Genetic Testing to Guide Screening for Pancreatic Ductal Adenocarcinoma: Results of a Microsimulation Model

Abstract

Background:

First-degree relatives (FDRs) of patients with pancreatic ductal adenocarcinoma (PDAC) have elevated PDAC risk, partially due to germline genetic variants. We evaluated the potential effectiveness of genetic testing to target MRI-based screening among FDRs.

Methods:

We used a microsimulation model of PDAC, calibrated to Surveillance, Epidemiology, and End Results (SEER) data, to estimate the potential life expectancy (LE) gain of screening for each of the following groups of FDRs: indviduals who test positive for each of eight variants associated with elevated PDAC risk (e.g., BRCA2, CDKN2A); individuals who test negative; and individuals who do not test. Screening was assumed to take place if LE gains were achievable. We simulated multiple screening approaches, defined by starting age and frequency. Sensitivity analysis evaluated changes in results given varying model assumptions.

Results:

For women, 92% of mutation carriers had projected LE gains from screening for PDAC, if screening strategies (start age, frequency) were optimized. Among carriers, LE gains ranged from 0.1 days (ATM+ women screened once at age 70) to 510 days (STK11+ women screened annually from age 40). For men, LE gains were projected for all mutation carriers, ranging from 0.2 days (BRCA1+ men screened once at age 70) to 620 days (STK11+ men screened annually from age 40). For men and women who did not undergo genetic testing, or for whom testing showed no variant, screening yielded small LE benefit (0 – 2.1 days).

Conclusions:

Genetic testing of FDRs can inform targeted PDAC screening by identifying which FDRs may benefit.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal malignancy, often diagnosed at a late stage when no longer surgically resectable.¹ There is significant interest in methods of early detection and prevention to increase survival. Germline genetic mutations have been identified in 4–19% of patients with PDAC.^2–8 The role of genetic testing in pancreatic cancer is evolving, and recent National Comprehensive Cancer Network (NCCN) and American Society of Clinical Oncology (ASCO) guidelines have proposed germline genetic testing for all patients with PDAC.^{9, 10}

PDAC risk factors are challenging to identify. Family history moderately elevates PDAC risk relative to the general population (relative risk (RR) = 1.7–4.6); the extent of such elevation depends on the number of affected family members.^{11, 12} Several known germline genetic pathologic variants that are associated with increased PDAC risk include Lynch syndrome (MLH1, MSH2, MSH6, PMS2, EPCAM)^{13, 14}, hereditary breast and ovarian cancer (HBOC; BRCA1, BRCA2, PALB2)^{3, 15–18}, Peutz-Jeghers syndrome (STK11)¹⁹, and CDKN2A.¹⁸, ^20–25

The Cancer of the Pancreas Screening (CAPS) Consortium guidelines recommend annual screening via EUS and/or MRI starting at age 50 or 55 for patients with ATM, BRCA1, BRCA2, PALB2, or Lynch syndrome mutations who also have a first-degree family member with PDAC. Patients with STK11 or CDKN2A mutations are also recommended to undergo screening regardless of family history and starting at a younger age (30–40).²⁶ This is a small group of patients, and the extent to which screening within this group is implemented is currently unknown. Recent clinical studies of screening with EUS or MRI in high-risk populations have had varied results. Some studies have shown a high rate of false-positive results with attendant potential harms.^{27, 28} However, a Danish study showed an EUS-based screening strategy to be both clinically effective and cost-effective in patients with hereditary pancreatitis and first-degree relatives of patients with familial pancreatic cancer.²⁹

Using disease-modeling techniques, we have previously shown that MRI-based screening for PDAC may provide life expectancy (LE) benefits in some high-risk populations with a genetic predisposition.^{30, 31} In the current study, we utilized an updated disease model that simulates the natural history of PDAC to evaluate the potential for genetic testing to improve screening decisions and life expectancy in first-degree relatives of PDAC patients.

Methods

PDAC Simulation Model: An Overview

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) cancer registry and the published literature.^{1, 28, 30–45} The current analysis was performed on an expanded and updated version of this model. The model, written in C++, 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. Each cohort is assigned a relative risk of PDAC. Each model cycle, an individual may remain in their current health state (e.g., PanIN3) or progress to a new health state (e.g., undetected Stage I cancer) on the basis of calibrated transition probabilities.

