Introduction
Pancreatic neuroendocrine tumors (PanNETs) are rare and account for between 1 % and
5 % of pancreatic mass lesions [1]
[2]. Most PanNETs are low grade and run an indolent course but recent data suggest a
significant increase in incidence [1]
[3]. PanNETs can be classified in a number of ways including functional status and may
present with clinical symptoms dependent on the hormone being secreted; however, about
40 % of PanNETs are non-functional [2]. Management and subsequent prognosis of PanNETs is guided by the Ki67 proliferative
index (PI) which has been validated in a number of studies and incorporated in to
international guidelines [4]
[5]
[6]
[7]. This is particularly important in patients with non-functioning PanNETs less than
2cm as the risk:benefit equation of surgery versus observation is harder to judge.
In these circumstances, Ki67 PI is crucial in guiding discussion between clinicians
and patients as to the most suitable management plan [5].
Endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) has become the procedure
of choice for obtaining tissue from pancreatic lesions. With respect to PanNETs, EUS-FNA
has been used to provide cytological material for pathological analysis for many years
[8]. Previous studies have shown that EUS-FNA is useful in diagnosing PanNET but the
ability to provide an estimate of Ki67 PI appears variable [9]
[10]
[11]
[12]
[13]. This may be due to a number of factors including lack of cellular cohesion in cytological
preparations and adequacy of material for sufficient staining of relevant cells [11]
[13]. Several studies have examined the role of EUS-FNA in patients with PanNETs but
there are important limitations in each of them [8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]. The main limitation in most of these studies is the small numbers of patients included,
retrospective nature and lack of a gold standard used for comparison i. e. surgery
[10]
[11]
[13]
[14]
[16]
[17]. EUS-FNA appears to underestimate Ki67 PI, leading to under-grading of PanNETs with
some studies showing EUS-FNA samples labelled as grade 1 being upgraded to grade 2
on surgical resection histology [16]
[17]. The “holy grail” would be to obtain histological quality samples to provide the
pathologist with sufficient material to make a firm diagnosis of PanNET, accurately
grade the lesion, and subtype using immunohistochemistry. Recently, EUS-guided fine-needle
biopsy (EUS-FNB) has become available with several needle types designed to take core
samples and preserve tissue architecture for histological assessment [18]
[19]
[20]. Some studies have shown improved accuracy for tissue sampling via EUS with some
core needle types [18]
[20] although a recent meta-analysis showed no improvement in sample adequacy or diagnostic
accuracy for one type of core needle [21]. Despite this there are few studies comparing EUS-FNA with EUS-FNB particularly
in circumstances where additional information would be crucial to patient management
such as those with suspected PanNETs < 2 cm. A pilot study showed fewer passes were
required for diagnostic material when using an FNB needle compared to an FNA needle,
however, no assessment was made of the ability to determine Ki67 PI and grade and
no gold standard was used [22]. The aim of this study was to assess whether Ki67 PI and grading can be more accurately
determined using FNB compared to FNA using surgical excision histology as the gold
standard. Subgroup analysis of different needle types was also performed.
Patients and methods
The Freeman Hospital is a tertiary referral center for hepatobiliary and pancreatic
diseases in the North East region of England providing services to approximately 3.5
million people. EUS is provided by three experienced pancreaticobiliary endosonographers
(JSL, MKN, KWO) with a minimum of 3000 EUS performed individually.
Retrospective analysis of all pancreatic pathology reports for neuroendocrine tumors
was performed for the period January 2009 to June 2017. The pathology database was
searched using SNOMED codes for all surgical resections of PanNET lesions that also
had preoperative EUS-guided sampling. Patient demographics, lesion size, and location
were noted. FNA and FNB reports were examined and Ki67 PI and grade recorded. Surgical
histology reports were examined and time from EUS to surgery, operation performed,
TNM stage, Ki67 and grade recorded and compared using correlation coefficient and
proportional analysis.
