Beyond sensitivity and specificity: Redefining the era connotation of “low-risk” in pancreatic cancer screening

Abstract

Pancreatic cancer remains a highly lethal malignancy, primarily due to late-stage diagnosis. Current screening paradigms, which focus exclusively on high-risk individuals, leave the vast “low-risk” population unscreened. This conventional binary risk stratification, based predominantly on family history and known genetic syndromes, fails to incorporate emerging risk dimensions such as polygenic risk scores, lifestyle factors, and novel biomarkers. We propose a paradigm shift from this static model towards a dynamic, multidimensional risk stratification framework. By integrating genetic susceptibility (e.g., newly identified variants in NOC2L, HNF4G), lifestyle metrics (e.g., new-onset diabetes), and liquid biopsy biomarkers (e.g., circulating tumor DNA, carbohydrate antigen 19-9 dynamics), we can reclassify a subset of “low-risk” individuals who may benefit from targeted screening. The integration of artificial intelligence for prospective validation, as seen in ongoing trials, is crucial for implementing this approach.

Key Words: Pancreatic cancer screening; Low-risk redefinition; Dynamic risk stratification; Precision prevention; Multimodal integration

Core Tip: This commentary argues that the conventional “low-risk” label for pancreatic cancer, based solely on family history and known genetic syndromes, is obsolete. We propose a paradigm shift towards a dynamic, multidimensional risk model. This framework continuously integrates polygenic risk scores, liquid biopsy biomarkers (e.g., circulating tumor DNA), lifestyle factors, and artificial intelligence to enable real-time, personalized risk stratification. This approach can reclassify a significant portion of the “low-risk” population, paving the way for precision screening and potentially transforming early detection strategies for this lethal malignancy.

INTRODUCTION

The most prevalent histological type of pancreatic cancer (PC) is pancreatic ductal adenocarcinoma (PDAC), which constitutes 80%-90% of all pancreatic malignancies[1]. The origin of this neoplasm is the epithelial cells that line the pancreatic ducts, leading to a high degree of invasiveness and a poor prognosis for patients[2,3]. PDAC is projected to become the second leading cause of cancer-related deaths by 2030, with a 5-year survival rate of only 12%[4,5]. This dismal outlook is largely a consequence of late diagnosis, highlighting the critical need for effective early detection strategies. However, current screening paradigms exclusively target high-risk individuals, leaving the vast “low-risk” general population without surveillance options[6,7], explicitly recommend against screening asymptomatic, non-high-risk individuals. This recommendation hinges on the low incidence of PDAC in the general population and the suboptimal positive predictive value of current screening tools.

The conventional definition of “low-risk” is predominantly based on the absence of known hereditary syndromes (e.g., breast cancer susceptibility gene 2, cyclin-dependent kinase inhibitor 2A, serine/threonine kinase 11 mutations) or a strong family history of PC[7,8]. However, this definition is increasingly outdated. Landmark studies, such as the cancer of pancreas screening trials, have demonstrated that screening high-risk individuals with magnetic resonance imaging and endoscopic ultrasonography can detect early-stage PDAC and its precursors, improving survival[9,10]. Meanwhile, advancements in molecular profiling have unveiled novel genetic risk variants beyond the classic syndromes[11-13]. For instance, a recent genome-wide association study identified five new PDAC susceptibility loci, each increasing risk by 15%-25%[12]. Concurrently, blood-based biomarkers, including circulating tumor DNA (ctDNA) and refined carbohydrate antigen 19-9 (CA19-9) dynamics, show promise for earlier detection[11,14]. This commentary posits that the era of binary risk stratification is obsolete. We advocate for a new, dynamic model that continuously integrates genetic, biomarker, and lifestyle data to redefine “low-risk” and enable precision screening[15].

THE LIMITATIONS OF THE TRADITIONAL “LOW-RISK” DEFINITION

The prevailing model for PC screening is not only binary but fundamentally static, relying on a one-time assessment of a limited set of criteria (Figure 1). Individuals are categorized as “high-risk” if they meet specific criteria, such as having two or more first-degree relatives with PC, carrying a pathogenic mutation in a gene like cyclin-dependent kinase inhibitor 2A or breast cancer susceptibility gene 2, or having Peutz-Jeghers syndrome[6,7,9]. While this approach effectively targets a narrow subset of the population, it overlooks up to 90% of PC cases that occur in individuals classified as “low-risk” due to the absence of these specific criteria[12].