We specifically simulate progression to PDAC through cystic and solid pathways. In our model, 10% of PDAC develops via the cystic pathway, and 90% via the solid pathway.⁴⁰ In the cystic pathway, patients may develop low-risk cysts, which may progress to high-risk cysts, undetected cancer, or detected cancer. Cancers are detected at AJCC Stages I-IV. The prevalence of cysts in our model (age-specific) was based on a study by Laffan et al, in which cysts were detected by CT.⁴⁶ Prevalence estimates were adjusted to account for false-negative cyst detection, as below (see Screening Strategies). The proportion of high-risk cysts was based on the proportion of patients with cysts who underwent surgical resection in a single-arm multimodal screening study, as previously described.^{31, 33} For the purposes of the simulation, data from unstaged cancers were excluded. In the solid pathway, patients may develop PanIN precursor lesions (PanIN 1–3), which then may progress to the same AJCC stages. These unobservable transitions are calibrated using multiple observable targets, as described below and in a prior publication.⁴⁵ Once diagnosed, patients are subject to stage-dependent mortality rates derived from SEER data.³⁶

Calibration of the Simulation Model

Some transition probabilities address events that cannot be observed, such as a PanIN 1 to PanIN 2 transition, and were estimated through calibration. In the calibration process, the model is used to estimate unobservable variables and calculates the resulting development of PDAC in the simulated population, comparing the results to a set of known targets. Iterative adjustment of the unobservable variables allows the calculated results to get closer to the set of targets, finding an optimal solution. The χ² goodness of fit (GOF) metric was used to identify optimal values for each variable, using previously described methods.^{45, 47, 48}

Calibration targets for the simulation model are shown in Table 1 and include: the lifetime risk of PDAC, PDAC incidence by age, the lifetime risk of PDAC-associated death, the incidence of pancreatic cysts by age, the lifetime risk of pancreatic cysts, the proportion of cysts that are high-risk vs low-risk, PanIN prevalence by grade at ages 50 and 80, and stage of PDAC at diagnosis.^{1, 33, 36, 38–41} To simulate cohorts with increased risk for PDAC, key calibration targets –the lifetime risk of PDAC, and the lifetime risk of PDAC-associated death – are multiplied by the cohort’s elevated relative risk (relative to an average-risk population), and the transitions between pre-cancerous states are recalibrated. Age-specific incidence of PDAC and of cysts are removed from the calibration targets at higher PDAC risk levels, as the age distribution of cysts and cancer are likely to be different at higher risk levels. The simulation model is recalibrated at each PDAC risk level up to a relative risk of 50 with the reduced set of calibration targets.

Table 1:

Calibration Targets and Model Inputs

Calibration Target / Model Input	Female	Male	Source
Lifetime prevalence of PDAC (%)	1.17	1.48	1, 36
Incidence of PDAC (age 20–29) (#/100K)	0.15	0.05	1
Incidence of PDAC (age 30–39)	0.49	0.53	1
Incidence of PDAC (age 40–49)	2.6	3.2	1
Incidence of PDAC (age 50–59)	9.7	13.5	1
Incidence of PDAC (age 60–69)	26.2	35.9	1
Incidence of PDAC (age 70–79)	49.3	60.3	1
Proportion of PDAC from solid lesions (%)	90	90	40
Prevalence of PanIN 1 lesions (age 50) (%)	20	20	38, 39
Prevalence of PanIN 1 lesions (age 80)	60	60	38, 39
Prevalence of PanIN 2 lesions (age 80)	15	15	38, 39
Prevalence of PanIN 3 lesions (age 80)	2	2	38, 39
Lifetime prevalence of pancreatic cysts (%)	4.19	4.19	46
Prevalence of cysts (age 20–29) (%)	0	0	46
Prevalence of cysts (age 30–39) (%)	0	0	46
Prevalence of cysts (age 40–49) (%)	2.22	2.22	46
Prevalence of cysts (age 50–59) (%)	2.51	2.51	46
Prevalence of cysts (age 60–69) (%)	4.67	4.67	46
Prevalence of cysts (age 70–79) (%)	6.77	6.77	46
Prevalence of cysts (age 80–89) (%)	14.21	14.21	46
Proportion of pancreatic cysts considered high-risk (%)	5.95	5.95	33
Proportion of PDAC diagnosed at Stage I (%)	8.75	7.40	1
… at Stage II	30.2	28.8	1
… at Stage III	10.1	9.47	1
… at Stage IV	50.9	54.3	1
Lifetime risk of PDAC-associated death (%)	1.08	1.39	1, 36
Annual mortality rates (%)	Age-specific	Age-specific	78
PDAC mortality from detected Stage I disease (monthly probability of death)	0.02403	0.02449	1
…from detected Stage II disease	0.03601	0.03508	1
…from detected Stage III disease	0.05816	0.05665	1
…from detected Stage IV disease	0.07466	0.07498	1
MRI/EUS Sensitivity (%)	62	62	27–29, 32, 33, 37, 42, 43, 49–56
MRI/EUS Specificity (%)	96	96	27–29, 32, 33, 37, 42, 43, 49–56
Surgical Mortality (False-Positive Cases) (%)	2.0	2.0	34, 35, 41, 44