All EUS procedures were undertaken after obtaining informed patient consent. Hitachi
EUB-7500 or Preirus US workstations (Hitachi Medical Systems, Wellingborough, UK)
and Pentax linear echoendoscopes (Pentax, Slough, UK) were used to perform EUS. Sampling
of the lesion was performed under real-time EUS guidance using the techniques described
hereafter. All procedures were performed using the slow pull technique followed by
10 mL of suction during the passes and fanning of the needle throughout the lesion
to optimize tissue acquisition. The number of passes into the lesion was determined
by the individual endosonographer but a minimum of three passes was the target during
the study period. Rapid On Site Evaluation (ROSE) was available during the study period
but only applied to the FNA samples and we have previously shown no increase in diagnostic
yield in an established EUS service [23].
EUS-FNA technique
FNA was performed using either 22- or 25-gauge needles (Cook Medical, Limerick, Ireland)
and tissue from each pass expressed onto glass slides to create smear slides which
were air-dried and subsequently stained with Giemsa-based stains. The remaining tissue
was expressed into cytofix red solution (BD Surepath, Bioscience Healthcare, Nottingham,
UK) and used for Surepath Liquid Based Cytology (LBC). Any macroscopically intact
tissue fragments were processed as a cell block and stained with hematoxylin and eosin.
All slides were analyzed by experienced cytopathologists. Cytological diagnosis of
well-differentiated neuroendocrine tumor was based on standard features (such as mostly
dispersed cell population with eccentric cytoplasm, smooth nuclear outline with salt-and-pepper
chromatin, inconspicuous nucleoli and naked nuclei). Immunocytochemistry was performed
on the LBC or cell block (cell block used in preference if contained enough tumor
cells), with the neuroendocrine markers synaptophysin, chromogranin and CD56, plus
Ki67 when sufficient material was available. Ki67 was assessed by estimation or cell
counting and tumors were given a provisional grade.
EUS-FNB technique
EUS-FNB was performed using either ProCore needles (Cook Medical, Limerick, Ireland)
or Sharkcore needles (Covidien/Medtronic, Whiteley, UK). All passes were placed in
one container of neutral buffered formalin which was sent for standard histopathological
processing and assessment was by experienced histopathologists. Diagnosis of well
differentiated neuroendocrine tumor was made on the basis of histomorphological characteristics
(i. e. presence of tumor tissue fragments of loosely cohesive, fairly monotonous cells
with regular nuclei and homogenous/finely stippled chromatin) and immunohistochemical
confirmation through positive staining for pancytokeratin, synaptophysin and chromogranin
and/or CD56. In a proportion of cases specific hormones and cytokeratin 19 were performed
to provide additional prognostic and management information. All cases were stained
for Ki67 to assess proliferation index and inform histological grade. Ki67 PI was
determined by direct cell count under high magnification with or without aid of a
manual cell counter.
Outcome measures
The primary objective was to determine whether EUS-FNA or FNB samples correlate with
surgical resection histology samples with particular reference to Ki67 PI and grading
in all PanNETs.
The secondary objective was to assess the degree of mismatch between EUS-guided sampling
and surgical resection histology with reference to under- and over-grading. Subgroup
analysis on purely solid PanNETs and of the different FNB needle types was also performed.
Ki67 proliferation index and grading
Grade of PanNETs is based on mitotic activity and Ki67 PI. The WHO classification
2010 provides a three-tier system defined as follows: grade 1: < 2 mitoses/10 hpf
(2 mm2), Ki67 < 2 %; grade 2: 2 to 20 mitoses/10 hpf (2 mm2), Ki67 3 % to 20 %; grade 3: > 20 mitoses/10 hpf (2 mm2), Ki67 > 20 % [22]. The boundary between grade 1 and 2 PanNETs was changed in the UK in 2012: grade
1: < 2 mitoses/10 hpf (2 mm2), Ki67 ≤ 5 %; grade 2: 2 to 20 mitoses/10 hpf (2 mm2), Ki67 > 5 % to 20 % [24]. However, EUS-guided sampling and surgical histology samples were assessed using
the grading system used during the relevant clinical period. Therefore, the threshold
for the grade was compared at the relevant time and not reallocated according to the
most current guidelines.
