Simple Algorithm for Blood Morphology Analysis (SAMBA): An Artificial Intelligence-Assisted Approach to Identify Peripheral Blood Phenotypes in a Cohort of Patients with Pancreatic Ductal Adenocarcinoma (PDAC)

Introduction: Eosinophils have become important therapeutic targets in several allergic diseases. This study aimed to apply a similar approach—using peripheral blood findings to identify features of type 2 (T2) inflammation—to oncology. We used Simple Algorithm for Blood Morphology Analysis (SAMBA) to describe blood count profiles in a cohort of patients with PDAC independently of relative ratios and propose a potential classification. Materials and Methods: We included 153 treatment-naïve patients with PDAC confirmed by postoperative histopathology. Of these, n = 68 (44.4%) underwent curative surgery, whereas n = 85 (55.6%) received palliative treatment or underwent surgical biopsy only. SAMBA was applied using laboratory reference ranges and overall survival (OS; months) as the analytical context. Results: The subgroups defined according to laboratory reference ranges did not differ significantly in disease stage, grade, or resection status. Based on correlation analyses, Kaplan–Meier curves, and a classification and regression tree (CART) analysis of OS, we propose four blood count phenotypes: exhausted, lymphopenic, T2-low, and T2-high (median OS: 1.5 months, n = 22; 4.5 months, n = 12; 8.0 months, n = 25; and 10.0 months, n = 94, respectively). Conclusions: We propose the following preliminary interpretations of these phenotypes: exhausted, potentially associated with concomitant infection or marked inflammation; lymphopenic, characterized by a low lymphocyte count and potentially reduced antitumor immune activity; eosinopenic (T2-low), characterized by a relatively balanced inflammatory profile but limited T2-associated features; and eosinophilic (T2-high), characterized by no major abnormalities in the other blood count parameters and a more favorable prognosis. SAMBA is a transparent, readily interpretable algorithm that may help standardize and accelerate preliminary research in immuno-oncology.