Screening Strategies

In our analysis, screening was implemented either once at a given age, or annually from a given age until 75. Screening was assumed to include MRI with referral to EUS in case of any positive findings. We updated our prior work to take into account new studies and information, and we assumed 62% sensitivity and 96% specificity for MRI/EUS detection of Stage I-III cancers; these values were also applied to cystic precursors, incorporating false-negative designations of low-risk cysts, high-risk cysts, and cancers, as well as the probability of low-risk cysts being misclassified as high-risk cysts, as previously.^{27–29, 31–33, 37, 42, 43, 49–56} For metastatic malignancy, perfect detection was assumed. If the simulated patient had a low-risk cystic lesion detected at the time of screening, we assumed 10 years of annual surveillance, overriding the timing and frequency of the initially planned screening strategy.⁵⁷ During this annual surveillance period, we assumed perfect detection of progression to a high-risk cyst. If a high-risk cystic was detected, either at initial screening or during annual surveillance of a low-risk cyst, the patient was assumed to undergo surgical resection (with a 2% surgical mortality^{34, 35, 41, 44}). If PDAC was detected, PDAC-associated mortality was assigned based on SEER data for the appropriate stage of disease.¹ To avoid the potential penalty of lead-time, we applied mortality based on the stage at detection but not until the age at which the simulated patient would have been detected in the absence of screening.

Genetic Mutations

Germline pathogenic variants have been identified in 4–19% of patients with PDAC regardless of family history of PDAC or other malignancy.^{2–8, 58–60} We therefore estimated the overall frequency of germline mutations in the general population of patients with PDAC to be 9%, the median of these varied studies. We assumed a 4.5% chance of a genetic test finding a pathologic variant among FDRs of PDAC patients. The RR of PDAC for individuals with a single FDR diagnosed with PDAC is estimated to be 2.5.^{11, 12} The relative prevalence of each genetic mutation and the RR of PDAC for each mutation was estimated from the literature (Table 2).^{3, 4, 11, 12, 14–25, 60–74} Given the wide range of reported results, often with a small number of reported cases, we opted to combine information from these studies to develop a composite estimate of risk of disease. If a study specific to a mutation cohort was available, we estimated the RR of PDAC for that mutation in this more specific and larger study. Where this was not feasible, we estimated the RR of PDAC via a weighted average of cases from all available studies. Given the relative risk of PDAC for each mutation among the 4.5% of FDRs who test positive, the 95.5% of FDRs who test negative were calculated to have a RR of PDAC of 2.35 in order for the overall average RR of PDAC for an FDR to be 2.5.

