Risk of malignancy in small pancreatic cysts decreases over time.

Abstract

Background:

Pancreatic cysts < 15mm without worrisome features have practically no risk of malignancy at the time of diagnosis but this can change over time. Optimal duration of follow-up is a matter of debate. We evaluated predictors of malignancy and attempted to identify a time to safely discontinue surveillance.

Methods:

Bi-centric study utilizing prospectively collected databases of patients with pancreatic cysts measuring < 15mm and without worrisome features who underwent surveillance at the Massachusetts General Hospital (1988–2017) and at the University of Verona Hospital Trust (2000–2016). The risk of malignant transformation was assessed using the Kaplan-Meier method and parametric survival models, and predictors of malignancy were evaluated using Cox regression.

Results:

806 patients were identified. Median follow-up was 58 months (6–347). Over time, 58 (7.2%) cysts were resected and of those, 11 had high grade dysplasia (HGD) or invasive cancer. Three additional patients had unresectable cancer for a total rate of malignancy of 1.7%. Predictors of development of malignancy included an increase in size ≥ 2.5mm/year (HR=29.54, 95% CI: 9.39–92.91, P < 0.001) and the development of worrisome features (HR=9.17, 95% CI: 2.99–28.10, P = 0.001). Comparison of parametric survival models suggested that the risk of malignancy decreased after three years of surveillance and was lower than 0.2% after five years.

Conclusions:

Pancreatic cysts < 15mm at the time of diagnosis have a very low risk of malignant transformation. Our findings indicate the risk decreases over time. Size increase of ≥ 2.5mm/year is the strongest predictor of malignancy.

Introduction

Incidental identification of pancreatic cystic lesions has become common with the increasing use of cross-sectional imaging and improvement in the resolution of both computed tomography (CT) and magnetic resonance (MR) imaging. A recent study by Kromrey et al. [1] reported a 49.1% prevalence of pancreatic cysts by MR, of which only 6% had a size > 1 cm. These results are in line with autopsy studies that have shown a high prevalence of small cystic pancreatic lesions in older patients [2]. In the clinical setting, the most common incidentally discovered pancreatic cyst is a branch duct intraductal papillary mucinous neoplasm (BD-IPMN) [3]. The natural history of BD-IPMNs is not completely understood, but they are considered as premalignant lesions, albeit with a low risk of progression [4,5]. At present, it is unclear how many of these increasingly identified small pancreatic cysts represent true BD-IPMNs vs non-neoplastic lesions.

The management of incidentally discovered pancreatic cysts is based on surveillance protocols that have been described over the years in several guidelines [6–10]. However, an endpoint for surveillance in unchanged lesions has not been uniformly defined. The American Gastroenterological Association (AGA) guidelines recommend discontinuing surveillance after 5 years if the cysts show no significant change during follow-up [9]. By contrast, the International Association of Pancreatology (IAP) and the American College of Gastroenterology (ACG) guidelines do not offer surveillance discontinuation guidelines in patients with pancreatic cysts that lack concerning features for high-grade dysplasia or invasive cancer [6,8]. European guidelines recommend continued patient follow-up so long as patients remain fit for surgery [7]. In a previous study, we showed that patients with BD-IPMNs that remained ≤ 15 mm for a period of 5 years had a risk of developing pancreatic malignancy that was no different than that of the general population [11], suggesting that follow-up for small cysts could be discontinued at that time.

This study evaluates the risk of progression of small pancreatic cysts in patients followed in two referral distinct centers. Our goal was to evaluate the likelihood of progression, to identify factors associated with a higher risk of malignant transformation, and to determine if there is a time to safely discontinue surveillance.

