Progression to pancreatic ductal adenocarcinoma from pancreatic intraepithelial neoplasia: Results of a simulation model

Abstract

Objectives:

To gain insight into the natural history and carcinogenesis pathway of Pancreatic Intraepithelial Neoplasia (PanIN) lesions by building a calibrated simulation model of PanIN progression to pancreatic ductal adenocarcinoma (PDAC)

Methods:

We revised a previously validated simulation model of solid PDAC, calibrating the model to fit data from the National Cancer Institute’s Surveillance, Epidemiology, and End Results program and published literature on PanIN prevalence by age. We estimated the likelihood of progression from PanIN states (1, 2, and 3) to PDAC and the time between PanIN onset and PDAC (dwell time). We evaluated a hypothetical intervention to test for and treat PanIN 3 lesions to estimate the potential benefits from PanIN detection.

Results:

We estimated the lifetime probability of progressing from PanIN 1 to PDAC to be 1.5% (men), 1.3% (women). Progression from PanIN 1 to PDAC took 33.6 years and 35.3 years, respectively, and from PanIN 3 to PDAC took 11.3 years and 12.3 years. A hypothetical test for PanIN 3 detection and treatment could provide a maximum, average life expectancy gain of 40 days.

Conclusions:

Our modeling analysis estimates PanINs have a relatively indolent course to PDAC, supporting the feasibility of potential future early detection strategies.

1. Introduction

Defining the evolution of pancreatic ductal adenocarcinoma (PDAC) from precursor pancreatic intraepithelial neoplasia (PanIN) lesions is critically important. In particular, knowledge of the time periods between the onset of PanIN lesions and the development of PDAC – defined as “dwell times” – provide necessary insight into the feasibility of future early detection strategies that are focused on PanIN identification and treatment.

The rate and degree of dysplasia in PanINs has been observed to correlate with the presence of malignancy [1–3], and clinical information from retrospective studies of pancreatectomy specimens yields important insight into the natural history of PanINs. Published data on pancreatectomy and/or autopsy specimens suggests that, in general, higher-grade PanINs (e.g. PanIN 3) are more prevalent in: 1) older individuals [4]; 2) patients with PDAC [1]; and 3) patients at higher risk for PDAC (e.g. from chronic pancreatitis) [1,5]. Autopsy studies demonstrate that most patients with PanIN 3 lesions also have lower-grade PanINs, and confirm other observational data that indicate a higher likelihood of PanIN 3 with increasing age [4,6]. Most likely, these observations concerning PanINs correlate with a progressively increasing mutational burden [1,7–10].

PanIN lesions are not yet directly observable except in pathologic specimens; as such, observational data that elucidates their progression is absent. By using simulation modeling methods, it is possible to leverage the available data on PanIN prevalence [4,6], combined with SEER cancer registry data on age-based PDAC incidence and death [11,12], to infer PanIN dwell times and rates at which PanIN precursors transform. Our purpose was to gain quantitative insight into the natural history of PanIN lesions by building a calibrated simulation model of PanIN progression to PDAC.

2. Materials and methods

2.1. Model structure

We previously developed a simulation model of the natural history of PDAC in hypothetical populations, calibrated to data from the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program and published literature [4,6,11–16]. The current analysis was performed on an expanded version of this model. In our original model, we simulated progression along cystic and solid pathways to PDAC for both men and women, but did not explicitly consider PanIN precursor states [17]. In our updated model, we simulated PanIN development using three health states, defined by the highest grade of PanINs in the subject. The model, written in the programming language C++ (Microsoft Visual Studio 2015, Microsoft Corporation, Redmond WA), is a microsimulation model with a 1-month cycle length. It simulates cohorts of 10 million hypothetical patients from age 20 until death or age 100. A simplified schematic of the model is provided in Fig. 1.

Fig. 1. Model Schematic.

Simplified schematic of the Markov state transition model. In this pathway to solid-derived PDAC, subjects progress from Normal to having at least one PanIN 1 lesion. Subjects can transition to the PanIN 2 and then PanIN 3 cohorts if they develop at least one higher-grade PanIN lesion. From PanIN 3, subjects can transition to Undetected PDAC, which can then progress to wider Undetected disease or be Detected at each stage. Subjects with Detected PDAC are subject to stage-dependent estimates of cancer mortality.

