher in ACT versus TCT for both methods, considerable overlap was observed in Ki-67 values for these cases.
Receiver-operating characteristic analysis of Ki-67 labeling index cutoffs for predicting atypical carcinoid tumor histology, based on biopsy using tumor tracing (A), resection using tumor tracing (B), biopsy using hot spot method (C), and resection using hot spot method (D).
Receiver-operating characteristic analysis of Ki-67 labeling index cutoffs for predicting atypical carcinoid tumor histology, based on biopsy using tumor tracing (A), resection using tumor tracing (B), biopsy using hot spot method (C), and resection using hot spot method (D).
Distribution of Ki-67 labeling values in typical and atypical pulmonary carcinoid tumors. Cutoff values determined by receiver-operating characteristic analysis are illustrated with red lines. Pictured data include biopsy samples using tumor tracing (A), resection specimens using tumor tracing (B), biopsy samples using hot spot method (C), and resection specimens using hot spot method (D).
Distribution of Ki-67 labeling values in typical and atypical pulmonary carcinoid tumors. Cutoff values determined by receiver-operating characteristic analysis are illustrated with red lines. Pictured data include biopsy samples using tumor tracing (A), resection specimens using tumor tracing (B), biopsy samples using hot spot method (C), and resection specimens using hot spot method (D).
Of the 7 patients who experienced distant metastasis, the range of hot spot Ki-67 on the biopsy was 2.7% to 7.2%, with only 1 patient below 3.5%. The range of tumor tracing Ki-67 in the biopsies of these patients was 1.1% to 3.7%, with only 1 patient below 2.1%. On subsequent resection specimens, the range of hot spot Ki-67 was 0.9% to 12.6%, with 1 patient below 3.6%. The range of tumor tracing Ki-67 on the resections was 0.2% to 5.0%, with 3 tumors showing values 1.0% or less. Three patients with ACT had hot spot Ki-67 less than 3.5% on resection; 1 of these patients (hot spot Ki-67 on resection 0.9%; hot spot Ki-67 on biopsy 3.9%) experienced distant metastases and died of disease, whereas the other 2 patients did not experience recurrence or distant metastases. Six patients with ACT had tumor tracing Ki-67 less than 1.03%; 1 had distant metastases and died of disease, 1 had distant metastases with unknown outcome, 1 had lymph node metastases with no other distant disease, and 3 had no recurrence or metastases.
DISCUSSION
Ki-67 labeling index is a marker of cell proliferative activity, similar to mitotic count. However, it is more sensitive than mitotic count because it detects all cells in the G1, G2, and S phases of the cell cycle as well as cells in mitosis.27 Ki-67 labeling index has been incorporated into routine practice for grading of gastrointestinal neuroendocrine tumors.8 Although manual Ki-67 reading can be tedious because of the high cell count required for accurate assessment, automated Ki-67 labeling index has been widely adopted in the grading of gastrointestinal neuroendocrine tumors, where it has shown to correlate very well with manual counting.8,28 Automated counting has also been shown to have good to excellent concordance with manual counts in lung neuroendocrine tumors.16,19,29,30 Ki-67 may have the additional advantage of higher interobserver agreement than mitotic count.29 However, data on the value of adding Ki-67 labeling index to evaluation of lung carcinoid tumors have been mixed. Some data support that it does not add additional prognostic value in addition to the WHO grading system,18 or that Ki-67 is of borderline significance, with stage and histologic typing providing more clearly significant information.13,18–20 Conversely, other studies have found that Ki-67 is an independent predictor of outcome, whereas mitotic count and in some instances even stage and histologic type are not; some of these studies have proposed that Ki-67 may be a better predictor of true tumor biologic potential.2,4,5,11,24,31 The best grading system may incorporate both WHO classification and Ki-67. In 1 large study, incorporation of Ki-67, necrosis, and mitotic count provided very good separation of survival in a 3-tiered grading system.16 However, another study showed that this system did not outperform the current WHO classification.19 Unfortunately, we had too few events in our cohort to perform a robust statistical evaluation of the prognostic value of Ki-67 values independent of WHO classification. Our follow-up data are somewhat limited in length as well, which limits our ability to perform robust survival analysis, because pulmonary carcinoid tumors have an indolent course that requires very long follow-up for optimal assessment of survival.
