The age-dependent association of risk factors with pancreatic cancer

Abstract

Background:

Pancreatic cancer presents as advanced disease in >80% of patients; yet, appropriate ages to consider prevention and early detection strategies are poorly defined. We investigated age-specific associations and attributable risks of pancreatic cancer for established modifiable and non-modifiable risk factors.

Patients and methods:

We included 167 483 participants from two prospective US cohort studies with 1190 incident cases of pancreatic cancer during >30 years of follow-up; 5107 pancreatic cancer cases and 8845 control participants of European ancestry from a completed multicenter genome-wide association study (GWAS); and 248 893 pancreatic cancer cases documented in the US Surveillance, Epidemiology, and End Results (SEER) Program. Across different age categories, we investigated cigarette smoking, obesity, diabetes, height, and non-O blood group in the prospective cohorts; weighted polygenic risk score of 22 previously identified single nucleotide polymorphisms in the GWAS; and male sex and black race in the SEER Program.

Results:

In the prospective cohorts, all five risk factors were more strongly associated with pancreatic cancer risk among younger participants, with associations attenuated among those aged >70 years. The hazard ratios comparing participants with three to five risk factors with those with no risk factors were 9.24 [95% confidence interval (CI) 4.11–20.77] among those aged ≤60 years, 3.00 (95% CI 1.85–4.86) among those aged 61–70 years, and 1.46 (95% CI 1.10–1.94) among those aged >70 years (P_{heterogeneity} = 3×10⁻⁵). These factors together were related to 65.6%, 49.7%, and 17.2% of incident pancreatic cancers in these age groups, respectively. In the GWAS and the SEER Program, the associations with the polygenic risk score, male sex, and black race were all stronger among younger individuals (P_{heterogeneity} ≤0.01).

Conclusions:

Established risk factors are more strongly associated with earlier-onset pancreatic cancer, emphasizing the importance of age at initiation for cancer prevention and control programs targeting this highly lethal malignancy.

INTRODUCTION

Pancreatic cancer is the third leading cause of cancer-related death in the USA, with 5-year survival of only 10%.¹ While 50% of pancreatic cancers are diagnosed after age 70 years,² a rising incidence of pancreatic cancer has been noted particularly in younger populations.^3–5 Due to the lethality of pancreatic cancer, younger cases contribute greatly to the societal burden of this disease by causing substantial loss of potential life years.⁶ In contrast to other major cancers (e.g. colon, lung), only 30%−40% of pancreatic cancers are attributable to genetic predisposition and lifestyle factors,^7,8 complicating prevention and early detection of this highly fatal disease.

Cancer development is characterized by a stepwise accumulation of driver gene mutations leading to malignant transformation. Notably, mistakes made during normal stem cell division may contribute a substantial fraction of driver mutations, including for pancreatic cancer.^9,10 As people advance in age, an increasing number of these mutations develop proportional to the lifetime number of stem cell divisions within an organ. Consequently, pancreatic cancer that develops at an older age may harbor fewer driver events caused by inherited and lifestyle factors and more from random mutations, such that inherited and lifestyle factors may be more strongly associated with development of earlier-onset versus later-onset pancreatic cancer. Several prior studies have suggested an enrichment of risk factors for pancreatic cancer among younger cases,^11–14 which partly supports the differential etiology of pancreatic cancer by age at diagnosis.

In the current study, we comprehensively investigated age-specific associations and attributable risks of pancreatic cancer for inherited and lifestyle risk factors, leveraging detailed epidemiological data from two prospective US cohort studies with repeated exposure measurements over several decades, a completed genome-wide association study (GWAS) of pancreatic cancer, and the Surveillance, Epidemiology, and End Results (SEER) Program. We demonstrate an age-dependent pattern for nearly all pancreatic cancer risk factors, whereby associations with pancreatic cancer were strongest among younger participants and substantially weakened as individuals aged.

METHODS

Study populations

Prospective cohort studies.