Table 2:

Genetic mutation assumptions

Germline Genetic Mutation	Prevalence among PDAC Patients	Relative Risk of PDAC (Range)	Source
ATM	1.0 %	2.4 (2.41 – 5.71)	73, 74
BRCA1	0.7 %	1.6 (1.0 – 2.8)	3, 15, 16, 63, 68, 73, 79
BRCA2	2.4 %	3.2 (2.13 – 6.8)	3, 17, 63, 68, 73, 79
CDKN2A	2.0 %	12.7 (12.3 – 47.8)	21–25, 66, 73
Lynch Syndrome (MLH1, MSH2, MSH6, PMS2, EPCAM)	1.5 %	5.9 (5 – 10.7)	14, 73, 80
PALB2	1.0 %	3.2 (2.37 – 4)	81
STK11	0.01 %	24 (up to 132)	19
TP53	0.4 %	6.3 (3 – 7.3)	69, 73
Overall	9.0 %		3, 4, 11, 12, 14–25, 60–74. 79–81

Decision Model

Figure 1 illustrates the decision model structure. The decision model describes 10 simulated populations of FDRs, each described by genetic testing outcomes. One modeled population did not undergo genetic testing (no-test). Another modeled population underwent genetic testing with no pathologic mutation found (test-negative arm). Other populations underwent genetic testing and found one of the eight modeled mutations (test-positive arm). These outcomes, described by the RR of PDAC and a proposed screening strategy (e.g., one-time screening at age 50) are then used as inputs to the microsimulation model.

Figure 1: Decision Model structure.

A first-degree relative (FDR) of a patient with PDAC may undergo – or not undergo – genetic testing. If the FDR undergoes genetic testing, the FDR could test positive or test negative for a germline mutation. If the FDR tests positive for one of the eight specified germline mutations, he or she would enter the simulation model with assumptions for PDAC risk that are based on the specific mutation found (Table 2).

The output of the microsimulation model is the average life expectancy (LE) of simulated populations for each of the 10 potential outcomes. Note that if screening did not provide a LE gain (i.e., LE with screening is lower than the LE without), screening is not applied in our analysis. The LE gain represented for each simulation population derives from earlier detection of precancerous and cancerous lesions, predominantly a stage shifting effect, weighed against the potential surgical mortality associated with resection of false-positive findings.

Our base-case screening strategy is one-time screening at age 50. Consider the top-most branch of the decision tree (Figure 1), a population of FDRs who opted for genetic testing and were found to have an ATM mutation. Based on a summary of literature, the RR of PDAC for FDRs with an ATM mutation is 2.4 (Table 2). We therefore ran the simulation model assuming a population with a RR for PDAC of 2.4, applying a screening strategy of one-time at age 50 to determine whether there was an associated LE gain. If LE with screening was higher than LE without, we assumed that population would be screened. If not, we assumed that population would not be screened.

We tested 17 screening strategies in addition to the base-case strategy of one-time screening at age 50. One-time screening was evaluated in individuals aged 35–75, in 5-year increments (i.e., we evaluated one-time screening at age 30, at age 35, etc.). Annual screening, from the year of initiation to age 75, was evaluated in individuals aged 30–75 in 5-year increments (i.e., beginning at age 30, at age 35, etc.). For each genetic testing outcome, the screening strategy that yielded the highest LE gain (LE with screening minus LE without screening) was considered the optimal screening strategy. If no strategy yielded a LE gain, we assumed no screening would take place.

Sensitivity Analysis

Sensitivity analysis was performed for several key assumptions. We varied the assumed RR of PDAC for an FDR from 1.0 to 5.0 (with corresponding recalculation of the RR of test-negative patients). We varied assumptions regarding the probabilities of specific genetic mutations (Table 2). We grouped the 8 most common mutations into three categories: low risk (ATM and BRCA1); medium risk (BRCA2, PALB2, Lynch syndrome, and TP53); and highest risk (CDKN2A and STK11). In the base case, the proportion of test-positive individuals in each group was 19% (low risk), 59% (moderate), and 22% (high). We varied the probability of being in the highest-risk group from 0% to 40%, renormalizing the probabilities of the other risk groups. We varied screening sensitivity from 50–100% and specificity from 95–100%. We simultaneously varied specificity and surgical mortality to better evaluate the effects of false-positive screen results. We tested a range of surgical mortality (consequent to false-positive test results) from 1–5%. We tested the impact of stopping screening at age 70, 75, or 80. Finally, we tested the uncertainty of the calibration process by completing 100 calibration runs with different randomly generated starting parameters to determine if the calibration process yielded similar results.