Increased Ki-67 labeling index has been associated with other markers of aggressive behavior, including increased apoptotic index, aneuploidy, BCL2 expression, and mutant p53 expression.9,10,21 Ki-67 may be an indicator of adverse prognosis in TCT when greater than 5%, where it is associated with higher stage.18,32 Atypical carcinoid tumors with Ki-67 values higher than 10% or mitotic rate of 6 or more per 10 high-power fields also appear to have a particularly dismal prognosis.3,5,20,33 Interestingly, metastatic carcinoid tumors tend to show significantly higher Ki-67 values compared with the primary tumor.18
Ki-67 index is generally very dichotomous between low/intermediate-grade and high-grade pulmonary neuroendocrine neoplasms. It can be useful to avoid the diagnostic pitfall of overcalling small cell carcinoma in the setting of a crushed carcinoid tumor, where confident morphologic distinction may be impossible.34,35 High Ki-67 index above 20% has also been proposed as an indicator of high-grade neuroendocrine carcinoma on small biopsies.25 However, the discrimination between TCT and ACT is unfortunately much more difficult. Sensitivity and specificity are not particularly robust and discriminatory capacity is only moderate for Ki-67 in predicting TCT versus ACT.16,25 This is also true of mitotic count, where the distinction between low/intermediate and high grade is often clear, whereas the distinction between low and intermediate grade requires more rigorous evaluation and still does not provide the same discriminatory capacity.16 Therefore, it is not surprising that there is substantial interobserver disagreement in the application of the WHO classification of pulmonary neuroendocrine tumors, especially for the diagnosis of ACT.36 Ki-67 has been suggested as a way to increase diagnostic agreement.36
Confident pathologic grading of pulmonary carcinoid tumors using the current WHO system usually requires surgical excision, because necrosis is patchy and can be missed because of sampling, and mitotic activity can show “hot spots” of increased activity that may not be represented in a biopsy sample.25 Therefore, the increased sensitivity for proliferative activity detected by Ki-67 is particularly attractive as a potential tool to predict preliminary grade in pulmonary carcinoid tumors, but its utility in this setting has not been well studied. One study of Ki-67 on tissue microarrays, which mimic the scenario of a small biopsy, did show Ki-67 index higher than 2.5% was associated with worse prognosis.4 Another study has looked at manual hot spot Ki-67 correlation between biopsy and resection specimens and has found good correlation.25 In our study, we found a small but statistically significant difference in Ki-67 between biopsy and resection, when using both hot spot and tumor tracing methods. Hot spot Ki-67 was lower in the biopsy, which could underestimate preliminary grade, and the Ki-67 determined by tumor tracing was higher in the biopsy compared with resection, which could overestimate preliminary grade. These limitations of determining Ki-67 labeling index on biopsy specimens largely make sense when considering each method. Determination of hot spot Ki-67 is limited in small samples, because “hotter” areas of the tumor may exist in unsampled regions, hence the tendency to underestimation of grade. It is possible that the increased hot spot values in the resections could be at least partially due to the biopsy procedure itself, with induction of a hot spot due to tissue damage and regeneration (ie, perhaps the biopsy actually provides better assessment, because the resection may overestimate grade because of “artificial” hot spots); however, it is also logical that true hot spots would be better detected when a whole tumor section is available for analysis. The number of cells counted by tumor tracing in a resection specimen is generally very large, which leads to a stable estimate of overall tumor proliferative activity. However, this method would be predicted to be susceptible to regional variability in proliferative activity when only a small area of the tumor is sampled via biopsy. The reason for the propensity to have a higher tumor tracing Ki-67 in biopsy is not entirely clear; one could imagine that the surface of a tumor accessible by bronchoscopy could have reactive changes that would lead to relatively higher proliferative activity, but it seems there must also be other factors at play.
The cutoff value of Ki-67 that might be useful to divide TCT from ACT is not standardized. Proposed cutoffs have varied widely in different studies, which is not surprising given the wide variety of methods that have been employed to determine this index (number of cells counted, manual versus automated methods, differing definitions of hot spots, hot spot counting versus random counting, different statistical methods to determine the “best” cutoff, etc). Although 1 study found that there was no overlap between the Ki-67 index observed between TCT and ACT,31 we found significant overlap in our study, which has been observed in other studies as well.14,18–20 Proposed Ki-67 cutoffs to divide TCT from ACT in the literature include 1%,19,22 2%,5 2.5%,12,19 4%,2,14,16,20,24 5%,18,36 7%,31 and 7.5%.17 Although it would be ideal to use the same Ki-67 cutoff as the gastrointestinal tract to increase uniformity and reduce confusion, there is at least some evidence that this provides suboptimal discrimination of pulmonary TCT versus ACT, with less separation of the survival curves than the WHO system.16 Although disappointing, it is not necessarily surprising that site-specific factors may impact the Ki-67 index, which could include changes of prior biopsy, exposure to the environment within the bronchus, and potential traumatization due to cough, etc. In our study, a Ki-67 hot spot cutoff value of 3.5% provided optimal sensitivity and specificity for distinguishing TCT from ACT, which was applicable to both biopsy and resection. Tumor tracing method required different cutoff values in biopsy (2.1%) versus resection (1.03%), which is a limitation of this method. The AUC was also slightly higher for the hot spot method when resected tumors were considered. Therefore, the hot spot method may be better suited to distinguish pulmonary TCT from ACT.
In summary, Ki-67 labeling index in pulmonary carcinoid tumors determined on biopsy versus subsequent resection shows a small but statistically significant difference using automated hot spot and tumor tracing methods. Hot spot method has a tendency toward undergrading based on biopsy, whereas tumor tracing has the opposite tendency toward overgrading. Tumor tracing has the drawback of requiring different cutoff values for biopsy versus resection. These potential shortcomings should be kept in mind when deciding when and how to use Ki-67 labeling index in clinical practice.