We used this updated model in the current evaluation of the dominant, solid pathway of PDAC development [18]. Subjects could develop PanINs starting at grade 1 (i.e. “PanIN 1”) and progress through PanIN 2 and PanIN 3 to malignancy. We assumed that PanIN lesions could not regress or improve to less dysplastic forms. Once a subject reached the PanIN 3 category, they could transition to undetected malignancy, which could then be detected at the localized, regional, or distant disease stages. The “Undetected” malignancy state was defined as one in which malignant cells were present, but was not detectable through clinical (e.g. symptomatic) presentation. We assumed that there is a larger population of undetected PDAC than detected PDAC, given this broad definition of undetected cancer. Each model cycle (one month), a subject could remain in their current health state or progress to a new health state on the basis of calibrated transition probabilities. If diagnosed with PDAC, patients were subject to stage-dependent survival rates derived from SEER [11]. Patients were also subject to age and gender-specific all-cause mortality risks (i.e. non-PDAC-related) each cycle. Model outcomes were averaged over 10 million simulated subjects.

2.2. Model calibration

In our model, some transition probabilities address events that cannot be observed, such as a PanIN 1 to PanIN 2 transition, and therefore require estimation through a calibration process. In calibration, the model is run iteratively using multiple possible parameter estimates for unobservable parameters; the resulting outputs are compared to the calibration targets. Preferred parameter sets are determined by lowest χ² goodness of fit (GOF) metric using a calibration method based on simulated annealing process with some modifications. Please see the Appendix for details of the calibration process.

Calibration targets for our model are shown in Table 1 and included: lifetime risk of PDAC, incidence of PDAC by age, lifetime risk of PDAC-associated death, the proportion of PDAC that arises from the solid pathway, PanIN prevalence by grade at ages 50 and 80, and stage of PDAC at diagnosis [4,6,11,13–15,18]. Some calibration targets were constant with age, while others varied. The development of PanIN 1 lesions was assumed to exponentially increase with age, reflecting the increase in prevalence observed in older age groups. Linear progression and power functions were tested and did not produce superior calibration results. Transitions from PanIN 1 to 2, PanIN 2 to 3, and PanIN 3 to Undetected PDAC were all modeled as linear functions with the freedom to increase with age; however, the calibration results were best if these transitions were constant.

Table 1.

Calibration targets and model inputs.

Calibration Target/Model Input	Female	Male	Source
Lifetime prevalence of PDAC (%)	1.06	1.13	11, 12
Incidence of PDAC (age 20–29) (#/100K)	0.11	0.11	11
Incidence of PDAC (age 30–39)	0.44	0.56	11
Incidence of PDAC (age 40–49)	2.89	3.67	11
Incidence of PDAC (age 50–59)	10.67	15.44	11
Incidence of PDAC (age 60–69)	27.89	38.44	11
Incidence of PDAC (age 70–79)	53.89	65.22	11
Prevalence of PanIN 1 lesions, age 50 (%)	20	20	4, 6
Prevalence of PanIN 1 lesions, age 80	60	60	4, 6
Prevalence of PanIN 2 lesions, age 80	15	15	4, 6
Prevalence of PanIN 3 lesions, age 80	2	2	4, 6
Proportion of PDAC diagnosed as localized disease (%)	10.00	8.00	11
… as regional disease	28.00	27.00	11
… as distant disease	50.00	56.00	11
… as unstaged disease	12.00	9.00	11
Proportion of PDAC from solid lesions (%)	90	90	18
Lifetime risk of PDAC-associated death (%)	0.88	0.94	11, 12
PDAC mortality from detected localized disease (monthly probability of death)	0.0228	0.0242	11
… from detected regional disease	0.0393	0.0393	11
… from detected distant disease	0.0631	0.0631	11
… from detected unstaged disease	0.0522	0.0522	11

One of the key uncertainties in our model is the estimate of PanIN prevalence at ages 50 and 80. Available information to support these estimates is relatively less robust than information regarding cancer incidence and prevalence. To address this limitation, we tested two alternative cases for PanIN prevalence. We tested a low scenario, in which all PanIN prevalence estimates were adjusted down 30%, and a high scenario, in which these estimates were increased 30%. See Appendix for further details on the calibration process and results.

2.3. Outcome measures

The primary model outputs were the average life expectancy (LE) of the simulated population, the prevalence by age and gender of the population when stratified by highest-grade PanIN lesion, and the prevalence of undetected and detected PDAC. For the current analysis, primary measures of interest were the probabilities of progression from “normal” (i.e. no PanIN) to PanIN 1, PanIN 1 to 2, PanIN 2 to 3, PanIN 3 to Undetected PDAC, and Undetected to Detected PDAC, as well as the time elapsed between PanIN development (each grade) and Detected PDAC, an estimation of dwell time.