Results

Base Case: One-time screening at age 50

We found that one-time screening at age 50 provided a LE gain for men with a RR of PDAC > 2.7, and for women with a RR of PDAC > 3.9.

Among male test-positive FDRs, six of the eight common genetic mutations evaluated had a RR of PDAC > 2.7, and therefore a LE gain associated with screening, ranging from 1.3 days (BRCA2, RR=3.2) to 79 days (STK11, RR=24). ATM and BRCA1 had RR below the threshold for LE gain from screening. In total, 81% of test-positive male FDRs were projected to achieve LE gains if screened. The average LE gain for all test-positive male FDRs was 9.5 days. Test-negative and no-test male FDRs had RR of PDAC < 2.7, below the threshold for LE gain from screening. In total, 4% of all male FDRs who underwent genetic testing would be recommended to undergo one-time screening at age 50. The average LE gain among all male FDRs who underwent genetic testing was 0.43 days.

Among female test-positive FDRs, only four of the eight common genetic mutations evaluated had a LE gain associated with screening, ranging from 4.5 days (Lynch syndrome, RR=5.9) to 61 days (STK11). 44% of test-positive female FDRs were projected to achieve LE gains if screened. The average LE gain for all test-positive female FDRs was 6.0 days. Test-negative and no-test female FDRs had no LE gain from screening. In total, 2% of female FDRs who underwent genetic testing would be recommended to undergo one-time screening at age 50. The average LE gain among all female FDRs who underwent testing was 0.27 days.

These results are shown in the first column of Tables 3 and 4 for men and women, respectively.

Table 3:

Life Expectancy (days) Gained for Selected Screening Strategies (Men)

Scenario	Base Case: One-Time Screening Age 50	One-time Screening Age 60	Annual Screening Age 50	Optimal Screening Strategy	Optimal Screening Strategy
	LE Gain in days	LE Gain in days	LE Gain in days		LE Gain in days
ATM	(0.8)	1.8	(21)	One-time age 65	1.9
BRCA1	(2.8)	(0.3)	(40)	One-time age 70	0.2
BRCA2	1.3	3.8	(1.2)	Annual age 60	8.7
CDKN2A	31	32	240	Annual age 45	260
Lynch	9.0	12	68	Annual age 50	68
PALB2	1.3	3.8	(1.2)	Annual age 60	8.7
STK11	79	66	500	Annual age 40	620
TP53	10	13	80	Annual age 50	80
Test Positive (Average)	9.5	11	69		78
Test Negative	(1.0)	1.6	(22)	One-time age 65	1.9
Testing (Average)	0.4	2.1	3.1		5.3
No testing	(0.6)	2.0	(18)	One-time age 65	2.1

This table shows a subset of the 17 potential screening strategies tested. One-time screening could be performed from ages 30 to 75 in 5-year increments. Annual screening from the age of initiation to age 75 could be performed starting from ages 30 to 70 in 5-year increments.

Table 4:

Life Expectancy (days) Gained for Selected Screening Strategies (Women)

Scenario	Base Case: One-Time Screening Age 50	One-time Screening Age 60	Annual Screening Age 50	Optimal Screening Strategy	Optimal Screening
	LE Gain in days	LE Gain in days	LE Gain in days		LE Gain in days
ATM	(3.3)	(0.6)	(46)	One-time age 70	0.1
BRCA1	(4.9)	(2.2)	(63)	No screening	0
BRCA2	(1.6)	1.1	(29)	One-time age 65	1.5
CDKN2A	22	24	180	Annual age 45	190
Lynch	4.5	7.4	31	Annual age 55	34
PALB2	(1.6)	1.1	(29)	One-time age 65	1.5
STK11	61	54	430	Annual age 40	510
TP53	5.6	8.5	42	Annual age 55	43
Test Positive (Average)	6.0	7.6	49		52
Test Negative	(3.4)	(0.7)	(47)	No screening	0
Testing (Average)	0.3	0.3	2.2		2.4
No testing	(3.1)	(0.3)	(43)	One-time age 65	0.3

This table shows a subset of the 17 potential screening strategies tested. One-time screening could be performed from ages 30 to 75 in 5-year increments. Annual screening from the age of initiation to age 75 could be performed starting from ages 30 to 70 in 5-year increments.