We examined risk factors for pancreatic cancer in two prospective US cohorts: the Nurses’ Health Study (NHS)¹⁵ that enrolled 121 700 female registered nurses aged 30–55 years in 1976 and the Health Professionals Follow-Up Study (HPFS)¹⁶ that enrolled 51 529 male health professionals aged 40–75 years in 1986. Since enrollment, participants were biennially mailed questionnaires inquiring about their medical history and lifestyle factors. We excluded participants with prior history of cancer at enrollment for an analysis population of 167 483 participants. Incident cases of pancreatic cancer were identified from self-report or during follow-up of participant deaths ascertained through reports from the next of kin, the US Postal Service, or the National Death Index.¹⁷ Physicians confirmed the diagnosis by reviewing medical records, death certificates, or cancer registry data. Cases with non-adenocarcinoma histology were excluded. The study protocol was approved by the institutional review boards of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health, and those of participating registries as required.

Pancreatic cancer GWAS.

We examined a polygenic risk score (PRS) for pancreatic cancer in 5107 pancreatic cancer cases and 8845 control participants of European ancestry from a completed multicenter GWAS, PanScan/PanC4 I-III, led by the Pancreatic Cancer Cohort Consortium and Pancreatic Cancer Case-Control Consortium (Supplementary Methods, available at https://doi.org/10.1016/j.annonc.2022.03.276)^18–20 Each participating study obtained written informed consent from participants and approval from the local institutional review board.

The SEER Program.

We examined incidence rates of pancreatic cancer by sex and race in SEER 21,²¹ which collected cancer incidence data from 21 population-based cancer registries since 2000, covering 36.7% of the US population. Using SEER*Stat (version 8.3.8), we selected pancreatic cancers diagnosed at ages of ≥30 years without non-adenocarcinoma tumors, resulting in 248 893 cases diagnosed during 2000–2017 (Supplementary Methods, available at https://doi.org/10.1016/j.annonc.2022.03.276).

Exposure assessment

In the prospective cohorts, we collected data on five established risk factors for pancreatic cancer, cigarette smoking,²² obesity,²³ diabetes,²⁴ tall height,²³ and non-O blood group,²⁵ as well as three suspected risk factors, inadequate exercise,²³ excessive alcohol use,²⁶ and poor diet quality²⁷ (Supplementary Figure S1, available at https://doi.org/10.1016/j.annonc.2022.03.276). Established risk factors were defined as follows: (i) at cohort enrollment, participants were asked detailed information about their current and past smoking. After enrollment, information on smoking status and smoking intensity was reported biennially. Pack-years of smoking were computed to assess cumulative exposure to cigarettes. (ii) Current weight was reported at enrollment and biennially thereafter. Body mass index (BMI) was calculated as weight in kilograms divided by square of height in meters. For main analyses, the cumulative average of BMI was computed by using all available questionnaires until the previous survey cycle. Obesity was defined as a BMI of ≥30 kg/m². (iii) At enrollment and biennially thereafter, participants were asked if they had diabetes diagnosed by a physician. Because a subset of pancreatic cancer patients develop diabetes before their cancer diagnosis due to a subclinical pancreatic tumor,²⁸ we only considered diabetes for >2 years as a risk factor. (iv) Height was reported at enrollment. Tall height was defined as a height of ≥167 cm for women and ≥181 cm for men, which were the 75th percentiles of height in the National Health and Nutrition Examination Survey 1999–2016.²⁹ (v) In 1996, participants were asked their blood group: O, A, B, AB, or unknown, which has been validated previously.²⁵ Information on assessment of the suspected risk factors is provided in Supplementary Methods, available at https://doi.org/10.1016/j.annonc.2022.03.276.

Genotyping and imputation methods in the PanScan/PanC4 I-III GWAS have been described previously.²⁰ We calculated a PRS for pancreatic cancer by summing risk alleles in 22 previously identified single nucleotide polymorphisms (SNPs) in European populations,^{18–20,30–32} weighted by reported effect sizes (Supplementary Table S1, available at https://doi.org/10.1016/j.annonc.2022.03.276)³³ In the SEER Program, sex and race were obtained from the registry databases.