Statistical analysis
Parametric data were described using mean and standard deviation whereas non-parametric
data were described using median and interquartile range. Proportional analysis was
performed using either Chi-square or Fishers exact test. To compare Ki67 PI in EUS-guided
samples and surgical resection histology, correlation coefficients were calculated
using Pearson’s r. Pearson’s r values were interpreted as 0 meaning no relationship, > + 0.30
as a weak positive relationship, > + 0.50 as a moderate positive relationship, > + 0.70
as a strong positive relationship and + 1.0 as a perfect positive relationship. To
assess agreement in tumor grading (categorical data) between EUS-guided samples and
surgical resection histology, Cohen’s Kappa was calculated. Kappa values were interpreted
as < 0 indicating no agreement, 0 to 0.20 as poor, 0.21 to 0.40 as fair, 0.41 to 0.60
as moderate, 0.61 to 0.80 as good and 0.81 to 1 as almost perfect agreement. Subgroup
analysis was performed for solid lesions, lesions < 2 cm and different needle types.
A P value < 0.05 was considered statistically significant. Data were analyzed using MedCalc
11.2.1.0 (MedCalc Software, Ostend, Belgium).
Ethical approval
As per United Kingdom National Health Service research ethics guidance, ethical approval
from an institutional review body was not required for this study. Institutional authorization
to hold a prospective patient database for use for quality improvement was obtained.
Written informed consent was obtained from all patients prior to the procedure.
Results
One hundred sixty-four patients were diagnosed with PanNET in our center over the
study period of whom 57 underwent surgical resection (mean age 55.6, 30 males, mean
size 28.3 mm (standard error ±2.3). Twenty-two lesions were located in the head, 10
lesions in the body and 25 lesions in the tail of the pancreas with no significant
difference in size between the locations (P = 0.13). 41 were solid, eight cystic and eight mixed morphology. Surgical procedures
included distal pancreatectomy with splenectomy (n = 25), spleen-preserving distal
pancreatectomy (n = 9), Whipples' procedure (n = 8), enucleation (n = 6), pylorus-preserving
pancreaticoduodenectomy (n = 5), surgical bypass and biopsy (n = 2), duodenectomy
(n = 1) and distal pancreatectomy with splenectomy, sleeve gastrectomy, left nephrectomy,
left adrenalectomy, splenic flexure colectomy and diaphragm resection (n = 1). On
surgical histology 33 lesions were grade 1, 22 were grade 2, one was grade 3 and one
was mixed neuroendocrine-acinar.
Thirty-five lesions underwent FNA which was performed using 22-gauge needles in 24
cases, 25-gauge needles in four cases and not stated in seven cases. Twenty-six lesions
underwent FNB (4 lesions underwent both FNA and FNB) of which 12 were by Procore needles
and 14 usiing Sharkcore needle. For Procore, a 22-gauge needle was used in six cases,
25 gauge needle in five cases and combined 25 and 19 gauge in one case. For Sharkcore,
a 22-gauge needle was used in 11 cases and a 25 gauge needle in three cases. [Table 1] shows the demographics of those undergoing EUS sampling with the only significant
difference found in distribution of lesion morphology between the FNA and FNB groups.
This was to be expected as an FNB needle would not be used for the puncture of a cystic
lesion.
Table 1
Demographics of PanNETs in those undergoing EUS-FNA and EUS-FNB.
|
Variable
|
FNA group
|
FNB group
|
P
|
|
N
|
35
|
26
|
–
|
|
Mean age (years)
|
55.2
|
56.2
|
0.80
|
|
Sex
|
|
|
0.61
|
|
|
20
|
13
|
|
|
|
15
|
13
|
|
|
Mean lesion size (mm + /- standard error)
|
25.7 (3.1)
|
32.5 (3.6)
|
0.16
|
|
Lesion location
|
|
|
0.39
|
|
|
11
|
12
|
|
|
|
6
|
5
|
|
|
|
19
|
9
|
|
|
Lesion morphology
|
|
|
0.015
|
|
|
21
|
23
|
|
|
|
8
|
0
|
|
|
|
6
|
3
|
|
|
Median number of needles passes (IQR)
|
2.5 (1.0 – 3.0)
|
3.0 (2.0 – 4.0)
|
0.14
|
|
Median number of days from EUS to surgery (IQR)
|
79.0 (50.5 – 125.5)
|
65.5 (48.0 – 99.0)
|
0.37
|
PanNET, pancreatic neuroendocrine tumor; EUS-FNA, endoscopic ultrasound-guided fine-needle
aspiration; EUS-FNB, endoscopic ultrasound-guided fine-needle biopsy; IQR, interquartile
range.