Methods

This is a retrospective, observational, study of patients with either a suspected or presumed branch duct IPMN [3,12]. Data were collected from prospectively maintained IPMN registries of two referral centers: patients evaluated at Massachusetts General Hospital (MGH) between 1988–2017 and at the Department of General and Pancreatic Surgery, University of Verona Hospital Trust, from 2000 to 2016. For inclusion in our study, patients had to have at least one cyst < 15 mm at diagnosis, and at least 6 months of follow-up. Demographics, epidemiological and radiological data were collected, including cyst size, number of cysts, presence of thick walls, mural nodules, dilation of main pancreatic duct (MPD), septae and evidence of communication with the MPD. Patients with worrisome features (WF) and/or high-risk stigmata (HRS) at the baseline observation were excluded, patients with hereditary pancreatic cancer syndromes were excluded as well. The length of follow-up was calculated as the difference between the first and the last observation. The difference in terms of cyst size and development of any change in the characteristics of the cyst, such as presence of new cysts and septae or appearance of WF were collected for all observations.

The surveillance of these patients was based on abdominal MR or CT scans after an average of 12 months from the diagnosis and then every two years or yearly for patients with a more recent diagnosis, in accordance with the latest guidelines available [13,14]. Also, only imaging techniques of the same type were compared for the measurement of the cyst size and to report any change. Endoscopic ultrasound (EUS) with or without fine-needle aspiration (FNA) was performed mostly when there was suspicion of malignancy in imaging studies or if there was doubt regarding the diagnosis of IPMN. Patients who developed HRS were resected if fit for surgery.

The primary endpoint of our analysis was development of malignancy both high-grade dysplasia and carcinoma arising in IPMN defined according to Baltimore and Verona consensus meetings [15,16]. The presence of a concurrent pancreatic adenocarcinoma (PDAC) was included as well [17].

Statistical analysis

Categorical and continuous variables are presented as percentages and medians with interquartile ranges, respectively. For the univariate analysis, relationships between categorical characteristics were evaluated using the Chi-squared test and Fisher’s exact test, and between continuous variables using the Student’s t-test and Mann-Whitney U test, as appropriate. The overall annual growth rate of the pancreatic cysts was calculated as the difference in size between the first and the last observation divided by the length of the follow-up. For multicystic lesions, calculation of the annual growth rate was based on the largest cyst at the time of the diagnosis. We categorized the annual growth rate measurements as ≥ 2.5 versus < 2.5 mm/year based on the assessment of the latest IAP guidelines for management of BD-IPMN, where a cyst growth rate ≥ 5 mm over 2 years has been included among the WF [6].

Cumulative incidence of malignancy was calculated using the Kaplan-Meier method and computing time from cyst diagnosis to malignancy or last follow-up. To evaluate whether the risk (i.e. hazard) of malignancy was increasing, decreasing, or remained constant following diagnosis, we compared parameter estimates from exponential and Weibull failure time models (without covariates). The exponential model assumes that the hazard rate is constant over time, while the Weibull model is more flexible and incorporates a scale parameter that allows the hazard rate to increase or decrease with time. Because the Weibull model is equivalent to the exponential model when the scale parameter equals one, comparing these two nested models permits statistical assessment of whether the risk of malignancy is changing versus constant during follow-up. The adequacy of the exponential model/constant hazard assumption was assessed via the Lagrange test, and the estimated hazard function was plotted. The impact of cyst characteristics including size at diagnosis, change in size per year, development of septae and new cysts, and presence of WF on risk of malignancy were evaluated using univariate Cox proportional hazards models incorporating time-varying covariates for repeated cyst assessments. A P-value less than 0.05 was considered statistically significant. SPSS version 24 and SAS version 9.3 were used for the analysis.

Results

A total of 806 patients were identified, including 308 from MGH and 498 from Verona Hospital. A preliminary analysis to compare the cohorts from the two institutions was performed and no statistically significant differences were found between the two groups except for a longer median time of follow-up (59 vs 42 months, P < 0.001) and a more frequent positive family history for pancreatic cancer (14.4 vs 5.4%, P < 0.001) in patients from MGH.

The median age for the entire cohort was 65 years (range 17–88) and 244/806 (30.3%) of the patients were males. The median time of follow-up was 58 months (range 6–347) and 330 patients (49%) were followed for more than 5 years.

