Abstract A123: AI-PACED: A prospective feasibility study of automated risk stratification, AI-augmented serial imaging, and biobanking for early detection of sporadic pancreatic cancer
Khurram Khaliq Bhinder, Baloy Jyoti. Talukdar, Angela Ammirabile, Armin Zarrintan, Ahmed Khan Jadoon, Subhosree Dey, Takeru Yamaguchi, Sovanlal Mukherjee, Ajit H. GoenkaAbstract
Purpose:
Sporadic pancreatic ductal adenocarcinoma (PDA) is diagnosed almost exclusively after symptom onset, often when curative resection is no longer feasible. Glycemically-defined new-onset diabetes (gNOD) with an ENDPAC score ≥3 identifies a population at markedly elevated short-term risk, and radiomics-based AI models have identified sub-visual, pre-diagnostic imaging signatures of PDA that precede symptomatic presentation. Translating these signatures into a prospective early detection pathway requires establishing recruitment yield, imaging and biospecimen adherence, and integration of a radiomics-based AI into real-world workflows. We designed the AI-Augmented Pancreas Cancer Early Detection (AI-PACED) study, a prospective, dual-cohort feasibility trial (NCT07324096) evaluating serial AI-analysis of CT and longitudinal blood biobanking in a gNOD, ENDPAC-enriched high-risk population.
Methods:
Individuals with gNOD, identified through weekly semi-automated EMR-based risk stratification and confirmed by manual chart review, self-select into Cohort A1 (serial CT and blood biobanking), Cohort A2 (blood biobanking alone), or passive EMR surveillance (Cohort B). Target enrollment is 100 (70 in A1/A2, 30 in Group B). Cohorts A1 and A2 undergo blood collection at baseline, 6, and 12 months. Cohort A1 additionally undergoes paired portal venous phase contrast-enhanced CT, concentrated during the first year after gNOD onset, when PDA risk is highest, to prioritize detection while limiting incremental radiation exposure. All CTs are interpreted by non-study radiologists per standard institutional protocol. The AI analyses is applied post-hoc, firewalled from clinical care. Feasibility endpoints include gNOD identification yield, recruitment, and cohort self-selection among eligible high-risk individuals. The protocol is IRB-approved and open to accrual at Mayo Clinic, Rochester, Minnesota since March 2026.
Results:
Semi-automated EMR screening identified 253 candidate individuals, of whom manual chart review confirmed 175 (mean age: 66.6 years, range 50-84; 50.9% females) as true gNOD, comprising 16 with ENDPAC ≥3, 139 with ENDPAC <3, and 20 in whom ENDPAC could not be calculated. The remaining 78 were excluded as false positives (e.g., active malignancy, inaccurate glycemic values). This confirmation step distinguishes true, actionable gNOD from EMR artifact prior to outreach.
Conclusions:
AI-PACED supports the feasibility of a two-stage algorithmic and manual workflow for distinguishing true gNOD from EMR artifact, while quantifying the false-positive burden that will inform refinement of a fully automated pipeline. Recruitment into asymptomatic, imaging-intensive surveillance is likely to remain the principal constraint, particularly because most glycemic elevation is managed outside quaternary centers. These constraints motivate extending enrollment infrastructure beyond a single center to inform a multi-institutional trial evaluating lead-time advantage of an AI-driven strategy for earlier, potentially curative detection of sporadic PDA.
Citation Format:
Khurram Khaliq Bhinder, Baloy Jyoti. Talukdar, Angela Ammirabile, Armin Zarrintan, Ahmed Khan Jadoon, Subhosree Dey, Takeru Yamaguchi, Sovanlal Mukherjee, Ajit H. Goenka. AI-PACED: A prospective feasibility study of automated risk stratification, AI-augmented serial imaging, and biobanking for early detection of sporadic pancreatic cancer [abstract]. In: Proceedings of the AACR Conference on Pancreatic Cancer: New Frontiers in Biology and Therapeutic Development; 2026 Sep 25-28; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(18_Suppl_2):Abstract nr A123.