Both cytology and histopathology, confirmed a diagnosis of PanNET using morphological
features and this is demonstrated in [Fig. 1]. However, only 23 of 35 FNA samples (66 %) were able to report a Ki67 PI and grade
compared to all 26 FNB samples (100 %) (P = 0.0006). For those with reportable values at EUS-FNA, the correlation of Ki67 PI
compared to surgical resection histology was poor (r = – 0.08, 95 % confidence interval
–0.50 – 0.38, P = 0.74). For FNB samples, the correlation of Ki67 PI with surgical histology was
moderate (r = 0.65, 95 % confidence interval 0.35 – 0.83, P = 0.0004). With respect to tumor grading, FNA samples showed a poor correlation with
a kappa value of 0.026 whereas FNB samples showed a moderate correlation with a kappa
value of 0.474. [Table 2] shows the mismatch in categorization for both FNA and FNB samples. Of the 23 FNA
samples with a reported grading, 22 were grade 1 on cytology of which seven of 22
were grade 2 on surgical resection and one was grade 3 on cytology but was grade 1
on surgical resection. Of the 26 FNB samples with a reported grading, 17 were grade
1 at FNB but six of 17 were grade 2 on surgical resection and nine were grade 2 at
FNB of which one of nine was grade 1 on surgical resection. There was no statistically
significant difference between FNA and FNB regarding grade 1 mismatch (P = 1.0).
Fig. 1 Examples of EUS-FNA and EUS-FNB samples of PanNETs. a FNA sample Papanicolaou x40 magnification. b FNA sample synaptophysin staining x40 magnification. c FNB sample H&E x40 magnification. d FNB sample synaptophysin staining high magnification.
Table 2
EUS-FNA and EUS-FNB sample tumor grading compared to surgical resection histology.
|
Surgical resection sample
|
|
Grade 1
|
Grade 2
|
Grade 3
|
Unable
|
|
FNA grade 1
FNB grade 1
|
15
11
|
7
6
|
0
0
|
0
0
|
|
FNA grade 2
FNB grade 2
|
0
1
|
0
8
|
0
0
|
0
0
|
|
FNA grade 3
FNB grade 3
|
1
0
|
0
0
|
0
0
|
0
0
|
|
FNA unable
FNB unable
|
8
0
|
3
0
|
1
0
|
0
0
|
EUS-FNA, endoscopic ultrasound-guided fine-needle aspiration; EUS-FNB, endoscopic
ultrasound-guided fine-needle biopsy
Lesions < 2 cm
Overall there were 25 lesions < 2 cm of which 19 were sampled by FNA and eight by
FNB (3 were sampled by both). For FNA samples, the correlation of Ki67 PI compared
to surgical resection histology was poor (r = –0.149, 95 % CI – 0.66 – 0.53, P = 0.75). For FNB samples, the correlation of Ki67 PI compared to surgical resection
histology was moderate (r = 0.57, 95 % CI – 0.23 – 0.91, P = 0.14). With respect to tumor grading, FNA samples showed a poor correlation with
a kappa value of 0.003 and FNB samples showed a good correlation with a kappa of 0.72. Of
the 11 of 19 FNA samples with a reported grading, 10 were grade 1 and one was grade
3 on cytology but on surgical resection histology 17 were grade 1, two were grade
2 and none were grade 3. Of the eight FNB samples, six were grade 1 and two were grade
2 but at surgical resection five were grade 1 and three were grade 2.