The median cyst size at the time of the diagnosis was 10 mm (range 2–14) and 440/806 (55%) had more than one cyst. 54 patients (6.7%) had cysts with septae at the first imaging, and only 15 (1.9%) were found to have a mild dilation of the main pancreatic duct (MPD) (< 5 mm).

During follow-up, 45% of patients (359/806) had an increase in cyst size. The median annual growth rate in these patients was 1.2 mm (range 0.1–44.0).

Table 1 shows the characteristics of the cohort and a comparison between patients who developed malignancy during surveillance and those who did not. None of the epidemiological, radiological, or clinical parameters at the first observation were significantly associated with the development of malignancy. The development of WF (61.5% vs 7.7%, P < 0.001), the development of septae or new cystic lesions (38.5% vs 6.6%, P < 0.001) and the presence of an annual growth rate ≥ 2.5 mm/year (71.4% vs 15.4%, P < 0.001) were the only significant predictors of malignancy.

Table 1.

Univariate analysis of predictors for malignancy in small pancreatic cysts

Characteristics	Malignancy (No)	Malignancy (Yes)	P value
n	792	14
Age in years (range)	65 (17–88)	66 (44–79)	0.834
Male gender (%)	242 (30.6)	2 (14.3)	0.249
Family history (%)	54 (8.3)	1 (9.1)	1.000
Symptoms (%)	442 (55.8)	6 (42.9)	0.334
Diabetes (%)	54 (13.6)	0	0.605
Multifocal IPMN (%)	438 (55.4)	6 (42.9)	0.351
Septae (%)	52 (6.6)	2 (14.3)	0.240
Site of the main cyst			0.122
head (%)	308 (43.9)	3 (23.1)
body (%)	287 (41.0)	9 (69.2)
tail (%)	106 (15.1)	1 (7.7)
Number of cysts			0.290
1(%)	172 (57.1)	6 (85.7)
2(%)	46 (15.3)	0
>2 (%)	83 (27.6)	1 (14.3)
Dilation of MPD (< 5 mm) (%)	14(1.8)	1 (7.1)	0.233
Unilocular (%)	740 (96.0)	13 (100)	1.000
Cyst size (range)	10 (2–14)	10 (3–12)	0.349
Development of WF (%)	61 (7.7)	8 (61.5)	<0.001
Development of new cysts and/or septae (%)	52 (6.6)	5 (38.5)	<0.001
Growth rate > 2.5 mm/year	122 (15.4)	10 (71.4)	<0.001

Overall 67 patients (8.3%) developed WF during follow-up, of whom 17 (2%) occurred after 5 years of surveillance; and only 3 patients (0.4%) showed HRS. The median time to develop WF was 43 months (6–325). Regarding the annual growth rate, 132 patients (16.4%) had cysts that grew ≥ 2.5 mm/year and in the majority of these patients (79%) this growth occurred within the first 5 years of surveillance.

During surveillance, 58 (7.2%) patients underwent surgery for the development of WF or HRS, and of those patients, 11 (1.4% of the entire cohort) had malignancy, including 2 cases of high-grade dysplasia, 4 cases with confirmed invasive cancer arising in IPMN and 5 patients with pancreatic cancer, although it is unclear if they developed the carcinoma within the IPMN or if it was a distinct PDAC. Additionally, another 3 patients were considered unresectable when malignancy was detected, for a total rate of malignancy of 1.7%. Among these 3 patients, one had a suspicious concurrent pancreatic adenocarcinoma (PDAC). The median time to develop malignancy was 65 months (10–351); after resection, in 3 of the 11 patients who had malignancy, recurrence was identified, and they eventually died from pancreatic cancer. The remaining 8 patients have had no evidence of recurrence.

Analysis of malignant predictors over time

Table 2 shows results from univariate Cox regression models for the association between risk of malignancy and 4 variables: 1) size of the cyst at the diagnosis, 2) development of WF during follow-up, 3) annual growth rate and 4) development of septae and/or new cysts. Cyst size at the first observation and the development of any change did not show an association with development of malignancy. In contrast, an annual growth rate of ≥ 2.5 mm was found to be strongly associated with the risk of malignancy with a nearly 30-fold increased risk (HR = 29.54, 95% CI: 9.39–92.91, P < 0.001). The development of WF was also associated with a marked increase in risk of malignancy (HR = 9.17, 95% CI: 2.99–28.10, P = 0.001).