Statistical analyses

We categorized participants by age into three groups (≤60, 61–70, and >70 years), or two groups (≤70 and >70 years) when sample size was inadequate, as in the USA, about 20% of pancreatic cancers develop before age 60 years, and 50% before age 70 years.² Random-effects meta-regression was used to test for heterogeneity of the association magnitude between oldest and youngest age groups.³⁴

In the prospective cohorts, person-years for participants were calculated from when the exposure was first surveyed until pancreatic cancer diagnosis, death, or end of follow-up, whichever came first. During follow-up, participants could contribute person-years to different age groups (Supplementary Figure S2, available at https://doi.org/10.1016/j.annonc.2022.03.276). The Andersen-Gill data structure was used to handle time-varying covariates, which created a record for every 2-year survey cycle at which a participant was at risk, with covariates updated at the time of questionnaire return. Cox proportional hazards regression was used to examine age-stratified associations between risk factors and pancreatic cancer risk. To avoid arbitrary determination of age cut-offs, we also used an alternative to Cox regression with age as the time scale and modeled the association as a smooth function of age approximated by B-spline (R, version 4.0.3)³⁵ Logistic regression and Poisson regression were used to examine associations of pancreatic cancer risk with the PRS in the GWAS and demographic factors in the SEER Program, respectively. Information on covariate adjustment is provided in Supplementary Methods, available at https://doi.org/10.1016/j.annonc.2022.03.276.

In the prospective cohorts, we computed age-specific population attributable fractions (PAFs) for the proportion of incident pancreatic cancers in the study population that theoretically would not have occurred if the risk factor had been absent.³⁶ To provide contemporary estimates for potentially modifiable factors, we also derived PAFs by using contemporary, nationally representative data on exposure prevalence (Supplementary Methods, available at https://doi.org/10.1016/j.annonc.2022.03.276). All analyses were carried out with SAS, version 9.4 (SAS Institute Inc., Cary, NC), unless specified otherwise. A two-sided P < 0.05 indicated statistical significance.

RESULTS

Risk factors in the prospective cohorts

During approximately 5.0 million person-years of follow-up in a study population of 118 148 women from NHS and 49 335 men from HPFS, 1190 incident cases of pancreatic cancer were identified, including 146 cases aged ≤60 years (incidence rate, 6 per 100 000 person-years), 336 aged 61–70 years (24 per 100 000 person-years), and 708 aged >70 years (60 per 100 000 person-years). Table 1 shows age-specific characteristics for study participants.

Table 1.

Age-specific characteristics among participants from two prospective cohorts

Characteristica	Full study population			Pancreatic cancer casesb
	≤60 years	61–70 years	>70 years	≤60 years	61–70 years	>70 years
Women/NHS
No. person-years or cases	2 002 737	1 070 675	817 529	110	223	435
White, %	96.8	96.8	96.7	97.3	96.4	96.8
Ever-smoking, %	56.1	56.0	53.9	67.3	63.8	52.1
Obesityc, %	10.1	13.9	14.4	17.6	20.3	18.2
Diabetes, >2 years, %	2.1	7.3	12.8	10.8	9.5	21.6
Tall heightd, %	35.1	34.1	32.4	44.5	36.9	33.6
Men/HPFS
No. person-years or cases	364 923	357 963	363 730	36	113	273
White, %	90.8	90.7	90.7	88.9	86.7	93.0
Ever-smoking, %	46.0	51.8	55.4	70.6	63.9	59.4
Obesityc, %	9.2	10.6	8.9	19.4	11.8	7.1
Diabetes, >2 years, %	1.5	5.0	9.4	9.1	6.8	16.9
Tall heightd, %	35.7	32.3	26.2	55.6	33.6	24.5
	≤70 years		>70 years	≤70 years		>70 years
Women/NHS
Non-O blood group, %	56.9		56.9	69.7		62.1
Inadequate exercisee, %	42.7		39.4	49.6		41.0
Excessive alcohol usef, %	12.8		12.6	15.3		15.9
AHEI-2010, mean (SD)	51 (10)		53 (9)	51 (9)		53 (9)
Men/HPFS
Non-O blood group, %	57.6		56.4	71.2		59.0
Inadequate exercisee, %	29.4		27.6	40.9		25.3
Excessive alcohol usef, %	10.8		11.7	15.5		16.0
AHEI-2010, mean (SD)	53 (11)		56 (10)	54 (10)		56 (11)

AHEI, alternate healthy eating index; HPFS, Health Professionals Follow-Up Study; MET, metabolic equivalent; NHS, Nurses’ Health Study; SD, standard deviation.