Additional Screening Strategies

We tested 17 screening strategies in addition to the base case strategy of one-time testing at age 50, including one-time and annual screening strategies at a range of ages (see Methods).

For male FDRs, the no-screening strategy would be selected for RR of PDAC < 1.5. All of the eight common genetic mutations evaluated had a LE gain associated with screening, ranging from 0.22 days (BRCA1, one-time screening at age 70) to 620 days (STK11, annual screening starting at age 40). Test-negative FDRs achieved a small LE gain with a one-time screen at age 65. No-test male FDRs also achieved a small LE gain with a one-time screen at age 65. While all test-positive male FDRs could achieve LE gains from screening, optimizing the frequency and starting age would yield an average LE gain of 78 days. The average LE gain among FDR males who underwent genetic testing was 5.3 days. If we subtract the LE gain among no-test men (2.1 days), the benefit of genetic testing for all male FDRs was 3.2 days if using optimal screening stratgies. This is an improvement on the base-case screening strategy of a one-time screen at age 50, for which the LE gain for genetic testing was 0.43 days (as described above). A schematic of the decision model with these values is shown in Figure 2 for illustration.

Figure 2: Optimal Screening Strategy (Men).

For men, all of the eight common genetic mutations evaluated had a positive LE gain associated with screening, ranging from 0.2 days (BRCA1, one-time screening at age 70) to 620 days (STK11, annual screening starting at age 40). Test-negative men achieved a small LE gain of 1.9 days with a one-time screen at age 65. No-test men achieved a small LE gain of 2.1 days with a one-time screen at age 65. Overall, all test+ men would undergo screening, with an average LE gain of 78 days. The average LE gain among tested men was 5.3 days (4.5% * 78 + 95.5% * 1.9). If we subtract the LE gain among no-test men of 2.1 days, the net benefit of genetic testing was 3.2 days if using optimal screening strategies.

For female FDRs, the no-screening strategy would be selected for RR of PDAC < 2.35. Seven of the eight common genetic mutations evaluated had a LE gain associated with screening, ranging from 0.11 days (ATM, one-time screen at age 70) to 510 days (STK11, annual screening starting at age 40). Test-negative FDRs had RR of PDAC just below the threshold for testing. No-test FDRs achieved a small LE gain with a one-time screen at age 65. Overall, 92% of test-positive female FDRs could achieve LE gains from screening, and optimizing the frequency and starting age would yield an average LE gain of 52 days. The average LE gain among FDR females who underwent genetic testing was 2.4 days. If we subtract the LE gain among no-test women (0.26 days), the benefit of genetic testing for all female FDRs was 2.1 days. This is an improvement on the base-case screening strategy of a one-time screen at age 50, for which the LE gain for genetic testing was 0.27 days (as described above). A schematic of the decision model with these values is shown in Figure 3.

Figure 3: Optimal Screening Strategy (Women).

For women, seven of the eight common genetic mutations evaluated had a positive LE gain associated with screening, ranging from 0.1 days (ATM, one-time screen at age 70) to 510 days (STK11, annual screening starting at age 40). Test-negative women no LE gain with screening. No-test women achieved a small LE gain of 0.3 days with a one-time screen at age 65. Overall, 92% of test+ women would undergo screening, with an average LE gain of 52 days. The average LE gain among tested women was 2.4 days (4.5% * 52). If we subtract the LE gain among no-test women of 0.3 days, the net benefit of genetic testing was 2.1 days if using optimal screening strategies.

Sensitivity Analysis

Likelihood of Mutation and RR of PDAC:

We varied the probability of having one of the highest risk mutations (CDKN2A and STK11) from 0% to 40%. If the probability of these highest risk mutations was 0%, the LE benefit of genetic testing under the base-case screening strategy dropped from 0.43 to 0.15 days (men) and from 0.27 to 0 days (women). If the probability was 40%, the LE benefit of genetic testing rose to 0.62 days (men) and 0.38 days (women).