Table 2.

Cox regression analysis of malignant transformation predictors

Characteristic	Univariate HR	95% CI	P value
Size at the diagnosis (mm)	0.87	0.722–1.054	0.158
Development of WF	9.17	2.992–28.103	< 0.001
Annual growth rate ≥ 2.5 mm/year	29.54	9.391–92.910	< 0.001
Development of new cysts and/or septae	2.89	0.624–13.336	0.175

Analysis of risk of malignancy

Cumulative incidence of malignancy was found to be 0.94% and 3.37% at 5 and 10 years (Table 3). Using the Lagrange test to assess the adequacy of the exponential model relative to the Weibull, the constant hazard assumption of the exponential model was rejected in favor of the Weibull model (P = 0.019), meaning that the risk of malignancy was not constant. The positive estimate of the Weibull scale parameter suggests that the risk of malignancy decreases over time (Weibull scale parameter = 107.19, 95% CI: 35.50–322.66). Fig. 1 shows how the risk of malignancy calculated from the Weibull model decreases steeply for the entire population during the initial 3 years and then continues to decrease at a slower pace before effectively leveling off below 0.2% after 5 years.

Table 3.

Cumulative incidence of malignant transformation in small pancreatic cysts

Years from diagnosis	Cumulative incidence	95 % CI
1	0.14%	0.02%−0.97%
2	0.43%	0.14%−1.34%
3	0.94%	0.42%−2.10%
4	0.94%	0.42%−2.10%
5	0.94%	0.42%−2.10%
6	1.49%	0.73%−3.04%
7	2.16%	1.13%−4.13%
8	2.60%	1.38%−4.86%
9	2.60%	1.38%−4.86%
10	3.37%	1.74%−6.47%
11	4.15%	2.17%−7.88%

Fig 1.

Weibull model shows how the risk of malignancy in small pancreatic cysts decreases over time.

Discussion

Surveillance of pancreatic cysts without WF or HRS has been the subject of several studies [18–22] and they all show that these lesions follow an indolent course in the majority of patients. However, the natural history is incompletely understood and although it is a fact that some can become malignant, estimating the future risk for an individual patient has proven to be a challenge. Recently, two large-cohort studies have been published on this topic. Marchegiani et al. [23] analyzed the fate of 1036 patients with presumed BD-IPMNs and suggested a strategy of possible discontinuation of follow-up in patients with stable cysts for at least 5 years who were older than 65 years of age. In contrast, Oyama et al. [5] showed in a cohort of 1404 patients with BD-IPMNs that the risk of malignant transformation was higher than in the general population and remained higher even after 5 years of surveillance. However, both studies included patients with presumed BD-IPMNs of all sizes.

In the current study, we chose to focus on small pancreatic cysts. Based on our preliminary analysis that suggested that cysts that remained ≤ 15 mm for 5 years had a risk of pancreatic cancer that was no different than the general population [11], we chose to study patients who met this size threshold. This is particularly relevant since the vast majority of incidentally discovered pancreatic cysts are very small (in the study by Kromrey 94% of all cysts were ≤ 1 cm [1]). While previous studies only considered presumed BD-IPMN, we decided to include also suspected BD-IPMN for two reasons; first, in small pancreatic cysts is not always easy to identify the communication between the cyst and the MPD at the imaging, second we wanted to evaluate the risk of malignancy in a cohort which was as close as possible to the clinical practice. We found that indeed the overall risk of malignancy was only 1.7% for the entire cohort over a median follow-up of 5 years. The cumulative incidence of malignancy was 0.94% at 5 years and 3.37% at 10 years. It is not unexpected that the cumulative incidence increases over time, since by definition it increases at every timepoint when there is a new case; however, this is not inconsistent with our findings that show how the risk of malignancy is not constant over the time and decreases sharply after 3 years, levelling off at 5 years, which suggests a deceleration in terms of cumulative incidence. This suggests that perhaps in most patients with small, incidentally discovered cysts, which do not develop WF and do not show an annual growth rate ≥ 2.5 mm/year, follow-up can be discontinued at 3 years. It is unclear what the optimal time for the first follow-up should be, but perhaps initial repeat imaging 6 months after, and if stability is found, which will be the case in the vast majority of patients, then one more study one year later and a final one at 3 years to determine if further follow-up is warranted.