2.3. Exploratory analysis

In an exploratory analysis, we introduced a hypothetical strategy, defined by “perfect” detection and intervention, to determine the maximum life expectancy gains that could be achieved – at the population level – if future technologies were capable of PanIN 3 detection. In this strategy, subjects who have a PanIN 3 lesion were identified without the possibility of false-positive or false-negative findings, and an intervention was applied that was capable of removing their risk of developing PDAC without associated complication, morbidity, or mortality.

We specifically considered the following testing scenarios: a test that could identify PanIN3 lesions at the time they developed (i.e. any age); and a one-time test applied at age 50, 60 or 70. We considered a perfect treatment scenario: if a PanIN 3 lesion or Undetected PDAC was present at the time of testing, the hypothetical intervention removed the subject’s PDAC risk. We ran our test-and-treat analysis first without consideration of surgical mortality, and then again under the assumption that the hypothetical intervention would be associated with a 2% mortality rate, an estimate derived from published mortality rates associated with pancreatic surgery [19–21].

3. Results

3.1. Calibration results

The calibration process was successful; we were able to achieve a well-calibrated model under the constraints of the defined calibration targets. Specific results of the calibration process are detailed in the Appendix, including comparisons of calibration targets and model projections.

3.2. PanIN and PDAC prevalence

Fig. 2 shows the model-estimated prevalence of PanIN 1, PanIN 2, PanIN 3, and solid-pathway undetected and detected PDAC by age. The total prevalence of PanIN lesions increases slowly with age, such that by age 80 almost 74% of both men and women have at least a single PanIN 1 lesion.

Fig. 2. Pancreatic Intraepithelial Neoplasia (PanIN) and Pancreatic Ductal Adenocarcinoma (PDAC) Prevalence by Age (Male).

The prevalence of PanIN lesions (A) and the prevalence of undetected and detected PDAC (B) by age for males, as calculated by the simulation model. The incidence and lifetime prevalence of detected PDAC are calibrated to SEER data, PanIN prevalence at ages 50 and 80 are calibrated to available literature. All other prevalence values are calculated by the simulation model.

3.3. Progression probability

For men, over the course of a lifetime, we estimated the probability of transitioning from the PanIN 1 to the PanIN 2 cohort to be 24%, from the PanIN 2 to the PanIN 3 cohort to be 18%, and from the PanIN 3 cohort to Detected malignancy to be 33%. For women, these probabilities were 26%, 18%, and 28%, respectively. This information is summarized in Fig. 3. For both men and women, a substantial percentage of subjects in the PanIN 1 cohort did not progress to PanIN 2 (76% (men) and 74% (women)), and similarly a substantial percentage of subjects in the PanIN 2 cohort did not progress to PanIN 3 (82% (men and women)). The overall chance of progressing from PanIN 1 to Detected malignancy was 1.5% for men, and 1.3% for women.

Fig. 3. Conditional Probability of Progression.

This figure shows the lifetime probability of progression conditional on a patient being in each PanIN state, from no PanIN to PanIN 1 (A), from PanIN 1 to PanIN 2 (B), from PanIN 2 to PanIN 3 (C), and from PanIN 3 to Detected PDAC (D). For example, for a woman, chart C shows the PanIN 2 population, for whom there is an 18% lifetime chance of progression to PanIN 3 and an 82% chance of remaining in PanIN 2. Undetected PDAC is not shown in these charts. The lifetime cumulative probability of progression from PanIN1 to Detected PDAC is shown in (E). Females are shown in the top sequence of charts, males in the bottom sequence.

3.4. Dwell times

On average, men spent 16.5 years (IQR Q1-Q3 6.00–24.3) in the PanIN 1 cohort before transitioning to the PanIN 2 cohort; women spent 17.4 (6.50–25.6) years. Men then spent 13.3 (4.58–19.3) years in the PanIN 2 cohort before transitioning to PanIN 3 (women 13.9 (4.83–20.3) years), and 9.0 (2.83–12.8) years in the PanIN 3 cohort before transitioning to undetected PDAC (women 10.0 (3.25–14.3) years). PDAC remained undetected for 2.56 (0.92–3.50) years (men) and 2.65 (1.00–3.58) years (women). If we include those subjects who died of other causes prior to progression, the time spent in the PanIN 1, 2, and 3 cohorts was 22.3 (9.42–32.8) years, 16.4 (6.00–24.1) years, and 10.0 (3.25–14.3) years for men) and 23.7 (10.3–34.9) years, 17.5 (6.50–25.7) years, and 11.3 (3.75–16.3) years for women, respectively. In total, men who developed PDAC spent 33.6 (24.0–42.6) years from the time they developed their first PanIN lesion until a PDAC diagnosis (women 35.3 (25.6–44.7) years). This dwell time analysis is summarized in Table 2.