^aObesity, inadequate exercise, excessive alcohol use, and AHEI-2010 assessed by the cumulative average of body mass index, physical activity, alcohol consumption, and AHEI-2010, respectively.

^bTime-varying variables assessed at the time of cancer diagnosis.

^cDefined as a body mass index of ≥30 kg/m².

^dDefined as a height of ≥167 cm for women and ≥181 cm for men (i.e. the 75th percentiles of height in the National Health and Nutrition Examination Survey 1999–2016).

^eDefined as aerobic exercise of <75 vigorous-intensity or 150 moderate-intensity minutes (i.e. 500 MET-minutes/week), according to the Physical Activity Guidelines for Americans, 2nd edition.

^fDefined as alcohol use of >1 drink/day for women and >2 drinks/day for men, according to the 2015–2020 Dietary Guidelines for Americans.

Association between risk factors and pancreatic cancer

To test the hypothesis that risk factors may be more strongly associated with pancreatic cancer developed at younger ages, we first examined age-stratified associations of pancreatic cancer risk with five established risk factors: cigarette smoking, obesity, diabetes, tall height, and non-O blood group (Table 2).

Table 2.

Age-stratified association of pancreatic cancer risk with established risk factors among participants from two prospective cohorts

Risk factor	Age ≤60 years			Age 61–70 years			Age >70 years			P heterogeneity b
	Person-years	No.cases	Hazard ratio (95% CI)a	Person-years	No.cases	Hazard ratio (95% CI)a	Person-years	No.cases	Hazard ratio (95% CI)a
Cigarette smoking
Never	1 056 294	46	1 [Reference]	629 200	119	1 [Reference]	528 219	311	1 [Reference]
Ever	1 265 695	98	1.84 (1.29–2.62)	764 156	210	1.45 (1.16–1.82)	619 032	378	1.04 (0.89–1.21)	0.004
Former: ≤39 pack-years	689 728	45	1.40 (0.93–2.12)	500 702	109	1.12 (0.87–1.46)	425 156	240	0.96 (0.81–1.14)
Former: ≥40 pack-years	39 285	5	1.79 (0.70–4.55)	83 192	31	1.69 (1.13–2.52)	113 928	79	1.10 (0.86–1.41)
Current: ≤39 pack-years	415 911	28	2.34 (1.44–3.79)	80 955	24	1.87 (1.20–2.92)	28 333	15	1.04 (0.62–1.76)
Current: ≥40 pack-years	120 770	20	3.38 (1.98–5.79)	99 306	46	2.83 (2.00–4.01)	51 615	44	1.58 (1.14–2.17)
Obesityc
No	1 857 150	109	1 [Reference]	1 218 066	274	1 [Reference]	1 025 127	606	1 [Reference]
Yes	201 523	24	1.85 (1.19–2.88)	183 973	58	1.41 (1.06–1.87)	146 839	98	1.18 (0.95–1.46)	0.07
Diabetes, >2 years
No	2 231 090	121	1 [Reference]	1 239 278	278	1 [Reference]	948 837	514	1 [Reference]
Yes	41 165	14	3.85 (2.07–7.18)	88 214	26	1.12 (0.73–1.72)	124 844	127	1.89 (1.54–2.33)	0.03
Tall heightd
No	1 531 714	77	1 [Reference]	946 917	215	1 [Reference]	822 601	495	1 [Reference]
Yes	833 464	69	1.68 (1.21–2.33)	480 195	120	1.12 (0.90–1.41)	357 342	213	1.04 (0.88–1.22)	0.01
		Age ≤70 years					Age >70 years
		Person-years	No. cases	Hazard ratio (95% CI)a			Person-years	No. cases	Hazard ratio (95% CI)a
ABO blood group
O		345 376	45	1 [Reference]			307 014	160	1 [Reference]
Non-O		459 562	106	1.84 (1.30–2.61)			402 869	251	1.18 (0.97–1.44)	0.03
A		295 242	64	1.73 (1.18–2.54)			250 853	144	1.09 (0.87–1.37)
B		108 326	27	1.95 (1.20–3.14)			91 365	68	1.40 (1.05–1.86)
AB		55 994	15	2.19 (1.21–3.94)			60 651	39	1.22 (0.85–1.73)