As our estimates cover a range of relative risk from 1.6 to 24, should some mutations prove to have higher or lower risk of disease, an alternative case could be applied to estimate screening value. STK11 in particular has been proposed to have a much higher relative risk of PDAC that would further enhance screening value; at a RR of 50, annual screening starting at age 35 would be recommended, with a LE benefit of 1,300 days (men) and 1,100 days (women).

We varied the assumed RR of PDAC for no-test FDRs from 1 to 5 (base-case RR = 2.5). Varying RR of PDAC for no-test FDRs from 1 to 2.5 did not change the analysis results. At RR of PDAC = 5, all FDRs would undergo screening regardless of genetic testing results, therefore the overall benefit of genetic testing decreased, as the information gained from the genetic testing would be minimal. Within this range of relative risk of PDAC, there was no improvement in the LE benefit of screening.

Screening Sensitivity and Specificity, and Surgical Mortality:

We varied MRI sensitivity from 50% to 100%; this did not substantially change the LE benefit of genetic testing. Screening specificity was varied from 95% to 100%. At 100% specificity, screening provided LE gains at all RR levels, and the benefit of genetic testing therefore dropped to 0.20 days (men) and 0.17 days (women). Surgical mortality was varied from 1% to 5%; at a surgical mortality rate of 5%, only the highest risk groups would undertake screening (44% of test-positive male and 23% of test-positive female FDRs, respectively), with lower LE gains (1.6 days (men), 0.85 days (women)). At a 1% surgical mortality rate, all test-positive male and female FDRs would undertake screening, with slightly lower incremental LE gains (1.2 days (men), 1.5 days (women)).

We varied screening specificity and surgical mortality (for false-positive test results) together to test the downside of a false-positive screen. If screening specificity is low but surgical mortality high, fewer test-positive male and female FDRs would undertake screening, and the LE gain of genetic testing would drop substantially. Sensitivity analysis results are shown in Table 5.

Table 5:

Sensitivity Analysis

			Men		Women
MRI/EUS Sensitivity (%)	MRI/EUS Specificity (%)	Surgical Mortality (%)	FDRs with Genetic Mutations Screened (%)	LE Gain of Testing and Screening (days)	FDRs with Genetic Mutations Screened (%)	LE Gain of Testing and Screening (days)
62 %	96 %	2 %	100 %	3.2	92 %	2.1
50 %	96 %	2 %	92 %	3.1	81 %	2.3
75 %	96 %	2 %	100 %	3.2	92 %	2.1
100 %	96 %	2 %	100 %	3.1	100 %	2.1
62 %	95 %	2 %	92 %	3.4	81 %	2.6
62 %	100 %	2 %	100 %	0.20	100 %	0.17
62 %	96 %	1 %	100 %	1.2	100 %	1.5
62 %	96 %	2.5 %	92 %	2.7	81 %	1.9
62 %	96 %	5 %	44 %	1.6	23 %	0.85
50 %	95 %	2 %	92 %	3.3	81 %	2.5
50 %	100 %	2 %	100 %	0.18	100 %	0.13
100 %	95 %	2 %	100 %	3.5	92 %	2.5
100 %	100 %	2 %	100 %	0.23	100 %	0.15
62 %	95 %	1 %	100 %	1.2	100 %	1.8
62 %	100 %	1 %	100 %	0.19	100 %	0.17
62 %	95 %	5 %	44 %	1.6	23 %	0.90
62 %	100 %	5 %	100 %	0.15	100 %	0.14

The top row of the table shows the base case set of assumptions and results. If we test alternative assumptions shown in the first three columns, the changes in the results are shown in the following four columns. In the second row, in which we varied MRI/EUS sensitivity to 50% (from the base case assumption of 62%), leaving all other assumptions the same, the change in results for men was: small drop in the percentage of FDRs screened, LE gain of testing and screening essentially unchanged. The most consequential changes are in surgical mortality, and in particular surgical mortality when varied with specificity of MRI/EUS.