Table 2.

Dwell time: Average time spent in each PanIN state.

	All Femalesa	Females Who Progress to the Next Stateb	All Malesa	Males Who Progress to the Next Stateb
Time in PanIN 1 State	23.7	17.4	22.3	16.5
Average years (IQR Q1-Q3)	(10.3–34.9)	(6.50–25.6)	(9.42–32.8)	(6.00–24.3)
Time in PanIN 2 State	17.5	13.9	16.4	13.3
Average years (IQR Q1-Q3)	(6.50–25.7)	(4.83–20.3)	(6.00–24.1)	(4.58–19.3)
Time in PanIN 3 State	11.3	10.0	10.0	9.0
Average years (IQR Q1-Q3)	(3.75–16.3)	(3.25–14.3)	(3.25–14.3)	(2.83–12.8)
Time in Undetected PDAC		2.65		2.56
Average years (IQR Q1-Q3)		(1.00–3.58)		(0.92–3.50)
PanIN 1 to Detected PDAC		35.3		33.6
Average years (IQR Q1-Q3)		(25.6–44.7)		(24.0–42.6)
PanIN 3 to Detected PDAC		12.3		11.3
Average years (IQR Q1-Q3)		(5.75–16.8)		(5.25–15.4)

^aThe “All Females” and “All Males” columns show the average (IQR Q1-Q3) amount of time spent in the state for all subjects who enter that state, whether they progress to the next state or not.

^bThe “Females Who Progress to the Next State” and “Males Who Progress to the Next State” columns show the average (IQR Q1-Q3) amount of time spent in each state for those subjects who eventually progress to the next state.

We tested the uncertainty surrounding PanIN prevalence at ages 50 and 80 using high and low scenarios. In the low scenario, the lifetime prevalence of a PanIN was decreased to 55%, with the result that the percentage of people with PanIN who eventually developed cancer increased to 1.7%. In this low scenario, the time from PanIN 1 development to Detected PDAC was slightly lower at 34.0 years, with each of the dwell times from PanIN 1 to cancer slightly shorter. In the high scenario, the lifetime prevalence of a PanIN was increased to 92%, with the result that the percentage of people with PanIN who eventually developed cancer decreased to 1.1%. The time from PanIN 1 development to Detected PDAC was slightly lower at 33.1 years, and each of the dwell times from PanIN 1 to cancer was again slightly shorter. The overall impact of changing PanIN prevalence estimates on dwell time results was small.

3.5. Exploratory analysis results: hypothetical “perfect” test and treatment

Our hypothetical perfect test-and-treat strategy for PanIN 3 lesions that allowed for continuous 100% risk reduction would have an expected benefit of 40 and 39 additional days of life for men and women, respectively. If such a strategy were applied only once (at age 50, 60, or 70), the expected risk reduction would drop to 12%, 23%, or 28% (ages 50, 60 or 70) for men and 11%, 23%, or 30% (ages 50, 60 or 70) for women. This would provide an attenuated life expectancy benefit of 7, 10, or 7 days (ages 50, 60, or 70) for men; 7, 10, or 8 days (ages 50, 60, or 70) for women. When introducing the risk of surgical mortality, the life expectancy benefit drops by approximately half a day across all cases. These trade-offs are summarized in Table 3.

Table 3.

Hypothetical “test and treat” intervention.^a.

Reduction in Lifetime Risk of PDAC (%)b	Female	Male
One time testing at age 50	−11%	−12%
One time testing at age 60	−23%	−23%
One time testing at age 70	−30%	−28%
Continuous testing	−100%	−100%
Life Expectancy Gains (days)
With 0% Surgical Mortality c	Female	Male
One time testing at age 50	6.9	7.1
One time testing at age 60	10	10
One time testing at age 70	8.8	7.9
Continuous testing	39	40
With 2% Surgical Mortality d	Female	Male
One time testing at age 50	6.4	6.7
One time testing at age 60	9.2	9.3
One time testing at age 70	7.7	7.0
Continuous testing	35	37

^aAll benefits of the intervention are averaged across the entire simulated population.

^bIn the first section, we show the average reduction in the lifetime risk of PDAC in the simulated population given a hypothetical perfect PanIN 3 test followed by perfect risk-reduction for future PDAC.