Figure 1 Traditional pancreatic cancer screening pathways and their limitations. The core decision point in traditional screening models is a straightforward binary judgement process. A small number of individuals meeting stringent family history/genetic criteria are directed onto a high-risk screening pathway, receiving early detection (green arrow). The vast majority of individuals failing to meet these strict criteria are uniformly categorised as “low risk” and excluded from the screening system. The prominent yellow arrow clearly indicates the adverse outcome associated with this pathway, highlighting the core flaw of the traditional model: The failure to detect a substantial number of sporadic cases at an early stage. FPC: Familial pancreatic cancer; PJS: Peutz-Jeghers syndrome; CDKN2A: Cyclin-dependent kinase inhibitor 2A; BRCA2: Breast cancer susceptibility gene 2; MRI: Magnetic resonance imaging; EUS: Endoscopic ultrasound.

This approach is flawed for several reasons. First, a significant proportion of PDAC cases occur in individuals without a recognized family history or known genetic syndrome. Relying solely on these factors misses a substantial at-risk population. Second, the model fails to account for the cumulative impact of lower-penetrance genetic variants. The polygenic nature of PDAC risk means that an individual with several common risk variants, each conferring a modest increase in risk, may have a lifetime risk comparable to someone with a single high-penetrance mutation. Third, non-genetic factors, such as new-onset diabetes (NOD), chronic pancreatitis, and obesity, are not adequately integrated into the initial risk assessment[7,16]. NOD, in particular, is a well-established early indicator of PDAC, yet it is often not acted upon until symptoms develop.

EMERGING BIOMARKERS AND THE POTENTIAL FOR LIQUID BIOPSY

The limitations of current imaging modalities underscore the need for complementary, non-invasive tools. Liquid biopsy has emerged as a promising frontier, offering a multifaceted view of the disease through several complementary components: (1) ctDNA: The detection of kirsten rat sarcoma viral oncogene homolog and tumor protein 53 mutations in ctDNA boasts high specificity for PDAC. Emerging technologies now allow for the identification of these mutations in plasma and even duodenal fluids with improving and clinically informative sensitivity for early-stage disease[11,17]. Moreover, the presence of post-operative ctDNA is a powerful predictor of recurrence and poor survival[18,19]; (2) Novel protein and glycomic biomarkers: Multi-marker panels that combine CA19-9 with other analytes, such as mucin 1 glycoprotein or specific glycomic profiles, are under active investigation[14,20]. For instance, the proximity-assisted click-mediated antibody-neutralization assay, an artificial intelligence (AI)-driven model combined with CA19-9, reportedly achieved high sensitivity for stage I PDAC[21], highlighting the power of integrated diagnostic approaches; and (3) Multimodal blood tests: Prospective studies are evaluating dual-screening strategies that combine clinical risk models with multi-analyte blood tests (e.g., including circulating microRNAs, extracellular vesicles) to identify high-risk individuals within general population cohorts[22]. The true potential of these biomarkers lies not in single measurements but in dynamic monitoring. Serial assessment of ctDNA variant allele frequency or CA19-9 kinetics can provide unparalleled insights into tumor evolution, thereby serving as the cornerstone for a continuously updated, dynamic risk model.

INTEGRATING GENETIC BACKGROUND, LIFESTYLE, AND AI FOR A DYNAMIC MODEL

A modern risk stratification model must be multidimensional, quantifiable, and inherently dynamic, capable of evolving with new patient data (Figure 2). This involves several key components: (1) Polygenic risk scores (PRS): A PRS aggregates the effects of numerous common genetic variants into a single metric. An individual with a high PRS but no family history could be reclassified into a higher-risk category[23]; (2) Lifestyle and clinical factors: Factors like long-term smoking, high body mass index, and especially NOD after age 50 should be formally incorporated into risk algorithms[7]. The model should assign quantitative weights to these factors; and (3) AI and deep learning: AI serves as the essential computational engine for this model, capable of integrating these complex, non-linear data streams to generate predictive, dynamic risk scores[24-26]. For example, deep learning models applied to electronic health records have demonstrated high accuracy in predicting future cancer risk[27]. Such tools are ideally suited for the continuous re-evaluation required by our proposed framework.