CI, confidence interval.

^aFrom Cox proportional hazards regression conditioned on age (in months) and calendar year of the survey cycle and sex/cohort and adjusted for race/ethnicity (white, black, other, or unknown), pack-years of smoking (never, ≤19, 20–39, ≥40, or missing; unadjusted when cigarette smoking was examined), and body mass index in kg/m² (<25.0, 25.0–29.9, 30.0–34.9, ≥35.0, or missing; unadjusted when obesity was examined).

^bHeterogeneity between the age groups of ≤60 years (or ≤70 years for non-O blood group) and >70 years tested by using the random-effects meta-regression method.

^cDefined as a body mass index (cumulative average) of ≥30 kg/m².

^dDefined as a height of ≥167 cm for women and ≥181 cm for men (i.e. the 75th percentiles of height in the National Health and Nutrition Examination Survey 1999–2016).

For cigarette smoking, the hazard ratios (HRs) comparing ever smokers to never smokers were 1.84 [95% confidence interval (CI) 1.29–2.62] among participants aged ≤60 years, 1.45 (95% CI 1.16–1.82) among those aged 61–70 years, and 1.04 (95% CI 0.89–1.21) among those aged >70 years (P_{heterogeneity} = 0.004; Table 2). Among participants aged >70 years, only current smokers who had smoked ≥40 pack-years in their lifetime had an increased risk for pancreatic cancer. To assess cumulative tobacco exposure with different latency periods before cancer diagnosis, we examined lifetime pack-years, as well as pack-years during or beyond the most recent 10 years (Supplementary Table S2, available at https://doi.org/10.1016/j.annonc.2022.03.276). Regardless of latency period, cigarette smoking was more strongly associated with pancreatic cancer risk among participants aged ≤60 years than those aged >70 years.

As with cigarette smoking, obesity and diabetes were more strongly associated with pancreatic cancer risk at a younger age (P_{heterogeneity} = 0.07 and 0.03, respectively; Table 2). For obesity and diabetes, the HRs were 1.85 (95% CI 1.19–2.88) and 3.85 (95% CI 2.07–7.18) among participants aged ≤60 years, in contrast to 1.18 (95% CI 0.95–1.46) and 1.89 (95% CI 1.54–2.33) among those aged >70 years. As an individual’s BMI can vary greatly over the life course, we examined obesity at different timepoints in life and restricted the study population to participants who were enrolled before age 60 years (Supplementary Table S3, available at https://doi.org/10.1016/j.annonc.2022.03.276). In all these analyses, obesity was more strongly associated with earlier-onset than late-onset pancreatic cancer.

We next examined tall height, which was associated with increased risk of pancreatic cancer among participants aged ≤60 years (HR, 1.68; 95% CI 1.21–2.33), but not among older participants (P_{heterogeneity} = 0.01; Table 2). Due to an inadequate number of cases aged ≤60 years, the association with ABO blood group was examined by pooling participants aged ≤70 years. Non-O blood group was more strongly associated with pancreatic cancer risk among participants aged ≤70 years (HR, 1.84; 95% CI 1.30–2.61) than those aged >70 years [HR, 1.18 (95% CI 0.97–1.44); P_{heterogeneity} = 0.03; Table 2].

