Surface-enhanced Raman scattering (SERS) nanoprobe-based immunoassay is an emerging liquid biopsy modality for the detection of blood biomarkers. However, quantitative detection of blood biomarkers using SERS nanoprobe immunoassays remains challenging, primarily due to intrinsic stochastic fluctuations of SERS signals. Herein, Ag nanogap shells (AgNGS) encoded with Raman labels are prepared as SERS nanoprobes with single-particle sensing resolution and long-term structural and signal stability. Then, digital SERS nanoprobe immunoassays using AgNGS nanoprobes and magnetic nanoparticles conjugated with detection and capture antibodies for carbohydrate antigen 19-9 (CA19-9) and apolipoprotein A1 (APOA1) are developed for their quantitative detection in the clinical serum samples of pancreatic cancer (PC). Furthermore, digitization of the SERS intensity-based assay outcomes into On/Off states allows significant improvement in limits of detection and quantification for biomarker detection. Finally, the digital AgNGS nanoprobe immunoassay enables quantitative detection of both CA19-9 and APOA1 from the 150 clinical serum samples of PC patients, which are integrated with machine learning analysis via bootstrap sampling and logistic regression modeling for the diagnosis of early-stage PC, achieving a high AUC of 0.988. AgNGS nanoprobe immunoassays offer an effective liquid biopsy route for profiling various blood biomarkers associated with many diseases.