Figure 2 Dynamic risk stratification model based on multi-dimensional data integration. This framework enables continuous acquisition of multidimensional data: Whole-genome polygenic risk scores, circulating tumor DNA/carbohydrate antigen 19-9 dynamics, and new-onset diabetes or lifestyle modifications. An artificial intelligence engine integrates these non-linear variables to generate a quantitative risk score that updates over time. Individuals are stratified in real-time into average-, medium-, or high-risk categories, triggering tiered interventions. Feedback from each downstream test or clinical event refines individual risk profiles and recalibrates the model. This adaptive design transforms screening from a disposable qualification process into a dynamic, evidence-based monitoring continuum, promising earlier detection while minimising over-testing. ctDNA: Circulating tumor DNA; KRAS: Kirsten rat sarcoma viral oncogene homolog; CA19-9: Carbohydrate antigen 19-9; MUC1: Mucin 1; BMI: Body mass index; AI: Artificial intelligence; EUS: Endoscopic ultrasound; MRI: Magnetic resonance imaging.

Implementing this model necessitates a shift in clinical practice. It begins with an initial risk assessment (e.g., PRS + baseline clinical data around age 50). This risk score is continuously updated with longitudinal data, such as the onset of NOD, a significant weight change, or a rise in CA19-9. When an individual’s dynamic risk score crosses a predefined threshold, it triggers eligibility for imaging surveillance (e.g., endoscopic ultrasonography/magnetic resonance imaging). This transforms screening from a one-time qualification into a continuous, evidence-based monitoring process. The following graph illustrates how these emerging biomarkers can be integrated with traditional and novel risk factors to create a dynamic risk model (Figures 1 and 2).

CHALLENGES AND FUTURE DIRECTIONS

Despite the promise of dynamic risk stratification, several challenges must be addressed before widespread clinical implementation. First, the ethical and psychological implications of reclassifying “low-risk” individuals based on genetic or biomarker data require careful consideration. Labeling patients as “intermediate” or “high-risk” may induce anxiety and lead to over-surveillance, underscoring the need for robust patient education and shared decision-making frameworks[28]. Second, the logistical and economic hurdles associated with genetic testing, repeated biomarker assays, and AI-based analytics remain substantial. However, cost-effectiveness analyses suggest that targeting screening to a redefined high-risk group could prove economically viable by focusing resources more efficiently[29]. Prospective validation in diverse populations is critical. Ongoing initiatives like the PC Early Detection (PRECEDE) Consortium and the PC Cohort Consortium (PanScan) are addressing this by developing multi-ethnic cohorts and adjusted risk algorithms[30]. Looking ahead, the convergence of continuous biomarker monitoring, AI-driven risk integration, and patient-specific screening intervals holds the potential to transform PC from a lethal disease into a detectable one. Future research should prioritize the standardization of assays, the refinement of AI algorithms through federated learning across institutions, and the execution of large-scale, risk-adaptive randomized trials to demonstrate the ultimate impact of dynamic risk stratification on PC mortality.

CONCLUSION

The definition of “low-risk” for PC is no longer fit for purpose in the modern era. Clinging to a binary, static model based solely on family history and a limited set of genetic syndromes inevitably misses opportunities for early detection in a significant portion of the population. The tools for a revolution are at hand: An expanding catalog of genetic risk variants, sophisticated liquid biopsy biomarkers, and powerful AI-based integration tools. By embracing a dynamic, multidimensional risk stratification model that continuously updates an individual’s risk based on genetics, biomarkers, and lifestyle, we can transition from a reactive to a proactive screening paradigm. This approach holds the promise of redefining “low-risk”, ultimately allowing us to extend the benefits of early detection to a broader population and finally change the trajectory of this devastating disease.