^cIn the second section, we show the average change in life expectancy for the simulated population under each hypothetical “test and treat” scenario, with the assumption that there is no surgical mortality.

^dIn the third section, we show the average change in life expectancy for the simulated population under the same hypothetical “test and treat” scenarios, with the addition of 2% surgical mortality.

4. Discussion

We found that PanIN 3 lesions are likely to have a higher probability of progression relative to PanIN 1 and 2 lesions, and to progress more rapidly. This would imply that PanIN 3 lesions, in addition to having a higher degree of dysplasia, are on average also more aggressive. We estimated the average dwell time from Undetected to Detected PDAC in our model to be just over three years, defining a potential window for cancer screening. It is important to note that not all undetected PDAC lesions may be within the detection threshold for imaging modalities, therefore the realistic window for clinical detection may be shorter. Also, the majority of PDACs are initially diagnosed within the pancreas or liver, rather than in vulnerable locations such as the CNS that might drive detection at a smaller size. This period of time is similar to the dwell time estimated for lung cancer in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) dataset (male smokers: 1.82 years, female smokers: 1.84 years) [22] and for colorectal cancer from the German screening cohort (4.5–5.8 years for all age groups in both men and women) [23]. Finally, the results of our model indicate a relatively long period from the development of PanIN 3 to malignancy (on average 11.3 years for men and 12.3 years for women). If true, this would allow a broad window for early detection.

Our hypothetical test-and-treat intervention – a test strategy that is perfect for the detection of PanIN 3 lesions – resulted in a gain of 9 days of life expectancy if applied once at age 60, about 30% of the value of a perfect test-and-treat applied continuously. Such an intervention would allow for the detection of a large number of pre-cancerous lesions prior to the onset of PDAC. At the age of 60, just under 50% of the population would have a PanIN lesion, but very few (0.1%) would have developed cancer. The mortality of surgical resection approaches 2% in most case series [19–21], and carries ongoing morbidity due to impaired insulin production, pancreatic insufficiency, and post-operative discomfort [24,25]. However, this must be weighed against the fact that most PDAC is discovered at an advanced stage when curative resection is not possible [26]. By way of comparison, colorectal cancer screening with colonoscopy has been shown to increase life expectancy 170 days among 50–54 year olds, and 41 days among 70–74 year olds [27]. Mammography has been estimated to increase life expectancy anywhere from 4–36 days depending on the frequency and age of screening [28]. These results indicate that the benefit of detecting and intervening on PanIN lesions may be within a reasonable range for a public health intervention if the performance characteristics of such a test were sufficiently high.

Our analysis has specific limitations that merit mention. First, we did not explicitly model the development of individual PanIN lesions, but rather the presence of at least one PanIN lesion within each cohort. It has been observed that some PDAC are associated with a relatively larger number of associated PanINs; the correlation between the number of PanINs and the relative risk of PDAC is unknown [29]. Our analysis does not allow for consideration of variability in the number of PanIN lesions, or of the correlation between the number of PanINs and the rate of progression. Second, it has been proposed that the medical community move from a three-tiered classification of PanIN lesions to a two-tiered system. In this revised description, PanIN 1 and 2 lesions (per WHO 2010) are grouped as low-grade PanINs, and PanIN 3 is classified as high-grade. This distinction was suggested to reflect the relatively low malignant potential of PanIN 1 and 2 lesions, relative to the higher risk behavior of PanIN 3 lesions [30]. In our analysis, we needed to retain the three-tiered classification because it was used in our primary data sources [4,6]. However, our results can be easily interpreted in the new classification schema by combining the PanIN 1 and PanIN 2 groups.

Finally, our analysis assumes an average-risk population. Several subpopulations of individuals have varied levels of elevated risk for PDAC. For example, smoking, type II diabetes, BRCA2 mutation positivity, and Peutz-Jeghers disease have been shown to confer risks for PDAC that are 2x, 2x, 3x, and 69x that of the average population [31–34]. Such individuals ultimately are likely to benefit more from PDAC screening than are average risk patients [17,35,36]. The current study does not address differences in the natural history of solid-derived PDAC or PanIN precursor lesions that are present in individuals at higher risk, and does not make corresponding projections in such subpopulations. However, the presented methods and analysis can be used to develop such projections in future work.

In conclusion, our simulation model describes the development of PDAC from PanIN lesions. PanIN lesions are a precursor to most PDAC, so knowledge of the dwell times between PanIN onset and the development of PDAC is critical to shaping future, successful screening strategies. We have shown that given this relatively long period for screening, an intervention that could detect and mitigate the risk from PanIN lesions could provide a population health benefit.