Solid PanNETs
When cystic and mixed cystic/solid PanNETs were excluded, this left 41 pure solid
lesions of which 21 were obtained using FNA and 23 using FNB (3 lesions were sampled
using both techniques). There were no significant differences in mean size (FNA 27.5 mm
vs. FNB 32.8 mm, P = 0.38), number of needle passes (3.0 vs. 3.0, P = 0.91) or lesion location (P = 0.25). Ki67 PI could be reported on 14 of 21 FNA samples compared to all 23 FNB
samples (P = 0.003). For FNA samples, the correlation of Ki67 PI compared to surgical resection
histology was poor (r = – 0.102, 95 % confidence interval – 0.66 – 0.53, P = 0.77). For FNB samples, the correlation of Ki67 PI with surgical resection histology
was moderate (r = 0.641, 95 % confidence interval 0.31 – 0.83, P = 0.001). With respect to tumor grading, FNA samples showed a poor correlation with
a kappa value of 0.044 and FNB samples showed a moderate correlation with a kappa
of 0.58. Of the 14 FNA samples with a reported grading, all were grade 1 on cytology
but six of 14 were grade 2 on surgical resection histology. Of the 23 FNB samples
with a reported grading, 15 were grade 1 and eight grade 2. At surgical resection
five of 15 which were grade 1 at EUS-FNB were found to be grade 2 but all of those
that were grade 2 at EUS-FNB were confirmed as grade 2 at surgical resection. There
was no statistically significant difference between FNA and FNB regarding grade 1
mismatch (P = 0.71).
FNB needle subgroup analysis
Ki67 PI and grading could be reported on all FNB samples irrespective of the two needle
types (Procore and Sharkcore). Twelve samples had been obtained using the Procore
needle (reverse bevel) and 14 samples obtained using the Sharkcore needle (Fork tip).
For Procore samples, the correlation of Ki67 PI with surgical resection histology
was moderate (r = 0.521, 95 % confidence interval – 0.07 – 0.84, P = 0.08). For Sharkcore samples, the correlation of Ki67 PI with surgical resection
histology was good (r = 0.788, 95 % confidence interval 0.42 – 0.93, P = 0.0013). With respect to tumor grading, Procore samples showed a moderate correlation
with a kappa of 0.47 and Sharkcore samples similarly showed a moderate correlation
with a kappa of 0.435.
Discussion
The current study shows that FNB needles are significantly more likely to provide
sufficient pathological material that can be used to characterize PanNETs. Furthermore,
when compared to surgical resection histology, FNB samples had a closer correlation
with respect to Ki67 PI and tumor grading than FNA samples. This difference was maintained
when cystic PanNETs were excluded. This subgroup analysis was performed as there was
a significant difference in the distribution of lesion morphology when comparing FNA
to FNB. Crucially, in lesions < 2 cm, FNB samples correlated significantly better
than FNA samples but was only moderate compared to the final surgical resection sample.
When looking at the different subtypes of FNB needles, Sharkcore samples appeared
to have the closest correlation of Ki67 PI to surgical resection histology but no
difference in agreement with overall grading. This reflects the closer correlation
of the Ki67 PI in Sharkcore samples to the final surgical histology compared to the
other needle type. However, the range of Ki67 PI for grade 2 is large (e. g. 2 % to
20 %) and therefore if an endoscopic sample was reported as 2.5 % and the surgical
sample 18 % this would be grade 2 for both samples. Clinically, whilst the grade is
helpful, a more precise Ki67 PI supports clinical decision-making.
Both FNA and FNB appeared to under-grade lesions that were found to be grade 2 on
surgical histology. In fact, the vast majority of FNA samples were reported as grade
1, none as grade 2 and one as grade 3. Several of those classified as grade 1 were
upgraded at resection and the grade 3 sample was found to be grade 1 at surgery. In
the FNB group there were more confident classifications into grade 2 although there
was still a proportion classified as grade 1 that were ultimately grade 2 on surgical
resection.
PanNETs are rare but carry a more favorable prognosis when compared to pancreatic
adenocarcinoma [1]
[2]. Prognosis and management in PanNETs is closely linked to Ki67 PI and tumor grade,
therefore, accurate sampling of such lesions is essential [4]
[5]
[6]. Many PanNETs are sampled pre-peratively using EUS but the accuracy of EUS- FNA
has been shown to vary [9]
[10]
[11]
[12]
[13]. The recent addition of FNB needles to the endosonographers’ armamentarium may improve
the diagnostic accuracy [18]
[20] and tumor grading but there is a paucity of data concerning their use for PanNETs.