When considering a combinatorial score that sums the five established risk factors based on presence or absence of each individual factor, we observed a pronounced attenuation with age in the association with pancreatic cancer risk. The HRs comparing participants with three to five risk factors to those with no risk factors were 9.24 (95% CI 4.11–20.77) among those aged ≤60 years, 3.00 (95% CI 1.85–4.86) among those aged 61–70 years, and 1.46 (95% CI 1.10–1.94) among those aged >70 years (P_{heterogeneity} = 3 × 10⁻⁵; Figure 1A). Such attenuation with increasing age was similar by sex/cohort and birth cohort (Supplementary Table S4, available at https://doi.org/10.1016/j.annonc.2022.03.276). We also estimated the HR per additional risk factor as a smooth function of age, whereby the association again weakened as individuals aged (Figure 1B).

Figure 1. Age-specific association of pancreatic cancer risk with combinatory risk factors among participants from two prospective cohorts.

Five established risk factors for pancreatic cancer were considered: ever-smoking, obesity, >2-year diabetes, tall height, and non-O blood group. Participants with missing information on more than one of the factors were excluded. (A) Age-stratified hazard ratios by number of risk factors from Cox proportional hazards regression conditioned on age (in months) and calendar year of the survey cycle and sex/cohort and adjusted for race/ethnicity (white, black, other, or unknown). Only lower limits of 95% CIs are shown. Heterogeneity between the age groups of ≤60 years and >70 years was tested by using the random-effects meta-regression method, comparing participants with 3–5 risk factors to those with no risk factors. (B) Age-varying hazard ratios per additional risk factor from Cox B-spline piecewise regression with age as the time scale and conditioned on sex/cohort and calendar year of the survey cycle and adjusted for race/ethnicity (white, black, other, or unknown). Solid curve represents point estimates and dashed curves represent 95% CIs.

CI, confidence interval.

In exploratory analyses, we examined several suspected risk factors for pancreatic cancer (Supplementary Table S5, available at https://doi.org/10.1016/j.annonc.2022.03.276). Notably, inadequate exercise was associated with pancreatic cancer risk among participants aged ≤70 years (HR, 1.37; 95% CI 1.11–1.69) but not among those aged >70 years [HR, 0.98 (95% CI 0.83–1.15); P_{heterogeneity} = 0.01]. In contrast, excessive alcohol use was not differentially associated with pancreatic cancer risk by age, and diet quality was not associated with pancreatic cancer risk in either age group.

Attributable risk of pancreatic cancer for established risk factors

We next examined the proportion of incident pancreatic cancers by age that could be attributed to the five established risk factors individually and in combination (Figure 2 and Supplementary Table S6, available at https://doi.org/10.1016/j.annonc.2022.03.276). PAFs were greater for earlier-onset versus late-onset pancreatic cancer for each of the risk factors, aside from diabetes, which increased in prevalence as individuals aged. The PAFs for combinatory risk factors were 65.6%, 49.7%, and 17.2% among participants aged ≤60 years, 61–70 years, and >70 years, respectively, indicating a greater proportion of earlier-onset pancreatic cancers that were attributable to these risk factors.

Figure 2. Age-specific proportion of incident pancreatic cancers attributable to established risk factors among participants from two prospective cohorts.

Population attributable fraction estimates the proportion of incident pancreatic cancers in the study population that theoretically would not have occurred if the risk factor had been absent and demographic factors (age, sex, and race/ethnicity) remained unchanged.

Prevention potential of pancreatic cancer

We next considered attributable risks by age for potentially modifiable factors, namely, ever-smoking, obesity, inadequate exercise, and excessive alcohol use (Supplementary Table S7, available at https://doi.org/10.1016/j.annonc.2022.03.276). Together, PAFs for these factors were 37.7% and 6.5% among those aged ≤70 years and >70 years, respectively. Estimates were similar when national prevalence of exposure was considered (Supplementary Table S8, available at https://doi.org/10.1016/j.annonc.2022.03.276). Given this finding, we also examined the ages at which individuals quit smoking or became obese in relation to pancreatic cancer risk (Supplementary Table S9, available at https://doi.org/10.1016/j.annonc.2022.03.276). Notably, smoking cessation at earlier ages was associated with more pronounced risk reduction for pancreatic cancer (P_trend = 0.001), and individuals who quit smoking before age 50 years had the same risk reduction as individuals who never smoked. Similarly, obesity onset before age 50 years rather than at later ages was associated with increased risk of pancreatic cancer. Furthermore, weight gain between 18–21 years and 50 years of age rather than between 50 and 65 years was associated with increased pancreatic cancer risk.