Discussion

We found that the use of genetic testing may increase the value of MRI/EUS screening for PDAC, and that screening strategies may be optimized based on genetic testing results and individuals’ RR of PDAC. Our analysis suggests that genetic testing can narrow the population of screening candidates, improving feasibility. In addition, our analysis generates hypotheses concerning optimal screening strategies for different cohorts of high-risk individuals.

Multiple recent papers have documented the germline mutation rate among unselected PDAC patients to be at least 4%,^2–8 and guidelines encourage genetic testing for newly diagnosed PDAC patients.^{9, 10} Our analysis focuses on the family members of PDAC patients rather than the patients themselves, as the primary value of genetic testing may be in the identification and screening of high-risk family members. Finally, it takes into account the potential negative effects of screening, i.e., false-positive results and associated surgical interventions.

Genetic testing can have two beneficial effects – flagging high-risk individuals who should be screened, and removing lower-risk individuals from the screening population. Based on our results, on average, FDRs of PDAC patients would gain only limited benefit from universal screening, but testing identified a subset of FDRs who could benefit from screening. Although the magnitude of potential LE gain from screening varied across sensitivity analysis assumptions, genetic testing added LE gains throughout almost the entire range of possible RR assumptions. If we compare these results to other cancer screening programs, the USPSTF cites a range of 27–56 days LE gain from mammography for women over 50,⁷⁵ and 25 days LE gain from low-dose CT for lung cancer screening for those who qualify.⁷⁶ This is similar to the magnitude of benefit our analysis would suggest for the higher-risk genetic mutations.

There are limitations to this analysis. Any simulation is a rough approximation of reality, dependent on the assumptions made. Most observable precancerous lesions are cysts, but solid lesions make up the majority of PDAC. We have used an assumption that 10% of PDAC derives from cystic lesions.⁴⁰ If the true rate of cystic PDAC was higher, the value of screening would also be higher, as MRI is most useful for identifying and following cystic lesions. To simulate the natural history of cysts in our model, we used CT imaging data to inform cyst prevalence,⁴⁶ and surgical resection as a proxy for defining high-risk cysts.³³ We did so in the absence of robust histologic/pathologic data to inform these estimates. Our use of CT data (rather than MRI) may have resulted in underestimating cyst prevalence in our model. In a recent study by Kromrey et al reporting cyst prevalence seen on MRI⁷⁷, the proportion of reported cysts was substantially higher; however, a large proportion was very small (<5 mm) and may not have represented cystic precursors to cancer. Further understanding of the natural history of pancreatic cysts, stratified by size, will be critically important for determining the true value of PDAC screening in all populations. The performance of MRI as a screening modality for PDAC has not been established in randomized controlled trials. As screening practices evolve, it is hoped that information about the natural history of precursor lesions and their surveillance will be improved. Further, our knowledge of the genetic risk factors for PDAC and the relative risk associated with these genetic mutations continues to evolve. Our evaluation does not explicitly take into account competing risks of mortality specific to the genetic mutations addressed, which could attenuate the LE benefit of pancreatic cancer screening in these populations. In addition, some of the targets available for model calibration (cyst prevalence, PanIN prevalence) are not easily available by sex, such that the differentiation of results between men and women may be less accurate.

Of note, this analysis does not take into account other factors of consideration in screening for cancer. The surgical mortality associated with resection is included, but not the more chronic health impacts of living with diabetes or pancreatic insufficiency. The patient anxiety associated with knowing genetic testing results, or of being part of a screening program, are not specifically incorporated. In particular, the costs associated with testing, screening, and cancer diagnosis are not addressed in this analysis.

In conclusion, we have developed a microsimulation model of the natural history of pancreatic precursor lesions and PDAC in a variety of populations at varied risk for PDAC, and have applied a decision model of genetic testing and screening options to these populations. We have identified subgroups of high-risk individuals for whom screening may provide a LE benefit. In order to minimize the risk of surgical mortality from unnecessary interventions while maximizing diagnosis of culprit precursor lesions, we have hypothesized several different frequencies of screening that could be applied to different risk groups. Our evaluation shows that germline genetic testing of FDRs of PDAC patients could add LE benefit by identifying those high-risk family members who would most benefit from screening, and that the timing and frequency of screening can be adapted based on genetic testing results to maximize this benefit.