By Pooja Toshniwal PahariaReviewed by Lauren HardakerJan 28 2026A plasma protein signature accurately stratifies cancer risk in patients with diffuse symptoms, offering a potential blood-based triage tool to guide faster, more targeted diagnostic workups.
Image credit: Ken stocker/Shutterstock.com
A recent study published in Nature Communications reports promising cancer risk stratification among individuals presenting with diffuse clinical symptoms using plasma protein profiling. The model distinguished cancer from non-malignant conditions with overlapping presentations and maintained performance in independent replication cohorts.
These findings suggest that proteome profiling could support triage and referral decisions for earlier diagnostic evaluation, potentially helping to reduce diagnostic delays.
Early cancer diagnosis remains challenging in individuals with broad or non-localized symptom profiles, as they often fall outside established organ-specific diagnostic pathways. In addition, population-based screening programs are largely limited to cervical, breast, and colorectal cancers. As a result, alternative quick-turnaround diagnostic strategies are needed to minimize delays in diagnosis.
However, referring all individuals with heterogeneous symptom presentations to advanced imaging modalities such as positron emission tomography-computed tomography (PET-CT) may overburden healthcare systems and expose individuals to unnecessary investigations. Readily accessible, minimally invasive blood-based markers capable of distinguishing cancer from other conditions in high-risk symptomatic populations could help refine referral decisions and facilitate earlier diagnosis.
Non-specific presentations challenge organ-based cancer diagnosis
In the present study, researchers designed a triage model to identify cancer patients with unclear clinical presentations using plasma protein profiling. The model was based on a 29-protein signature selected for inclusion in a penalized multivariable prediction model, identified using proximity extension assay-based proteomics.
For analysis, researchers obtained 1,463 plasma proteins from 456 individuals (discovery cohort from the MEDECA study, median age, 71 years, 55 % female) presenting with diffuse symptoms sampled before cancer diagnosis. Symptoms included extreme fatigue, malaise, poor appetite with unexplained weight loss, unexplained pain, prolonged fever, abnormal laboratory findings, and increased healthcare consultations.
Radiological findings suggested metastasis without a clinically apparent primary source of malignancy. These patients received referral for fast-track multidisciplinary cancer diagnostic evaluation at a hospital in Sweden. They underwent several biochemical investigations, imaging assessments, and diagnostic biopsies. They were followed for six months post-evaluation.
The team tested the model in an age- and sex-matched replication cohort (ALLVOS study) of 238 individuals with comparable symptoms. These individuals had received a referral to comparable rapid diagnostics at another hospital. Referral was based on similar, non-specific symptom criteria, but excluded individuals referred solely for radiological signs of malignancy without an identifiable primary tumor.
The researchers characterized the differentially expressed proteins as cancer-related, tissue-enriched, or secreted based on the Human Protein Atlas annotations. They used multivariate logistic regression with penalization to identify the proteome signature of cancer. Principal component analysis (PCA) enabled differentiation between cancer and non-cancer samples.
Protein signature stratifies cancer risk across diffuse symptoms
Proteomic analysis revealed a cancer-associated plasma protein signature in individuals exhibiting diffuse symptoms. Based on the proteomic signature, the model stratified cancer (n=160) and non-cancer (n=296) cases with an area under the receiver operating characteristic curve (AUC) of 0.80. Hematologic malignancies were the most prevalent (28 %), followed by adenocarcinomas of the pancreas, gallbladder, and bile ducts (11 %). Clinical assessments indicated metastatic disease in 102 of 115 individuals with solid malignancies.
The penalized regression-based model maintained performance (AUC, 0.82) in the replication cohort when distinguishing cancer from non-cancer conditions overall, including inflammatory diseases and infections, although discrimination varied by disease category. The team observed good inter-panel correlation for assessments applied as controls. Correlation coefficients for interleukin-6 (IL-6) and C-X-C motif chemokine ligand 8 (CXCL-8) ranged from 0.97 to 0.99 and were slightly lower for tumor necrosis factor (TNF), ranging from 0.95 to 0.96.
Over six months of follow-up, 35 patients received cancer diagnoses. The most frequently diagnosed cancers included hematologic malignancies, neuroendocrine tumors, and lung squamous cell carcinomas. Metastasis was detected in approximately half of individuals with solid malignancies.
Proteins with the strongest statistical associations with cancer status included ribonucleotide reductase subunit M2 (RRM2), poly ADP-ribose polymerase 1 (PARP1), keratin 19 (KRT-19), PC4 and SFRS1 interacting protein 1 (PSIP1), as well as additional high-importance features within the predictive model such as anterior gradient 2 (AGR2) and carcinoembryonic antigen-related cell adhesion molecule 5 (CEACAM5).
AGR2 promotes tumor growth by inhibiting p53 activity, while keratin-19 is associated with epithelial proliferation and radiation-resistant colorectal tumors. CEACAM5 contributes to tumor angiogenesis and metastatic progression and is widely used for monitoring gastrointestinal and other malignancies.
Deoxyribonucleic acid (DNA) repair-related proteins RRM2 and PARP1 have been linked to poor prognosis and are therapeutic targets in multiple cancers, whereas PSIP1 has been linked to increased tumorigenicity, particularly in breast cancer.
Blood-based triage may accelerate diagnosis in complex cases
The study findings indicate that plasma protein profiling may help prioritize high-risk individuals for rapid, sensitive diagnostic evaluation using a minimally invasive blood-based approach.
While the model is not yet ready for clinical implementation and was developed in cohorts with a high prevalence of advanced and metastatic disease, early identification through such triage tools could enable more timely investigation and reduce the burden of hard-to-detect cancers.
Future studies should assess performance in lower-prevalence primary care populations, compare outcomes with other blood-based diagnostic platforms, and evaluate clinical utility and cost-effectiveness to support translation into routine care.
Download your PDF copy now!
Journal Reference
Wannberg, F. et al. (2025). Plasma protein profiling predicts cancer in patients with non-specific symptoms. Nature Communications, 17(1), 151. DOI: 10.1038/s41467-025-67688-3, https://www.nature.com/articles/s41467-025-67688-3
AI-Designed CytoTape Records Cellular Events Over Time With High Resolution