PRS in the PanScan/PanC4 I-III GWAS

We next examined whether a PRS that summarized the cumulative effect of 22 independent genome-wide significant SNPs would have differential associations with pancreatic cancer risk by age (Table 3). As with the risk factors described in the preceding text, the strongest association of the PRS with pancreatic cancer risk was noted among participants aged ≤60 years, with lesser associations identified in older participants (P_{heterogeneity} = 0.01). Comparing participants in the top versus bottom 10% of the PRS, the odds ratios for pancreatic cancer were 6.91 (95% CI 4.60–10.40) among those aged ≤60 years and 4.12 (95% CI 3.08–5.52) among those aged >70 years.

Table 3.

Age-stratified association of pancreatic cancer risk with polygenic risk score among cases and controls from the PanScan/PanC4 I-III genome-wide association study

Polygenic risk scorea	Age ≤60 years			Age 61–70 years			Age >70 years			P heterogeneity c
	No. cases	No. controls	Odds ratio (95% CI)b	No. cases	No. controls	Odds ratio (95% CI)b	No. cases	No. controls	Odds ratio (95% CI)b
Bottom 10% (low-risk)	46	189	1 [Reference]	75	405	1 [Reference]	100	290	1 [Reference]
>10%−50%	324	746	1.75 (1.20–2.54)	509	1577	1.48 (1.11–1.98)	608	1215	1.50 (1.15–1.96)
>50%−90%	567	726	3.35 (2.32–4.84	802	1610	2.20 (1.65–2.92)	979	1203	2.43 (1.88–3.16)
Top 10% (high-risk)	292	161	6.91 (4.60–10.40)	371	399	4.13 (3.02–5.66)	434	324	4.12 (3.08–5.52)
Per SD increase	1229	1822	1.71 (1.57–1.87)	1757	3991	1.50 (1.41–1.60)	2121	3032	1.50 (1.41–1.60)	0.01

CI, confidence interval; SD, standard deviation.

^aCategorization and standardization based on distribution among all control participants (score range: bottom 10%, 1.949–3.491; >10–50%, 3.492–4.121; >50–90%, 4.122–4.751; top 10%, 4.751–6.454).

^bFrom logistic regression adjusted for age subgroups (<51 or 51–60 years among those aged ≤60 years; 71–80 or >80 years among those aged >70 years), sex, and study site.

^cHeterogeneity between the age groups of ≤60 years and >70 years tested by using the random-effects meta-regression method.

Sex and race in the SEER Program

Male sex and black race are known risk factors for pancreatic cancer in the USA.^37,38 Thus, we examined whether these risk factors had differential associations with pancreatic cancer incidence by age (Supplementary Table S10, available at https://doi.org/10.1016/j.annonc.2022.03.276). As with the aforementioned exposures, the male-to-female incidence rate ratio (IRR) was greater in the age group of <60 years (1.41; 95% CI 1.39–1.44) than in the age group of ≥70 years (1.14; 95% CI 1.12–1.15; P_{heterogeneity} < 1 × 10⁻⁸). Similarly, black-to-white IRRs declined monotonically with age (P_{heterogeneity} < 1 × 10⁻⁸), as did joint associations when considering sex and race together.

DISCUSSION

More than 80% of patients with pancreatic cancer present with advanced disease, when cure is uncommon and the disease rapidly acquires resistance to available treatments. To make significant progress in patient outcomes, pancreatic cancer prevention and earlier detection will be critical. Nevertheless, how risk factors for pancreatic cancer change with age has not been systematically explored. In the current study, we show a remarkably consistent pattern across known pancreatic cancer risk factors, whereby risk factors play a relatively large role in development of earlier-onset pancreatic cancer and a limited role in late-onset disease. Furthermore, behavioral changes earlier in life appear to be required to reduce cancer risk from those factors that are modifiable, such as cigarette smoking and obesity. These results demonstrate the value of understanding how risk factors over a lifetime impact disease development, while emphasizing the importance of timing in public health measures for prevention and early detection.

