Pancreatic Cancer Prediction Model May Reduce Unnecessary Biopsies
A pancreatic cancer prediction model using common blood markers has the potential to improve early detection and reduce unnecessary biopsies by six percent.
Source: Getty Images
By - Researchers have developed and externally validated a prediction model to detect early-stage pancreatic cancer using routinely collected blood biomarkers, according to a recent study published in JAMA Network Open.
The research team underscored that accurate risk prediction for pancreatic cancer could facilitate early detection and prevent unnecessary diagnostic tests, such as biopsies, for low-risk patients.
They further noted that diagnosing pancreatic cancer typically relies on a time-consuming, invasive combination of modalities: clinical symptoms, carbohydrate antigen 19-9 (CA19-9) serum levels, radiological findings, and pathologic confirmation of the diagnosis by fine-needle aspiration or brush cytology.
These tests still often yield inconclusive results, which must then be confirmed via biopsy or resection of a pancreatic tumor.
The complexity and invasive nature of these methods led the researchers to explore a more efficient approach that prevents unnecessary diagnostic measures and reduces time to treatment for pancreatic cancer patients.
Blood-based biomarkers, such as CA19-9 and bilirubin levels, are routinely measured in patients being screened for pancreatic cancer and may provide insights into cancer risk.
To evaluate this hypothesis, the research team developed a risk prediction model that leverages these biomarkers to discriminate between early-stage pancreatic cancer and benign periampullary diseases.
To do so, the researchers pulled data on adult patients with pancreatic cancer or benign periampullary disease treated from 2014 to 2022 at four academic hospitals in Italy, the Netherlands, and the United Kingdom. Serum levels of CA19-9 and bilirubin from each patient collected at diagnosis and prior to the start of medical intervention were evaluated.
The study sample included 249 patients in the development cohort and 296 in the validation cohort. Model performance was measured in terms of its discrimination, measured using area under the curve. The tool’s discrimination was then compared to that of each biomarker on its own.
During external validation, the model achieved an area of 0.89 when tasked with differentiating between early-stage pancreatic cancer and benign periampullary diseases. The tool also significantly outperformed both CA19-9 and bilirubin.
Further, in a subset of patients without elevated tumor marker levels, the model achieved an area under the curve of 0.84.
The researchers also found that at a risk threshold of 30 percent, decision curve analysis showed that performing biopsies based on the prediction model was equivalent to reducing unnecessary biopsies by six percent, without missing any patients with early-stage pancreatic cancer.
The research team concluded that these findings suggest the model’s potential to assess the added diagnostic and clinical value of novel biomarkers while preventing potentially unnecessary invasive diagnostic procedures for low-risk patients.
This research highlights recent efforts to use advanced technologies to forecast pancreatic cancer risk.
In May, investigators from Harvard Medical School and the University of Copenhagen created an artificial intelligence (AI) model capable of detecting pancreatic cancer up to three years before diagnosis.
The model uses routinely collected clinical data to make its predictions, expanding the potential for broader, population-based screening for pancreatic cancer.
The researchers hope that such a model could improve the detection and treatment of the disease, as pancreatic cancer is particularly aggressive and a significant driver of cancer mortality.