Our study also aimed to identify features that would help predict progression in these very small cysts. While in series that include cysts of all sizes a larger size indeed is associated with a higher likelihood of developing WF and malignancy [12,24], we did not observe this in a cohort of cysts ranging in size from 2 to 14 mm, and in fact no baseline characteristic was able to predict development of malignancy. We did find that the strongest predictor of malignancy was the annual growth rate of the cyst. Patients whose cysts grew ≥ 2.5 mm/year had a 30-fold increase of malignant transformation. This is consistent with the study by Marchegiani et al. [23] that identified the growth rate of the cyst as the most important predictor of pancreatic cancer development with a HR of nearly 20 at the same 2.5 mm/year threshold. Other studies [25,26], have also shown an association between the annual growth rate in pancreatic cysts and risk of malignancy, although with different cut-offs. The current analysis confirmed these results and shows that even in a selected cohort of small pancreatic cysts this is the most relevant predictor. An annual growth rate of ≥ 2.5 mm was identified in 71.4% of the cysts that developed malignancy, but only in 15.4% of those who did not. This finding might suggest the necessity to modify the follow-up strategy according to the presence of this predictor which appears to be independent from other WF.

To the best of our knowledge, this is the first study to use a parametric survival model to understand how the risk of malignancy is distributed over time in this disease setting. According to our analysis, the probability of a small pancreatic cyst developing high-grade dysplasia or invasive carcinoma is higher in the first 3 years of surveillance and is close to zero after 5 years of follow-up. Most of the previous studies regarding the natural history of BD-IPMN attempted to define the safety of surveillance discontinuation with different statistical models.[5,27,28] Cumulative Incidence, Standard incidence ratio and Cox regression analysis have been commonly used with this purpose. However, these statistical functions do not provide useful information about the shape of the hazard function which is how the time from diagnosis to malignancy is distributed in a specific cohort. The aim of our study was to estimate a time to safely discontinue follow-up in patients with small pancreatic cysts; using the Weibull model, it appears that if after 3 years no change has occurred the risk of malignancy is negligible. It is important to highlight that the risk will never be zero, since in the general population (with or without known pancreatic cysts), the lifetime risk of pancreatic cancer is estimated to be 1.6% [29].

This study has several limitations including its retrospective and bi-centric nature. While the inclusion of patients from two different institutions increased the sample size and the power of the study, on the other hand it introduced potential bias associated with follow-up techniques and surveillance timing in each center. Another limitation is that although we included both suspected and presumed IPMN, we lack a definitive diagnosis in most cases, and therefore may have in this cohort diseases with different biological behavior [3]. Finally, these findings, although statistically significant, are based on a very small number of malignant cases (which explains the large variance of the HR in the Cox regression analyses). Additionally, in only one patient was a concurrent PDAC definitively identified: some studies have suggested that the risk of concurrent PDAC in patients with IPMN, mainly BD-IPMN, can be as high as 7%; however, in these cohorts the mean size of the IPMNs that are being followed is much greater [30].

In conclusion, pancreatic cysts < 15 mm showed a very low risk of malignant transformation, which is similar to the risk of pancreatic cancer in the general population. The strongest predictor of malignancy was annual cyst growth ≥ 2.5 mm. Analysis of this cohort shows that the risk of malignancy in these small pancreatic cysts is higher in the first 3 years and then becomes considerably lower. If this observation is confirmed in other studies, it would allow for discontinuation of follow-up in the vast majority of patients with incidentally discovered pancreatic cysts.