Previous studies have suggested that pancreatic cancer cases with a predisposing family history or germline mutations,^11,13,14 cigarette smoking,¹³ or excess body weight,¹² had earlier age of onset than those without these factors. By utilizing four distinct, large, population-based studies, we examined more than 10 risk factors to estimate the association and attributable risk across a range of ages. In two prospective cohorts, we found stronger associations and greater attributable risk for pancreatic cancer among younger participants for each of the established risk factors assessed, including cigarette smoking, obesity, diabetes, tall height, and non-O blood group. In the GWAS, we utilized a large multi-institutional dataset to examine a weighted genetic risk score and again identified a stronger association among younger individuals. Finally, we examined male sex and black race in relation to pancreatic cancer incidence in a nationwide cancer database. Across these different patient populations and considering a multitude of different exposures, a similar pattern emerged, whereby risk factors were more strongly associated with pancreatic cancer risk at younger ages; that is, the relative risk conferred by a risk factor for a younger adult was greater than that for an older adult, although incidence rates for pancreatic cancer are higher in older adults.

These findings have several important implications. First, to facilitate disease prevention, interventions for modifiable risk factors will likely require implementation at younger ages, as these factors were not strongly associated with risk after age 70 years. Conversely, early detection programs may have greater potential to reduce pancreatic cancer mortality in older populations, as 50% of pancreatic cancer cases are diagnosed after age 70 years. Second, future studies to investigate risk factors and prediction models for pancreatic cancer will need to consider age not only as a risk marker, but also as a stratification variable that may modify the association or predictive ability of other factors. Third, given the relatively weak associations between traditional risk factors and late-onset pancreatic cancer, detection strategies that include early identification of tumor-induced symptoms or signs may be needed, such as with machine learning tools applied to patient records³⁹ or regular blood-based assessments.^40–42

This study has important strengths. In the two prospective cohorts, participants were followed longitudinally starting in mid-life for >30 years, minimizing selection bias and reducing the impact of exposures differing by birth cohort. Furthermore, modifiable factors were assessed not only at enrollment, but in a repeated manner throughout follow-up. Thus, we were able to assess participants’ exposure every 2 years and model-specific risk factors over their lifetime, rather than at a single timepoint. Because all data were collected prospectively and before disease development, misclassification and recall bias were minimized.

The current study also has limitations. Although we examined many of the known risk factors for pancreatic cancer, several were not available in our patient populations, including family history of pancreatic cancer and personal history of chronic pancreatitis. Although we examined race in the SEER dataset, the prospective cohorts and GWAS included predominantly white participants. In the SEER Program, we were unable to control for potential confounders as data were unavailable, such as cigarette smoking or BMI.

In conclusion, inherited and lifestyle factors are more strongly associated with earlier-onset pancreatic cancer, emphasizing the importance of considering age in design of cancer prevention and control programs for this highly lethal malignancy.

PanScan/PanC4 I-III Consortium:

The members are L. T. Amundadottir; E. Ardanaz; A. A. Arslan; L. E. Beane-Freeman; P. M. Bracci; B. Bueno-de-Mesquita; M. Du; S. Gallinger; G. G. Giles; P. J. Goodman; V. A. Katzke; A. P. Klein; C. Kooperberg; P. Kraft; D. Li; N. Malats; L. L. Marchand; M. L. McCullough; R. L. Milne; J. P. Neoptolemos; S. Perdomo; G. M. Petersen; H. A. Risch; X. O. Shu; R. Z. Stolzenberg-Solomon; S. K. Van Den Eeden; K. Visvanathan; E. White; B. M. Wolpin; and W. Zheng. Their affiliations appear in Supplementary Material, available at https://doi.org/10.1016/j.annonc.2022.03.276.