EUS has become the modality of choice for assessment and sampling of PanNETs with
a reported sensitivity of up to 80 % to 90 % [9]. Previous studies have shown that EUS-FNA samples can be used to report Ki67 PI
but the correlation with surgical resection histology is unclear. Surgical resection
histology was not the gold standard in all of these studies and many were small (n < 20)
[10]
[11]
[13]
[14]
[15]
[16]
[17]. Two recent studies have shown that cytology tends to underestimate grade 2 lesions
and the concordance was affected specimen cellularity [16]
[17].
One recent study compared FNA to FNB in patients with suspected PanNETs (n = 20) which
showed that FNB required fewer passes to obtain a diagnosis. There was no reporting
of KI67 PI or grading or comparison with surgical histology in this study [22].
There are limitations to the current study. Primarily, this is a retrospective analysis
in patients with a positive diagnosis. Many patients referred to our unit will have
had a single sampling either by FNA or FNB but no other tissue samples taken to make
comparison, mainly due to advanced stage on imaging negating the need. There were
also a number of patients who proceeded straight to surgery without preoperative biopsy
and there is a possibility that this contributed some bias. Both of these factors,
however, would be expected to affect both FNA and FNB equally. During the study period
the classification scheme for grading PanNETs changed [24]
[25]. However, to minimize the bias this might have caused, each sample, whether endoscopic
or surgical, was allocated according to the grading system used at that time period.
This meant that allocation to a grade was consistent for each lesion. There were also
very few lesions that had undergone simultaneous FNA and FNB, therefore, the differences
between the groups may have been affected by this. Ideally, a prospective study using
both FNA and FNB needles in this group of patients would be required to determine
which is optimal for sampling. The most recent guidelines for assessing PanNETs recommend
a count of 500 to 2000 cells in the area of highest labelling in intact tumor tissue
to provide the most accurate assessment of Ki67 [24]
[25]. One of the difficulties of EUS-guided samples is that only a small part of the
lesion is sampled and often fewer than 500 cells may be available for counting. This
means that there is not a representative "hotspot" to assess, therefore, the lesion
may be under-graded. Furthermore, current recommendations are to photograph the "hotspot"
and then manually count 2000 cells. This practice was not consistently used throughout
the study period. Future studies should set a minimum number of cells present before
Ki67 assessment can be performed, recording of magnification used, intensity of staining
seen and which aids were used to facilitate cell counts. These factors apply to both
histology and cytology samples and were not standardized during the study period.
Also, with respect to cytology samples particularly, cellular dissociation may affect
assessment of tumor and non-tumor cells following staining, which is especially seen
in liquid-based cytology samples. Cell blocks may mitigate this effect, however, samples
were reported overall rather than by the individual components. There was a tendency
to perform more passes with the FNB needle (3.0 vs 2.5) which may have affected the
study outcome despite not being statistically significantly different. In a study
with larger numbers this may have contributed to improved pathological samples. Furthermore,
different needle calibers were used depending upon endosonographer preference, which
may have also had an effect on tissue acquisition [26]. Also, because our study was conducted over several years, it is possible that either
the endosonographers or pathologists have moved up the learning curve and therefore
affected the results. This may be relevant as FNB needles have only become available
in the last few years. Finally, the majority of lesions were grade 1 or 2, therefore,
the accuracy of FNA or FNB in grade 3 lesions cannot be commented upon in this study.
However, most grade 3 lesions would be put forward for surgical resection or systemic
therapy and the area of most interest being whether it is safe to truly classify a
lesion as grade 1 using solely endoscopic sampling. Strengths of the current study
include representation of one of the largest endoscopic series in patients presenting
with PanNETs with FNA and FNB from a high-volume center with experienced endosonographers
and pathologists. Also, EUS-guided samples KI67 PI and grade were compared to surgical
resection histology as the gold standard unlike in most previous studies [10]
[11]
[13]
[14]
[16]
[17].
Conclusion
In summary, in the current study we have shown that FNB is superior to FNA for accurate
characterization of grading of PanNET lesions with the best correlation to surgical
resection histology seen using the Sharkcore needle. A randomized, controlled, multicenter,
crossover trial in which lesions are sampled using both FNA and FNB with blinding
of the pathologists will be required to fully elucidate which needle type is optimal.
